Hydrology Research by IWA Publishing

Vol 51 | Issue 5 | October 2020

Hydrology Research Advances in Eco-hydrology and Watershed Water Resources Management in China

ISSN Online 1234-5678 iwaponline.com/hr

Editorial Board Editors-in-Chief

US Mail Identification Statement

Bjørn Kløve Water, Energy and Environmental Engineering Research Unit, Faculty of Technology, 90014 University of Oulu, Finland E-mail: bjorn.klove@oulu.fi

Hydrology Research is published six times a year, in February, April, June, August, October and December. IWA Publishing is a company registered in the UK, no. 3690822.

Nevil Wyndham Quinn Department of Geography & Environmental Management, Faculty of Environment & Technology, University of the West of England, Bristol, Bristol BS16 1QY, UK E-mail: Nevil.Quinn@uwe.ac.uk

US Publishing office: Hydrology Research, c/o Mercury Airfreight International Ltd, 365 Blair Road, Avenel, NJ 07001, USA Periodicals postage paid at Rahway, NJ, USA Postmaster: send address corrections to Hydrology Research, c/o Mercury Airfreight International Ltd, 365 Blair Road, Avenel, NJ 07001, USA

Editor Emeritus Chong-Yu Xu Department of Geosciences, University of Oslo, Sem Sælands vei 1, 0316 Oslo, Norway E-mail: c.y.xu@geo.uio.no

Associate Editors Andrew Barton, Federation University, Australia Jie Chen, Wuhan University, China Harry Dixon, UK Centre for Ecology and Hydrology, UK Michał Wojciech Kawecki, UK Emmanouil Varouchakis, Technical University of Crete, Greece Elena Volpi, Università Roma Tre, Italy Przemysław Wachniew, AGH University of Science and Technology, Poland Ke Zhang, Hohai University, China

Young Career Researcher Editorial Board Tanveer Adyel, Monash University, Australia Zheng Duan, Lunds Universitet, Sweden Masoud Irannezhad, Southern University of Science and Technology, China Mahboobeh Kiani-Harchegani, Yazd University, Iran Hong Li, Oslo Metripolitan University, Norway Kabir Rasouli, Environment and Climate Change Canada, Canada Prabin Rokaya, University of Saskatchewan, Cananda Hossein Tabari, KU Leuven, Belgium Cristina Valhondo, University de Montpellier, France

Editorial Board Azadeh Ahmadi, Isfahan University of Technology, Iran Paul Bates, University of Bristol, UK Daniela Biondi, Università della Calabria, Italy Andrew R. Black, University of Dundee, UK Günter Blöschl, Technical University of Vienna, Austria Michael Bruen, University College Dublin, Ireland Markus C. Casper, University of Trier, Germany Lu Chen, Huahzong University of Science and Technology, China Hannah Cloke, University of Reading, UK Barry F.W. Croke, The Australian National University, Australia Christophe Cudennec, INRA, France Ravinesh Deo, University of Southern Queensland, Australia Weili Duan, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, China Guangtao Fu, University of Exeter, UK Efi Foufoula-Georgiou, University of Minnesota, USA Sigurdur M. Gardarsson, University of Iceland, Iceland Rien van Genuchten, Federal University of Rio de Janeiro, Brazil Jamie Hannaford, Centre for Ecology and Hydrology, UK Nilgun B. Harmancioglu, Dokuz Eylul University, Turkey Denis Hughes, Rhodes University, South Africa Thomas Kjeldsen, University of Bath, UK Ozgur Kisi, Ilia State University, Georgia Harri Koivusalo, Aalto University, Finland George Kuczera, The University of Newcastle, Australia Ian Littlewood, IGLEnvironment, UK Mario Martina, School of Advanced Studies (IUSS), Italy Neil McIntyre, The University of Queensland, Australia Eva Nora Paton, TU Berlin, Germany Domenico Pianese, Universita' degli studi di Napoli, Italy John W. Pomeroy, University of Saskatchewan, Canada Roberto Ranzi, University of Brescia, Italy Liliang Ren, Hohai University, China Noori Roohollah, University of Tehran, Iran Dan Rosbjerg, Technical University of Denmark, Denmark K.D. Sharma, National Institute of Hydrology, India Kuniyoshi Takeuchi, ICHARM, PWRI, Japan Valery Vuglinsky, State Hydrological Institute, Russia Guoqiang Wang, Beijing Normal University, China Howard Wheater, University of Saskatchewan, Canada Stefanos Xenarios, Nazarbayev University, Kazakhstan Xing Yuan, Nanjing University of Information Science and Technology, China Qi Zhang, Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences, China

Aims and Scope It is the principal aim of the Journal to conform to the highest possible scientific and technical standards. Hydrology Research publishes articles within hydrology in its widest sense, with an emphasis on water quantity and quality aspects of the hydrological cycle. Papers that draw upon adjoining sciences are embraced as well. Hydrology Research is intended to be a link between basic hydrological research and practical application of scientific results within the broad field of water management.

Submission All manuscripts should be submitted electronically through our Online Submission and Peer Review System: http://www.editorialmanager.com/hydrology. Authors should prepare their papers in accordance with the current Instructions for Authors, available at https://iwaponline.com/hr/pages/Instructions_for_authors or from the Editorial Office at the address given inside the back cover.

Published in partnership with: Nordic Association for Hydrology Nordic Association for Hydrology (NHF) is an independent body aiming at promoting hydrology as a science, increasing the understanding of hydrology, and enhancing the application of hydrological methods within applied sciences and water management, generally and in the North. The Association strongly supports international hydrological co-operation. More information is available at http://www.nhf-hydrology.org/

British Hydrological Society The British Hydrological Society (BHS) was formed in 1983 in response to a clear need in the UK for a new, broad-based, national society for the advancement of hydrology. The Society caters for all those with an interest in the inter-disciplinary subject of hydrology. It aims to promote interest and scholarship in scientific and applied aspects of hydrology, and to foster the involvement of its members in national and international activities. More information is given at http://www.hydrology.org.uk/

Official Journal of: The Italian Hydrological Society (IHS): http://www.sii-ihs.it/ The German Hydrological Society (DHG): http://www.dhydrog.de/

Hydrology Research An International Journal volume 51 |

issue 5 | October 2020

Contents Advances in Eco-hydrology and Watershed Water Resources Management in China 833 Editorial: Advances in Eco-hydrology and Watershed Water Resources Management in China Guoqiang Wang 834 An improved routing algorithm for a large-scale distributed hydrological model with consideration of underlying surface impact Jingjing Li, Haoyuan Zhao, Jun Zhang, Hua Chen, Chong-Yu Xu, Lu Li, Jie Chen and Shenglian Guo 854 Identification of regional water security issues in China, using a novel water security comprehensive evaluation model Jiping Yao, Guoqiang Wang, Baolin Xue, Gang Xie and Yanbo Peng 867 Copula-based drought severity-area-frequency curve and its uncertainty, a case study of Heihe River basin, China Zhanling Li, Quanxi Shao, Qingyun Tian and Louie Zhang 882 Effects of antecedent soil water content on infiltration and erosion processes on loessial slopes under simulated rainfall Lan Ma, Junyou Li and Jingjing Liu 894 Assessment of hydrological drought based on nonstationary runoff data Xueli Sun, Zhanling Li and Qingyun Tian 911 Spatiotemporal distributions and ecological risk assessment of pharmaceuticals and personal care products in groundwater in North China Jin Wu, Jingchao Liu, Zenghui Pan, Boxin Wang and Dasheng Zhang 925 Impacts of bias nonstationarity of climate model outputs on hydrological simulations Yu Hui, Yuni Xu, Jie Chen, Chong-Yu Xu and Hua Chen 942 Drought prediction models driven by meteorological and remote sensing data in Guanzhong Area, China Jianzhu Li, Siyao Zhang, Lingmei Huang, Ting Zhang and Ping Feng

Contents

Hydrology Research

51.5

2020

959 The influences of sponge city construction on spring discharge in Jinan city of China Kangning Sun, Litang Hu and Xiaomeng Liu 976 Quantification of climate change and land cover/use transition impacts on runoff variations in the upper Hailar Basin, NE China Yuhui Yan, Baolin Xue, Yinglan A, Wenchao Sun and Hanwen Zhang 994 Analysis for spatial-temporal matching pattern between water and land resources in Central Asia Ying Zhang, Zhengxiao Yan, Jinxi Song, Anlei Wei, Haotian Sun and Dandong Cheng 1009 Different runoff patterns determined by stable isotopes and multi-time runoff responses to precipitation in a seasonal frost area: a case study in the Songhua River basin, northeast China Jie Li, Wei Dai, Yang Sun, Yihui Li, Guoqiang Wang and Yuanzheng Zhai 1023 Water balance changes in response to climate change in the upper Hailar River Basin, China Junfang Liu, Baolin Xue, Yinglan A, Wenchao Sun and Qingchun Guo 1036 The spatial pattern of periphytic algae communities and its corresponding mechanism to environmental variables in the Weihe River Basin, China Yixin Liu, Jiaxu Fu, Dandong Cheng, Qidong Lin, Ping Su, Xinxin Wang and Haotian Sun 1048 Spatiotemporal variation and tendency analysis on rainfall erosivity in the Loess Plateau of China Yongsheng Cui, Chengzhong Pan, Chunlei Liu, Mingjie Luo and Yahui Guo 1063 Potential impact of water transfer policy implementation on lake eutrophication on the Shandong Peninsula: a difference-in-differences approach Jia He, Jiping Yao, Aihua Li, Zhongxin Tan, Gang Xie, Huijian Shi, Xuan Zhang, Wenchao Sun and Peng Du 1077 Succession of phytoplankton in a shallow lake under the alternating influence of runoff and reverse water transfer Qing Li, Guoqiang Wang, Zhongxin Tan and Hongqi Wang 1091 A framework for event-based flood scaling analysis by hydrological modeling in data-scarce regions Jianzhu Li, Kun Lei, Ting Zhang, Wei Zhong, Aiqing Kang, Qiushuang Ma and Ping Feng 1104 Response of redox zonation to recharge in a riverbank filtration system: a case study of the Second Songhua river, NE China Xiaosi Su, Yaoxuan Chen, Hang Lyu, Yakun Shi, Yuyu Wan and Yiwu Zhang 1120 Coincidence probability of streamflow in water resources area, water receiving area and impacted area: implications for water supply risk and potential impact of water transfer Xingchen Wei, Hongbo Zhang, Vijay P. Singh, Chiheng Dang, Shuting Shao and Yanrui Wu 1136 The exploration of a Temporal Convolutional Network combined with Encoder-Decoder framework for runoff forecasting Kangling Lin, Sheng Sheng, Yanlai Zhou, Feng Liu, Zhiyu Li, Hua Chen, Chong-Yu Xu, Jie Chen and Shenglian Guo

iii

Contents

Hydrology Research

51.5

2020

1150 Impact of urbanization on variability of annual and flood season precipitation in a typical city of North China Peijun Li, Depeng Zuo, Zongxue Xu, Xiaoxi Gao, Dingzhi Peng, Guangyuan Kan, Wenchao Sun, Bo Pang and Hong Yang 1170 Effect of initial plant density on modeling accuracy of the revised sparse Gash model: a case study of Pinus tabuliformis plantations in northern China Yiran Li, Xiaohua Liu, Chuanjie Zhang, Zedong Li, Ye Zhao and Yong Niu 1184 Entropy weight method coupled with an improved DRASTIC model to evaluate the special vulnerability of groundwater in Songnen Plain, Northeastern China Bin Wang, Yanguo Teng, Huiqun Wang, Rui Zuo, Yuanzheng Zhai, Weifeng Yue and Jie Yang 1201 A field investigation on rill development and flow hydrodynamics under different upslope inflow and slope gradient conditions Pei Tian, Chengzhong Pan, Xinyi Xu, Tieniu Wu, Tiantian Yang and Lujun Zhang

833

51.5

2020

Editorial: Advances in Eco-hydrology and Watershed Water Resources Management in China

The past decades have witnessed China’s rapid economic

research and management of hydrology, water resources,

development, urbanization and the improvement of people’s

and the water eco-environment. This Special Issue is

well-being. The water shortages caused by population

composed of 25 original articles contributed by the

growth, water pollution and/or climate extremes (drought

participants

and ﬂood), as well as the degradation of aquatic ecosystems,

achievements in the ﬁeld of ‘Advances in Eco-hydrology

who

ofﬁcially

presented

their academic

place pressure on the social development of the whole

and Watershed Water Resources Management in China’.

country. In order to fuel sustainable development, it is

These articles consist of various topics including climate

necessary and urgent to introduce feasible strategies and

change inﬂuences on hydrological processes, evaluation and

policies concerning effective water resources management.

predication of drought and ﬂooding, water pollution control

Therefore, the China Water Forum gathered professional

(surface and ground water), water security and water

brains to motivate sparkling ideas which are recorded in

resources shortages, as well as water environmental man-

this Special Issue.

agement modeling by process-based and deep learning

The China Water Forum, organized by the China

methods. The results and conclusions are expected to

Society of Natural Resources, is one of the most useful

guide global water quantity-quality-ecology management

platforms for scientists to discuss current water-related

and climate change adaptation.

problems in China. The 17th China Water Forum hosted by Beijing Normal University (BNU), China, was held on

Guest Editor

8‒10 November 2019. Over 1,000 professors, engineers

Guoqiang Wang

and graduate students from more than 80 domestic univer-

College of Water Sciences, Beijing Normal University,

sities/institutes attended this forum, who are related to the

Beijing 100875, China

This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying, adaptation and redistribution, provided the original work is properly cited (http://creativecommons.org/licenses/by/4.0/). doi: 10.2166/nh.2020.200

834

51.5

2020

An improved routing algorithm for a large-scale distributed hydrological model with consideration of underlying surface impact Jingjing Li, Haoyuan Zhao, Jun Zhang, Hua Chen, Chong-Yu Xu

, Lu Li,

Jie Chen and Shenglian Guo

ABSTRACT Large-scale hydrological models are important tools for simulating the hydrological effect of climate change. As an indispensable part of the application of distributed hydrological models, large-scale flow routing methods can simulate not only the discharge at the outlet but also the temporal and spatial distribution of flow. The aggregated network-response function (NRF), as a scale-independent routing method, has been tested in many basins and demonstrated to have good runoff simulation performance. However, it had a poor performance and produced an unreasonable travel time when it was applied to certain basins due to a lack of consideration of the influence of the underlying surface. In this study, we improve the NRF routing method by combining it with a velocity function using a new routing parameter b to reflect the wave velocity’s sensitivity to slope. The proposed method was tested in 15 catchments at the Yangtze River basin. The results show that it can provide better daily runoff simulation performance than the original routing model and the calibrated travel times in all catchments are more reasonable. Therefore, our proposed routing method is effective and has great potential to be applied to other basins. Key words

| DEM-based routing method, distributed hydrological model, NRF routing method, velocity function, WASMOD, Yangtze River basin

HIGHLIGHTS

• • • • •

This study coupled the aggregated network-response function (NRF) ﬂow routing method with a velocity function. The improved NRF method reﬂects the sensitive wave velocity to slope and gets better runoff simulation performance. The calibrated travel time is closer to benchmark value after improvement. The improved NRF method adapts to basins of various underlying surfaces. The improved NRF method gets more reasonable wave velocity after improvement.

Jingjing Li Haoyuan Zhao Hua Chen (corresponding author) Jie Chen Shenglian Guo State Key Laboratory of Water Resources and Hydropower Engineering Science, Wuhan University, Wuhan, China E-mail: chua@whu.edu.cn Jun Zhang Bureau of Hydrology, Changjiang Water Resources Commission, Wuhan, China Chong-Yu Xu Department of Geosciences, University of Oslo, Oslo, Norway Lu Li NORCE Norwegian Research Centre, Bjerknes Centre for Climate Research, Jahnebakken 5, 5007 Bergen, Norway

835

J. Li et al.

The improvement study of the NRF ﬂow routing method

Hydrology Research

51.5

2020

INTRODUCTION Climate change has received widespread attention from the

water to the total land area can be calculated based on the

scientiﬁc community and the public, as it has caused a series

water storage in each cell, it will contribute to land–atmosinteraction

simulation

(Olivera

).

of serious problems, including natural disasters and extreme

phere

climate events (Kharin et al. ; Chen et al. ; Lu & Qin

However, the choice of a proper spatial resolution is a

; Ragettli et al. ). As tools for estimating regional

dilemma for the cell-to-cell routing method. In distributed

and global water resources and predicting the hydrological

hydrological models, high resolution can reﬂect the hetero-

response

al.

hydrological

geneity of the hydrological characteristics of the basin and

models have become a hot topic recently (Müller Schmied

improve the accuracy of calibrated parameters. On the

& Döll ; Wang et al. ; Abou Rafee et al. ; Li

other hand, higher-resolution DEM will lead to a signiﬁcant

et al. ). A large-scale routing method is an important

increase in the amount of required computations, especially

part of a large-scale hydrological model, with which a dis-

when coupled with a meteorological model because the

tributed hydrological model can simulate the change in

meteorological model has vast computational cost. However,

climate change,

large-scale

runoff at temporal and spatial scales (Beven ). Many

a coarser resolution may lead to a decrease in model perform-

routing methods based on a digital elevation model (DEM)

ance due to not considering routing within the cell (Arora

have been developed (Wen et al. ; Li et al. ; Lu

et al. ; Gong et al. ). Besides this, there may be a

et al. ; Huang & Lee ; Ling et al. ; Fan et al.

great difference between the network that is generated at a

). The main steps of runoff routing in distributed hydro-

coarse resolution and the true one, which is another disad-

logical models are (1) extracting the ﬂow network and (2)

vantage of the cell-to-cell routing method. Attempts have

calculating the runoff routing according to the ﬂow path

been made to bring a network that is generated at a coarse

(Wen et al. ). Generally, runoff routing based on a

resolution closer to the real one. However, a gap still exists

DEM is calculated in one of two ways (Olivera et al. ):

(Fekete et al. ; Guo et al. ).

cell-to-cell and source-to-sink. The cell-to-cell method calcu-

Some of the source-to-sink routing methods have the

lates the water movements from each grid to its adjacent

same disadvantages as the cell-to-cell method, including

downstream grid until the outlet grid of the basin is reached.

the routing models by Ducharne et al. () and Guo

The outﬂow of each cell is calculated based on the inﬂow and

et al. (), which will provide worse performance as the

the river channel’s storage. The source-to-sink method, under

cell resolution becomes coarser. The routing model by

the assumption that the water is rigid, directly calculates the

Ducharne et al. () uses the mean travel time of each

water’s movement from the grid where runoff is generated to

large routing cell, so the water in a cell will arrive at the

the outlet of the basin.

outlet in the same day. This can be unreasonable, especially

The ﬁrst of these two kinds of routing methods is based

when the cell size is large. The routing model by Guo et al.

on the mass conservation equation and the channel storage

() uses the mean ﬂow path length for each cell, making

function equation. It is widely used due to its simplicity

the routing method scale-dependent. Therefore, it would

(Miller et al. ; Sausen et al. ; Arora & Boer ;

beneﬁt a hydrological modeler to ﬁnd a routing method

Huang & Lee ). In addition to its simplicity, the cell-

with the capability to be scale-independent, which will pre-

to-cell routing method has other advantages. It has the

serve the spatially distributed travel time information from

potential to include the nonlinear losses from the surface

a ﬁner-resolution DEM for coarser resolution cells in the

water to the groundwater because it considers the feedback

large-scale hydrological model. The disadvantages of some

between the ﬂow and the water storage in each cell (Naden

source-to-sink routing methods have been overcome by

et al. ). Besides this, with the cell-to-cell routing method,

some model developers. For instance, Wen et al. () pro-

the water storage in any cell of interest can be queried at

posed a multiscale routing framework that can reduce the

each time step. If the ratio of land area covered by surface

impact of the spatial scale by using histograms for the ﬂow

836

J. Li et al.

The improvement study of the NRF ﬂow routing method

path lengths. The aggregated network-response function

Hydrology Research

51.5

2020

STUDY AREA AND DATA

(NRF) routing method (Gong et al. ) can preserve all of the spatially distributed travel time information in

Study area

high-resolution Hydro1 k DEM data for cells with different resolutions by aggregating the pixel-response function to

The study area is the Yangtze River basin (24 270 –35 540 N,

the cell-response function (CRF). Thus, NRF has the

90 330 –122 190 E). It has a drainage area of 1,800,000 km2.

unique advantage of being able to obtain a stable result

As shown in Figure 1, the Yangtze River is the longest river

when moving from a ﬁner to a coarser resolution, making

in China and the third longest river in the world. It has a

it a preferable routing method for large-scale hydrological

total length of 6,397 km.

models. Another advantage of source-to-sink methods is

The climate, terrain, land cover and land use, and runoff

the computation efﬁciency (Naden et al. ; Olivera

characteristics vary greatly among the catchments in the

et al. ; Gong et al. ), which enables them to be

Yangtze River basin. Fifteen catchments in the Yangtze

widely applied (Olivera et al. ; Ducharne et al. ;

River basin were selected in this study in order to evaluate

Gong et al. ; Lu et al. ).

the performance of the improved routing method over a

NRF, as a source-to-sink routing method, has been

wide range of representative catchments. The characteristics

applied in basins of China, North America, and Africa

of the terrain in those catchments are shown in Table 1.

(Gong et al. , ; Li et al. , , ; Ngongondo

‘Elevation difference’ means the height difference between

et al. ). The results showed that NRF provides good daily

the highest point and the lowest point in a catchment.

runoff simulation performance. However, in some cases, the

Table 1 also presents the shape factor and the river length

calibrated wave velocity is too large to be physically realistic,

of the catchments. The basin shape factor (Morisawa )

and NRF will fail to provide good daily runoff simulation

is the ratio of the actual basin length to the perimeter of a

performance if the wave velocity becomes smaller (Gong

circle whose area is the same as that of the basin, and

et al. ; Li et al. ). In this method, the underlying fac-

measures the general runoff concentration behavior of the

tors that affect the routing process are not comprehensively

basin. Only the ﬁrst-order stream, second-order stream,

considered. In this respect, the NRF routing method needs

and third-order stream (Wang et al. ) were included in

to be improved.

the stream length in this study. A drainage map of the

This study aims to improve NRF by considering the

river system and drainage stations can be found in Figure 1.

inﬂuence of the underlying surface in order to obtain a

The Yangtze River basin has very complex terrain, as it

more reasonable wave velocity and travel time and better

ﬂows through mountains, plateaus, hills, and plains. The

daily runoff simulation performance. To achieve this aim,

overall distribution of elevation at the Yangtze River is

the original NRF routing method is coupled with a velocity

high in the west and low in the east. The mean elevation

function proposed by Sircar et al. () in order to take the

of the source of the Yangtze River, which is located in the

underlying surface’s inﬂuence into consideration, in which a

hinterland of the Qinghai–Tibet Plateau, is more than

new routing parameter b is added to reﬂect how sensitive

4,000 m above sea level (a.s.l.). The middle reaches lie

the wave velocity is to slope. The remainder of this paper

mostly in low mountains areas and the downstream lies

is organized as follows. Following a description of the

mostly in a hilly plain area.

study area, we present the original and improved NRF rout-

The mean annual precipitation is 1,110 mm. Due to the

ing methods. Then, the results of a daily runoff and peak

vast amount of territory, the complex terrain, and the typical

ﬂow simulation by the model before and after the improve-

monsoon climate, the spatial distribution and the temporal

ment

routing

distribution of annual precipitation in the Yangtze River

parameters and the factors that inﬂuence them are analyzed

are very uneven (Xu et al. ). The main source of river

and compared in different catchments to explore the

runoff in the basin is the rainfall, but snowmelt and ground-

relationship between catchment characteristics and model

water are also important sources of runoff in the upper

parameters. Finally, we present our conclusions.

reaches of the Yangtze River basin (Yao et al. ). The

are

compared.

The

travel

time

and

837

Figure 1

J. Li et al.

The improvement study of the NRF ﬂow routing method

Hydrology Research

51.5

2020

Location of the sub-basins and discharge stations in the Yangtze River basin.

annual mean air temperature in the Yangtze River basin is

Geological Survey’s Earth Resources Observing System.

generally high in the east, low in the west, high in the

The projection of HYDRO1 k is Lambert azimuthal equal-

south, and low in the north. The source area of the basin

area projection, whose transformation equations (Weisstein)

experiences the lowest temperatures (Su et al. ).

enable each grid to have an equal area.

It is of great signiﬁcance to simulate the runoff in the

Flow direction and ﬂow accumulation databases based

Yangtze River basin, as it accounts for 18% of the total

on Hydro1 k are also available, which can be used with

area of China, and has more than 400 million people

the discharge stations to obtain the basin’s boundary.

living in it. The Yangtze River basin is also the most impor-

Then, the estimated basin boundary can be evaluated and

tant agricultural production base in China because of its

corrected based on the basin boundary that was delineated

warm and humid climate, which provides good conditions

using the ﬁner DEM at the resolution of 90 × 90 m.

for crop growth (Tian et al. ).

Daily precipitation, temperature, and relative humidity data for the period 1969–2008 were taken from the daily data-

Data and processing

set of surface climatological data of China (V 3.0) in National Meteorological Information Center (http://data.cma.cn/

Hydro1 K (USGS ) was used in this study to delineate

data/cdcdetail/dataCode/SURF_CLI_CHN_MUL_DAY_V3.

the basin’s boundary. It is a global DEM with a resolution

0.html). The meteorological data were interpolated using the

of 1 × 1 km and was produced by the United States

inverse distance weighted method (Shepard ) to be

J. Li et al.

838

Table 1

The improvement study of the NRF ﬂow routing method

Hydrology Research

51.5

2020

Characteristics of the study area

Mean discharge (m3/s)

Area (km2)

Jinsha River basin Downstream of Shigu Pingshan Upstream of Shigu Shigu

4,394.49 1,275.61

Mintuo River basin

Fushun Gaochang

Jialing River basin Jialing River Wu River basin

Wujiang

Han River basin Dongting Lake basin Poyang Lake basin

Sub-basin

Mean Slope elevation (m)

Elevation difference (m)

River length (km)

252,217 3.10 213,003 3.66

7.45 3,042 4.87 4,569

5,620 4,124

5,430 3,559

360.91 2,750.02

23,096 2.39 134,622 2.47

1.98 590 8.88 2,949

4,523 7,087

724 2,924

Beibei

1,841.34

157,478 2.44

6.29 1,148

2,948

3,124

Wulong

1,415.63

80,099

2.71

3.91 1,104

2,628

1,237

Hanjiang

Baihe

686.30

54,085

2.06

6.29 1,148

2,948

1,462

Xiangjiang Yuanjiang Lishui

Xiangtan Taoyuan Shimen

1,990.02 2,028.19 456.86

78,659 86,172 15,118

2.19 2.55 1.96

3.02 338 3.32 622 5.12 722

1,857 2,378 2,204

2,200 2,210 332

Gangjiagn Fuhe Xinjiang Leanhe Changhe

Waizhou Lijiadu Meigang Shizhenjie Dufengkeng

2,032.18 376.80 557.69 296.61 155.62

83,532 16,023 15,592 8,294 4,938

2.27 2.04 2.08 1.78 1.73

2.46 2.33 3.17 2.74 3.01

2,026 1,415 2,040 1,421 1,361

1,454 432 323 224 156

River

Tuojiang Minjiang

Station

Shape factor

287 227 279 205 246

gridded data, which were used to drive the hydrological

described in Gong et al. (), the main components of

model at resolutions of 3, and 10 arc-minutes in the study.

WASMOD are the computation of snow accumulation and

The observed daily discharge during 1964–2008 of all catch-

melting, evapotranspiration, slow ﬂow, fast ﬂow, and the

ments, which was used to calibrate and validate the model,

total runoff. Compared with the equations in Gong et al.

was provided by the Bureau of Hydrology of the Changjiang

(), we set two more parameters in this study: c5 is set

Water Resources Commission.

to correct the precipitation bias, which results from the

The travel time of a ﬂood between major hydrological

lack of snow measurements; c4 is set in the potential evapor-

stations in the Yangtze River basin was taken from Zhang

ation equation. As shown in Table 2, there are ﬁve to seven

et al. (), which was calculated based on a great deal of

parameters with or without the snow module. The snow

data and some proven methods. In this study, it was used

modules are only applied to Jialing River, Jingsha River,

as a benchmark to compare the travel time calculated by

and Mintuo River where snowmelting is one of the most

the original routing method with the travel time calculated

important sources of runoff. If the snow module is not

by the improved routing method.

taken into consideration, parameters a1 and a2 will not be

METHODS

Table 2

The parameters of the runoff-generation model and their prior range for calibration

Water and Snow Balance Modeling The Water and Snow Balance Modeling (WASMOD) system is a conceptual modeling system that was developed by Xu (). Different versions of WASMOD have been used in different regions of the world (Widén-Nilsson et al. ; Gong et al. ; Kizza et al. ; Li et al. ; Yang et al. , , ). In this study, the daily version of WASMOD was used as a runoff-generation model. As

Parameter

About

Prior range for calibration

a1 ( C)

Snowfall

[0 to 6]

a2 ( C)

Snowmelt

[ 6 to 0]

a4 (–)

Actual evaporation

[0.1 to 0.999]

c1 (1/mm)

Fast runoff

[0 to 0.1]

c2 (1/mm)

Slow runoff

[0 to 0.1]

c4 (mm day 1 C 2) Potential evaporation [6 × 10 6 to 6 × 10 5] c5 (–)

Precipitation

[0.5 to 1.5]

839

J. Li et al.

The improvement study of the NRF ﬂow routing method

Hydrology Research

51.5

2020

For a given grid, the travel time can be calculated as

calibrated in the hydrological model.

follows: pt ¼ c 5 ptd

(1)

where ptd is precipitation data described in the Study area and data section and pt is the corrected precipitation. pett ¼ c4 ( max (tmpt þ 25, 0))2 (100 rht )

8 k P > > > <

0:5 i¼1 v45 (tan (ci )) ti ¼ k > P li > > : 0:5 i¼1 v45 (tan (c0 ))

ci > c0 (5) ci c0

(2) where ci is the slope, vi is the wave velocity, ti is the travel

where pett is the potential evaporation, tmpt is the temperature, and rht is the relative humidity.

time of the grid, c0 is the slope threshold, which is set to be a constant to prevent ti from being inﬁnite, and k is the number of grids through the ﬂow path. The slope threshold is 0.1 in this study.

The original routing method

Other details about the NRF routing methods can be Gong et al. () developed a routing method, named the aggregated NRF, which was developed with the source-tosink concept. It builds the relationship between each grid in DEM and the outlet grid. There are five steps in the improved NRF routing method: (1) extract the flow direction and flow net based on DEM; (2) calculate the travel time ti between each grid and the outlet of the basin according to the flow

seen in Gong et al. (). In this study, the computation cell size in each catchment was determined by its area to ensure a modest number of cells. The cell size in a larger catchment is larger. Conversely, the cell size is smaller in a smaller catchment. The cell size was set to be 100 at catchments whose area is greater than 50,000 km2. It was set to be 30 at catchments whose area is less than 10,000 km2 and 60 otherwise.

path; (3) construct the pixel-response function (PRF) based on ti; (4) calculate CRF based on PRF with the way of linear averaging method; and (5) calculate the discharge

The improved routing method

at the outlet of the basin based on the result of the runoffgeneration model and CRF. The main advantage of NRF

The NRF routing method performed well in the Dongjiang

is that the routing accuracy is independent of cell size,

basin and the Willamette basin; however, the calibrated par-

because the CRF in NRF is not affected by the cell size

ameter (v45 ) in the Hydro1 k-driven NRF routing method

(Gong et al. ).

showed an unrealistic value (greater than 25 m/s) (Gong

For the wave velocity in grid, Beven & Kirkby ()

et al. , ). To improve the NRF, we combine the orig-

proposed that the overland ﬂow velocity could be calculated

inal routing method with a wave function given by Sircar

based on the topography as follows:

et al. () and modiﬁed the equation as:

vi ¼ v45 tan (ci )

(3)

vi ¼ v45 ( tan (ci ))b

where vi is the wave velocity of the grid, and ci is the slope of the grid. v45 is the wave velocity in the grid whose slope is

45 . In the study of Gong et al. (), the authors calculate the wave velocity as follows: vi ¼ v45

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi tan (ci )

ti ¼

8 k P > > > <

i¼1 v45

k > P > > :

i¼1 v45

li (tan (ci ))b li (tan (c0 ))b

(6) ci > c0 (7) ci c0

where b is a parameter that reﬂects how sensitive the wave velocity is to slope. The value of the parameter b is related to (4)

the condition of the underlying surface of the catchment.

J. Li et al.

840

The improvement study of the NRF ﬂow routing method

Hydrology Research

51.5

2020

Compared with Equation (4), Equation (6) is more gen-

To measure the peak ﬂow simulation performance, the

eral and more physically realistic. Because different basins

time lag and relative error of peak ﬂow were also calculated.

have different underlying conditions, it cannot be known how sensitive the wave velocity is to slope. In this case, b varies in different basins. Equation (6) reflects the relationship between wave velocity and slope more flexibly and can more easily reflect the characteristics of different basins. Thus, Equation (5) is substituted with Equation (7) in the improved NRF routing method. There are two parameters in the improved NRF routing method (referred to as ‘after’ in the following) and one parameter in the original NRF routing method (referred to as ‘before’ in the following). Table 3 shows the routing parameter value sets for calibration. They were chosen based on computer capability limitations and the physical meaning of each parameter. There are 7 routing parameter value sets before and 42 sets after, composed of six kinds of parameter

Qre ¼

ΔT ¼ Tsim Tobs

warm-up period was 1 year for all catchments. The Nash–Sutcliffe (NSE) efﬁciency (Nash & Sutcliffe ) was used as the criterion for evaluating overall runoff simulation performance in this study.

the simulated maximum peak flow, Qobs max is the observed and Tobs is the day when Qobs max appears. ΔT is the time lag of the simulated peak flow relative to the observed peak flow. Qre and ΔT of all selected flood events in a catchment were averaged to reflect the overall simulation performance of the peak flow in that catchment as follows:

Qre ¼

l 1X jQre,j j l j¼1

(11)

ΔT ¼

l 1X jΔTj j l j¼1

(12)

where ΔT is the averaged time lag of the peak flow in the catchment, Qre is the averaged relative error of the peak flow in the catchment, and l is the number of flood events in the catchment. The parameters of the routing method were calibrated

NSE ¼ 1

(10)

maximum peak ﬂow, Tsim is the day when Qsim max appears,

Evaluation criteria

ment was calibrated and validated in all catchments. The

(9)

where Qre is the relative error of the peak ﬂow, Qsim max is

b and seven kinds of parameter v45 .

The model before and after the routing method improve-

Qsim max Qobs max Qobs max

2 i¼1 (Oi Si ) Pn 2 i¼1 (Oi Oi )

(8)

using the Monte Carlo algorithm (Barraquand & Latombe ), and the parameters of the runoff-generation model were calibrated using Covariance Matrix Adaption Evol-

where Oi is the observed runoff, Si is the simulated runoff,

ution (CMAES) (Hansen & Ros ). The Monte Carlo

and n is the length of the time step.

algorithm was used to search for the best routing parameter

Table 3

The parameters of the routing methods and their prior range for calibration

set. First, 300 runoff-generation parameter value sets were produced using LATIN-Hypercube sampling (Mckay et al.

Parameter

Explanation

Values

) according to the prior range of the parameters.

Before

v45 (m/s)

The wave velocity of a grid whose slope is 45

4, 5, 6, 7, 8, 9, 10

Table 3 shows the routing parameter sets. Second, for

After

b (–)

0.2, 0.25, 0.3, Power exponent reﬂecting 0.35, 0.4, how sensitive is the v to 0.45 slope v45 (m/s) The wave velocity of a grid 4, 5, 6, 7, 8, 9, whose slope is 45 10 The combination of all values yields 42 parameter sets

ameter value sets were used to drive the hydrological model,

each routing parameter set, all of the runoff-generation parand then the NSE was calculated based on the simulated runoff and the observed runoff. Third, the best routing parameters were selected based on the NSE. Then, the Covariance Matrix Adaption Evolution algorithm was used

841

J. Li et al.

The improvement study of the NRF ﬂow routing method

Hydrology Research

51.5

2020

to search for the best runoff-generation parameters under

period of the 15 catchments is 0.11, with a maximum of

the condition that the routing parameters are ﬁxed.

0.44 in the calibration period and 0.43 in the validation period. From Figure 2, it can also be found that the improvement in NSE varies greatly in the 15 catchments of the

RESULTS AND DISCUSSION

Yangtze River basin after the improvement of the routing method.

Modeling performance assessment with the improved

In order to demonstrate the difference in the simulated and observed discharge hydrographs of the original and

NRF

improved routing methods, daily hydrographs of a represenBoth routing methods were taken as the routing module of

tative year 1968 for the Shimen catchment are plotted in

the WASMOD. To all catchments, the model was calibrated

Figure 3. It shows that with the original NRF routing

by using 7 years’ data series and validated by using another 4

method, the simulated ﬂood peak is often later than the

years’ data series during the period of 1964–2008. Figure 2

observed ﬂood peak. The improved NRF can better simulate

shows the performance of the daily runoff simulation

the ﬂood peak. Tables 4 and 5 show the optimal parameters

before and after the improvement.

of the hydrological model with the original and the

Figure 2 shows that the model with the improved NRF

improved routing method.

routing method performs better than the model with the

In order to assess the ﬂood simulation performance of

original NRF routing method for all of the study catchments

the model with the improved NRF routing method, the aver-

to varying extents. Taking the Baihe catchment and the

aged relative errors Qre and averaged time lags ΔT of all

Waizhou catchment as examples, the Baihe catchment has

ﬂood events in each catchment were calculated using

an NSE of 0.87 in the calibration period and 0.74 in the vali-

Equations (11) and (12) and are shown in Figure 4. The

dation period after the improvement, while the NSE is 0.43

ﬂood events whose peak ﬂow was greater than three times

in the calibration period and 0.31 in the validation period

the average runoff of catchment were selected in this

before the improvement. The Waizhou catchment has an

study. It can be seen that the model with the improved

NSE of 0.82 in the calibration period and 0.78 in the vali-

NRF routing method provides better ﬂood simulation per-

dation period after the improvement, while the NSE is

formance than the model with the original NRF routing

0.58 in the calibration period and 0.59 in the validation

method. After the improvement, the averaged time lags of

period before the improvement. The mean improvement in

the peak ﬂow in almost all catchments are less than one

the NSE in both the calibration period and the validation

day, except for the Shigu and Waizhou catchments (Figure 4,

Figure 2

NSE in (a) the calibration period and (b) the validation period in all catchments by the NRF routing method before and after the improvement.

J. Li et al.

842

Figure 3

Table 4

The improvement study of the NRF ﬂow routing method

Hydrology Research

51.5

2020

The simulated and observed discharge hydrographs for the Shimen catchment in 1968 by the NRF routing method before and after the improvement.

The optimal parameter values of the hydrological model with the original routing method

Catchment

v45

Pingshan

0.5

6.00

0.00

0.65

0.0075

0.00005

0.10

0.58

Shigu

0.5

6.00

0.00

0.99

0.0124

0.00008

0.27

0.50

Fushun

0.5

6.00

1.83

0.93

0.0092

0.00055

0.35

0.78

Gaochang

0.5

5.78

1.51

1.00

0.0064

0.00018

0.10

0.64

Beibei

0.5

0.05

5.00

0.81

0.0145

0.00017

0.12

0.59

Wulong

0.5

—

0.88

0.0087

0.00022

0.41

0.91

Baihe

0.5

—

0.71

0.1000

0.00008

0.15

0.69

Xiangtan

0.5

—

0.95

0.0015

0.00002

0.66

1.11

Taoyuan

0.5

—

0.94

0.0040

0.00007

1.00

1.14

Shimen

0.5

—

0.88

0.0879

0.00575

1.00

0.97

Waizhou

0.5

—

0.96

0.0088

0.00003

0.26

0.71

Lijiadu

0.5

—

0.96

0.0017

0.00001

1.00

1.10

Meigang

0.5

—

0.96

0.0026

0.00001

1.00

1.20

Shizhenjie

0.5

—

0.93

0.0010

0.00001

1.00

1.36

Dufengkeng

0.5

—

0.94

0.0089

0.00046

1.00

0.90

left panel). The averaged relative error of the peak ﬂow in all

distributions histogram of time lags ΔT in each catchment

catchments decreases in nearly all of the catchments

with the original and improved NRF routing method. Fre-

(Figure 4, right panel). The amelioration is most evident in

quency is the number of ﬂood events that ΔT in a certain

the Waizhou catchment, where Qre equals 0.32 with the

value is divided by the total number of ﬂood events in this

original routing method and 0.17 with the improved routing

catchment. When the time lag ΔT is positive, it means that

method. The relative error of the peak ﬂow decreases to the

the simulated ﬂood peak is later than the actual one. On

minimum error range of 0.1 to 0.1 after the improvement

the contrary, when the time lag ΔT is negative, it means

in the routing method in all of the catchments except for

that the simulated ﬂood peak is earlier than the actual

the Dufengkeng and Fushun catchments.

one. When ΔT equals zero, the hydrological model simulates

To evaluate the ability of simulating the occurrence of

the time of ﬂood peak accurately. Figure 5 shows that with

ﬂood peaks, the time lags ΔT in all selected ﬂood events in

the original routing method, the frequency of positive time

each catchment were analyzed. Figure 5 plots the frequency

lag ΔT is larger than the frequency of negative time lag ΔT.

J. Li et al.

843

Table 5

The improvement study of the NRF ﬂow routing method

Hydrology Research

51.5

2020

The optimal parameter values of the hydrological model with the improved routing method

Catchment

v45

Pingshan

0.2

6.00

0.00

0.47

0.0038

0.00015

0.11

0.58

Shigu

0.45

6.00

0.00

0.99

0.0125

0.00012

0.26

0.50

Fushun

0.45

0.99

2.32

0.12

0.0148

0.00224

0.14

0.66

Gaochang

0.2

2.88

6.00

1.00

0.0059

0.00028

0.91

0.64

Beibei

0.3

5.45

1.81

0.50

0.0120

0.00054

0.27

0.74

Wulong

0.45

—

0.88

0.0082

0.00026

0.40

0.90

Baihe

0.3

—

0.81

0.0090

0.00051

1.00

1.14

Xiangtan

0.25

—

0.22

0.0039

0.00012

0.21

0.79

Taoyuan

0.2

—

0.94

0.0040

0.00019

1.00

1.09

Shimen

0.25

—

0.91

0.0123

0.00080

1.00

1.19

Waizhou

0.35

—

0.97

0.0038

0.00007

0.91

0.89

Lijiadu

0.4

—

0.94

0.0044

0.00015

1.00

0.91

Meigang

0.4

—

0.94

0.0081

0.00043

1.00

0.96

Shizhenjie

0.25

—

0.95

0.0078

0.00085

1.00

0.89

Dufengkeng

0.45

—

0.76

0.0124

0.00104

0.24

0.78

Figure 4

The averaged time lag of peak ﬂow (left panel) and relative error of peak ﬂow (right panel) in all catchments by the NRF routing method before and after the improvement.

That indicates that the simulated peak ﬂow is often delayed

model with the original routing method. Those phenomena

compared with the observed peak ﬂow with the original

indicate that the improved routing methods can better serve

NRF routing method. There is a higher frequency of zero

the ﬂood simulation.

time lag of peak ﬂow (ΔT equals zero) in most of the catch-

These results indicate that the model with the improved

ments with the improved routing method than that with the

routing method performs better in both peak ﬂow simu-

original routing method. Particularly, in the Gaochang,

lation and long-term runoff simulation than the model

Shizhengjie, and Meigang catchments, the frequency of

with the original routing method. Besides this, NRF’s advan-

zero time lag of peak ﬂow is greater than 0.8 in the model

tage of being able to preserve the travel time information in

with the improved routing method and less than 0.4 in the

the PRF is maintained in the improved model.

844

Figure 5

J. Li et al.

The improvement study of the NRF ﬂow routing method

The frequency distributions of time lags (day) for all of the catchments by the NRF routing method before and after the improvement.

Hydrology Research

51.5

2020

845

J. Li et al.

The improvement study of the NRF ﬂow routing method

Hydrology Research

51.5

2020

Analysis of the routing time calculated by the improved

by the original routing method and the improved routing

NRF

method. The runoff travel times were compared with the travel time provided by Zhang et al. () as a benchmark.

Figure 6(a) and 6(b) show the distributions of the travel time

For most catchments, the travel time between grids that

of all grids in all catchments that were calculated by the orig-

was calculated by the improved routing method was shorter

inal and the improved routing methods. Figure 6(d) plots the

than that calculated by the original routing method

mean grid travel time and mean slope of all of the catchments.

(Figure 6(a) and 6(b)). Besides this, the travel time between

The simulated travel time was evaluated based on two hydro-

the speciﬁc upstream hydrological station and the outlet

logical stations in each catchment: the outlet station and the

station was shorter and closer to the benchmark value for

speciﬁc upstream hydrological station that made the distance

all the catchments (Figure 6(c)). The main reason why the

between the two hydrological stations as large as possible.

travel time that was calculated by the improved routing

Figure 6(c) plots the runoff travel times between the upstream

method was longer than the benchmark value in some

hydrological station and the outlet station that were calculated

catchments is that the routing parameters v45 and b are

Figure 6

The evaluation of the travel times. (a) Box plots of the travel time in all catchments calculated by the model with the NRF routing method before the improvement and (b) after the improvement; (c) the averaged travel time in all of the catchments simulated by the model before and after the improvement compared with observations.

846

J. Li et al.

The improvement study of the NRF ﬂow routing method

Hydrology Research

51.5

2020

spatially constant over the whole catchment. However, rout-

the runoff travel times in these catchments, as calculated

ing refers to the process in which rainfall in a catchment

using the original routing method, were unrealistic and at

evolves along a certain path into discharge at the outlet.

least twice as long as the runoff travel times that were calcu-

The rainfall is ﬁrst routed by a hillslope and then routed

lated using the improved routing model. On the other hand,

by the river before being transferred into discharge. Hill-

for the Dufengkeng, Fushun, and Wulong catchments, we

slope routing and river routing are two forms of routing

found no obvious change in the travel times that were calcu-

(Kirkby & Beven ). Usually, the runoff travels faster in

lated using the original and improved routing methods. The

river routing than in hillslope routing. However, the hill-

original and improved routing methods also provided a simi-

slope travel time and the river travel time are not

lar NSE in these catchments.

separated in both the original and improved NRF routing methods. The parameters v45 and b reﬂect the state of the

Analysis of the wave velocity calculated by the

whole basin; however, the benchmark value reﬂects the

improved NRF

travel time that is needed in the selected river section. Thus, the travel time in the river that is calculated by the

Figure 7 shows the theoretical curves of wave velocity versus

improved routing method will be a little longer than the

slope. Figure 8(a) and 8(c) plot the calibrated curves that

benchmark value. However, the travel time that was calculated in the Shigu catchment is much longer than the benchmark value, which is due to the Shigu catchment’s underlying surface. The shape factor of the Shigu catchment is the largest of all of the studied catchments, which means that the shape of the Shigu catchment is narrow and long and the water system’s development is immature. The above factors lead to much more hillslope routing and a longer travel time in the Shigu catchment. In addition, the snowmelt and groundwater are the main sources of runoff in the Shigu catchment. The snowmelt runoff moves more slowly down the hillslope than the rainfall runoff, so the optimal routing parameters that were calibrated using the observed runoff lead to a longer travel time in the Shigu catchment compared with the benchmark value, which only considers routing in the river. The travel time that was calculated using the original routing method produces a late peak ﬂow and a smaller amount of ﬂow (Figure 4). An unreasonable travel time is one of the main causes of a hydrological model’s poor performance in long-term runoff simulation. The improved routing method provides a more realistic travel time between the grids compared with the benchmark value. When analyzed in conjunction with Figures 2 and 6, it can be found that the NSE and the travel time in all catchments have a close relationship. From Figure 2, it can be seen that the NSE that was calculated by the original NRF is lower in some catchments, i.e., the Waizhou, Shimen, Beibei, and Baihe catchments. As shown in Figure 6(c),

Figure 7

The theoretical curve of velocity and slope: (a) wave velocity changes with slope with the parameter v45 before the improvement and (b) wave velocity changes with slope with the two parameters v45 and b after the improvement.

847

J. Li et al.

The improvement study of the NRF ﬂow routing method

Hydrology Research

51.5

2020

were calculated by the original routing method and the

larger NSE value. Thus, the calibrated value of v45 reaches

improved routing method (Table 3), respectively. From the

the upper limit of the prior range for calibration in 13

analysis in the section Analysis of the routing time calcu-

catchments, resulting in only three kinds of curves between

lated by the improved NRF, it is known that the improved

slope and calibrated wave velocity in 15 catchments. More-

routing method is able to achieve a more reasonable travel

over, in many basins, the optimal value of v45 in cali2 is

time. The reason for this is that the improved routing

larger than 10 m/s and has a maximum of 31 m/s; thus,

method has a more ﬂexible wave velocity–slope curve

the grid wave velocity is too large to be realistic

(Figure 7) that can better reﬂect the real world.

(Figure 8(e)). Figure 8(b), 8(d) and 8(f) show the distri-

It can be seen from Figure 7(b) that the wave velocity of

butions of the wave velocity of all grids in all catchments

a grid is determined by the parameters b and v45 together. If

before the improvement with the strict prior range of par-

the parameter b remains constant, the wave velocity

ameter v45 for calibration; after the improvement with

increases as v45 increases. If the parameter v45 remains con-

the strict prior range of parameter v45 for calibration;

stant, the wave velocity increases as b decreases at grids with

before the improvement with the wider prior range of par-

a slope of less than 45 ; however, the wave velocity

ameter v45 for calibration. It also shows that the grid wave

increases as b increases at grids with a slope of greater

velocity is too large to be realistic before the improvement

than 45 . The effect that slope has on the wave velocity

with the wider prior range of parameter v45 for calibration.

varies across catchments due to the differences in the

The model can provide good runoff simulation perform-

hydrology of the underlying surface, which can be adjusted

ance

by the parameter b. The bigger b is, the more sensitive the

compensates for the problem of the routing time being

wave velocity is to slope. When b is equal to zero, the

too long due to the unreasonable distribution of the

because

the

excessively

large

value

v45

slope has no effect on wave velocity, which will be a con-

grids’ wave velocity. This indicates that the original routing

stant. A larger wave velocity results in a shorter runoff

method cannot adapt to the variation in the underlying sur-

travel time in a catchment.

face of the different basins. This is consistent with the

Each routing parameter value set refers to a certain

results from Gong et al. (), who found that the optimal

curve between slope and calibrated wave velocity. The orig-

value of v45 was physically unrealistic as it was determined

inal NRF only produces three kinds of curves for the 15

to be greater than 26 m/s with the original Hydro1 k-driven

catchments (Figure 8(a)). The curves of 13 catchments

NRF routing method.

coincide because they have the same optimal value of

The original NRF routing method uses a constant (0.5)

v45 . However, the improved NRF produces a different cali-

to reﬂect the inﬂuence that the slope has on the wave vel-

brated curve for each catchment (Figure 8(c)). Tables 6 and

ocity; however, it fails to adapt to the variation in the

7 present the optimal values of v45 that were calibrated by

underlying surface of the different basins. The improved

the original routing method with a strict range of 4–10 m/s

routing method makes it possible to ﬁt the actual physical

(cali1), as in Table 3, and with a wider prior range of

relationship between the slope and wave velocity in a

4–31 m/s (cali2), respectively. Due to computer capability

more ﬂexible way. The reasonable distribution of wave vel-

limitations, the increment of parameter v45 above 10 m/s

ocity leads directly to a more accurate simulation of the

is 3 m/s. Figure 8(e) plots the calibrated curves between

travel time, which in turn leads to a better simulation of

slope and calibrated wave velocity in cali2. In combination

the runoff.

with the optimal value of v45 shown in Tables 6 and 7, it

The improved NRF routing method belongs to hydrolo-

can be seen that the optimal value of v45 in cali1 is the

gical methods rather than hydraulics methods. There are

same as 10 m/s, reaching the upper limit of the prior

some other hydrological routing methods calculating the

range for nearly all of the catchments. When the prior

travel time of each grid and then routing the runoff to dis-

range for calibration is ﬁxed to be 4–10 m/s, the larger

charge. But the parameter to calibrate and inﬂuence

the value of v45 , the shorter the travel time, which is

factors that have been taken into consideration are different

closer to the reality in most catchments. This results in a

in different methods.

848

Figure 8

J. Li et al.

The improvement study of the NRF ﬂow routing method

Hydrology Research

51.5

2020

Curves and box plots of calibrated wave velocity versus slope (a) and (b) before the improvement with the strict prior range of parameter v45 for calibration, (c) and (d) after the improvement with the strict prior range of parameter v45 for calibration, and (e) and (f) before the improvement with the wider prior range of parameter v45 for calibration.

For the routing method in Ducharne et al. (), only the

also the hydraulic radius, the roughness of the cell and the

slope between two grids will inﬂuence the velocity. For the

scale factor about the cell size will inﬂuence the wave velocity.

routing method in Guo et al. (), not only the slope, but

For the routing method in Bunster et al. (), it proposed a

J. Li et al.

849

Table 6

The improvement study of the NRF ﬂow routing method

Hydrology Research

51.5

2020

The values of the optimal routing parameter v45 with the prior range of 4–10 m/s

Catchment

Pingshan

Shigu

Fushun

Gaochang

Beibei

Optimal value (m/s)

NSE

0.88

0.78

0.80

0.81

0.61

Catchment

Wulong

Baihe

Xiangtan

Taoyuan

Shimen

Optimal value (m/s)

NSE

0.88

0.43

0.77

0.82

0.60

Catchment

Waizhou

Lijiadu

Meigang

Shizhenjie

Dufengkeng

Optimal value (m/s)

NSE

0.88

0.43

0.77

0.82

0.60

Table 7

The values of the optimal routing parameter v45 with a wider prior range of 4–31 m/s

Catchment

Pingshan

Shigu

Fushun

Gaochang

Beibei

Optimal value

NSE

0.88

0.82

0.80

0.89

0.87

Catchment

Wulong

Baihe

Xiangtan

Taoyuan

Shimen

Optimal value

NSE

0.88

0.83

0.85

0.79

Catchment

Waizhou

Lijiadu

Meigang

Shizhenjie

Dufengkeng

Optimal value

NSE

0.80

0.85

0.93

0.81

0.72

travel time formulation that accounts for the dynamics of the

Factors that inﬂuence the wave velocity calculation

upstream contributions, compared to the previous methods. In that method, the dynamics of the upstream contributions and

The wave velocity of a grid is affected not only by the grid’s

slope will inﬂuence the wave velocity. Du et al. () pro-

slope but also by many other factors, including the basin

posed a time variant routing method, the spatially distributed

shape factor, rainfall intensity, vegetation cover, branching

direct hydrograph travel time method (SDDH), in which

ratio, basin area, climate, and form of drainage. The factors

time variant runoff, grid length, the roughness of the cell,

that inﬂuence the travel time are so numerous that they may

and slope are used to calculate the wave velocity.

compensate for each other to some extent. It is difﬁcult to

Some data such as hydraulic radius may not be available

calculate the wave velocity of each grid precisely. So, these

when applying the large-scale hydrological model. Thus, the

inﬂuencing factors were generalized to be reﬂected by the

routing method requiring too much input information may

parameter v45 , the parameter b, and slope in this study.

not preferable in a large-scale hydrological model. In the

The parameter b varies across basins due to differences in

improved NRF routing method, all the necessary infor-

basin characteristics, and it represents the underlying

mation can be taken from the DEM. Compared with the

surface.

routing methods that only take the slope into consideration,

The underlying surface inﬂuences the wave velocity and

the improved NRF can be adapted to more basins. Overall,

travel time in many ways. There is no strict and determinis-

the improved NRF routing method is simple and effective

tic functional relationship between basin characteristic and

for a large-scale hydrological model.

mean wave velocity in a large-scale hydrological model.

850

J. Li et al.

The improvement study of the NRF ﬂow routing method

Hydrology Research

51.5

2020

Multivariate regression analysis (Alexopoulos ; Zhang

There are other factors that are not included in Equation

et al. ) is a mathematical analysis method that uses the

(13), and for R 2 it is not equal to 1. However, we have yet to

least squares method to model the relationship between a

ﬁnd more accurate and detailed rules. The mean wave vel-

dependent variable and multiple independent variables. In

ocity that was predicted by Equation (13) was compared

this study, we used it to examine the relationship between

with the mean wave velocity that was calculated with the

the mean wave velocity of grids v (m/s) calculated with

improved routing method as shown in Figure 9, which

the improved routing method and several inﬂuencing fac-

shows an overall good result.

tors. The result equation will not be used to calculate the

In the future study, more data on the factors that inﬂu-

wave velocity directly. But this method can identify basin

ence wave velocity, such as soil, vegetation cover, and the

characteristics that have a greater impact on the mean

branching ratio, should be inputted into the method in

wave velocity. Some variables that are related to wave vel-

order to examine the relationship between routing par-

ocity are chosen (Table 1): mean discharge at the outlet

ameters and basin characteristics. If appropriate, the

station Q (m3/s); area A (km2); shape factor F; mean slope

improved routing model has the potential to be applied to

S ( ); river length Lr (m); and range of elevation ΔH (m).

data-sparse areas by a parameter transformation based on

The area, shape factor, and river length are basin geometric

the distribution in a subzone of climate, soil, and vegetation,

factors which affect the path and sequence of runoff and

or even to calculate the travel time without parameters,

then inﬂuence the mean wave velocity. The slope and

which would enable this model to be used to simulate the

range of elevation inﬂuence the wave velocity, because

hydrological effect of climate and land-use change.

potential energy will be converted into kinetic energy. In the process of runoff routes, runoff will continually merge and diverge, and the speed between water will affect each

CONCLUSIONS

other. Thus, the mean discharge at the outlet station is In this paper, a ﬂow routing algorithm was developed by

also taken as an independent variable. Based on data of the 15 studied catchments, multiple

combining an aggregation NRF with a velocity function.

linear regression analysis was done using the above factors as independent variables and mean wave velocity as the dependent variable. The value of R 2 was found to be equal to 0.73 and the significance level of the F-test was 0.051, which indicates that the mean wave velocity is associated (at the 90% level) with the above factors. The final regression equation is shown as Equation (13). The standardized regression coefficients of those variables are 0.23 (Q), 1.37 (A), 0.44 (F), 1.42 (S), 0.55 (Lr ), and 0.79 (ΔH). It can be inferred that the slope has the greatest positive effect on wave velocity and the area has the most negative effect.

v ¼ 0:0001884Q 0:00001691A þ 0:8231F þ 0:6517S 0:0003146ΔH þ 0:0004925Lr

(13)

where v (m/s) is the mean wave velocity of the grids; Q (m3/s) is the mean discharge at the outlet station; A (km2) is the area; F is the shape factor; S ( ) is the mean slope; Lr (m) is the river’s length; and ΔH (m) is the range of elevation.

Figure 9

A comparison between the predicted and calculated mean wave velocity of grids.

851

J. Li et al.

The improvement study of the NRF ﬂow routing method

Hydrology Research

The improved routing NRF method uses a parameter b that

51.5

2020

REFERENCES

can reflect how sensitive the wave velocity is to a slope. The routing methods before and after the improvement were applied to 15 catchments in the Yangtze River using a daily WASMOD-M model based on Hydro1 k. The main conclusions can be summarized as follows: (1) The original NRF routing method was found to provide unsatisfactory runoff simulation performance in most of the studied catchments, with an unreasonable calibrated travel time and wave velocity of the grids, for it cannot adapt to the different underlying conditions in the different catchments. (2) After the improvement in the NRF routing method, the relationship between wave velocity and slope was found to be more flexible. It can simulate the complexity of the wave velocity distribution more effectively in all types of basins, resulting in a reasonable runoff travel time. (3) The runoff travel time that was calculated by the model with the improved NRF routing method was found to be shorter and more reasonable, which leads to better runoff simulation performance. (4) The model with the improved NRF routing method was found to yield better results with respect to long-term daily runoff time series and peak flow amounts than the model with the original NRF routing method. The improved model was found to perform better in catchments with different characteristics. The improved model also retains the best advantage of NRF; i.e., it records all of the travel time information from the PRF.

ACKNOWLEDGEMENT This study was partly supported by the National Natural Science Fund of China (grant no. 51539009) and the National Key Research and Development Program (grant no. 2017YFA0603702). The authors are grateful to the National Meteorological Information Center for providing the meteorology datasets and the Bureau of Hydrology of the

Changjiang

Water

Resources

Commission

providing the datasets for the Yangtze River basin.

for

Abou Rafee, S. A., Uvo, C. B., Martins, J. A., Domingues, L. M., Rudke, A. P., Fujita, T. & Freitas, E. D.  Large-scale hydrological modelling of the Upper Paraná River Basin. Water 11 (5), 882. doi:10.3390/w11050882. Alexopoulos, E. C.  Introduction to multivariate regression analysis. Hippokratia 14 (Suppl. 1), 23. doi:10.1111/j.13652362.2009.02222.x. Arora, V. K. & Boer, G. J.  A variable velocity flow routing algorithm for GCMs. Journal of Geophysical Research: Atmospheres 104 (D24), 30965–30979. doi:10.1029/ 1999jd900905. Arora, V., Seglenieks, F., Kouwen, N. & Soulis, E.  Scaling aspects of river flow routing. Hydrological Processes 15 (3), 461–477. doi:10.1002/hyp.161. Barraquand, J. & Latombe, J. C.  A Monte-Carlo algorithm for path planning with many degrees of freedom. IEEE International Conference on Robotics and Automation. doi:10.1109/ROBOT.1990.126256. Beven, K.  How far can we go in distributed hydrological modelling? Hydrology and Earth System Sciences 5 (1), 1–12. doi:10.5194/hess-5-1-2001. Beven, K. J. & Kirkby, M. J.  A physically based, variable contributing area model of basin hydrology/Un modèle à base physique de zone d’appel variable de l’hydrologie du bassin versant. Hydrological Sciences Journal 24 (1), 43–69. doi:10.1080/02626667909491834. Bunster, T., Gironás, J. & Niemann, J. D.  On the influence of upstream flow contributions on the basin response function for hydrograph prediction. Water Resources Research 55 (6), 4915–4935. doi:10.1029/2018WR024510. Chen, L., Chang, J., Wang, Y. & Zhu, Y.  Assessing runoff sensitivities to precipitation and temperature changes under global climate-change scenarios. Hydrology Research 50 (1), 24–42. doi:10.2166/nh.2018.192. Du, J., Xie, H., Hu, Y., Xu, Y. & Xu, C.-Y.  Development and testing of a new storm runoff routing approach based on time variant spatially distributed travel time method. Journal of Hydrology 369 (1–2), 44–54. doi:10.1016/j.jhydrol.2009. 02.033. Ducharne, A., Golaz, C., Leblois, E., Laval, K., Polcher, J., Ledoux, E. & de Marsily, G.  Development of a high resolution runoff routing model, calibration and application to assess runoff from the LMD GCM. Journal of Hydrology 280 (1–4), 207–228. doi:10.1016/S0022-1694(03)00230-0. Fan, M., Mawuko, D. O., Shibata, H. & Ou, W.  Spatial conservation areas for water yield hydrological ecosystem services with their economic values effects under climate change: a case study of Teshio watershed located in northernmost of Japan. Hydrology Research 50 (6), 1679–1709. doi:10.2166/nh.2019.009. Fekete, B. M., Vörösmarty, C. J. & Lammers, R. B.  Scaling gridded river networks for macroscale hydrology:

852

J. Li et al.

The improvement study of the NRF ﬂow routing method

development, analysis, and control of error. Water Resources Research 37 (7), 1955–1967. doi:10.1029/2001wr900024. Gong, L., Widen-Nilsson, E., Halldin, S. & Xu, C.-Y.  Largescale runoff routing with an aggregated network-response function. Journal of Hydrology 368 (1–4), 237–250. doi:10. 1016/j.jhydrol.2009.02.007. Gong, L., Halldin, S. & Xu, C.-Y.  Global-scale river routing – an efficient time-delay algorithm based on HydroSHEDS high-resolution hydrography. Hydrological Processes 25 (7), 1114–1128. doi:10.1002/hyp.7795. Guo, J., Xu, L. & Leung, L. R.  A new multiscale flow network generation scheme for land surface models. Geophysical Research Letters 31 (23), 345–357. doi:10.1029/ 2004GL021381. Hansen, N. & Ros, R.  Benchmarking a weighted negative covariance matrix update on the BBOB-2010 noisy testbed. In: Conference Companion on Genetic and Evolutionary Computation. doi:10.1145/1830761.1830788. Huang, P. C. & Lee, K. T.  Efficient DEM-based overland flow routing using integrated recursive algorithms. Hydrological Processes 31 (5), 1007–1017. doi:10.1002/hyp.11080. Kharin, V., Flato, G., Zhang, X., Gillett, N., Zwiers, F. & Anderson, K. . Earth’s Future 6 (5), 704–715. doi:704715. 10.1002/2018EF000813. Kirkby, M. J. & Beven, K.  Channel Network Hydrology. John Wiley & Sons, New York. Kizza, M., Rodhe, A., Xu, C.-Y. & Ntale, H. K.  Modelling catchment inflows into Lake Victoria: uncertainties in rainfall–runoff modelling for the Nzoia River. Hydrological Sciences Journal 56 (7), 1210–1226. doi:10.1080/02626667. 2011.610323. Li, L., Xia, J., Xu, C.-Y. & Singh, V. P.  Evaluation of the subjective factors of the GLUE method and comparison with the formal Bayesian method in uncertainty assessment of hydrological models. Journal of Hydrology 390 (3), 210–221. doi:10.1016/j.jhydrol.2010.06.044. Li, Z., Xu, Z. & Li, Z.  Performance of WASMOD and SWAT on hydrological simulation in Yingluoxia watershed in northwest of China. Hydrological Processes 25 (13), 2001–2008. doi:10.1002/hyp.7944. Li, L., Xu, C.-Y. & Jain, S.  Comparison of the global TRMM and WFD precipitation datasets in driving q large-scale hydrological model in Southern Africa. Hydrology Research 44 (5), 770–788. doi:10.2166/nh.2012.175. Li, H., Beldring, S. & Xu, C.-Y.  Implementation and testing of routing algorithms in the distributed Hydrologiska Byråns Vattenbalansavdelning model for mountainous catchments. Hydrology Research 45 (3), 322–333. doi:10.2166/nh.2013.009. Li, L., Shen, M., Hou, Y., Xu, C.-Y., Lutz, A. F., Chen, J., Jain, S. K., Li, J. & Chen, H.  Twenty-first-century glaciohydrological changes in the Himalayan headwater Beas River basin. Hydrology Earth System Sciences 23 (3), 1483–1503. doi:10.5194/hess-23-1483-2019. Ling, Z., Nan, Z., Xu, L., Yi, X., Hernández, F. & Li, L.  Application of the MacCormack scheme to overland flow

Hydrology Research

51.5

2020

routing for high-spatial resolution distributed hydrological model. Journal of Hydrology 558, 421–431. doi:10.1016/ j.jhydrol.2018.01.048. Lu, W. & Qin, X. S.  Integrated framework for assessing climate change impact on extreme rainfall and the urban drainage system. Hydrology Research 51 (1), 77–89. doi:10. 2166/nh.2019.233. Lu, G., Liu, J., Wu, Z., Hai, H., Xu, H. & Lin, Q.  Development of a large-scale routing model with scale independent by considering the damping effect of sub-basins. Water Resources Management 29 (14), 5237–5253. doi:10.1007/ s11269-015-1115-7. Mckay, M. D., Beckkman, R. J. & Conover, W.  Comparison of three methods for selecting values of input variables in the analysis of output from a computer code. Technometrics 21 (2), 266–294. doi:10.2307/1271432. Miller, J. R., Russell, G. L. & Caliri, G.  Continental-scale river flow in climate models. Journal of Climate 7 (6), 914–928. doi:10.1175/1520-0442(1994)007<0914:CSRFIC> 2.0.CO;2. Morisawa, M.  Measurement of drainage-basin outline form. Journal of Geology 66 (5), 587–591. doi:10.1086/626538. Müller Schmied, H. & Döll, P.  Assessment of the terrestrial water balance using the global water availability and use model WaterGAP – status and challenges. In: EGU General Assembly Conference. pp. 2921. Naden, P., Broadhurst, P., Tauveron, N. & Walker, A.  River routing at the continental scale: use of globally-available data and an a priori method of parameter estimation. Hydrology and Earth System Sciences 3 (1), 109–123. Nash, J. E. & Sutcliffe, J. V.  River flow forecasting through conceptual models part I – a discussion of principles. Journal of Hydrology 10 (3), 282–290. doi:10.1016/0022-1694(70) 90255-6. Ngongondo, C., Gong, L., Xu, C.-Y. & Alemaw, B. F.  Flood frequency under changing climate in the upper Kafue River basin, southern Africa: a large scale hydrological model application. Stochastic Environmental Research and Risk Assessment 27 (8), 1883–1898. doi:10.1007/s00477-013-0724-z. Olivera, F., Famiglietti, J. & Asante, K.  Global-scale flow routing using a source-to-sink algorithm. Water Resources Research 36 (8), 2197–2207. doi:10.1029/2000WR900113. Ragettli, S., Tong, X., Zhang, G., Wang, H., Zhang, P. & Stähli, M.  Climate change impacts on summer flood frequencies in two mountainous catchments in China and Switzerland. Hydrology Research. In press. doi:10.2166/nh.2019.118. Sausen, R., Schubert, S. & Dümenil, L.  A model of river runoff for use in coupled atmosphere-ocean models. Journal of Hydrology 155 (3–4), 337–352. doi:10.1016/0022-1694(94) 90177-5. Shepard, D.  A two-dimensional interpolation function for irregularly-spaced data. In: Proceedings of the ACM National Conference. pp. 517–524. doi:10.1145/800186.810616. Sircar, J. K., Ragan, R. M., Engman, E. T. & Fink, R. A.  A GIS based geomorphic approach for the digital computation of time-area curves. In: Civil Engineering Applications of

853

J. Li et al.

The improvement study of the NRF ﬂow routing method

Remote Sensing in Water Resources Engineering (D. B. Stafford, ed.). ASCE, New York, pp. 28–296. Su, B. D., Jiang, T. & Jin, W. B.  Recent trends in observed temperature and precipitation extremes in the Yangtze River basin, China. Theoretical and Applied Climatology 83 (1–4), 139–151. doi:10.1007/s00704-005-0139-y. Tian, Z., Jing, Q., Dai, T., Jiang, D. & Cao, W.  Effects of genetic improvements on grain yield and agronomic traits of winter wheat in the Yangtze River Basin of China. Field Crops Research 124 (3), 417–425. doi:10.1016/j.fcr.2011.07.012. USGS  HYDRO1 k Elevation Derivative Database. Available from: https://www.usgs.gov/centers/eros/ science/usgs-eros-archive-digital-elevation-hydro1 k?qtscience_center_objects=0#qt-science_center_objects (accessed 23 November). doi:10.5066/F77P8WN0. Wang, F., Ducharne, A., Cheruy, F., Lo, M. H. & Grandpeix, J. Y.  Impact of a shallow groundwater table on the global water cycle in the IPSL land–atmosphere coupled model. Climate Dynamics 50 (9–10), 3502–3522. doi:10.1007/ s00382-017-3820-9. Wang, Y., Ni, J., Yue, Y., Li, J., Borthwick, A. G. L., Cai, X., Xue, A., Li, L. & Wang, G.  Solving the mystery of vanishing rivers in China. National Science Review 6 (6), 1239–1246. doi:10.1093/nsr/nwz022. Weisstein, E. W. ‘Lambert Azimuthal Equal-Area Projection.’ From MathWorld – A Wolfram Web Resource. Available from: http://mathworld.wolfram.com/ LambertAzimuthalEqual-AreaProjection.html (accessed 23 November). Wen, Z., Xu, L. & Yang, S.  A new multiscale routing framework and its evaluation for land surface modeling applications. Water Resources Research 48 (8), 8528. doi:10. 1029/2011WR011337. Widén-Nilsson, E., Halldin, S. & Xu, C.-Y.  Global waterbalance modelling with WASMOD-M: parameter estimation

Hydrology Research

51.5

2020

and regionalisation. Journal of Hydrology 340 (1–2), 105–118. doi:10.1016/j.jhydrol.2007.04.002. Xu, C.-Y.  WASMOD – the water and snow balance modeling system. In: Mathematical Models of Small Watershed Hydrology Applications (V. P. Singh & D. Frevert, eds). Water Resources Publication, Highlands Ranch, Colo, pp. 555–590. Xu, J., Yang, D., Yi, Y., Lei, Z., Chen, J. & Yang, W.  Spatial and temporal variation of runoff in the Yangtze River basin during the past 40 years. Quaternary International 186 (1), 32–42. doi:10.1016/j.quaint.2007.10.014. Yang, X., Magnusson, J., Rizzi, J. & Xu, C.-Y.  Runoff prediction in ungauged catchments in Norway: comparison of regionalization approaches. Hydrology Research 49 (2), 487–505. doi:10.2166/nh.2017.071. Yang, X., Magnusson, J. & Xu, C.-Y.  Transferability of regionalization methods under changing climate. Journal of Hydrology 568, 67–81. doi:10.1016/j.jhydrol.2018.10.030. Yang, X., Magnusson, J., Huang, S., Beldring, S. & Xu, C.-Y.  Dependence of regionalization methods on the complexity of hydrological models in multiple climatic regions. Journal of Hydrology 582, 124357. doi:10.1016/j.jhydrol.2019.124357. Yao, Z., Liu, Z., Huang, H., Liu, G. & Wu, S.  Statistical estimation of the impacts of glaciers and climate change on river runoff in the headwaters of the Yangtze River. Quaternary International 336, 89–97. doi:10.1016/j.quaint. 2013.04.026. Zhang, Y., Min, Y., Zheng, J., Wang, G. & Niu, D.  Compilation of Flood Forecasting Plan for the Yangtze River Basin. Hydrology Bureau of Yangtze River Water Resources Commission, Ministry of Water Resources, Wuhan (in Chinese). Zhang, G., Xie, H., Yao, T., Li, H. & Duan, S.  Quantitative water resources assessment of Qinghai Lake basin using Snowmelt Runoff Model (SRM). Journal of Hydrology 519, 976–987. doi:10.1016/j.jhydrol.2014.08.022.

First received 25 November 2019; accepted in revised form 15 March 2020. Available online 23 June 2020

854

51.5

2020

Identiﬁcation of regional water security issues in China, using a novel water security comprehensive evaluation model Jiping Yao, Guoqiang Wang

, Baolin Xue, Gang Xie and Yanbo Peng

ABSTRACT In order to solve regional water security issues, such as shortage of water resources, the aggravation of water pollution, the destruction of the ecological environment, etc., this study proposed the ﬂood control security index, resource security index and ecological security index, respectively, according to the construction principle of human development index. Based on the above security indexes, a novel water security comprehensive evaluation model is established by combining the coupling coordination degree model and the state space model. The proposed model has the advantage of simple operation and fast data speed, which is convenient for water security evaluation in different periods and regions. Taking China as an example, the water security conditions were evaluated from 2007 to 2016 for 31 provincial-level administrative regions in China, including ﬂood control security

Jiping Yao (corresponding author) Guoqiang Wang Baolin Xue Beijing Key Laboratory of Urban Hydrological Cycle and Sponge City Technology, College of Water Sciences, Beijing Normal University, Beijing 100875, China E-mail: wanggq@bnu.edu.cn Gang Xie Yanbo Peng Shandong Academy for Environmental Planning, Jinan 250000, China

index, resource security index, ecological security index and water security level of each region, and the speciﬁc problems of water security in each region were obtained. The evaluation results are consistent with the actual situation in each region, which provides the scientiﬁc basis for the local government authorities to formulate the corresponding regional water security policy. Key words

| ecological security index, ﬂood control security index, regional water security, resources security index, water security evaluation model

INTRODUCTION Water is essential for maintaining the balance of life and the

et al. , a; Yang et al. ; Yao et al. c). How-

living environment. It is a basic natural resource and stra-

ever, in recent years, frequent ﬂoods, water shortage,

tegic resource to promote human economic development

pollution and water ecological damage have become serious

and social progress (Masseroni et al. ; Wang et al.

global challenges (Parry et al. ; Emam et al. ; Awan

b; Yao et al. a, b). Water security refers to the

et al. ; Han et al. ; Liu et al. a; Ledingham et al.

capacity of water resources with quantity and quality

). In 1972, the United Nations Conference on Environ-

guarantee required for human survival and development,

ment and Development predicted that a water crisis would

which can maintain the basin sustainability and human

take place following the oil crisis (Biswas ). In 2000,

and ecological environment health, and ensure people’s

the Hague and the World Water Week took ‘water security

life and property from water disasters (ﬂoods, landslides

in the 21st century’ as the theme of the world Ministerial

and droughts) (Ren et al. ; Fang et al. a; Wang

Level Conference (Falkenmark ). In 2009, United Nations Educational, Scientiﬁc and Cultural Organization

This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying,

pointed out in its World Water Resources Development

adaptation and redistribution, provided the original work is properly cited

Report that the contradiction between supply and demand

(http://creativecommons.org/licenses/by/4.0/).

of water resources in human society is more prominent,

doi: 10.2166/nh.2020.014

855

J. Yao et al.

Identiﬁcation of regional water security issues in China

Hydrology Research

51.5

2020

coupled with climate degradation and rapid population

global sustainable development challenges. Actually, the

growth, which leads to more serious water security problems

water security issues are caused by the comprehensive

(Russo et al. ; Jiang ; Kumar ; Li et al. ;

inﬂuence of society, economy, resources, environment

Fang et al. b). Therefore, it is necessary to study water

and ecology (Awan et al. ). However, at present, the

security and propose solutions to related issues from differ-

water security evaluation is only based on the character-

ent perspectives.

istics of the research area or the focus of the research

Many scholars study water security and related issues

problem, and puts forward the methods to solve the

from different perspectives. Harris & Kennedy ()

water security problem from different angles. The theory

pointed out that urban water supply should be integrated

and methods of water security evaluation are not compre-

into the actual urban development planning from the

hensive enough, and the universality of the proposed

perspective of water supply, and further explored the evol-

evaluation method is relatively low.

ution of urban water security. Rijsberman & van de Ven

In the 1990 Human Development Report, the United

() revealed the development of urban water security

Nations proposed the Human Development Index (HDI),

from a new perspective of water resources carrying capacity.

which comprehensively reﬂects the level of human develop-

Sullivan () proposed a water poverty index similar to the

ment among different countries and regions by using three

Consumer Price Index to reﬂect the impact of water short-

variables: human life index, education level index and

age on human beings. Falkenmark & Lundqvist ()

GDP index (Kawada et al. ). The index plays an impor-

paid attention to the water quantity problem, comprehen-

tant role in guiding the development strategies of developing

sively considering the shortage of natural water resources

countries. Therefore, this paper puts forward a ﬂood control

and water quality shortage caused by water pollution, and

security index, resource security index and ecological secur-

provided a more objective and real basis for local relevant

ity index similar to HDI from ﬂood control, resources and

departments to formulate water security protection policies.

ecology, and establishes an evaluation model that can com-

Ou et al. () used the established entropy weight-fuzzy

prehensively reﬂect the regional water security situation.

matter element model to evaluate the rural water safety situ-

The water security of the 31 provinces of China is analyzed

ation, which effectively reduced the inﬂuence of uncertainty

and evaluated by the proposed water security evaluation

and fuzziness in the evaluation process on the authenticity

model, and the distributions of ﬂood control security

of water safety. Tian & Gang () used the pressure state

index, the resource security index and the ecological

response conceptual model to evaluate regional water

security index are respectively obtained. Additionally, the

security from the perspective of ecological security.

distributions of water security index which can comprehen-

Norman et al. () evaluated the water security situation

sively reﬂect the water security status of each region are

of a community in Canada with the method of water

obtained. All of the above indexes provide a scientiﬁc and

security index evaluation, and obtained good results. Con-

reliable basis for a comprehensive understanding of water

sidering the uncertainty of drought events, Dong & Xia

security problems in various regions of China and for the

() introduced the Dempster Shafer evidence theory

formulation of targeted strategies to solve water security

and evidence reasoning algorithm to assess the risk of

problems.

water security during drought. Gain et al. () put forward a framework that can quantitatively reﬂect the impact of human activities and natural water resources

MATERIALS AND METHODS

on water security by combining the factors that affect freshwater resources with the factors of social and econ-

Study area and data

omic development. Larson () took water security as the leading mode to study natural resource policy, and

According to the three key factors of water security, namely

comprehensively explored the situation of human water

ﬂood control security, resource security and ecological

security by combining climate change with other pressing

security, the current situation and changes of water security

856

J. Yao et al.

Identiﬁcation of regional water security issues in China

Hydrology Research

51.5

2020

in China from 2007 to 2016 are comprehensively evaluated.

index and ecological security index respectively. The three

In this paper, 31 administrative regions in mainland China

indexes are used to establish the water security evaluation

are taken as the research units of regional water security

model. The speciﬁc calculation methods are discussed below.

to study the temporal and spatial evolution of water security

The key points of ﬂood control security are casualties

in China (Figure 1), and the key factors affecting the water

caused by ﬂood, economic loss of water conservancy facili-

security status of China and each research unit are analyzed.

ties and disaster area loss caused by ﬂood in the region.

Due to the lack of relevant data, Taiwan, Hong Kong and

Therefore, the ﬂood control security index (FS) established

Macao were not included in the study. The dataset of this

includes three data sets: loss rate of disaster area Ra, regional

study came from the China Water Conservancy Bulletin,

ﬂood disaster population rate Rc and the ratio of economic

China Water Development Statistical Yearbook, China

loss of water-saving facilities to direct economic loss Re,

Statistical

which reﬂects the regional social development water. The

Yearbook,

China

Environmental

Statistical

Yearbook and the China Flood and Drought Disaster

speciﬁc calculation formula is as follows:

Bulletin during the period 2007–2016 (Liu et al. a).

Methods Determination of security indexes of the regional water security evaluation model

8 Ra > > < Rc > > : Re

¼ ¼ ¼

Rad Ras Rcf Rcp Rew Red

(1)

where Rad denotes ﬂood damage areas (km2), Ras denotes Based on the three key factors of ﬂood control safety,

ﬂood disaster areas (km2), Rcf denotes the number of

resource security and ecological security, which can compre-

ﬂood-affected population (10,000 people), Rcp denotes the

hensively reﬂect the regional water security, this paper

number of population affected by disasters (10,000

establishes the ﬂood control security index, resource security

people), Rew denotes economic loss of water-saving facilities

Figure 1

Overview of the study area.

857

J. Yao et al.

Identiﬁcation of regional water security issues in China

Hydrology Research

51.5

2020

in ﬂoods (100,000,000 yuan), Red denotes direct economic

as the lower limit of per capita water resources in this

losses in ﬂoods (100,000,000 yuan). Since ﬂoods are

study (Liu et al. b).

driven by disasters caused by natural and human events,

The ecological security index (ES) focuses on the

the ﬁve-year moving average value is calculated based on

environmental conditions related to water, which reﬂects

the above data to reduce the impact of the annual ﬂood con-

the security of water ecology through the ecological con-

trol security index. The regional ﬂood control security index

ditions of rivers, lakes and reservoirs. The index includes

(FS) is deﬁned as:

the ratio of the length of a river above Class III to total

8 1, Ri 0:01 > > P > > lg Ri > > < 0:2 0:4 × , P3 FS ¼ > lg Ri > > 0:3 0:3 × , > > 3 > : 0, Ri > 1

river length (RW) and the ratio between the number of non-eutrophication lakes to the total number of lakes 0:01 < Ri 0:1

(RE). The calculation formula for the ecological security (i ¼ a, c, e)

index (ES) is as follows:

0:01 < R1 1 (2)

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi RW 2 þ RE2 ES ¼ 2

(5)

The resource security index (RS) reﬂects the coordination between the per capita water resources, urban and rural development level and water supply capacity. The index shows the relative guarantee strength among regional

Construction of regional water security model

basic resources, economic development level and water

Based on the ﬂood control security index, resource security

supply capacity. The resource security index (RS) includes

index and ecological security index, the water security index

per capita water resources, urbanization rate and per

is proposed to reﬂect regional water security. Due to the

capita water demand, and its calculation formula is as follows:

and ecology (Wilkinson & Bathurst ), this paper introduces the coupling coordination model (Cheng et al. )

8 > < 1, Wr 2W 1=2 RS ¼ 1 2 2 2 > 1 × [(1 W) þ (1 U) þ (1 C) ] , Wr < 2W : 3 (3) where W denotes the normalized water resource factor (see Equation (4)), U denotes the urbanization rate (%), and C is the per capita water storage capacity. 8 1, Wr > 2W > > < W ¼ lg Wr =lg 2W , > Wrmin > : Wrmin 0, Wr Wrmin

complementary relationship among ﬂood control, resources

to characterize the interaction between flood control, resources and ecology, and the specific calculation formula is as follows: qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 8 1=2 > > DFS↔RS ¼ [FS × RS=(FS þ RS)2 ] × (a × FS þ b × RS) > > < qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1=2 DRS↔ES ¼ [RS × ES=(RS þ ES)2 ] × (b × RS þ c × ES) > > q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi > > 1=2 : DFS↔ES ¼ [FS × ES=(FS þ ES)2 ] × (a × FS þ b × ES) (6)

Wrmin < Wr 2W

(4)

where DFS↔RS , DRS↔ES and DFS↔ES respectively represent the coupling coordination degree of the interaction between the ﬂood control security and the resource security, the

where Wr denotes the per capita water resources volume, W 3

coupling coordination degree of the interaction between

is the global per capita water resources volume (6,123 m ),

the resource security and the ecological security, and the

and Wrmin is the lower limit of global per capita water

coupling coordination degree of the ﬂood control security

resources volume. The per capita water resources of Israel

and the ecological security interaction. a, b and c are the

(a country with a serious water shortage), 97 m3, is taken

undetermined coefﬁcients of the ﬂood control security

J. Yao et al.

858

Identiﬁcation of regional water security issues in China

Hydrology Research

index, resource security index and ecological security index 1 respectively, the value of which should be a ¼ b ¼ c ¼ 2 (Cheng et al. ). On the basis of the above-mentioned security interactions, the state space model (Li et al. ) is used to obtain a water security index (WSI) that can comprehen-

Table 2

51.5

2020

Average values of ﬂood control security index in the study area from 2007 to 2016

Region

Beijing

0.5150

0.5031

0.2132

0.4258

Tianjin

0.7343

0.1358

0.1840

0.4737

Hebei

0.5656

0.0428

0.1871

0.5344

Shanxi

0.6055

0.0531

0.0718

0.5636

resource security and ecological security interaction on

Inner Mongolia

0.6150

0.2719

0.0954

0.4797

water security. Based on the above steps, a water security

Liaoning

0.6164

0.1722

0.2212

0.4629

evaluation model is established. The speciﬁc calculation for-

Jilin

0.5247

0.0649

0.2144

0.5137

mula is as follows:

Heilongjiang

0.6584

0.0154

0.0943

0.6020

Shanghai

0.3654

0.0321

0.0611

0.6145

Jiangsu

0.3190

0.0010

0.0613

0.7709

Zhejiang

0.4632

0.0084

0.1338

0.6285

Anhui

0.5147

0.0137

0.2187

0.5813

sively consider the impact of ﬂood control security,

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi D2FS↔RS þ D2RS↔ES þ D2FS↔ES WSI ¼ 3

(7)

In order to scientiﬁcally and effectively show the advan-

Fujian

0.4454

0.0580

0.1926

0.5303

tages and disadvantages of water security in various regions,

Jiangxi

0.5247

0.0234

0.2297

0.5549

this paper classiﬁes water security into ﬁve levels: unsafe,

Shandong

0.5712

0.0092

0.0783

0.6385

relatively unsafe, general security, relative security and

Henan

0.2643

0.0064

0.2314

0.6406

security according to the water security classiﬁcation stan-

Hubei

0.3644

0.0454

0.1936

0.5494

dard (Ren et al. ) (see Table 1).

Hunan

0.5185

0.0356

0.2318

0.5369

RESULTS AND DISCUSSION

Guangdong

0.4352

0.0585

0.1526

0.5410

Guangxi

0.3543

0.0485

0.1351

0.5634

Hainan

0.5439

0.0399

0.0818

0.5750

Chongqing

0.4053

0.0854

0.1641

0.5246

Sichuan

0.5295

0.0612

0.1723

0.5253

Considering that the ﬂood control security index, resource

Guizhou

0.4664

0.1050

0.1298

0.5197

security index and ecological security index are affected by

Yunnan

0.5745

0.1935

0.1744

0.4712

interannual changes, only relying on the evaluation results

Tibet

0.4157

0.2733

0.1922

0.4661

of a single year to determine the current situation of water

Shaanxi

0.6509

0.1643

0.1837

0.4707

security in each region of the study area will lead to a large

Gansu

0.7150

0.1307

0.1260

0.4929

deviation between the evaluation results and the actual situ-

Qinghai

0.5609

0.6281

0.2317

0.4088

ation. Consequently, based on the average values of the water

Ningxia

0.4540

0.2043

0.3703

0.4464

security evaluation indexes data for 2007–2016, the proposed

Xinjiang

0.6254

0.3694

0.3534

0.4088

water security evaluation model is used to obtain the ﬂood control security index, the resource security index, the ecological security index and the water security index

water security status in 31 administrative regions of the

(Tables 2–4). The spatial distribution characteristics of

study area were analyzed (Figures 2–5).

Table 1

Characteristics of ﬂood control security in the study

Water security evaluation standards

area Water security

Relatively

Generally

Relatively

level

Unsafe unsafe

safe

Safe

Level threshold

[0,0.2) [0.2,0.4)

[0.4,0.6)

[0.6,0.8)

[0.8,1.0]

Table 2 shows the average value of ﬂood control security index (FS) of each administrative region in the study area

J. Yao et al.

859

Table 3

Identiﬁcation of regional water security issues in China

Hydrology Research

Average values of resources security index in the study area from 2007 to 2016

Table 4

51.5

2020

Average values of ecological security index in the study area from 2007 to 2016

Region

Beijing

0.0869

0.8584

0.1194

0.2630

Beijing

0.8155

1.0000

0.9124

Tianjin

0.0523

0.8030

0.0834

0.2303

Tianjin

0.0916

0.0000

0.0648

Hebei

0.1883

0.4642

0.1379

0.2496

Hebei

0.4353

0.4219

0.4286

Shanxi

0.2657

0.4687

0.0943

0.2602

Shanxi

0.2689

0.0000

0.1901

Inner Mongolia

0.6706

0.5648

0.2053

0.4434

Inner Mongolia

0.7445

0.0000

0.5264

Liaoning

0.2861

0.6396

0.4130

0.4273

Liaoning

0.4391

1.0000

0.7723

Jilin

0.5264

0.5404

0.5955

0.5531

Jilin

0.6747

0.4290

0.5654

Heilongjiang

0.6373

0.5673

0.3549

0.5050

Heilongjiang

0.6690

0.4026

0.5521

Shanghai

0.2041

0.8885

0.0022

0.2603

Shanghai

0.4786

0.0000

0.3384

Jiangsu

0.4157

0.6121

0.0219

0.3051

Jiangsu

0.3497

0.1389

0.2661

Zhejiang

0.6721

0.6215

0.4005

0.5490

Zhejiang

0.7189

0.9088

0.8194

Anhui

0.5612

0.4551

0.2649

0.4141

Anhui

0.6167

0.5489

0.5838

Fujian

0.7374

0.6237

0.2529

0.4938

Fujian

0.7980

0.7488

0.7738

Jiangxi

0.7847

0.5178

0.3356

0.5100

Jiangxi

0.9369

1.0000

0.9690

Shandong

0.1185

0.5785

0.1037

0.2345

Shandong

0.4693

0.4599

0.4646

Henan

0.2341

0.4623

0.2264

0.2990

Henan

0.5095

0.6689

0.5946

Hubei

0.5916

0.5692

0.9986

0.6572

Hubei

0.8050

0.7883

0.7967

Hunan

0.6946

0.5069

0.3699

0.5056

Hunan

0.9821

0.0000

0.6945

Guangdong

0.6066

0.6307

0.2099

0.4476

Guangdong

0.7891

0.7387

0.7643

Guangxi

0.8142

0.4814

0.6873

0.6343

Guangxi

0.9585

1.0000

0.9795

Hainan

0.6451

0.5645

0.4140

0.5313

Hainan

0.9370

0.6689

0.8141

Chongqing

0.5656

0.6057

0.1991

0.4268

Chongqing

0.9928

0.0000

0.7020

Sichuan

0.6843

0.4768

0.2387

0.4364

Sichuan

0.8846

0.6645

0.7823

Guizhou

0.7236

0.4272

0.4199

0.5030

Guizhou

0.8142

1.0000

0.9118

Yunnan

0.7638

0.4638

0.7863

0.6399

Yunnan

0.8693

0.6594

0.7716

Tibet

1.0000

0.2796

0.4480

1.0000

Tibet

0.9948

1.0000

0.9974

Shaanxi

0.4558

0.4824

0.1298

0.3364

Shaanxi

0.6797

1.0000

0.8550

Gansu

0.3851

0.3781

0.1976

0.3148

Gansu

0.7485

1.0000

0.8832

Qinghai

0.9571

0.4614

0.9819

0.6879

Qinghai

0.9749

1.0000

0.9875

Ningxia

0.0542

0.5006

0.2218

0.2363

Ningxia

0.3671

0.0090

0.2597

Xinjiang

0.7689

0.4353

0.4179

0.5131

Xinjiang

0.9969

1.0000

0.9984

from 2007 to 2016, which is obtained from ﬂood control

are more developed in society and economy, have more invest-

security evaluation factors (Ra, Rc, Re). Based on Table 2,

ment in infrastructure, and are more advanced in technology,

the spatial distribution of the average ﬂood control security

so they have better performance in this regard (Liu et al.

index in the study area from 2007 to 2016 was obtained

b). On the contrary, the ﬂood control security index of

(Figure 2). The ﬂood control security index for Heilongjiang,

other provincial administrative areas is lower than 0.6,

Shanghai, Jiangsu, Zhejiang, Henan and Shandong exceeds

which indicates that the ﬂood control security level in these

0.6, which indicates that the ﬂood control security level of

areas is lower. In particular, the ﬂood control security level

these areas is higher. This is mainly due to the fact that

of Xinjiang and Qinghai in the west of the study area is the

these areas are prone to more rainfall and ﬂood events,

lowest at 0.4088. This low level is due to the rare rainfall,

860

Figure 2

J. Yao et al.

Identiﬁcation of regional water security issues in China

Hydrology Research

51.5

2020

Spatial distribution of annual average ﬂood control security index in the study area from 2007 to 2016.

rare flood and insufficient flood control infrastructure in wes-

Characteristics of resources security in the study area

tern China (especially the northwest) (Zhang et al. ). In this case, people are more vulnerable to ﬂoods and suffer

Resource security reﬂects the coordination of regional

more losses. These areas require a higher level of ﬂood control

basic resources, economic development and water supply

alarms. Relevant government departments of the study area

capacity. Table 3 shows the average value of the resource

need to strengthen the uniﬁed management of water resources

security index (RS) for each administrative region in the

and ﬂood control, and speed up the preparation or revision of

study area from 2007 to 2016, which was obtained from

emergency operation plans for ﬂood control operation of

resource security evaluation factors (W, U, C ). Based on

important rivers. In addition, relevant departments need to

Table 3, the spatial distribution of the annual average

speed up the formulation of water distribution plans and

resource security index in the study area from 2007 to

water conservancy renovation plans for major rivers, improve

2016 was obtained (Figure 3). In all administrative regions

the uniﬁed dispatching management scheme, clarify the dis-

of the study area, the resource security index (RS) of

patching project and authority, and strengthen the ﬂood risk

Tibet is equal to 1, which indicates that the level of resource

management.

security in the region is very high. This is mainly due to the

861

Figure 3

J. Yao et al.

Identiﬁcation of regional water security issues in China

Hydrology Research

51.5

2020

Spatial distribution of annual average resource security index in the study area from 2007 to 2016.

relatively high per capita water resources in Tibet, which is

water resources are scarce (W < 0.12), which leads to the

twice as large as that in the rest of the world. In addition, the

lack of per capita water resources. Especially, the per capita

low urbanization rate, sparse population and relatively high

water resources of Tianjin are lower than that of the world

per capita water resources in this area make Tibet have a

minimum standard of water resources per capita (97 m3).

superior resource security effect (Liu et al. b). The resource

Additionally, the water supply capacity of Hebei, Shanxi,

security index (RS) of Qinghai, Hubei, Guangxi and Yunnan is

Shanghai, Jiangsu, Shandong, Shaanxi and Gansu is very

higher than 0.6, which indicates that the resource security

low (C < 0.14). This indicates that the above-mentioned

water in these areas is higher. The resource security indexes

administrative regions are facing great challenges in

of Beijing, Tianjin, Hebei, Shanxi, Shanghai, Jiangsu, Shan-

resource security, these regions need to reasonably adjust

dong, Henan, Shaanxi and Gansu are lower than 0.4, which

the industrial structure, optimize the industrial layout, and

indicates that the resource security level of these areas is rela-

improve the efﬁciency of water resource allocation accord-

tively low. This is mainly because the urbanization rate of

ing to the natural conditions and relevant development

Beijing and Tianjin is very high (U > 0.8), and the per capita

plans of water resources in each region.

862

Figure 4

J. Yao et al.

Identiﬁcation of regional water security issues in China

Hydrology Research

51.5

2020

Spatial distribution of annual average ecological security index in the study area from 2007 to 2016.

Characteristics of ecological security in the study area

Xinjiang all exceed 0.8, which indicates that the ecological security of these areas is at a high level. According to the

Ecological security fully reﬂects the water quality and eco-

data in the China Water Development Statistical Yearbook

logical characteristics of rivers, lakes and reservoirs.

and China Water Resources Bulletin, the results of regional

Table 4 shows the average value of the ecological security

ecological security assessment during the period of 2007–

index (ES) for each administrative region in the study area

2016 show that the above-mentioned areas have better

from 2007 to 2016, which was obtained from ecological

environmental governance, and the proportion of rivers

security evaluation factors (RW, RE). Based on Table 4,

above class III and non-eutrophication lakes and reservoirs

the spatial distribution of the annual average ecological

that meet the requirements of RW and RE is relatively large.

security index in the study area from 2007 to 2016 was

However, the ecological security index related to water qual-

obtained (Figure 4). The results show that the ecological

ity of rivers, lakes and reservoirs in Tianjin and Shanxi is

security indexes of Beijing, Zhejiang, Jiangxi, Guangxi,

lower than 0.2. This indicates that the ecological security

Hainan, Guizhou, Shaanxi, Tibet, Gansu, Qinghai and

of Tianjin and Shanxi is at a low level, which may be

863

Figure 5

J. Yao et al.

Identiﬁcation of regional water security issues in China

Hydrology Research

51.5

2020

Spatial distribution of annual average water security index in the study area from 2007 to 2016.

caused by the serious water pollution and eutrophication of

departments of environmental protection need to formulate

lakes and reservoirs in these two areas. The ecological secur-

effective measures to reduce the total amount of pollutants,

ity index of Shanghai, Jiangsu and Ningxia varies between

control the movement rate of pollutants, improve the efﬁ-

0.2 and 0.4 and this indicates that the ecological security

ciency of sewage treatment, realize the comprehensive

of these areas is at the general level. In addition, the ecologi-

treatment of major rivers, and improve the ecological secur-

cal security index of the remaining areas is more than 0.5,

ity early warning system.

which indicates that the ecological security of these areas is at a relatively high level. In general, since the release of

Characteristics of water security in the study area

the Water Pollution Control Action Plan by the State Council in April 2015, the water ecological security situation in

Table 5 shows the average value of the water security index

the study area has been improved, and the water ecological

(WSI) for each administrative region in the study area from

environment protection has been strengthened (Liu et al.

2007 to 2016, which was obtained from FS, RS and ES.

b). In order to ensure the sustainable and healthy devel-

Based on Table 5, the spatial distribution of the annual aver-

opment of ecological security in the study area, the relevant

age water security index in the study area from 2007 to 2016

J. Yao et al.

864

Table 5

Identiﬁcation of regional water security issues in China

Average values of water security index in the study area from 2007 to 2016

WSI

Security level

Hydrology Research

51.5

2020

the water security of these regions is in an unsafe state. This is mainly due to the low level of resource security and ecologi-

Region

Beijing

0.4258 0.2630 0.9124 0.5337 Generally safe

be caused by the serious water shortage and low control rate

Tianjin

0.4737 0.2303 0.0648 0.2563 Relatively unsafe

of ecological water functional areas in these areas.

Hebei

0.5344 0.2496 0.4286 0.4042 Generally safe

Shanxi

0.5636 0.2602 0.1901 0.3380 Relatively unsafe

Inner Mongolia 0.4797 0.4434 0.5264 0.4832 Generally safe

cal security in Tianjin, Hebei, Shanxi and Ningxia, which may

To sum up, the northern part of the study area (Beijing, Tianjin, Hebei, Shanxi and Inner Mongolia) suffers from serious water shortage, and the water quality of lakes and

Liaoning

0.4629 0.4273 0.7723 0.5541 Generally safe

Jilin

0.5137 0.5531 0.5654 0.5441 Generally safe

Heilongjiang

0.6020 0.5050 0.5521 0.5530 Generally safe

Shanghai

0.6145 0.2603 0.3384 0.4044 Generally safe

Jiangsu

0.7709 0.3051 0.2661 0.4474 Generally safe

Zhejiang

0.6285 0.5490 0.8194 0.6656 Relatively safe

Anhui

0.5813 0.4141 0.5838 0.5264 Generally safe

Fujian

0.5303 0.4938 0.7738 0.5993 Relatively safe

Jiangxi

0.5549 0.5100 0.9690 0.6780 Relatively safe

Shandong

0.6385 0.2345 0.4646 0.4459 Generally safe

Jiangsu, Zhejiang, Anhui, Fujian, Shanxi and Shandong)

Henan

0.6406 0.2990 0.5946 0.5114 Generally safe

Hubei

0.5494 0.6572 0.7967 0.6678 Relatively safe

has strong ﬂood control capacity and a high-water security

Hunan

0.5369 0.5056 0.6945 0.5790 Generally safe

security level of Shanghai and Jiangsu are relatively low.

Guangdong

0.5410 0.4476 0.7643 0.5843 Generally safe

The water security indexes of the central regions (Henan,

Guangxi

0.5634 0.6343 0.9795 0.7257 Relatively safe

Hubei and Hunan) in the study area are similar to that of

Hainan

0.5750 0.5313 0.8141 0.6402 Relatively safe

Chongqing

0.5246 0.4268 0.7020 0.5511 Generally safe

Sichuan

0.5253 0.4364 0.7823 0.5813 Generally safe

Guizhou

0.5197 0.5030 0.9118 0.6448 Relatively safe

Yunnan

0.4712 0.6399 0.7716 0.6276 Relatively safe

Tibet

0.4661 0.4622 0.9974 0.6419 Relatively safe

Shaanxi

0.4707 0.3364 0.8550 0.5540 Generally safe

Gansu

0.4929 0.3148 0.8832 0.5636 Generally safe

Qinghai

0.4088 0.6879 0.9875 0.6947 Relatively safe

Ningxia

0.4464 0.2363 0.2597 0.3141 Relatively unsafe

Xinjiang

0.4088 0.5131 0.9984 0.6401 Relatively safe

reservoirs is poor, which is reﬂected by the lower resource security index and ecological security index in these areas. The above problems are the main reasons for the unsafe water security in these areas. The northeast of the study area (Heilongjiang, Jilin and Liaoning) is relatively balanced in ﬂood control security, resource security and ecological security, which leads to the overall high-water security index in these areas, and the water security is in a general state of security. The east part of the study area (Shanghai,

index. However, the resource security level and ecological

the eastern regions. Henan is the most populous area in the study area, which leads to the most serious problem of resource shortage. The south (Guangdong, Guangxi, Hainan) and southwest (Chongqing, Sichuan, Guizhou, Yunnan, Tibet) of the study area have a high water security index, and the lakes, rivers and reservoirs in these areas have good water quality. The water security situation in the northwest of the study area (Shaanxi, Ningxia, Gansu, Qinghai, Xinjiang) is generally good. However, Ningxia is facing relatively unsafe water resources and environmental conditions, which need more attention. The above conclusions are consistent with the published articles (Liu

was obtained (Figure 5). The results show that the water

et al. b; Wang et al. c; Zhao et al. ), which indi-

security index of Guangxi is the highest among all administra-

cates that the water security evaluation model proposed in

tive regions in the study area (WSI ¼ 0.7257), which indicates

this paper has scientiﬁc and reliable application value.

that the water security in this area is in a safe state. The water security indexes of Zhejiang, Jiangxi, Hubei, Hainan, Guizhou, Yunnan, Tibet, Qinghai and Xinjiang vary from 0.6 to

CONCLUSIONS

0.8, which indicates that these areas are in a relatively safe state, while the water security indexes of Tianjin, Hebei,

Based on ﬂood control security, resource security and ecologi-

Shanxi and Ningxia are less than 0.4, which indicates that

cal security, which can comprehensively reﬂect the water

865

J. Yao et al.

Identiﬁcation of regional water security issues in China

security situation, this paper establishes the corresponding

Hydrology Research

51.5

2020

REFERENCES

security indexes, and uses the coupling coordination model and state space model to organically combine these security indexes, and constructs a water security comprehensive evaluation model. The proposed model is used to study the flood control security, resource security and ecological security of each administrative region in 2007–2016 in China. The water security level of each administrative region in the study area is characterized according to the above-mentioned security and the water security issues between each region are analyzed and compared by using the water security level. The results show that, in terms of flood control security, except for the relative security of Heilongjiang, Shandong, Henan, Jiangsu, Shanghai, Zhejiang, the flood control security capacity of other administrative regions needs to be further improved. In terms of resource security, Beijing, Tianjin, Hebei, Shanxi, Shaanxi, Gansu and Ningxia have a relatively low level of resource security due to their low resource base conditions, these areas need to improve the efficiency of resource utilization and establish a sound resource security system. In addition, the uncoordinated level of resources, population and economic development in Shanghai, Jiangsu, Shandong and Henan also leads to the relatively low level of resource security, these areas need to establish appropriate resource optimization allocation programs to improve the coordination between social and economic development and resource utilization. In terms of ecological security, the level of ecological security in Tianjin, Shanxi, Shanghai, Jiangsu and Ningxia is relatively low, so it is urgent to strengthen the restoration and protection of ecological water environment in these areas. Compared with the traditional water security evaluation system, which needs multi-layer design, a complex calculation method and different evaluation indexes, the water safety evaluation model proposed in this paper has clear theory, simple structure, easy interpretation, strong robustness, and good promotion and application.

ACKNOWLEDGEMENTS This study was supported by the National Natural Science Foundation of China (Grant No. 51679006 and Grant No. 51879006).

Awan, U. K., Liaqat, U. W., Choi, M. & Ismaeel, A.  A SWAT modeling approach to assess the impact of climate change on consumptive water use in Lower Chenab Canal area of Indus basin. Hydrol. Res. 47, 1025–1037. Biswas, A. K.  Water crisis – current perceptions and future realities. Water Int. 24, 363–367. Cheng, K., Yao, J. & Ren, Y.  Evaluation of the coordinated development of regional water resource systems based on a dynamic coupling coordination model. Water Sci. Technol. Water Supply 19, 565–573. Dong, Q. & Xia, L.  Risk assessment of water security in Haihe River Basin during drought periods based on D-S evidence theory. Water Sci. Eng. 7, 119–132. Emam, A. R., Kappas, M. & Hosseini, S. Z.  Assessing the impact of climate change on water resources, crop production and land degradation in a semi-arid river basin. Hydrol. Res. 46, 854–870. Falkenmark, M.  Water security for the 21st century – Building bridges through dialogue. In: Proceedings of the 11th Stockholm Water Symposium, Stockholm, Sweden, 13–16 August 2001. Water Sci. Technol. 45, VII–VIII. Falkenmark, M. & Lundqvist, J.  Towards water security: political determination and human adaptation crucial. Econ. Water Resour. 2, 489–503. Fang, Q., Wang, G., Liu, T., Xue, B.-L. & Yinglan, A. a Controls of carbon ﬂux in a semi-arid grassland ecosystem experiencing wetland loss: vegetation patterns and environmental variables. Agric. For. Meteorol. 259, 196–210. Fang, Q., Wang, G., Xue, B., Liu, T. & Kiem, A. b How and to what extent does precipitation on multi-temporal scales and soil moisture at different depths determine carbon ﬂux responses in a water-limited grassland ecosystem? Sci. Total Environ. 635, 1255–1266. Gain, A. K., Giupponi, C. & Wada, Y.  Measuring global water security towards sustainable development goals. Environ. Res. Lett. 11, 124015. Han, D., Wang, G., Liu, T., Xue, B.-L., Kuczera, G. & Xu, X.  Hydroclimatic response of evapotranspiration partitioning to prolonged droughts in semiarid grassland. J. Hydrol. 563, 766–777. Harris, J. M. & Kennedy, S.  Carrying capacity in agriculture: global and regional issues. Ecol. Econ. 29, 443–461. Jiang, Y.  China’s water security: current status, emerging challenges and future prospects. Environ. Sci. Policy 54, 106–125. Kawada, Y., Nakamura, Y. & Otani, S.  An axiomatic foundation of the multiplicative human development index. Rev. Income Wealth 65, 771–784. Kumar, P.  Hydrocomplexity: addressing water security and emergent environmental risks. Water Resour. Res. 51, 5827–5838. Larson, R. B.  Water security. Northwest. Univ. Law Rev. 112, 139–200.

866

J. Yao et al.

Identiﬁcation of regional water security issues in China

Ledingham, J., Archer, D., Lewis, E., Fowler, H. & Kilsby, C.  Contrasting seasonality of storm rainfall and flood runoff in the UK and some implications for rainfall-runoff methods of flood estimation. Hydrol. Res. 50, 1309–1323. Li, J., Guan, X., Wang, G. & Jiang, H.  Assessment on agricultural non-point source pollution in a drinking water source using remote sensing images. Chin. J. Ecol. 35, 3382–3392. Li, W., Yao, Y. & Shekhar, D. K.  State-space model for transient behavior of membrane-based liquid desiccant dehumidifier. Int. J. Heat Mass Transfer 144, 118711. Liu, F., Song, X., Yang, L., Han, D., Zhang, Y., Ma, Y. & Bu, H. a Predicting the impact of heavy groundwater pumping on groundwater and ecological environment in the Subei Lake basin, Ordos energy base, Northwestern China. Hydrol. Res. 49, 1156–1171. Liu, H., Jia, Y., Niu, C., Gan, Y. & Xu, F. b Evaluation of regional water security in China based on dualistic water cycle theory. Water Policy 20, 510–529. Masseroni, D., Ercolani, G., Chiaradia, E. A. & Gandolfi, C.  A procedure for designing natural water retention measures in new development areas under hydraulic-hydrologic invariance constraints. Hydrol. Res. 50, 1293–1308. Norman, E. S., Dunn, G., Bakker, K., Allen, D. M. & de Albuquerque, R. C.  Water security assessment: integrating governance and freshwater indicators. Water Resour. Manage. 27, 535–551. Ou, B., Fu, S., Wang, Y. & Wang, L.  The comprehensive evaluation of rural drinking water security in Yunnan Province. In: 2012 International Conference on Structural Computation And Geotechnical Mechanics (P. Yang & X. Jiang, eds). People’s Republic of China, Kunming, pp. 155–158. Parry, S., Hannaford, J., Lloyd-Hughes, B. & Prudhomme, C.  Multi-year droughts in Europe: analysis of development and causes. Hydrol. Res. 43, 689–706. Ren, L., Li, Q. & Yuan, F.  The hydrological cycle and water security in a changing environment in China. Hydrol. Res. 43, 1–2. Ren, Y., Yao, J., Xu, D. & Wang, J.  A comprehensive evaluation of regional water safety systems based on a similarity cloud model. Water Sci. Technol. 76, 594–604. Rijsberman, M. A. & van de Ven, F. H. M.  Different approaches to assessment of design and management of sustainable urban water systems. Environ. Impact Assess. Rev. 20, 333–345. Russo, T., Alfredo, K. & Fisher, J.  Sustainable water management in urban, agricultural, and natural systems. Water 6, 3934–3956. Sullivan, C.  Calculating a water poverty index. World Dev. 30, 1195–1210.

Hydrology Research

51.5

2020

Tian, J. & Gang, G.  Research on regional ecological security assessment. In: 2012 International Conference on Future Energy, Environment And Materials (G. Yang, ed.). Pt B. Hong Kong, People’s Republic of China, pp. 1180–1186. Wang, G., Hu, X., Zhu, Y., Jiang, H. & Wang, H.  Historical accumulation and ecological risk assessment of heavy metals in sediments of a drinking water lake. Environ. Sci. Pollut. Res. 25, 24,882–24,894. Wang, G., Li, J., Sun, W., Xue, B., Yinglan, A. & Liu, T. a Nonpoint source pollution risks in a drinking water protection zone based on remote sensing data embedded within a nutrient budget model. Water Res. 157, 238–246. Wang, P., Yao, J., Wang, G., Hao, F., Shrestha, S., Xue, B., Xie, G. & Peng, Y. b Exploring the application of artificial intelligence technology for identification of water pollution characteristics and tracing the source of water quality pollutants. Sci. Total Environ. 693, 133440. Wang, Q., Liu, Y.-y., Zhang, Y.-z., Tong, L.-j., Li, X., Li, J.-l. & Sun, Z. c Assessment of spatial agglomeration of agricultural drought disaster in China from 1978 to 2016. Sci. Rep. 9, 14393. Wilkinson, M. E. & Bathurst, J. C.  A multi-scale nested experiment for understanding flood wave generation across four orders of magnitude of catchment area. Hydrol. Res. 49, 597–615. Yang, H., Wang, G., Wang, L. & Zheng, B.  Impact of land use changes on water quality in headwaters of the Three Gorges Reservoir. Environ. Sci. Pollut. Res. 23, 11,448–11,460. Yao, J., Wang, G., Xue, B., Wang, P., Hao, F., Xie, G. & Peng, Y. a Assessment of lake eutrophication using a novel multidimensional similarity cloud model. J. Environ. Manage. 248, 109259. Yao, J., Wang, G., Xue, W., Yao, Z. & Xue, B. b Assessing the adaptability of water resources system in Shandong Province, China, using a novel comprehensive co-evolution model. Water Resour. Manage. 33, 657–675. Yao, J., Wang, P., Wang, G., Shrestha, S., Xue, B. & Sun, W. c Establishing a time series trend structure model to mine potential hydrological information from hydrometeorological time series data. Sci. Total Environ. 698, 134227. Zhang, Y., Wang, Y., Chen, Y., Liang, F. & Liu, H.  Assessment of future flash flood inundations in coastal regions under climate change scenarios – a case study of Hadahe River basin in northeastern China. Sci. Total Environ. 693, 133550. Zhao, L., Gong, D., Zhao, W., Lin, L., Yang, W., Guo, W., Tang, X. & Li, Q.  Spatial-temporal distribution characteristics and health risk assessment of heavy metals in surface water of the Three Gorges Reservoir, China. Sci. Total Environ. 704, 134883.

First received 19 January 2020; accepted in revised form 25 March 2020. Available online 11 May 2020

867

51.5

2020

Copula-based drought severity-area-frequency curve and its uncertainty, a case study of Heihe River basin, China Zhanling Li, Quanxi Shao, Qingyun Tian and Louie Zhang

ABSTRACT Copulas are appropriate tools in drought frequency analysis. However, uncertainties originating from copulas in such frequency analysis have not received signiﬁcant consideration. This study aims to develop a drought severity-areal extent-frequency (SAF) curve with copula theory and to evaluate the uncertainties in the curve. Three uncertainty sources are considered: different copula functions, copula parameter estimations, and copula input data. A case study in Heihe River basin in China is used as an example to illustrate the proposed approach. Results show that: (1) the dependence structure of drought severity and areal extent can be modeled well by Gumbel; Clayton and Frank depart the most from Gumbel in estimating drought SAF curves; (2) both copula parameter estimation and copula input data contribute to the uncertainties of SAF curves; uncertainty ranges associated with copula input

Zhanling Li (corresponding author) Qingyun Tian School of Water Resources and Environment, MOE Key Laboratory of Groundwater Circulation and Environmental Evolution, China University of Geosciences, Beijing 100083, China E-mail: zhanling.li@cugb.edu.cn Quanxi Shao Louie Zhang CSIRO Data 61, Private Bag 5, Wembley, WA 6913, Australia

data present wider than those associated with parameter estimations; (3) with the conditional probability decreasing, the differences in the curves derived from different copulas are increasing, and uncertainty ranges of the curves caused by copula parameter estimation and copula input data are also increasing. These results highlight the importance of uncertainty analysis of copula application, given that most studies in hydrology and climatology use copulas for extreme analysis. Key words

| copula, drought, input data, parameter estimation, SPEI, uncertainty

INTRODUCTION Drought is a stochastic natural phenomenon that arises

Previous literature has been most concerned with the

from considerable deﬁciency in precipitation. With the

severity and the duration of drought due to their ability to

global temperature increasing, it becomes more common

characterize drought from the temporal scale (e.g., Lee &

and has great effects on terrestrial ecosystems (Mishra &

Kim ; Beguería et al., ; Rahmat et al. ; Tan

Singh ; Fang et al. a; Han et al. ). The frequency

et al. ; Afshar et al. ; Homdee et al. ; Fang

analysis of drought is important since it is capable of quan-

et al. b; Hameed et al. ). However, the spatial

tifying the drought characteristics (Rahmat et al. ).

extent of drought is also very important in drought assess-

Bivariate or even multivariate frequency analysis is more

ments. Recent researches have shown that the drought

informative than univariate analysis due to the complexity

area has been signiﬁcantly increasing globally and regionally

and multivariate features of drought (Yu et al. ; Jiang

during the last several decades. The percentage of areas

et al. ; Rahmat et al. ; Amirataee et al. ).

under drought had increased by about 1.74% per decade from 1950 to 2008 globally (Dai ). The drought-affected

This is an Open Access article distributed under the terms of the Creative

area increased nearly 12-fold from 1950 to 2009 in China

Commons Attribution Licence (CC BY-NC-ND 4.0), which permits copying

(Wang et al. ). A signiﬁcant increase in drought area

and redistribution for non-commercial purposes with no derivatives, provided the original work is properly cited (http://creativecommons.org/ licenses/by-nc-nd/4.0/). doi: 10.2166/nh.2020.173

has been detected in the northwest of China, with an increasing rate of 3.96% per decade since the late 1990s

868

Z. Li et al.

Copula-based drought SAF curve and its uncertainty

Hydrology Research

51.5

2020

(Yu et al. ). Therefore, taking both severity and areal

functions based on Bayesian framework; whereas the

extent into account is very necessary for better understand-

ﬁxed drop-down menu options in the toolbox and the non-

ing drought from both a temporal and spatial scale, and it

editable display for the outputs restrict its wide application

would be valuable for drought-based decision-making.

to some extent.

Applying the concept of quantifying the frequency

This paper aims to develop a drought SAF curve with

analyses of both severity and areal extent based on the

copula theory and to evaluate the uncertainties in the

copula theory, the drought severity-area-frequency (SAF)

curve

curves can be derived. These curves are very useful in

parameter estimations, and copula input data. The following

drought planning and management since they can provide

issues will be addressed, namely: (i) to compare and select

the probabilities and return periods of the drought areal

the copula function with best performance in modeling

extent with different severities over a region (Reddy &

the dependence structure of drought severity and drought

Ganguli ; Amirataee et al. ). However, uncertainties

areal extent; (ii) to construct a drought SAF curve based

are usually involved in the estimation of drought SAF curves

on the conditional probability distribution in the study

due to the effects of different copula functions, marginal

area; and (iii) to quantify the uncertainties in drought SAF

distributions, input data, parameter estimations, etc.

curves from three uncertainty sources.

caused

different

copula functions,

copula

The former uncertainty sources, copula functions and marginal distributions, are of major concern in previous literatures. For example, Halwatura et al. () compared four marginal distribution functions and two types of

METHODS

copulas, and assessed the impacts of these uncertainties on the estimations of recurrence intervals (RIs) of drought

Drought index and its characteristics

events. They stated that no differences were found between the RIs derived from the different marginal distributions

A number of indices have been developed so far for drought

for mild drought events, and the effects of different copulas

quantiﬁcation, monitoring, and analysis, such as PDSI

on RIs can nearly be negligible in their study. However,

(Palmer Drought Severity Index), SPI (Standardized

Zhao et al. () drew a slightly different conclusion by

Precipitation Index), and SPEI (Standardized Precipitation-

comparing six marginal distributions and three Archime-

Evapotranspiration Index), etc. Under the current climate

dean copulas in a case study of the Weihe River basin.

warming conditions, many studies show that, SPEI, devel-

They pointed out that the return period of drought varies

oped by Vicente-Serrano et al. (), could give a better

depending on the selected marginal distributions and

performance in drought assessments due to the fact that it

copula functions, and the differences become larger with

combines the multi-scalar character with the capacity of

the increase of return periods. Zhang et al. () evaluated

involvement of temperature effects on droughts (e.g., Tan

the inﬂuences of marginal distributions to the uncertainty of

et al. ; Homdee et al. ). Thus, SPEI was selected

joint distribution by employing a Bayesian framework, and

and 12-monthly scale of SPEI will be analyzed in this

the results showed that the stronger the tail of marginal

study. The temperature-based Thornthwaite method is

distribution, the greater the uncertainty of joint distribution,

used to calculate potential evapotranspiration (PET) as it

especially for extreme events. All the above studies made

only requires monthly mean temperature data and the latitu-

great contributions in enriching the uncertainty analysis

dinal coordinate of the location, which are readily available

in drought frequency curves. However, the uncertainties

at most meteorological stations. SPEI is computed with the

caused by copula parameter estimations and copula input

assumption that P-PET (precipitation minus potential eva-

data are rarely regarded. Sadegh et al. () presented a

potranspiration) series follow the Log-logistic distributions.

multivariate copula analysis toolbox which has the ability

Detailed information about SPEI and its calculation can

to quantify the uncertainties in drought frequency curves

be found in the studies of Vicente-Serrano et al. (),

associated with copula parameter estimations and copula

Beguería et al. (), and Hameed et al. ().

Z. Li et al.

869

Copula-based drought SAF curve and its uncertainty

Hydrology Research

Speciﬁcally, drought characteristics used here include drought severity (S) and drought areal extent (A). The sever-

(IDW)

spatial

interpolation

technology

51.5

2020

generally

employed to obtain the spatial distribution of SPEI values.

ity of drought events is deﬁned as the average of SPEI values of all cells in the study area, which is located below the drought threshold (Amirataee et al. ). It is computed using the equation:

Suppose X and Y are two continuous random variables with marginal distribution functions CDFs u1 and u2 , then there

PN St ¼

Copula-based conditional distribution functions

i¼1

I[SPEIi,t SPEIthr ] × SPEIi,t for t ¼ 1, . . . T NCt

exists a copula C such that: (1)

where I[:] is the logical indicator function with the value 1 if its argument is true, and the value 0 otherwise. SPEIi,t is the

FXY (x, y) ¼ P(X x, Y y) ¼ C(u1 , u2 )

(3)

The conditional density functions are related to C(u1 , u2 ; θ) through:

value of SPEI in cell i and period t, SPEIthr is the threshold of SPEI. NCt represents the total number of drought cells in period t, N means the total number of cells in the study

C(u2 ju1 ; θ) ¼ P(U2 u2 jU1 ¼ u1 ) ¼

@ 2 C(u1 , u2 ; θ) @u1 @u2

(4)

area, T means the data record length. Positive SPEI values indicate wet periods, while negative values indicate dry

and when it is integrated with respect to u2 ,

periods compared with the normal conditions of the area (Vicente-Serrano et al. ). Light drought ( 0.5 to 0.99), moderate drought ( 1.0 to 1.49), severe drought ( 1.5 to 1.99), and extreme drought ( 2.0) are deﬁned

uð2

C(u2 ju1 ; θ) ¼

tive and falls below a certain threshold value. A threshold

C(vju1 ; θ)dv ¼ 0

according to SPEI values (Vicente-Serrano et al. ). Drought events occur when the SPEI is continuously nega-

uð2

@ @u1

0 uð2

@ 2 C(u1 , v; θ) dv @u1 @v

@C(u1 , v; θ) @C(u1 , u2 ; θ) dv ¼ @v @u1

(5)

value of 0.5 is selected here. In order to represent the extreme condition and to analyze the associated risk of droughts using the exceedance probability, the negative drought severity is transformed to positive values for fitting to the available distributions (Mishra & Singh ; Amirataee et al. ), thus, there is a negative sign in Equation (1). The areal extent of drought is defined as a specified percent of total area in period t in which the value of SPEI is below the specified threshold (Reddy & Ganguli ; Amirataee et al. ). It is computed using the following equation: PN At ¼

i¼1

I[SPEIi,t SPEIthr ] × ai for t ¼ 1, . . . T PN i¼1 ai

where (U1 , U2 ) ∼ C, and C is a bivariate copula distribution function with parameter θ, representing the bivariate dependence structure of variables X and Y. C(u2 ju1 ; θ) is the numeric vector of the conditional distribution function of the copula with parameter θ evaluated at u2 given u1 . Different copulas differ in describing the dependence structures. Comparisons of copulas could be necessary before bivariate drought frequency analysis. Ten copulas are selected and given in Table 1, including not only oneparameter Archimedean families such as Clayton, Gumbel, Frank, and Joe, but also two-parameter copulas, BB1

(2)

(Clayton–Gumbel), BB6 (Joe–Gumbel), BB7 (Joe–Clayton), and BB8 (Joe–Frank). The more ﬂexible structures of twoparameter copulas allow for different non-zero lower

where ai is the area of cell i. Since SPEI is calculated orig-

and upper tail dependence coefﬁcients (Brechmann &

inally at a limited number of sites where observations on

Schepsmeier ). Many of the above copulas are com-

climate variables are available, inverse distance weighting

monly used in hydrological and climatological applications

Z. Li et al.

870

Table 1

Copula-based drought SAF curve and its uncertainty

Hydrology Research

51.5

2020

The cumulative density functions and parameter ranges of alternative copulas

No.

Copula family

C(u1 , u2 )

Normal

Φθ (Φ 1 (u1 ), Φ 1 (u2 ))

1 θ 1

Student-t

tθ, δ (t 1 δ (u1 ),

θ>0

Clayton

Gumbel

Parameter space

t 1 δ (u2 ))

θ>0

θ θ u θ 1 þ u2 1 1 exp [( lnu1 )θ þ ( lnu2 )θ ]θ 1

θ 1

Frank

1 (e θu1 1)(e θu2 1) ln 1 þ θ e θ 1

Joe

1 [(1 u1 )θ þ (1 u2 )θ (1 u1 )θ (1 u2 )θ ]θ

θ 1

BB1

1δ 1 θ θ θ δ 1 þ [(u δ 1 1) þ (u2 1) ]

δ > 0, θ 1

BB6

1θ 1 δ δ δ 1 1 exp {[ ln(1 (1 u1 )θ )] þ [ ln(1 (1 u2 )θ )] }

δ > 0, θ 1

BB7

1 1δ θ δ δ 1 1 [(1 (1 u1 )θ ) þ (1 (1 u2 )θ ) 1]

δ > 0, θ 1

BB8

θ≠0 1

0 " #1θ 1 1@ (1 (1 δu1 )θ )(1 (1 δu2 )θ ) A 1 1 δ 1 (1 δ)θ

δ > 0, θ 1

Note: For Normal, Φθ is the bivariate joint normal distribution with linear correlation coefﬁcient θ, and Φ is standard normal marginal distribution. Formulas in this table are from Chen & Khashanah (2015).

(e.g., Xu et al. ; Zhang et al. , ; Zhao et al. ;

The AIC and BIC are deﬁned as:

Ayantobo et al. ). Copula parameter estimation and goodness of ﬁt

AIC ¼ 2D 2‘

(6)

BIC ¼ Dlnn 2‘

(7)

The copula parameter θ (or θ and δ) in Table 1 helps to measure the extent of relationship between variables

where D is the number of parameters in the model, ‘ is the

(Ayantobo et al. ). Its estimation from sample data is

log-likelihood value of the best parameter set, and n rep-

not straightforward. One of the commonly used approaches

resents the sample size. The derivations of both equations

is maximum likelihood estimation (MLE) (Brechmann &

appear to be similar. A lower AIC or BIC value associates

Schepsmeier ). Although the solution to MLE involves

with a better copula model. The AIC takes into account

the roots of a cubic equation, and cannot be written in a

both the complexity of the model and minimization of

simple form, the MLE can be easily found using numerical

model error residuals and provides a more robust measure

routines. The standard errors for MLE copula parameter

of quality of model predictions. It avoids the problem of

estimations are based on inversion of the Hessian matrix

over-parameterization by adding a penalty term based on

(Brechmann & Schepsmeier ).

the number of parameters (Sadegh et al. ; Sun et al.

The best copula is selected according to the Akaike and

). The BIC equation differs from the AIC equation

Bayesian information criteria (AIC and BIC, respectively).

with respect to its ﬁrst term that depends on the sample

871

Z. Li et al.

Copula-based drought SAF curve and its uncertainty

Hydrology Research

51.5

2020

size. The penalty term of BIC is larger than that of AIC,

2. Using the probability distribution type which has been

when the number of samples is too large, it can effectively

pre-speciﬁed according to the original severity and areal

prevent the model complexity from being too high due to

extent series, to ﬁt the Sj* and Aj* for obtaining their opti-

the excessive model accuracy (Brechmann & Schepsmeier

mal parameter estimates and the corresponding marginal

).

distributions, CDF(Sj*) and CDF(Aj*). 3. Using CDF(Sj*), CDF(Aj*) and the best candidate copula pre-selected according to the original marginal distrij butions, to estimate the copula parameters θ^ (i ¼ 1, or

Uncertainty analysis

i ¼ 1, 2).

According to MLE, the optimal value of copula parameter

4. Calculating drought severities at given drought areal

can be derived, together with its standard errors which

extents according to the conditional probability func-

can be used to represent the parameter uncertainty ranges

tions, and then obtaining the drought curve SAFj*.

(without considering the uncertainty in marginal distributions in severity and areal extent). The effects of copula parameter estimations to drought SAF curves are quantified by random generating copula parameters based on the asymptotic normality first, then calculating the drought curves according to the new generating copula parameter values, and finally identifying the α=2th and (1 α)=2th

5. The above procedure is repeated m times to generate m bootstrap samples, m estimations of marginal distributions for severity and areal extent, m sets of copula j parameters θ^ , and m drought curves SAFj* ( j ¼ 1, 2, i

…, m). 6. From drought curves SAFj* ( j ¼ 1, …, m), identifying the α=2th and (1 α)=2th percentiles of drought curves

percentiles of drought curves which are deﬁned as the

which is deﬁned as the lower and upper limits of conﬁ-

lower and upper limits of conﬁdence intervals with given

dence intervals with given probabilities, and then

probabilities. These conﬁdence intervals are the uncertainty

analyzing the contributions of copula input data to the

ranges of drought curves caused by copula parameter

uncertainty of drought curves.

estimations. To quantify the effects of copula input data uncertainty

The main steps of this study are as follows. The values of

to drought SAF curves, the non-parametric bootstrap

12-month SPEI at the stations are calculated ﬁrst. Then, the

method, which has usually been used in quantifying the

drought severity and drought areal extent are determined

sample uncertainty in previous studies (e.g., Hu et al. ;

according to IDW spatial interpolation technique. The two

Vergni et al. ), is selected. The idea behind this

drought variables are ﬁtted by a variety of theoretical

method is that since the distribution from which the

probability distributions. After that, copulas are used to

sample has been taken is unknown, the values in that

compare the performances of capturing the dependence

sample are the best guide to the true distribution. The theor-

structures of the two drought variables with goodness of

etical details of the bootstrap approach can be found in

ﬁt tests. Drought SAF curves are then extracted with the

Efron (). The copula input data in this case consists of

best copula. Finally, the uncertainties in SAF curves

n sets of drought severity (S) and drought areal extent (A),

caused by different copulas, copula parameter estimations,

denoted as (S1, A1), (S2, A2), …, (Si, Ai), …, (Sn, An), in

and copula input data, are investigated.

which n means the data length. The Si and Ai series are ﬁtted by theoretical probability distributions and the best distribution is selected based on goodness of ﬁt tests. The

STUDY AREA AND DATA

speciﬁcs of bootstrap procedure are as follows: 1. In the jth bootstrap, resampling the original n set values

The Heihe River basin, which falls between 97.1 –102.0 E

of drought variables with replacement, and generating a

and 37.7 –42.7 N, is the second largest typical inland river

bootstrapped sample of the same size n as ((S1 , (S2j*, A2j*), ……, (Sij*, Aij*), ……, (Snj*, Anj*).

A1j*),

basin in the northwest of China with a controlled catchment area of 143,000 km2 (Figure 1). It is one of the most drought-

Z. Li et al.

872

Figure 1

Copula-based drought SAF curve and its uncertainty

Hydrology Research

51.5

2020

Locations of the meteorological stations in Heihe River basin.

prone river basins in China (Wang et al. ; Ayantobo

The locations and the detailed information of stations are

et al. ; Niu et al. ) with an annual precipitation of

shown in Figure 1 and Table 2.

approximately 150 mm across the basin, an annual average air temperature of 6–8 C, and annual potential evaporation of 1,000–2,000 mm. The runoff of the basin mainly comes

RESULTS AND ANALYSIS

from the upper reach, and consumed by large areas of irrigation in the middle reach and oasis ecosystem in the lower

Drought characteristics and their dependence

reach. Severe drought events would lead to inadequate

structures

water supply and thus affect the local agriculture and vulnerable oasis ecosystem. In this study, monthly precipitation

Drought severity and areal extent series over the study area

and temperature data spanning from 1960 to 2015 were col-

determined by Equations (1) and (2) are plotted in Figure 2.

lected from nine meteorological stations within the basin.

It is clear that, before 1997, the drought was mainly

Table 2

Climatic and geographical characteristics of stations used in the study during 1960–2015

No.

Station

Longitude

Latitude

Elevation (mm)

Annual precipitation (mm)

Annual mean temperature ( C)

Tuole

98.25

38.48

3,367

298

2.5

Yeniugou

99.36

38.26

3,320

418

2.8

Qilian

100.15

38.11

2,787

409

1.1

Shandan

101.05

38.48

1,765

201

6.6

Zhangye

100.17

39.05

1,483

128

7.5

Jiuquan

98.29

39.46

1,477

7.6

Gaotai

99.50

39.22

1,332

108

7.9

Dingxin

99.31

40.18

1,177

8.5

Ejinaqi

101.04

41.57

941

9.0

873

Z. Li et al.

Copula-based drought SAF curve and its uncertainty

Hydrology Research

51.5

2020

previous studies (e.g., Loukas & Vasiliades ; Serinaldi et al. ; Reddy & Ganguli ; Afshar et al. ; Amirataee et al. ), ten univariate probability distributions

(Exponential,

Normal,

Lognormal,

Logistic,

Loglogistic, Gamma, Weibull, Log-Pearson III, Burr, and GEV distribution) are tested for ﬁtting drought severity (S). Since the drought areal extent has deﬁnite ranges from 0 to 1, the bounded distributions, Beta, Power function, Pert, and Uniform, are tested as its candidate margins. Figure 2

Amirataee et al. () stated that Beta distribution Drought characteristics of Heihe River basin during the period of 1960–2015.

performed satisfactorily in ﬁtting the drought area extent in their study. Kolmogorov–Smirnov (K-S), Anderson–

characterized by light drought conditions. However, several

Darling (A-D), and Chi-squared (C-S) are used to test the

moderate or even severe droughts were captured after 1997,

goodness of ﬁt.

and furthermore, drought events with stronger severity and larger areal extent occurred more frequently than before.

Results are shown in Table 3. The critical values for the three tests are 0.061, 2.502, and 15.507 at 0.05 signiﬁcance

It is necessary to evaluate the dependency between vari-

level, respectively. As shown, GEV is selected ﬁnally as

ables prior to the application of copula when analyzing the

the marginal distribution for drought severity, because it

bivariate joint probability distribution of such random vari-

not only passes the K-S, A-D, and C-S tests at 0.05 signiﬁ-

ables (Amirataee et al. ). For this purpose, Pearson’s,

cance levels (all the statistic values are lower than the

Spearman’s, and Kendall’s correlation coefﬁcients are

critical values for the three tests), but shows the best

applied to explore the dependency between drought severity

performance with the smallest statistic values. This

and drought areal extent. These correlation coefﬁcients are

distribution was also proved to be suitable in ﬁtting the

often used in measuring the monotone association. The

drought severity in many other studies (e.g., Lee & Kim

ﬁrst belongs to the parametric method and sensitive to out-

; Reddy & Ganguli ). As to the variable of drought

liers, and the latter two belong to the non-parametric

areal extent, none of the distributions passes the tests.

methods and usually suggested for non-normally distributed

Then, we perform the ﬁtting process based on its pseudo

data (Shong ).

observations (Ghoudi & Rémillard ). In light of the

The correlation coefﬁcients between the two drought

references of Ghoudi & Rémillard () and Kojadinovic

variables are 0.678, 0.722, and 0.549, respectively, and all

et al. (), it is acceptable to replace the unknown mar-

the corresponding p-values are less than the signiﬁcance

ginal by their empirical counterparts obtained from the

level of 0.05. It is concluded that there is a signiﬁcant corre-

pseudo observations.

lation between the variables of drought severity and drought areal extent at 95% conﬁdence level.

Goodness of ﬁt of copula functions

Marginal distributions and copula goodness of ﬁt

The results of modeling the joint probability distribution of the two drought variables by using ten alternative copula functions are summarized in Table 4, including the statistic

Determination of marginal distribution for drought variables

values of goodness of ﬁt tests (AIC, BIC, and log-likelihood), the optimal parameter estimations, and their standard errors.

The first step in copula fitting consists of the identification of

Clearly, nearly all the copulas have close AIC, BIC, and log-

appropriate marginal distributions for the two drought vari-

likelihood values except for Frank which shows the greatest

ables. Since they are typically unknown, we ﬁrst ﬁt them by

departure from the others, indicating that most of the selected

using

copulas

several

theoretical

distributions.

According

have

similar

performance

capturing

the

Z. Li et al.

874

Table 3

Copula-based drought SAF curve and its uncertainty

Hydrology Research

51.5

2020

Goodness of ﬁt of theoretical probability distributions for drought severity (S) and drought areal extent (A)

Drought variable

No.

Probability distribution

Kolmogorov–Smirnov (K-S)

Anderson–Darling (A-D)

Chi-squared (C-S)

Drought severity (S)

1 2 3 4 5 6 7 8 9 10

Exponential Normal Lognormal Logistic Loglogistic Gamma Weibull Log-Pearson III Burr GEV

0.4475 0.0680 0.0434 0.0776 0.0580 0.0452 0.0841 0.0431 0.0481 0.0387

125.74 3.87 1.55 5.28 2.93 1.44 7.45 1.60 2.10 1.30

810.61 10.14 10.50 24.32 24.79 9.48 23.22 12.14 17.74 8.79

Drought areal extent (A)

1 2 3 4

Beta Power function Pert Uniform

0.1806 0.1266 0.1424 0.1767

35.46 38.91 37.16 16.43

101.81 112.88 188.62 48.12

Note: The minimum statistic values for K-S, A-D, and C-S tests are in bold italics.

Table 4

Goodness of ﬁt of different copulas and their optimal parameter estimations

No.

Copula

Akaike information criteria (AIC)

Bayesian information criteria (BIC)

Log likelihood

Optimal parameter (standard error)

Normal

465.60

461.38

233.80

0.79 (0.006)

Student-t

465.43

460.00

234.72

0.79, 7.98 (0.019, 6.082)

Clayton

442.62

438.41

222.31

2.01 (0.415)

Gumbel

473.45

469.24

237.73

2.27 (0.114)

Frank

364.49

360.27

183.24

6.24 (1.637)

Joe

433.61

429.40

217.81

2.77 (0.190)

BB1

490.71

482.28

247.36

0.92, 1.60 (0.138, 0.096)

BB6

471.45

463.01

237.72

1.00, 2.27 (0.230, 0.350)

BB7

514.65

506.21

259.33

2.09, 1.82 (0.127, 0.151)

BB8

435.22

426.79

219.61

2.89, 0.99 (0.145, 0.004)

Note: The minimum AIC, BIC and log likelihood values are in bold italics.

dependence structures of drought severity and drought areal

of 5,000 samples are generated, and the scatter plot of gener-

extent in this study case. Comparatively, Gumbel, among the

ated samples, together with the observed datasets are shown

one-parameter copula functions, provides the best ﬁt with the

in Figure 3. Different copula functions perform a little differ-

smallest AIC ( 473.45) and BIC ( 469.24) values. Among the

ently. The Normal, Student-t, and Frank are symmetric and

two-parameter copulas, BB7 performs the best. Student-t and

can be used to model symmetric correlation patterns of vari-

BB8 copulas even perform inferior to Gumbel according to

ables. Gumbel and Clayton are asymmetric and can model

the corresponding AIC and BIC values.

the asymmetrical tail correlation patterns. Gumbel exhibits

To better understand the performance of each copula in modeling the drought variables in this study, the generated

greater correlation in the upper tail, whereas Clayton exhibits greater correlation in the lower tail (Aas et al. ).

samples derived from the estimated theoretical copula by

From the scatterplot, the observed data in the two oppo-

Monte Carlo technique are compared with the observed

site corners appear clustering, seemingly displaying both

data for checking whether they have similar features. A set

upper- and lower-tail dependence, while the dependence

875

Figure 3

Z. Li et al.

Copula-based drought SAF curve and its uncertainty

Hydrology Research

51.5

2020

Generated and observed CDF for drought severity and drought areal extent based on different copula functions. Gray dots present the Monte Carlo simulations from each copula and blue dots present the observed data. Please refer to the online version of this paper to see this ﬁgure in color: http://dx.doi.10.2166/nh.2020.173.

structures in the tails appear slightly asymmetric and a little

Drought SAF curves

skewed. Therefore, asymmetric copulas may be a good choice. Combined with the results of goodness of ﬁt pre-

The drought SAF curve, developed by using conditional

sented in Table 3, Gumbel is then selected as the

probability function based on Gumbel copula, is plotted in

benchmark copula in this study. The two-parameter copulas,

Figure 4. The horizontal axis represents the percentage of

BB1, BB6, and BB7, are not selected herein since, on one

drought areal extent and the vertical axis represents the

hand, they show similar performance with Gumbel and,

drought severity. Only droughts with conditional probability

on the other hand, the higher model complexity (with

of less than 0.2 (corresponds to return period of more than

more parameters) would make a stimulation over condition-

ﬁve years) are considered in the analysis. It can be seen

ing of the model to some extent (Sadegh et al. ).

that the drought event with larger areal extent is generally

Actually, the optimality of copula functions could be dataset- and case-specific (Afshar et al. ). For example, Reddy & Ganguli () evaluated three copula families in modeling the joint dependence between the drought severity and drought areal extent in western Rajasthan of India, and found that the Gumbel–Hougaard copula best represents the drought properties in their study. Xu et al. () constructed a trivariate joint distribution based on the copula theory and concluded that the Joe and Gumbel copulas are more suitable to estimate the joint distribution of drought duration, affected area, and severity in southwest China. Amirataee et al. () regarded the Frank copula as the most appropriate in modeling the joint probability of drought severity and percent of area in the Lake Urmia basin, Iran. In our case, with considering the results of goodness of fit and the greater dependence in upper tails of the data, the one-parameter Gumbel is finally selected as the benchmark copula.

Figure 4

Drought severity-area-frequency (SAF) curves derived from Gumbel copula over the study area with six conditional probabilities (0.005, 0.01, 0.02, 0.05, 0.1, and 0.2).

876

Z. Li et al.

Copula-based drought SAF curve and its uncertainty

Hydrology Research

51.5

2020

associated with higher severity. For example, with a con-

Clayton are 442.62 and 438.41 (Table 3), very close to

ditional probability of 0.1 (at the return period of ten

those for Gumbel ( 473.45 and 469.24). However, as

years), when the drought areal extent amounts to 20%, the

shown in Figure 3, the drought SAF curves derived from

average drought severity is 0.98, belonging to light drought;

these two copulas have a considerably larger difference,

while when the drought areal extent increases to 60% and

and Clayton gives larger possibilities of drought severity

80%, the severity increases to 1.17 and 1.30, resulting in

than Gumbel.

moderate droughts.

BB6 gives quite close drought severity estimations with

The drought SAF curve based on the conditional prob-

Gumbel no matter how much the conditional probability is.

ability can allow a water planner to assess the probability of

BB1, BB7, and Student-t copulas give similar curve shapes

a drought with a certain areal extent and severity occurring

with Gumbel, whereas they overestimate the drought severity

simultaneously. For example, for a drought areal extent reach-

overall. A similar conclusion can be drawn from Normal

ing 40%, the occurrence probability is 0.05 when the severity

when the drought areal extent is higher than 10%, overesti-

is 1.12, and 0.02 when the severity increases to 1.21. This

mating the drought severities comparatively. Joe and BB8

information cannot be obtained from the univariate fre-

overestimate the drought severity when the drought area is

quency analysis of droughts, in which only the probability

lower than about 10%, but underestimate the drought sever-

of univariate occurrence can be derived. The copula-based

ity when the drought areal extent becomes higher.

drought SAF curve can provide decision-makers with prob-

In summary, the drought SAF curves are inﬂuenced by

ability maps of drought severity and useful information on

copula choices, and the curves have different conditional

drought areal extent in the study area, yielding important

probabilities of occurrence depending on which copula is

information for water management purposes.

used. BB6 gives quite close estimations in drought severity to Gumbel. Normal, Student-t, BB1, BB7 overestimate the

Drought SAF curve uncertainties

drought severities, while Joe and BB8 underestimate the drought severities as a whole. Clayton and Frank copulas

Uncertainty caused by different copula functions

depart the most, and overestimate the drought severity much more. As the conditional probability decreases, the

In this subsection, whether different copulas lead to signiﬁ-

differences in drought severity estimations from different

cantly different assessments of SAF curves is investigated.

copulas increase, although most of the differences are not

Figure 5 shows the conditional probability curves derived

large. It implies that the choice of copula should be

from different copula functions. Conditional probabilities of

implemented with caution since different copulas show

0.005, 0.01, 0.02, 0.05, 0.1, and 0.2 are compared. Gumbel

different performances in tail behaviors, especially when

copula is selected as the benchmark as stated above.

the focus is on extreme events (with small probabilities).

With a conditional probability of 0.2, most of the curves are close to each other, and only slight differences are found

Uncertainty caused by copula parameter estimations

between the benchmark and the alternatives, except for the two curves derived from Clayton and Frank. As the con-

After estimating the copula parameter(s) with the MLE

ditional probability decreases (conditional return period

method, 1,000 copula parameters are randomly generated

increases), these differences increase in general, being con-

from the Normal distribution with the mean and standard

sistent with the ﬁndings of Zhao et al. (), who pointed

error derived from MLE. The estimated optimal parameter

out that with an increase of the return period, the drought

value and standard error for Gumbel copula is 2.27 and

differences from different copula functions become larger.

0.114 (see Table 3). Such range of standard error in par-

Such differences are not very large in this study as a

ameter estimation is then translated to drought SAF

whole, except for those from Clayton and Frank.

curves, rendering the analysis of uncertainty. Combining

As to Clayton and Frank, large differences are detected.

with the conditional probability function and the generated

Interestingly, it is also noted that the AIC and BIC values for

copula parameters, a set of conditional probability isolines

Z. Li et al.

877

Figure 5

Copula-based drought SAF curve and its uncertainty

Hydrology Research

51.5

2020

Drought severity-area-frequency (SAF) curves derived from different copula functions over the study area with six conditional probabilities (0.005, 0.01, 0.02, 0.05, 0.1, and 0.2).

are then derived, which will be used to analyze the SAF

SAF curves become wider with the conditional prob-

curve uncertainties caused by copula parameter estimations.

ability decreasing. This is because the lower the

Figure 6 presents the 95% CIs of conditional prob-

probability of the drought event, the less the number of

ability isolines associated with the uncertain parameter

the observed data, and the less information available

estimations of Gumbel. To avoid the curve overlapping,

for the analysis.

the six conditional probabilities are separated into two

Kendall’s tau method is also used to estimate the par-

plots. Obviously, the uncertainty in copula parameter

ameter in Gumbel, yielding an optimal parameter value of

estimation has a certain effect on the conditional

2.22 with a standard error of 0.130, which does not deviate

probability isolines. The uncertainty ranges of drought

much from the results of the MLE method. The corresponding

878

Figure 6

Z. Li et al.

Copula-based drought SAF curve and its uncertainty

Hydrology Research

51.5

2020

Uncertainty ranges of drought severity-area-frequency (SAF) curve derived from copula parameter estimations over the study area with six conditional probabilities (0.005, 0.01, 0.02, 0.05, 0.1, and 0.2).

uncertainty ranges of conditional probability isolines are also quite close to the results from the MLE method. Uncertainty caused by copula input data Using bootstrap technique to generate 1,000 set of drought samples ((S1j*, A1j*), (S2j*, A2j*), …, (Sij*, Aij*), …, (S501j*, A501j*) (j ¼ 1, , 1,000)). To help visual comparisons of dependence structures among the original and the generated copula input data, the marginal distributions of the original samples and bootstrapped samples of drought severity and drought areal extent are plotted in Figure 7. The density of points in any area represents the density of the copula. It is observed that the bootstrapped samples well preserve

Figure 7

Dependence structures between drought severity and drought areal extent from 1,000 bootstrapped samples and the original samples.

the dependence structure between the drought severity and drought areal extent. Uncertainty bands in Figure 8 reveal the effects of copula input data uncertainty to the estimation of conditional drought severity. Obviously, with the conditional probability decreasing, the uncertainty bands are increasing, similar ﬁndings as those from the effects of copula parameter uncertainty. Table 5 presents the lower (2.5%

by copula input data present more widely. This reﬂects that the uncertainty of input data brings more uncertainty to drought SAF curves than that of copula parameter estimations. In other words, to improve the quality of copula input data can reduce the uncertainty of drought SAF curves more effectively.

percentile) and upper (97.5% percentile) limits and the ranges of estimated drought severity at a certain drought areal extent owing to both input data uncertainty and par-

CONCLUSIONS

ameter estimation uncertainty. The ranges resulting from both uncertainty sources grow remarkably as the con-

In this study, 12-month SPEI is calculated over the Heihe

ditional probability decreases. Compared with those

River basin in northwest China, the drought SAF curve is

caused by copula parameter, the uncertainty bands caused

developed, and then the uncertainties in the curve

Z. Li et al.

879

Figure 8

Table 5

Copula-based drought SAF curve and its uncertainty

Hydrology Research

51.5

2020

Uncertainty ranges of drought severity-area-frequency (SAF) curve derived from copula input data over the study area with six conditional probabilities (0.005, 0.01, 0.02, 0.05, 0.1, and 0.2).

Uncertainty bands of the estimated drought severity at given drought areal extent over the study area

Conditional probability Areal extent

0.005

0.01

0.02

0.05

0.1

0.2

20%-A

0.158 [1.088, 1.245] 0.133 [1.021, 1.155] 0.112 [0.964, 1.077] 0.088 [0.883, 0.971] 0.073 [0.820, 0.893] 0.059 [0.755, 0.814]

20%-B

0.108 [1.096, 1.204] 0.098 [1.030, 1.128] 0.083 [0.970, 1.053] 0.066 [0.888, 0.954] 0.050 [0.828, 0.879] 0.037 [0.763, 0.800]

40%-A

0.161 [1.117, 1.278] 0.140 [1.048, 1.188] 0.117 [0.991, 1.108] 0.088 [0.910, 0.999] 0.072 [0.848, 0.919] 0.058 [0.781, 0.839]

40%-B

0.114 [1.125, 1.240] 0.104 [1.060, 1.163] 0.087 [0.996, 1.083] 0.062 [0.917, 0.978] 0.049 [0.857, 0.906] 0.034 [0.791, 0.825]

60%-A

0.162 [1.134, 1.296] 0.141 [1.065, 1.206] 0.120 [1.006, 1.126] 0.090 [0.926, 1.016] 0.073 [0.863, 0.936] 0.058 [0.796, 0.854]

60%-B

0.109 [1.143, 1.252] 0.097 [1.079, 1.176] 0.081 [1.017, 1.098] 0.064 [0.933, 0.997] 0.043 [0.873, 0.915] 0.031 [0.807, 0.837]

80%-A

0.166 [1.148, 1.314] 0.146 [1.079, 1.224] 0.123 [1.019, 1.143] 0.093 [0.939, 1.032] 0.074 [0.877, 0.950] 0.059 [0.810, 0.868]

80%-B

0.119 [1.158, 1.277] 0.092 [1.093, 1.185] 0.082 [1.032, 1.114] 0.062 [0.949, 1.011] 0.042 [0.888, 0.930] 0.030 [0.822, 0.852]

Note: A and B mean the estimated drought severity uncertainties caused by copula input data (A) and copula parameter estimations (B); numbers in the brackets indicate the lower (2.5% percentile) and upper (97.5% percentile) limits, and those outside mean the ranges of 95% CIs.

associated with different copula functions, copula parameter

even presents quite close AIC and BIC values with

estimations, and copula input data are investigated.

Gumbel. The differences in drought SAF curves caused by

Results show that GEV is ﬁnally selected as the marginal

different copulas are increasing with the conditional prob-

distribution for drought severity, and pseudo observations

ability decreasing. As to the parameter estimations and

for drought areal extent. Gumbel shows the best perform-

input data, both of them result in the drought SAF curves

ance in capturing the dependence structure of drought

uncertainty. The uncertainty ranges caused by copula

severity and drought areal extent and thus is selected as

input data are larger than those caused by copula parameter

the benchmark copula. The estimations of drought SAF

estimations, and consequently, to improve copula input data

curves are inﬂuenced by copula choices. BB6 is capable of

quality is recommended for developing the reliability of such

giving close drought severity as Gumbel; Normal, Student-t,

frequency analysis. Results also show that the uncertainty

BB1, BB7 overestimate the drought severities; Joe and

bands in the drought SAF curves increase as the conditional

BB8 underestimate the drought severities when the drought

probability decreases. It indicates that larger uncertainties

areal extents become higher; Clayton and Frank produce the

exist in the frequency analysis of extreme values (with

most deviation in drought SAF curves although Clayton

small probability). This highlights the importance of

880

Z. Li et al.

Copula-based drought SAF curve and its uncertainty

uncertainty analysis in frequency study, given that most studies in hydrology and climatology focus on extreme values. Couple-based frequency analysis of drought with considering the uncertainty issues is capable of providing more beneﬁcial probability information to water managers, and aiding in the development of mitigation strategies and disaster preparedness plans for different levels of drought.

ACKNOWLEDGEMENTS The ﬁrst author is grateful for ﬁnancial support from the China Scholarship Council and China University of Geosciences (Beijing), as well as the support provided by CSIRO during her visit in Australia. This study is supported by the Fundamental Research Funds for the Central Universities (No. 35832015028) and NSFC (No. 41101038). The authors also thank the anonymous referees for their comments and suggestions that have led to the quality of the paper being improved.

REFERENCES Aas, K., Czado, C., Frigessi, A. & Bakken, H.  Pair-copula constructions of multiple dependence. Insurance: Mathematics and Economics 44 (2), 182–198. Afshar, M. H., Sorman, A. U. & Yilmaz, M. T.  Conditional copula-based spatial-temporal drought characteristics analysis – a case study over Turkey. Water 8 (10), 426. Amirataee, B., Montaseri, M. & Rezaie, H.  Regional analysis and derivation of copula-based drought severity-areafrequency curve in Lake Urmia basin, Iran. Journal of Environmental Management 206, 134–144. Ayantobo, O. O., Li, Y., Song, S., Javed, T. & Yao, N.  Probabilitic modelling of drought events in China via 2dimensional joint copula. Journal of Hydrology 559, 373–391. Beguería, S., Vicente-Serrano, S. M., Reig, F. & Latorre, B.  Standardized precipitation evapotranspiration index (SPEI) revisited: parameter ﬁtting, evapotranspiration models, tools, datasets and drought monitoring. International Journal of Climatology 34 (10), 3001–3023. Brechmann, E. C. & Schepsmeier, U.  Modeling dependence with C- and D-Vine copulas: the R package CDVine. Journal of Statistical Software 52 (3), 1–27. Chen, K. H. & Khashanah, K.  Analysis of systemic risk: a dynamic Vine copula-based ARMA-EGARCH model. In: Transactions on Engineering Technologies, World Congress

Hydrology Research

51.5

2020

on Engineering and Computer Science (S.-I. Ao, H. K. Kim & M. A. Amouzegar, eds). Springer Nature, Singapore. Dai, A.  Characteristics and trends in various forms of the palmer drought severity index during 1900–2008. Journal of Geophysical Research Atmospheres 16. https://doi.org/10. 1029/2010JD015541. Efron, B.  Bootstrap methods: another look at the jackknife. Annals of Statistics 7, 1–26. Fang, Q. Q., Wang, G. Q., Xue, B. L., Liu, T. X. & Kiem, A. a How and to what extent does precipitation on multitemporal scales and soil moisture at different depths determine carbon flux responses in a water- limited grassland ecosystem? Science of the Total Environment 635, 1255–1266. Fang, Q. Q., Wang, G. Q., Liu, T. X., Xue, B. L. & Yinglan, A. b Controls of carbon flux in a semi-arid grassland ecosystem experiencing wetland loss: vegetation patterns and environmental variables. Agricultural and Forest Meteorology 259, 196–210. Ghoudi, K. & Rémillard, B.  Empirical processes based on pseudo-observations II: the multivariate case. Asymptotic Methods in Stochastics 44, 381–406. Halwatura, D., Lechner, A. M., McIntyre, N. & Arnold, S.  Uncertainties in estimating design droughts. 8th International Congress on Environmental Modelling and Software. International Environmental Modelling and Software Society (iEMSs), Toulouse, France. http://www. iemss.org/society/index.php/iemss-2016-proceedings. Hameed, M., Ahmadalipour, A. & Moradkhani, H.  Apprehensive drought characteristics over Iraq: results of a multidecadal spatiotemporal assessment. Geosciences 8, 58. doi:10.3390/geosciences8020058. Han, D. M., Wang, G. Q., Liu, T. X., Xue, B. L., Kuczera, G. & Xu, X. Y.  Hydroclimatic response of evapotranspiration partitioning to prolonged droughts in semiarid grassland. Journal of Hydrology 563, 766–777. Homdee, T., Pongput, K. & Kanae, S.  A comparative performance analysis of three standardized climatic drought indices in the Chi River basin, Thailand. Agriculture and Natural Resources 50 (3), 211–219. Hu, Y., Liang, Z., Liu, Y., Wang, J., Yao, L. & Ning, Y.  Uncertainty analysis of SPI calculation and drought assessment based on the application of Bootstrap. International Journal of Climatology 35 (8), 1847–1857. Jiang, C., Xiong, L., Xu, C. & Guo, S.  Bivariate frequency analysis of nonstationary low-flow series based on the time-varying copula. Hydrological Processes 29 (6), 1521–1534. Kojadinovic, I., Yan, J., Leeuw, J. D. & Zeileis, A.  Modeling multivariate distributions with continuous margins using the copula R package. Journal of Statistical Software 34 (9), 1346–1352. Lee, J. H. & Kim, C. J.  Derivation of drought severity-durationfrequency curves using drought frequency analysis. Journal of Korea Water Resources Association 44 (11), 889–902.

881

Z. Li et al.

Copula-based drought SAF curve and its uncertainty

Loukas, A. & Vasiliades, L.  Probabilistic analysis of drought spatiotemporal characteristics in Thessaly region, Greece. Natural Hazards & Earth System Sciences 4 (5/6), 719–731. Mishra, A. K. & Singh, V. P.  Analysis of drought severity-areafrequency curves using a general circulation model and scenario uncertainty. Journal of Geophysical Research Atmospheres 114(D6). https://doi.org/10.1029/2008JD010986. Niu, J., Kang, S., Zhang, X. & Fu, J.  Vulnerability analysis based on drought and vegetation dynamics. Ecological Indicators 105, 329–336. https://doi.org/10.1016/j.ecolind. 2017.10.048. Rahmat, S. N., Jayasuriya, N. & Bhuiyan, M.  Development of drought severity-duration-frequency curves in Victoria, Australia. Australian Journal of Water Resources 19 (1), 31–42. Reddy, M. J. & Ganguli, P.  Spatio-temporal analysis and derivation of copula-based intensity-area-frequency curves for droughts in western Rajasthan (India). Stochastic Environmental Research and Risk Assessment 27 (8), 1975–1989. Sadegh, M., Ragno, E. & Aghakouchak, A.  Multivariate copula analysis toolbox (MvCAT): describing dependence and underlying uncertainty using a Bayesian framework. Water Resources Research 53, 5166–5183. Serinaldi, F., Bonaccorso, B., Cancelliere, A. & Grimaldi, S.  Probabilistic characterization of drought properties through copulas. Physics & Chemistry of the Earth 34 (10–12), 596–605. Shong, N.  Pearson’s Versus Spearman’s and Kendall’s Correlation Coefﬁcients for Continuous Data. Master of Science Thesis, University of Pittsburg, Pittsburg, PA, USA. http://d-scholarship.pitt.edu/8056/1/Chokns_etd2010.pdf Sun, P., Zhang, Q., Yao, R., Singh, V. P. & Song, C.  Low ﬂow regimes of the Tarim River Basin, China: probabilistic behavior, causes and implications. Water 10 (4), 470. Tan, C., Yang, J. & Li, M.  Temporal-spatial variation of drought indicated by SPI and SPEI in Ningxia Hui Autonomous Region, China. Atmosphere 6 (10), 1399–1421.

Hydrology Research

51.5

2020

Vergni, L., Lena, B. D., Todisco, F. & Mannocchi, F.  Uncertainty in drought monitoring by the standardized precipitation index: the case study of the Abruzzo region (central Italy). Theoretical and Applied Climatology 128 (1), 13–26. Vicente-Serrano, S. M., Beguería, S. & López-Moreno, J. I.  A multiscalar drought index sensitive to global warming: the standardized precipitation evapotranspiration index. Journal of Climate 23 (7), 1696–1718. Wang, X. J., Zhang, J. Y., Shahid, S., Elmahdi, A., He, R. M., Bao, Z. X. & Ali, M.  Water resources management strategy for adaptation to droughts in China. Mitigation and Adaptation Strategies for Global Change 17 (8), 923–937. Wang, Y., Xu, Z., Zhang, B. & Li, Q.  Monitoring the meteorological drought in the middle reaches of Heihe River basin based on TRMM precipitation data. 2016 IEEE International Geoscience and Remote Sensing Symposium, Beijing, China, pp. 4313–4316. doi: 10.1109/IGARSS.2016. 7730124. Xu, K., Yang, D., Xu, X. & Lei, H.  Copula based drought frequency analysis considering the spatio-temporal variability in Southwest China. Journal of Hydrology 527, 630–640. Yu, M., Li, Q., Hayes, M. J., Svoboda, M. D. & Heim, R. R.  Are droughts becoming more frequent or severe in China based on the standardized precipitation evapotranspiration index: 1951–2010? International Journal of Climatology 34 (3), 545–558. Zhang, Q., Xiao, M. & Singh, V. P.  Uncertainty evaluation of copula analysis of hydrological droughts in the East River basin, China. Global and Planetary Change 129, 1–9. Zhang, D., Chen, P., Zhang, Q. & Li, X.  Copula-based probability of concurrent hydrological drought in the Poyang lake-catchment-river system (China) from 1960–2013. Journal of Hydrology 553, 773–784. Zhao, P., Lü, H., Fu, G., Zhu, Y., Su, J. & Wang, J.  Uncertainty of hydrological drought characteristics with copula functions and probability distributions: a case study of Weihe River, China. Water 9 (5), 334.

First received 29 November 2019; accepted in revised form 4 March 2020. Available online 6 May 2020

882

51.5

2020

Effects of antecedent soil water content on inﬁltration and erosion processes on loessial slopes under simulated rainfall Lan Ma, Junyou Li and Jingjing Liu

ABSTRACT Soil texture and antecedent soil water content (ASWC) are primary factors governing hillslope hydrological and erosion processes. We used simulated rainfall to investigate the runoff and erosion processes on sloped plots with three loessial soils and analyzed the effects of soil texture and ASWC on the hydrological processes. The results demonstrated that the average infiltration rate decreased with increasing clay content (i.e., Ansai (AS) loamy sand > Fuxian (FX) clay loam > Yangling (YL) clay). ASWC had little effect on infiltration processes for the YL clay but exerted a significant influence on

Lan Ma (corresponding author) Junyou Li Jingjing Liu Key Laboratory of State Administration of Forestry and Grassland on Soil and Water Conservation, School of Soil and Water Conservation, Beijing Forestry University, Beijing 100083, China E-mail: mlpcz@sina.com

infiltration for the FX and AS soils; this implies that infiltration models for loamy soils must consider the effects of ASWC. The Horton model was found to describe infiltration processes in these loessial soils better than the Kostiakov or Philip models. The YL clay yielded much less sediment than the FX and AS soils, and its sediment yield rate gradually decreased with the rainfall duration. There was a negative relationship between clay content and sediment yield under high ASWC, but no clear relation under low ASWC. These erosion differences derived from the splash erosion for the YL clay, and the depressions or rills occurred on the loamy soil plots. Key words

| antecedent soil water content, erosion, inﬁltration, simulated rainfall, soil texture

INTRODUCTION The Loess Plateau of China has an arid to semi-arid climate

important role in sustaining vegetation growth. In particular,

and has long experienced severe soil erosion due to its erod-

antecedent soil water content (ASWC) affects runoff gener-

ible soil texture, steep hillslopes, and frequent rain storms

ation and related soil erosion processes. Inﬁltration, runoff,

(Tang ). In fact, in some regions of steep terrain

and erosion processes within the Loess Plateau have been

within the plateau, soil loss rates reach up to 5,000–10,000

investigated previously to help conserve and use scarce

Mg km 2 a 1; such pronounced soil loss causes severe nutri-

soil and water resources across the region (Pan et al. ,

ent loss, induces widespread land degradation, and results in

; Wei et al. ; Mei et al. ).

frequent severe ﬂooding along the Yellow River (Fu et al.

Soil inﬁltration processes under natural rainfall con-

). In arid and semi-arid areas, soil moisture plays an

ditions are complex, owing primarily to spatiotemporal variability in soil properties and rainfall (Geiger & Durnford

This is an Open Access article distributed under the terms of the Creative

; Dong et al. ). Typically, soil inﬁltration processes

Commons Attribution Licence (CC BY-NC-ND 4.0), which permits copying

are investigated by subjecting disturbed or undisturbed soil

and redistribution for non-commercial purposes with no derivatives, provided the original work is properly cited (http://creativecommons.org/

columns to a stable hydraulic head or simulated rainfall;

licenses/by-nc-nd/4.0/).

the inﬁltrated water volume and the development of the

doi: 10.2166/nh.2020.013

883

L. Ma et al.

Effects of soil moisture on inﬁltration and erosion processes

Hydrology Research

51.5

2020

wetting front are commonly recorded (Youngs ). These

However, the inﬁltration of rainfall into soils is typically

soil column tests, particularly being subjected to ponding

greater for soils with more coarse particles than for those

hydraulic heads, only focus on a hydrological point, in con-

with ﬁner particles under the same rainfall conditions.

trast to soil inﬁltration processes that occur on ﬁeld

ASWC has an important impact on both rainfall–runoff

hillslopes during rainfall events (Boers et al. ).

and soil erosion processes, yet little information is available

Soil texture tends to become coarser from south to north on the Loess Plateau, while precipitation decreases from

regarding the effects of ASWC on the relationship between soil texture and erodibility.

600 to 200 mm a 1 (Tang ). Such differences in soil tex-

The primary objectives of the present study are as fol-

ture and precipitation have a pronounced effect on ASWC

lows: (1) to evaluate the effectiveness of three inﬁltration

and erosion processes on hillslopes (Lado et al. ).

models and (2) to investigate the differences in inﬁltration

Numerous well-established soil inﬁltration models have

and erosion processes between the three loessial soils. The

been reported previously, including the Horton, Philip,

results presented here will form a useful dataset for the

and Green–Ampt models. Both the selection of an appropri-

future assessment of soil erodibility and will help provide

ate model and the quantiﬁcation of model parameters are

theoretical guidance for erosion control in the loess area.

necessary to describe hillslope rainfall–runoff processes. Previous studies have conducted soil column tests to elucidate the effects of soil properties or ASWC on inﬁltration

DATA AND METHODS

processes (Marcus et al. ; Makoto et al. ), and it is worth further comparing these ﬁndings of such previous

Experimental set-up

studies with inﬁltration processes on sloped plots under rainfall conditions.

Experiments were conducted in an indoor rainfall simulator

The Universal Soil Loss Equation (USLE) and the Water

with a side-sprinkle rainfall set-up. With this set-up, rainfall

Erosion Prediction Project (WEPP) model are two popular

intensity could be controlled by adjusting spray nozzle size

predictive models for soil erosion. Soil erodibility (K ) is an

and water pressure, and rainfall uniformity exceeded 85%.

important factor in the USLE and is typically assessed

The fall height of raindrops was approximately 16 m,

based on soil texture (Wischmeier & Mannering ).

which ensured kinetic energy similar to that of natural

Extensive ﬁeld studies have been conducted by Simanton

rainfall.

et al. () and Elliot et al. () to develop methods of pre-

Six experimental plots were constructed, each with

dicting erodibility for cropland and rangeland soils based on

length, width, and depth of 2.00, 0.55, and 0.35 m, respect-

soil properties for use in the WEPP model. Nevertheless, as

ively; the plot size was selected considering the size of a

Laflen () suggested, the future expansion of databases

standard ﬁeld runoff plot (5 × 20 m) and avoiding marginal

detailing soil properties will be essential for continued veri-

effects. The plots were constructed using steel, with small

ﬁcation of predictive models. In particular, establishing a

holes at the bottoms of the plots to allow soil water percola-

relationship between sediment yield and soil texture will

tion. The experimental slope was adjusted to 10 ; this is a

be important for the development and veriﬁcation of such

common gradient for cultivated farmlands in the Loess Pla-

models.

teau of China. The tested soils were taken from Ansai

For a sloped plot, overland ﬂow occurs when rainfall

County (AS, 36 520 N, 109 190 E), Fuxian County (FX,

intensity exceeds soil inﬁltration capacity. Soil erosion pro-

35 530 N, 108 360 E), and Yangling District (YL, 34 160 N,

cesses are controlled primarily by soil erodibility and

108 050 E), which are located along the north to south of

rainfall–runoff–erosion

raindrop

the Loess Plateau of China. The particle size distributions

dynamics,

including

splash and overland ﬂow scour (Vermang et al. ).

of these soils are presented in Figure 1. The contribution

When soil detachment is controlled primarily by overland

by the weight of clay particles (<2 μm) to the total soil

ﬂow, detachment rates tend to decrease with increasing

weight was 28%, 20%, and 12% for YL, FX, and AS soils,

(Wischmeier & Mannering ).

respectively. The YL, FX, and AS soils belong to the clay

soil clay content

884

L. Ma et al.

Effects of soil moisture on inﬁltration and erosion processes

Hydrology Research

51.5

2020

concentration was determined as the ratio of the total dry sediment mass collected to the total runoff volume, and the runoff coefﬁcient was determined as the ratio of surface runoff to corresponding rainfall. A soil’s instantaneous inﬁltration rate ( fi) can be calculated according to the following formula (Pan et al. ):

fi ¼ I cos θ

Figure 1

10Ri St

(1)

where I is rainfall intensity (mm min 1), θ is slope gradient Particle size distributions for tested soils.

( ), t is the time interval required to collect the runoff sample (min), Ri is the runoff collected during the ith time

soil, clay loam soil, and loamy sand categories, respectively, according to the international soil texture classiﬁcation system (Hillel ).

interval (mL), and S is the area of the plot (cm2). In empirical inﬁltration models, the Kostiakov equation is typically used to describe inﬁltration processes in exper-

The soils were gently crushed before being screened

imental soils (Kostiakov ):

with a 10 mm sieve. In each plot, the soil was packed into three layers with a thickness of 10 cm and a total depth of 30 cm and a soil density of 1.20 g cm 3. To avoid discontinu-

f(t) ¼ i0 t a

(2)

ities between neighboring layers in each plot, each soil layer was lightly raked before the next soil layer was packed. The

where f(t) is the instantaneous inﬁltration rate (mm min 1),

outlet of each plot was constructed like a shutter, with eight

i0 is the initial inﬁltration rate (mm min 1), t is the duration

steel sheets (0.55 × 0.02 m) attached to each plot at intervals

(min), and a is an empirical parameter associated with soil

of 2 cm. Rainfall I was simulated for a dry run with low

texture.

ASWC (approximately 0.13 m3 m 3). Rainfall II was applied 3 days after Rainfall I in the same soil plots for a wet run with high ASWC (0.3–0.4 m3 m 3).

Here, physically based models including the Horton and Philip models were selected to discuss inﬁltration processes for the three soils considered. The Horton equation can be written as follows (Horton ):

Data measurement and analysis Both rainfalls with the intensity of about 100 mm h 1 were

f(t) ¼ fc þ (f0 fc )e αt

(3)

administered for about 70 min, in line with typical storm characteristics for the Loess Plateau. Two replicates of

where f(t) is the instantaneous inﬁltration rate (mm min 1),

each treatment were subjected to the simulated rainfall sim-

fc is the ultimate inﬁltration capacity or stabilized inﬁltration

ultaneously. The time until runoff initiation was recorded for

rate (mm min 1), f0 is the initial inﬁltration rate (mm min 1),

each treatment; after runoff initiation, runoff and sediment

t is the duration (min), and α is the decay parameter. The

were collected for each test at 3-min intervals throughout

parameters fc and α depend primarily on soil properties

the rainfall event. When sediment had been deposited

and antecedent moisture. This equation assumes ponding

enough, the sediment was separated from water, oven

conditions for the soil throughout the inﬁltration process

dried for 24 h at 105 C, and weighed. Inﬁltration rates

due to the generation of overland runoff.

were calculated by subtracting the measured runoff rates from

the

rainfall

intensity.

The

average

sediment

Philip () suggested that soil inﬁltration rates decrease with time according to a power-law equation and

L. Ma et al.

885

Effects of soil moisture on inﬁltration and erosion processes

Hydrology Research

51.5

2020

order: YL clay (0.80) > FX clay loam (0.56) > AS loamy

can be expressed as follows:

sand (0.40). 1 f(t) ¼ A þ St 0:5 2

Under high ASWC (0.30–0.40 m3 m 3; Rainfall II), the

(4)

average inﬁltration rates and runoff coefﬁcients for the three soils were 0.31–0.50 mm min 1 and 0.71–0.81, respectively.

where A is the constant inﬁltration rate (cm min ), t is

The AS loamy sand had a higher average and stabilized inﬁltra-

the duration (min), and S is the soil suction wetting rate

tion rate than the YL clay and FX clay loam, and there was no

(cm min 0.5), which can be obtained by ﬁtting the cumulative

signiﬁcant difference (p ¼ 0.05) between the YL clay and FX

inﬁltration amount and t 0.5 at the beginning of inﬁltration.

clay loam (Table 1). The average and stabilized runoff coefﬁ-

This equation assumes ponding inﬁltration into deep soil

cients for the three soil types decreased in the following

with uniform antecedent water content.

order: YL clay (0.81) > FX clay loam (0.78) > AS loamy sand

Analysis of variance (ANOVA) and paired t-tests were

(0.71). These results demonstrate that soil inﬁltration rates

used to analyze possible differences in soil inﬁltration and

increase as clay content decreases and indicate that the

sediment yield processes between the three soils and between

runoff coefﬁcient decreases with decreasing clay content

high and low ASWC. Correlation and regression analyses

under similar rainfall conditions. This is in accordance with pre-

were also undertaken to investigate the relationship between

vious ﬁndings based on soil column tests (Makoto et al. ).

sediment yield and runoff. The above analyses were per-

Based on the comparison between Rainfall I and II,

formed using SPSS 19.0 (by SPSS Inc., an IBM Company).

ASWC had no significant effect (p ¼ 0.05) on the average infiltration rate and runoff coefficient for the YL clay. This behavior is similar to that reported by Benito et al. (),

RESULTS AND DISCUSSION

who demonstrated little difference in runoff or soil inﬁltration rates between dry and wet periods for clay soils. This

Runoff and inﬁltration

can be attributed to the low porosities typical of clay soils. In contrast, the inﬁltration rates for the AS loamy sand 3

Under low ASWC conditions (0.13 m m ; Rainfall I), the

and FX clay loam under low ASWC were both about two

average inﬁltration rates for the YL clay, FX clay loam,

times greater than those under high ASWC. This result

agrees with both Cerdà () and Jones (), who found

and AS loamy sand were 0.34, 0.66, and 0.97 mm min , respectively. Signiﬁcant differences between the three soils

that ﬁeld soils with sandy or loamy characteristics exhibit

were found at the p ¼ 0.05 level (Table 1). The runoff coefﬁ-

signiﬁcantly greater inﬁltration rates during dry periods

cients for the different soil types decreased in the following

than during wet periods.

Table 1

Runoff, inﬁltration, and sediment yield characteristics for YL clay, FX clay loam, and AS loamy sand plots under low and high ASWC The stabilized valuea

The average value during the rainfall process Soil

ASWC (m3 m

)

IR b

SYR

0.80 a

5.13 a

3.76 a

0.27 a

0.84 a

2.04 a

1.43 a

0.13

0.34 a

0.13

0.66 b

0.56 b

9.26 b

11.21 b

0.40 b

0.73 b

9.68 b

9.07 b

0.13

0.97 c

0.40 c

2.37 c

3.77 a

0.80 c

0.56 c

1.88 a

2.11 a

0.30

0.31 a

0.81 a

4.36 a

3.30 a

0.31 a

0.81 a

2.44 a

1.86 a

0.35

0.34 a

0.78 a

32.72 b

27.36 b

0.25 a

0.86 a

12.75 b

9.70 b

0.40

0.50 b

0.71 b

35.69 c

29.76 c

0.43 b

0.75 b

–

IR, inﬁltration rate (mm min ); RC, runoff coefﬁcient; SYR, sediment yield rate (g m a The average of the last four observed values during the later phase of rainfall. b

min ); SC, sediment concentration (kg m ).

The same letter represents no signiﬁcant differences at the p ¼ 0.05 level among the three soil using ANOVA.

L. Ma et al.

886

Effects of soil moisture on inﬁltration and erosion processes

Hydrology Research

51.5

2020

The stable inﬁltration rates and runoff coefﬁcients

decreasing clay content. The regressed power equations

attained at the later stage of rainfall exhibited variation simi-

were converted into logarithmic lines, and the differences

lar to that observed for the average values presented in the

between the three soil types were further analyzed. Under

preceding paragraph. Signiﬁcant differences in these par-

low ASWC, signiﬁcant differences (p ¼ 0.05) in cumulative

ameters between low and high ASWC conditions were

inﬁltration processes were found between the three soils;

found for the FX clay loam and AS loamy sand, but not

under high ASWC, the AS loamy sand exhibited a signiﬁ-

for the YL clay. These results provide further evidence that

cantly higher inﬁltration volume than both the FX clay

ASWC likely has little effect on clay soils but has a signiﬁ-

loam and YL clay. Low ASWC generated a greater inﬁltra-

cant effect on loamy or sandy soils.

tion volume than high ASWC for both FX clay loam and

For the three soils considered, the cumulative inﬁltra-

AS loamy sand. In contrast, the cumulative inﬁltration

tion volume (F(t)) increased with rainfall duration; this

volume varied little between low and high ASWC for the

relationship can be described well by the power-law

YL clay (Figure 2 and Table 1), although the low ASWC cor-

equation F(t) ¼ atb (Figure 2 and Table 2). The exponent b

responded to a higher a-value in the regressed equations.

in the regressed equations was found to be in the ranges

This indicates that, for clay soils, ASWC may have a greater

0.52–0.77 and 0.74–0.83 for low and high ASWC, respect-

inﬂuence on initial inﬁltration processes than later period.

ively; moreover, values of b were found to increase with

Changes in the infiltration rate throughout the simulated rainfall event are shown in Figure 3 for all three soils and both low and high ASWC. Under low ASWC conditions (Rainfall I), overland sheet flow occurred 2.5, 5, and 7 min after the initiation of simulated rainfall for the YL clay, FX clay loam, and AS loamy sand, respectively. The infiltration rate for the YL clay decreased quickly to reach 0.3 mm min 1 after the initial 8.5 min and then stabilized; conversely, the infiltration rate decreased continuously throughout the rainfall duration for both the FX clay loam and AS loamy sand. However, under high ASWC conditions, the infiltration rates of all three soils decreased abruptly, with all soil types reaching stable values within 10 min. The Kostiakov, Horton, and Philip models were adopted

Figure 2

Table 2

Cumulative inﬁltration processes for YL clay, FX clay loam, and AS loamy sand

to ﬁt the inﬁltration processes illustrated in Figure 3 using

plots under low (L) and high (H) ASWC.

the least square method (Table 3).

Best-ﬁt equations for relationship between cumulative inﬁltration volume (F(t)), cumulative sediment yield (S(t)), and rainfall time (t) for soils considered under low and high ASWC F(t) 3

ASWC (m3 m

0.13

)

S(t)

Soil

Number

The best-ﬁt equation

F(t) ¼ 2.290t 0.517

0.974

S(t) ¼ 59.970t 0.430

0.949

0.668

0.998

S(t) ¼ 0.083t 2.145

0.992

0.998

S(t) ¼ 61.61ln(t) 105.9

0.967

0.760

0.978

S(t) ¼ 43.850t

0.995

0.998

S(t) ¼ 800.50ln(t) 1,202.0

0.971

0.995

S(t) ¼ 88.68e

0.965

0.13

F(t) ¼ 2.719t

0.13

F(t) ¼ 2.619t 0.767

0.30

F(t) ¼ 0.741t

0.35

F(t) ¼ 0.968t 0.743

F(t) ¼ 0.981t

0.40

0.829

0.446

0.051t

L. Ma et al.

887

Effects of soil moisture on inﬁltration and erosion processes

Hydrology Research

51.5

2020

sand and YL clay were found to have greater f0-values than the FX clay loam. The greater f0-value obtained for the YL clay may be related to the abrupt decline in the infiltration rate during the initial phase of rainfall (Figure 3). The parameters A and S in the Philip model represent the stabilized and initial infiltration rates, respectively. Both A and S increased with increasing clay content. However, the Philip model generated a significantly lower stabilized and initial infiltration rate than the Horton model (i.e., A < fc and S < f0; Table 3). Under the simulated rainfall conditions, it was difficult to accurately observe the initial soil infiltration rate owing Figure 3

Inﬁltration rates vs. rainfall time for YL clay, FX clay loam, and AS loamy sand plots under low (L) and high (H) ASWC.

to the relatively low rainfall intensity. However, the Horton model produced stabilized inﬁltration rates ( fc) that were close to the observed values (Tables 1 and 3),

Based on the coefﬁcient of determination (R 2), the

and higher R 2 values were obtained for the Horton model

Kostiakov model can describe inﬁltration processes better

than the Kostiakov and Philip models. These results suggest

for the FX clay loam and AS loamy sand than for the YL

that the Horton model is more suited than the Kostiakov or

clay. The i0-values in the Kostiakov model were found to

Philip models to describing the inﬁltration processes of loes-

for low

sial soils under the rainfall conditions considered here. Zhao

and high ASWC, respectively. Moreover, the FX clay loam

et al. () found that both the Horton and Philip models

and AS loamy sand exhibited signiﬁcantly higher i0-values

can describe soil inﬁltration processes well for the loessial

than the YL clay. This ﬁnding supports the assertion that initial

soils of the Loess Plateau based on soil column tests.

inﬁltration rates decrease with increasing clay content.

Additionally, Genachte et al. () conducted experiments

vary in the ranges 0.77–2.93 and 0.62–1.0 mm min

The Horton model can give both initial ( f0) and stabil-

on the soil inﬁltration behaviors of Arenosol and Ferralsol

ized inﬁltration rates ( fc). For the three soils, the ﬁtted

soils in tropical rain forests and suggested that the effective-

fc-values were found to decrease with increasing clay con-

ness of inﬁltration models depends partly on soil properties.

tent and were close to the observed values (Tables 1 and

In particular, Genachte et al. () found that the Philip

3). The ﬁtted f0-values of the three soils were in the ranges

and Horton models could describe inﬁltration processes

under low (Rainfall I)

for the Arenosol better than the Kostiakov model, while

and high ASWC (Rainfall II), respectively. The AS loamy

all of the models considered (i.e., Philip, Kostiakov, and

1.50–2.51 and 1.44–1.75 mm min

Table 3

Fitted inﬁltration models for YL clay, FX clay loam, and AS loamy sand under low and high ASWC

Kostiakov model 3

Soil

ASWC (m3 m

0.13

)

Horton model

Philip model

Number

f (t) ¼ i0 t a

f (t) ¼ fc þ (f0 fc )e αt

f(t) ¼ 0.772t 0.29 0.46

0.44

f(t) ¼ 0.262 þ 2.254e 0.375t

0.91

f(t) ¼ 0.40 þ 1.096e

0.94

f(t) ¼ 0.675 þ 1.115e 0.051t

0.06t

0.13

f(t) ¼ 2.836t

0.13

f(t) ¼ 2.934t 0.34 0.21

0.37

f(t) ¼ 0.28 þ 1.471e

0.76

f(t) ¼ 0.293 þ 1.143e 0.289t

0.70

0.381t

0.30

f(t) ¼ 0.622t

0.35

f(t) ¼ 1.017t 0.35

0.21

0.40

f(t) ¼ 0.998t

0.463t

f(t) ¼ 0.467 þ 1.261e

1 f (t) ¼ A þ St 0:5 2

0.94

f (t) ¼ 0.025 þ 1.387t 0.5

0.72

0.94

f (t) ¼ 0.069 þ 2.838t 0.5

0.94

0.96

f (t) ¼ 0.322 þ 3.210t 0.5

0.95

0.5

0.94

f (t) ¼ 0.11 þ 0.943t

0.91

f (t) ¼ 0.115 þ 1.026t 0.5

0.92

0.93

0.5

0.88

f (t) ¼ 0.295 þ 0.958t

0.77

888

L. Ma et al.

Effects of soil moisture on inﬁltration and erosion processes

Hydrology Research

51.5

2020

Horton models) could predict inﬁltration behaviors well for

behavior with an exponential distribution, with the cumulat-

the Ferralsol.

ive sediment yield accelerating during the later phase of the

The clay contents of the YL, FX, and AS soils were

rainfall. This behavior differs markedly from that of the

approximately 28%, 20%, and 13%, respectively (Figure 1),

other soil types (Figure 4). The YL clay exhibited similar

and their corresponding average and steady inﬁltration

sediment yield distributions for both high and low ASWC.

rates decreased in the following order: AS > FX > YL. This

In contrast, the obtained cumulative curves are markedly

provides further evidence that the soil inﬁltration rate

different under high and low ASWC for the FX and AS

decreases with increasing clay content. For both the FX

soils with high ASWC generating signiﬁcantly greater sedi-

and AS soils, high ASWC generated a lower steady inﬁltra-

ment yields than low ASWC. These results indicate that

tion rate; in contrast, ASWC had little effect on the

the sediment yield may not be predicted effectively based

inﬁltration rate for the YL clay (Figure 3 and Table 2).

on rainfall duration under different ASWC conditions. Under low ASWC (Rainfall I), the FX clay loam

This indicates that, for soils with high clay content (i.e., >30%), ASWC has little effect on soil inﬁltration processes

plot

exhibited 2

average

sediment

yield

rate

during rainfall events. However, for loamy soils (i.e., clay

9.26 g m

content <20%), inﬁltration models should consider the

than the values obtained for the YL clay and AS loamy

effects of ASWC. These results may reﬂect the higher poros-

sand, respectively. Thus, the average sediment yield rate

ity of loamy soils relative to clay soils, allowing loamy soils

was lowest for the AS loamy sand, likely because this soil

to both hold more water and induce greater expansion of

exhibited the highest average inﬁltration rate (Table 1) and

soil volume.

the lowest average runoff rate. No signiﬁcant difference

min ; this value is 1.8 and 3.9 times greater

(p ¼ 0.05) was found between the average sediment concenSediment yield

trations for the AS loamy sand and YL clay (Table 1). The higher sediment yield for the FX clay loam plot can be attrib-

Cumulative sediment yields increased with rainfall duration, exhibiting power-law, exponential, or logarithmic linear distributions (Figure 4 and Table 2). The differences found in the best-ﬁt equations indicate that sediment yield processes (or rather, change processes in sediment yield with the rainfall duration) varied with soil texture and ASWC. Under high ASWC, the AS loamy sand exhibited sediment yield

uted to the occurrence and development of rills. In contrast, the difference in sediment yields between the AS loamy sand and YL clay can be attributed primarily to differences in soil infiltration processes. The AS loamy sand exhibited an average infiltration rate that was approximately three times greater than that of the YL clay. Therefore, the AS loamy sand plot can be considered to have had lower runoff erosion energy, even though its soil erodibility would have been higher than that of the YL clay (Wischmeier & Mannering ; Elliot et al. ). During the later phase of rainfall, the FX clay loam exhibited a stabilized sediment yield rate of 9.26 g m 2 min 1, which is almost five times that obtained for the YL clay and AS loamy sand (Figure 5). These results demonstrate that differences in the sediment yield rate between the three soil types are considerably greater during the later stage of rainfall than during the earlier stage. Under high ASWC (Rainfall II), FX clay loam and AS loamy sand exhibited similar sediment yield rates (32.7 and 35.7 g m 2 min 1, respectively) and sediment concen-

Figure 4

Cumulative sediment yield vs. rainfall duration for YL clay, FX clay loam, and AS loamy sand plots under low (L) and high (H) ASWC.

trations (27.36 and 29.76 kg m 3, respectively). These values are eight and nine times those of the YL clay,

889

L. Ma et al.

Effects of soil moisture on inﬁltration and erosion processes

Hydrology Research

51.5

2020

sediment yield rates under low ASWC. In contrast, initial sediment yield rates were lower than steady sediment yield rates for the FX soil; this could be attributed to the development of depressions under low ASWC for this soil type (Figure 5). Similar results have been reported previously for the YL and AS soils (Pan et al. ) and may reflect expansion and shrinkage of the surface soil layer and the greater raindrop splash during the initial rainfall stage. The FX clay loam and AS loamy sand plots generated significantly higher sediment yields under high ASWC (Rainfall II) than under low ASWC (Rainfall I). In contrast, no clear difference in sediment yield between low and high ASWC was observed for the YL clay plot (Table 1). This suggests that ASWC has a more significant effect on loamy or sandy soils than clay soils. Differences in sediment yield processes between low and high ASWC were considered using paired t-tests. No significant difference was found for the YL clay. However, high ASWC led to sediment yields that were 3.5 and 15 times higher than those under low ASWC for the AS loamy sand and FX clay loam, respectively. Benito et al. () found sediment concentration to be greater during wet periods than dry periods for all water-repellent soil sites considered. Here, the sediment yields obtained for the AS loamy Figure 5

(a) Sediment yield rate and (b) sediment concentration vs. rainfall duration for YL clay, FX clay loam, and AS loamy sand plots under low (L) and high (H) ASWC.

sand and FX clay loam under different ASWC support the ﬁndings of Benito et al. (), although the YL clay results do not. This highlights the importance of the effects of soil texture and ASWC on soil erosion.

respectively (4.36 g m 2 min 1 and 3.3 kg m 3) (Table 1).

Under low ASWC (Rainfall I), the sediment concen-

The greater sediment yields of the FX clay loam and AS

tration for the FX clay loam increased gradually, reaching

loamy sand are in accordance with their relatively low soil

approximately 17.5 kg m 3 after 35 min, and then stabilized

erodibility compared to the YL clay. Moreover, during the

at 10–15 kg m 3 (Figure 5). This behavior may be associated

later phase of rainfall, the sediment yield rates of the YL

with the development of rills. In contrast to the FX clay

clay and FX clay loam remained relatively stable, while

loam, both sediment yield rate and sediment concentration

the AS loamy sand plot exhibited signiﬁcant increases in

for the YL clay and AS loamy sand plots increased rapidly,

sediment concentration and the sediment yield rate

peaking after 3 min, and then decreased slowly to reach

(Figure 5).

stabilized values during the later phase of rainfall (Figure 5).

In the present study, no clear relationship was observed

No rill development was observed in the YL clay and AS

between clay content and sediment yield under low ASWC,

loamy sand plots. This, combined with the occurrence of

although a negative relationship was observed under high

peak sediment yields during the initial phase of rainfall for

ASWC (Figures 4 and 5). This could be a result of different

these soils, supports the assertion that soil erosion in the

erosional forms, and depressions or rills tended to occur on

YL clay and AS loamy sand plots results primarily from rain-

the loamy soil plots with high ASWC. For the YL and AS

drop splash. Moreover, easily-eroded soils are typically

soils, initial sediment yield rates were greater than steady

transported during the initial phases of rainfall events,

890

L. Ma et al.

Effects of soil moisture on inﬁltration and erosion processes

Hydrology Research

51.5

2020

such that their supply is exhausted by the time later runoff phases occur. This phenomenon has been observed frequently on vegetated slopes or on relatively small plots of bare soil (Pan & Shangguan ). Under high ASWC (Rainfall II), notable differences in sediment yield processes were observed between the different soil types. The sediment yield rate and sediment concentration for the YL clay plot decreased gradually, while those for the AS loamy sand plot increased continuously. The decreasing trend observed for the YL clay plot can be attributed primarily to the dominance of raindrop splash erosion processes. Conversely, the continuous increase in erosion for the AS loamy sand plot can be attributed to soil collapse in the downslope area under high ASWC. Both the sediment

Figure 6

Relationship between the runoff rate and the sediment yield rate for YL clay, FX clay loam, and AS loamy sand plots under low (L) and high (H) ASWC.

yield rate and sediment concentration were found to increase linearly for the FX clay loam plot, peaking after 15 min,

inﬁltration rate initially decreased abruptly before remaining

before decreasing gradually to reach a constant value. This

relatively constant during the later phase of runoff (Figure 3).

behavior can be attributed primarily to the accelerated devel-

These different relationships under low and high ASWC indi-

opment of rills during the initial phase of rainfall and their

cate that ASWC has an important effect on the correlation

relative stability during later phases.

between the runoff rate and the sediment yield rate.

For the YL clay, no signiﬁcant difference (p ¼ 0.05) in

When both low and high ASWC are considered

sediment yield processes was observed between low and

together, the results presented here indicate that the sedi-

high ASWC (i.e., between Rainfall I and II). In particular,

ment yield rate decreased linearly with the runoff rate for

the sediment concentration decreased rapidly from an initial

the YL clay; in contrast, higher runoff rates led to increased

after 12 min; then, it

sediment yield rate (as described by a power-law function)

decreased gradually to reach a stable value of 2.0 kg m 3

for both the AS loamy sand and FX clay loam (Figure 6).

during the later stage of rainfall. However, for the FX clay

In the present study, the plot surfaces were observed after

loam and AS loamy sand, high ASWC (Rainfall II) yielded

each rainfall. Scattered depressions were observed for the

signiﬁcantly more sediment than low ASWC (Rainfall I)

AS and FX soils, while neither rills nor depressions were

(Figure 5). Although the AS loamy sand and FX clay loam

observed for the YL clay under either low or high ASWC.

yielded similar volumes of sediment overall, differences in

This suggests that raindrop splash dominates erosion

sediment processes between these two soil types were very

processes for the YL clay, while runoff scouring likely

signiﬁcant (Figure 5).

dominates the AS and FX soils. For the YL clay, more soil

peak value of 23.6–3.5 kg m

particles were eroded by raindrop splash and delivered to Relationship between the sediment and runoff

the outlet at the beginning of the rainfall than later. Correspondingly, a negative relationship was observed between

Under low ASWC (Rainfall I), a signiﬁcant positive relation-

the sediment yield rate and the runoff rate for the YL clay.

ship was observed between the sediment yield rate and the

Previous studies have observed a similar negative relation-

runoff rate for the FX clay loam, with a correlation coefﬁcient

ship between erosion and the runoff rate for soils with

(R) of 0.76; conversely, a negative relationship (R ¼ 0.74) was

high clay content (Pan et al. ).

found for both the YL clay and AS loamy sand. However,

Under the same runoff rates, high ASWC led to greater

under high ASWC (Rainfall II), the relatively stable runoff

sediment yield rates for the FX clay loam and AS loamy

rate corresponded to a greater variation in the sediment

sand. This can be attributed primarily to the development

yield rate (Figure 6); this likely occurred because the soil

of rills and the collapse of the soil mass. However, for the

891

L. Ma et al.

Effects of soil moisture on inﬁltration and erosion processes

YL clay, low ASWC was found to generate greater sediment

Hydrology Research

Table 4

soil particles being more detachable under low ASWC.

51.5

2020

Best-ﬁt equations between cumulative sediment yield (S) and cumulative runoff volume (R) for soils considered under low and high ASWC

yield rates than high ASWC during the initial phase of rainfall for the same runoff rates. This result can be attributed to

ASWCs (m3 m

)

The tested soils

The best-ﬁt equations

Number

The positive relationship between the runoff rate and

0.13

YL clay

S ¼ 80.85R 0.350

0.982

the sediment yield rate for the FX clay loam and AS

0.13

FX clay loam

S ¼ 13.84R 66.68

0.987

0.364

0.941

0.998

loamy sand plots supports a popular viewpoint that increas-

0.13

AS loamy sand

S ¼ 45.32R

ing the runoff rate tends to enhance the soil detachment rate

0.30

YL clay

S ¼ 45.05R 0.408

owing to increases in erosional force or runoff energy

0.35

FX clay loam

S ¼ 34.06ln(R) þ 36.2

0.952

(Laflen ). However, for the YL clay plot, increasing

0.40

AS loamy sand

S ¼ 95.27e (0.042R)

0.963

the runoff rate led to decreasing the sediment yield rate (Figure 6). These results suggest that inter-rill erosion dominated by raindrop splash (as occurred for the YL clay) may have little relation to overland runoff. For a rainfall depth of approximately 100 mm, the AS loamy sand and FX clay loam plots produced approximately 82 mm of runoff under high ASWC (Rainfall II). This is signiﬁcantly greater than the runoff depths generated under low ASWC (Rainfall I), which were 44 and 60 mm for the AS loamy sand and FX clay loam, respectively. Under high ASWC, the FX clay loam and AS loamy sand plot yielded 2,159 and 2,355 g m 2 sediments, respectively; these values are 3 and 13 times, respectively, the corresponding values observed under low ASWC (Rainfall I; Figure 7). In contrast, runoff volume and sediment yield for the YL clay varied little between low and high ASWC. Regression analysis was undertaken to predict the relationship between cumulative sediment yield and cumulative runoff (Table 4). Generally, a linear or power-law equation

was found to effectively describe this relationship, except for the FX clay loam and AS loamy sand under high ASWC (Rainfall II). For the YL clay plot, and for the AS loamy sand plot under low ASWC only, the cumulative sediment yield increased with total runoff volume and the exponent of the regressed power equation was in the range of 0.35– 0.41. In this context, an exponent smaller than 1.0 indicates that the erosion rate would decrease gradually with rainfall duration. Such behavior is in line with that expected to occur when inter-rill erosion processes are dominated by raindrop splash. A linear relationship in this context indicates that the erosion rate remains relatively constant throughout the rainfall period; this is considered to be in line with the development of rill erosion within the FX clay loam plot under low ASWC (Table 4). Linear or power-law relationships between runoff and sediment load have been reported previously at the watershed scale (García-Ruiz et al. ; López-Tarazón et al. ; Tuset et al. ). However, the exponential equation obtained for the AS loamy sand plot under high ASWC indicates that the soil detachment rate accelerated gradually during the rainfall simulation; this behavior can be explained by soil collapse in the downslope area. In general, the relationship between runoff and sediment depends primarily on soil erosion processes, which in turn are controlled partly by soil texture and ASWC.

CONCLUSIONS The simulated rainfall experiments in sloped soil plots were Figure 7

Cumulative sediment yield vs. cumulative runoff for YL clay, FX clay loam, and AS loamy sand plots under low (L) and high (H) ASWC.

used to investigate the effects of soil texture and ASWC (0.13

and

0.35 m3 m 3)

inﬁltration

and

erosion

892

L. Ma et al.

Effects of soil moisture on inﬁltration and erosion processes

processes. The clay (<2 μm) contents of the YL, FX, and AS soils in this test corresponded to approximately 28%, 20%, and 12% by weight, respectively. Under the same rainfall conditions, average soil infiltration rates decreased with increasing clay content. ASWC had little effect on infiltration processes for the YL clay but had significant effects for the FX and AS soils. The Horton model was found to more effectively describe infiltration processes for these three loessial soils than the Kostiakov or Philip models. Under both low and high ASWC, the YL clay had a decreasing sediment yield rate with rainfall duration and generated much lower sediment yield than the FX and AS soils. The low sediment yield for the YL clay was mainly attributed to the dominance of raindrop splash erosion. The FX and AS soils had greater sediment yields under high ASWC than under low ASWC, which derived from the development of rills. There was no constant relationship between the sediment yield rate and the runoff rate, implying the importance of erosion form in predicting the soil erosion process on a hillslope.

ACKNOWLEDGEMENTS This study was supported by the National Natural Science Foundation of China (Grant No. 51779004). We would like to express great gratitude to the reviewers for their valuable suggestions made to improve the manuscript.

AUTHOR CONTRIBUTIONS L.M. conceived and designed the experiments; L.M., J.L., and J.L. performed the experiments and analyzed the data; L.M. wrote the paper.

REFERENCES Benito, E., Santiago, J. L., Blas, D. E. & Varela, M. E.  Deforestation of water-repellent soils in Galicia (NW Spain): effects on surface runoff and erosion under simulated rainfall. Earth Surface Processes and Landforms 28, 145–155. Boers, T. M., Deurzen, V. F. J. M. P., Eppink, L. A. A. J. & Ruytenberg, R. E.  Comparison of inﬁltration rates with

Hydrology Research

51.5

2020

an infiltrometer, a rainulator and a permeameter for erosion research in S E Nigeria. Soil Technology 5, 13–26. Cerdà, A.  Seasonal variability of infiltration rates under contrasting slope conditions in southeast Spain. Geoderma 69, 217–232. Dong, H., Huang, R. Q. & Gao, Q. F.  Rainfall infiltration performance and its relation to mesoscopic structural properties of a gravelly soil slope. Engineering Geology 230, 1–10. Elliot, W. J., Laflen, J. M. & Kohl, K. D.  Effect of Soil Properties on Soil Erodibility. ASAE/CSAE, St. Joseph, MI, Paper No. 89-2150. Fu, B. J., Hu, C. X., Chen, L. D., Honnay, O. & Gulinck, H.  Evaluating change in agricultural landscape pattern between 1980 and 2000 in the Loess hilly region of Ansai County, China. Agriculture, Ecosystems & Environment 114 (2–4), 387–396. García-Ruiz, J. M., Regüés-Muñoz, D., Alvera, B., Lana-Renault, N., Serrano-Muela, M. P., Nadal-Romero, E., NavasIzquierdo, A., Latrón, J., Bono, C. E. M. & Arnáez-Vadillo, J.  Flood generation and sediment transport in experimental catchments affected by land use changes in the central Pyrenees. Journal of Hydrology 356, 245–260. Geiger, S. L. & Durnford, D. S.  Infiltration in homogeneous sands and a mechanistic model of unstable flow. Soil Science Society of America Journal 64, 460–469. Genachte, G. V., Mallants, D., Ramos, J., Deckers, J. A. & Feyen, J.  Estimating infiltration parameters from basic soil properties. Hydrological Processes 10, 687–701. Hillel, D.  Introduction to Soil Physics. Academic Press, New York. Horton, R. E.  An approach toward a physical interpretation of infiltration-capacity 1. Soil Science Society of America Journal 5 (C), 399–417. Jones, J. A. A.  Global Hydrology: Processes, Resources and Environmental Management. Longman, Harlow. Kostiakov, A. N.  On the dynamics of the coefficient of water percolation in soils and on the necessity for studying it from a dynamic point of view for purposes of amelioration. In: Transactions of 6th Committee International Society of Soil Science, Russia, Part A, pp. 17–21. Lado, M., Ben-Hur, M. & Shainberg, I.  Soil wetting and texture effects on aggregate stability, seal formation, and erosion. Soil Science Society of America Journal 68, 1992–1999. Laflen, J. M.  WEPP-erosion prediction technology. In: Soil Erosion and Dryland Farming (J. M. Laflen, J. L. Tian & C. H. Huang, eds). CRC Press, New York, pp. 557–566. López-Tarazón, J. A., Batalla, R. J., Vericat, D. & Balasch, J. C.  Rainfall, runoff and sediment transport relations in a mesoscale mountainous catchment: the River Isabena (Ebro basin). Catena 82, 23–34. Makoto, H., Heinz, G. S. & Daiki, A.  Removal of saline water due to road salt applications from columns of two types of sand by rainwater infiltration: laboratory experiments and model simulations. Water, Air, & Soil Pollution 230, 305.

893

L. Ma et al.

Effects of soil moisture on inﬁltration and erosion processes

Marcus, A. H., William, E. C., Richard, B. D., Greg, H., Shaun, L. & Kathrin, M.  Effect of antecedent soil moisture on preferential flow in a texture-contrast soil. Journal of Hydrology 398, 191–201. Mei, X. M., Zhu, Q. K., Ma, L., Zhang, D., Wang, Y. & Hao, W. J.  Effect of stand origin and slope position on infiltration pattern and preferential flow on a Loess hillslope. Land Degradation & Development 29, 1353–1365. Pan, C. Z. & Shangguan, Z. P.  Runoff hydraulic characteristics and sediment generation in sloped grassplots under simulated rainfall conditions. Journal of Hydrology 331, 178–185. Pan, C. Z., Shangguan, Z. P. & Lei, T. W.  Influences of grass and moss on runoff and sediment yield on sloped loess surfaces under simulated rainfall. Hydrological Processes 20, 3815–3824. Pan, C. Z., Ma, L. & Wainwright, J.  Particle selectivity of sediment deposited over grass barriers and the effect of rainfall. Water Resources Research 52, 7963–7979. Philip, J. R.  Theory of infiltration. In: Advances in HydroScience, Vol. 5 (T. H. Chow, ed.). Academic Press, New York, pp. 215–296. Simanton, J. R., Weltz, M. A., West, L. T. & Wingate, G. D.  Rangeland Experiments for Water Erosion Prediction Project. ASAE/CSAE, St. Joseph, MI, Paper No. 87-2545.

Hydrology Research

51.5

2020

Tang, K. L.  Soil and Water Conservation in China. Science Press, Beijing. Tuset, J., Vericat, D. & Batalla, R. J.  Rainfall, runoff and sediment transport in a Mediterranean mountainous catchment. Science of the Total Environment 546, 114–132. Vermang, J., Demeyer, V., Cornelis, W. M. & Gabriels, D.  Aggregate stability and erosion response to antecedent water content of a loess soil. Soil Science Society of America Journal 73 (3), 718–726. Wei, W., Jia, F. Y., Yang, L., Chen, L. D., Zhang, H. D. & Yu, Y.  Effects of surficial condition and rainfall intensity on runoff in a loess Hilly area, China. Journal of Hydrology 513 (1), 115–126. Wischmeier, W. H. & Mannering, J. V.  Relation of soil properties to its erodibility. Soil Science Society of America Journal 33 (1), 131–137. Youngs, E. G.  Infiltration measurements – a review. Hydrological Processes 5, 309–320. Zhao, J. B., Zhang, Y., Chen, B. Q. & Dong, Z. B.  Law of water infiltration of lower part of middle Pleistocene loess in Luochuan of Shaanxi. Acta Pedologica Sinica 46, 965–972.

First received 21 January 2020; accepted in revised form 10 March 2020. Available online 15 April 2020

894

51.5

2020

Assessment of hydrological drought based on nonstationary runoff data Xueli Sun, Zhanling Li and Qingyun Tian

ABSTRACT A nonstationary standardized runoff index (NSRI) is proposed by using the GAMLSS framework to assess the hydrological drought under nonstationary conditions. The deﬁnition of the NSRI is similar to that of SRI, but using a nonstationary Gamma distribution by incorporating meteorological variables and antecedent runoff as covariates to describe the characteristics of runoff series. The new drought index is then applied to the upper reach of the Heihe River basin. Four models are developed, in which one is stationary, and the other three are nonstationary with one, two and three covariates, respectively. Results show that, for the nonstationary runoff series, the nonstationary models are more robust and reliable than the stationary one. Among these models,

Xueli Sun Zhanling Li (corresponding author) Qingyun Tian MOE Key Laboratory of Groundwater Circulation and Environmental Evolution, China University of Geosciences (Beijing), Beijing 100083, China and School of Water Resources and Environment, China University of Geosciences (Beijing), Beijing 100083, China E-mail: zhanling.li@cugb.edu.cn

the model with two covariates performs the best. For the model with one covariate, the precipitation shows better in the ﬁtting as a covariate in rainy seasons, and the antecedent runoff shows better in dry seasons. The NSRI identiﬁes more drought events than SRI does, and the drought conditions in our case are mainly affected by precipitation. It is proved that the proposed new drought index is a more effective method for drought assessments under nonstationary conditions. Key words

| GAMLSS, hydrological drought, nonstationary, SRI

HIGHLIGHTS

• • • •

A nonstationary standardized runoff index is developed. Six alternative covariates and three kinds of nonstationary models are compared. The nonstationary model with two covariates performs the best. Hydrological drought conditions in this case is mainly affected by precipitation.

INTRODUCTION Drought is a kind of universal natural disaster which

the ecosystems more vulnerable (Zhang et al. ; Cheng

can bring serious ecological, environmental and social

; Fang et al. a; Han et al. a). Thus, drought

consequences. It may inhibit the growth of vegetation, accel-

has attracted wide attention from the researchers. The

erate the degradation of grassland and the disappearance of

accurate and timely drought assessment is crucial for

oasis, aggravate the reduction of rivers and lakes, and lead to

devising plans and carrying out the necessary mitigation measures especially under the background of climate

warming. Being a robust tool for drought assessment, many drought indices have been developed during the last decades

895

X. Sun et al.

Assessment of hydrological drought based on nonstationary data

Hydrology Research

51.5

2020

(e.g., Nalbantis & Tsakiris ; Vicente-Serrano et al. ;

covariates herein. In addition, evapotranspiration is an

Bloomfield & Marchant ; Tabari et al. ). One of the

important process of water transfer in the hydrosphere

most widely used indices worldwide in the hydrological

and atmosphere, playing a vital role in the hydrological

drought assessment is the standardized runoff index (SRI;

cycle and the runoff processes (Han et al. b; Aa et al.

Luciano et al. ; Jiang et al. ; Zhou et al. ),

). The antecedent runoff (runoff in the previous

which is proposed by Shukla & Wood () based on the

month) contributes to the runoff changes at that month

theory of the standard precipitation index (SPI; McKee

especially in non-ﬂood seasons (Ding et al. ). Conse-

et al. ). It remains the most broadly accepted index due

quently, the evapotranspiration and the antecedent runoff

to being simple to calculate, the limited data required, and

are also taken into account as alternative covariates. As

the different time scales of hydrological drought assessed.

the actual evapotranspiration data are not available

However, the stationarity assumption of the SRI is often

and not easy to calculate, potential evapotranspiration is

violated in the context of global warming and strong

used here. In summary, the six variables, precipitation,

human disturbance (Xiong & Guo ; Villarini et al.

temperature, relative humidity, wind speed, potential evapo-

; Liu et al. ). For example, Xiong & Guo ()

transpiration, and antecedent runoff, are incorporated in

found that the annual mean runoff series in the Yangtze

developing the NSRI.

River basin in China exhibited a clear decreasing trend

This study, taking the upper reach of the Heihe River

from 1882 to 2001. Villarini et al. () found that the

basin (the second largest inland river basin in northwest of

runoff series in the continental United States had signiﬁcant

China) as the study area, ﬁrst tests the stationarity of

change points during the 20th century. These mean that

runoff series over the study area and then develops an

many runoff series are not stationary any more, and the SRI

NSRI to assess hydrological droughts under nonstationary

with the assumption of the runoff data to be analyzed

conditions by using GAMLSS. GAMLSS is one of the tech-

keeping stationary is no longer suitable for ﬁtting such non-

niques for simulating nonstationary time series and capable

stationarity data (Wang et al. b). Therefore, a modiﬁed

of providing a high degree of ﬂexibility for solving nonsta-

SRI, namely, a nonstationary standardized runoff index

tionary modelling (e.g., Villarini et al. ; Zheng et al.

(NSRI), will be proposed in this study with incorporating

). It can describe the linear or nonlinear relationship

the nonstationary characteristic of runoff series.

between any statistical parameter of the random variable

NSRI is deﬁned similarly with the SRI, but using a non-

sequence and the explanatory variable, and the description of

stationary Gamma distribution by incorporating covariates.

the random variable distribution function has a wide range of

Seeking appropriate covariates is the ﬁrst, also one of the

types (Stasinopoulos & Rigby ). Many hydrologists

vital steps in constructing the nonstationary drought index

have applied this technique to analyze the nonstationary

(Vu & Mishra ). A single covariate is usually considered

hydrological sequences (Villarini et al. ; Zheng et al.

in previous literatures. For example, time or climatic vari-

). Six variables are considered as covariates in total and

ables are usually used to describe the precipitation

three nonstationary models are developed in terms of different

changes in developing the nonstationary meteorological

combinations of covariates. To verify the reliability of the

drought index (Russo et al. ; Li et al. ; Wang et al.

NSRI, it is ﬁnally compared with the traditional SRI according

a; Javad & Somayeh ). However, as known, the

to the historical droughts.

runoff changes are always affected by the recharge sources.

The outline of this paper is as follows. The study area,

The recharge sources for the upper reach of the Heihe River

data description, and the methodology of SRI and NSRI

basin include precipitation, glacial and snow melting, frozen

by using GAMLSS are covered in the section ‘Materials

soil thawing and groundwater (Ding et al. ; Wang et al.

and methods’. The modelling results and the choice of differ-

). These sources are affected by meteorological elements

ent combinations of covariates are discussed in the section

like temperature, relative humidity and wind speed. There-

‘Results and discussion’, followed by the conclusions in

fore, these meteorological elements are considered as

the section ‘Conclusion’.

896

X. Sun et al.

Assessment of hydrological drought based on nonstationary data

MATERIALS AND METHODS

Hydrology Research

51.5

2020

Monteith method (Allen et al. ), which is suggested by the Food and Agriculture Organization of the United

Materials

Nations. The locations of the stations are shown in Figure 1 and the detailed information are displayed in Table 1. These

Study area

datasets were obtained from the China Meteorological Data Network (http://data.cma.cn/) and the Hydrology

The Heihe River originates from Qilian County in the

Yearbook.

northeastern part of Qinghai Province. The river basin, between 97 370 E–102 060 E and 37 440 N–42 400 N, is located in the middle of Hexi Corridor in the northwest of China, with an area of 13,40,000 km2. The circulation

Methods

of the westerly belt in the middle and high latitudes and the polar air mass make the climate across the basin

Standardized runoff index

dry, strong wind and sunshine, large temperature differences, less precipitation and great evaporation, resulting in the scarcity of water resources in this basin (Gao & Zhao ). The basin is divided into three reaches, the upper reach, the middle reach and the lower reach. The annual precipitation for the upper reach is about 350 mm, with 70% occurring from June to October. The water resources over the basin are mainly generated from the upper reach and consumed in the middle and lower reaches. The water

Shukla & Wood () proposed the SRI, analogous to the SPI (McKee et al. ). Many studies have shown the efficacy of SRI in depicting the hydrological droughts (e.g., Luciano et al. ; Jiang et al. ). To obtain this index, fit the runoff series for a certain period by using Gamma distribution firstly and then transform it into a standard normal distribution through an equal probability transformation. The specific calculation process is as follows:

resources over the basin is 2.62 billion m3, supporting for

1. Let x(n) represents the runoff data at the nth month. Fifty-

1.5 million people and 384,000 ha of farmland irrigation

four years (1961–2014) of runoff data are available here,

(Ling ). The amounts and variations of water resources

thus n ranges from 1 to 648. For a given time scale of k

in the upper reach have great effects on agriculture, econ-

months (a time scale of 12-month is considered here),

omic development and ecological environment for the

the cumulative runoff xk (n) is calculated as:

middle and lower reaches. Thus, it is particularly necessary to carry on the researches about hydrological droughts in the upper reach.

xk (n) ¼

n X

(1)

x(i)

i¼n kþ1

Data 2. Fitting a two-parameter Gamma distribution, denoted as The monthly runoff data at three hydrological stations

xk (n) ∼ Gamma (μ, σ):

(Qilian, Zhamushike and Yingluoxia) and the meteorological data at two national meteorological stations (Qilian and Yeniugou) in the upper reach are used. The datasets span from 1961 to 2014. Runoff data are used to calculate the hydrological drought index, whereas meteorological

(xk (n))σ 2 1 exp[ (xk (n))=(σ 2 μ)] f(xk (n)jμ, σ) ¼ 2 1=σ Γ(1=σ 2 ) (σ 2 μ) 1

xk (n) > 0

(2)

data, including precipitation, temperature, relative humidity and wind speed, are used as covariates to establish the

where μ and σ are the location and scale parameters in

NSRI. The meteorological data are also used for calculating

Gamma, μ > 0 and σ > 0. Γ( ) is the mathematical

the potential evapotranspiration by employing the Penman–

Gamma function.

X. Sun et al.

897

Figure 1

Table 1

Assessment of hydrological drought based on nonstationary data

Hydrology Research

51.5

2020

Locations of hydrological and meteorological stations in the upper reach of the Heihe River basin.

Hydrological and meteorological stations in the upper reach of the Heihe River basin

4. Converting cumulative probability to a standard normal distribution function:

Station name

Longitude and

DEM

Station type

(abbreviation)

latitude

(m)

Hydrological station

Qilian (QL)

100.25 E, 38.18 N 99.98 E, 38.23 N 100.18 E, 38.8 N

2,708

Zhamushike (ZMSK) Yingluoxia (YLX) Meteorological station

Qilian (QL) Yeniugou (YNG)

100.25 E, 38.18 N 99.61 E, 38.83 N

2,982 1,682 2,708 3,227

SRI ¼

8 > > > > > > > > > > > > > > > > < > > > > > > > > > > > > > > > > :

c0 þ c1 ω þ c2 ω2 1 þ d1 ω þ d2 ω2 þ d3 ω3 vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi #ffi u " u 1 0 F(xk (n)) 0:5 ω ¼ tln F(xk (n))2

ω þ

c0 þ c1 ω þ c2 ω2 1 þ d 1 ω þ d 2 ω2 þ d 3 ω3 vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi #ffi u " u 1 0:5 < F(xk (n)) 1 ω ¼ tln (1 F(xk (n))2 ω

(4)

3. The cumulative probability for a given time scale can be calculated as:

where c0 ¼ 2.515517, c1 ¼ 0.802853, c2 ¼ 0.010328, d1 ¼

ðx F(xk (n)) ¼ f(xk (n))dx 0

(3)

1.432788, d2 ¼ 0.189269, and d3 ¼ 0.001308 (Shukla & Wood ).

898

X. Sun et al.

Assessment of hydrological drought based on nonstationary data

Hydrology Research

51.5

2020

Akaike information criterion (AIC) (Akaike ), together

Nonstationary standardized runoff index

with the worm plot, is used to test its goodness of ﬁt. When modeling time series over 30 years, we should pay

Since the NSRI proposed here is normalized in the same

attention to the nonstationarity of the time series which is

way as the traditional SRI, the same drought-level standards

potentially caused by climate change (Russo et al. ).

for both indices are suggested (Guttman ). As shown in

The NSRI is then deﬁned based on a nonstationary

Table 2, positive NSRI values indicate wet conditions, and

Gamma distribution with its location and scale parameters

negative values indicate dry conditions. The higher the

varying with covariates by using the GAMLSS framework

NSRI value, the wetter it is, and the lower the value, the

(Rigby & Stasinopoulos ; Stasinopoulos & Rigby ).

drier it is.

The GAMLSS framework was introduced by Rigby and Stasinopoulos (, ) and Akantziliotou et al. (), as a way to overcome some of the limitations associated with

RESULTS AND DISCUSSION

generalized linear models (GLMs) and generalized additive models (GAMs) (Nelder & Wedderburn ; Hastie & Tibshirani ). This framework is expanded to allow modeling not only the mean (or location) but other parameters of the distribution for a response variable as linear and/or nonlinear functions of explanatory variables and/ or random effects. In GAMLSS, the cumulative runoff xk (n) is modeled as xk (n) ∼ Gamma (μn , σ n ). It is assumed that the changes of runoff with covariates obey the following distribution parameters:

Stationary test Since it is the cumulative runoff series ﬁtted by a distribution, the stationarity of cumulative runoff series is to be tested. The augmented Dickey–Fuller (ADF) test, an augmented version of the original Dickey–Fuller test (Dickey & Fuller ), is then performed to the 12-month cumulative runoff series from 1961 to 2014 to verify their stationarities. The computed Dickey–Fuller statistic is compared with the corresponding critical value at a certain

μn ¼ a0 þ

I X

ai zi (n)

(5)

i¼1

P-value. For example, when P ¼ 0.05, the critical value of the ADF test result is 3.5 (with the sample size of 50). If the computed Dickey–Fuller statistic is greater than this

σ n ¼ b0 þ

I X

critical value, it indicates that the null hypothesis of the bi zi (n)

(6)

i¼1

where μn and σ n are the location and scale parameters in

ADF test is accepted at the 0.05 signiﬁcance level, that is, the sequence to be analyzed is nonstationary. In this study, we used the package of ‘tseries’ in r to do the ADF test which can output the P-values and the Dickey–Fuller

nonstationary Gamma distribution, a0 and b0 are constant terms, zi (i ¼ 1, 2,…, I ) is the covariate, and i is the number of covariates. Here, six variables are selected as alternative covariates, namely precipitation (P), temperature (T ), relative humidity (H ), wind speed (W ), potential evapotranspiration (E) and antecedent runoff (R). The RS algorithm is used to fit the nonstationary Gamma model in the GAMLSS framework (Rigby & Stasinopoulos , ). After fitting the nonstationary Gamma model, the remaining calculation process of NSRI is the same as SRI, converting the corresponding cumulative probability of nonstationary Gamma model to a standard normal distribution function. To prevent the model overfitting,

Table 2

Drought-level standards and the threshold values

Index value

Category

>2.00

Extreme wet

1.99 to 1.50

Very wet

1.49 to 1.00

Moderate wet

0.99 to 0.00

Near normal

0.00 to 0.99

Mild drought

1.00 to 1.49

Moderate drought

1.50 to 1.99

Severe drought

2.00

Extreme drought

X. Sun et al.

899

Table 3

Assessment of hydrological drought based on nonstationary data

Hydrology Research

51.5

2020

Results of the ADF test for cumulative runoff data in the study area

ZMSK

YLX

Month

P-value

Dickey–Fuller

P-value

Dickey–Fuller

P-value

Dickey–Fuller

0.02

3.84

0.26

2.78

0.07

3.34

0.02

3.84

0.26

2.77

0.07

3.35

0.02

3.81

0.26

2.77

0.07

3.34

0.03

3.75

0.26

2.79

0.07

3.35

0.03

3.70

0.28

2.73

0.11

3.16

0.02

3.95

0.30

2.68

0.17

3.01

0.02

3.86

0.28

2.74

0.21

2.90

< 0.01

4.23

0.24

2.83

0.22

2.89

< 0.01

4.16

0.24

2.84

0.10

3.17

< 0.01

4.15

0.27

2.75

0.09

3.26

0.01

4.14

0.27

2.76

0.08

3.28

0.01

4.14

0.27

2.76

0.08

3.29

statistic values directly. The P-value of this test in this pack-

stations, the results are opposite. Therefore, we conclude

age is obtained through interpolating from the critical value

that the cumulative runoff at these two stations can be con-

table of Banerjee et al. (). If the computed P-value is

sidered as nonstationary series. To model such series, the

larger than a given signiﬁcance level, such as 0.05, the null

NSRI needs to be developed.

hypothesis is accepted. Otherwise, it is rejected. As shown in Table 3, the Dickey–Fuller statistics at the

Correlation test

QL station are less than 3.5, and the corresponding P-values are less than 0.05, which means that, at the 0.05

The selection of a suitable covariate is of great importance in

signiﬁcance level, the cumulative runoff series at the QL

developing a NSRI. To test whether the runoff and the

station exhibit a stationary state. While at ZMSK and YLX

alternative variables over the study area are related to

Table 4

The correlation coefﬁcients between runoff and the six variables in the study area

Pearson

Kendall

Spearman

Station

Variable

Sig.

ZMSK

Precipitation (P) Temperature (T ) Relative humidity (H ) Wind speed (W ) Potential evapotranspiration (E) Antecedent runoff (R)

0.892 0.786 0.759 0.138 0.654 0.706

0.000 0.000 0.000 0.000 0.000 0.000

0.667 0.728 0.611 0.054 0.538 0.615

0.000 0.000 0.000 0.038 0.000 0.000

0.862 0.911 0.807 0.079 0.761 0.825

0.000 0.000 0.000 0.042 0.000 0.000

YLX

Precipitation (P) Temperature (T ) Relative humidity (H ) Wind speed (W ) Potential evapotranspiration (E) Antecedent runoff (R)

0.918 0.788 0.813 0.041 0.654 0.718

0.000 0.000 0.000 0.294 0.000 0.000

0.693 0.727 0.641 0.056 0.532 0.618

0.000 0.000 0.000 0.033 0.000 0.000

0.879 0.911 0.839 0.094 0.759 0.830

0.000 0.000 0.000 0.016 0.000 0.000

900

X. Sun et al.

Assessment of hydrological drought based on nonstationary data

Hydrology Research

51.5

2020

each other, a correlation test is carried out by using Spear-

Comparing the results of different numbers of covari-

man, Pearson and Kendall correlation tests. The test

ates, it can be found that the AIC values for M2 (with two

results are shown in Table 4, in which r indicates the corre-

covariates) show a little lower than those for M1 (with one

lation coefﬁcient, sig. indicates the signiﬁcance level. If sig.

covariate), also lower than those for M3 (with three covari-

<0.05, there is a signiﬁcant correlation between the two

ates) (Table 5). Thus, M2 are selected for our analysis ﬁnally.

variables, and the higher the absolute value of r, the stronger

Speciﬁcally, M21 (precipitation and temperature as covari-

the correlation. Among the six variables, ﬁve of them

ates) and M22 (precipitation and relative humidity as

(precipitation, temperature, relative humidity, potential

covariates) are selected for ZMSK and YLX stations,

evapotranspiration and antecedent runoff) are strongly cor-

respectively, according to the lowest AIC values.

related with runoff with correlation coefﬁcients greater than

To further evaluate the quality of ﬁt, taking August

0.5, and wind speed shows weak correlations. Therefore,

runoff series at the YLX station as an example, the worm

except for wind speed, all variables are selected as covari-

plots of M0 and the best nonstationary M22 are provided

ates to develop the NSRI.

in Figure 2. The worm plot was ﬁrst used by van Buuren

Nonstationary modeling with GAMLSS

residual point is within the acceptable area surrounded by

() to determine the model ﬁt based on whether the the elliptic curve. According to Figure 2(a), it can be Goodness of ﬁtting

observed that although most residual points are within the acceptable area, minority points are at 95% conﬁdence

For better understating the interactions of different factors

level boundary and even outside the acceptable area,

and ﬁnding the main inﬂuencing factors of runoff variation

which means that M0 does not meet the reliability require-

in different seasons, four different types of models are devel-

ments quite well. While, in Figure 2(b), the M22 residual

oped with adding different variables to each nonstationary

points are all in the acceptable area, indicating that this

cases, Model 0 (M0), Model 1 (M1), Model 2 (M2) and

model satisﬁes the reliability and ﬁts well. The performance

Model 3 (M3). M0 is the stationary case with constant

of the nonstationary model with meteorological elements as

Gamma parameters, and the remaining three are all non-

covariates was thus conﬁrmed for August runoff series at the

stationary cases. M1, M2 and M3 take one, two and three

YLX station.

variables as covariates, respectively. Different combinations of covariates are shown in Table 5 for details, with 16 differ-

Percentile estimation

ent models developed in total. M11–M15 represent the nonstationary models with one covariate, M21–M27 rep-

Figure 3, also taking the cumulative runoff series in August at

resent the nonstationary models with two covariates and

the YLX station as an example, shows the ﬁve estimated per-

M31–M34 represent the nonstationary models with three

centile results (5th, 25th, 50th, 75th and 95th) from M0 and

covariates. The AIC values corresponding to each model

M22, as well as the observed runoff series. As shown in

are also displayed in the table. The model with the smallest

Figure 3(a), the quantile grayscale image is horizontally

AIC value is regarded as the best model.

linear, and the dynamic change of runoff series under climate

As shown, the stationary model, M0 has larger AIC

change conditions cannot be described well. Figure 3(b)

values, presenting the poor performance in the ﬁtting.

reﬂects that the percentiles are constantly changing with

Comparatively, the nonstationary models, M1–M3, with

the observed data, and M22 with two meteorological factors

assuming that location and scale parameters in Gamma

as covariates shows better performance in describing the

change with the covariates, perform better in terms of the

nonlinear variation in the runoff data. It also implies that

lower AIC values. This is also largely supported by the

the inﬂuence of meteorological factors to runoff sequence

result of the stationary test. That is, the nonstationary time

cannot be neglected.

series show poor performance with stationary models and better performance with nonstationary models.

Overall, in the context of climate change, the nonstationary

Gamma

distribution

with

considering

X. Sun et al.

901

Table 5

Assessment of hydrological drought based on nonstationary data

Hydrology Research

51.5

2020

Summary of AIC values for stationary (M0) and the nonstationary Gamma models (M1, M2, M3) for runoff data in the study area

Station

Model

ZMSK

M0 M1

YLX

M0 M1

Covariate

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

M0 M11 M12 M13 M14 M15 M21 M22 M23 M24 M25 M26 M27 M31 M32 M33 M34

Null P T H E R PT PH PE PR TH TE HE PTH PTE PHE THE

507 502 512 510 506 494 492 497 503 503 505 507 510 497 503 498 510

507 502 510 516 515 498 494 498 506 503 514 512 510 499 507 501 514

508 507 509 515 517 497 496 507 505 507 512 517 514 506 505 505 513

507 506 514 514 512 494 494 505 507 506 512 512 503 505 507 506 504

507 504 515 506 515 505 494 504 506 504 505 514 501 504 506 505 500

511 507 506 506 503 503 497 507 508 507 507 507 503 507 507 508 506

514 510 514 516 519 503 500 510 510 510 515 519 518 510 510 510 518

513 499 502 523 520 501 499 509 509 507 516 519 519 509 509 509 521

506 501 497 512 512 503 490 500 500 495 510 497 511 500 500 501 508

507 499 508 510 515 500 498 500 498 499 508 513 509 499 497 499 511

509 503 515 516 515 493 493 504 502 503 511 512 514 504 502 503 513

510 504 509 508 512 494 493 500 505 503 496 518 499 500 505 501 494

M0 M11 M12 M13 M14 M15 M21 M22 M23 M24 M25 M26 M27 M31 M32 M33 M34

Null P T H E R PT PH PE PR TH TE HE PTH PTE PHE THE

509 495 511 514 515 491 497 486 501 495 515 513 513 489 502 494 506

509 497 503 508 513 495 498 489 502 497 512 514 513 491 503 496 500

510 498 508 512 510 499 499 497 501 498 512 509 512 498 502 501 511

510 490 510 515 516 498 490 488 492 491 515 515 507 488 493 490 508

508 508 514 510 515 503 508 499 505 508 509 514 501 509 505 506 501

510 503 503 494 508 503 503 494 499 501 506 512 514 506 499 502 513

513 510 509 509 518 501 510 499 512 511 509 517 518 509 512 511 519

515 509 498 520 517 499 509 497 510 507 520 519 519 507 510 510 522

507 490 501 512 511 491 490 495 483 506 511 506 506 495 482 488 501

510 491 507 513 513 498 492 490 494 499 513 511 516 492 495 491 515

511 497 520 511 513 500 497 500 496 500 502 511 499 500 494 499 497

512 499 516 502 512 495 499 494 499 500 500 523 508 498 500 499 504

Note: P, precipitation; T, temperature; H, relative humidity; E, potential evapotranspiration; R, antecedent runoff. The smallest AIC value in each type of model is denoted in bold font.

Figure 2

Worm plots of M0 (a) and M22 (b) for August runoff series at the YLX station.

902

Figure 3

X. Sun et al.

Assessment of hydrological drought based on nonstationary data

Hydrology Research

51.5

2020

Results of M0 (a) and M22 (b) for 12-month aggregated runoff series in August at the YLX station. The light grey region represents the area between the 0.05 and 0.95 quantiles, the dark grey region represents the area between the 0.25 and 0.75 quantiles, and the black line represents the median (0.5 quantile). The black dots are the observed values.

Figure 4

Twelve-month SRI and NSRI from 1961 to 2014 in the study area.

903

X. Sun et al.

Assessment of hydrological drought based on nonstationary data

Hydrology Research

51.5

2020

meteorological factors as covariates tends to more robust in

several signiﬁcant differences between SRI and NSRI esti-

describing the variation behavior exhibited in the runoff

mates can also be observed (Figure 4). For example, from

data in our study case.

August 1984 to May 1985, August 1999 to June 2000, and September 2010 to June 2011, SRI detects normal, while NSRI detects them as mild droughts.

Hydrological drought assessment Drought characteristics Estimation of the SRI and the NSRI The occurrence frequency of different drought grades is calFigure 4 shows the time series of drought index calculated

culated from both the SRI and the NSRI. As shown in

from the observed runoff data at three hydrological

Figure 5, compared with the results of the SRI, moderate

stations. For the QL station, only stationary SRI sequences

and severe droughts occur more frequently, while mild

are calculated from 1961 to 2014. This study deﬁnes a

and extreme droughts are less frequent according to the

drought event as a period of time when the index value

NSRI for the ZMSK station. For the YLX station, the fre-

is continuously negative and reaches a threshold of 1.0

quency of mild drought is higher, and the frequencies of

or lower. In this respect, the dry periods for the QL station

moderate and severe droughts are slightly lower in terms

were from July 1970 to September 1971, from November

of the NSRI.

1972 to August 1974, from June 1978 to June 1980, from

It also can be found that, in Figure 5, the two stations

September 1982 to May 1983, and from May 1992 to

present completely opposite results in the frequency of

August 1992. For ZMSK and YLX stations, both SRI and

drought occurrence, especially at mild and moderate

NSRI sequences are demonstrated in the ﬁgure. Overall,

grades. For the ZMSK station, the frequency of mild

the SRI and NSRI series at both stations provide close

drought decreased and that of moderate drought increased,

drought index in most cases, while also deﬁning several

while for the YLX station, the results are the opposite. This

differences.

can be explained from the locations of the two stations. As

For the ZMSK station, the period of October 1990 to

displayed in Figure 1, the ZMSK station is located

January 1991 is detected as normal state from the SRI, but

upstream of the YLX station. The moderate drought in

moderate drought from the NSRI, the period of June to

the upstream (where the ZMSK station is located) may

July 2011 is detected as wet from the SRI, while drought

merely cause the weaker drought (mild drought) in the

from the NSRI. For the YLX station, both SRI and NSRI

downstream (where the YLX station is located), since

recorded extreme drought events from August 1973 to July

there are other tributaries ﬂowing into the main stream

1974. The results obtained are consistent with historical

and then alleviating the drought grade in the downstream

records and related literature (Yang et al. ). However,

to some extent.

Figure 5

Drought occurrence frequency based on SRI and NSRI in the study area.

X. Sun et al.

904

Table 6

Assessment of hydrological drought based on nonstationary data

Hydrology Research

51.5

2020

Drought characteristics from the SRI and the NSRI in the study area

SRI Station

Number

Peak

Longest duration (months)

Maximum severity

2.92

59.83

ZMSK

2.29

26.69

YLX

2.63

43.73

NSRI Station

Number

Peak

Longest duration (months)

Maximum severity

ZMSK

2.42

33.62

YLX

2.47

37.66

The run-length theory can be used to extract drought

Discussion

charateristics, including the peak, duration and severity (Yevjevich ). Drought peak is deﬁned as the minimum

Number of covariates

of the drought index during a drought event. Drought duration refers to the duration of drought from the start to the

Many meteorological factors show strong effects on

end of a drought event. Drought severity refers to the cumu-

droughts under the global warming in recent decades (e.g.,

lative sum of the differences between the drought index and

Núñez et al. ; Zarch et al. ), so many of these factors

the threshold during the drought (Yevjevich ; Gu et al.

are taken into account in the study of hydrological droughts.

). A threshold of 1.0 was set to identify the occurrence

Here, six variables are considered, and three nonstationary

of drought events in this study. Drought peak, duration and

models are developed. When only one variable is con-

severity were then extracted from the drought events accord-

sidered, M1, in which the variable of precipitation is

ing to the run-length theory. As shown in Table 6, the QL

taken as a covariate, shows the best for 9 of 24 series, and

station identiﬁes 11 drought events, of which the drought

the series are concentrated basically in April, May, August,

peak reaches 2.92, the longest drought duration lasts for

September and October (Table 5). It indicates that precipi-

25 months and the maximum drought severity is 59.83.

tation is the main factor affecting runoff changes during

The ZMSK station identiﬁes more drought events (18

these months. It is understandable since these months are

events) based on the NSRI, with higher drought peak

usually with more precipitation in the study area. M15,

( 2.42) and higher drought severity ( 33.62), compared

taking the variable of antecedent runoff as a covariate,

with the results of SRI (12 events, 2.29, 26.69). The

shows the best for 13 of 24 series, and most of them are con-

YLX station also identiﬁes more drought events (19

centrated from November to about next April. This ﬁnding

events) based on the NSRI, while the drought peak

keeps consistent with the river recharge resources in differ-

( 2.47) and the maximum drought severity ( 37.66) are

ent seasons. In rainy seasons (e.g., summer), precipitation

lower than those from the SRI ( 2.63 and 43.73). The

contributes the runoff most, while in dry seasons (e.g.,

longest drought durations are quite close from both the

winter and spring), the river is mostly recharged by glacial

NSRI and the SRI for this station.

and snow melting water, permafrost melting water and

Table 7

The AIC values for nonstationary models with time as a covariate for runoff data in the study area

Station

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

ZMSK

497

498

497

496

497

505

498

490

499

494

496

YLX

495

496

498

497

495

501

498

485

492

494

495

905

X. Sun et al.

Assessment of hydrological drought based on nonstationary data

Hydrology Research

51.5

2020

groundwater over the study area (Li et al. ; Wang et al.

M1, seven in M2 and four in M3). It is observed that in

). Pei et al. () and Wang et al. () concluded that

Figure 6(a) and 6(c), in the case of taking the variable of pre-

about 75% of runoff are recharged by precipitation in rainy

cipitation as the covariate, no matter what kinds of

seasons and more than 75% are recharged by snow melting

nonstationary models are (M21, M22, M23 and M24 with

in dry seasons in the Heihe River basin.

two covariates or M31, M32 and M33 with three covariates),

In the case of taking two variables as covariates (M2

the derived drought index are quite close with those from

models), all the corresponding AIC values show slightly

M11 (in which only the variable of precipitation is selected

lower than those for M1 models (Table 5). It indicates that

as the covariate). That is, as long as the combinations of pre-

two variables are better in describing the nonstationary

cipitation and other variables are taken as covariates for the

characteristics of the runoff data. In the case of taking

nonstationary model, the derived drought index curves are

three variables as covariates (M3 models), however, the

quite close. It indicates that the hydrological drought in

AIC values have gone up instead. This may be the result of

this study is mainly affected by precipitation, and the contri-

over-ﬁtting caused by the over-complexity of the model.

bution of precipitation to hydrological droughts is great over

Generally speaking, when the number of covariates

the study area. This conclusion is also conﬁrmed by Li et al.

increases, the complexity of the model increases, the likeli-

() and Wang et al. (, b). When other variables,

hood function also increases, it leads to AIC value

rather than precipitation, are considered as covariates

decreases. Whereas when the number is too large, the

(M12, M13, M14, M15, M25, M26, M27 and M34), the

growth rate of likelihood function slows down, which thus

derived drought index curves ﬂuctuate a great deal, and

results in the increase of AIC (Akaike ). In addition,

large differences are found between them (Figure 6(b) and

as stated by Agilan & Umamahesh (), the increase of

6(d)). This indicates that the results of drought assessment

covariates would bring greater uncertainty in the ﬁnal

differ greatly without considering precipitation. All these

assessment. Thus, M2, with two variables as covariates,

mean that, although other variables have certain effects to

are ﬁnally selected for our analysis. Note that many studies also included the time as a cov-

hydrological droughts, precipitation is still the major inﬂuence factor in hydrological drought in our case.

ariate in a nonstationary analysis (Russo et al. ; Wang et al. a; Javad & Somayeh ), although some studies

Performance of the SRI and the NSRI

have proved that time is not suitable in their cases due to its linear monotonic trends (Li et al. ). It is considered

According to Figure 4, the SRI and the NSRI at ZMSK and

as an additional covariate for comparison to discuss the

YLX stations display the close drought curves in most cases,

nonstationary runoff process in this study. As illustrated in

while there are still several differences. These differences are

Table 7, the corresponding AIC value is moderate, slightly

attempted to be explained by the meteorological variables.

higher than those from the selected models (M21 for the

According to Jiang et al. (), the hydrological drought

ZMSK station and M22 for the YLX station) and lower

in the upper reach of Heihe River basin lag behind the

than those from the stationary models (M0). Besides, since

meteorological drought by about 4 months. Therefore, we

using time as an explanatory variable has no physical mean-

deduce that the hydrological drought situation of a certain

ing, it can only describe the general trend of a hydrological

period is possibly related with the precipitation that slides

sequence over time (Vu & Mishra ). Thus, the time is

forward for 4 months.

not considered as a covariate in our nonstationary analysis.

For the ZMSK station, the period of October 1990 to January 1991 is detected as normal state from the SRI,

Effects of covariates to drought index

while moderate drought from the NSRI. It is found that the precipitation sliding forward for 4 months (from June

To discuss the effects of different covariates to the drought

to September in 1990) was 328.7 mm, lower than the

index, we compare the drought index calculated from

multi-year mean value of 333.7 mm for the same period.

three types and 16 nonstationary models (ﬁve models in

Less precipitation is more likely to cause drought, and

906

Figure 6

X. Sun et al.

Assessment of hydrological drought based on nonstationary data

Hydrology Research

51.5

2020

Differences in the NSRI with different covariates in the study area (the models in the legend are the same as those in Table 5).

thus the conclusion drawn from the NSRI (drought) seems

March 2011 (6.4 mm) is less than the multi-year mean

to be more reliable. For the period of June to July 2011, it

value (11.0 mm), and the mean temperature (9.04 C)

is found that the observed precipitation from February to

shows higher than the multi-year mean value (8.28 C).

907

X. Sun et al.

Assessment of hydrological drought based on nonstationary data

Hydrology Research

51.5

2020

Owing to the less precipitation and the higher temperature

of the SRI, but corresponds to mild drought (May 1964),

during this period, it is concluded that drought is more

normal state (June 1975) and moderate drought (December

likely to occur, and thus the detection from the NSRI

2004) in terms of the NSRI. A monthly runoff of 183 mm is

(drought) is more reasonable than that from the SRI (wet).

classiﬁed as normal state from the SRI, while classiﬁed as

For the YLX station, the drought conditions for periods

moderate wet (February 1982), normal (August 1999) and

of August 1984 to May 1985, August 1999 to June 2000 and

mild drought (September 2010) states from the NSRI.

September 2010 to June 2011 are different from the SRI and

In a changing environment, hydrological drought can no

the NSRI. It is found that the precipitation sliding forward

longer be evaluated solely on runoff. Zhang et al. ()

for 4 months in these three periods (336.8, 327.8 and

investigaed the impacts of both climate variations and

333.6 mm) are less than their multi-year mean value for

human activities to hydrological droughts in the middle

the same period (364.9, 358.3 and 345.5 mm), and the aver-

reach of Yangtze River in China. Wang et al. (b) incor-

age relative humidity of these three periods (54.50, 53.48

porated the climate-driven and human-induced variables

and 51.76%) are also lower than the multi-year mean

as covariates in the nonstationary model to indicate the non-

value (55.47, 54.27 and 52.76%). Therefore, the occurrence

stationary features of runoff series. In this study, the

of drought (drawn from the NSRI), rather than the normal

determination of NSRI depends not only on runoff data

state (drawn from SRI), is regarded as more reasonable.

but also on the key factors affecting runoff. Variables includ-

Another observation is that the same runoff value corre-

ing precipitation and temperature are considered as

sponds the same drought classiﬁcation based on the SRI,

covariates in characterizing the nonstationary features in

which is because its calculation solely depends on runoff

the runoff data, and thus drought state detected by the

data (Shukla & Wood ), and the drought state detected

NSRI varies with the inﬂuence of covariates.

from the SRI keeps stable. However, this conclusion is not

We also calculated the NSRI with time as a covariate

applicable to the nonstationary model. The NSRI leads to

(denoted as the TSRI here) for comparison (Figure 7). It is

differences in drought classiﬁcation with the same runoff

observed that TSRI differs a great deal with SRI especially

value. For example, a monthly runoff of 150 mm for the

for the ﬁrst and last several years. This is because the location

YLX station corresponds to a mild drought state in terms

parameter is constant and close to its mean value in the SRI,

Figure 7

Twelve-month SRI and TSRI from 1961 to 2014 in the study area.

908

X. Sun et al.

Assessment of hydrological drought based on nonstationary data

Hydrology Research

51.5

2020

while it is time-dependent in the TSRI. This condition leads to

In general, the current study solves the problems of

more differences between TSRI and SRI for the early and last

hydrological drought assessment based on nonstationary

years of the record period and less differences for the middle

runoff data. It is proved that the proposed new drought

part of the period (Javad & Somayeh ). Furthermore, the

index, NSRI, is capable of providing more reliable results

linear monotonic trend of time also makes it not as appropri-

for drought assessment over the study area. Revelant ﬁnd-

ate as other meteorological variables in characterizing the

ings are also very helpful in developing strategies for

nonstationary processes of runoff series (Li et al. ).

coping with local droughts and water resource risk manage-

These results support that, compared with the tra-

ment. It should be noted that, although the ﬁndings here are

ditional SRI, the new developed NSRI, with considering

in the context of NSRI at 12-month scale and limited

meteorological variables as covariates, is proved to be

stations, the framework and the methods in this study

more robust and more applicable in this drought assessment.

could be extended over further scales and regions. This paper mainly considers the relation of several meteorological elements to hydrological droughts. In fact, hydrological droughts are also induced by factors like the

CONCLUSIONS

underlying surface of the basin, vegetation coverage and human activities (such as reservoir opening and irrigation)

Appropriate drought indices are often important tools for

(Shukla & Wood ; Fang et al. b; Zhang et al.

regional drought assessments. In the context of changing

; Wang et al. a, b, c). The use of appropriate

environment, the traditional SRI which is deﬁned based

quantitative factors to characterize human activities will be

on a stationary Gamma distribution is no longer suitable

considered in the future research for making the nonstation-

due to the nonstationary characteristics of runoff series

ary drought index more reliable under the complex

(Villarini et al. ; Russo et al. ). Thus, the NSRI is

conditions. Additionally, to identify the hydrological

developed in this study within the GAMLSS framework.

drought propagation and its key inﬂuencing factors is also

The proposed new drought index considers climate change

a hot topic and can be included in the future studies.

and incorporates six variables into calculation, accounting for the nonstationary features of runoff data. As a quantitative indicator, just like the SRI, the NSRI can assess drought

ACKNOWLEDGEMENTS

events at different time scales (e.g., 3, 6, 9 and 12 months) and allow comparison of climatic conditions at different

This study is supported by the Fundamental Research Funds

locations. The 12-month time-scale NSRI is applied to the

for the Central Universities (No. 39435832015028).

upper reach of the Heihe River basin. The conclusions can be summarized as follows. For the study area, (1) the ﬁtting results of the nonstationary models

DATA AVAILABILITY STATEMENT

to those nonstationary runoff data (ZMSK and YLX stations) are much better than the stationary models according to AIC values; (2) for the nonstationary models with one covariate, precipitation shows better in the ﬁtting as a covariate in rainy seasons, and antecedent runoff shows better in dry seasons, which is highly related with the recharge sources in

The meteorological dataset is available from the China Meteorological Data Network (http://data.cma.cn/). The hydrological dataset is available from the Hydrological Yearbook of the People’s Republic of China.

different seasons; the nonstationary models with two covariates perform better than those with one or three covariates;

REFERENCES

(3) the NSRI identiﬁes more drought events than the SRI, and precipitation is the major factor affecting the hydrological drought conditions over the study area.

Aa, Y., Wang, G., Liu, T., Xue, B. & Kuczera, G.  Spatial variation of correlations between vertical soil water and

909

X. Sun et al.

Assessment of hydrological drought based on nonstationary data

evapotranspiration and their controlling factors in a semiarid region. Journal of Hydrology 574, 53–63. Agilan, V. & Umamahesh, N. V.  Covariate and parameter uncertainty in non-stationary rainfall IDF curve. International Journal of Climatology: A Journal of the Royal Meteorological Society 38 (1), 365–383. Akaike, H.  A new look at the statistical model identification. IEEE Trans Autom Control 19 (6), 716–723. Akantziliotou, K., Rigby, R. A. & Stasinopoulos, D. M.  The R implementation of generalized additive models for location, scale and shape. In: Statistical Modelling in Society: Proceedings of the 17th International Workshop on Statistical Modelling (M. Stasinopoulos & G. Touloumi, eds). Chania, pp. 75–83. Allen, R. G., Pereira, L. S., Raes, D. & Smith, M.  Crop Evapotranspiration Guidelines for Computing Crop Water Requirements. Food and Agriculture Organization of the United Nations Rome, Italy. Banerjee, A., Dolado, J. J., Galbraith, J. W. & Hendry, D. F.  Cointegration, Error Correction, and the Econometric Analysis of Non-Stationary Data. Oxford University Press, Oxford. Bloomfield, J. P. & Marchant, B. P.  Analysis of groundwater drought building on the standardised precipitation index approach. Hydrology and Earth System Sciences 17 (12), 4769–4787. Cheng, M.  Impact and Assessment of Drought Effects on Natural Ecosystem: A Case Study of Hei River Basin. Nanjing University of Information Science and Technology, Nanjing, China. Dickey, D. A. & Fuller, W. A.  Distribution of the estimators for autoregressive time series with a unit root. Journal of the American Statistical Association 74 (366), 427. Ding, Y. J., Ye, B. S. & Liu, S. Y.  Effect of climatic factors on streamflow in the alpine catchment of the qilian mountains. Acta Geographica Sinica 54 (05), 431–437. Fang, Q., Wang, G., Liu, T. & Xue, B. A. Y. a Controls of carbon flux in a semi-arid grassland ecosystem experiencing wetland loss: vegetation patterns and environmental variables. Agricultural and Forest Meteorology 259, 196–210. Fang, Q., Wang, G., Xue, B., Liu, T. & Kiem, A. b How and to what extent does precipitation on multi-temporal scales and soil moisture at different depths determine carbon flux responses in a water-limited grassland ecosystem? The Science Of The Total Environment 635, 1255–1266. Gao, X. C. & Zhao, L.  The prediction and analysis of population carrying capacity in Heihe basin. Northwest Population Journal 31 (03), 120–123 þ 129. Gu, J. S., Li, Q. F., Niu, M. Y., Chen, Q. H., He, P. F., Zhou, Z. M., Zeng, T. S., Du, Y., Song, Y. & Han, X. Y.  An analysis of meteorological drought characteristics in the Upper Reaches of Huaihe River based on multidimensional copula function. China Rural Water and Hydropower 61 (08), 83–87 þ 92. Guttman, N. B.  Accepting the standardized precipitation index: a calculation algorithm. Journal of the American Water Resources Association 35 (2), 311.

Hydrology Research

51.5

2020

Han, D., Wang, G., Xue, B., Liu, T. A. Y. & Xu, X. a Evaluation of semiarid grassland degradation in North China from multiple perspectives. Ecological Engineering 112, 41–50. Han, D., Wang, G. & Liu, T. b Hydroclimatic response of evapotranspiration partitioning to prolonged droughts in semiarid grassland. Journal of Hydrology 563, 56. Hastie, T. J. & Tibshirani, R. J.  Generalized Additive Models. Chapman and Hall, London. Javad, B. & Somayeh, H.  A non-stationary reconnaissance drought index (NRDI) for drought monitoring in a changing climate. Water Resources Management 32 (8), 2611–2624. Jiang, Y. W., Zhang, X. F., Yang, L. X. & He, C. S.  Analysis and comparison of spatial and temporal patterns of meteorological and hydrological drought indices in the Upper Reach of the Heihe River Watershed. Resources Science 36 (09), 1842–1851. Jiang, S. H., Wang, M. H., Ren, L. L., Xu, C. Y., Yuan, F., Liu, Y. & Shen, H. R.  A framework for quantifying the impacts of climate change and human activities on hydrological drought in a semiarid basin of Northern China. Hydrological Processes 33 (7), 1075–1088. Li, L., Wang, Z. N. & Wang, Q. C.  Influence of climatic change on flow over the Upper Reaches of Heihe River. Scientia Geographica Sinica 26 (01), 40–46. Li, J. Z., Wang, Y. X., Li, S. F. & Hu, R.  A nonstationary standardized precipitation index incorporating climate indices as covariates. Journal of Geophysical Research 120 (23), 12082–12095. Ling, P. Q.  The Coordinated Development Reseach of Population, Resource, Environment and Economy of Heihe River Basin. LanZhou University, Lanzhou, China. Liu, S. Y., Huang, S. Z., Xie, Y. Y., Wang, H., Leng, G., Huang, Q., Wei, X. & Wang, L.  Identification of the non-stationarity of floods: changing patterns, causes, and implications. Water Resources Management 33 (3), 939–953. Luciano, T., Michele, L. & Ignacio, L. M.  Investigation of scaling properties in monthly streamflow and Standardized Streamflow Index (SSI) time series in the Ebro basin (Spain). Physica A: Statistical Mechanics and its Applications 391 (4), 1662–1678. McKee, T. B., Doesken, N. J. & Kleist, J.  The relationship of drought frequency and duration to time scales preprints. In 8th Conference on Applied Climatology. pp. 179–184. Nalbantis, I. & Tsakiris, G.  Assessment of hydrological drought revisited. Water Resources Management 23 (5), 881–897. Nelder, J. A. & Wedderburn, R. W. M.  Generalized linear models. Journal of the Royal Statistical Society A 135, 370–384. Núñez, J., Rivera, D., Oyarzún, R. & Arumí, J. L.  On the use of standardized drought indices under decadal climate variability: critical assessment and drought policy implications. Journal of Hydrology 517, 458–470. Pei, H., Fang, S. F., Liu, Z. H. & Qin, Z. H.  Design and application of distributed snowmelt-runoff model. Resources Science 30 (03), 454–459.

910

X. Sun et al.

Assessment of hydrological drought based on nonstationary data

Rigby, R. A. & Stasinopoulos, D. M.  The GAMLSS project: a flexible approach to statistical modelling. In: New Trends in Statistical Modelling: Proceedings of the 16th International Workshop on Statistical Modelling (B. Klein & L. Korsholm, eds). Odense, pp. 249–256. Rigby, R. A. & Stasinopoulos, D. M.  Generalized additive models for location, scale and shape. Applied Statistics 54, 507–554. Russo, S., Dosio, A., Sterl, A., Barbosa, P. & Vogt, J.  Projection of occurrence of extreme dry-wet years and seasons in Europe with stationary and nonstationary standardized precipitation indices. Journal of Geophysical Research [Atmospheres] 118, 7628–7639. Shukla, S. & Wood, A. W.  Use of a standardized runoff index for characterizing hydrologic drought. Geophysical Research Letters 35, L02405. Stasinopoulos, D. M. & Rigby, R. A.  Generalized additive models for location scale and shape (GAMLSS) in R. Journal of Statistical Software 23 (7), 1–46. Tabari, H., Nikbakht, J. & Hosseinzadeh, T. P.  Hydrological drought assessment in Northwestern Iran based on streamflow drought index (SDI). Water Resources Management 27 (1), 137–151. van Buuren, Y.  Basic back: whether or not you’re among the 60 to 80% of us who experience severe and even debilitating back pain, you can have a healthy and pain-free back forever. Canadian Living 26 (4), 79. Vicente-Serrano, S. M., López-Moreno, J. I., Beguería, S., LorenzoLacruz, J., Azorin-Molina, C. & Morán-Tejeda, E.  Accurate computation of a streamflowdrought index. Journal of Hydrologic Engineering 17 (2), 318–332. Villarini, G., Serinaldi, F., Smith, J. A. & Krajewski, W. F.  On the stationarity of annual flood peaks in the continental United States during the 20th century. Water Resources Research 45 (8), W08417. Vu, T. M. & Mishra, A. K.  Nonstationary frequency analysis of the recent extreme precipitation events in the United States. Journal of Hydrology 575, 999–1010. Wang, S. L., Liu, X. D., Jin, M., Zhang, X. L., Che, Z. X. & Wang, R. X.  The impact of temperature and precipitation on the streamflow in the middle part of the Qilian Mountains, Northwestern China. Journal of Arid Land Resources and Environment 25 (01), 162–165. Wang, G., Liu, S. & Liu, T. a Modelling above-ground biomass based on vegetation indexes: a modified approach for biomass estimation in semi-arid grasslands. International Journal of Remote Sensing 40 (10), 3835–3854.

Hydrology Research

51.5

2020

Wang, Y. X., Duan, L. M., Liu, T. X., Li, J. Z. & Feng, P. b A non-stationary standardized streamflow index for hydrological drought using climate and human-induced indices as covariates. Science of The Total Environment 699, 134278. Wang, G., Li, J., Sun, W., Xue, B. A. Y. & Liu, T. c Non-point source pollution risks in a drinking water protection zone based on remote sensing data embedded within a nutrient budget model. Water Research 157, 238–246. Wang, Y. X., Li, J. Z., Feng, P. & Hu, R. a A time-dependent drought index for non-stationary precipitation series. Water Resources Management 29 (15), 5631–5647. Wang, Y. H., Yang, D. W., Lei, H. M. & Yang, H. B. b Impact of cryosphere hydrological processes on the river runoff in the upper reaches of Heihe River. Journal of Hydraulic Engineering 46 (09), 1064–1071. Xiong, L. H. & Guo, S. L.  Trend test and change-point detection for the annual discharge series of the Yangtze River at the Yichang hydrological station. Hydrological Sciences Journal 49 (1), 99–112. Yang, L. X., Zeng, S. X., Jiang, Y. W., Gu, J. & He, C. S.  Comparison of drought characteristics in the upstream of the Heihe River basin based on palmer drought severity index and standard precipitation index. Research of Soil and Water Conservation 24 (02), 132–136. Yevjevich, V.  An Objective Approach to Definitions and Investigations of Continental Hydrologic Droughts, Vol. 6. Colorado State University, pp. 23–30. Zarch, M. A. A., Sivakumar, B. & Sharma, A.  Droughts in a warming climate: a global assessment of standardized precipitation index (SPI) and reconnaissance drought index (RDI). Journal of Hydrology 526, 183–195. Zhang, X. Q., Sun, Y., Zheng, D. & Mao, W. Y.  Regional response of temperature change in the arid regions of China to global warming. Arid Zone Research 27 (4), 592–599. Zhang, D., Zhang, Q., Qiu, J., Bai, P., Liang, K. & Li, X.  Intensification of hydrological drought due to human activity in the middle reaches of the Yangtze River, China. Science of the Total Environment 637, 1432–1442. Zheng, J. T., Chen, F. L., Zhang, X. H., Long, A. H. & Liao, H.  Analysis of design annual runoff of Manas River based on GAMLSS model. Climate Change Research 14 (03), 257–265. Zhou, H. H., Wang, Y. Q., Fang, G. H., Ye, Z. X. & Li, W. H.  Application of standardized runoff index on hydrological drought characteristics identification in Aksu River. Journal of Water Resources and Water Engineering 30 (2), 6–11 þ 18.

First received 13 March 2020; accepted in revised form 10 August 2020. Available online 2 September 2020

911

51.5

2020

Spatiotemporal distributions and ecological risk assessment of pharmaceuticals and personal care products in groundwater in North China Jin Wu, Jingchao Liu, Zenghui Pan, Boxin Wang and Dasheng Zhang

ABSTRACT The contamination of surface water by pharmaceuticals and personal care products (PPCPs) has attracted widespread attention, but data regarding their impacts on groundwater (GW) are sparse. In river–GW interaction areas, rivers are likely an important source of PPCPs in aquifers, especially rivers impacted by sewage treatment plant effluent. Understanding the characterization, transport, and risk is valuable for the effective protection of vital aquatic ecosystem services, environmental health, and drinking water supplies. To attain this objective, statistics with spatial analysis and ecological risk were used to assess the effects of artificial recharge (AR) engineering on 16 PPCPs in aquifers in North China. The results indicated that 15 PPCPs were detected in unconfined and confined aquifers, with a few PPCPs being detected up to 1,000 ng/L. The most frequently detected

Jin Wu Jingchao Liu Zenghui Pan Boxin Wang Dasheng Zhang (corresponding author) Hebei Institute of Water Science, Shijiazhuang 050051, China E-mail: skyzhangdasheng@126.com Jin Wu Jingchao Liu College of Architecture and Civil Engineering, Beijing University of Technology, Beijing 100124, China

PPCPs were sulfisoxazole, sulfachloropyridazine, sulfamerazine, sulfamethazine, sulfamethoxazole, and ibuprofen. In addition, the spatial and seasonal variations in most PPCPs were significant. Furthermore, the maximum concentrations were compared to the predicted no-effect concentrations to evaluate the ecological risk, and four PPCPs were found to be of medium or high ecological risk. This study highlights that AR engineering has a significant ecological effect on GW. Key words

| ecological risk, groundwater, PPCPs, river–groundwater interaction, spatiotemporal distributions

INTRODUCTION China is one of the largest producers and consumers of

46% of the antibiotics were ultimately released into rivers

pharmaceuticals and personal care products (PPCPs) (Liu

through sewage efﬂuent (Zhang et al. ). PPCPs have

& Wong ; Hanna et al. ). Due to widespread

been detected in surface water (SW), groundwater (GW),

consumption and inefﬁcient treatment, using PPCPs has

sediment, soil, and vegetables around China, some of

put pressure on the environment, resulting in drinking

which are considered environmentally pseudopersistent

water resources in China being threatened (Thomas et al.

with continuous discharge (Kostich et al. ; Peng et al.

; Wang et al. a). Approximately 92,700 tons of anti-

; Wang et al. ; Zhang et al. ). Signiﬁcant atten-

biotics were consumed in 2013 in China, and approximately

tion is being paid to the presence of PPCP residues in the environment due to their potential adverse effects on ecosys-

This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying,

tems and human health. Speciﬁcally, studies have shown

adaptation and redistribution, provided the original work is properly cited

that mixtures of PPCPs exhibit greater effects than those of

(http://creativecommons.org/licenses/by/4.0/).

the individual compounds (Cleuvers ; Jiang et al. ;

doi: 10.2166/nh.2020.001

912

J. Wu et al.

Pharmaceuticals and personal care products in groundwater in North China

Hydrology Research

51.5

2020

Brandt et al. ). It is anticipated that environmental

the inﬁltration of sewage by artiﬁcial aquifer recharge and

legislation will be widened to cover a range of PPCPs.

bank ﬁltration is expected to be the main source of PPCPs

However, the knowledge of their fate within water cycles

in unconﬁned aquifers (Teijon et al. ; Yang et al. ).

is still insufﬁcient.

However, such inﬁltration processes are not expected to

GW maintains ﬂows and levels in rivers and lakes, is

impact unconﬁned aquifers and deteriorate GW quality.

essential for the health of GW-dependent ecosystems, and

A few works have investigated the occurrence of PPCPs

is regarded as the most important source of drinking water

in conﬁned aquifers, examining the selected PPCPs

in many parts of the world (Barnes et al. ; Lapworth

(diethyltoluamide, crotamiton, ethenzamide, propyphena-

et al. ). The riparian zone adjacent to rivers is often an

zone, carbamazepine, and caffeine) in conﬁned aquifers

important area for residents in China, where the interactions

(up to 500 m in depth) in Tokyo due to leakage from decre-

between rivers and GW are active (Fang et al. ; Aa et al.

pit sewer networks (Kuroda et al. ). Generally, the GW

). The subsurface geochemistry in the riparian zone is

in unconﬁned aquifers is used as a drinking water source

changed greatly due to either the leakage of river water to

in North China and is considered a key factor for the

recharge the underlying GW or vice versa (Yang et al.

steady development of the national economy (Bo et al.

; Zhang et al. ). Increasing evidence shows that

; Li et al. ). Therefore, in developing management

rivers are one of the main sources of PPCPs in riverside

strategies to control GW pollution by PPCPs, the contami-

GW (Kuroda et al. ; Yang et al. a). Artiﬁcial recharge

nation and ecological risk of PPCPs in unconﬁned and

(AR) is considered a promising method to alleviate GW

conﬁned aquifers both need to be assessed.

depletion by using reclaimed water, particularly in arid

Considering the issues raised above, the objectives of

areas (Grünheid et al. ; Li et al. ; Singh et al. ;

this study were to investigate the occurrence of PPCPs

Zheng et al. ). The PPCPs in rivers can enter GW

in different types of aquifers, including unconﬁned and con-

aquifers through the process of GW–SW exchange (i.e., via

ﬁned aquifers, and to characterize the ecological risk for the

bank ﬁltration or AR) (Buerge et al. ; Dougherty et al.

development of resistance to PPCPs in the GW as well as to

; Petrie et al. ). For example, Einsiedl et al. ()

optimize the management of GW recharging.

studied the transport of pharmaceuticals in karst GW systems affected by wastewater treatment. Díaz-Cruz & Barceló () reviewed PPCPs in different source waters

MATERIALS AND METHODS

used for artiﬁcial aquifer recharge purposes. In view of risk management and control, the ecological impact of AR

Study area

engineering on GW quality needs more attention since previous studies regarding ecological effects are sparse.

The studied river–GW interaction (RGI) area is located in

The ecological effect of PPCPs in GW, as well as their

North China (Figure 1). The area is characterized by hot,

relationship with environmental factors, is as yet poorly

humid summers and generally cold, windy, dry winters.

understood compared to those of SW. Previously, the detec-

Its annual temperature is ∼11.5 C, and the average precipi-

tion frequency and the ecological risk index have been

tation is ∼600 mm/a. AR engineering is using reclaimed

widely employed to describe the spatiotemporal character-

water which is produced by a membrane bioreactor with

istics of PPCPs in rivers and lakes (Sun et al. ; Yang

wastewater treatment plant efﬂuent. Then the reclaimed

et al. b; Hanna et al. ; Bexfield et al. ). Since

water, about 38,000,000 m3/a, is introduced to the river.

more PPCPs are likely to have GW threshold values in the

Besides, some water containing treated and untreated

coming decades, more data related to occurrence and

municipal water from nearby factories and villages also

ecological effects are required. Moreover, studies seldom

ﬂows into the river. The GW near the released reclaimed

discuss PPCPs in unconﬁned and conﬁned aquifers simul-

water area is only recharged by reclaimed water in the wet

taneously recharged by reclaimed water. Among these

season. The mean concentrations of NH4þ, total nitrogen,

GW–SW exchange areas with sewage efﬂuent discharge,

and total phosphorus in reclaimed water were 4.30, 10.64,

913

Figure 1

J. Wu et al.

Pharmaceuticals and personal care products in groundwater in North China

Hydrology Research

51.5

2020

GW sampling locations (a) and hydrogeological cross-sections of the study area (b).

and 0.78 mg/L, which indicate bad water quality. While

study area are widely distributed in plains and mountain

other water quality parameters ranged from medial to

valleys. The characteristics of the sedimentary layers are:

good quality according to the data analysis of 43 holes col-

north thin south thick and east thin west thick. The hydro-

lected in 2013 (Figure 1). The quaternary sediments in the

geological characteristics in the study area are that the

914

J. Wu et al.

Pharmaceuticals and personal care products in groundwater in North China

Hydrology Research

51.5

2020

depth of buried bedrock increases from north to south, the

control procedures were followed, usually including solvent

thickness of the quaternary strata is between 50 and

blank, procedure blank, and independent check standard.

300 m, and gradually increases from northeast to southwest.

Method detection limits (limits of detection, LODs) and

The lithology changes from coarse particles to ﬁne particles,

quantiﬁcation limits (limits of quantiﬁcation, LOQs) were gen-

and the layers change from a single layer to multiple layers.

erally determined as the minimum detectable amount of an

The quaternary strata mainly include a sand pebble layer, a

analyte with a signal-to-noise ratio. Recoveries obtained by

sand gravel layer, and a silty clay layer.

spiking the analytes into GW ranged from 65 to 128%. The

Two subzones of the study area were identiﬁed, includ-

LOQs

were

0.2–6 ng/L

for

pharmaceuticals

GW.

ing the northern part (N zone) and southern part (S zone),

The concentrations of 16 PPCPs were determined for

based on the hydrogeological conditions. The N zone of

all GW samples. Sixteen PPCPs were classed and abbreviated

the study area was dominated by gravel and sand with

good permeability, while the S zone consisted of silty clay

sulfamonomethoxine (SMM), sulfathiazole (STZ), N4-acetyl-

with poor permeability. The lithology of the aquifer changed

sulfamethoxazol

follows:

enroﬂoxacin

(EFX),

(N4AcSMX),

erythromycin sulﬁsoxazole

(ETM), (SFS),

from sandy gravel to ﬁne sand from north to south. The shal-

sulfachloropyridazine (SCP), sulfamerazine (SMR), sulfa-

low aquifer in the N zone is an unconﬁned aquifer (UA-N).

methazine (SMZ), sulfamethoxazole (SMX), trimethoprim

However, there are multiple layers of aquifers in the S

(TMP), caffeine (CAF), chloramphenicol (CAP), ibuprofen

zone with an unconfined aquifer (UA-S), first confined

(IBU), triclosan (TCS), and diﬂoxacin (DIF). The main phys-

aquifer (FCA-S), and second conﬁned aquifer (SCA-S). The

icochemical properties of the 16 target PPCPs are shown in

thicknesses of UA-N and UA-S ranged between 0–30 and

the Supplementary material, Table S1.

0–80 m, respectively. The thicknesses of FCA-S and SCA-S ranged between 30–50 and 50–80 m, respectively.

Leaching potential assessment

Sampling and analysis

Leaching potential assessment models were adopted to

GW samples were collected from 47 monitoring wells of riverside sections by the QED low-ﬂow sampling equipment (Sample Pro™ sampling pump) in May 2016 (summer, wet season) and December 2016 (winter, dry season) (Figure 1). Twenty-four samples were collected in the N zone to represent the NS aquifer with the sampling depths at 50 m, and 20 samples were collected in the S zone to represent the UA-S, FCA-S, and SCA-S aquifers with the sampling depths at 30, 50, and 80 m, respectively. Detailed information about

the analysis procedure was

provided

elsewhere (Sui et al. ; Yang et al. b; Chen et al. ). In short, GW samples were pumped into 2 L glass bottles using a stainless-steel submersible pump. All samples were kept in precleaned containers at a cool temperature

assess the leaching potential of the selected PPCPs in the vadose zone. The model provides a quantitative value for representing the leaching potential. The model is described by Equation (1) by considering both the mobility and the persistence of chemicals. GUS ¼ log t1=2 (4 log KOC )

(1)

where GUS is the groundwater ubiquity score, KOC is the organic carbon partition coefﬁcient, and t1/2 is the degradation half-life in the soil (days). The adjusted criteria were as follows: low leaching potential (GUS 1.8), moderate leaching potential (1.8 < GUS 2.8), and high leaching potential (2.8 GUS).

and then immediately transported to the laboratory for treatment. In the laboratory, the water samples were commonly

Ecological risk assessment

concentrated by preconditioned solid-phase extraction. The target antibiotics were subsequently analyzed using high/

The ecological risk assessment has been used to quantify the

ultra/ultrahigh-performance liquid chromatography–tandem

ecological effect exposed by environmental pollutants in

mass spectrometry. Appropriate quality assurance and quality

previous studies. In this study, risk quotients (RQs) were

915

J. Wu et al.

Pharmaceuticals and personal care products in groundwater in North China

Hydrology Research

51.5

2020

applied to assess the ecological risk of PPCPs in GW; a high

whole area, except for EFX. Six PPCPs need priority atten-

RQ suggests a high ecological risk and vice versa. The RQs

tion, including IBU, SMX, SMZ, SMR, SFS, and SCP,

for each PPCP in the GW sample were calculated with the

whose frequencies of detection are higher than 50% in

following equation:

each sampling season. This result was consistent with a previous study in which sulfonamides were the most extensively

RQ ¼

MEC PNEC

(2)

detected PPCPs in adjacent watersheds (Zou et al. ). However, sulfonamides in other countries, such as France, Germany, and Korea, had relatively lower concerns

where MEC is the measured concentration, and PNEC is the

regarding GW (Supplementary material, Table S3). This

predicted no-effect concentration. In this study, the chronic or

phenomenon can be explained by the different use patterns

acute toxicity data of the target antibiotics were collected from

of PPCPs between countries. In China, sulfonamides are

previous studies (Supplementary material, Table S2). The

extensively used in poultry and aquaculture because of

PNEC was obtained from the lowest no observed effect con-

their low price (Chen et al. ). The maximum concen-

centration (NOEC) with bold marks in Table S2. According

trations of each PPCP ranged from below the detectable

to the recommendations of the European technical guidance

limit to 1,186 ng/L, spanning several orders of magnitude.

document (TGD), NOECs were priority toxicity data (EC

Only TCS in FCA-S in the dry season had a peak con-

). Three levels of ecological risks were classiﬁed: between

centration of >1,000 ng/L. Large variations in PPCP

0.01 and 0.1 is low risk; between 0.1 and 1 is medium risk;

concentrations in this study may be attributed to variations

and larger than 1 is high risk.

in residence time and attenuation processes, such as dilution, adsorption to aquifer material, and degradation.

Statistical analysis

Therefore, the physicochemical properties of PPCPs also affect the migration and transformation in aquifers. By com-

Statistical tests were performed to identify major factors that

parison with other counties, the maximum concentration

likely affect the occurrence and contribution of PPCPs in

of PPCPs was at a relatively high level in a global context

GW. Most results for individual PPCPs were non-detections

(Table S3).

in this study, and the datasets do not conform to any distribution. Therefore, nonparametric statistical methods were

Temporal and spatial variations in PPCPs in GW

used to perform hypothesis testing. A paired samples t-test was conducted to compare the temporal differences

Temporal variations in PPCPs

between different seasons. Independent sample tests were conducted to compare the spatial differences between two

The type of PPCP detection frequency can be categorized

zones and among different aquifers. Statistical analyses

into three classes. The CAF is different from other PPCPs

were performed with SPSS Statistics V20.0 (SPSS, Inc.

with a higher detection frequency in the dry season

Quarry Bay, HK). The p-value used to indicate statistical sig-

(Table 1). For ETM, SMM, SCP, SMR, SMX, CAP, IBU,

niﬁcance for all tests was 0.05.

and DIF, the detection frequencies in different aquifers were all higher in the wet season. For STZ, N4AcSMX, SFS, SMZ, TMP, and TCS, the highest detection frequencies

RESULTS AND DISCUSSION

in different aquifers were different. For instance, STZ had a high detection frequency in the dry season in the N zone

Overview of PPCPs in GW

and a high detection frequency in the wet season in the S zone. IBU was the most abundant PPCP in the GW in

The frequency of detection and the maximum concen-

both the dry and wet seasons for the whole study area, indi-

trations of the PPCPs in aquifers are summarized in

cating that GW pollution by IBU was not a sporadic but

Table 1. All the PPCPs were detected at least once in the

continuous event. Figure 2 shows the mean concentrations

916 J. Wu et al.

Maximum concentration (MC, ng/L) and frequency of detection (FD, %) of PPCPs in the dry and wet season in different aquifers of RGI

UA-N

UA-S

Dry

Wet

FCA-S

Dry

Wet

SCA-S

Dry

Wet

Dry

Wet

PPCPs

EFX

BLD

0.0

BLD

0.0

BLD

0.0

BLD

0.0

BLD

0.0

BLD

0.0

BLD

0.0

BLD

0.0

ETM

1.55

6.3

1.71

73.3

BLD

0.0

0.37

5.0

BLD

0.0

0.74

13.6

BLD

0.0

BLD

0.0

SMM

3.43

43.8

12.39

73.3

7.83

40.0

3.53

60.0

10.77

9.5

12.92

90.9

4.24

54.5

9.32

71.4

STZ

3.86

68.8

2.24

50.0

2.28

60.0

54.40

85.0

3.64

38.1

23.67

90.9

4.39

36.4

12.57

71.4

N4AcSMX

22.73

25.0

51.69

73.3

77.74

85.0

15.58

35.0

48.81

33.3

35.96

36.4

56.06

63.6

4.37

28.6

1.51

50.0

16.73

73.3

255.07

80.0

3.16

70.0

155.34

57.1

21.29

95.5

144.04

72.7

2.93

76.2

SCP

3.25

56.3

6.76

73.3

6.77

55.0

2.71

80.0

4.45

52.4

5.13

90.9

12.40

68.2

2.48

76.2

SMR

2.49

62.5

0.69

73.3

2.19

55.0

1.74

65.0

2.89

52.4

1.03

86.4

2.31

59.1

0.66

66.7

SMZ

0.49

56.3

0.83

73.3

56.47

95.0

19.50

95.0

12.57

76.2

9.49

86.4

8.58

95.5

33.44

81.0

SMX

19.29

100.0

11.01

100.0

54.19

75.0

8.12

65.0

15.36

95.2

7.45

81.8

20.56

77.3

15.08

71.4

TMP

1.03

81.3

3.23

100.0

0.54

95.0

0.23

45.0

0.86

90.5

0.64

50.0

4.89

100.0

2.01

52.4

CAF

782.62

56.3

461.87

26.7

249.29

5.0

BLD

0.0

398.51

19.0

BLD

0.0

235.52

4.5

BLD

0.0

CAP

0.55

43.8

0.16

66.7

9.66

60.0

1.07

95.0

9.23

47.6

1.01

95.5

8.37

22.7

0.48

85.7

IBU

24.50

100.0

42.99

100.0

30.92

90.0

34.35

100.0

64.21

95.2

23.19

95.5

190.14

77.3

7.91

90.5

TCS

18.05

25.0

109.90

26.7

5.19

5.0

BLD

0.0

1,186.38

19.0

BLD

0.0

10.81

4.5

BLD

0.0

DIF

BLD

0.0

22.05

73.3

BLD

0.0

BLD

0.0

BLD

0.0

5.59

13.6

BLD

0.0

BLD

0.0

Average FD

55.33

48.4

46.52

66.0

47.38

50.0

9.05

50.0

119.56

42.9

9.26

58.0

43.89

46.0

5.70

48.2

Hydrology Research

SFS

Pharmaceuticals and personal care products in groundwater in North China

Table 1

| 51.5

| 2020

J. Wu et al.

917

Figure 2

Pharmaceuticals and personal care products in groundwater in North China

Hydrology Research

51.5

2020

The mean concentration of the PPCPs in the unconﬁned aquifer (N and S zones) in different seasons. The concentration of EFX is not listed due to no available data. Bar graphs mean ± standard deviation (paired t-test, *P < 0.05; **P < 0.01).

for the PPCPs in the unconﬁned aquifer in different seasons.

Previous studies have shown that many factors inﬂuence

The mean concentrations of SMM, STZ, SCP, N4AcSMX,

the seasonal variations in PPCPs in aquatic environments,

TMP, ETM, IBU, and DIF in the N zone or the S zone

such as the consumption pattern, physicochemical proper-

were signiﬁcantly higher in the wet season than those in

ties, and the ﬂow conditions of rivers (Sun et al. ;

the dry season. For SFS, N4AcSMX, SMX, and CAP, the

Wang et al. b). Some PPCPs have seasonal uses, indicat-

mean concentrations were signiﬁcantly lower in the wet

ing that their inﬂuent load will vary throughout the year. It is

season. Figure 3 shows the spatial distribution of the cumu-

anticipated that high consumption of some PPCPs in the wet

lative concentration of PPCPs in GW in the RGI area. From

season will lead to high occurrence and concentration in

the viewpoint of total concentration, PPCPs concentrations

rivers containing sewage. However, in the dry season,

in most sites were higher in the dry season than those in the

some factors will reduce the attenuation rate of PPCPs in

wet season. The higher occurrence of PPCPs in the dry

rivers, such as lower runoff, decreased river ﬂow rate,

season (winter) was in accordance with previous studies

slower photolysis, thermal degradation, and biodegradation

(Sun et al. ).

in the cold season. As a result, the PPCPs in GW are directly

918

Figure 3

J. Wu et al.

Pharmaceuticals and personal care products in groundwater in North China

The distribution of PPCPs in GW in the RGI area.

Hydrology Research

51.5

2020

J. Wu et al.

919

Pharmaceuticals and personal care products in groundwater in North China

Hydrology Research

51.5

2020

and/or indirectly affected by the PPCPs in rivers. Therefore,

leaching potential group, EFX, SMM, and CAF were

the seasonal variation in PPCPs in aquifers in the RGI area

most likely to leach because of their high t1/2 and low

was affected by multiple factors simultaneously. Monitoring

KOC. TMP and IBU had moderate potential for leaching.

PPCPs at least twice a year (dry season/wet season) is essen-

TCS would be the compound most resistant to leaching. It

tial to obtain representative sample data to reﬂect the

is noted that due to relatively small chemical inputs, the tar-

dynamics of the seasonal change.

geted PPCPs are not expected to remain in the vadose zone for a long time. The different patterns of the 15 PPCPs in the

Spatial variations in PPCPs in different subzones In general, PPCP concentrations decreased from upstream (N zone) to downstream (S zone) (Figure 3). In the dry season, the mean concentration of CAF in the unconfined aquifer in the N zone was significantly higher than that in the S zone. In the wet season, the mean concentrations of SMM, TMP, IBU, CAF, TCS, N4AcSMX, ETM, DIF, and SMX were significantly higher in the N zone than those in the S zone. It is obvious that the dominant component in the N zone is CAF. In the S zone, SFS, IBU, and CAF were all abundant. CAF is known to be labile and has been suggested as a highly sensitive indicator of immediate contamination by sewage. The high CAF levels in the northern part were probably due to the high burden of sewage in the river water. The high concentrations of SFS and IBU in the southern part may be attributed to effluent from treated/untreated wastewater. As shown in Table 2, most of the targeted PPCPs appeared to have high leaching potential. Among the high

two subzones indicated that other factors would affect the occurrence in addition to the dilution of river flow due to precipitation. Dilution from sewage, degradation within the upstream sewer, rainfall, sampling mode, and pollution source characterization all contribute to this variability. Spatial variations in PPCPs in different aquifers Because information on confined aquifers in the northern part was not available, spatial variations in the PPCPs in different aquifers were analyzed in the southern part. As shown in Table 1, PPCPs in the dry season were more frequently detected in unconfined aquifers (50% for average frequency) than in the first confined aquifers (42% for average frequency in confined aquifers) and in the second confined aquifers (46% for average frequency). With regard to PPCPs in the wet season, the average frequency of detection in the second confined aquifers (48%) was lower than that in the unconfined aquifers (50%). For PPCPs in the first confined aquifers, the average frequency of detection (58%) was higher than that in the unconfined aquifers (50%). Figure 4 shows the mean concentrations of

Table 2

PPCPs in the conﬁned aquifer in different seasons. Generally,

The logt1/2, logKoc, and GUS of PPCPs in soils

Name

logt1/2

logKOC

GUS

Leaching potential

EFX

2.56

1.174

7.22

High

ETM

2.56

2.754

3.19

High

SMM

1.88

1.678

4.35

High

STZ

1.88

2.207

3.36

High

SMR

1.88

2.076

3.61

High

SMZ

1.88

2.282

3.22

High

SMX

1.88

2.412

2.98

High

TMP

1.89

2.857

2.16

Middle

CAF

1.48

4.43

High

IBU

1.48

2.626

2.03

Middle

TCS

2,880

4.369

0.77

Low

DIF

2.56

2.743

3.21

High

PPCPs in confined aquifers have seldom been investigated, possibly because they were supposed to be protected from pollution by the vadose zone and the upper confining bed. However, it can be seen that the mean concentrations of some PPCPs in the two confined aquifers were relatively higher than those in the unconfined aquifer, suggesting insignificant attenuation effects by the upper confining bed. For many PPCPs, there may be multiple pathways into the GW, and difficulties in understanding these processes are compounded by the paucity of information compared to that for SW. The environmental behavior of PPCPs in aquifers depends not only on the specific hydrogeochemical conditions (Wang et al. b) but also on the physicochemical properties of the PPCPs in the GW system (Koh et al.

920

Figure 4

J. Wu et al.

Pharmaceuticals and personal care products in groundwater in North China

Hydrology Research

51.5

2020

The mean concentration of the PPCPs in aquifers (S zone) in different seasons. The concentration of EFX is not listed due to no available data. Mean contribution to each heavy metal concentration from each source.

). One explanation for the phenomenon of PPCPs in

defects in the well casing or via the backﬁlling sand

two conﬁned GW samples is that those aquifers are con-

around the well casing, driven by the difference in the poten-

sidered to be recharged from the upper aquifers. Another

tial of aquifers. The GW was recharged by the river ﬂows

reason involves contamination by unconﬁned GW via

deeper than 80 m in the area of the studied well.

J. Wu et al.

921

Pharmaceuticals and personal care products in groundwater in North China

Ecological risk of PPCPs

Hydrology Research

51.5

2020

are insufﬁcient to reduce the ecological risk of all PPCPs to below the high or medium level in these areas. These pri-

The highest concentration and lowest PNEC were simul-

ority PPCPs should be given special attention to be strictly

taneously used as a worst-case scenario assuming ecological

regulated during GW recharging.

risk. Due to the lack of predictive toxicity data on the chronic effects of SMM N4AcSMX and DIF, their RQs were ignored. As shown in Table 3, eight PPCPs were detected in aquifers at

CONCLUSIONS

low-risk levels. The ecosystem risks of SFS, CAF, IBU, and TCS were found to be at least middle risk by using a RQ. Similar

With increasing artiﬁcial sewage recharge activities in the

to the concentration of PPCPs, the potential ecological risks

past several years, the detection frequencies and concen-

also exhibited spatial and seasonal variations to some extent.

tration levels of PPCPs in aquifers in North China were

Overall, the potential ecological risks posed by PPCPs to GW

generally higher than those reported in global GW. As a

could be a serious issue. The results of this study indicated

whole, IBU, SMX, SMZ, SMR, SFS, and SCP were widely

that PPCPs in GW in the RGI area also need to be monitored

detected in aquifers. The seasonal dynamics of some

for regulation and control by legislation due to their wide distri-

PPCPs showed statistically signiﬁcant changes. Signiﬁcant

bution and signiﬁcant adverse ecological effects.

spatial variations in PPCPs in different subzones and in

From the perspective of ecological risk, SFS, CAF, IBU,

different aquifers were also found. However, the reasons

and TCS can be viewed as priority PPCPs because of middle

for the spatiotemporal variation were supposed to be com-

or high ecological risk. These results indicated that AR was a

plex and most likely caused by the combination of speciﬁc

source of PPCPs in GW, particularly in GW with short resi-

hydrogeochemical conditions and the physicochemical

dence times, and poses a threat to ecological systems. In

properties of the PPCPs in the GW system. Risk assessment

addition, another GW–SW exchange engineering method,

revealed that four antibiotics (DIF, IBU, CAF, and SFS)

river bank ﬁltration, has been proved to be an important,

posed at least a medium risk in GW under the worst scen-

effective, and inexpensive technique for SW puriﬁcation

arios. Finally, eight PPCPs (IBU, SMX, SMZ, SMR, SFS,

due to ﬁltration, adsorption, and biodegradation (Munz

SCP, CAF, and TCS) were identiﬁed as priority control anti-

et al. ; Wang et al. c). However, these processes

biotics in the RGI area in North China.

Table 3

The ecological risk value of PPCPs in the dry and wet season in different aquifers of RGI

UA-N

UA-S

PPCPs

Dry

Wet

ETM

7.75 × 10 4

8.55 × 10 4

STZ

3.86 × 10

SFS

2.44 × 10 3

SCP

1.31 × 10

SMR SMZ

1.64 × 10

SMX TMP

3.55 × 10

CAF

78.26 × 10

CAP

7.4 × 10

IBU

2.45 × 10

TCS

FCA-S

Dry

Wet

Dry

1.85 × 10 4

NA 4

SCA-S

NA 4

2.24 × 10

2.28 × 10

5.44 × 10

3.64 × 10

2.7 × 10 4

4.1 × 10 1

5.1 × 10 3

2.5 × 10 1

2.72 × 10

2.73 × 10

1.09 × 10

1.8 × 10

3.66 × 10 3

1.02 × 10 3

3.22 × 10 3

2.56 × 10 3

Wet

Dry

Wet

3.7 × 10 4

2.37 × 10 3

4.39 × 10 4

1.25 × 10 3

3.4 × 10 2

2.3 × 10 1

4.73 × 10 3

1 × 10 3

2.07 × 10

5 × 10

4.25 × 10 3

1.52 × 10 3

3.4 × 10 3

3.16 × 10

2.86 × 10

1.11 × 10 4

2.77 × 10

1.88 × 10

6.5 × 10

2.05 × 10 3

1.17 × 10 3

5.77 × 10 3

8.63 × 10 4

1.63 × 10 3

7.93 × 10 4

2.19 × 10 3

1.61 × 10 3

9.02 × 10

4.19 × 10

9.7 × 10 4

1.12 × 10

1.85 × 10

8.09 × 10

2.96 × 10

46.18 × 10

24.93 × 10

39.85 × 10

2.17 × 10

1.3 × 10

4.3 × 10

3.09 × 10

5.5 × 10

2.59 × 10

1.45 × 10

1.25 × 10

3.43 × 10

6.42 × 10

5.93 × 10

2.19 × 10

1.69 × 10

6.92 × 10 3

23.55 × 10

1.37 × 10

1.13 × 10

6.44 × 10 9

2.32 × 10

19.01 × 10

7.9 × 10 1

5.4 × 10

922

J. Wu et al.

Pharmaceuticals and personal care products in groundwater in North China

Despite the uncertainties associated with sampling, geology, hydrology, and environment and parameters, the results obtained from this study are valuable. The environmental risk of PPCPs in GW recharged by river water has been questioned due to dilution, filtration, and degradation. However, the concentration and associated ecological risk indicate a significant threat by PPCPs to the RGI area, particularly to riverbank organisms. This work will help us gain insights into the contamination and monitoring of PPCPs in a GW–river interactive system and provide a basis to propose an appropriate mitigation strategy for PPCP management in similar watersheds around the world. Further studies should focus on the influencing factors of PPCPs during SW and GW exchange processes.

ACKNOWLEDGEMENTS This study was ﬁnancially supported by the study on the watershed water management system of the Sino-German cooperation project and research on the key technology of water quality safety assurance in groundwater extraction reduction in Miyun district, Beijing. The authors would like to thank the editor and anonymous reviewers for their valuable comments that greatly improved the work. Author contributions as follows: J.W.: conceptualization, methodology, software, and writing the original draft; J.L.: data curation and visualization; Z.P.: investigation and writing – reviewing and editing; B.W.: investigation and resources; and D.Z.: supervision and funding acquisition.

SUPPLEMENTARY MATERIAL The Supplementary Material for this paper is available online at https://dx.doi.org/10.2166/nh.2020.001.

REFERENCES Aa, Y., Wang, G., Liu, T., Xue, B. & Kuczera, G.  Spatial variation of correlations between vertical soil water and evapotranspiration and their controlling factors in a semiarid region. J. Hydrol. 574, 53–63. https://doi.org/10.1016/ j.jhydrol.2019.04.023.

Hydrology Research

51.5

2020

Barnes, K. K., Kolpin, D. W., Furlong, E. T., Zaugg, S. D., Meyer, M. T. & Barber, L. B.  A national reconnaissance of pharmaceuticals and other organic wastewater contaminants in the United States – I) groundwater. Sci. Total Environ. 402 (2–3), 192–200. https://doi.org/10.1016/j.scitotenv.2008. 02.021. Bexfield, L. M., Toccalino, P. L., Belitz, K., Foreman, W. T. & Furlong, E. T.  Hormones and pharmaceuticals in groundwater used as a source of drinking water across the United States. Environ. Sci. Technol. 53 (6), 2950–2960. https://doi.org/10.1021/acs.est.8b05592. Bo, Z., Mei, H., Yongsheng, Z., Xueyu, L., Xuelin, Z. & Jun, D.  Distribution and risk assessment of fluoride in drinking water in the west plain region of Jilin province, China. Environ. Geochem. Health 25 (4), 421–431. https://xs.scihub. ltd/https://doi.org/10.1023/B:EGAH.0000004560.47697.91. Brandt, K. K., Amezquita, A., Backhaus, T., Boxall, A., Coors, A., Heberer, T. & Zhu, Y. G.  Ecotoxicological assessment of antibiotics: a call for improved consideration of microorganisms. Environ. Int. 85, 189–205. https://doi.org/ 10.1016/j.envint.2015.09.013. Buerge, I. J., Buser, H. R., Kahle, M., Muller, M. D. & Poiger, T.  Ubiquitous occurrence of the artificial sweetener acesulfame in the aquatic environment: an ideal chemical marker of domestic wastewater in groundwater. Environ. Sci. Technol. 43 (12), 4381–4385. https://doi.org/10.1021/es900126x. Chen, H., Jing, L., Teng, Y. & Wang, J.  Characterization of antibiotics in a large-scale river system of China: occurrence pattern, spatiotemporal distribution and environmental risks. Sci. Total Environ. 618, 409–418. https://doi.org/10.1016/ j.scitotenv.2017.11.054. Cleuvers, M.  Mixture toxicity of the anti-inflammatory drugs diclofenac, ibuprofen, naproxen, and acetylsalicylic acid. Ecotox. Environ. Safe 59 (3), 309–315. https://doi.org/10. 1016/S0147-6513(03)00141-6. Díaz-Cruz, M. S. & Barceló, D.  Trace organic chemicals contamination in ground water recharge. Chemosphere 72 (3), 333–342. https://doi.org/10.1016/j.chemosphere.2008. 02.031. Dougherty, J. A., Swarzenski, P. W., Dinicola, R. S. & Reinhard, M.  Occurrence of herbicides and pharmaceutical and personal care products in surface water and groundwater around Liberty Bay, Puget Sound, Washington. J. Environ. Qual. 39 (4), 1173–1180. https://dl.sciencesocieties.org/ publications/jeq/abstracts/39/4/1173. EC (European Commission)  Technical Guidance Document in Support of Commission Directive 93//67/EEC on Risk Assessment for New Notified Substances and Commission Regulation (EC) No. 1488/94 on Risk Assessment for Existing Substances. Einsiedl, F., Radke, M. & Maloszewski, P.  Occurrence and transport of pharmaceuticals in a karst groundwater system affected by domestic wastewater treatment plants. J. Contam. Hydrol. 117 (1–4), 26–36. https://doi.org/10.1016/j.jconhyd. 2010.05.008.

923

J. Wu et al.

Pharmaceuticals and personal care products in groundwater in North China

Fang, Q., Wang, G., Liu, T., Xue, B.-L. & Aa, Y.  Controls of carbon flux in a semi-arid grassland ecosystem experiencing wetland loss: vegetation patterns and environmental variables. Agric. Forest. Meteorol. 259, 196–210. Grünheid, S., Amy, G. & Jekel, M.  Removal of bulk dissolved organic carbon (DOC) and trace organic compounds by bank filtration and artificial recharge. Water Res. 39 (14), 3219–3228. https://doi.org/10.1016/j.watres.2005.05.030. Hanna, N., Sun, P., Sun, Q., Li, X., Yang, X., Ji, X., Zou, H., Ottoson, J., Nilsson, L. E., Berglund, B., Dyar, O. J., Tamhankar, A. J. & Stålsby Lundborg, C.  Presence of antibiotic residues in various environmental compartments of Shandong province in eastern China: its potential for resistance development and ecological and human risk. Environ. Int. 114, 131–142. https://doi.org/10.1016/j.envint. 2018.02.003. Jiang, Y. H., Li, M. X., Guo, C. S., An, D., Xu, J., Zhang, Y. & Xi, B. D.  Distribution and ecological risk of antibiotics in a typical effluent–receiving river (Wangyang River) in north China. Chemosphere 112, 267–274. https://doi.org/10.1016/ j.chemosphere.2014.04.075. Koh, Y. K., Chiu, T. Y., Boobis, A. R., Scrimshaw, M. D., Bagnall, J. P., Soares, A., Pollard, S., Cartmell, E. & Lester, J. N.  Influence of operating parameters on the biodegradation of steroid estrogens and nonylphenolic compounds during biological wastewater treatment processes. Environ. Sci. Technol. 43, 6646–6654. https://doi.org/10.1021/es901612v. Kostich, M. S., Batt, A. L. & Lazorchak, J. M.  Concentrations of prioritized pharmaceuticals in effluents from 50 large wastewater treatment plants in the US and implications for risk estimation. Environ. Pollut. 184, 354–359. https://doi. org/10.1016/j.envpol.2013.09.013. Kuroda, K., Murakami, M., Oguma, K., Muramatsu, Y., Takada, H. & Takizawa, S.  Assessment of groundwater pollution in Tokyo using PPCPs as sewage markers. Environ. Sci. Technol. 46 (3), 1455–1464. https://doi.org/10.1021/es202059 g. Lapworth, D. J., Baran, N., Stuart, M. E. & Ward, R. S.  Emerging organic contaminants in groundwater: a review of sources, fate and occurrence. Environ. Pollut. 163, 287–303. https://doi.org/10.1016/j.envpol.2011.12.034. Li, Z., Li, M., Liu, X., Ma, Y. & Wu, M.  Identification of priority organic compounds in groundwater recharge of China. Sci. Total Environ. 493, 481–486. https://doi.org/10. 1016/j.scitotenv.2014.06.005. Li, J., Yang, Y., Huan, H., Li, M., Xi, B., Lv, N. & Yang, J.  Method for screening prevention and control measures and technologies based on groundwater pollution intensity assessment. Sci. Total Environ. 551, 143–154. https://doi.org/ 10.1016/j.scitotenv.2015.12.152. Liu, J. L. & Wong, M. H.  Pharmaceuticals and personal care products (PPCPs): a review on environmental contamination in China. Environ. int. 59, 208–224. https://doi.org/10.1016/ j.envint.2013.06.012. Munz, M., Oswald, S. E., Schäfferling, R. & Lensing, H. J.  Temperature-dependent redox zonation, nitrate removal and

Hydrology Research

51.5

2020

attenuation of organic micropollutants during bank filtration. Water Res. https://doi.org/10.1016/j.watres.2019.06.041. Peng, X., Ou, W., Wang, C., Wang, Z., Huang, Q., Jin, J. & Tan, J.  Occurrence and ecological potential of pharmaceuticals and personal care products in groundwater and reservoirs in the vicinity of municipal landfills in China. Sci. Total Environ. 490, 889–898. https://doi.org/10.1016/j.scitotenv. 2014.05.068. Petrie, B., Barden, R. & Kasprzyk-Hordern, B.  A review on emerging contaminants in wastewaters and the environment: current knowledge, understudied areas and recommendations for future monitoring. Water Res. 72, 3–27. https://doi.org/10.1016/j.watres.2014.08.053. Singh, L. K., Jha, M. K. & Chowdary, V. M.  Multi-criteria analysis and GIS modeling for identifying prospective water harvesting and artificial recharge sites for sustainable water supply. J. Clean Prod. 142, 1436–1456. https://doi.org/10. 1016/j.jclepro.2016.11.163. Sui, Q., Huang, J., Deng, S., Chen, W. & Yu, G.  Seasonal variation in the occurrence and removal of pharmaceuticals and personal care products in different biological wastewater treatment processes. Environ. Sci. Technol. 45 (8), 3341–3348. https://doi.org/10.1021/es200248d. Sun, Q., Li, Y., Li, M., Ashfaq, M., Lv, M., Wang, H., Hu, A. & Yu, C.  PPCPs in Jiulong River estuary (China): spatiotemporal distributions, fate, and their use as chemical markers of wastewater. Chemosphere 150, 596–604. https:// doi.org/10.1016/j.chemosphere.2016.02.036. Teijon, G., Candela, L., Tamoh, K., Molina-Díaz, A. & Fernández-Alba, A. R.  Occurrence of emerging contaminants, priority substances (2008/105/CE) and heavy metals in treated wastewater and groundwater at Depurbaix facility (Barcelona, Spain). Sci. Total Environ. 408 (17), 3584–3595. https://doi.org/10.1016/j.scitotenv. 2010.04.041. Thomas, R., Gough, R. & Freeman, C.  Linear alkylbenzene sulfonate (LAS) removal in constructed wetlands: the role of plants in the treatment of a typical pharmaceutical and personal care product. Ecol. Eng. 106, 415–422. https://doi. org/10.1016/j.ecoleng.2017.06.015. Wang, Z., Zhang, X. H., Huang, Y. & Wang, H.  Comprehensive evaluation of pharmaceuticals and personal care products (PPCPs) in typical highly urbanized regions across China. Environ. Pollut. 204, 223–232. https://doi.org/ 10.1016/j.envpol.2015.04.021. Wang, G., Li, J., Sun, W., Xue, B., Aa, Y. & Liu, T. a Non-point source pollution risks in a drinking water protection zone based on remote sensing data embedded within a nutrient budget model. Water Res. 157, 238–246. https://doi.org/10. 1016/j.watres.2019.03.070. Wang, P., Rene, E. R., Yan, Y., Ma, W. & Xiang, Y. b Spatiotemporal evolvement and factors influencing natural and synthetic EDCs and the microbial community at different groundwater depths in the Chaobai watershed: a long-term field study on a river receiving reclaimed water.

924

J. Wu et al.

Pharmaceuticals and personal care products in groundwater in North China

J. Environ. Manage. 246, 647–657. https://doi.org/10.1016/j. jenvman.2019.05.156. Wang, P., Yao, J., Wang, G., Hao, F., Shrestha, S., Xue, B., Xie, G. & Peng, Y. c Exploring the application of artificial intelligence technology for dentification of water pollution characteristics and tracing the source of water quality pollutants. Sci. Total Environ. 693, 133440. Yang, Z., Cai, Y. & Mitsch, W. J.  Ecological and hydrological responses to changing environmental conditions in China’s river basins. Ecol. Eng. 76, 1–6. https://doi.org/10.1016/ j.ecoleng.2014.12.007. Yang, L., Cheng, Q., Lin, L., Wang, X., Chen, B., Luan, T. & Tam, N. F.  Partitions and vertical profiles of 9 endocrine disrupting chemicals in an estuarine environment: effect of tide, particle size and salinity. Environ. Pollut. 211, 58–66. https://doi.org/10.1016/j.envpol.2015.12.034. Yang, Y., Ok, Y. S., Kim, K. H., Kwon, E. E. & Tsang, Y. F. a Occurrences and removal of pharmaceuticals and personal care products (PPCPs) in drinking water and water/sewage treatment plants: a review. Sci. Total Environ. 596, 303–320. https://doi.org/10.1016/j.scitotenv.2017.04.102. Yang, L., He, J. T., Su, S. H., Cui, Y. F., Huang, D. L. & Wang, G. C. b Occurrence, distribution, and attenuation of pharmaceuticals and personal care products in the riverside groundwater of the Beiyun River of Beijing, China. Environ.

Hydrology Research

51.5

2020

Sci. Pollut. Res. 24 (18), 15838–15851. https://xs.scihub.ltd/ https://doi.org/10.1007/s11356-017-8999-0. Zhang, Q.-Q., Ying, G.-G., Pan, C.-G., Liu, Y.-S. & Zhao, J.-L.  Comprehensive evaluation of antibiotics emission and fate in the river basins of China: source analysis, multimedia modeling, and linkage to bacterial resistance. Environ. Sci. Technol. 49, 6772–6782. https://doi.org/10.1021/acs.est.5b00729. Zhang, Y., Lv, T., Carvalho, P. N., Zhang, L., Arias, C. A., Chen, Z. & Brix, H.  Ibuprofen and iohexol removal in saturated constructed wetland mesocosms. Ecol. Eng. 98, 394–402. https://doi.org/10.1016/j.ecoleng.2016.05.077. Zhang, P., Cai, Y., Yang, W., Yi, Y., Yang, Z. & Fu, Q.  Multiple spatio-temporal patterns of vegetation coverage and its relationship with climatic factors in a large dam-reservoirriver system. Ecol. Eng. 138, 188–199. https://doi.org/10. 1016/j.ecoleng.2019.07.016. Zheng, G., Cao, J. R., Cheng, X. S., Ha, D. & Wang, F. J.  Experimental study on the artiﬁcial recharge of semiconﬁned aquifers involved in deep excavation engineering. J. Hydrol. 557, 868–877. https://doi.org/10.1016/j.jhydrol.2018.01.020. Zou, S. C., Xu, W. H., Zhang, R. J., Tang, J. H., Chen, Y. J. & Zhang, G.  Occurrence and distribution of antibiotics in coastal water of the Bohai Bay, China: impacts of river discharge and aquaculture activities. Environ. Pollut. 159, 2913–2920. https://doi.org/10.1016/j.envpol.2011.04.037.

First received 1 January 2020; accepted in revised form 15 March 2020. Available online 20 April 2020

925

51.5

2020

Impacts of bias nonstationarity of climate model outputs on hydrological simulations Yu Hui, Yuni Xu, Jie Chen, Chong-Yu Xu and Hua Chen

ABSTRACT Bias correction methods are based on the assumption of bias stationarity of climate model outputs. However, this assumption may not be valid, because of the natural climate variability. This study investigates the impacts of bias nonstationarity of climate models simulated precipitation and temperature on hydrological climate change impact studies. The bias nonstationarity is determined as the range of difference in bias over multiple historical periods with no anthropogenic climate change for four different time windows. The role of bias nonstationarity in future climate change is assessed using the signal-to-noise ratio as a criterion. The results show that biases of climate models simulated monthly and annual precipitation and temperature vary with time, especially for short time windows. The bias nonstationarity of precipitation plays a great role in future precipitation change, while the role of temperature bias is not important. The bias nonstationarity of climate model outputs is ampliﬁed when driving a hydrological model for hydrological simulations. The increase in the length of time window can mitigate the impacts of bias nonstationarity for streamﬂow projections.

Yu Hui Jie Chen (corresponding author) Hua Chen State Key Laboratory of Water Resources and Hydropower Engineering Science, Wuhan University, Wuhan, China E-mail: jiechen@whu.edu.cn Yu Hui Changjiang Institute of Survey, Planning, Design and Research, Wuhan, China Yuni Xu Bureau of Hydrology, ChangJiang Water Resources Commission, Wuhan, China

Thus, a long time period is suggested to be used to calibrate a bias correction method for hydrological climate change impact studies to reduce the inﬂuence of natural climate variability. Key words

| bias nonstationarity, climate change signal, climate model outputs, hydrology, natural climate variability

Chong-Yu Xu Department of Geosciences, University of Oslo, Oslo, Norway

HIGHLIGHTS

• • •

The biases of GCM precipitation and temperature vary with time, due to natural climate variability. The bias nonstationarity of precipitation plays a great role in future precipitation change, while the role of temperature bias is not important. The bias nonstationarity of precipitation and temperature has great considerable impacts on future streamﬂow changes.

INTRODUCTION The assessment of climate change impacts on the hydrologi-

Marhaento et al. ; Shen et al. ; Lu & Qin ;

cal cycle has been widely investigated during recent years

Ragettli et al. ). Global climate models (GCMs) can pro-

(Graham et al. ; Li et al. ; Chen et al. a;

vide climate variables (e.g. precipitation and temperature) for the future period used as inputs of hydrological models

for hydrological impact studies. However, the coarse resolution of GCMs does not meet the need of high resolution for the hydrological models (Maraun et al. ). In parallel,

926

Y. Hui et al.

Impacts of bias nonstationarity on hydrology

Hydrology Research

51.5

2020

GCMs are imperfect representations of reality with systema-

of climate models simulated precipitation are nonstationary

tic biases found between climate model simulations and

even for two close historical periods, while temperature

observations. To resolve these problems, a number of down-

biases are relatively stationary. This study attributed the

scaling methods have been developed during the last two

bias nonstationarity of precipitation to natural climate varia-

decades. Especially, bias correction becomes a standard pro-

bility (multi-decadal climate variability). A similar study was

cedure when using regional climate model (RCM) outputs

also carried out by Wang et al. (), who tested bias non-

for hydrological impact studies (van Pelt et al. ; Teutsch-

stationarity of precipitation in the eastern United States

bein & Seibert ; Olsson et al. ; Zhuan et al. ).

determined by a skill score, which compares the errors of

During recent years, bias correction methods are also com-

a downscaling method over the validation period with the

monly used for GCM outputs. Since the resolution of

errors of observations between calibration and validation

climate model outputs is lower than that of observations,

periods. The results show that precipitation biases are non-

bias correction methods also act as downscaling methods.

stationary at most of the stations, especially for the annual

The usually used bias correction methods range from

extreme precipitation. Taking into account multi-decadal cli-

simple mean-based scaling to sophisticated distribution-

mate variability, Nahar et al. () investigated the bias

based mapping and multivariate or/and multisite correction.

stationarity of six GCMs simulated monthly and seasonal

The commonly used bias correction methods are usually

precipitation and temperature over multi-decadal time

based on an assumption that biases of climate models out-

periods from 1900 to 1999 in Australia. One hundred

puts are stationary. In other words, these bias correction

samples of the observations and GCM simulations gener-

methods assume climate model outputs present the same

ated using bootstrapping was used to calculate uncertainty

biases in magnitude and direction between historical and

in biases with a 95% conﬁdence, which represents a possible

future periods. However, this assumption may not be

range of bias if bias stationarity exists. When the actual bias

valid, as pointed out in a few recent studies (e.g. Buser

goes beyond this uncertainty, the bias was assumed to be

et al. ; Ehret et al. ; Gutierrez et al. ; Maurer

nonstationary. This study showed that biases of precipitation

et al. ; Velázquez et al. ; Dixon et al. ). For

and temperature are not stationary for some regions in Aus-

example, Christensen et al. () evaluated bias nonstatio-

tralia (e.g. east coast of Australia), because of the natural

narity on simulated monthly precipitation and temperature

climate variability.

from an ensemble of 13 RCMs over Europe. They found

For a historical period with no anthropogenic forcing,

that biases of simulated precipitation and temperature vary

the bias nonstationarity is attributed to natural climate varia-

as a second-order function of observations, suggesting that

bility. Natural climate variability refers to variations in the

bias nonstationarity exists in different climatic regimes.

mean state and other statistics of the climate, due to natural

Maraun () veriﬁed the bias stationary assumption in

internal processes or natural external forcing. It involves a

RCMs for European seasonal mean temperature and pre-

wide range of time scales, from one day to the next, as

cipitation sums using a pseudo-reality approach, which

well as from one year or multi-decade to the next. For the

considers one climate model as pseudo observation to com-

application of a bias correction method in climate change

pare with other climate model simulations. The results

impact studies, the decadal and multiple decadal time

showed that biases are relatively stable in general, but bias

scales are widely and implicitly used. At the decadal or

nonstationarity was identiﬁed in some regions where

multi-decadal time scales, the natural climate variability

changes in potentially relevant physical variables are signiﬁ-

includes some natural modes of decadal or multi-decadal cli-

cant. This study was conducted in the climate model world,

mate variability, such as El Niño/Southern Oscillation

the transferability to real world needs to be further investi-

(ENSO), Atlantic Multidecadal Oscillation (AMO), Paciﬁc

gated. To test the bias nonstationarity in real-world

Decadal Oscillation (PDO), and Interdecadal Paciﬁc Oscil-

climate, Chen et al. () compared the biases between cli-

lation (IPO). The phase of these modes has been found to

mate model simulations and corresponding observations

be linked with changes in precipitation and temperature.

over two historical periods. The results showed that biases

For example, the PDO warm (cold) phase periods link to

927

Y. Hui et al.

Impacts of bias nonstationarity on hydrology

Hydrology Research

51.5

2020

the decrease (increase) in precipitation in the majority of

propagation of bias nonstationarity in hydrological impact

China (Ouyang et al. ). Precipitation decreases during

studies. The ﬁrst objective is achieved by separating a long

IPO warm phase periods, while it increases during IPO

historical climate series (100 years) to decadal and multi-

cool phase periods in east Australia (Power et al. ).

decadal time periods to calculate the difference in bias

However, since the unpredictability and complexity of natu-

among different periods. Precipitation and temperature,

ral dynamic processes make it difﬁcult for climate models to

which are primary variables for hydrological simulations,

accurately capture the characteristics, GCMs from Coupled

are investigated. The second objective is achieved by com-

Model Intercomparison Project phase 5 (CMIP5) have been

paring the range of biases among multiple periods to

found to poorly represent the phase in natural modes of

climate change signals between future and historical

decadal or multi-decadal climate variability (Polade et al.

periods. The last objective is achieved by analyzing stream-

; Ruiz-Barradas et al. ; Bellenger et al. ;

ﬂow time series simulated by a hydrological model using

Fuentes-Franco et al. ). This poor representation for

the above two approaches.

GCMs could lead to biases between simulations and obser-

The rest of the paper is organized as follows: the follow-

vations when sampling in a ﬁnite decadal or multi-decadal

ing section introduces the study area and hydro-climate data

time, and further lead to changes in bias among different

followed by the methodology. The main results are then pre-

decadal or multi-decadal periods (Nahar et al. ).

sented, and the discussion and conclusions are presented in

In climate change impact studies, natural climate varia-

the ﬁnal section.

bility used to be estimated based on multi-decadal climate model simulations in the absence of anthropogenic-induced climate change (Hulme et al. ; Arnell ; Chen et al.

STUDY AREA AND DATA

b). In these methods, the differences among several separated multi-decadal periods obtained from a long time

Study area

series of unforced simulations were supposed to represent the multi-decadal natural climate variability. Similarly, to

The case study was conducted at the Hanjiang River basin

evaluate bias nonstationarity of climate model outputs, the

(Figure 1), located in south-central China, which is the lar-

estimate of bias nonstationarity is depicted as the range of

gest tributary of the Yangtze River basin. The Hanjiang

the biases among multi-decadal historical periods, while

River ﬂows through Shaanxi and Hubei Provinces with a

the timescale of each period may have an effect on the

drainage area of 159,000 km2 and a length of 1,567 km.

results of estimation. The different temporal scales of

The watershed above the Danjiangkou Reservoir, with a

sampling can lead to changes in the magnitude and direc-

sub-basin area of 95,200 km2 and a length of 918 km, was

tion of bias, due to different phases between GCM outputs

used in this study. The Danjiangkou reservoir is the water

and observations. Therefore, the impacts of different

source of the Middle Route of the South-to-North Water

temporal scales of natural climate variability on bias

Transfer Project in China. The watershed has a subtropical

nonstationarity of climate model outputs need to be investi-

monsoon climate. The mean annual precipitation is about

gated, especially for their propagation in hydrology.

840 mm, of which 70–80% of the total amount falls in the

Even though few studies (Maraun ; Chen et al. ;

wet season from May to September. The average maximum

Velázquez et al. ; Nahar et al. ) have investigated the

and minimum temperatures are 24–29 and 0–3 C, respect-

bias nonstationarity of climate model outputs, the role of

ively. The daily mean discharge of the Hanjiang River is

bias nonstationarity in future climate change was not

approximately 1,150 m3/s.

investigated, especially in hydrological impact studies. Accordingly, the objectives of this study are to investigate:

Data

(1) the bias nonstationarity of climate model outputs in the context of natural climate variability, (2) the role of

This study used both observed and GCM-simulated precipi-

bias nonstationarity in future climate change, and (3) the

tation and temperature for the Hanjiang watershed. The

928

Figure 1

Y. Hui et al.

Impacts of bias nonstationarity on hydrology

Hydrology Research

51.5

2020

Location map of the Hanjiang River basin. The upstream part above the Danjiangkou reservoir was used in this study.

inﬂow to the Danjiangkou reservoir calculated using a water

21st century. The other greenhouse gas emission scenarios

mass balance method was also used for hydrological model

(RCP8.5) are also presented and discussed in the Discussion

calibration and validation. The streamﬂow time series cover-

section. Since this study only used monthly and annual data,

ing the 1961–2000 period was provided by the Bureau of

the daily precipitation and temperature were summed (for

Hydrology of the Changjiang Water Resources Commission.

precipitation) or averaged (for temperature) to monthly and

The observed precipitation and temperature were

annual values. In order to run a lumped hydrological

obtained from the gridded Climatic Research Unit (CRU)

model, all gridded precipitation and temperature within and

Time-series (TS) data, produced by CRU at the University

surrounding the Hanjiang watershed were averaged to a

of East Anglia, UK. This dataset provides monthly gridded

single time series using the Thiessen polygon method for

data with a high resolution of 0.5 × 0.5 . The latest version

both observations and GCM simulations.

CRU TS version 4.01 (Harris et al. ) covering the period 1901–2000 was used in this study. The GCM-simulated daily precipitation and maximum

METHODOLOGY

and minimum temperatures were extracted from the database of CMIP5. In order to capture the uncertainty related to cli-

A climate time series for the past–present period may contain

mate models, 17 GCMs were used in this study (Table 1).

both natural climate variability and anthropogenic climate

The time period covers from 1901 to 2100. All GCMs’

change signal. In order to investigate the impacts of natural

simulations during the 1901–2005 period were driven by

climate variability on bias correcting climate model outputs,

historical climate forcing (including natural solar, volcanic

the climate change signal needs to be ﬁrst removed. A

variations and anthropogenic radiative forcing), while those

simple linear detrending method of Zhuan et al. () was

during the 2006–2100 period were generated under the

used to remove the anthropogenic climate change signal for

representative concentration pathway (RCP) 4.5. RCP4.5

observed and GCM-simulated monthly precipitation and

corresponds to the medium anthropogenic forcing for the

temperature for the 1901–2000 period. This detrending

Y. Hui et al.

929

Table 1

Impacts of bias nonstationarity on hydrology

Hydrology Research

51.5

2020

Basic information of the used 17 CMIP5 models

Resolution (Longitude × ID

Model name

Modeling center

Institution

Latitude)

ACCESS1.0

CSIRO-BOM

1.875 × 1.25

ACCESS1.3

Commonwealth Scientiﬁc and Industrial Research Organization (CSIRO) and Bureau of Meteorology (BOM), Australia

BCC-CSM1.1(m)

BCC

Beijing Climate Center, China Meteorological Administration

1.125 × 1.25

CNRM-CM5

CNRM-CERFACS

Centre National de Recherches Météorologiques/Centre Européen de Recherche et Formation Avancée en Calcul Scientiﬁque

1.4 × 1.4

CSIRO-Mk3.6.0

CSIRO-QCCCE

Commonwealth Scientiﬁc and Industrial Research Organization in collaboration with Queensland Climate Change Centre of Excellence

1.9 × 1.9

CanESM2

CCCMA

Canadian Centre for Climate Modelling and Analysis

2.8 × 2.8

GFDL-CM3

NOAA GFDL

NOAA Geophysical Fluid Dynamics Laboratory

2.5 × 2.0

GFDL-ESM2G

2.5 × 2.0

GFDL-ESM2M

2.5 × 2.0

INM-CM4

INM

Institute for Numerical Mathematics

2.0 × 1.5

IPSL-CM5A-LR

IPSL

L’Institut Pierre-Simon Laplace

3.75 × 1.9

IPSL-CM5A-MR

2.5 × 1.25

IPSL-CM5B-LR

3.75 × 1.9

MIROC-ESM-CHEM

MIROC-ESM

MIROC5

MRI-CGCM3

1.875 × 1.25

Japan Agency for Marine-Earth Science and Technology, Atmosphere and Ocean Research Institute (The University of Tokyo), and National Institute for Environmental Studies

2.8 × 2.8

MIROC

Atmosphere and Ocean Research Institute (The University of Tokyo), National Institute for Environmental Studies, and Japan Agency for Marine-Earth Science and Technology

1.4 × 1.4

MRI

Meteorological Research Institute

1.1 × 1.1

MIROC

2.8 × 2.8

method ﬁrst detects the trend and breakpoint using the

variability at the temporal scale, the 100-year precipitation

Mann–Kendall test (Mann ; Kendall ). If a trend

and temperature time series are divided into several inde-

exists, it is removed for the period after the breakpoint

pendent decadal and multi-decadal periods. The time

using a linear method. To preserve the seasonality, the

window consists of 10, 20, 33 and 50 years for the 100-

detrending method was applied for each month, respectively.

year period. For example, when the time window is 20 years, the 100-year period is divided into ﬁve 20-year periods

Calculation of difference in bias

(1901–1920, 1921–1940, 1941–1960, 1961–1980 and 1981– 2000). The ﬁrst period of each time window (e.g. 1901–

After detrending, the variation of observed and GCM-simu-

1910 for 10-year window) is used as the baseline period.

lated precipitation and temperature within the 1901–2000

For each decadal or multi-decadal period, the biases of

period is considered to be only attributed to the natural

GCM-simulated monthly and annual precipitation (BP)

climate variability. Since natural climate variability is

and temperature (BT) relative to observations are calculated

inherently complex and manifests itself over various tem-

using Equations (1) and (2), respectively:

poral and spatial scales, this study only investigates its impacts on bias correcting climate model outputs over dec-

BPi ¼ (Pmod,i Pobs,i )=Pobs,i

(1)

BTi ¼ Tmod,i Tobs,i

(2)

adal and multi-decadal temporal and watershed spatial scales. In order to investigate the impacts of natural climate

Y. Hui et al.

930

Impacts of bias nonstationarity on hydrology

Hydrology Research

51.5

2020

where i indicates the number of decadal or multi-decadal

where DBmax and DBmin indicate the maximum and mini-

historical periods, Pmod and Pobs indicate the mean values

mum values of differences in bias between historical

of simulated and observed precipitation over the period,

periods and the baseline period, respectively. The outliers

and Tmod and Tobs indicate the mean values of simulated

were removed when calculating the RB to ensure the rea-

and observed temperature.

sonability of estimation. The outliers are detected if they

The differences in bias of precipitation (DBP) and tempera-

are larger than Q75 þ 1.5 × (Q75 Q25) or smaller than

ture (DBT) between the baseline period and all other historical

Q25 1.5 × (Q75 Q25), where Q75 and Q25 indicate the

periods are then calculated for each GCM. The use of the differ-

25th and 75th percentiles, respectively. It is worth noting

ence in bias is beneﬁcial to compare with the results from

that when the time window length is 50 years, the RB is

different GCMs. The differences in bias represent the possible

just equal to the absolute value of difference in bias between

changes of GCM bias due to natural climate variability at dec-

the 1951–2000 period and the 1901–1950 baseline period.

adal and multi-decadal scales. The range of difference in biases

Meanwhile, the CCS of each future period is calculated

is used to judge the bias stationarity of climate model outputs.

relative to the historical baseline period. The 100-year

The impacts of natural climate variability on bias nonstationar-

period from 2001 to 2100 is also divided into several deca-

ity is investigated by using multiple time windows. The main

dal or multi-decadal periods as the same time window as

steps are illustrated in Figure 2 for a 20-year time window.

the historical period. By doing this, the SNR can be calculated for each time scale. The climate change signal for precipitation (CCSP) and temperature (CCST) was calcu-

Calculation of climate change signals relative to

lated using Equations (4) and (5):

differences in bias To identify the role of bias nonstationarity in future climate change, the signal-to-noise ratio (SNR) is used as a criterion

CCSPj ¼ (Pmod,j Pmod,b )=Pmod,b

(4)

CCSTj ¼ Tmod,j Tmod,b

(5)

to quantify the extent of bias nonstationarity relative to future climate change. The SNR is deﬁned as the ratio of cli-

where j indicates the number of decadal or multi-decadal

mate change signal (CCS) of future periods and the range of

periods from 2001 to 2100, and subscript b indicates the

bias (RB) among multiple historical periods.

baseline period (e.g. 1901–1910, 1901–1920, 1901–1933 or

RB is calculated based on the differences in bias among multiple periods using Equation (3): RB ¼

1901–1950). These precipitation and temperature time series without detrending were used to retain climate change tendency.

max (jDBmax j, jDBmin j) if DBmax DBmin > 0 if DBmax DBmin 0 jDBmax DBmin j

(3)

Then, SNR is calculated as the absolute ratio of CCS and RB for both monthly and annual precipitation and temperature. If the CCS is relatively smaller than the RB, e.g. the SNR being smaller than 1, the RB is considered to have large impacts on future climate changes and their hydrological impacts and vice versa.

Hydrological modeling The hydrological simulations were carried out using a Figure 2

Illustration of the calculation of the differences in bias between four multiple

lumped hydrological model named as the Two-Parameter

time periods and the baseline period. Precipitation at the 20-year time window

Monthly Water Balance Model (Xiong & Guo ). It is

is used as an example. BP indicates bias between GCM-simulated and observed precipitation, and DBP indicates the difference in bias between

a simple and monthly lumped rainfall-runoff model with

future 20-year period and the 1901–1920 baseline period.

only two physical parameters. The ﬁrst parameter is c,

Y. Hui et al.

931

Impacts of bias nonstationarity on hydrology

Hydrology Research

51.5

2020

which represents a coefﬁcient to take account of the effect of

periods for the Hanjiang watershed, indicating a good

the time scale change. The second parameter is SC, which

performance.

represents the ﬁeld capacity of a watershed. The monthly

Using the calibrated hydrological model, the streamﬂow time series are generated for each time window using the

runoff (Q) was simulated using Equations (6) and (7):

detrended observed and GCM-simulated precipitation and Et ¼ c × EPt × tanh (Pt =EPt )

(6)

Qt ¼ (St 1 þ Pt Et ) × tanh [(St 1 þ Pt Et )=SC]

(7)

temperature for the historical period. Meanwhile, the streamﬂow time series are also generated for each time window of the future periods. The propagation of precipitation and temperature bias nonstationarity in hydrology was investigated using the same methods as presented above under

where t indicates the number of months, E and EP indicate

‘Calculation of difference in bias’ and ‘Calculation of climate

the monthly actual and potential evapotranspiration,

change signals relative to differences in bias’.

respectively, P indicates the monthly areal precipitation, and S indicates the water content in the soil. The monthly potential

evapotranspiration

was

calculated

the

RESULTS

Thornthwaite method (Thornthwaite ; Xu & Singh ), using the monthly mean temperature. This model

Bias nonstationarity of precipitation and temperature

has been tested and proved to be highly efﬁcient in several watersheds with the monsoon-rainfall-dominated climate

Figure 3 presents the difference in bias of mean annual pre-

in the south of China (Xiong & Guo ; Guo et al. ;

cipitation between each historical period and the baseline

Chen et al. ; Bai et al. ).

period for four different time window lengths: 10, 20, 33

The observed monthly streamﬂow was used to model calibration

(1981–2000).

resent differences in bias of all historical periods relative to

Model calibration was done automatically using shufﬂed

the baseline period for all 17 GCMs, and the thick line is

complex evolution-University of Arizona (SCE-UA) algor-

their median value. The x axis represents the midpoints of

ithm (Duan et al. ). The optimal combination of

the time window. The results show that the difference in

parameters was chosen on the basis of the Nash–Sutcliffe

bias of mean annual precipitation varies with time over his-

efﬁciency (NSE) coefﬁcient (Nash & Sutcliffe ). The

torical periods, indicating that the bias of precipitation is

values of NSE are 0.79 for both calibration and validation

not stationary. The difference in bias reﬂects the impacts of

Figure 3

(1961–1980)

and

validation

and 50 years. For each time window length, the dash lines rep-

Difference in bias of mean annual precipitation between each historical period and the baseline period for 10-, 20-, 33- and 50-year windows. The dashed lines are for 17 GCMs and the thick line is their median value.

932

Y. Hui et al.

Impacts of bias nonstationarity on hydrology

Hydrology Research

51.5

2020

natural climate variability on bias nonstationarity, because

particular, the magnitude and uncertainty of differences in

the trend of precipitation change was removed. The results

bias are larger for short than long time windows. For

in Figure 3 show that the natural climate variability has

example, the median value of difference in temperature

great impacts on bias nonstationarity for mean annual pre-

bias changes between –0.7 and –0.05 C for the 10-year

cipitation, especially for the short time window. However,

window, between –0.5 and –0.05 C for the 20-year

this problem becomes less important as the time window

window, between –0.28 and –0.06 C for the 33-year

gets longer, as indicated by the fact that the difference in

window, and it is –0.13 C for the 50-year window. The nega-

bias decreases with the extension of the time window. For

tive values of the difference in bias indicate the bias over the

example, in terms of the median value over all GCMs, the

ﬁrst baseline period is larger than that over the following

difference in bias changes between –11.5 and 3.8% for the

several historical periods. The large uncertainty related to

10-year window, between 0.5 and 5.5% for the 20-year

climate models is observed, especially for the short time

window, between –2.7 and –0.4% for the 33-year window,

window. Since the temperature time series were detrended

and it is –4.4% for the 50-year window. This indicates that

before calculating the difference in bias, the variation

the decadal variability has more impacts on bias nonstatio-

resulted from natural climate variability. Even though the

narity than multi-decadal variability. Not only the median

bias varies with time due to natural climate variability, its

of difference in bias but also the uncertainty (the difference

impacts on future climate change need to be investigated

between maximum and minimum values over the whole

by comparing future climate change signals.

period) related to GCMs decreases with the longer time window. The uncertainty of difference in bias is 56.4% for

Impacts of bias nonstationarity on future climate

the 10-year window, 27% for the 20-year window, 17.5% for

changes

the 33-year window, and 15.8% for the 50-year window. Therefore, using a long reference period to ﬁlter out low fre-

In order to investigate the role of bias nonstationarity in

quency modes of variability may lessen the impacts of

future climate change, the SNR of mean annual precipi-

natural climate variability on bias correction methods. How-

tation and temperature is calculated for all four time

ever, if just looking at a single climate model, the difference in

windows. Figure 5 presents the SNR of mean annual precipi-

bias can reach 10%, even for the 50-year time window.

tation and temperature over multiple time periods from

Figure 4 presents the difference in bias of mean annual

2001 to 2100 for 10-, 20-, 33- and 50-year windows. Each

temperature between each historical period and the baseline

boxplot depicts the distribution of SNR from an ensemble

period for four different time windows. The results show

of 17 GCMs. The larger SNR values suggest that the bias

that the differences in temperature bias vary with time. In

range is relatively less important compared with the climate

Figure 4

Difference in bias of mean annual temperature between each historical period and the baseline period for 10-, 20-, 33- and 50-year windows. The dashed lines are for 17 GCMs and the thick line is their median value.

933

Figure 5

Y. Hui et al.

Impacts of bias nonstationarity on hydrology

Hydrology Research

51.5

2020

Signal-to-noise ratio (SNR) of (a) mean annual precipitation and (b) temperature from 2001 to 2100 for 10-, 20-, 33- and 50-year windows. The black dashed line indicates the value of SNR is equal to 1.

change signal and vice versa. The unity of SNR means that

length of time window. However, the median values of

the range of bias is equal to the climate change signal. In

SNR are still less than one, even for the 33- and 50-year

this case, the calculated climate change signal may just be

time window. The above results indicate that great attention

the difference in bias between two periods. Figure 5(a)

should be paid to natural climate variability when investi-

shows that SNR values of precipitation are consistently

gating the change of precipitation, in the face of the

less than one for the 10-year window of all future periods.

increase in the greenhouse gas emission scenario.

In this case, it is hard to judge the existence of climate

However, different results are observed for annual mean

change signal. The calculated climate change signal may

temperature. Even though the difference in bias varies with

be just a part of the range of bias caused by the natural cli-

time, the anthropogenic climate change is signiﬁcantly

mate variability, especially taking into account the fact

greater than the bias nonstationarity for annual mean temp-

that the SNR values are mostly constant over time. With

erature, especially for the future periods (Figure 5(b)). In

the increase in the length of time window, the bias nonsta-

other words, even though the bias nonstationarity can

tionarity of mean annual precipitation becomes less

affect the detection of the real anthropogenic climate

important, as indicated by some GCMs presenting SNR

change of annual mean temperature, the inﬂuence is limited.

values being greater than one. In particular, the uncertainty

Comparing to future climate change signal, the temperature

related to GCMs becomes larger with the increase in the

bias can be considered as stationary. However, for the ﬁrst

934

Y. Hui et al.

Impacts of bias nonstationarity on hydrology

Hydrology Research

51.5

2020

and second decades, the natural climate variability still plays

January to December. The median value of the SNR was cal-

an important role in anthropogenic climate change or the

culated over 17 GCMs for each period. Generally, the bias of

climate change signal may not exist, as indicated by the rela-

monthly precipitation is nonstationary, while the tempera-

tively small SNR. The SNR values increase with time, which

ture is stationary compared to future climate change. In

is caused by the increase in the anthropogenic climate

other words, the impacts of bias nonstationarity on future

change, as the range of bias is supposed to be constant.

climate changes are large for monthly precipitation, but

Meanwhile, the dispersion of boxplot is gradually ampliﬁed

it is limited for monthly temperature. Similar to annual

with time, resulting from the increase in the GCM uncer-

precipitation

tainty in the far future. Additionally, the magnitude and

nonstationarity on future climate changes lessen for the

dispersion of the SNR become larger with the extension of

longer time window, as indicated by the increase in the

time window. This implies that the impacts of bias nonstatio-

SNR. However, the monthly variability is observed for

narity on future climate changes are important for the long

these impacts. In terms of the monthly precipitation

time window.

(Figure 6(a)), the SNR in the wet season (May–September)

and

temperature,

the

impacts

bias

The SNR of precipitation and temperature is also calcu-

is larger than in that the dry season (October–April),

lated at monthly scale. Figure 6 shows the portrait diagram

especially for the mid and far future, while the SNR is still

of the median value of the SNR for mean monthly precipi-

smaller than one for 10-, 20- and 33-year windows. When

tation and temperature over multiple time periods from

the time window increases to 50 years, the SNR in the wet

2001 to 2100 with 10-, 20-, 33- and 50-year windows,

season becomes larger than 1. This indicates that the

respectively. The y-axis represents the 12 months from

impacts of bias nonstationarity on future climate changes

Figure 6

Medians of signal-to-noise ratio (SNR) of (a) mean monthly precipitation and (b) temperature from 2001 to 2100 for 10-, 20-, 33- and 50-year windows.

935

Y. Hui et al.

Impacts of bias nonstationarity on hydrology

Hydrology Research

51.5

2020

are less important in the wet season than in the dry season.

hydrological simulations. The results show that the impacts

Therefore, the impacts of bias nonstationarity in the dry

of natural climate variability are ampliﬁed from climate

season need to be paid more attention than in the wet

world to hydrological world. For example, the difference

season and annual scale. In terms of the mean monthly

in bias for all GCMs varies between –30 and 20% for

temperature (Figure 6(b)), the SNR from May to November

mean annual precipitation, while that varies between –60

is larger than other months for the mid and far future, and

and 60% for mean annual streamﬂow for the 10-year time

the maximum values of the SNR present between May and

window. This is because of the non-linear process from

August over each decadal or multi-decadal period. In

climate to hydrology. Similar to precipitation and tempera-

addition, the comparison of the SNR between annual and

ture, the extent of the impacts of natural climate variability

monthly temperature shows that the SNR at the monthly

gradually mitigates as the time window gets longer. For

scale is smaller than that at the annual scale, especially for

example, the median value of difference in streamﬂow

short time windows, while for the 50-year window, the

bias changes between –12.3 and 25.6% for the 10-year

SNR between June and October is comparable to or even

window, between 8.5 and 20.2% for the 20-year window,

larger than that at the annual scale. In the near future

between –0.9 and 5.9% for the 33-year window, and it is

(2001–2030), the SNR in each month is comparable to or

–6.2% for the 50-year window. The uncertainty of difference

smaller than one for the short time window, because of

in streamﬂow bias related to GCMs also reduces as the

the weak climate change signal.

extension of the time window. The uncertainty is 121.6% for the 10-year window, 58.2% for the 20-year window,

Propagation of bias nonstationarity in hydrology

35% for the 33-year window and 27.1% for the 50-year window.

The propagation of bias nonstationarity of climate models

The impacts of bias nonstationarity of climate model

simulated precipitation and temperature in hydrology is

outputs on future streamﬂow changes are also evaluated

investigated by running a hydrological model over the Han-

using the SNR as a metric. Figure 8 shows the SNR of

jiang watershed. Figure 7 presents the difference in bias of

mean annual streamﬂow over multiple periods from 2001

mean annual streamﬂow between each historical period

to 2100 for 10-, 20-, 33- and 50-year windows. In the

and the baseline period for 10-, 20-, 33- and 50-year time

future, the magnitude of the streamﬂow change between

windows. The difference in bias between GCM-driven and

the future and baseline period is much smaller than the

observation-driven streamﬂow also varies with time, which

range of bias, as indicated by the small SNR values,

reﬂects the impacts of natural climate variability on

especially for the 10- and 20-year time windows. This

Figure 7

Difference in bias of mean annual streamﬂow between each historical period and the baseline period for 10-, 20-, 33- and 50-year windows. The dashed lines are for 17 GCMs and the thick line is their median value.

936

Figure 8

Y. Hui et al.

Impacts of bias nonstationarity on hydrology

Hydrology Research

51.5

2020

Signal-to-noise ratio (SNR) of mean annual streamﬂow for 10-, 20-, 33- and 50-year windows. The black dashed line indicates the value of SNR is equal to 1.

indicates that the natural climate variability plays an impor-

50-year time window, the median values of SNR are smaller

tant role in future streamﬂow changes, especially for the

than one, indicating that the natural climate variability is

short time window. In other words, previous studies that

still more important than the anthropogenic climate

attribute the streamﬂow change to anthropogenic climate

change for future streamﬂow simulations. These results of

change may not be correct when taking into account the

streamﬂow are similar to those of mean annual precipi-

impacts of climate change. Similar to precipitation and

tation, but the SNR of the former is smaller than the latter,

temperature, the importance of natural variability reduces

due to the larger range of bias and lower streamﬂow change.

with the extension of the time window, as indicated by the

Figure 9 presents the median of SNR of mean monthly

increase in the SNR. For example, the median values of

streamﬂow across 17 GCMs for 10-, 20-, 33- and 50-year

the SNR change between 0.07 and 0.21 for the 10-year

windows. The results show that the bias nonstationarity

window, between 0.09 and 0.3 for the 20-year window,

has large impacts on the change of future monthly stream-

between 0.38 and 0.85 for the 33-year window, and between

ﬂow, even though the impacts mitigate with the extension

0.61 and 1.01 for the 50-year window. The uncertainty

of the time window. For example, the values of SNR are

related to GCMs becomes greater for the longer time

consistently smaller than one for 12 months and all

window. With the exception of the second period of the

future periods in 10- and 20-year windows. This is also

Figure 9

Medians of signal-to-noise ratio (SNR) of mean monthly streamﬂow for 10-, 20-, 33- and 50-year windows.

937

Y. Hui et al.

Impacts of bias nonstationarity on hydrology

Hydrology Research

51.5

2020

the case for most months and future periods in the 33-year

be enough for calibrating a bias correction method for

window. For the 50-year window, the values of SNR

impact studies.

increase to larger than 1 in the wet season, indicating that the anthropogenic climate change may be more impor-

Uncertainty of baseline observations

tant than the bias nonstationarity resulting from natural climate variability for the wet season. In addition, the

In this study, the CRU dataset was used as the baseline to calcu-

impact of bias nonstationarity on future climate change is

late the bias of GCM-simulated precipitation and temperature.

ampliﬁed in hydrological simulations, as indicated by the

The CRU data is gridded data obtained by interpolating gauged

fact that the SNR of streamﬂow is smaller than that of pre-

precipitation and temperature to regular grids. Due to limited

cipitation and temperature.

gauges used for interpolation, the CRU dataset may also be biased. In order to investigate the uncertainty related to the baseline dataset, another dataset produced by the University

DISCUSSION AND CONCLUSION

of Delaware (version 4.01, available at www.esrl.noaa.gov/ psd/data/gridded/data.UDel_AirT_Precip.html), with a spatial

Results summary

resolution of 0.5 from 1901 to 2000, the same as the CRU dataset, is also used to estimate the difference in bias between each

This study ﬁrst investigated the bias nonstationarity of GCM-

historical period and the baseline period. Figure 10 shows the

simulated precipitation and temperature in the context of

difference in bias for mean annual precipitation and tempera-

natural climate variability for multiple time windows. The

ture obtained from CRU and the University of Delaware

role of bias nonstationarity in future climate changes was

(USA) dataset for 10-, 20-, 33- and 50-year windows. Similarly,

estimated by comparing the range of bias in the historical

the difference in bias varies with time for all time windows.

periods to climate change signal in the future periods.

Meanwhile, the range of difference in bias decreases with the

Finally, the propagation of bias nonstationarity in hydrologi-

extension of the time window. Generally, the range of differ-

cal simulations is investigated by running a hydrological

ence in bias is similar for two datasets, even though the

model. The main ﬁndings are as follows: (1) the biases of

magnitude is not exactly the same.

GCM precipitation and temperature vary with time, due to natural climate variability, especially for the short time win-

Uncertainty of greenhouse gas emission scenarios

dows; (2) precipitation bias nonstationarity plays an important role in future precipitation changes at annual

The role of bias nonstationarity on future climate changes may

and monthly scales, while the importance of temperature

rely on greenhouse gas emission scenarios, as different scen-

bias nonstationarity in future temperature change is not sig-

arios predict different future climate change signal. All the

niﬁcant; (3) the bias nonstationarity of climate model

above results are based on the scenario of RCP4.5. In order

outputs is ampliﬁed when driving a hydrological model for

to investigate the uncertainty related to the greenhouse gas

hydrological simulations. In other words, the bias nonstatio-

emission scenario, the climate change signal of mean annual

narity of precipitation and temperature has considerable

precipitation and temperature predicted by RCP4.5 and

impacts on future streamﬂow changes for this speciﬁc water-

RCP8.5 are compared (Figure 11). The results showed that cli-

shed. This implies that the usually predicted hydrological

mate change signals predicted by RCP4.5 and RCP8.5 are

change in the future may be just the result of bias nonstatio-

comparable for precipitation, while the former is much smaller

narity due to natural climate variability. In addition, the

than the latter for temperature in terms of both median value

increase in the length of time window mitigates the impacts

and uncertainty related to GCMs for all time windows. Thus,

of bias nonstationarity on streamﬂow projections. Thus, it

the use of different greenhouse gas emission scenarios would

suggests using a long period (e.g. 50 years) for calibrating a

not change the conclusion that the role of precipitation bias

bias correction method in hydrological climate change

nonstationarity is important in future precipitation change.

impact studies. The climatology of 20 or 30 years may not

However, the impacts of temperature bias nonstationarity

938

Figure 10

Y. Hui et al.

Impacts of bias nonstationarity on hydrology

Hydrology Research

51.5

2020

The difference in bias of (a) mean annual precipitation, (b) temperature obtained from CRU (blue) and the University of Delaware dataset (red) for 10-, 20-, 33- and 50-year windows. The dash lines are for 17 GCMs, and the thick line is their median value. Please refer to the online version of this paper to see this ﬁgure in color: http://dx.doi.10. 2166/nh.2020.254.

may depend on emission scenarios. For a high emission scen-

deviation of bias over historical periods. The median value is

ario (e.g. RCP8.5), the bias nonstationarity will become even

obtained from all 17 GCMs. The results show that the range

less important, while for a low emission scenario (e.g.

of bias is much larger than the standard deviation of bias for

RCP4.5), it may become important for near future periods.

both precipitation and temperature, especially over the short time window. When using the standard deviation of bias

Methods to estimate natural climate variability

instead of the range of bias, the lower values of ‘noise’ can lead to larger SNR, indicating the impacts of bias nonstationar-

For the historical period, the difference in bias between two

ity would be less important. However, the impacts of

periods is caused by natural climate variability, because the

precipitation bias are still great for most GCMs in the near

anthropogenic climate change is pre-removed. Thus, to investi-

and mid future, as indicated by the SNR being smaller than 1.

gate the role of bias nonstationarity in future climate change, the range of bias is deﬁned as ‘noise’, i.e. the difference between

Future work

maximum and minimum values of biases across multiple periods with outliers deleted. An alternative approach using

This study only investigated the bias nonstationarity of precipi-

the standard deviation of bias over multiple periods can also

tation and temperature for the historical period. Thus, the bias

be used to estimate natural climate variability. Table 2 presents

nonstationarity only resulted from natural climate variability.

a comparison of the median of range of bias and standard

However, for a future period, bias nonstationarity resulted

Y. Hui et al.

939

Figure 11

Impacts of bias nonstationarity on hydrology

Hydrology Research

51.5

2020

The climate change signal of (a) mean annual precipitation and (b) temperature under RCP4.5 (blue) and RCP8.5 (red) for 10-, 20-, 33- and 50-year windows. The dashed lines are for 17 GCMs, and the thick line is their median value. Please refer to the online version of this paper to see this ﬁgure in color: http://dx.doi.10.2166/nh.2020.254.

Table 2

A comparison of the median of range of bias and standard deviation of bias over historical periods for 10-, 20-, and 33-year windows

Precipitation (%)

Temperature ( C)

Standard

the atmospheric physics, due to the limitation of inadequate knowledge of climate system, such as cloud and convective precipitation mechanisms, surface albedo feedback and land–atmosphere interactions. Thus different models simulate somewhat different responses to the same external forcing.

Range

deviation

Range

deviation

10-year window

25.63

6.35

0.89

0.28

This may lead to different biases between different climate

20-year window

12.31

5.29

0.57

0.22

models and observations. As the attribution of both natural

33-year window

6.13

3.41

0.38

0.19

climate variability and climate model sensitivity, the perform-

50-year window

4.67

3.31

0.16

0.11

ance of a bias correction method may be more deteriorated in the future climate change than historical period. This will

from both natural climate variability and climate model sensi-

transfer to hydrological simulations, implying that hydrologi-

tivity (Chen et al. ; Velázquez et al. ; Hui et al. ).

cal simulations forced by bias-corrected precipitation and

The climate model sensitivity is more signiﬁcant than natural

temperature may perform even worse when taking into

climate variability, especially for precipitation bias nonstatio-

account the impacts of climate model sensitivity. The impacts

narity (Hui et al. ). Climate models were proposed using

of climate model sensitivity on bias nonstationarity can be

different structures and parameterization schemes to represent

investigated in future studies.

940

Y. Hui et al.

Impacts of bias nonstationarity on hydrology

Due to the limited length of observed data, 100 years data were divided into 10, ﬁve, three and two nonoverlapping periods respectively corresponding to four different window lengths (10, 20, 33 and 50 years). For each time window, the biases over these ﬁnite periods were used to represent the range of biases (RB) caused by natural climate variability. However, small sample size, e.g. two non-overlapping periods for the 50-year window, may lead to large uncertainty of the estimated RB. To investigate the impacts of sample size on the estimation of bias variability, the use of multiple member GCMs may be a solution (Chen & Brissette ). In addition, this study only used the TwoParameter Monthly Water Balance Model to simulate monthly runoff. The hydrological model uncertainty was not considered. All these can be avenues for future studies.

ACKNOWLEDGEMENTS This work was partially supported by the National Natural Science Foundation of China (Grant No. 51779176 and 51539009) and the National Key R&D Program of China (2019YFC0408903). The authors would like to acknowledge the contribution of the World Climate Research Program Working Group on Coupled Modelling, which is responsible for CMIP. We wish to thank the climate modeling groups (listed in Table 1 of this paper) for producing and making available their model output.

DATA AVAILABILITY STATEMENT Data cannot be made publicly available; readers should contact the corresponding author for details.

REFERENCES Arnell, N. W.  Relative effects of multi-decadal climatic variability and changes in the mean and variability of climate due to global warming: future streamﬂows in Britain. J. Hydrol. 270 (3–4), 195–213. Bai, P., Liu, X. M., Liang, K. & Liu, C. M.  Comparison of performance of twelve monthly water balance models in different climatic catchments of China. J. Hydrol. 529, 1030–1040.

Hydrology Research

51.5

2020

Bellenger, H., Guilyardi, E., Leloup, J., Lengaigne, M. & Vialard, J.  ENSO representation in climate models: from CMIP3 to CMIP5. Clim. Dyn. 42 (7–8), 1999–2018. Buser, C. M., Kunsch, H. R., Luthi, D., Wild, M. & Schar, C.  Bayesian multi-model projection of climate: bias assumptions and interannual variability. Clim. Dyn. 33, 849–868. Chen, J. & Brissette, F. P.  Reliability of climate model multimember ensembles in estimating internal precipitation and temperature variability at the multi-decadal scale. Int. J. Climatol. 39, 843–856. Chen, H., Guo, S. L., Xu, C.-Y. & Singh, V. P.  Historical temporal trends of hydro-climatic variables and runoff response to climate variability and their relevance in water resource management in the Hanjiang basin. J. Hydrol. 344 (3–4), 171–184. Chen, J., Brissette, F. P. & Lucas-Picher, P.  Assessing the limits of bias-correcting climate model outputs for climate change impact studies. J. Geophys. Res. Atmos. 120 (3), 1123–1136. Chen, J., Brissette, F. P. & Lucas-Picher, P. a Transferability of optimally-selected climate models in the quantification of climate change impacts on hydrology. Clim. Dyn. 47, 3359–3372. Chen, J., St-Denis, B. G., Brissette, F. P. & Lucas-Picher, P. b Using natural variability as a baseline to evaluate the performance of bias correction methods in hydrological climate change impact studies. J. Hydrometeorol. 17, 2155–2174. Christensen, J. H., Boberg, F., Christensen, O. B. & Lucas-Picher, P.  On the need for bias correction of regional climate change projections of temperature and precipitation. Geophys. Res. Lett. 35, L20709. Dixon, K. W., Lanzante, J. R., Nath, M. J., Hayhoe, K., Stoner, A., Radhakrishnan, A., Balaji, V. & Gaitan, C. F.  Evaluating the stationarity assumption in statistically downscaled climate projections: is past performance an indicator of future results? Clim. Change 135, 395–408. Duan, Q., Sorooshian, S. & Gupta, V. K.  Effective and efficient global optimization for conceptual rain-runoff models. Water Resour. Res. 28 (4), 1015–1031. Ehret, U., Zehe, E., Wulfmeyer, V., Warrach-Sagi, K. & Liebert, J.  HESS opinions ‘Should we apply bias correction to global and regional climate model data?’. Hydrol. Earth Syst. Sci. 16, 3391–3404. Fuentes-Franco, R., Giorgi, F., Coppola, E. & Kucharski, F.  The role of ENSO and PDO in variability of winter precipitation over North America from twenty first century CMIP5 projections. Clim. Dyn. 46 (9–10), 3259–3277. Graham, L. P., Andreasson, J. & Carlsson, B.  Assessing climate change impacts on hydrology from an ensemble of regional climate models, model scales and linking methods – a case study on the Lule River basin. Clim. Change 81, 293–307. Guo, S. L., Wang, J. X., Xiong, L. H., Ying, A. W. & Li, D. F.  A macro-scale and semi-distributed monthly water balance model to predict climate change impacts in China. J. Hydrol. 268 (1–4), 1–15. Gutierrez, J. M., San-Martin, D., Brands, S., Manzanas, R. & Herrera, S.  Reassessing statistical downscaling

941

Y. Hui et al.

Impacts of bias nonstationarity on hydrology

techniques for their robust application under climate change conditions. J. Clim. 26 (1), 171–188. Harris, C. R. U., & Jones, I. C. & D, P.  Climatic Research Unit (CRU) Time-Series (TS) Version 4.01 of High-Resolution Gridded Data of Month-by-Month Variation in Climate (Jan. 1901–Dec. 2016). Centre for Environmental Data Analysis. Oxon, UK. Hui, Y., Chen, J., Xu, C.-Y., Xiong, L. H. & Chen, H.  Bias nonstationarity of global climate model outputs: the role of internal climate variability and climate model sensitivity. Int. J. Climatol. 39 (4), 2278–2294. Hulme, M., Barrow, E. M., Arnell, N. W., Harrison, P. A., Johns, T. C. & Downing, T. E.  Relative impacts of human-induced climate change and natural climate variability. Nature 397 (6721), 688–691. Kendall, M. G.  Rank Correlation Methods. Griffin, London, UK. Li, L., Diallo, I., Xu, C.-Y. & Stordal, F.  Hydrological projections under climate change in the near future by RegCM4 in Southern Africa using a large-scale hydrological model. J. Hydrol. 528, 1–16. Lu, W. & Qin, X. S.  Integrated framework for assessing climate change impact on extreme rainfall and the urban drainage system. Hydrol. Res. 51, 77–89. Mann, H. B.  Nonparametric tests against trend. Econometrica 13, 245–259. Maraun, D.  Nonstationarities of regional climate model biases in European seasonal mean temperature and precipitation sums. Geophys. Res. Lett. 39, L06706. Maraun, D., Wetterhall, F., Ireson, A. M., Chandler, R. E., Kendon, E. J., Widmann, M., Brienen, S., Rust, H. W., Sauter, T., Themeßl, M., Venema, V. K. C., Chun, K. P., Goodess, C. M., Jones, R. G., Onof, C., Vrac, M. & Thiele-Eich, I.  Precipitation downscaling under climate change: recent developments to bridge the gap between dynamical models and the end user. Rev. Geophys. 48 (3), RG3003. Marhaento, H., Booij, M. J. & Hoekstra, A. Y.  Attribution of changes in stream flow to land use change and climate change in a mesoscale tropical catchment in Java, Indonesia. Hydrol. Res. 48 (3–4), 1143–1155. Maurer, E. P., Das, T. & Cayan, D. R.  Errors in climate model daily precipitation and temperature output: time invariance and implications for bias correction. Hydrol. Earth Syst. Sci. 17 (6), 2147–2159. Nahar, J., Johnson, F. & Sharma, A.  Assessing the extent of non-stationary biases in GCMs. J. Hydrol. 549, 148–162. Nash, J. E. & Sutcliffe, J. V.  River flow forecasting through conceptual models. J. Hydrol. 10, 282–290. Olsson, T., Jakkila, J., Veijalainen, N., Backman, L., Kaurola, J. & Vehvilainen, B.  Impacts of climate change on temperature, precipitation and hydrology in Finland-studies using bias corrected Regional Climate Model data. Hydrol. Earth Syst. Sci. 19 (7), 3217–3238. Ouyang, R., Liu, W., Fu, G., Liu, C., Hu, L. & Wang, H.  Linkages between ENSO/PDO signals and precipitation,

Hydrology Research

51.5

2020

streamflow in China during the last 100 years. Hydrol. Earth Syst. Sci. 18 (9), 3651–3661. Polade, S. D., Gershunov, A., Cayan, D. R., Dettinger, M. D. & Pierce, D. W.  Natural climate variability and teleconnections to precipitation over the Pacific-North American region in CMIP3 and CMIP5 models. Geophys. Res. Lett. 40 (10), 2296–2301. Power, S., Casey, T., Folland, C., Colman, A. & Mehta, V.  Inter-decadal modulation of the impact of ENSO on Australia. Clim. Dyn. 15, 319–324. Ragettli, S., Tong, X., Zhang, G., Wang, H., Zhang, P. & Stähli, M.  Climate change impacts on summer flood frequencies in two mountainous catchments in China and Switzerland. Hydrol. Res. https://doi.org/10.2166/nh.2019.118. Ruiz-Barradas, A., Nigam, S. & Kavvada, A.  The Atlantic Multidecadal Oscillation in twentieth century climate simulations: uneven progress from CMIP3 to CMIP5. Clim. Dyn. 41, 3301–3315. Shen, M., Chen, J., Zhuan, M., Chen, H., Xu, C.-Y. & Xiong, L.  Estimating uncertainty and its temporal variation related to global climate models in quantifying climate change impacts on hydrology. J. Hydrol. 556, 10–24. Teutschbein, C. & Seibert, J.  Bias correction of regional climate model simulations for hydrological climate-change impact studies: review and evaluation of different methods. J. Hydrol 456–457, 12–29. Thornthwaite, C. W.  An approach toward a rational classification of climate. Geograph. Rev. 38 (1), 55–94. Van Pelt, S. C., Kabat, P., Ter Maat, H. W., Van den Hurk, B. J. J. M. & Weerts, A. H.  Discharge simulations performed with a hydrological model using bias corrected regional climate model input. Hydrol. Earth Syst. Sci. 13 (12), 2387–2397. Velázquez, J. A., Troin, M., Caya, D. & Brissette, F.  Evaluating the time-invariance hypothesis of climate model bias correction: implications for hydrological impact studies. J. Hydrometeorol. 16 (5), 2013–2026. Wang, Y., Sivandran, G. & Bielicki, J. M.  The stationarity of two statistical downscaling methods for precipitation under different choices of cross-validation periods. Int. J. Climatol. 38 (S1), e330–e348. Xiong, L. H. & Guo, S. L.  A two-parameter monthly water balance model and its application. J. Hydrol. 216 (1–2), 111–123. Xu, C.-Y. & Singh, V. P.  Evaluation and generalization of temperature-based methods for calculating evaporation. Hydrol. Process. 15 (2), 305–319. Zhuan, M. J., Chen, J., Shen, M. X., Xu, C.-Y., Chen, H. & Xiong, L. H.  Timing of human-induced climate change emergence from internal climate variability for hydrological impact studies. Hydrol. Res. 49 (2), 421–437. Zhuan, M. J., Chen, J., Xu, C.-Y., Zhao, C., Xiong, L. H. & Liu, P.  A method for investigating the relative importance of three components in overall uncertainty of climate projections. Int. J. Climatol. 39, 1853–1871.

First received 2 May 2020; accepted in revised form 25 August 2020. Available online 7 October 2020

942

51.5

2020

Drought prediction models driven by meteorological and remote sensing data in Guanzhong Area, China Jianzhu Li, Siyao Zhang, Lingmei Huang, Ting Zhang and Ping Feng

ABSTRACT Drought is an important factor that limits economic and social development due to its frequent occurrence and profound influence. Therefore, it is of great significance to make accurate predictions of drought for early warning and disaster alleviation. In this paper, SPEI-1 was confirmed to classify drought grades in the Guanzhong Area, and the autoregressive integrated moving average (ARIMA), random forest (RF) and support vector machine (SVM) model were established. Meteorological data and remote sensing data were used to derive the prediction models. The results showed the following. (1) The SVM model performed the best when the models were developed using meteorological data, remote sensing data and a combination of meteorological and remote sensing data, but the model’s corresponding kernel functions are different and include linear, polynomial and Gaussian radial basis kernel functions, respectively. (2) The RF model driven by the remote sensing data and the SVM model driven by the combined meteorological and remote sensing data were found to perform better than the model driven by the corresponding other data in the Guanzhong Area. It is difficult to accurately measure drought with the single meteorological data. Only by considering the combined factors can we more accurately monitor and predict drought. This study can provide an important scientific basis for regional drought warnings and predictions. Key words

| drought prediction, integrated autoregressive moving average model, random forest model, standardized precipitation evaporation index, support vector machine model

HIGHLIGHTS

• •

SPEI-1 was used to analyze the temporal distribution characteristics of drought and the main driving factors in Guanzhong Area, China. Drought grades were selected as the dependent variable, and the meteorological, geographical and vegetative factors were selected as the independent variables to establish an autoregressive integrated moving average (ARIMA) model, random forest (RF) model and support

• •

vector machine model. Meteorological data and remote sensing data were used as independent variables to derive prediction models, respectively. Comparing the models driven by remote sensing data only and the combination of meteorological and remote sensing data, the RF model driven by the remote sensing data and the SVM model driven by the combined meteorological and remote sensing data were found to

•

perform better than the model driven by the corresponding other data in Guanzhong Area. This study can provide an important scientiﬁc basis for regional drought warning and prediction.

Jianzhu Li Siyao Zhang Ting Zhang (corresponding author) Ping Feng State Key Laboratory of Hydraulic Engineering Simulation and Safety, Tianjin University, Tianjin, China E-mail: zhangting_hydro@tju.edu.cn Lingmei Huang Faculty of Water Resources and Hydroelectric Engineering, Xi’an University of Technology, Xi’an, China

943

J. Li et al.

Drought prediction by meteorological and remote sensing data

Hydrology Research

51.5

2020

INTRODUCTION As a severe natural disaster, drought not only affects

shortages in crops and has shown better performances in

economic development but also inﬂuences water resources,

capturing spatial and temporal characteristics (Yuan ).

agriculture, ecology and environments (Mazdiyasni &

Based on SPI and PDSI concepts, Vicente-Serrano et al.

Aghakouchak ; Malik et al. ; Zhang et al. a,

() proposed a standardized precipitation evaporation

b; Guo et al. ). Due to the uncertainty in when

index (SPEI) (Yuan & Zhou ; Vicente-Serrano et al.

droughts begin and end, it is very difﬁcult to predict drought.

), which can express the dry and wet conditions of the

Therefore, drought has become one of the critical factors

land surface on multiple scales with the potential evapor-

limiting the sustainable development of the economy and

ation included. Zhuang et al. () and Wang & Chen

society in many areas (Yuan & Zhou ; Zhang et al.

() investigated the application of the SPEI in China

a, b; Dai et al. ). As it increases in severity,

and found that this index has good applicability in China

drought has a profound impact on biogeochemical pro-

and can accurately represent the occurrence of drought.

cesses in terrestrial ecosystems (Fang et al. ). Drought

Hernandez & Annette () applied SPI and SPEI in con-

also affects the water absorption of vegetation by affecting

junction with precipitation and temperature projections

soil water content and increases the sensitivity of vegetation

from two general circulation models at six major urban cen-

to water energy, which plays an important role in terrestrial

ters of south Texas spanning ﬁve climatic zones. Both the

water, energy and carbon cycles (Fang et al. ; Yinglan

models predicted a progressively increasing aridity in the

et al. a, b). A warming climate is expected to perturb

region throughout the 21st century. Li et al. () investi-

the hydrological cycle, resulting in changes in both the fre-

gated the spatiotemporal characteristics of drought in the

quency and duration of drought (Han et al. ). To

Weihe River Basin by employing the SPEI index. Vega

alleviate drought effects, it is of great signiﬁcance to

et al. () investigated hydrological patterns in the Brazi-

strengthen the study of regional drought characteristics

lian rainforest through a 9-month SPEI series and

and to make more accurate drought predictions for early

determined the Hurst exponents from detrended time

warnings and disaster mitigation (Miao ).

series of days with precipitation and accumulated monthly

The drought index is a primary factor in drought predic-

rainfall. The researchers found that the Hurst exponent cor-

tion, and many indices were presented in the previous

related positively with the monthly mean rainfall. The SPEI

studies for different drought types. The standardized precipi-

not only considers temperature and precipitation but also

tation index (abbreviated as SPI, see McKee et al. ;

evapotranspiration (Lu ), and it has been conﬁrmed to

Bonaccorso et al. ) and the Palmer drought severity

be applicable and better than other drought indices in the

index (abbreviated as PDSI, refer to Palmer ) are two

Guanzhong Area (Xu ). Therefore, the SPEI was

common indicators used to characterize regional drought.

selected as the drought index to evaluate the drought

The SPI was proposed by McKee et al. () and has

events in the Guanzhong Area in this paper.

been widely used as an indicator of drought. This index

The traditional drought index is usually based on hydro-

only requires long-term (generally more than 30 years)

meteorological data measured at stations, and the spatial

precipitation data (Che & Li ; Li et al. ), but it

resolution of the index does not necessarily meet the

cannot reﬂect the seasonal distribution characteristics of

requirements of drought monitoring in large-scale regions.

precipitation (Dong & Xie ). The PDSI considers the

Meteorological satellites, which are widely used in drought

balance of water resources, including precipitation and

remote sensing monitoring (Di et al. ; Sahoo et al.

evaporation processes, additional runoff values, soil moist-

), can acquire multitemporal, multispectral, continuous

ure content and other conditions (Zargar et al. ), and

and complete data (Kogan ; Yilmaz et al. ). At pre-

the PDSI has been extensively used to monitor long-term

sent, the most commonly used and better-performing remote

drought (Liu et al. ). This index also considers water

sensing drought index is the normalized vegetation index

944

J. Li et al.

Drought prediction by meteorological and remote sensing data

Hydrology Research

51.5

2020

(NDVI) (Farrar et al. ; Bannari et al. ; Sun ),

The main aims of this study are: (1) to analyze the tem-

which can effectively monitor the dynamic changes in veg-

poral distribution characteristics of drought and the main

etation cover (Kogan ; Chen et al. ). Ichii et al.

driving factors in the Guanzhong Area, China; (2) to build

() suggested that there was a relative difference in the

an ARIMA model, RF model and SVM model, and predict

correlation between NDVI and natural meteorological fac-

drought grades by using gauge-based and remote sensing

tors at different geographical latitudes and that NDVI was

monitoring data, respectively; and (3) to select the best pre-

signiﬁcantly correlated with temperature as well as regional

diction method and the corresponding predictors.

precipitation. Yan et al. () studied the relationship between NDVI and climate data in coastal areas of the Jiangsu Province and conﬁrmed that both temperature and

STUDY AREA AND DATA

precipitation showed a signiﬁcantly positive correlation with NDVI data in the region.

Study area

To reduce the losses caused by drought, drought prediction is needed on the basis of real-time drought monitoring.

The Guanzhong Area is located in the middle of the

Commonly used prediction methods include time series

Shaanxi Province, between the Loess Plateau and the

analysis, artiﬁcial neural networks (Moody & Darken

Qinling Mountains, China. The geographical coordinates

), support vector machines (SVMs) (Ahmad et al. ;

are 33 340 N–35 520 N, 106 180 E–110 380 E, with an area of

Weng ) and random forests (RFs) (Breiman ;

36,000 km2. The administrative divisions in the Guanzhong

Gislason et al. ; Cong ). The autoregressive inte-

Area include the cities of Xi’an, Xianyang, Weinan, Baoji

grated moving average (ARIMA) (Box & Jenkins ;

and Tongchuan (Wei et al. ), and this area is an impor-

Shumway & Stoffer ) model proposed by Box & Jenkins

tant area connecting the eastern and western parts and

() is a common model used in time series analysis (Han

the northern and southern parts of China (Qiao ).

et al. ). Yurekli et al. () used the ARIMA model to

Additionally, this area is a key construction area of the

simulate the 5-year monthly runoff observation data of the

Belt and Road Initiative.

Kelkit River (Bai et al. ). Zhang et al. () used the

The topography and geomorphology of the Guanzhong

ARIMA model to predict the drought of the northern

Area are complex with elevations between 270 and

Haihe River in China, indicating that the ARIMA model

2,439 m. The altitudes are high in the western, northern

has a good prediction accuracy. Machine learning has

and southern regions and low in the eastern and middle

been widely used to predict drought. For example, Fan

areas (Figure 1). The topography descends in steps from

et al. () established a drought prediction model in

the mountainous area to the center of the basin and in

autumn in the Zhejiang Province based on the SVM

turn includes the piedmont alluvial plain, the loess plateau

method with a radial basis kernel function and a cross-vali-

and the terrace of the river valley (Qiao ).

dation approach. The optimal model parameters were then

The Guanzhong Area is located in the transitional zone

determined, and they concluded that the developed model

between arid and humid regions (Li et al. ), which

has high prediction accuracy (Fan et al. ). Wu et al.

belongs to a continental monsoon climate with cold winters,

() employed the RF model to analyze the drought

hot summers and distinct seasonal features. During the same

grades of 21 representative stations from 1962 to 2012 in

period of heat and rain, drought is more likely to occur in

the Huaihe River Basin. The researchers found that the over-

the Guanzhong Area. The average annual precipitation is

all average prediction accuracy is higher than the weather

500–700 mm, with precipitation concentrated mostly in

system’s weather prediction accuracy, suggesting acceptable

summer and autumn and little precipitation occurs in

prediction results. ARIMA, RF and SVM have been used in

winter (Qiao ).

drought prediction, but the performance of these three

The special topography results in climatic conditions of

methods needs to be further explored to improve the accu-

high temperature and low rainfall. The area is mostly plain

racy of drought prediction in the Guanzhong Area.

formed by loess sedimentation and river alluvial sediments

945

Figure 1

J. Li et al.

Drought prediction by meteorological and remote sensing data

Hydrology Research

51.5

2020

Location of the Guanzhong Area and the meteorological stations.

1976–1980

and

1994–1997

(Gao

).

with soft soil, poor water retention capacity, and therefore,

1971–1972,

the area is prone to drought. In terms of agriculture,

Droughts in the Guanzhong Area occurred mostly in

approximately 70% of the land on the Guanzhong Plain

spring and summer, and continuous drought events fre-

is covered by cropland, which mainly includes grain

quently occurred, lasting from spring to summer and from

crops, fruit woodlands and vegetables. More than 50% of

summer to autumn (Wu ).

the croplands in this area are rainfed. There are many irrigated croplands in the western and middle parts, while

Data

most of the croplands in the east are rainfed (Zhou et al. ). Due to the dense population and developed agricul-

The data used for drought prediction in the Guanzhong

ture, the increasing industrial and agricultural water

Area include gauge-based meteorological data and remote

supply led to a more severe drought situation. Thus, the

sensing data.

Guanzhong Area is known by the saying, ‘9 drought years out of 10 years’ (Yu ).

The gauge-based meteorological data include the daily wind speed, precipitation, air temperature, air pressure, sun-

According to the ‘Record of Natural Disasters in the

shine hours and relative humidity of 11 meteorological

Shaanxi Province’, from the 2nd century to 1949 AD,

stations in the Guanzhong Area, covering the period from

there were more than 600 drought events in the Shaanxi

January 1960 to December 2016. The data were downloaded

Province, and 326 in the Guanzhong Area, accounting

from the Meteorological Data Sharing Service System of the

for 54% of the total drought records (Meteorological

China Meteorological Administration (http://data.cma.cn/).

Station of Shaanxi Meteorological Bureau ). There

The remote sensing data are the land surface tempera-

were 22 large-scale drought events in the Guanzhong

ture of the Guanzhong Area and the NDVI from July 2002

Area from 1949 to present, especially the continuous

to June 2011. These data were downloaded from the Inter-

severe droughts that occurred in 1959–1961, 1966–1967,

national Scientiﬁc Data Mirror website of the Chinese

J. Li et al.

946

Drought prediction by meteorological and remote sensing data

Academy of Sciences Computer Network Information Center

(http://www.gscloud.cn).

The

digital

elevation

model (DEM) was also obtained from this website. Detailed information on the remote sensing data used in this study is shown in Table 1.

METHODS

Hydrology Research

Table 2

51.5

2020

The classiﬁcation of drought based on SPEI

SPEI

Drought grades

Number

( 0.5,0)

no drought

( 1, 0.5]

light drought

( 1.5, 1]

moderate drought

( 2, 1.5]

heavy drought

∞, 2]

severe drought

Calculation of the SPEI-1 series

Predictive models

The SPEI index is calculated as follows:

ARIMA model

(1) Calculate the monthly water surplus and deﬁciency Di, Di ¼ Pi – PETi, with i indicating the month, Pi indicating

The ARIMA model is one of the most commonly used

the monthly precipitation (mm) and PETi the monthly

models in time series analyses (Zhao et al. ). The data

potential evaporation (mm), which is calculated using

series are formed by the prediction index with time regarded

the Morton-type Penman formula (Mei ).

as a random sequence. The dependence relation of this

(2) Calculate the probability distribution of the monthly

random series reﬂects the continuity of the original data in

water surplus and deﬁcit Di by using the log–logistic

time, which has both the inﬂuence of external factors and

probability distribution function. (3) Standardize the log–logistic probability distribution function F(x) for monthly accumulated water surplus and deﬁcit Di (Li et al. ).

its own change laws (Wu ). The ARIMA modeling steps are as follows: (1) Determine if the time series is stationary. The time series can be differentiated to obtain a stationary series if it is a

The severity of drought can be graded according to the value of SPEI, as shown in Table 2 (Zhang et al. ).

non-stationary series. The difference operators can be explained as each observation minus the previous one,

According to the multi time scale characteristics of the

in which d represents the difference times. The autore-

SPEI, the SPEI on different time scales can reﬂect changes

gressive moving average (ARMA) model can be

in humidity and dryness in different periods (Yang et al.

described as ARMA (p,q), and the ARIMA model can

). SPEI-1 refers to the SPEI index on a monthly time

be described as ARIMA (p,d,q). When the time series

scale. The trend of SPEI values on different time scales is

is stationary, d ¼ 0, and the ARIMA model becomes

consistent overall, but the trend of humidity and dryness reﬂected by SPEI values on short time scales is more speciﬁc. Therefore, SPEI-1 is selected in this paper.

the ARMA model (Ghashghaie & Nozari ). (2) Determine p and q (p represents the lag order of the autoregressive processes, and q represents the lag order of the moving average processes (Ghashghaie &

Table 1

Nozari )). The autocorrelation function (ACF) Detailed information on the remote sensing data

graph and the partial ACF (PACF) graph are drawn, Spatial

Temporal

and the ARMA(p,q) models are judged based on the tail-

Production name

resolution

ing and truncation of the ACF and PACF graph (Wu

China 1 km land temperature monthly synthetic products (MYDLT1M)

1 km

month

).

China 500 m NDVI synthetic products (MYDND1M)

500 m

month

30 m resolution DEM data (GDEMV2)

30 m

–

(3) Fit the model and perform the normality test, autocorrelation test and white noise test on the residuals. (4) Use the ARIMA model to inspect and predict the monitoring series. For the seasonal non-stationary time series,

947

J. Li et al.

Drought prediction by meteorological and remote sensing data

Hydrology Research

51.5

2020

seasonal differences are also needed to obtain the

The SVM was originally proposed for the problem of

stationary time series and form the ARIMA (p,d,q)

binary pattern classiﬁcation in the case of linear separability.

(P, D,Q)S model, where D represents the order of seaso-

Given a set of observation samples S ¼ {(x1, y1), (x2, y2)…}

nal differences; P and Q represent the order of seasonal

⊂X × { 1, 1}, X⊂Rn is called the input space or the input

autoregression and the order of seasonal moving

characteristic space, and yi∈{ 1,1} is the sample class tag.

averages, respectively; and S represents the length of

The purpose of classification is to find a classification hyper-

the seasonal period.

plane that completely separates the two classes. Let G ¼ {ω x þ b ¼ 0| ω ∈ Rn, x∈X, b∈R} be all hyperplane sets

RF model

that can completely and correctly classify S. In all hyperplanes, the maximum interval classiﬁer is looking for an

The RF model uses the bootstrap resampling method to

optimal hyperplane that satisﬁes the two types of classiﬁ-

extract multiple samples from the original samples, builds

cation intervals (the sum of the sample-to-hyperplane

the decision tree for each bootstrap sample, and then com-

distance closest to each other from the hyperplane). The

bines the predictions of multiple decision trees to obtain

linear inseparable problem of the input space can be trans-

the ﬁnal prediction results by voting (Breiman ; Fang

formed into a linear separable problem by using the

et al. ). Specifically, the RF model is a combined classification model composed of many decision tree classification models (h(X,θi), i ¼ 1, 2, …, k), and the parameter set (θi) is an independent and identically distributed random vector. Given the independent variable X, each decision tree classi-

appropriate kernel functions to operate in two classes. Common kernel functions are linear kernel functions (Altman ), polynomial kernel functions, Gaussian radial basis kernel functions and sigmoid kernel functions (Wang ; Burges ).

fication model selects the optimal classification results by one-vote voting rights (Wang ). The modeling steps for the RF model are as follows: (1) Randomly extract M samples from the fitting data set T, and then fit the extracted data set. (2) For each instance, generate a decision tree, and at each node of the tree: a. randomly extract a subset of m variables from the p valid overall features; b. choose the best variables and the best partition from the set of m variables; and c. continue until the tree is fully generated (Wang ). (3) Perform RF predictions using all trees.

RESULTS AND DISCUSSION Calculation of the SPEI-1 series The SPEI-1 series of 11 meteorological stations in the Guanzhong Area are obtained based on monthly meteorological data. Here, we show only the results of Baoji Station in Figure 2. The SPEI-1 series of the Guanzhong Area was calculated according to the SPEI-1 series at the 11 stations, as shown in Figure 3. The Guanzhong Area is prone to drought disasters because of the frequent alternation of dry and wet conditions. In 1962, 1963, 1967, 1969, 1976, 1979, 1994– 2002 and 2007, the Guanzhong Area suffered severe

SVM model

drought events as shown in Figure 3 based on the drought classiﬁcation in Table 2.

The SVM model seeks the best compromise between the

The occurrence frequency and percentage of drought at

complexity of the model (learning-intensive reading of

all grades in 12 months at Baoji Station are acquired and

speciﬁc training samples) and the ability to learn (the ability

shown in Table 3. Moderate drought mainly occurs in

to identify any samples without error) based on limited

autumn, while the occurrence of heavy drought is concen-

sample information for better promotion ability (Wang

trated in spring and summer. In addition, Baoji Station

).

experienced no severe drought during 1960–2016.

J. Li et al.

948

Drought prediction by meteorological and remote sensing data

Figure 2

SPEI-1 series of the Baoji Station.

Figure 3

SPEI-1 series of the Guanzhong Area.

Table 3

Hydrology Research

51.5

2020

The number and proportion of droughts at all grades in 12 months at Baoji Station

Moderate drought

Heavy drought

Severe drought

12.26%

9.21%

4.76%

–

5.66%

6.58%

9.52%

–

10.38%

6.58%

9.52%

–

7.71%

11.32%

7.89%

7.14%

–

8.81%

5.66%

6.58%

11.90%

–

8.59%

6.60%

7.89%

9.52%

–

8.15%

9.43%

6.58%

11.90%

–

8.81%

5.66%

9.21%

9.52%

–

9.03%

4.72%

7.89%

9.52%

–

8.15%

9.43%

10.53%

4.76%

–

7.93%

8.49%

13.16%

4.76%

–

8.15%

10.38%

7.89%

7.14%

–

Month

No drought

Light drought

7.49%

9.03%

8.15%

6 7

J. Li et al.

949

Drought prediction by meteorological and remote sensing data

In the ‘History of Natural Disasters in the Shaanxi Province’ and ‘Report on China’s Disaster Situation 1949–1995’

Hydrology Research

51.5

2020

in the Guanzhong Area, with a predicted length of 12 months.

(Liu ), the Guanzhong Area suffered ﬂood disasters in the mid-1960s and early 1980s, and among them, 1964,

Drought prediction by the ARIMA model

1983 and 1984 were serious ﬂooding years. There were drought events in the early 1960s, mid-late 1970s and

The SPEI-1 series of the 11 meteorological stations in the

1990s, among which, the continuous drought events that

Guanzhong Area were used for ARIMA modeling and pre-

occurred in 1959–1961, 1966–1967, 1971–1972, 1976–

diction. The ARIMA model was ﬁtted using the SPEI-1

1980, 1994–1997 and 1999–2002 were severe. Drought

series from January 1960 to December 2015 and predicted

and ﬂoods occurred alternately from the mid-1980s to the

drought grades from January to December 2016 with a

early 1990s, and the climate was more humid in the 21st

length of 12 months. The ARIMA model parameters corre-

century (Cai et al. ; Lei et al. ). The wet and dry con-

sponding to each station are shown in Table 5, and the

ditions in Figure 3 are consistent with previous studies.

ﬁtting and prediction results of Baoji Station are shown in

Therefore, the SPEI-1 can reasonably reﬂect the occurrence

Figure 4. The ARIMA model parameters at different stations

of drought in the Guanzhong Area.

have regional heterogeneity, depending on the natural con-

The number and proportion of droughts at all grades were calculated according to the SPEI-1 series at the 11

ditions, such as the underlying surface, geographical location and climatic characteristics.

stations, as shown in Table 4. The proportion of droughts

According to Figure 4(a), the SPEI-1 ﬁtting series of the

at 11 stations is nearly 35%, mainly light and moderate

ARIMA model at each station is mostly consistent with the

droughts, and severe droughts hardly occur. In addition,

measuring series, which shows that the ARIMA model is

the number of droughts at all grades decreases with the

basically in line with reality and that the ﬁtting performs

severity of drought.

well. The prediction SPEI-1 series shown in Figure 4(b) ﬂuctuates in the range of 1 to 1, suggesting that the Guan-

Drought prediction by different models

zhong Area was in a no drought or light drought state in 2016. There is no sudden increase or decrease in SPEI-1

The three models previously mentioned in the ‘Predictive

values in 2016, and the change is relatively small. Among

models’ section were used to predict the drought grades

them, the prediction SPEI declined in 2016. The SPEI-1

Table 4

The number and proportion of droughts at all grades

Station

No drought

Light drought

Moderate drought

Heavy drought

Severe drought

Baoji

454

66.57%

108

15.84%

11.29%

6.30%

0.00%

Fengxiang Luochuan

451

66.13%

106

15.54%

13.20%

5.13%

0.00%

440

64.42%

122

17.86%

13.03%

4.39%

0.29%

Tongchuan

459

67.30%

106

15.54%

11.29%

5.57%

0.29%

Wugong

447

65.54%

108

15.84%

13.64%

4.84%

0.15%

Yaoxian

458

67.06%

110

16.11%

10.69%

6.00%

0.15%

Changwu

456

66.86%

102

14.96%

12.76%

4.99%

0.44%

Pucheng

457

67.01%

105

15.40%

12.32%

4.84%

0.44%

Longxian

461

67.60%

108

15.84%

12.76%

3.37%

0.44%

Yongshou

456

66.86%

14.52%

13.49%

4.99%

0.15%

Qindu

452

66.08%

118

17.25%

11.84%

3.80%

1.02%

average

454

66.49%

108

15.88%

12.39%

4.93%

0.31%

J. Li et al.

950

Table 5

Drought prediction by meteorological and remote sensing data

Hydrology Research

51.5

2020

The ARIMA model parameters corresponding to each station

Station

ARIMA(p,d,q)(P,D,Q)S

Station

ARIMA(p,d,q)(P,D,Q)S

Baoji

ARIMA(5,4,3)(1,0,1)12

Fengxiang

ARIMA(5,4,3)(1,0,2)12

Luochuan

ARIMA(3,1,2)(0,1,0)12

Tongchuan

ARIMA(2,2,3)(0,1,0)12

Wugong

ARIMA(3,1,2)(0,1,0)

Yaoxian

ARIMA(1,1,2)(0,1,0)12

Changwu

ARIMA(0,1,2)(1,1,0)12

Pucheng

ARIMA(1,2,2)(0,1,0)12

Longxian

ARIMA(1,1,2)(0,1,1)

Yongshou

ARIMA(4,3,3)(2,1,0)12

Qindu

ARIMA(1,1,2)(0,1,0)12

Guanzhong Area

ARIMA(0,1,2)(0,1,0)12

Figure 4

Fitting and prediction results of the ARIMA model at Baoji Station. (a) Fitting results of the ARIMA model in Baoji Station. (b) Prediction results of the ARIMA model in Baoji Station.

was lower after August 2016 when the drought was serious.

The ﬁtting and prediction of SPEI values were graded

The SPEI-1 was at a higher level before August 2016, that is,

according to Table 2, and the qualiﬁed rate of the drought

the drought degree was lighter.

grade was calculated. The qualiﬁed ﬁtting and prediction

J. Li et al.

951

Drought prediction by meteorological and remote sensing data

Hydrology Research

51.5

2020

rates of the ARIMA model at each station are shown in

temperature, average humidity, average wind speed, sunshine

Table 6. The drought prediction grades of the ARIMA

hours, rainfall and potential evaporation of the 11 meteorolo-

model can represent the actual situation to some extent,

gical stations in the Guanzhong Area were selected as the

and the average qualiﬁed ﬁtting and prediction rates are

independent variables. The data from January 1960 to

0.5337 and 0.6061, respectively. Except for the qualiﬁed pre-

December 2015 were used as the ﬁtting data, and the data

diction rate of the Longxian station, which is 0.7500 and is

from January 2016 to December 2016 were used as the pre-

signiﬁcantly higher than the average, the qualiﬁed rates of

diction data with a predicted length of 12 months.

other stations ﬂuctuate around the average. The case

The qualiﬁed ﬁtting and prediction rates corresponding

where the qualified fitting rate is lower than the qualified

to the 11 stations of the RF model constructed by the

prediction rate may be due to the uneven length of the ﬁtting

meteorological data were calculated and are shown in

and prediction data sets. The ﬁtting data were used from Jan-

Table 7. The average qualiﬁed ﬁtting and prediction rates

uary 1960 to December 2015 (56 years), and the prediction

are 0.7623 and 0.6515, respectively. The qualiﬁed prediction

data were used from January to December 2016 (1 year), so

rates are signiﬁcantly different from the average qualiﬁed

that the internal variance in the ﬁtting process was much

prediction rates at some stations, such as Baoji, Tongchuan,

larger than the prediction process, that is, the ﬁtting process

Wugong and Yongshou with prediction accuracies of

had a large error with respect to the prediction process.

0.5000, 0.7500, 0.8333 and 0.5000, respectively. The qualified fitting and prediction rates of the other stations fluctuate around the average. Due to the noise in the fitting

Drought prediction by the RF model

data, the overﬁtting phenomenon is obvious in the model. RF model driven by meteorological data and accuracy assessment. SPEI-1 grade (no drought: 1, light drought: 2,

RF model driven by remote sensing data and accuracy

moderate drought: 3, heavy drought: 4, special drought: 5)

assessment. After classifying the SPEI-1 grades of 11 stations

was selected as the dependent variable. The monthly mean

in the Guanzhong Area, kriging spatial interpolation was per-

values of daily maximum temperature, daily minimum

formed to obtain the SPEI-1 spatial distribution. The DEM,

Table 6

Qualiﬁed ﬁtting and prediction rate of the ARIMA model

Station

Fitting

Prediction

Station

Fitting

Prediction

Baoji

0.5417

0.5000

Changwu

0.5372

0.5833

Fengxiang

0.5461

0.6667

Pucheng

0.4940

0.6667

Luochuan

0.4717

0.5000

Longxian

0.6741

0.7500

Tongchuan

0.5134

0.6667

Yongshou

0.5655

0.5833

Wugong

0.4985

0.5833

Qindu

0.5149

0.5833

Yaoxian

0.5134

0.5833

Average

0.5337

0.6061

Table 7

Drought qualiﬁed ﬁtting and prediction rates of the RF models in 11 stations (meteorological data)

Station

Fitting

Prediction

Station

Fitting

Prediction

Baoji

0.7887

0.5000

Changwu

0.7708

0.5833

Fengxiang

0.7723

0.5833

Pucheng

0.7827

0.8333

Luochuan

0.7426

0.5833

Longxian

0.7455

0.6667

Tongchuan

0.7753

0.7500

Yongshou

0.7307

0.5000

Wugong

0.7857

0.8333

Qindu

0.7321

0.6667

Yaoxian

0.7589

0.6667

Average

0.7623

0.6515

J. Li et al.

952

Drought prediction by meteorological and remote sensing data

Hydrology Research

51.5

2020

slope, slope direction, rainfall, NDVI, daytime land tempera-

values of daily maximum temperature, daily minimum temp-

ture (LTD) and nighttime land temperature (LTN) were used

erature, average humidity, average wind speed, sunshine

as independent variables, and SPEI-1 grades were used as the

hours, rainfall and potential evaporation obtained from 11

dependent variables. The RF model was constructed with

meteorological stations in the Guanzhong Area were used

the data from July 2002 to June 2010 as the ﬁtting data and

as the independent variables, as well as DEM, slope, and

the data from July 2010 to June 2011 as the prediction data.

slope direction, rainfall, NDVI, LTD and LTN obtained

The average qualiﬁed ﬁtting and prediction rates of the RF

from remote sensing products. SPEI-1 grades were used as

model constructed by remote sensing data in the Guanzhong

the dependent variable. The data from July 2002 to June

Area are 0.6746 and 0.6836, respectively. The independent

2010 were used for ﬁtting the model, and the data from

variables corresponding to the 11 stations were extracted,

July 2010 to June 2011 were used for prediction. The quali-

and then, the qualiﬁed ﬁtting and prediction rates were calcu-

fied fitting and prediction rates corresponding to the 11 stations are shown in Table 9. The average qualified fitting

lated as shown in Table 8. As seen from Table 6, the qualiﬁed ﬁtting and prediction

and prediction rates are 0.6316 and 0.5388, respectively.

rates of the 11 stations are 0.6913 and 0.5616, respectively.

Except for the qualiﬁed prediction rate of Longxian station

Except for the qualiﬁed ﬁtting rate of Luochuan and the qua-

(0.6667), which is signiﬁcantly higher than the average qua-

liﬁed prediction rate of Qindu, the qualiﬁed rates of other

liﬁed prediction rate, the qualiﬁed rates of the other stations

stations ﬂuctuate around the average qualiﬁed rates. The

ﬂuctuate around the average.

model works well for the 11 stations, but the RF model built by using remote sensing data also presents a signiﬁcant

Drought prediction by the SVM model

overﬁtting problem due to data noise. SVM model driven by meteorological data and accuracy RF model driven by combined meteorological and remote

assessment. By using the same ﬁtting data and prediction

sensing data and accuracy assessment. The monthly mean

data as in the ‘RF model driven by meteorological data

Table 8

Drought qualiﬁed ﬁtting and prediction rates of the RF models at 11 stations (remote sensing data)

Station

Fitting

Prediction

Station

Fitting

Prediction

Baoji

0.6979

0.6563

Changwu

0.6458

0.4792

Fengxiang

0.6667

0.5625

Pucheng

0.7396

0.5313

Luochuan

0.5833

0.4896

Longxian

0.6979

0.6250

Tongchuan

0.7188

0.5000

Yongshou

0.6771

0.4896

Wugong

0.7500

0.6042

Qindu

0.7813

0.6771

Yaoxian

0.6458

0.5625

Average

0.6913

0.5616

Fitting

Prediction

Table 9

Drought qualiﬁed ﬁtting and prediction rates of the RF models in 11 stations (meteorological and remote sensing data)

Station

Fitting

Prediction

Station

Baoji

0.6667

0.6042

Changwu

0.5625

0.4583

Fengxiang

0.5938

0.4688

Pucheng

0.6667

0.5729

Luochuan

0.5625

0.5208

Longxian

0.6875

0.6667

Tongchuan

0.6042

0.4479

Yongshou

0.6354

0.4896

Wugong

0.6875

0.5938

Qindu

0.6875

0.6250

Yaoxian

0.5938

0.4792

Average

0.6316

0.5388

953

J. Li et al.

Drought prediction by meteorological and remote sensing data

and accuracy assessment’ section, four common kernel func-

Hydrology Research

Table 11

51.5

2020

The qualiﬁed rates of four common kernel functions (remote sensing data)

tions (linear kernel function, polynomial kernel function, Kernel

Linear kernel

Polynomial kernel

Gaussian radial basis kernel

Sigmoid kernel

function) were used to construct the SVM model as well

functions

function

Gaussian radial basis kernel function and sigmoid kernel as ﬁt and predict the drought grades. The average qualiﬁed

Fitting

0.7247

0.7795

0.7523

0.6680

ﬁtting and prediction rates at the 11 stations are shown in

Prediction

0.7269

0.7188

0.6548

0.7016

Table 10.

Difference

0.0022

0.0607

0.0975

0.0336

As shown in Table 10, the polynomial kernel function and the linear kernel function perform best in the ﬁtting and prediction, respectively. The Gaussian radial basis

function in the prediction ranks second, and it differs from

kernel function ranks second for both the qualiﬁed ﬁtting

the ﬁrst-ranked linear kernel function by only 0.0081. In

and prediction rates. The qualiﬁed ﬁtting and prediction

addition, the differences in the qualiﬁed ﬁtting and predic-

rates of the sigmoid kernel function differ by 0.0364,

tion rates of the four kernel functions are less than 0.1,

which is the smallest and the most stable among the four

which shows that they are relatively stable. Therefore, the

kernel functions, but the qualiﬁed rates of ﬁtting and predic-

polynomial kernel function is more suitable for drought pre-

tion are the lowest. The qualiﬁed rate of linear kernel

diction with remote sensing data driving the SVM model.

function ﬁtting is 0.7532, which is only 0.0437 smaller than the highest Gaussian radial basis kernel function, and the qualiﬁed rate is the highest in the prediction process.

SVM model driven by combined meteorological and remote

The difference between qualiﬁed ﬁtting and prediction

sensing data and accuracy assessment. Using the same ﬁt-

rates is 0.0498, which is only larger than the sigmoid

ting data and prediction data as in the ‘RF model driven

kernel function (ranks ﬁrst). Therefore, the linear kernel

by combined meteorological and remote sensing data and

function is considered to be more suitable for drought moni-

accuracy assessment’ section, four kernel functions were used to construct the SVM model as well as ﬁt and predict

toring in the Guanzhong Area.

the drought grades. The qualiﬁed ﬁtting and prediction SVM model driven by remote sensing data and accuracy

rates of the four kernel functions are shown in Table 12.

assessment. Using the same ﬁtting data and the prediction

The Gaussian radial basis kernel function and the linear

data shown in the ‘RF model driven by remote sensing

kernel function perform best in the ﬁtting and prediction

data and accuracy assessment’ section, four kernel functions

process, respectively. The Gaussian radial basis kernel func-

were used to construct the SVM model as well as ﬁt and pre-

tion ranks second with a qualiﬁed prediction rate of 0.7652

dict the drought grades. The qualiﬁed ﬁtting and prediction

and only differs from the linear kernel function (ranks ﬁrst)

rates of the four kernel functions are shown in Table 11.

by 0.0075. That is, the Gaussian radial basis kernel function

The polynomial kernel function and the linear kernel

performs well in both ﬁtting and prediction. Therefore, the

function perform best in the ﬁtting and prediction process,

Gaussian radial basis kernel function is more suitable for

respectively. The qualiﬁed rate of the polynomial kernel

drought prediction with the SVM model driven by combined meteorological and remote sensing data.

Table 10

The average qualiﬁed rates of four common kernel functions (meteorological data)

Table 12

The qualiﬁed rates of four common kernel functions (meteorological and remote sensing data)

Linear

Polynomial

Gaussian radial

Sigmoid

Kernel

kernel

basis kernel

kernel

functions

function

Fitting

0.7532

0.7969

0.7676

0.6273

Prediction

0.8030

0.6970

0.7121

Difference

0.0498

0.0999

0.0555

Linear

Polynomial

Gaussian radial

Sigmoid

Kernel functions

kernel function

basis kernel function

kernel function

0.5909

Fitting

0.8532

0.7727

0.9441

0.6799

0.0364

Prediction

0.7727

0.7348

0.7652

0.7273

954

J. Li et al.

Drought prediction by meteorological and remote sensing data

Comparison of the models

Hydrology Research

51.5

2020

The qualiﬁed prediction rates of the Gaussian radial basis kernel function rank second, the differences from the linear

Comparison of different models driven by meteorological data

radial basis kernel function which ranks ﬁrst are small, and the performance is excellent. Therefore, the SVM model (Gaussian radial basis kernel function) driven by combined

Comparing the qualiﬁed ﬁtting and prediction rates of the

meteorological and remote sensing data is more suitable for

ARIMA model, the RF model and the SVM model driven

drought prediction in the Guanzhong Area.

by meteorological data, it is found that among the three

According to the ‘Comparison of different models

models, the SVM model performs best in drought predic-

driven by meteorological data, Comparison of the models

tion, where the polynomial kernel function and linear

driven by remote sensing data, and Comparison of

kernel function perform best in the ﬁtting and prediction

models driven by combined meteorological and remote sen-

process, respectively. The qualiﬁed ﬁtting rate of the linear

sing data’ sections, the SVM model is superior to the other

kernel function is 0.7532, which is only 0.0437 smaller

two models for the following reasons. (1) The SVM model

than the highest Gaussian radial basis kernel function, and

performs better than the RF model when dealing with unba-

the qualiﬁed rate in the prediction is the highest. The differ-

lanced data. Drought grades in the Guanzhong Area are

ence between qualiﬁed ﬁtting and prediction rates is 0.0498,

unbalanced, with light and moderate drought as the main

which is relatively stable. Therefore, it is considered that the

types, and heavy and severe droughts are rare. The samples

SVM model (linear kernel function) is more suitable for

of drought levels are unbalanced. (2) The ARIMA model is

drought monitoring in the Guanzhong Area.

better in terms of linear prediction and less effective in terms of nonlinear prediction. The SVM model can also

Comparison of the models driven by remote sensing data

solve nonlinear problems through the application of kernel functions.

Comparing the qualiﬁed rates of the RF model and SVM model based on remote sensing data, the SVM model per-

Comparison of the model-driven data

forms better in the prediction of drought, in which the polynomial kernel function and linear kernel function per-

For the RF model, except for Longxian station, the qualiﬁed

form best in the ﬁtting and prediction, respectively. The

ﬁtting and prediction rates show that the performance of

polynomial kernel function ranks second in the qualiﬁed

remote sensing data in the RF model performed better than

prediction rate and only differs from the linear kernel func-

the combined meteorological and remote sensing data. The

tion (ranks ﬁrst) by 0.0081, and the difference between the

use of meteorological data will reduce the qualiﬁed ﬁtting

qualiﬁed ﬁtting and prediction rates is only 0.0607, which

and prediction rates of the RF model. However, the qualiﬁed

is relatively stable. Therefore, the SVM model (polynomial

ﬁtting and prediction rates of remote sensing data at the 11

kernel function) driven by remote sensing data is more suit-

stations are only slightly higher than those of the combined

able for drought monitoring in the Guanzhong Area.

meteorological and remote sensing data. Therefore, the RF model driven by the remote sensing data performed well

Comparison of models driven by combined meteorological and remote sensing data

for drought monitoring in the Guanzhong Area.

Comparing the qualiﬁed rates of the RF model and SVM

other three kernel functions based on the remote sensing

model based on combined meteorological and remote sen-

data are lower than those driven by the combined meteoro-

sing data, the SVM model performs better than the RF

logical and remote sensing data. The difference in the

model in the prediction of drought, in which the Gaussian

qualiﬁed rates of the four kernel function models based on

In the SVM model, except for the polynomial kernel function, the qualiﬁed ﬁtting and prediction rates of the

radial basis kernel function and linear kernel function per-

different data was calculated, which is shown in Table 13.

form best in the ﬁtting and prediction process, respectively.

The differences between the ﬁtting and prediction of the

955

Table 13

J. Li et al.

Drought prediction by meteorological and remote sensing data

Difference in the qualiﬁed rates of the SVM model for different data (absolute values)

Hydrology Research

51.5

2020

terms of drought prediction using meteorological data. The SVM model (polynomial kernel function) is

Linear

Polynomial

Gaussian radial

Sigmoid

Kernel

kernel

basis kernel

kernel

functions

function

Fitting

0.1285

0.0068

0.1918

0.0119

Prediction

0.0458

0.0160

0.1104

0.0257

superior if the models are driven by remote sensing data, while the SVM model (Gaussian radial basis kernel function) outperforms the other models if the predictors are a combination of meteorological and remote sensing data. (3) When using the remote sensing data and the combi-

polynomial and sigmoid kernel functions are mostly less

nation of meteorological and remote sensing data to

than 0.05, respectively, but large for the other kernel func-

build the RF model, the use of meteorological data has

tions. The use of meteorological data has different effects

a small reduction effect on the qualiﬁed ﬁtting and pre-

on the qualiﬁed ﬁtting and prediction rates for different

diction rate; the use of meteorological data has a

kernel functions and has little inﬂuence on the polynomial

different effect on qualiﬁed ﬁtting and prediction rates

and sigmoid kernel functions. The use of meteorological

of different kernel functions in the SVM model. In con-

data significantly improves the qualified fitting and predic-

clusion, the RF model driven by the remote sensing data

tion rate of the Gaussian radial basis kernel function and

and the SVM model driven by the combined meteorolo-

the qualiﬁed ﬁtting rate of the linear kernel function. The

gical and remote sensing data performed better than the

use of remote sensing data signiﬁcantly improved the quali-

model driven by the corresponding other data in the

ﬁed ﬁtting and prediction rates (except for the polynomial

Guanzhong Area.

kernel function for the ﬁtting and linear kernel functions

Different types and lengths of meteorological data and

for prediction). It is difﬁcult to accurately measure the pro-

remote sensing data used in this paper may affect the ﬁtting

cess of drought using a source of meteorological data. Only

and prediction accuracy of meteorological and remote sen-

by considering the factors of precipitation, temperature, veg-

sing data models, which should be discussed in the future.

etation growth and so on, can drought be monitored and predicted more accurately. Therefore, the SVM model driven by the combined meteorological and remote sensing data is better for drought monitoring in the Guanzhong Area.

ACKNOWLEDGEMENT This work is supported by National Natural Science

CONCLUSIONS Based on the calculation of the SPEI series of 11 meteorological stations in the Guanzhong Area, the applicability of the SPEI to the drought characterization in the Guanzhong Area was analyzed. Three different models were used to predict the drought grades, and the main conclusions are as follows: (1) The identiﬁed drought events based on SPEI-1 were in line with the recorded droughts, suggesting that the SPEI-1 data can reasonably reﬂect the occurrence of drought events in the Guanzhong Area. (2) The SVM model (linear kernel function) performs the best, while the ARIMA model performs the worst in

Foundation of China (No. 51479130).

REFERENCES Ahmad, S., Kalra, A. & Stephen, H.  Estimating soil moisture using remote sensing data: a machine learning approach. Advances in Water Resources 33 (1), 69–80. Altman, N. S.  An introduction to kernel and nearest-neighbor non-parametric regression. American Statistician 46 (3), 175–185. Bai, Z., Zhang, L., Wang, J. & Huang, Y.  Research on forecast of meteorological drought in Yunnan based on ARIMA Model. Yangtze River 46 (15), 6–9. Bannari, A., Morin, D., Bonn, F. & Huete, A. R.  A review of vegetation indices. Remote Sensing Reviews 13 (1–2), 95–120.

956

J. Li et al.

Drought prediction by meteorological and remote sensing data

Bonaccorso, B., Bordi, I., Cancelliere, A., Rossi, G. & Sutera, A.  Spatial variability of drought: an analysis of the SPI in Sicily. Water Resources Management 17, 273–296. Box, G. E. & Jenkins, G. M.  Time Series Analysis Forecasting and Control. Holden-Day, San Francisco. Breiman, L.  Random forests. Machine Learning 45 (1), 5–32. Burges, C. J.  A tutorial on support vector machines for pattern recognition. Data Mining and Knowledge Discovery 2, 127–167. Cai, X., Ye, D., Li, Q., Zhang, C. & Wang, N.  Temporal and spatial variation characteristics of drought in Shaanxi based on CI index. Agricultural Research in the Arid Areas 31 (5), 1–8. Che, S. & Li, C.  Time and space characteristics of drought in Shijiazhuang based on standardized precipitation index. Meteorological Science and Technology 38 (1), 66–70. Chen, T., De, J. R. A. M., Liu, Y., Van, D. W. G. R. & Dolman, A. J.  Using satellite based soil moisture to quantify the water driven variability in NDVI: a case study over mainland Australia. Ecological Informatics 5 (5), 400–409. Cong, L.  Analysis of Drought Evolution Characteristics and Drought Prediction in Semi-Arid Areas. Master’s Thesis, Shenyang Agricultural University, Shenyang. Dai, M., Huang, S. Z., Huang, Q., Leng, G. Y., Guo, Y., Wang, L. & Zheng, X. D.  Assessing agricultural drought risk and its dynamic evolution characteristics. Agricultural Water Management 231, 106003. Di, L. P., Rundquist, D. C. & Han, L. H.  Modeling relationships between NDVI and precipitation during vegetative growth cycles. International Journal of Remote Sensing 15 (10), 2121–2136. Dong, Q. & Xie, P.  Progress in hydrological drought research. Journal of China Hydrology 34 (4), 1–7. Fan, G., Zhang, Y., Liu, M. & Mao, Y.  Research on drought prediction based on support vector machine. Chinese Journal of Agrometeorology 32 (3), 475–478. Fang, K., Wu, J., Zhu, J. & Xie, B.  A review of random forest methods. Statistics & Information Forum 26 (3), 32–38. Fang, Q., Wang, G., Xue, B., Liu, T. & Kiem, A.  How and to what extent does precipitation on multi-temporal scales and soil moisture at different depths determine carbon flux responses in a water-limited grassland ecosystem? Science of the Total Environment 635, 1255–1266. Fang, Q., Wang, G., Liu, T., Xue, B., Sun, W. & Shrestha, S.  Unraveling the sensitivity and nonlinear response of water use efficiency to the water-energy balance and underlying surface condition in a semiarid basin. Science of the Total Environment 699, 134405. Farrar, T. J., Nicholson, S. E. & Lare, A. R.  The influence of soil type on the relationships between NDVI, rainfall, and soil moisture in semiarid Botswana.I. NDVI response to soil moisture. Remote Sensing of Environment 50 (2), 121–133. Gao, P.  Precipitation Forecast in Guanzhong Basin and Its Application in Drought Research. Master’s Thesis, Chang’an University, Xi’an.

Hydrology Research

51.5

2020

Ghashghaie, M. & Nozari, H.  Effect of dam construction on Lake Urmia: time series analysis of water level via ARIMA. Journal of Agricultural Science and Technology 20 (S), 1541–1553. Gislason, P. O., Benediktsson, J. A. & Sveinsson, J. R.  Random forests for land cover classification. Pattern Recognition Letters 27 (4), 294–300. Guo, Y., Huang, S. Z., Huang, Q., Leng, G. Y., Fang, W., Wang, L. & Wang, H.  Propagation thresholds of meteorological drought for triggering hydrological drought at various levels. Science of the Total Environment 712, 136502. Han, P., Wang, P. X., Tian, M., Zhang, Y., Liu, J. & Zhu, D.  Application of the ARIMA models in drought forecasting using the standardized precipitation index. In: International Conference on Computer and Computing Technologies in Agriculture. Springer, Berlin, Heidelberg, pp. 352–358. Han, D., Wang, G., Liu, T., Xue, B. L., Kuczera, G. & Xu, X.  Hydroclimatic response of evapotranspiration partitioning to prolonged droughts in semiarid grassland. Journal of Hydrology 563, 766–777. Hernandez, E. & Annette, U. V.  Standardized precipitation evaporation index (SPEI)-based drought assessment in semiarid south Texas. Environmental Earth Sciences 71 (6), 2491–2501. Ichii, K., Hashimoto, H., Nemani, R. & White, M.  Modeling the interannual variability and trends in gross and net primary productivity of tropical forests from 1982 to 1999. Global and Planetary Change 48 (4), 286. Kogan, F. N.  Remote sensing of weather impacts on vegetation in nonhomogeneous areas. International Journal of Remote Sensing 11 (8), 1405–1420. Kogan, F. N.  Droughts of the late 1980s in the United States as derived from NOAA polar-orbiting satellite data. Bulletin of the American Meteorological Society 76 (5), 655–668. Lei, X., Li, Q., Wang, J., Li, H. & Li, H.  The drought and flood evolution law in the Guanzhong Area and the analyses of the present century’s drought and flood characteristics. Journal of Catastrophology 31 (3), 101–109. Li, J., Mo, S., Shen, B., Si, H. & Wang, Y.  Analysis of drought characteristics in Weihe River Basin based on SPEI. Journal of Xi’an University of Technology 32 (1), 70–76. Liu, X.  Natural Disasters and Preventive Measures in the Northwest Region Since the Founding of the People’s Republic (1994–2000). Master’s Thesis, Tianjin University of Commerce, Tianjin. Liu, W., An, S., Liu, G. & Guo, A.  Further correction of Palmer’s drought pattern. Journal of Applied Meteorological Science 15 (2), 207–215. Lu, J.  Temporal and spatial variation characteristics of drought in Yungui area from 1960 to 2014 based on SPEI and run-length theory. Journal of Zhejiang University (Science Edition) 45 (3), 363–372. Malik, A., Kumar, A. & Singh, R. P.  Application of heuristic approaches for prediction of hydrological drought using multi-scalar streamflow drought index. Water Resources Management 33 (11), 3985–4006.

957

J. Li et al.

Drought prediction by meteorological and remote sensing data

Mazdiyasni, O. & Aghakouchak, A.  Substantial increase in concurrent droughts and heatwaves in the United States. Proceedings of the National Academy of Sciences 112 (37), 11484–11489. Mckee, T. B., Doesken, N. J. & Kleist, J.  The relationship of drought frequency and duration to time scales. In: Proceedings of the 8th Conference on Applied Climatology, January 17–22 Anaheim, pp. 179–184. Mei, X.  Temporal and Spatial Distribution of Actual Evapotranspiration in Guanzhong Agricultural Region of Shaanxi Province. Master’s Thesis, Xi’an University of Technology, Xi’an. Meteorological Station of Shaanxi Meteorological Bureau  Record of Natural Disasters in the Shaanxi Province. Shaanxi Meteorological Bureau, Xi’an. Miao, Z.  Analysis and Prediction of Meteorological Drought Characteristics in Baojixia Irrigation District. Master’s Thesis, Northwest A&F University, Yangling. Moody, J. & Darken, C. J.  Fast learning in networks of locally tuned processing units. Neural Computation 1 (2), 281–294. Palmer, W. C.  Meteorological Drought. Research Paper No.45 US Department of Commerce Weather Bureau, Washington, DC. Qiao, L.  Study on Drought Early Warning and Emergency Water Source Allocation in Guanzhong Area. Doctoral Dissertation, Chang’an University, Xi’an. Sahoo, A. K., Shefﬁeld, J., Pan, M. & Wood, E. F.  Evaluation of the tropical rainfall measuring mission multi-satellite precipitation analysis (TMPA) for assessment of large-scale meteorological drought. Remote Sensing of Environment 159, 181–193. Shumway, R. H. & Stoffer, D. S.  An approach to time series smoothing and forecasting using the EM algorithm. Journal of Time Series Analysis 3 (4), 253–264. Sun, Q.  Temporal and Spatial Variation Characteristics of Drought in China From 1981 to 2015 Based on VCI Index. Master’s Thesis, Jiangsu Normal University, Nanjing. Vega, H. M., Lima, J. R. & Cerniak, S. N.  SPEI and Hurst analysis of precipitation in the Amazonian Area of Brazil. Revista Brasileira de Meteorologia 34 (2), 325–334. Vicente-Serrano, S. M., Beguer, A. S. & Lpez-Moreno, J. I.  A multi-scalar drought index sensitive to global warming: the standardized precipitation evapo-transpiration index. Journal of Climate 23 (7), 1696–1718. Wang, Z.  Big Data Mining and Application. Tsinghua University Press, Beijing. Wang, L. & Chen, W.  Applicability analysis of standardized precipitation evapotranspiration index in drought monitoring in China. Plateau Meteorology 33 (2), 423–431. Wei, Z., Chen, S. & Huang, Y.  Temporal and spatial characteristics of potential evapotranspiration in Shaanxi from 1981 to 2000 and its response to climate factors. Scientia Geographica Sinica 35 (8), 1033–1041. Weng, Q.  Remote sensing of impervious surfaces in the urban areas: requirements, methods, and trends. Remoting Sensing of Environment 117 (2), 34–49.

Hydrology Research

51.5

2020

Wu, J.  SPSS Statistical Analysis Starts From Scratch. Tsinghua University Press, Beijing. Wu, H.  Study on the Characteristics of Water Resources Change and Drought Vulnerability in Guanzhong Plain. Doctoral Dissertation, Chang’an University, Xi’an. Wu, J., Chen, Y. & Yu, S.  Drought prediction based on random forest model. China Rural Water and Hydropower 11, 17–22. Xu, H.  Analysis of the Characteristics of Drought Time and Space Evolution and Vulnerability Assessment in Shaanxi Province. Doctoral Dissertation, Chang’an University, Xi’an. Yan, S., Lu, Q., Zhang, J., Zhang, Z. & Bai, S.  Evolution characteristics of NDVI in coastal areas of Jiangsu Province and its response to regional climate change. Journal of Nanjing Forestry University (Natural Sciences Edition) 36 (01), 43–47. Yang, S. Y., Meng, D., Li, X. J. & Wu, X. L.  Multi scale response of vegetation change to SPEI meteorological drought index in North China from 2001 to 2014. Journal of Ecology 38 (3), 1028–1039. Yilmaz, M. T., Hunt, E. R. & Jackson, T. J.  Remote sensing of vegetation water content from equivalent water thickness using satellite imagery. Remote Sensing of Environment 112 (5), 2514–2522. Yinglan, A., Wang, G., Liu, T., Shrestha, S., Xue, B. L. & Tan, Z. a Vertical variations of soil water and its controlling factors based on the structural equation model in a semi-arid grassland. Science of the Total Environment 691, 1016–1026. Yinglan, A., Wang, G., Liu, T., Xue, B. L. & Kuczera, G. b Spatial variation of correlations between vertical soil water and evapotranspiration and their controlling factors in a semi-arid region. Journal of Hydrology 574, 53–63. Yu, Y.  Comparison of Drought Monitoring and Drought Time and Space Characteristics in Guanzhong Region Based on CWSI and TVDI. Master’s Thesis, Shaanxi Normal University, Xi’an. Yuan, X.  Study on Establishment of Hydrological Drought Index and Calculation Method of Drought Frequency. Master’s Thesis, Taiyuan University of Technology, Taiyuan. Yuan, W. & Zhou, G.  Theoretical analysis and research prospects of drought index. Advance in Earth Sciences 19 (6), 982–991. Yurekli, K., Kurune, A. & Ozturk, F.  Application of linear stochastic models to monthly ﬂow data of Kelkit steam. Ecological Modeling 183, 67–75. Zargar, A., Sadiq, R., Naser, B. & Khan, F. I.  A review of drought indices. Environmental Reviews 19 (17), 333–349. Zhang, Y., Hao, Z., Wang, Y., Li, M. & Chen, E.  The relationship between multi-scale drought characteristics and climate index of Taiyuan based on SPEI and SPI index. Ecology and Environmental Sciences 23 (9), 1418–1424. Zhang, B., Wang, S. & Wang, Y. a Copula-based convectionpermitting projections of future changes in multivariate

958

J. Li et al.

Drought prediction by meteorological and remote sensing data

drought characteristics. Journal of Geophysical Research– Atmospheres 124 (14), 7460–7483. Zhang, X., Chen, N., Sheng, H., Ip, C., Yang, L. & Chen, Y. b Urban drought challenge to 2030 sustainable development goals. Science of the Total Environment 693, 133536. Zhang, Y., Li, W., Chen, Q., Pu, X. & Xiang, L.  Multi-models for SPI drought forecasting in the north of Haihe River Basin, China. Stochastic Environmental Research and Risk Assessment 31 (10), 1–11. Zhao, X., Zhu, N. & Huang, L.  Time series modeling algorithm and empirical analysis based on ARIMA model.

Hydrology Research

51.5

2020

Journal of Guilin University of Electronic Technology 32 (5), 410–415. Zhou, X. J., Peng, X. W., Tansey, K., Zhang, S. Y., Li, H. M. & Wang, L.  Developing a fused vegetation temperature condition index for drought monitoring at ﬁeld scales using Sentinel-2 and MODIS imagery. Computers and Electronics in Agriculture 168, 105144. Zhuang, S., Zuo, H., Ren, P., Xiong, G. & Li, B.  Application of standardized precipitation evaporation index in China. Climatic and Environmental Research 18 (5), 617–625.

First received 24 December 2019; accepted in revised form 29 March 2020. Available online 23 June 2020

959

51.5

2020

The inﬂuences of sponge city construction on spring discharge in Jinan city of China Kangning Sun, Litang Hu and Xiaomeng Liu

ABSTRACT In recent years, intense human activities have threatened to dry up the well-known karst springs in Jinan, China. Sponge city construction program was one of the measures aiming to improve the recharge to groundwater and also protect sources of spring discharge. An inﬂuence study of sponge city construction on groundwater is necessary while not fully evaluated. In this paper, a threedimensional numerical groundwater ﬂow model was developed to address this issue. Model calibration showed that the simulated groundwater level successfully reproduced the observed results. Then, 12 scenarios were established and predicted according to different precipitation conditions and the achieved degrees of sponge city construction. The results indicated that the sponge city construction was conducive to the rise of regional groundwater level after 20 years. However, the groundwater level around the spring groups would only increase by an average of 0.22 m, and the annual spring discharge would increase by approximately 9.00 million m3 after 20 years. Results revealed that the extent of spring discharge recovery was not evident in a short time frame. The proper positioning of sponge city construction was suggested to be considered further to

Kangning Sun Litang Hu (corresponding author) College of Water Sciences, Beijing Normal University, Beijing 100875, China; Beijing Key Laboratory of Urban Hydrological Cycle and Sponge City Technology, Beijing 100875, China; and Engineering Research Center of Groundwater Pollution Control and Remediation of Ministry of Education, Beijing 100875, China E-mail: litanghu@bnu.edu.cn Xiaomeng Liu The Government of Beishicao Town Shunyi District, Beijing 101300, China

balance the protection of springs with the issue of waterlogging. Key words

| fractured-karst aquifer, Jinan springs, numerical simulation, sponge city

INTRODUCTION Karst areas account for approximately 7–12% of the total

southern and northern regions of China, including the

land area of the Earth; however, these areas provide more

Yunnan, Guizhou, Sichuan, Chongqing, Guangxi, Hunan,

than 25% of the drinking water for the world’s population

and Hubei regions, which comprise a total area of approxi-

(Ford & Williams ; Wang et al. ). Owing to its

mately 1.76 × 106 km2 (Lu ).

rich reserves, large water yield, extensive distribution, and

Jinan, the capital city of Shandong Province in China, is

good water quality, karst water has become an especially

famous for its numerous karst springs, including four big

important groundwater resource for the world (Bakalowicz

spring groups called Baotu, Black Tiger, Five Dragon, and

). Besides, karst springs are also the sources for local

Pearl springs. However, sustained and rapid economic

rivers and lakes which are important for local ecosystems.

development that occurred in the city from 1975 to 2003

Additionally, the unique topography of karst areas is a sig-

caused a sharp increase in groundwater exploitation, result-

niﬁcant source of tourism for the development of local

ing in continuous groundwater level decline and the cut off

economies. Many karst catchment areas are present in

of springs. For example, the longest continuous cutoff period

This is an Open Access article distributed under the terms of the Creative

). The Chinese government closely monitored the

of Baotu spring was 926 days from 1999 to 2001 (Qian et al. Commons Attribution Licence (CC BY 4.0), which permits copying, adaptation and redistribution, provided the original work is properly cited

spring cutoff problem and conducted research on the restor-

(http://creativecommons.org/licenses/by/4.0/).

ation of the groundwater level and the protection of local

doi: 10.2166/nh.2020.008

960

K. Sun et al.

Inﬂuences of sponge city construction on spring discharge

Hydrology Research

51.5

2020

springs. Furthermore, various measures were taken to allevi-

() and Wang et al. (), aiming at estimation of the

ate severe situations. For example, the utilization of

causes of groundwater level decline and the drying up of

groundwater was partially replaced by surface water (such

springs. These investigations and studies provide good refer-

as water from the Yellow and Yangtze Rivers), many pump-

ences for our further analysis. To our knowledge, the

ing wells were shut down in spring catchment areas,

inﬂuence of sponge city construction on groundwater ﬂow

artiﬁcial groundwater recharge increased, and the inﬁltra-

has not been fully examined, especially the short-term and

tion

the

long-term inﬂuences of the project on groundwater level

implementation of these measures, the springs have been

and spring discharge. The purpose of this study was to

ﬂowing since 2003; however, due to low water levels and

develop a numerical model to evaluate the inﬂuence of

ﬂow, the springs are still at risk.

sponge city construction on the regional groundwater level

surface

water

also

increased.

Since

The sponge city program was announced by the Chinese

and spring discharge in consideration of the implementation

government in 2013 to address the problem of urban water-

of sponge city construction in Jinan. First, we established a

logging. This approach emphasized the use of natural

three-dimensional (3D) numerical model of groundwater

systems (e.g., changing the conditions of the land surface

ﬂow. Then, the model was calibrated based on the esti-

for precipitation inﬁltration, control, and storage). Orig-

mation of the parameters, and the factors controlling the

inally, the term ‘sponge city’ meant a city could function

groundwater discharge were analyzed. Finally, scenarios

like a sponge, demonstrating great resilience to environ-

portraying different construction and precipitation con-

mental changes and natural disasters. This approach was

ditions

were

prepared

predict

changes

the

similar to the best management practices (Ice ) and

groundwater level and spring discharge. This study will pro-

low impact development system of the United States (Xu

vide

et al. ), the water-sensitive urban design of Australia

construction and the protection of springs in Jinan and simi-

(Ahammed ), and the sustainable urban drainage

lar areas.

theoretical

reference

for

further

sponge

city

system of the UK (Ellis & Lundy ). Today, the concept of a sponge city has evolved into a modern stormwater management approach to help solve drainage issues, fully utilize

STUDY AREA

land resources, and promote sustainable development. From 2015 to 2016, two pilot batches of sponge cities (30 cities in

Background

total) were supported by the Chinese Central Government, and the total investment was about 42.3 billion RMB ( Jia

The study area (Figure 1) is located in Jinan, the capital city

et al. ). Jinan was among the ﬁrst batch of pilot sponge

of Shandong Province, with an area of approximately

cities in China in 2015, the purposes of which were to

1,500 km2 and coordinates of approximately 36 280 –

increase precipitation inﬁltration, raise the groundwater

36 460 N, 116 400 –117 140 E. Its eastern and western borders

level, and to inﬂuence the springs.

are the Dongwu and Mashan Faults, respectively. The

Since the 1950s, many ﬁeld investigations and studies

southern boundary is the groundwater watershed, and the

regarding aquifer properties, origin of springs, groundwater

northern boundary the zone of contact metamorphism and

quantity and quality have been carried out in Jinan using

magmatism. The elevation gradually decreases from south

borehole logging, long-term observation of groundwater

to north. The northern area is a piedmont alluvial plain

level and spring ﬂow, tracer test and numerical simulation.

and an intermountain plain with an elevation of approxi-

The earliest two-dimensional groundwater ﬂow model with

mately 20 m–50 m. The urban district is mainly distributed

one-year calibration was established in 1989 to explore the

in this area. To the south is a continuously low and hilly

balance of groundwater supply and spring protection, and

area with an elevation just over 200 m. Further south is a

then over ﬁve groundwater ﬂow models were progressively

steep lower-middle mountainous area with a maximum

constructed (Wang et al. ). Two typical three-dimen-

elevation of 800 m. The study area has a temperate continen-

sional groundwater models were developed by Qian et al.

tal climate. The annual average amount of precipitation

961

Figure 1

K. Sun et al.

Inﬂuences of sponge city construction on spring discharge

Hydrology Research

51.5

2020

Location and geology map of study area.

there is 670.47 mm, while the annual average amount of

(P) strata are buried and exposed only slightly in the north-

evaporation is approximately 1,475.6 mm. Moreover, pre-

west. The Quaternary (Q) strata is well exposed in the

cipitation is mainly concentrated during the months of

northern area. Several large faults are distributed through-

June to September, which accounts for 75% of the total

out the study area in northwest and northeast directions.

annual precipitation. The four main rivers in the study

The Dongwu Fault is impermeable in most areas, demon-

area are the Yellow, Xiaoqing, Dashahe, and Yufu Rivers.

strating weak permeability only in the northeastern part of the spring area; the Mashan Fault is impermeable in the

Geology and hydrogeology

southern mountainous area and permeable in the northern area. The Qianfoshan Fault is in the central part of the

As shown in Figure 1, the Archean Taishan Group (Art),

Jinan spring area and is impermeable in the south and per-

consisting of metamorphic rocks, is distributed in the

meable in the north. The remaining small faults are all

southern area and is the basement of the study area. The

permeable.

Cambrian (∈) strata is characterized by interbeds of lime-

Pore water in a Quaternary aquifer and fractured-karst

stone and shale, which are well exposed from south to

groundwater in an Ordovician aquifer are the main types

north. The Ordovician (O) strata is composed of thick-

of groundwater in the study area, which account for more

bedded limestone and mainly distributed in the middle of

than 80% of the total groundwater discharge, especially

the study area. Most of the Carboniferous (C) and Permian

the karst aquifer, which is the major recharge resource for

962

K. Sun et al.

Inﬂuences of sponge city construction on spring discharge

Hydrology Research

51.5

2020

the springs. Precipitation recharges the groundwater mainly

exploitation was transferred from urban areas to suburban

through surface ﬁssures that have developed in the meta-

areas, and the exploitation decreased suddenly from 30 ×

morphic rock and limestone in the southern and central

104 m3/d to 12 × 104 m3/d in the urban areas. However,

areas, which are the main recharge sources for groundwater

the groundwater level and spring discharge were still

(Figure 2). Surface water and irrigation inﬁltration are also

reduced, and the springs continued to dry up. The last

important components for groundwater recharge. Recently,

stage was the period that began post-2001, after which

artiﬁcial recharge has become another important recharge

more effective measures were taken by the Jinan govern-

source. The karst groundwater in the study area is mainly

ment to protect the springs. For example, water from the

discharged through groundwater exploitation, springs, and

Yellow and Yangtze Rivers and the Wohushan and Yuqin-

recharge of the Quaternary pore water, while the pore

ghu reservoirs was utilized, and the groundwater was

water is mainly discharged through evaporation, springs,

gradually replaced by these surface waters in the water

and groundwater exploitation.

supply system. At the same time, artiﬁcial recharge of the groundwater system was conducted. The implementation

Groundwater utilization

of these measures has ensured continuous ﬂowing of the springs since 2003. Furthermore, the groundwater level

Groundwater utilization can be classiﬁed into four stages (Figure 3). During the ﬁrst stage (pre-1965), groundwater exploitation was limited, and precipitation was the main factor affecting the groundwater regime, and so the highest

began to rise, and spring discharge gradually increased. However, the springs still face the risk of drying up and are in a low water level state, especially from April to June each year.

spring discharge reached 51 × 104 m3/d in 1962. The second stage was the period from 1965 to 1980. Groundwater exploitation gradually increased from 10 × 104 m3/d

METHODS AND DATA

in 1965 to approximately 30 × 104 m3/d in 1980 in the urban district. The total groundwater exploitation in the

Simulation method

spring area reached 60 × 104 m3/d in 1972 and then increased to 70 × 104 m3/d by 1980. Due to the overexploita-

Equivalent porous media method

tion of groundwater, the springs have begun to dry up during each dry season since 1972, and the number of dried-up days

Flow in a fractured-karst aquifer is usually a non-Darcy ﬂow.

increased correspondingly. The third stage was the period

Flow simulation in a fractured-karst medium is challenging.

from 1981 to 2000. For spring protection, groundwater

According to the condition of whether physical processes

Figure 2

Schematic diagram of genesis of Jinan karst spring (Wang et al. 2014).

K. Sun et al.

963

Figure 3

Inﬂuences of sponge city construction on spring discharge

Hydrology Research

51.5

2020

Changes in precipitation, spring discharge, and groundwater exploitation in urban district; average groundwater level and number of dried-up days of Baotu spring from 1958 to 2010.

are considered, these models can roughly be classiﬁed into

spatial difference, and it only measures the structural

two categories (Hartmann et al. ; Meng et al. ).

relationship between inputs and outputs. Some examples

The ﬁrst category is a physically based model. This type of

of a lumped model are the commonly used artiﬁcial neural

model usually applies a ﬁnite differential method or a

network model (Meng et al. ), linear or nonlinear reser-

ﬁnite element method to discretize a fractured-karst aquifer

voir model (Padilla & Pulido-Bosch ), wavelet analysis

into several small grids, where different hydrogeological par-

(Labat et al. ), and regression analysis (Felton & Currens

ameters and boundary, recharge, and discharge conditions

). Lumped models cannot reﬂect the hydraulic par-

are given. Owing to the high heterogeneity and anisotropy

ameters of an aquifer, the ﬂow direction of groundwater,

of karst aquifer media, it is difﬁcult to accurately character-

or the velocity of groundwater ﬂow.

ize the hydrogeological parameters of an aquifer; thus,

The geology structure in the study area is complex, and

aquifers are often generalized. Many simpliﬁcation methods

the aquifer contains karst, ﬁssures, and porous aquifers with

have been proposed, such as the equivalent porous media

different pore sizes, which are difﬁcult to generalize with

model (Scanlon et al. ; Dragoni et al. ), double

one another. In this study, a convenient and efﬁcient equiv-

media model (Robineau et al. ), triple media model

alent medium simulation method was chosen, which is

(Chen & Hu ), discrete fractured network model

widely used at present in physically based models. In par-

(Dverstorp et al. ), and discrete conduit network

ticular, when the main aquifer media are dissolution pores

model (Ghasemizadeh et al. ). The second category is

rather than caves and channels, this kind of model can be

a lumped model; this type of model takes a karst ground-

used to simulate water balances and trends of the regional

water system as a uniﬁed entity without considering

groundwater ﬂow (Scanlon et al. ). Therefore, this

964

K. Sun et al.

Inﬂuences of sponge city construction on spring discharge

Hydrology Research

51.5

2020

method could be employed in our study area. Likewise, it

permeable. The Qianfoshan Fault was represented as a

has been successfully applied in most of the karst aquifer

weakly permeable boundary in the southern section and a

systems in northern China (Kang et al. ).

permeable boundary in the northern part. The bottom boundary of the model was set as the bottom of the upper

Simulation program

Quaternary overburden and the lower Cambrian, and the total thickness of the model was approximately 500–

This study chose a polygon grid ﬁnite difference groundwater

600 m. There were three vertical model layers. The ﬁrst

modeling system (named PGMS) to establish a 3D ground-

layer mainly consisted of Quaternary, Cambrian, and Ordo-

water ﬂow numerical model. The PGMS was based on a

vician strata, and the thickness of this layer was

polygonal grid, which was more convenient, ﬂexible, and

approximately

accessible and had more advantages than a rectangular

included upper Cambrian strata with a thickness of approxi-

grid. Moreover, the PGMS could deal well with problems

mately 58–100 m. The third layer mainly consisted of

associated with karst groundwater and karst springs, and it

Middle to Lower Cambrian strata, and the thickness was

was applied in the Heihe River Basin (Hu et al. ) with

approximately 170–470 m. Among them, the ﬁrst and third

good results. In addition, the authors have the source code

layers were major aquifers. The model was represented as

of the PGMS, which could be further improved based on

a heterogeneous anisotropy with a 3D spatial structure

the need for other numerical simulations.

and transient groundwater ﬂow.

Uncertainty

analysis

software

called

185–380 m.

The

second

layer

mainly

Uncertainty

Quantiﬁcation Python Laboratory (UQ-PyL) (Wang et al.

Data preparation

) was chosen for parameter optimization. UQ-PyL coupled with the PGMS through ‘control ﬁles’ (‘template’

The main data sources concerning groundwater level and

and ‘driver’ ﬁles). The template ﬁle was a parameter control

spring discharge are listed in Table 1 and include topogra-

ﬁle that wrote the parameters in a separate ﬁle, and the

phy, precipitation, hydrogeology, observation wells, spring

model ﬁle generated by the PGMS needed this ﬁle to read

discharge, and many research reports (Wang et al. ).

the parameters automatically. The driver ﬁle contained all

Sources and sinks were processed in this study. The main

the adjustable parameters. The model parameter settings

recharge sources included were inﬁltration from precipi-

could be changed in this ﬁle, and the results of the model

tation, rivers, and artiﬁcial recharge, and the return ﬂow of

operation could also be processed.

irrigation water. The main rivers studied were the Yufu,

Conceptual model and data preparation

Table 1

Conceptual model

Lists of data sources used in the model

Data items

Descriptions

Topography

Spatial resolution of 16 m and over the study area, and resolution with 0.5 m on the northeast of spring area

Precipitation

Daily precipitation data from 2014 to 2016 at 9 stations, and annual precipitation data from 1956 to 2017 at 20 stations

Hydrogeology

Hydrogeological maps and borehole descriptions from previous reports (Wang et al. ), 10 hydrogeological cross sections

Observation wells

Annual groundwater level data from 1956 to 2017 at 100 wells and monthly water level data from 2014 to 2016 at 17 wells

Spring discharge

Annual springs discharge from 1958 to 2010 and monthly data from 2014 to 2017

The Jinan spring catchment was selected as the model area. The boundary conditions were set according to previous studies (Wang et al. ). The northern boundary of the model area was regarded as a no-flow boundary, but a permeable boundary was present where the Yellow River flows. The southern boundary was generalized as an impermeable boundary. The eastern boundary was represented as a no-flow boundary, except the northeastern section of the spring area, which was a weakly permeable boundary. The western boundary was represented as a noflow boundary, except the northern plains area, which was

965

K. Sun et al.

Inﬂuences of sponge city construction on spring discharge

Beidasha, Yellow, and Xiaoqing Rivers. Artiﬁcial ground-

Hydrology Research

51.5

2020

Model discretization and calibration

water recharge was primarily from two strong seepage zones (reaches of the Yufu and Xingji Rivers). The area of irrigation was determined by remote sensing images.

Model discretization and zonation of hydrogeological parameters

Groundwater evaporation was calculated from a model based on the depth of groundwater, soil types, and given

The study area was divided into three layers, and each layer

limits (Hu et al. ). Groundwater exploitation occurred

was divided into 6,488 auxiliary triangles for a total of

in local regions. Spring discharge was calculated based on

19,464 auxiliary triangles (Figure 4). The period of model

the groundwater level and the thickness and permeability

calibration was from January 2014 to December 2016. The

of underlying media, which included Baotu, Black Tiger,

time step was set as one month; therefore, there were a

Five Dragon, and Pearl springs. The critical water level

total of 36 time steps. The trial-and-error method was used

elevation of the spring groups was set according to the

to calibrate the model against observed data to achieve the

actual outﬂow elevation. The outcrops of Baotu, Black

smallest possible objection function, which was deﬁned as

Tiger, Five Dragon, and Pearl springs were 27.01 m,

the sum of the squares of the differences between the

27.30 m, 26.20 m, and 26.77 m, respectively.

observed and calculated results, including both the water

Figure 4

Three-dimensional (3D) network model sketch of research area.

K. Sun et al.

966

Inﬂuences of sponge city construction on spring discharge

Hydrology Research

51.5

2020

level and spring discharge. Then, UQ-PyL software was used

ﬁeld and spring discharge (Serial Zones 16 and 31,

to optimize the parameters.

Table 2). The zonation of the three aquifers is presented

Many aquifer parameters, including horizontal (Kxx and Kyy) and vertical (Kzz) hydraulic conductivities,

in Figure 5, and the optimized parameters are listed in Table 2.

speciﬁc yield (Sy), and speciﬁc storage (Ss), were involved due to the complexity of the aquifer system (Kzz ¼ ½ Kxx, Kyy ¼ Kxx). Based on the obtained borehole data, the hydrogeological

parameters

Comparison of observed and simulated groundwater level data

of the three simulation

layers were subdivided into 71 zones. The hydrogeological

Calibration targets were calculated as the root mean

parameters were initially assigned, and then they were

square of the difference between the observed and simu-

adjusted and ﬁxed after the model was calibrated. Consid-

lated groundwater levels (Δh). There was a total of 17

ering the changes of the hydrogeological parameters

observation wells (13 in the plains area and four in the

caused by sponge city construction, the ﬁrst layer of the

mountainous area), and the observation data were col-

sponge city pilot area was subdivided separately to

lected from 2014 to 2016. The results (Table 3)

better simulate the changes in the groundwater ﬂow

indicated that the number of absolute errors between

Table 2

Lists of calibrated and optimized hydrogeological parameters

Serial zone

Kxx (m/d)

Ss (10

0.02

Serial zone

Kxx (m/d)

Ss (10

0.40

0.10

40.00

0.60

0.10

20.00

0.05

0.60

0.20

0.03

0.40

0.20

0.06

1.00

0.20

0.05

0.60

0.04

0.60

0.02

0.05

Serial zone

Kxx (m/d)

Ss (10

7.00

0.05

10.00

0.10

0.01

5.00

0.05

5.00

0.10

0.01

80.00

5.00

0.05

0.10

0.20

0.05

70.00

7.00

0.05

0.50

0.10

30.00

7.00

0.05

0.10

0.20

0.15

0.25

1.00

0.01

0.10

0.15

0.20

10.00

7.00

0.01

0.10

0.20

0.10

0.60

0.001

50.00

0.50

0.005

0.05

1.00

0.10

0.60

0.20

5.00

1.00

0.05

0.01

1.00

0.05

0.60

0.20

40.00

0.10

0.001

0.01

0.50

0.10

0.01

1.00

0.05

30.00

0.10

0.005

5.00

2.00

0.01

0.008

1.00

0.05

0.10

1.00

0.01

5.00

2.00

0.05

50.00

7.00

0.05

0.10

1.00

0.05

0.10

0.50

0.01

30.00

7.00

0.01

1.00

0.50

0.10

0.20

0.01

30.00

4.00

0.10

50.00

5.00

0.10

1.00

0.01

30.00

4.00

0.005

1.00

0.10

2.00

0.05

)

10.00

4.00

0.20

10.00

5.00

0.10

2.00

0.05

40.00

7.00

0.05

10.00

5.00

0.10

40.00

1.00

0.01

30.00

7.00

0.04

3.00

2.00

0.10

40.00

1.00

0.01

30.00

7.00

0.10

10.00

5.00

0.10

40.00

1.00

0.10

5.00

1.00

0.20

3.00

1.00

0.10

40.00

1.00

0.05

4.00

1.00

0.10

1.00

0.10

0.50

0.01

1.00

0.10

10.00

0.50

0.10

0.20

0.10

5.00

7.00

0.05

10.00

0.50

0.05

K. Sun et al.

967

Figure 5

Table 3

Inﬂuences of sponge city construction on spring discharge

Hydrology Research

51.5

2020

Parameter zonation of three layers.

Statistics of absolute errors and percentages for 17 observation wells

Absolute errors (m)

Δh 0.5

0.5<Δh 1.0

1.0<Δh 3.0

Δh>3.0

Total

All observation holes

Frequency of occurrence Percentage (%)

146 27.29

107 20.00

170 31.78

112 20.93

535 100

Observation holes in plain area

Frequency of occurrence Percentage (%)

131 33.08

98 24.75

135 34.09

32 8.08

396 100

Observation holes in mountain area

Frequency of occurrence Percentage (%)

15 10.79

9 6.48

35 25.18

80 57.54

139 100

the simulated and observed values were less than 0.5 m,

Groundwater budget analysis

1.0 m, and 3.0 m, which accounted for 27%, 47%, and 79%, respectively. The relative errors within 5%, 10%,

There were four springs in total, and observation data were

and 20% accounted for 70%, 85%, and 94%, respectively.

collected from 2014 to 2016. The observed average annual

The goodness-of-ﬁt in the plains area was higher than

spring discharge for Pearl, Five Dragon, Baotu, and Black

that of the mountainous area. If the observation holes

Tiger springs was 0.06, 0.12, 0.18, and 0.16 × 108 m3, respect-

in the plains area were solely analyzed, the absolute

ively. The simulated average annual spring discharge for

errors were less than 0.5 m, 1.0 m, and 3.0 m, accounting

Pearl, Five Dragon, Baotu, and Black Tiger springs was

for 33%, 58%, and 92%, respectively; the relative errors

0.07, 0.14, 0.22, and 0.17 × 108 m3, respectively. For the

within 5%, 10%, and 20% accounted for 77%, 92%,

four springs, the observed and simulated annual total dis-

and 95%, respectively. The ﬁtting accuracy of the obser-

charges were 0.53 and 0.60 × 108 m3, respectively, and the

vation holes in the mountainous area was slightly poor,

absolute error of discharge was approximately 0.07 ×

but the relative errors of the simulated value that were

108 m3 with a relative error (the ratio of the absolute differ-

less than 20% also accounted for 91%, and more than

ence of observed and simulated discharge to the observed

half of them were less than 5%. Figure 6 shows the

results) of approximately 13%. Meanwhile, for the obser-

comparison

observed

vation wells around the four springs (Figure 6(e)), the

groundwater levels (corresponding with observation

simulated water level was consistent with the observed

wells a, b, c, d, e, and f in Figure 1). The results indicated

results.

between

the

simulated

and

that the prediction could adequately reﬂect the actual groundwater level.

The simulated groundwater budget in the study area from 2014 to 2016 was obtained. The main groundwater

968

Figure 6

K. Sun et al.

Inﬂuences of sponge city construction on spring discharge

Hydrology Research

51.5

2020

Comparison of observed and simulated groundwater levels: (a), (b), (c), (d), (e), and (f) represent observation wells a, b, c, d, e, and f from Figure 1.

recharge sources included inﬁltration from precipitation and

and 24% of the total recharge, respectively. From 2014 to

rivers (natural river inﬁltration and artiﬁcial recharge), and

2016, the total annual value of groundwater recharge was

the average recharge values were 3.24 and 0.77 × 108 m3

3.39, 3.93, and 5.21 × 108 m3, respectively, and the total

(accounting for 78% and 19% of the total recharge), respect-

annual value of groundwater discharge was 2.43, 2.49, and

ively. The main groundwater discharge sources included

2.56 × 108 m3, respectively. The changes in groundwater sto-

groundwater exploitation and springs, and the average dis-

rage from 2014 to 2016 were 0.96, 1.44, and 2.65 × 108 m3,

charge was 1.41 and 0.61 × 10 m , accounting for 57%

respectively, and the groundwater system was always in a

969

K. Sun et al.

Inﬂuences of sponge city construction on spring discharge

Hydrology Research

51.5

2020

positive balance. The increase in groundwater storage

and when combined with our own research purposes, this

should have a beneﬁcial impact on the recovery of spring

area was divided into two zones. Zone a was the develop-

discharge.

ment

and

construction

area

with

sponge

city

construction target of runoff coefﬁcient and annual runoff control rate of 0.7 and 75%, respectively. Zone b was the

MODEL APPLICATION AND RESULTS Construction of Jinan sponge city

mountainous area with a sponge city construction target of runoff coefﬁcient and annual runoff control rate of 0.4 and 85%, respectively. The total area of the pilot area was 39 km2, including 22 km2 in Zone a and 17 km2 in Zone b.

The sponge city pilot was located in the southeast region of the study area with mountains to the south; the pilot was

Scenarios set

approximately 1.8 km north of the springs (Figure 7). The elevation of the pilot area ranged from 23 m to 460 m. In

The most direct effect of sponge city construction on

this area, the mountains and plains were merged, and the

groundwater was an increase in the inﬁltration capacity of

plains area (with its low permeability) was primarily utilized

the underlying surface; namely, the degree of sponge city

for development and construction. According to the con-

construction could be represented by a precipitation inﬁltra-

struction and implementation plans for Jinan sponge city,

tion coefﬁcient. Twelve scenarios (S) were set, as shown in

the planning targets were to be set by each planning area,

Table 4. Scenarios S0–S3 corresponded to the inﬁltration

Figure 7

Sponge city pilot area.

K. Sun et al.

970

Table 4

Inﬂuences of sponge city construction on spring discharge

Hydrology Research

51.5

2020

Scenarios set for model prediction

Inﬁltration coefﬁcient scenarios S0

Zone a

Zone b

Zone a

Zone b

Zone a

Zone b

Zone a

Zone b

Precipitation scenarios

0.15

0.30

0.19

0.34

0.23

0.42

0.30

0.60

Dry year

D-S0

D-S1

D-S2

D-S3

Normal year

N-S0

N-S1

N-S2

N-S3

Wet year

W-S0

W-S1

W-S2

W-S3

conditions caused by different degrees of sponge city con-

The increment of the groundwater level under different

struction. Among them, S0 was the scenario when sponge

precipitation conditions in the pilot area was analyzed when

city construction had not yet begun, while S3 was the scen-

sponge city construction was completed according to the

ario when sponge city had been completed according to the

initial target. In successive ‘dry’ years, the areas where the

initial planning target.

groundwater level rose more than 2 m, 4 m, 6 m, and 8 m

Precipitation conditions were simulated according to

accounted for approximately 58%, 25%, 9%, and 3% of the

dry (D), normal (N), and wet (W) years (obtained by the

total area, respectively. In successive ‘normal’ years, the

P-III curve), based on the multi-year precipitation series.

areas where the groundwater level rose more than 2 m,

The amount of precipitation (P) in the wet year (P ¼ 25%),

4 m, 6 m, and 8 m accounted for approximately 71%, 40%,

normal year (P ¼ 50%), and dry year (P ¼ 75%) was

14%, and 7% of the total area, respectively. In successive

782.45 mm, 648.73 mm, and 534.87 mm, respectively. The

‘wet’ years, the areas where the groundwater level rose

precipitation conditions (D, N, or W) corresponded with

more than 2 m, 4 m, 6 m, and 8 m accounted for approxi-

the benchmark scenarios (S), and were represented by

mately 86%, 50%, 22%, and 11% of the total area,

D-S0, N-S0, and W-S0, respectively. The prediction time

respectively. No matter what the precipitation conditions,

frame was 20 years, and the time step was one month.

the groundwater level in most of the pilot area could rise more than 2 m after 20 years. Sponge city construction

Results analysis Regional water level distribution Under different precipitation conditions, the groundwater

generally raised the groundwater level of the pilot area and its surroundings, and the higher the degree of sponge city construction and the more precipitation there was, the larger the increment, and the wider the range of increment.

level changed in scenarios S1, S2, and S3, in contrast to that of scenario S0 (Figure 8). No matter what the scenario

Change in groundwater level around springs

was, it was demonstrated that the most rapid and largest increase of groundwater level was in the mountainous

We took observation well e as an example to demonstrate

area, and it gradually expanded to the surrounding areas

the inﬂuence of sponge city construction on the ground-

and formed a high value area of the regional groundwater

water level near the spring groups. Under different

level. When the inﬁltration conditions were the same, the

precipitation conditions, the increment of the groundwater

groundwater level increased gradually with an increase in

level from scenarios S1, S2, and S3 was compared to that

precipitation. When the precipitation conditions were the

of scenario S0 (Figure 9), and 240 data were collected for

same, we observed that the greater the inﬁltration capacity

each scenario. Compared with scenario D-S0, where the

of the underlying surface, the greater the increment of the

maximum increment of the groundwater level was 0.09,

groundwater level.

0.13, and 0.27 m, the average increment of the groundwater

971

Figure 8

K. Sun et al.

Inﬂuences of sponge city construction on spring discharge

Distribution of increased groundwater level for various scenarios in pilot area surroundings after 20 years.

Hydrology Research

51.5

2020

972

Figure 9

K. Sun et al.

Inﬂuences of sponge city construction on spring discharge

Hydrology Research

51.5

2020

Statistics for increment of groundwater level in various scenarios (S) in observation well e.

level was 0.03, 0.06, and 0.13 m in scenarios D-S1, D-S2,

under various precipitation conditions. It can be seen that

and D-S3, respectively. Compared with scenario N-S0,

when the inﬁltration conditions were the same, the spring

where the maximum increment of the groundwater level

discharge increased the most in wet years. When the precipi-

was 0.15, 0.23, and 0.40 m, the average increment of the

tation remained the same, the larger the inﬁltration capacity

groundwater level was 0.03, 0.08, and 0.18 m in scenarios

was, the more the spring discharge increased; 20 years later,

N-S1, N-S2, and N-S3, respectively. Compared with scenario

the sponge city construction could increase the annual

W-S0, where the maximum increment of the groundwater

spring discharge by 0.0672, 0.0766, and 0.0875 × 108 m3 in

level was 0.16, 0.23, and 0.46 m, the average increment of

a dry year, normal year, and wet year, respectively. The

the groundwater level was 0.04, 0.10, and 0.22 m in scen-

observed data indicated that the discharge of Pearl spring

arios W-S1, W-S2, and W-S3, respectively. The simulation

in 2014 was 0.058 × 108 m3. It could be seen that under

results indicated that when precipitation was sufﬁcient, the

the condition of continuous wet years, sponge city construc-

groundwater level near the spring groups would increase

tion could increase the annual spring discharge there by

by 0.46 m, and the average water level would increase by

approximately 1.5 times the Pearl spring discharge after 20

0.22 m compared to the results that would occur in the

years.

absence of the sponge city after 20 years. The higher the construction level of the sponge city and the more precipitation there is, the more the groundwater level near the spring groups should rise.

DISCUSSION Based on our analysis, we found that sponge city construc-

Change in spring discharge

tion, due to an increased precipitation inﬁltration capacity of the underlying surface, could affect the precipitation to

Figure 10 shows the increment of spring discharge in scen-

recharge the local groundwater, which would be beneﬁcial

arios S1, S2, and S3 compared with that of scenario S0

for the increment of the groundwater level in and around

973

Figure 10

K. Sun et al.

Inﬂuences of sponge city construction on spring discharge

Hydrology Research

51.5

2020

Increment of spring discharge over prediction time frame.

the sponge city area. After 20 years, in all scenarios, the

and Ordovician limestones were exposed over a large area

groundwater level should have increased. The higher the

(Figure 2). The pilot test area of sponge city construction

degree of sponge city construction, the greater the amount

was almost located in the local groundwater system of the

of precipitation and the more obvious the uplift effect

Jinan watershed and partially belonged to the recharge

would be. However, the effect of sponge city construction

area. Therefore, when the inﬁltration coefﬁcient of the

with regard to increasing the groundwater level and spring

groundwater in the sponge city pilot area increases, it

discharge near the spring groups was not proportional.

could affect the local groundwater ﬂow within a certain

The average increment of the groundwater level near the

range, which has a limited effect on spring restoration (the

spring groups should rise by about 0.22 m, the maximum

simulated maximum annual increment percentage was

increment of the groundwater level near the spring groups

approximately 5%). The construction of the sponge city

could rise by 0.46 m, and the annual spring discharge

was a project that necessitated both huge quantities in con-

should increase by approximately 0.09 × 10 m in continu-

struction and monetary funds. The construction of the Jinan

ous wet years after 20 years. In 20 continuous wet years,

sponge city cost approximately 7.831 billion yuan, and it

the proportion of annual increment gradually changes

involved the reconstruction of a building area, garden and

from 0.40% to 5.00% when compared with the simulated

green space, roadways, an urban water system, among

annual spring discharge for the four springs in the scenario

other aspects. To increase groundwater recharge to the

S0. As time goes on, the inﬂuence of sponge city construc-

greatest extent possible, so as to protect the springs by

tion on spring discharge is increasingly prominent.

means of sponge city construction, it would be necessary

Therefore, an immediate response by the increment of

to expand the pilot area; however, this method could require

spring discharge would not be observed in a short time

further investments. It could be more effective if the pilot

frame.

area were moved to the southern part of the Jinan water-

In the Jinan spring area, the main groundwater recharge

shed. Owing to the short distance between the recharge

source was from the southern region where the Cambrian

and discharge areas of the watershed and the large

974

K. Sun et al.

Inﬂuences of sponge city construction on spring discharge

Hydrology Research

51.5

2020

topographic slope when rainstorms occur, the runoff would

0.22 m, on average, and the maximum annual spring dis-

not be able to be intercepted in time, which would result in

charge would increase by approximately 0.09 × 108 m3.

urban waterlogging; therefore, how to balance the protec-

The increased maximum annual spring discharge after 20

tion of the springs with the issue of waterlogging will be a

years could account for approximately 5% of the simulated

question worthy of further consideration.

average annual spring discharge. It was not possible for water managers to observe an immediate response of the groundwater level and spring discharge in a short time

CONCLUSIONS

frame after the implementation of the project. To increase spring discharge after the implementation of

As the ﬁrst batch of pilot sponge cities in China in 2015, one

sponge city construction in Jinan, one method would be to

of the important purposes of sponge city construction in

further expand the pilot area, which would incur further

Jinan was to increase the precipitation inﬁltration and pro-

expenses; the other method, to increase spring discharge,

tect the springs. The effect of sponge city construction on

would be to move the pilot area to the southern region,

the groundwater system in Jinan was discussed in detail in

which would probably enhance the recharge sources of

this study. Groundwater types vary from the karst aquifer

the groundwater system. Jinan has a waterlogging issue

in the southern region to the porous media in the northern

due to its short distance from the recharge zone to the dis-

region of the study area. The groundwater mainly received

charge zone. The combined inﬂuence of waterlogging and

inﬁltration recharge from precipitation in the southern

sponge city construction on the groundwater level was pre-

area. Four well-known major springs were located in the

sent and not discussed in this article; however, the proper

northern part of the study area, and the sponge city pilot

positioning of a sponge city pilot area could depend on prop-

area in Jinan was located in the north part. A ﬁnite-differ-

erly balancing the protection of springs with the issue of

ence groundwater ﬂow model in the Jinan spring area was

waterlogging.

developed to evaluate the inﬂuences of the project on the groundwater system. The equivalent media method was used to characterize the karst and ﬁssure aquifers. Model calibration was conducted based on the observed ground-

ACKNOWLEDGEMENTS

water level and spring discharge data, using a trial-anderror method and optimized software (UQ-PyL). After

This work was supported by the National Natural Science

model calibration, the model was applied to evaluate the

Foundation

China

(Grant

Nos.

41877173

and

changes in the groundwater level and spring discharge

41831283)

and

the

National

Key

Research

and

after sponge city construction near the pilot area.

Development Program of China (2018YFC0407900). The

The results from the 12 scenarios showed that the larger

authors would also like to acknowledge the ﬁnancial

the inﬁltration capacity of the sponge city pilot area, the

support

more the groundwater level and spring discharge rose.

Hydrological Process Modeling of Jinan ‘Sponge City’ and

Additionally, the more precipitation that fell, the more the

Beijing Advanced Innovation Program for Land Surface

uplift of the groundwater level occurred. Sponge city con-

Science.

struction

was

beneﬁcial

the

increment

the

project

named

Water

Cycle

the

groundwater level in the pilot area. After 20 years, even under the condition of dry years, the groundwater level in most areas of the pilot could rise more than 2 m. However, the effect of the sponge city construction on the protection of the springs was not very obvious. After 20 years, under the condition of continuous wet years, the groundwater level around the spring groups would only increase by

REFERENCES Ahammed, F.  A review of water-sensitive urban design technologies and practices for sustainable stormwater management. Sustainable Water Resources Management 3 (3), 1–14. https://doi.org/10.1007/s40899-017-0093-8.

and

975

K. Sun et al.

Inﬂuences of sponge city construction on spring discharge

Bakalowicz, M.  Karst groundwater: a challenge for new resources. Hydrogeology Journal 13 (1), 148–160. https://doi. org/10.1007/s10040-004-0402-9. Chen, C. X. & Hu, L. T.  A review of the seepage-pipe coupling model and its application. Hydrogeology & Engineering Geology 35 (3), 70–75 (In Chinese). Dragoni, W., Mottola, A. & Cambi, C.  Modeling the effects of pumping wells in spring management: the case of Scirca spring (central Apennines, Italy). Journal of Hydrology 493, 115–123. https://doi.org/10.1016/j.jhydrol.2013.03.032. Dverstorp, B., Andersson, J. & Nordqvist, W.  Discrete fracture network interpretation of field tracer migration in sparsely fractured rock. Water Resources Research 28 (9), 2327–2343. https://doi.org/10.1029/92WR01182. Ellis, J. B. & Lundy, L.  Implementing sustainable drainage systems for urban surface water management within the regulatory framework in England and Wales. Journal of Environmental Management 183, 630–636. https://doi.org/ 10.1016/j.jenvman.2016.09.022. Felton, G. K. & Currens, J. C.  Peak flow rate and recessioncurve characteristics of a karst spring in the inner bluegrass, central Kentucky. Journal of Hydrology 162 (1–2), 99–118. https://doi.org/10.1016/0022-1694(94)90006-X. Ford, D. C. & Williams, P. W.  Karst Hydrogeology and Geomorphology. Wiley, Chichester, UK. Ghasemizadeh, R., Hellweger, F., Butscher, C., Padilla, I., Vesper, D., Field, M. & Alshawabkeh, A.  Review: groundwater flow and transport modeling of karst aquifers, with particular reference to the North Coast Limestone aquifer system of Puerto Rico. Hydrogeology Journal 20 (8), 1441–1461. https://doi.org/10.1007/s10040-012-0897-4. Hartmann, A., Goldscheider, N., Wagener, T., Lange, J. & Weiler, M.  Karst water resources in a changing world: review of hydrological modeling approaches. Reviews of Geophysics 52 (3), 218–242. https://doi.org/10.1002/2013RG000443. Hu, L. T., Chen, C. X., Jiao, J. J. & Wang, Z. J.  Simulated groundwater interaction with rivers and springs in the Heihe river basin. Hydrological Processes 21 (20), 2794–2806. https://doi.org/10.1002/hyp.6497. Ice, G.  History of innovative best management practice development and its role in addressing water quality limited waterbodies. Journal of Environmental Engineering 130 (6), 684–689. https://doi.org/10.1061/(ASCE)0733-9372(2004) 130:6(684). Jia, H. F., Yu, S. L. & Qin, H. P.  Low impact development and sponge city construction for urban stormwater management. Frontiers of Environmental Science & Engineering 11 (4), 20. https://doi.org/10.1007/s11783-017-0989-4. Kang, F. X., Jin, M. G. & Qin, P. R.  Sustainable yield of a karst aquifer system: a case study of Jinan springs in Northern

Hydrology Research

51.5

2020

China. Hydrogeology Journal 19 (4), 851–863. https://doi.org/ 10.1007/s10040-011-0725-2. Labat, D., Ababou, R. & Mangin, A.  Introduction of wavelet analyses to rainfall/runoffs relationship for a karstic basin: the case of Licq-Atherey karstic system (France). Groundwater 39 (4), 605–615. https://doi.org/10.1111/ j.1745-6584.2001.tb02348.x. Lu, Y. R.  Karst water resources and geo-ecology in typical regions of China. Environmental Geology 51 (5), 695–699. https://doi.org/10.1007/s00254-006-0381-3. Meng, X. M., Yin, M. S., Ning, L. B., Liu, D. F. & Xue, X. W.  A threshold artificial neural network model for improving runoff prediction in a karst watershed. Environmental Earth Sciences 74 (6), 1–10. https://doi.org/10.1007/s12665-0154562-9. Padilla, A. & Pulido-Bosch, A.  Simple procedure to simulate karstic aquifers. Hydrological Processes 22 (12), 1876–1884. https://doi.org/10.1002/hyp.6772. Qian, J. Z., Zhan, H. B., Wu, Y. F., Li, F. L. & Wang, J. Q.  Fractured-karst spring-flow protections: a case study in Jinan, China. Hydrogeology Journal 14 (7), 1192–1205. https://doi. org/10.1007/s10040-006-0061-0. Robineau, T., Tognelli, A., Goblet, P., Renard, F. & Schaper, L.  A double medium approach to simulate groundwater level variations in a fissured karst aquifer. Journal of Hydrology 565, 861–875. https://doi.org/10.1016/j.jhydrol. 2018.09.002. Scanlon, B. R., Mace, R. E., Barrett, M. E. & Smith, B.  Can we simulate regional groundwater flow in a karst system using equivalent porous media models? Case study, Barton Springs Edwards aquifer, USA. Journal of Hydrology 276 (1–4), 137–158. https://doi.org/10.1016/S0022-1694(03)00064-7. Wang, Q. B., Yan, J. S., Duan, X. M. & Gao, Z. D.  Study on Groundwater Circulation and Sustainable Development in Jinan Spring Area of Karst in North China. Jinan Publishing House, Jinan, China (In Chinese). Wang, C., Duan, Q. Y., Tong, C. H., Di, Z. H. & Gong, W.  A GUI platform for uncertainty quantification of complex dynamical models. Environmental Modelling & Software 76, 1–12. https://doi.org/10.1016/j.envsoft.2015.11.004. Wang, G., Liu, S., Liu, T., Fu, Z., Yu, J. & Xue, B.  Modelling above-ground biomass based on vegetation indexes: a modified approach for biomass estimation in semi-arid grasslands. International Journal of Remote Sensing 40 (10), 3835–3854. Xu, C. Q., Jia, M. Y., Xu, M., Long, Y. & Jia, H. F.  Progress on environmental and economic evaluation of low-impact development type of best management practices through a life cycle perspective. Journal of Cleaner Production 213, 1103–1114. https://doi.org/10.1016/j.jclepro.2018.12.272.

First received 8 January 2020; accepted in revised form 6 April 2020. Available online 20 May 2020

976

51.5

2020

Quantiﬁcation of climate change and land cover/use transition impacts on runoff variations in the upper Hailar Basin, NE China Yuhui Yan, Baolin Xue, Yinglan A, Wenchao Sun and Hanwen Zhang

ABSTRACT Quantiﬁcation of runoff change is vital for water resources management, especially in arid or semiarid areas. This study used the Soil and Water Assessment Tool (SWAT) distributed hydrological model to simulate runoff in the upper reaches of the Hailar Basin (NE China) and to analyze quantitatively the impacts of climate change and land-use change on runoff by setting different scenarios. Two periods, i.e., the reference period (before 1988) and the interference period (after 1988), were identiﬁed based on long-term runoff datasets. In comparison with the reference period, the contribution rates of both climate change and land-use change to runoff change in the Hailar Basin during the interference period were 83.58% and 16.42%, respectively. The simulation analysis of climate change scenarios with differential precipitation and temperature changes suggested that runoff changes are correlated positively with precipitation change and that the impact of precipitation change on runoff is stronger than that of temperature. Under different economic

Yuhui Yan Baolin Xue (corresponding author) Yinglan A Wenchao Sun Hanwen Zhang College of Water Sciences, Beijing Normal University, Beijing 100875, China E-mail: xuebl@bnu.edu.cn Baolin Xue Yinglan A Wenchao Sun Beijing Key Laboratory of Urban Hydrological Cycle and Sponge City Technology, Beijing, China

development scenarios adopted, land use was predicted to have a considerable impact on runoff. The expansion of forests within the basin might induce decreased runoff owing to enhanced evapotranspiration. Key words

| climate change, Hailar Basin, land cover/use change, runoff, SWAT model

INTRODUCTION Water resources form the foundation of sustainable socio-

Piao et al. ). The water cycle in many watersheds has

economic development (Barnett et al. ). However,

been affected considerably, and runoff in major watersheds

owing to continued global extreme climatic events and the

has shown a rapid decrease (Bao et al. ; Wang et al.

negative inﬂuence of human activities, China is facing

a). Human activities affect the hydrological cycle and

water resource problems that are becoming increasingly

the formation process of water resources by changing the

severe and that represent an important constraining factor

mode of land use to varying degrees (Wheater & Evans

on China’s future sustainable socioeconomic development

; Sterling et al. ), which can result in a series of pro-

(Barnett et al. ; Piao et al. ; Voeroesmarty et al.

blems such as wetland shrinkage and groundwater funneling

). In recent decades, under the inﬂuence of global cli-

(Kaushal et al. ; Sterling et al. ; Li et al. ). There-

mate change, the climate in northern China has shown an

fore, it is of considerable scientiﬁc and practical importance

obvious trend of warming and drying (Fang et al. ;

to investigate the impact of climate change and land-use change on the hydrological cycle to resolve water resources

problems in a changing environment. Runoff is a vital link in the hydrological cycle that is also important in relation to the allocation of water resources

977

Y. Yan et al.

Climate and land-use impacts on runoff variation

Hydrology Research

51.5

2020

within a basin (Milly et al. ). Changes in runoff directly

The impact of climate change and land-use change on

affect life and production activities in a basin (Piao et al.

runoff varies spatially, and therefore, research is usually con-

). Therefore, it is of considerable importance to under-

ducted in speciﬁc watersheds (A et al. a). Many studies

take quantitative research on the impact of climate and

have assessed the impact of climate change and land-use

land-use changes on runoff. Methods commonly used for

change on runoff in various watersheds in China. Research

the quantitative analysis of the impact of environmental

in an agricultural catchment of a tributary of the Jinghe

changes on runoff can be divided into three categories: com-

River on the Loess Plateau highlighted that climate variabil-

parative basin tests, statistical analysis methods, and

ity had greater inﬂuence than land-use change on the

hydrological model simulations (Mishra & Singh ).

surface hydrological processes during 1981–2000 (Li et al.

The comparative watershed method is used for the manual

). Results of a study in the Dongjiang River Basin indi-

evaluation of the human impact on runoff by changing the

cated that climate variability and human activities each

natural geographical conditions (one or more watershed

accounted for approximately 50% of the runoff change in

characteristics) of the test watershed. However, this

the low-ﬂow period (Zhou et al. ). Analysis in three

approach is usually considered best for studying the effect

main tributary subcatchments in the Yellow River Basin

of climate change in small watersheds but it is difﬁcult to

illustrated that climate change accounts for only 8% of the

ﬁnd two similar medium- or large-sized watersheds, and

total decline in mean annual runoff, whereas human activi-

even the same watershed might undergo notable changes

ties account for 92% (Li et al. a). However, few studies

at different stages (Lorup et al. ). Statistical analysis

have conducted relevant research on the Hailar Basin.

can be used to analyze the trends of the change of hydrome-

The Hailar Basin is a typical northern inland river basin

teorological data, but it cannot consider the spatial

in NE China. In recent years, it has encountered increas-

heterogeneity of watersheds or the mechanisms via which

ingly serious water shortage problems (Chi et al. ;

land-use change and climate change might affect runoff

Wang et al. b). Therefore, it is of considerable urgency

change in watersheds (Gampe et al. ; Ishida et al.

that research be conducted to alleviate the water shortage

). Therefore, comprehensive physically based tools are

problems in this basin (Wang et al. a). This study used

needed to obtain as much information as possible from

the SWAT model to simulate runoff in the Hailar Basin,

limited existing data (Li et al. ). Hydrological models

and the impact of climate change and land-use change on

provide a framework for conceptualizing and studying

runoff was analyzed quantitatively. The main objectives of

relationships (Wang & Xu ). By linking model par-

this study were as follows: (1) to determine abrupt change

ameters

surface

points in runoff according to runoff time series, (2) to ana-

features, hydrological models can establish relationships

lyze quantitatively the contribution rates of climate change

among climate, human activities, and runoff (Leavesley

and land-use change to runoff, and (3) to investigate runoff

; Jothityangkoon et al. ). For example, Cuo et al.

response under different climate change and land-use

() used the variable inﬁltration capacity hydrological

change scenarios. The ultimate aim was to provide support

model to analyze quantitatively the impact of climate

for strategic planning and allocation of water resources in

change and land-use change on runoff in the upper reaches

the Hailar Basin.

directly

with

physically

observable

of the Yellow River Basin (China). Mango et al. () used the Soil and Water Assessment Tool (SWAT) distributed hydrological model to analyze quantitatively the impact of

STUDY AREA AND DATA

climate change and land-use change on runoff in the Mala River Basin (Kenya). In hydrological model simulations,

Study area

the mechanism of runoff formation is consistent with or without environmental change and no detailed data of

The Hailar Basin (47 380 –50 160 N, 117 430 –122 20 E) is

human activities are needed (Legesse et al. ). Therefore,

located in the northeast of the Inner Mongolia Autonomous

this study chose the hydrological model simulation method.

Region to the southwest of the city of Hulunbeier (Figure 1).

978

Figure 1

Y. Yan et al.

Climate and land-use impacts on runoff variation

Hydrology Research

51.5

2020

Map showing location of the upper Hailar River Basin.

It has a temperate continental monsoon climate, and it

period is derived primarily from snowmelt and precipitation.

covers an area of 54,500 km . The basin is located at the

The summer ﬂood season usually occurs during June–

junction of the western slopes of the Daxing’anling moun-

October, which is the period with the most concentrated

tains and the northeastern edge of the Inner Mongolia

precipitation (Xue et al. b; A et al. b).

High Plain (Wang et al. b). It is a fan-shaped basin that has large topographic ﬂuctuation (Han et al. a).

Data sources

The general trend of elevation (range: 536–1,706 m) is from low in the west to high in the east. The upper reaches

The data used in this study were divided into two parts. Part

of the Hailar Basin constitute the main area of the basin

1 data were used to analyze the water resources situation of

(Han et al. b). There are two ﬂood seasons annually: a

the upper reaches of the Hailar Basin and to determine the

spring ﬂood season and a summer ﬂood season. The

relationship between runoff and climate factors. Part 2 were

spring ﬂood season that usually occurs during March–May

the input data required by the SWAT model. The input data

reaches its peak in May (Fang et al. b). Runoff in this

of the SWAT model also included two main parts: spatial

979

Y. Yan et al.

Climate and land-use impacts on runoff variation

Hydrology Research

51.5

2020

data and attribute data. Spatial data mainly included digital

to soil parameters of the United States Geological Survey

elevation model data (90 × 90 m), land use/cover data

standard. The soil distribution data in the Hailar Basin are

(1,000 × 1,000 m), soil distribution data, and spatial distri-

shown in Figure 2(b). This study used daily meteorological

bution data of the hydrological stations and meteorological

and runoff data from 1980 to 2012. Meteorological data

stations. Attribute data mainly included soil type data,

included daily precipitation, maximum and minimum temp-

meteorological data, and hydrological data. Index tables of

erature, humidity, and wind speed.

land-use type and soil type were established in the modeling process to ensure the required data in the model corresponded to each database. Digital elevation model data: ﬁrst, geometric correction

METHODOLOGY

was performed, and then clipping and projection transformation were undertaken using ArcGIS. Land use/cover data:

Mutation analysis

ﬁrst, the original data were downloaded from the Resources and Environment Science Data Center of the Chinese

Mann–Kendall mutation test

Academy of Sciences, and reclassiﬁcation and projection transformation performed using ArcGIS. The data were

The Mann–Kendall (M-K) test is a widely used non-

divided into six categories: forest land, grassland, water

parametric test method recommended by the World

area, urban land, unused land, and cultivated land, as

Meteorological Organization. In recent years, many studies

shown in Figure 2(a). Soil type data and soil attribute data:

have adopted the M-K method to analyze the changes of

these data were obtained from the Harmonized World Soil

trends in time series of precipitation, runoff, temperature,

Database, which contains a large number of soil parameters.

and water quality (Hamed & Rao ; Yue et al. ).

The data are presented in a gridded format using the

In this study, the M-K catastrophe test was used to ana-

WGS1984 coordinate system. The soil classiﬁcation system

lyze the catastrophe of the runoff time series. The M-K

adopted is mainly FAO-90. As the data provided in the data-

mutation test deﬁnes statistical variables by constructing

base are international standards, the SPAW software was

order columns. Assuming time series X ¼ {x1 , x2 , . . . , xn },

required to convert the data from international standards

ri represents the cumulative number of the ith sample, and

Figure 2

(a) Distribution of land use/cover types and (b) distribution of soil types.

Y. Yan et al.

980

Climate and land-use impacts on runoff variation

Hydrology Research

k X

ri ¼

1 0

2020

For a time series, the principle of the sliding t-test is to ri

(k ¼ 2, 3, 4, . . . , n)

(1)

i¼1

51.5

Sliding t-test technique

xi is greater than xj (1 j i), we have

sk ¼

extract two subsequences of the main time series and then to test whether there is a signiﬁcant difference between

xi > xj (j ¼ 2, 3, . . . , i) other

the average values of those two subsequences (Machiwal (2)

& Jha ). If there is a signiﬁcant difference, the sequence is considered to have mutation (Zhao et al. ). Assuming

Under the assumption that the time series is random and

time series X ¼ {x1 , x2 , . . . , xn }. Then, by taking a certain time in the series as a reference point and taking

independent, we have the following:

subsequence sk E(sk ) UFk ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Var(sk )

and

subsequence

X2 ¼ {x1 , x2 , . . . , xn } forward and backward, respectively, (k ¼ 1, 2, . . . , n)

(3)

based on the reference point, statistic t can be obtained based on the two subsequences as follows:

When the elements x1 , x2 , . . . , xn are independent of t¼

each other and continuously and uniformly distributed: E(sk ) ¼ k(k 1)=4

(4)

Var(sk ) ¼ k(k 1)(2k þ 5)=72

(5)

UFi is a standard normal distribution. Given signiﬁcance level α, if |UFi| > U, there is an obvious trend change in the sequence, and the critical value of UF and UB is ±1.96. Arranging time series X in the reverse order and then performing the calculation according to the above equation,

2 1 x x rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 1 s þ n1 n2

(7)

P 1 1 ¼ (1=n1 ) ni¼1 2 ¼ (1=n2 ) In the above equation, x xi , x q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi Pn2 (n1 s21 þ n2 s22 )=(n1 þ n2 2). i¼1 xi , and s ¼ Then, statistic t obeys the t distribution with n1 þ n2 2 degrees of freedom. The specific steps of the sliding t-test for mutation are described in the following. 1. On the basis of the determination of the reference point, the lengths of subsequences n1 and n2 are

we have

X1 ¼ {x1 , x2 , . . . , xn }

determined. Normally, the lengths of the two subseUBk ¼ UFk 0 k ¼ nþ1 k

(k ¼ 1, 2, . . . , n)

(6)

quences are taken as the same, i.e., n1 ¼ n2. Then, subsequences n1 and n2 are taken forward and backward, respectively.

By analyzing the statistical sequences UFk and UBk, the

2. By sliding the reference point backward in turn, taking

trend change of sequence X can be analyzed further, which

out the corresponding subsequences, and calculating

allows the mutation time to be deﬁned and the mutation

the

region to be identiﬁed. If the value of UFk is >0, it indicates

statistics can be obtained.

corresponding

statistics,

the

n (n1 þ n2) þ 1

that the sequence shows an upward trend and vice versa.

3. The obtained statistics are arranged in sequence to obtain

When the value exceeds the critical line, it indicates that

the statistics sequence, the signiﬁcance level @ is selected,

the upward or downward trend is signiﬁcant. If the UFk

and the corresponding standard statistics t@ are obtained

and UBk curves have intersection points and the intersection

from a lookup table. The statistics sequence is drawn into

points are between the critical straight lines, then the time

a polyline graph. If jti j > t@ , the datum point can be

corresponding to the intersection points is the time when

adjudged the abrupt change point of the sequence

the abrupt change is considered to start (Hamed ).

(Dittmer ; Zhao et al. b).

981

Y. Yan et al.

Climate and land-use impacts on runoff variation

Hydrology Research

SWAT model

51.5

2020

reaches of the Hailar Basin based on three indices: the decision coefﬁcient R 2, Nash–Sutcliffe coefﬁcient Ens, and

The SWAT model is a distributed hydrological model devel-

relative error RE (Bhatta et al. ). The expressions of

oped by the Agricultural Research Center of the United

the three indices are as follows:

States Department of Agriculture (Arnold et al. ). The SWAT model is used widely for the simulation of hydrological cycle processes because its strong physical basis gives it

2 sim (Qobs Qobs Qsim ave )(Qi ave ) i 2 Pn 2 obs sim Qobs Qsim ave ) ave ) i¼1 (Qi i¼1 (Qi

the ability to simultaneously integrate the effects of topography, soil, land use, and weather (Yang et al. ). The model can simulate runoff change under different climate

"P Ens ¼ 1

change conditions, land-use conditions, soil conditions, and watershed management conditions; thus, the SWAT model was selected for the runoff simulations conducted

i¼1

R2 ¼ P n

Pn RE ¼

n obs i¼1 (Qi Pn obs i¼1 (Qi

Qsim i )

(9)

Qobs ave )

(Qobs Qsim ) Pn i obs i i¼1 Qi

i¼1

(10)

(11)

in this study (Moriasi et al. ). The model is calculated is the measured runoff value, In the above equations, Qobs i

according to the following water balance equation:

Qsim i SWt ¼ SW0 þ

t X

(Rday Qsurf Ea Wseep Qgw )

(8)

i¼1

where SW0 is the initial soil water content on the ith day, Rday is the precipitation on the ith day, Qsurf is the surface runoff on the ith day, Ea is the soil evaporation and plant transpiration on the ith day, Wseep is the seepage ﬂow on the ith day, and Qgw is the amount of groundwater on the ith day.

is the simulated runoff value, Qobs ave is the measured aver-

age runoff value, and Qsim ave is the simulated average runoff value. The R 2 values (range: 0–1) represent the ﬁtting degree between the simulated and measured values. The Nash–Sutcliffe coefﬁcient (range: 0–1) is used to assess the quality of the hydrological model simulation results. Only when Ens > 0.5 can the simulation results be accepted. A value of Ens close to 1 indicates that the model simulation quality is good and its credibility is high. If the value of the relative error RE is controlled to within 25%, the simulation results of the model can be considered within an acceptable

Model calibration and validation

range (Moriasi et al. ; Asl-Rousta et al. ).

In this study, the simulation of daily runoff was performed

Quantiﬁcation of climate and land-use contributions to runoff change

ﬁrst. Then, the SWAT-CUP software was used to calibrate and verify the model. The data used included land-use data in 1980 and 2000 and meteorological data from 1980

Based on the SWAT model, this study used scenario analysis

to 2012. Based on runoff data recorded at the Bahou

to separate the inﬂuences of various factors on runoff, i.e.,

Station, the Latin Hypercube One factor At a Time

assuming that climatic factors or human activity factors

(LH-OAT) method was used to analyze the sensitivity of

remain constant and that another factor changes accord-

the parameters, and those parameters with strong sensitivity

ingly, the contribution rate of this factor to runoff can be

were selected as adjustment parameters (van Griensven

analyzed quantitatively.

et al. ; Arnold et al. ). The SUFI-2 algorithm was

This study used the M-K mutation test and sliding t-test

used to determine the optimal ranges and values of the par-

to determine the time point of runoff mutation in the water-

ameters through iterative calculation, and the optimal

shed. The year before the time point of abrupt change was

values of the parameters were introduced into the model

deﬁned as a period of natural runoff, i.e., runoff was

through internal adjustment (Anoh et al. ). In this

assumed undisturbed by climate change and land-use

study, the effective evaluation method was used to evaluate

change, whereas runoff after the time point of abrupt

the simulation accuracy of the SWAT model in the upper

change was considered disturbed by both climate change

Y. Yan et al.

982

Table 1

Climate and land-use impacts on runoff variation

Hydrology Research

51.5

2020

based on the following scenarios: current precipitation

Model simulation scenario settings

unchanged, 10% and 20% increase in precipitation, 10%

Scenario

Land use/cover data

Meteorological data

1980s

1980–1988

unchanged, 1 C and 2 C increase in temperature, and

1980s

1989–2012

1 C and 2 C decrease in temperature.

2000s

1980–1988

2000s

1989–2012

and 20% decrease in precipitation, current temperature

To explore the inﬂuence of land use on runoff, this study used the cellular automata–Markov (CA–Markov) model to simulate land-use development scenarios based on existing land-use datasets of the Hailar Basin. Land use in the

and land-use change. To quantify the contributions of land-

upper reaches of the Hailar Basin was simulated for 2030

use change and climate change to runoff change, four scen-

and 2050, and the associated runoff was investigated accord-

arios were established (Table 1). A comparison of the simulated runoff under scenario 2 with that under scenario

ing to historical development trends and existing problems within the Hailar Basin.

1 allowed the analysis of the impact of climate change on runoff. Similarly, a comparison of the simulated runoff under scenario 3 with that under scenario 1 allowed the

Land use/land cover prediction

analysis of the impact of land-use change on runoff. A comparison of the runoff simulated under scenario 4 with that

This study used the CA–Markov model to predict future land

under scenario 1 allowed the analysis of the impact of

use/land cover. The CA–Markov model is a powerful tool

both climate change and land-use change on runoff. Thus,

that combines the advantages of the Markov chain and cel-

the contribution rates of climate change and land-use

lular automata.

change to runoff change were obtained.

The CA–Markov model describes land-use change from

In this study, Q1, Q2, Q3, and Q4 were used to represent

one period to another, which allows predictions of the

the average annual runoff simulated under scenarios 1, 2, 3,

future trend of land-use change. The following formula can

and 4, respectively. Thus, Q2 Q1 represents the impact of

be used to predict land use:

climate change on runoff, Q3 Q1 represents the impact of land use/cover change (land-use change) on runoff, and

Stþi ¼ Pij St

(14)

Q4 Q1 represents the total change of runoff within the basin. In this paper, we deﬁne the following:

where St and Stþi are the states of the land-use structure at time t and t þ i, respectively and Pij is the state-transition

Q2 Q1 ωcl ¼ Q4 Q1

(12)

Q3 Q1 Q4 Q1

(13)

ωh ¼

matrix. A cellular automaton represents a gridded dynamics model with strong spatial computing capability. The CA– Markov model can be expressed as follows: Stþi ¼ f(St , N)

(15)

Scenario setting and model analysis where S is a ﬁnite discrete state set element, N is a cellular To investigate the impact of climate change on runoff in the

neighborhood, t and t þ i are different moments, and f is

upper reaches of the Hailar Basin, the change values of pre-

the cell transformation rule in local space.

cipitation and temperature were determined according to

The model can simulate the temporal and spatial evol-

the possible range of future climate change. In simulating

utions of complex systems, and it has been used widely in

runoff in the upper reaches of the Hailar Basin, 25 different

many studies (Mitsova et al. ; Sang et al. ). The

climate change combination programs were considered

detailed parameters and steps of land-use prediction using

Y. Yan et al.

983

Climate and land-use impacts on runoff variation

Hydrology Research

51.5

2020

the CA–Markov model are as follows. (1) Data format con-

upward or downward trend is considered signiﬁcant. The

version and reclassiﬁcation are performed to obtain ﬁxed

range beyond the critical line is determined as the time

land-use types. (2) A state-transition probability matrix and

zone in which the abrupt change occurs. If there is an inter-

a transition area matrix are obtained using the CA–

section point between the UFk and UBk curves and that

Markov module. (3) A transitional ﬁtness image set is estab-

intersection point lies between the critical lines, the time

lished. (4) The CA ﬁlter and number of cycles are

corresponding to the intersection point is the time at

determined. (5) The accuracy of the predicted image is eval-

which the abrupt change starts. As shown in Figure 4(a),

uated according to the actual image. In this study, a map of

the UFk value after 1995 is <0, which indicates that the

land use in 2000 was used as the basic image, and an assem-

runoff sequence at the Bahou station began to decline con-

bly transition probability matrix of land-use maps in 2005

tinuously after 1995. In 2003, the UFk value was lower

and 2015 was used to predict the land-use situation in

than 1.96, indicating that runoff had a signiﬁcant decreas-

2030 and 2050.

ing trend based on the signiﬁcance level test of 0.05. As shown in Figure 4(a), during 1980–1995, the UFk and UBk curves had intersection points between the critical lines at

RESULTS

around 1980, 1982, 1986, 1988, and 1992. Therefore, the sliding t-test method was used to determine the mutation year in this study.

Runoff trend analysis

In this study, the sliding t-test method was also used to The linear analysis and 5-year moving average curve of

analyze the Bahou Station runoff time series. To avoid vari-

runoff during 1980–2012 at the Bahou Station at the outlet

able point drift caused by different subsequence lengths, the

of the upper reaches of the Hailar Basin are shown in

sliding length was changed repeatedly. Finally, it was deter-

Figure 3. In the past 30 years, annual runoff at the Bahou

mined that the length of the two sliding sequences should be

Station has ﬂuctuated but generally diminished with a rate

4, i.e., n1 ¼ n2 ¼ 4, and 33 statistics relating to the Bahou

that has become more obvious since 1990. Annual runoff

Station formed the corresponding statistical sequence. If

at the Bahou Station has ranged from a maximum value of

the signiﬁcance level @ ¼ 0:05 were selected, the statistic

69.82 × 10 m 8

in 1984 to a minimum value of 10.28 × 8

10 m in 2007 with a trend of 0.24 × 10 m /a.

t–t(6) and the critical value were found to be t0.05(6) ¼ 1.943, and each statistical sequence and critical value are

In this study, the M-K method and sliding t-test method

plotted on the graph for the mutation analysis. The analysis

were used to test for catastrophe in the runoff sequence. UFk

of Figure 4(b) reveals that for the annual runoff series, the t

values >0 indicate that the sequence shows an upward trend

statistics of the Bahou Station exceed the critical values in

and vice versa. When the values exceed the critical line, the

two places: 1987 and 1988, both of which are possible abrupt years. As the M-K mutation test identiﬁed 1988 as the intersection point of the UFk and UBk curves, it can be considered that mutation occurred in the annual runoff at the Bahou Station in 1988. As water resources are affected by various climate change factors, e.g., rising temperature and extreme weather, runoff of the major rivers in northern China currently shows an overall downward trend. Therefore, the identiﬁed sudden change is probably attributable to climate change because temperature rise leads to increased evapotranspiration. Based on these two methods, the mutation point was set to 1988, and the study period was divided into the reference period (before 1988) and the inter-

Figure 3

Results of the trend analysis of runoff at the Bahou Station.

ference period (after 1988).

Y. Yan et al.

984

Figure 4

Climate and land-use impacts on runoff variation

Hydrology Research

51.5

2020

Results of (a) the M-K mutation test and (b) the moving t-test.

Model calibration and validation

Table 2

Parameter sensitivity analysis and model calibration are clo-

Parameter

sely linked, and both are essential processes for runoff

name

simulation using hydrological models. The SWAT-CUP soft-

GW_DELAY Groundwater delay factor

(30,450) 45.75

ALPHA_BF

Base-ﬂow recession constant

(0,1)

SMFMN

Snowmelt coefﬁcient on 3 December 21

( 10,10) 2.42

CANMX

Maximum canopy storage

(0,100)

76.25

SOL_AWC

Available water capacity 5 of the soil layer

(0,1)

0.15

SMFMX

Snowmelt coefﬁcient on 6 June 21

( 10,10) 6.75

GWQMN

Threshold depth of water in the shallow aquifer required for return ﬂow to occur

Parameter sensitivity analysis

The most

ware can be used to perform sensitivity analysis and calibration on relevant parameters for a runoff simulation. Combined with the actual situation of the basin, the LHOAT method provided in SWAT-CUP was used in this study to select parameters related to runoff simulation in the northern region for sensitivity analysis and calibration, as shown in Table 2. Through

the

sensitivity analysis

the

SWAT

model, 11 parameters with high sensitivity were selected to calibrate and verify the model (Table 2). In this study, the preheating period of the model was taken as 1988–1989, 1990–2006 was taken as the calibration period, and 2007–2012 was taken as the validation period. The SWAT-CUP software was used to calibrate the model parameters, and the monthly runoff at the Bahou Station was used to calibrate and verify the

Rank Ranges

(0,2)

value

0.90

0.96

CN2

Curve number

(20,100) 46

SMTMP

Snow melting

( 10,10) 2.75

ESCO

Soil evaporation compensation coefﬁcient

(0.01,1)

SFTMP

Snow temperature

( 10,10) 3.63

model. In the calibration period, the R 2, Ens, and RE values of

suitable Variable name

0.46

the simulated monthly runoff at the Bahou Station were 0.75, 0.71, and 20.8%, respectively, while the corresponding values in the validation period were 0.74, 0.70, and

Contribution of land use and climate change to runoff

20.3%, respectively, proving the suitability of the SWAT

variation

model for runoff simulation in the upper reaches of the Hailar Basin. The simulation results are shown in

According to the abrupt change years, the study period was

Figures 5 and 6.

divided into the reference period (before 1988) and

985

Y. Yan et al.

Climate and land-use impacts on runoff variation

Hydrology Research

Figure 5

Correlation between simulated and measured values of monthly streamﬂow at the Bahou Station: (a) calibration period and (b) validation period.

Figure 6

Measured and simulated monthly streamﬂow graphs (Bahou Station) for the calibration period and the validation period.

interference period (after 1988). Using the meteorological

Table 3

data and land-use data of different periods and by establishing different scenarios, the contribution rates of land-use change and climate change to runoff in the upper reaches of the Hailar Basin were calculated. As shown in Tables 3 and 4, the average annual runoff simulated under scenarios 1–4 was 125.43, 104.05, 121.23, and 99.85 m3/s, respect-

51.5

2020

Comprehensive scenario simulation results

Inﬂuence of land-use

Inﬂuence of climate

Scenario runoff (m3/s)

change on runoff (m3/s)

125.43

–

104.05

–

21.38

121.23

4.2

–

99.85

4.2

21.38

Annul average

ively. The analysis of scenarios 1 and 2 suggests that under the inﬂuence of climate change, annual average runoff has

suggests that under the inﬂuence of land-use change,

decreased by 21.38 m3/s. The analysis of scenarios 1 and 3

annual average runoff has decreased by 4.2 m3/s. The

Y. Yan et al.

986

Table 4

Climate and land-use impacts on runoff variation

Hydrology Research

51.5

2020

Simulated response of land-use change and climate change

Time

Simulated annul average runoff (m3/s)

Varied runoff (m3/s)

Inﬂuence of land-use change on runoff (m3/s)

Inﬂuence of climate change on runoff(m3/s)

Reference period

1980–1988

125.43

–

Interference period

1989–2012

99.85

25.58

4.2(16.42%)

21.38(83.58%)

Period

analysis of scenarios 1 and 4 suggests that under the joint

Table 6

Relative variation of mean annual runoff for different scenarios (%)

inﬂuence of climate change and land-use change, annual average runoff has decreased by 25.58 m3/s. Speciﬁcally, Q2 Q1 ¼ 21.38 m3/s,

Q3 Q1 ¼ 4.2 m3/s,

Q4 Q1 ¼

Variations in precipitation Variations in temperature

ΔP ¼ 20%

ΔP ¼ 10%

ΔP ¼ 0

ΔP ¼ 10%

ΔP ¼ 20%

25.58 m3/s, ωcl ¼ 83:58%, and ωh ¼ 16:42%. The contri-

ΔT ¼ þ2

19.82

7.15

7.25

16.86

34.70

bution rate of climate change (land-use change) to runoff

ΔT ¼ þ1

23.59

10.72

6.50

13.96

30.86

change in the basin is 83.58% (16.42%); consequently, it

ΔT ¼ 0

26.30

13.34

11.40

27.59

can be concluded that climate change has a greater impact

ΔT ¼ 1

28.47

16.08

7.20

9.45

25.20

than land-use change on runoff change in the basin.

ΔT ¼ 2

29.79

18.38

12.46

7.92

23.14

Scenario simulation of climate change and land

greater the increase in runoff. Values when ΔP ¼ 0 indicate

use/cover change

the response of runoff to temperature change when precipitation is constant. It can be seen that runoff decreases as

Climate change scenarios

temperature increases, and the more the temperature increases,

As shown in Table 5, under the condition of constant basin temperature, surface runoff will increase with the increase of precipitation. Under the condition of unchanged precipitation, surface runoff will change with the change of temperature. In all temperature-drop scenarios, surface runoff shows an increasing trend, whereas in all temperature-rise scenarios, surface runoff shows a decreasing trend. As shown in Table 6, values when ΔT ¼ 0 represent the

the more runoff decreases. Values when neither ΔT nor ΔP are zero indicate the runoff response effect when both temperature and precipitation change. It can be seen that when temperature decreases and precipitation increases, the increase in runoff is greatest. When temperature increases and precipitation decreases, a decrease in runoff is greatest. When precipitation increases by more than 10%, runoff increases markedly, indicating that precipitation has a more signiﬁcant impact on runoff.

response of runoff to changed precipitation at constant temperature. It can be seen that the increase of precipitation

Land-use change scenarios

increases runoff, and the more precipitation increases, the According to the trend of land-use change in the Hailar Table 5

Relative variation of mean annual runoff for different scenarios (m3/s)

with current economic development and ecological protection in this region, this study simulated land-use scenarios

Variations in precipitation Variations in temperature P(1 þ 20%) P(1 þ 10%) P

Basin, and with reference to the problems encountered

P(1–10%) P(1–20%)

T þ 2 C

24.11

8.70

8.82 20.51 42.20

T þ 1 C

28.69

13.04

7.91 16.98 37.57

31.99

16.23

0.00

13.87 33.56

T 1 C

34.63

19.56

8.76

11.49 30.65

T 2 C

36.23

22.36

15.16 9.63

28.15

in 2030 and 2050 using the CA–Markov model and analyzed the impact of future land-use change on watershed runoff. In simulating and predicting future land-use situations, three development models were established: natural development, ecological protection, and economic development. The distribution maps of land use in 2030 and 2050 under these three economic development models are shown in Figure 7.

Y. Yan et al.

987

Figure 7

Climate and land-use impacts on runoff variation

Hydrology Research

51.5

2020

Distribution of land-use types under different economic development scenarios: (a) natural growth scenario in 2030, (b) natural growth scenario in 2050, (c) ecological protection scenario in 2030, (d) ecological protection scenario in 2050, (e) economic development scenario in 2030, and (f) economic development scenario in 2050.

Y. Yan et al.

988

Table 7

Climate and land-use impacts on runoff variation

Hydrology Research

51.5

2020

Proportion of land-use types in different scenarios (%)

Natural growth Land-use type

Ecological protection

Economic development

2030 (%)

2050 (%)

2030 (%)

2050 (%)

2030 (%)

2050 (%)

Cultivated land

7.902

7.983

7.728

7.658

7.937

8.228

Forest

34.294

34.501

34.294

35.966

34.190

34.094

Meadow

50.456

50.027

50.560

50.108

50.454

49.864

Water

0.720

0.895

0.721

0.895

0.686

0.895

Constructed land

0.639

0.779

0.743

0.860

0.744

1.104

Unused land

5.989

5.815

5.954

4.513

5.989

5.815

It can be seen from Table 7 that future areas of various

the proportion of the grassland area will still decrease in

land-use types will increase or decrease to varying degrees.

2030 and 2050; however, in comparison with the natural

In the case of natural growth, from 2030 to 2050, the pro-

growth scenario, both an increase in the proportion of the

portions of cultivated land, forest land, water areas, and

forest area and a decrease in runoff will be greater. Under

construction land will increase, while the proportions of grass-

the condition of economic development, the proportion of

land and unused land will decrease. In the case of ecological

the forest area will increase slightly in 2030 and 2050, the

protection, from 2030 to 2050, the proportions of forest land,

proportion of the grassland area will decrease greatly, and

water areas, and construction land will increase, while the pro-

the proportion of cultivated land will increase markedly;

portions of cultivated land, grassland, and unused land will

however, runoff in the basin will still show a decreasing

decrease. In the context of economic development, from

trend. This shows that the increase/decrease of the forest

2030 to 2050, the proportions of cultivated land, water areas,

land area has a reasonably direct impact on the increase/

and construction land will increase, while the proportions of

decrease of runoff. Forest land can improve both the

forest land, grassland, and unused land will decrease.

regional climate and the soil environment while reducing

The runoff simulation results under the different land-

the surface temperature and direct evaporation of water to

use scenarios are shown in Table 8. Under the natural

a certain extent, which can play a role in conserving water

growth scenario, ecological protection scenario, and econ-

resources and intercepting runoff (Xue et al. b). In

omic development scenario, runoff shows a decreasing

future planning, the proportion of ecological land (e.g.,

trend, but the degree of runoff reduction differs between

forest land and grassland) should be adjusted appropriately

the scenarios. Under the condition of natural growth, the

to alleviate the trend of decreasing runoff within the basin.

proportion of the forest area will increase, the proportion of the grassland area will decrease, and runoff will decrease in 2030 and 2050. Under the condition of ecological protec-

DISCUSSION

tion, the proportion of the forest area will still increase and According to the derived results, runoff in the Hailar Basin has shown a downward trend over the previous 30 years, Table 8

2030

2050

Change of runoff under different land-use scenarios

Land-use change scenarios

Rate of change in runoff (%)

Natural growth scenario Ecological protection scenario Economic development scenario

0.515 0.704 0.371

Natural growth scenario Ecological protection scenario Economic development scenario

0.942 1.408 0.152

with a more obvious downward trend since around 1990. A decrease in runoff in the basin is attributable to both decreased rainfall and changes of the underlying surface conditions (Xue et al. a; Zhao et al. a), e.g., increase of forest land and grassland areas, increase of forest land coverage rate, greater retention of surface runoff, and increase of surface evaporation in the basin (Wang et al. ). According to the results of the abrupt change test,

989

Y. Yan et al.

Climate and land-use impacts on runoff variation

Hydrology Research

51.5

2020

annual runoff in the basin experienced an abrupt change in

). Precipitation has a direct effect on runoff change,

1988. According to the UF curve, runoff has shown a decreas-

whereas the effect of temperature on runoff change is indir-

ing trend since 1988. Moreover, the UF curve exceeds the

ect (Fu et al. ; Li et al. b). The simulation scenarios

0.05 critical line after 2003, indicating that the trend of

of land-use change showed that in 2030 and 2050, under the

decrease is very signiﬁcant. The main reason might be that

three different scenarios of natural development, ecological

climate change has caused rainfall in the basin to change

development, and economic development, runoff in the

abruptly (Fang et al. a). However, since China

basin will show a decreasing trend to differing degrees. The

implemented the policy of returning farmland to forest in

main reason might be an increase in the proportion of the

the 1990s, the rate of coverage of forest land in the basin

forest land area and a decrease in the proportion of the grass-

has increased, which will have increased interception by veg-

land area (Xue et al. a). An increase in the forest land

etation and delayed runoff (Wang et al. ; Xue et al. a).

area has the functions of conserving water and ﬁxing soil,

Based on the determined catastrophe years, this study

which have a certain effect on intercepting runoff (Wang

considered annual runoff data during 1980–2012 as the

et al. ; Xu et al. ). Moreover, an increase in the

research object and divided the research period into the

forest land area leads to increased evapotranspiration,

reference period (before 1988) and the interference period

which can improve both the regional microclimate and the

(after 1988). By setting different scenarios, it was established

soil environment to a certain extent, reducing watershed

that the contribution rate of climate change (land-use

runoff (Wang et al. ; El Kateb et al. ). Compared

change) to runoff change in the basin was 83.58%

with land-use change, climate change has a more signiﬁcant

(16.42%). The main land-use types in the upper reaches of

impact on runoff change in the Hailar Basin. To optimize

the Hailar Basin are grassland and forest land, and there

water resources allocation in the basin, the layout of land

has been little change in the underlying surface for many

use must be optimized to cope with the challenges brought

years. Moreover, the area is located in the grassland area

by climate change (Fang et al. ). It will be important to

of northern China, where the population density is low

consider changing the proportion of ecological land (e.g.,

and there is little impact from human activities (Xue et al.

forest land and grassland) to regulate and control runoff in

b; Wang et al. ). Consequently, given the global

the basin (Xue et al. b; Zhang et al. ).

increase in temperature and regional decrease of precipi-

Certain problems encountered in this research should

tation, climate change can be considered the main reason

be resolved in future work. First, this study used only

for runoff change in the Hailar Basin.

runoff data from 1980 to 2012 for the mutation analysis,

In order to further explore the runoff variations under

and the time series was incomplete. The next step would

climate change and land-use change, this paper sets up

be to collect a dataset that is more complete for specifying

different future scenarios for research based on the results

the mutation year. Second, the predictions of land use in

of runoff variation attribution analysis. In this study, differ-

2030 and 2050 were based on the CA–Markov model and

ent scenarios were established to explore the changes of

relied mainly on the transformation matrix derived using

watershed runoff under the effects of both climate change

land-use maps from 2005 to 2015. However, the implemen-

and land-use change. The climate change scenarios

tation of land-use management strategies by the Chinese

showed that runoff in the basin will increase with the

government might change in the future, which would have

increase of precipitation. For every 10% increase in precipi-

a certain impact on the land use/coverage scenarios pre-

tation, the runoff will increase by 13.34% on average. Runoff

dicted in this study for 2030 and 2050. Third, different

in the basin will decrease with an increase in air tempera-

combinations of temperature and rainfall were used to

ture. For every 1 C increase in air temperature, runoff will

drive the SWAT model. The climate change scenario was

decrease by 6.5% on average. Overall, the inﬂuence on

reasonably simple and did not start from actual conditions.

runoff of precipitation change is greater than that of temp-

In the next step, the climate forecast could be corrected

erature change. The main reason might be that runoff in

based on actual local climate conditions to provide the

this basin is replenished primarily by precipitation (A et al.

improved simulation of future runoff.

990

Y. Yan et al.

Climate and land-use impacts on runoff variation

Hydrology Research

CONCLUSIONS

51.5

2020

will be greater than the amount of runoff derived from snowmelt, which will lead to an overall reduction in

This study employed the SWAT distributed hydrological

runoff. A signiﬁcant positive correlation was found

model to assess the hydrological response of the Hailar

between the change in precipitation and runoff

Basin to climate change and land-use change, and to analyze

change. Moreover, it was found that when precipitation

its impact quantitatively. First, the runoff change trend at the

increases

Bahou Station on the upstream reaches of the Hailar Basin

(decreases)

and

temperature

decreases

(increases), an increase (decrease) in runoff is greatest.

over the previous 30 years was analyzed, and the year of

(5) The projected impact of land use on runoff under differ-

abrupt change in runoff was determined. The SWAT

ent development scenarios was different. Runoff under

model was used to simulate runoff, and the contribution

the three development modes of natural development,

rates of climate change and land-use change to surface

ecological protection, and economic development

runoff were separated. Based on the principle of the water

showed a decreasing trend. In comparison with the

balance, the contributions of these two factors to runoff

natural development scenario, the reduction in runoff

were calculated. Using combinations of different climate

under ecological protection will be greater. Under the

and land-use scenarios, the natural runoff under the inﬂu-

economic development scenario, runoff will still show

ence

a decreasing trend. It was established that forest land

climate

change

and

land-use

change

was

simulated. The derived results are presented below:

has a certain effect on decreasing runoff ﬂow, i.e., an increase in the forest area will lead to decreased surface

(1) Annual runoff at the Bahou Station generally showed a

runoff. In future planning, the proportion of ecological

downward trend that exhibited diminishing volatility.

land (e.g., forest land and grassland) should be adjusted

The long-term runoff sequence changed abruptly in

appropriately to alleviate the trend of decreasing runoff

1988. This study divided the runoff sequence into the

within the basin.

reference period (before 1988) and the interference period (after 1988). (2) The SWAT model demonstrated satisfactory applica-

ACKNOWLEDGEMENTS

bility to simulating runoff at the Bahou Station. The R 2, Ens, and RE values of the monthly runoff simulated

This study was supported by the National Natural Science

in the calibration period were 0.75, 0.71, and 20.8%,

Foundation of China (Grant No. 31670451) and the

respectively, while the corresponding values in the vali-

Fundamental Research Funds for the Central Universities

dation period were 0.74, 0.70, and 20.3%, respectively,

(No. 2017NT18). We thank James Buxton MSc from

proving that the model was suitable both for the calcu-

Liwen Bianji, Edanz Group China (www.liwenbianji.cn./

lation of the runoff contribution rate and for the

ac), for editing the English text of this manuscript.

simulation of runoff under different scenarios. (3) Compared with land-use change, climate change was found to have a more signiﬁcant impact on runoff

COMPETING FINANCIAL INTERESTS

change in the river basin. The contribution rate of climate change (land-use change) to runoff change in

The authors declare no competing ﬁnancial interest.

the river basin was 83.58% (16.42%). (4) According to different combinations of temperature and rainfall change, temperature decrease will lead to evaporation and snowmelt runoff decrease, which will result in increased surface runoff in the watershed. A rise in temperature will increase both snowmelt runoff and evaporation; however, the amount of evaporation

REFERENCES A, Y., Wang, G., Sun, W., Xue, B. & Kiem, A.  Stratiﬁcation response of soil water content during rainfall events under different rainfall patterns. Hydrological Processes 32, 3128–3139.

991

Y. Yan et al.

Climate and land-use impacts on runoff variation

A, Y., Wang, G., Liu, T., Shrestha, S., Xue, B. & Tan, Z. a Vertical variations of soil water and its controlling factors based on the structural equation model in a semi-arid grassland. Science of the Total Environment 691, 1016–1026. A, Y., Wang, G., Liu, T., Xue, B. & Kuczera, G. b Spatial variation of correlations between vertical soil water and evapotranspiration and their controlling factors in a semiarid region. Journal of Hydrology 574, 53–63. Anoh, K. A., Koua, T. J. J., Kouame, K. J., Jourda, J. P. & Laurent, F.  Modelling water flow in a complex watershed in humid a tropical area using SWAT: a case study of Taabo watershed in Ivory Coast. International Journal of River Basin Management 16, 157–167. Arnold, J. G., Srinivasan, R., Muttiah, R. S. & Williams, J. R.  Large area hydrologic modeling and assessment – part 1: model development. Journal of the American Water Resources Association 34, 73–89. Arnold, J. G., Moriasi, D. N., Gassman, P. W., Abbaspour, K. C., White, M. J., Srinivasan, R., Santhi, C., Harmel, R. D., van Griensven, A., Van Liew, M. W., Kannan, N. & Jha, M. K.  SWAT: model use, calibration, and validation. Transactions of the ASABE 55, 1491–1508. Asl-Rousta, B., Mousavi, S. J., Ehtiat, M. & Ahmadi, M.  SWAT-based hydrological modelling using model selection criteria. Water Resources Management 32, 2181–2197. Bao, Z., Zhang, J., Wang, G., Fu, G., He, R., Yan, X., Jin, J., Liu, Y. & Zhang, A.  Attribution for decreasing streamflow of the Haihe River basin, northern China: climate variability or human activities? Journal of Hydrology 460, 117–129. Barnett, T. P., Adam, J. C. & Lettenmaier, D. P.  Potential impacts of a warming climate on water availability in snowdominated regions. Nature 438, 303–309. Bhatta, B., Shrestha, S., Shrestha, P. K. & Talchabhadel, R.  Evaluation and application of a SWAT model to assess the climate change impact on the hydrology of the Himalayan River Basin. Catena 181, 104082. Chi, D., Wang, H., Li, X., Liu, H. & Li, X.  Estimation of the ecological water requirement for natural vegetation in the Ergune River basin in Northeastern China from 2001 to 2014. Ecological Indicators 92, 141–150. Cuo, L., Zhang, Y., Gao, Y., Hao, Z. & Cairang, L.  The impacts of climate change and land cover/use transition on the hydrology in the upper Yellow River Basin, China. Journal of Hydrology 502, 37–52. Dittmer, K.  Changing streamflow on Columbia basin tribal lands – climate change and salmon. Climatic Change 120, 627–641. El Kateb, H., Zhang, H., Zhang, P. & Mosandl, R.  Soil erosion and surface runoff on different vegetation covers and slope gradients: a field experiment in Southern Shaanxi Province, China. Catena 105, 1–10. Fang, J. Y., Chen, A. P., Peng, C. H., Zhao, S. Q. & Ci, L.  Changes in forest biomass carbon storage in China between 1949 and 1998. Science 292, 2320–2322.

Hydrology Research

51.5

2020

Fang, Q., Wang, G., Liu, T., Xue, B.-L. & A, Y. a Controls of carbon flux in a semi-arid grassland ecosystem experiencing wetland loss: vegetation patterns and environmental variables. Agricultural and Forest Meteorology 259, 196–210. Fang, Q., Wang, G., Xue, B., Liu, T. & Kiem, A. b How and to what extent does precipitation on multi-temporal scales and soil moisture at different depths determine carbon flux responses in a water-limited grassland ecosystem? Science of the Total Environment 635, 1255–1266. Fang, Q., Wang, G., Liu, T., Xue, B., Sun, W. & Shrestha, S.  Unraveling the sensitivity and nonlinear response of water use efficiency to the water-energy balance and underlying surface condition in a semiarid basin. Science of the Total Environment 699, 134405. Fu, Z., Jiang, H., Wang, G., A, Y., Xue, B. & Wang, H.  Effects of soil properties on plant community structure in a semi-arid grassland. Chinese Journal of Ecology 37, 823–830. Gampe, D., Nikulin, G. & Ludwig, R.  Using an ensemble of regional climate models to assess climate change impacts on water scarcity in European river basins. Science of the Total Environment 573, 1503–1518. Hamed, K. H.  Trend detection in hydrologic data: the MannKendall trend test under the scaling hypothesis. Journal of Hydrology 349, 350–363. Hamed, K. H. & Rao, A. R.  A modified Mann-Kendall trend test for autocorrelated data. Journal of Hydrology 204, 182–196. Han, D., Wang, G., Liu, T., Xue, B.-L., Kuczera, G. & Xu, X. a Hydroclimatic response of evapotranspiration partitioning to prolonged droughts in semiarid grassland. Journal of Hydrology 563, 766–777. Han, D., Wang, G., Xue, B., Liu, T., A, Y. & Xu, X. b Evaluation of semiarid grassland degradation in North China from multiple perspectives. Ecological Engineering 112, 41–50. Ishida, K., Ercan, A., Trinh, T., Kavvas, M. L., Ohara, N., Carr, K. & Anderson, M. L.  Analysis of future climate change impacts on snow distribution over mountainous watersheds in Northern California by means of a physically-based snow distribution model. Science of the Total Environment 645, 1065–1082. Jothityangkoon, C., Sivapalan, M. & Farmer, D. L.  Process controls of water balance variability in a large semi-arid catchment: downward approach to hydrological model development. Journal of Hydrology 254, 174–198. Kaushal, S. S., Groffman, P. M., Band, L. E., Elliott, E. M., Shields, C. A. & Kendall, C.  Tracking nonpoint source nitrogen pollution in human-impacted watersheds. Environmental Science & Technology 45, 8225–8232. Leavesley, G. H.  Modeling the effects of climate change on water resources – a review. Climatic Change 28, 159–177. Legesse, D., Vallet-Coulomb, C. & Gasse, F.  Hydrological response of a catchment to climate and land use changes in Tropical Africa: case study South Central Ethiopia. Journal of Hydrology 275, 67–85.

992

Y. Yan et al.

Climate and land-use impacts on runoff variation

Li, Z., Liu, W.-z., Zhang, X.-c. & Zheng, F.-l.  Impacts of land use change and climate variability on hydrology in an agricultural catchment on the Loess Plateau of China. Journal of Hydrology 377, 35–42. Li, X., Cheng, G., Ge, Y., Li, H., Han, F., Hu, X., Tian, W., Tian, Y., Pan, X., Nian, Y., Zhang, Y., Ran, Y., Zheng, Y., Gao, B., Yang, D., Zheng, C., Wang, X., Liu, S. & Cai, X.  Hydrological cycle in the Heihe River Basin and its implication for water resource management in endorheic basins. Journal of Geophysical Research – Atmospheres 123, 890–914. Li, H., Shi, C., Zhang, Y., Ning, T., Sun, P., Liu, X., Ma, X., Liu, W. & Collins, A. L. a Using the Budyko hypothesis for detecting and attributing changes in runoff to climate and vegetation change in the soft sandstone area of the middle Yellow River basin, China. Science of the Total Environment 703, 135588. Li, Q. , Wang, G., Wang, H., Shrestha, S., Xue, B-L., Sun, W. & Yu, J. b Macrozoobenthos variations in shallow connected lakes under the influence of intense hydrologic pulse changes. Journal of Hydrology 584, 124755, http://doi.org/ 10.1016/j.jhydrol.2020.124755. Lorup, J. K., Refsgaard, J. C. & Mazvimavi, D.  Assessing the effect of land use change on catchment runoff by combined use of statistical tests and hydrological modelling: case studies from Zimbabwe. Journal of Hydrology 205, 147–163. Machiwal, D. & Jha, M. K.  Comparative evaluation of statistical tests for time series analysis: application to hydrological time series. Hydrological Sciences Journal 53, 353–366. Mango, L. M., Melesse, A. M., McClain, M. E., Gann, D. & Setegn, S. G.  Land use and climate change impacts on the hydrology of the upper Mara River Basin, Kenya: results of a modeling study to support better resource management. Hydrology and Earth System Sciences 15, 2245–2258. Milly, P. C. D., Dunne, K. A. & Vecchia, A. V.  Global pattern of trends in streamflow and water availability in a changing climate. Nature 438, 347–350. Mishra, A. K. & Singh, V. P.  A review of drought concepts. Journal of Hydrology 391, 204–216. Mitsova, D., Shuster, W. & Wang, X.  A cellular automata model of land cover change to integrate urban growth with open space conservation. Landscape and Urban Planning 99, 141–153. Moriasi, D. N., Arnold, J. G., Van Liew, M. W., Bingner, R. L., Harmel, R. D. & Veith, T. L.  Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Transactions of the ASABE 50, 885–900. Piao, S., Fang, J., Ciais, P., Peylin, P., Huang, Y., Sitch, S. & Wang, T.  The carbon balance of terrestrial ecosystems in China. Nature 458, 1009–1013. Piao, S., Ciais, P., Huang, Y., Shen, Z., Peng, S., Li, J., Zhou, L., Liu, H., Ma, Y., Ding, Y., Friedlingstein, P., Liu, C., Tan, K., Yu, Y., Zhang, T. & Fang, J.  The impacts of climate

Hydrology Research

51.5

2020

change on water resources and agriculture in China. Nature 467, 43–51. Sang, L., Zhang, C., Yang, J., Zhu, D. & Yun, W.  Simulation of land use spatial pattern of towns and villages based on CAMarkov model. Mathematical and Computer Modelling 54, 938–943. Sterling, S. M., Ducharne, A. & Polcher, J.  The impact of global land-cover change on the terrestrial water cycle. Nature Climate Change 3, 385–390. van Griensven, A., Meixner, T., Grunwald, S., Bishop, T., Diluzio, A. & Srinivasan, R.  A global sensitivity analysis tool for the parameters of multi-variable catchment models. Journal of Hydrology 324, 10–23. Voeroesmarty, C. J., McIntyre, P. B., Gessner, M. O., Dudgeon, D., Prusevich, A., Green, P., Glidden, S., Bunn, S. E., Sullivan, C. A., Liermann, C. R. & Davies, P. M.  Global threats to human water security and river biodiversity. Nature 467, 555–561. Wang, G. & Xu, Z.  Assessment on the function of reservoirs for flood control during typhoon seasons based on a distributed hydrological model. Hydrological Processes 25, 2506–2517. Wang, G., Song, J., Xue, B.-L., Xu, X. & Otsuki, K.  Land use and land cover change of Hulun Lake Nature Reserve in Inner Mongolia, China: a modeling analysis. Journal of the Faculty of Agriculture Kyushu University 57, 219–225. Wang, G., Yang, H., Wang, L., Xu, Z. & Xue, B.  Using the SWAT model to assess impacts of land use changes on runoff generation in headwaters. Hydrological Processes 28, 1032–1042. Wang, Z., Lai, C., Chen, X., Yang, B., Zhao, S. & Bai, X.  Flood hazard risk assessment model based on random forest. Journal of Hydrology 527, 1130–1141. Wang, Z., Deng, X., Song, W., Li, Z. & Chen, J.  What is the main cause of grassland degradation? A case study of grassland ecosystem service in the middle-south Inner Mongolia. Catena 150, 100–107. Wang, G., Hu, X., Zhu, Y., Jiang, H. & Wang, H. a Historical accumulation and ecological risk assessment of heavy metals in sediments of a drinking water lake. Environmental Science and Pollution Research 25, 24882–24894. Wang, W., Fang, Q., Wang, G., Li, R., Xue, B. & Wang, H. b Simulation of CO2 source, sink, and flux temporal and spatial distributions in Hulun Buir grassland. Acta Ecologica Sinica 38, 7288–7299. Wang, G., Li, J., Sun, W., Xue, B., A, Y. & Liu, T. a Non-point source pollution risks in a drinking water protection zone based on remote sensing data embedded within a nutrient budget model. Water Research 157, 238–246. Wang, G., Liu, S., Liu, T., Fu, Z., Yu, J. & Xue, B. b Modelling above-ground biomass based on vegetation indexes: a modified approach for biomass estimation in semi-arid

993

Y. Yan et al.

Climate and land-use impacts on runoff variation

grasslands. International Journal of Remote Sensing 40, 3835–3854. Wheater, H. & Evans, E.  Land use, water management and future flood risk. Land Use Policy 26, S251–S264. Xu, X., Yang, D., Yang, H. & Lei, H.  Attribution analysis based on the Budyko hypothesis for detecting the dominant cause of runoff decline in Haihe basin. Journal of Hydrology 510, 530–540. Xue, B.-L., Wang, L., Li, X., Yang, K., Chen, D. & Sun, L. a Evaluation of evapotranspiration estimates for two river basins on the Tibetan Plateau by a water balance method. Journal of Hydrology 492, 290–297. Xue, B.-L., Wang, L., Yang, K., Tian, L., Qin, J., Chen, Y., Zhao, L., Ma, Y., Koike, T., Hu, Z. & Li, X. b Modeling the land surface water and energy cycles of a mesoscale watershed in the central Tibetan Plateau during summer with a distributed hydrological model. Journal of Geophysical Research – Atmospheres 118, 8857–8868. Xue, B.-L., Guo, Q., Otto, A., Xiao, J., Tao, S. & Li, L. a Global patterns, trends, and drivers of water use efficiency from 2000 to 2013. Ecosphere 6, 174. Xue, B.-L., Li, Z., Yin, X.-A., Zhang, T., Iida, S. i., Otsuki, K., Ohta, T. & Guo, Q. b Canopy conductance in a two-storey Siberian boreal larch forest, Russia. Hydrological Processes 29, 1017–1026. Xue, B.-L., Guo, Q., Hu, T., Wang, G., Wang, Y., Tao, S., Su, Y., Liu, J. & Zhao, X. a Evaluation of modeled global vegetation carbon dynamics: analysis based on global carbon flux and above-ground biomass data. Ecological Modelling 355, 84–96. Xue, B.-L., Guo, Q., Hu, T., Xiao, J., Yang, Y., Wang, G., Tao, S., Su, Y., Liu, J. & Zhao, X. b Global patterns of woody residence

Hydrology Research

51.5

2020

time and its inﬂuence on model simulation of aboveground biomass. Global Biogeochemical Cycles 31, 821–835. Yang, J., Reichert, P., Abbaspour, K. C., Xia, J. & Yang, H.  Comparing uncertainty analysis techniques for a SWAT application to the Chaohe Basin in China. Journal of Hydrology 358, 1–23. Yue, S., Pilon, P. & Cavadias, G.  Power of the Mann-Kendall and Spearman’s rho tests for detecting monotonic trends in hydrological series. Journal of Hydrology 259, 254–271. Zhang, M., Liu, N., Harper, R., Li, Q., Liu, K., Wei, X., Ning, D., Hou, Y. & Liu, S.  A global review on hydrological responses to forest change across multiple spatial scales: importance of scale, climate, forest type and hydrological regime. Journal of Hydrology 546, 44–59. Zhao, F. F., Xu, Z. X., Huang, J. X. & Li, J. Y.  Monotonic trend and abrupt changes for major climate variables in the headwater catchment of the Yellow River basin. Hydrological Processes 22, 4587–4599. Zhao, G., Tian, P., Mu, X., Jiao, J., Wang, F. & Gao, P. a Quantifying the impact of climate variability and human activities on streamﬂow in the middle reaches of the Yellow River basin, China. Journal of Hydrology 519, 387–398. Zhao, Y., Zou, X., Zhang, J., Cao, L., Xu, X., Zhang, K. & Chen, Y. b Spatio-temporal variation of reference evapotranspiration and aridity index in the Loess Plateau Region of China, during 1961–2012. Quaternary International 349, 196–206. Zhou, Y., Lai, C., Wang, Z., Chen, X., Zeng, Z., Chen, J. & Bai, X.  Quantitative evaluation of the impact of climate change and human activity on runoff change in the Dongjiang River Basin, China. Water 10, 511.

First received 22 February 2020; accepted in revised form 3 April 2020. Available online 13 May 2020

994

51.5

2020

Analysis for spatial-temporal matching pattern between water and land resources in Central Asia Ying Zhang, Zhengxiao Yan, Jinxi Song, Anlei Wei, Haotian Sun and Dandong Cheng

ABSTRACT Central Asia, the pioneering place of the ‘Belt and Road’, is under the threat of prominent water issues. Based on the Gini coefficient model and the matching index, the amount of the total renewable water resources and the cultivated land area were introduced to evaluate the matching pattern between the water and land resources in Central Asia. The water problem of Kazakhstan, being the most prominent, shows low water resources per unit area with the highest reclamation rate. The matching degree for the upstream countries of the Aral Sea (Kyrgyzstan, Tajikistan) was better than those of the downstream countries (Turkmenistan, Uzbekistan, Kazakhstan). The Gini coefficient in Central Asia was 0.32, smaller than that of the global average value (0.59). The overall water available for use and the matching cultivated land resources was reasonable. Large differences exist in the matching degree in water distribution and utilization among Central Asian countries. The matching index of water and land resources in Central Asia was 1.25, similar to the matching degree estimated from the Gini coefficient model. Moreover, rational measures are suggested to alleviate the issue of water and land resources matching in Central Asia. Key words

Ying Zhang Zhengxiao Yan Jinxi Song (corresponding author) Anlei Wei Haotian Sun Dandong Cheng Shaanxi Key Laboratory of Earth Surface System and Environmental Carrying Capacity, College of Urban and Environmental Sciences, Northwest University, Xi’an 710127, China E-mail: jinxisong@nwu.edu.cn Jinxi Song State Key Laboratory of Soil Erosion and Dryland Farming on the Loess Plateau, Institute of Soil and Water Conservation, Chinese Academy of Sciences and Ministry of Water Resources, Yangling 712100, China

| ArcGIS, spatial and temporal pattern, water and land resources matching, water distribution and utilization

HIGHLIGHTS

• • • • •

The Gini coefﬁcient model is to evaluate the overall level of matching degree. The matching index model is introduced for evaluating the spatial matching pattern. The matching situation for the upstream countries of the Aral Sea was better than those of the downstream countries. Two models perform similarly in matching pattern. The matching index of water and land resources in Central Asia was 1.25.

INTRODUCTION Central Asia, located in the hinterland of the Eurasian con-

world (Unger-Shayesteh et al. ). The economic beneﬁts

tinent, lacks water resources (Gafurov et al. ). It is one of

of water use in Central Asia are lower than those in other

the regions threatened by serious water problems in the

areas of Asia (Lee & Jung ). Although the total amount of water resources is seemingly large in this

This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying,

region, in fact, only 24.4% is available for use by humans

adaptation and redistribution, provided the original work is properly cited

(Yang et al. ). On top of that, multiple transboundary

(http://creativecommons.org/licenses/by/4.0/).

rivers exist in Central Asia (Zhupankhan et al. ).

doi: 10.2166/nh.2020.177

995

Y. Zhang et al.

Matching pattern between water and land resources in Central Asia

Hydrology Research

51.5

2020

However, the mismatched spatial distributions of water and

conﬂicts. The correct evaluation of the formation and con-

land resources, along with the intense human activities (e.g.,

sumption of water resources and its spatial matching with

overexploitation of water resources), has ultimately led to

land resources is critical to alleviating water problems

the serious water crisis in Central Asia’s river basins. This

(Yang et al. ). Most of the current state-of-the-art studies

is the main reason for ongoing water conﬂicts in the region’s

have focused on the water problems in Central Asia from the

transboundary rivers and the ecological disaster of the Aral

aspects of the state of water resources, the water cycle pro-

Sea (Chen et al. ).

cesses, the transboundary river management, and so on

Kyrgyzstan and Tajikistan, located in the upstream of

(Yang et al. ). Although previous studies exist for quan-

the Aral Sea Basin, have the most abundant water

titative analysis, they mainly focus on the impact of

resources and are called the ‘water towers’ of Central

climate change on the water cycle and the water environ-

Asian countries. However, due to insufﬁcient water conser-

ment of the basin (Zhu et al. ). However, there is a

vancy engineering measures, water resources have ﬂowed

lack of study on the matching patterns between water and

into downstream countries, so the amount of available

land resources. Therefore, a quantitative study on spatial dis-

water resources in upstream countries is limited. Mean-

tribution difference and the matching pattern change of

while, the downstream countries of the Aral Sea Basin

water and land resources is the basis of a spatial optimal

(Kazakhstan, Turkmenistan, and Uzbekistan) enjoy more

allocation of regional water resources, offering important

available water resources, which ﬂow from upstream

guidance for efﬁcient utilization of water and land resources

countries, than their actual domestic water storage (Deng

in Central Asia.

et al. ). Owing to the limited water storage, the down-

Spatial-temporal dynamics of water and land resources

stream countries are also water-deﬁcient, especially the

matching have attracted many researchers’ attention. Most

oil-rich Turkmenistan, which has the saying that ‘water is

researchers have studied the overall matching of water and

more expensive than oil’.

land resources in the region on the macro-scale of time

In the former Soviet Union era, the upstream countries

and space. Liu et al. () constructed the matching analy-

focused on the construction of water conservancy facilities,

sis model of agricultural water and land resources using the

providing downstream countries with water resources and

total water resources as a parameter and evaluated the

power resources such as farmland irrigation water, while

matching pattern of agricultural water and land resources

downstream countries provided more cultivated land and

in Northeast China. Liu et al. () established the Gini

agricultural products for the upstream (Jalilov et al. ).

coefﬁcient model for agricultural water and land resources

Since the independence of Central Asian countries, a host

allocation and the model for measuring the matching

of international projects has been launched to achieve sus-

index of the broadly deﬁned agricultural water resources

tainable water management, but only a few have been

and land resources (MIBAWRLR) to assess the matching

implemented (Abdullaev et al. ). Uncompensated

degree of the agricultural water and land of the Jiansanjiang

water use patterns for downstream countries have caused

Administration and 15 farms in Heilongjiang Province,

dissatisfaction in upstream countries. The issue of ‘energy-

China. Based on the water and land resources and agricul-

water-irrigation’ has been plaguing relations between Cen-

tural output value, Huang et al. () used the data

tral Asian countries (Bernauer & Siegfried ). Water

envelopment analysis (DEA) model to analyze the situation

and land resources, as the basic materials for production

of water and land resources in Sichuan Province, China.

and life, are the basis for ensuring national food security

Only a few researchers analyzed the equilibrium state of

(Zhang et al. a). For the ﬁve Central Asian countries

supply and demand for agricultural land and water

dominated by irrigated agriculture, the shortage of water

resources at the micro-scale. Zhang et al. (b) con-

and land resources has delayed the process of agricultural

structed a water and land resources matching index

modernization (Riquelme & Ramos ). Generally speak-

calculation model in micro-scale and studied the temporal

ing, the mismatching between the formation area and

and spatial dynamics of crop water and land resource

consumption area of water resources will lead to water

matching in the Naoli River Basin, China.

996

Y. Zhang et al.

Matching pattern between water and land resources in Central Asia

The matching of water and land resources can be calcu-

Hydrology Research

51.5

2020

MATERIALS AND METHODS

lated from three aspects: calculating the matching index of water and land resources with the total water resources

Description of study area

volume per hectare as the index; evaluating the overall matching situation of water and land resource in the

Central Asia refers to the eastern part of the Caspian Sea,

region by the Gini coefﬁcient model, based on DEA to

south of Western Siberia, north of Afghanistan and Central

analyze the matching characteristics of water and land

Asia region to the west of Xinjiang, including Kazakhstan,

resources. Although these models are widely used to study

Uzbekistan,

the dynamics of regional water and land resources match-

(56 340 E–87 190 E, 35 080 N–55 260 N). With a total area of

ing, each model has its shortcomings. For the DEA model,

nearly 4.00 × 108 hm2 (excluding coastal waters) and a total

the water and land resources are used as the input indicators

population of about 7.08 × 107 in 2017, Central Asia is one

and the agricultural output as the output index to study the

of the most sparsely populated areas in the world today.

matching degree of water and land resources (Huang et al.

The total renewable water resources in this area are 2.28 ×

). However, multiple factors have been found to affect

1011 m3 and the cultivated area is 3.84 × 107 hm2. The rivers

agricultural output value: the amount of agricultural

in Central Asia are mostly inland rivers and lakes, mainly

equipment, labor force, water and land resources, etc.

including the Amu Darya, the Syr River, the Chu River, the

Therefore, the results of the DEA model are not satisfactory

Aral Sea, and so on (Figure 1). Its terrain is high in the east

when only water resources and land resources are taken as

and low in the west, dominated by plains and hills. The

research variables. For the Gini coefﬁcient model, different

total area of drylands is 5.17 × 106 km2 (above 80% of the

division degrees of units in the region can yield different

global total temperate desert area) in Central Asia, belonging

results. Meanwhile, it can only reﬂect the overall matching

to the temperate desert climate and grassland continental cli-

situation of water and land resources in the study area and

mate (Li et al. ). The vegetation is mainly grassland and

the relative matching of each unit in the region. For the

desert. In most areas, the climate is dry with strong evapor-

matching index model, although it reveals the overall match-

ation (Yao et al. ). The annual precipitation is below

ing status of water and land resources in the region, the

300 mm, and the annual precipitation in the desert near the

Kyrgyzstan,

Tajikistan,

and

Turkmenistan

spatial and temporal distribution and dynamic character-

Aral Sea and Turkmenistan is only 75–100 mm. The territory

istics of water and land resources matching at the micro-

area of the ﬁve Central Asian countries varies greatly, with

scale are still unclear. At the same time, the model does

Kazakhstan’s territory being vast, twice the total area of the

not take into account the differences in the use of water

other four countries. However, although Uzbekistan has a

resources between different types of cultivated land use.

total area of only one-sixth of Kazakhstan, its population is

However, this model can make up for the errors caused by

nearly 40% greater than that of Kazakhstan.

the division of the unit based on the Gini coefﬁcient, reﬂect the relative space–time ratio of water and land resources matching. To achieve comprehensive results, we use the

Data sources

Gini coefﬁcient model and the matching index model to analyze the matching situation of water and land resources

Since the data of water and land resources in Central Asia

in Central Asia.

are not continuous, this paper uses the data of 1992, 1997,

Due to the complexity of transboundary water resources in

2002, 2007, 2012, and 2016 from the Food and Agriculture

Central Asia, this paper takes the ﬁve countries of Central Asia

Organization of the United Nations (FAO a). The culti-

as the basic unit, then uses total renewable water resource and

vated area is used as the parameter of cultivated land

total water withdrawal as the water resource parameters, and

resources, including arable land and permanent crops. For

the cultivated area as the cultivated land resource parameter

water resources, the amount of agricultural water, total

to study the spatial–temporal matching relationship of water

water withdrawal, and renewable water resources are

and land resources in ﬁve Central Asian countries.

used. The renewable water resource includes the amount

997

Figure 1

Y. Zhang et al.

Matching pattern between water and land resources in Central Asia

Hydrology Research

51.5

2020

Research site map of Central Asia.

of surface water and groundwater available in the country.

volume per hectare. It is a quantitative relationship that

The land cover types are obtained from the global land

reﬂects the spatial and temporal distribution of water

cover data product with a resolution of 500 m (MCD12Q1

resources and cultivated land resources available for agri-

data set) of the Land Process Distributed Active Archive

cultural production in a speciﬁc region. It focuses on

Center (LPDAAC ). The area of cropland, grassland,

regional water resources and cultivated land, and coordi-

and forest land are collected by the FAO (FAO b).

nation and suitability of regional water resources and cultivated land resources in space–time distribution (Liu

Theory of water and land resource matching

et al. ). The higher the matching degree, the better the fundamental conditions of agricultural production are.

Matching index model of water and land resources

2. Calculating model: The matching level of water resources and cultivated areas with ﬁve countries in Central Asia

1. Basic concept: The matching index of water and land resources is expressed by the total water resources

are selected as the basic units. The agricultural water withdrawal

determined

the

proportion

998

Y. Zhang et al.

Matching pattern between water and land resources in Central Asia

Hydrology Research

51.5

2020

agricultural water used to the whole water consumption

Smaller G value indicates relatively even distribution,

(agricultural water, industrial water, and municipal

while larger G value indicates relatively uneven distribution.

water) in Central Asia (Liu et al. ). The calculation formula is as follows:

Because the spatial distribution of water resources is diversiﬁed, the Gini coefﬁcient is also applicable in the study of the patterns of water and land resource matching.

Rwl i ¼ Wi αi =Li ,

(i ¼ 1, 2, 3, 4, 5)

(1)

According to the spatial distribution characteristics of water and land resources, the Gini curve construction

is the matching index of water and land where Rwl i resources of the ith country in Central Asia; Wi is the

model for the matching problem of water and land resources in Central Asia is as follows (Liu et al. ):

total renewable water resource of the ith country, 108 m3; Li is the area of the cultivated area of the ith country, 104 hm2; αi is the proportion of agricultural water in the ith country to the total amount of social and economic water. The matching index of water and land resources in Central Asia reﬂects the mean value of the matching index of water and land resources among countries in Central Asia. The calculation formula is as follows:

1. Take the ﬁve countries in Central Asia as the basic calculation units, and calculate the total water resources volume per hectare at each country, then according to the size of this index, the ﬁve Central Asian countries are ranked in ascending order. 2. Calculate the percentage of total water resources and cultivated land resources in each unit. 3. The cumulative proportion of water and land resources in each country is calculated following step 1.

Rpj ¼

n X

Rwl i =n

(2)

i¼1

4. Draw XY scatter plots, the X coordinate being the cumulative proportion of the cultivated area of each unit, and the Y coordinate is the cumulative proportion

is the matching index of water and land

of the water resources parameters of each unit, then

resources in area j; Rwl i is the matching index of water

the Lorenz curve of water and land resources is

where

Rpj

and land resources of the ith country in area j; and n is the number of i countries in area j.

constructed. 5. Obtain the area of the ﬁgure between the Lorentz curve and the slash with an angle of 45 , and an integral

Gini coefﬁcient model

between 0 and 1. Then, the Gini coefﬁcient is twice the ﬁgure area.

The Gini coefﬁcient was proposed by Italian economist Corrado Gini in 1921 when studying income inequality. It has been recognized and applied in many ﬁelds. The Gini

If the G value is between 0 and 1, and the smaller the G

coefﬁcient (G), also known as the Lorentz coefﬁcient, is

value is, the higher the degree of matching. When G ¼ 0, the

determined by calculating the area of the Lorentz curve

Lorenz curve completely coincides with the 45 line, and

graph using MATLAB software. First, the Lorentz curve

the water and soil resources are completely matched. On

needs to be ﬁtted. Assuming that the area between the

the contrary, when G ¼ 1, the matching degree of water

actual distribution curve of income and the absolute equal

and land resources is extremely poor, that is, the cultivated

distribution curve is A, and the area between the actual dis-

land resources in this area are very rich, and the water

tribution curve of income and the coordinate axis is B, then

resources are in serious shortage.

the Gini coefﬁcient is calculated as follows:

The internationally accepted model for dividing the Gini coefﬁcient is as follows. For example, G ∈ (0, 0.2) indicates

G¼

A AþB

(3)

that the degree of matching between water resources and land resources is high (Table 1).

Y. Zhang et al.

999

Table 1

Matching pattern between water and land resources in Central Asia

Hydrology Research

51.5

2020

higher than domestic water resources. Kyrgyzstan and Taji-

Gini coefﬁcient division standard (Dorfman 1979)

Matching results

(0,0.2)

High

[0.2,0.3)

Consistent

[0.3,0.4)

Reasonable

[0.4,0.5)

Inconsistent

[0.5,1)

Poor

kistan are net exit water resources countries, of which, Kyrgyzstan’s net exit water volume accounts for 76% of domestic water resources. The climate in Central Asia is highly arid with annual precipitation generally below 300 mm. Due to topography and climate, upstream Kyrgyzstan and Tajikistan are the main water sources in Central Asia, with an exit water volume of 76 billion m3, but the main water users are Uzbekistan and Turkmenistan.

RESULTS The structure and spatial distribution of water and land

Composition and utilization of land resources

resources in Central Asia The total area of Central Asia is nearly 4.00 × 108 hm2. Composition and utilization of water resources

Regarding land types, grassland is the largest proportion (279.5 × 104 km2) and bare areas the second (67.7 ×

The total freshwater in the ﬁve Central Asian countries is 3

above 1 trillion m , but the real available water resources 3

104 km2). Nearly half of the areas show a natural landscape of desert and semi-desert, and the vegetation coverage in the

are only 206 billion m , of which, surface water is 187 billion

north and southeast is higher (Figure 2). The highest pro-

m3 and the non-repetitive groundwater is 19 billion m3

portions of cropland, forest land, and grassland are in

(Table 2). Kazakhstan has the largest water resources,

Uzbekistan (11.21%), Turkmenistan (8.78%), and Kazakh-

accounting for 36.69% of Central Asia. Turkmenistan

stan (69.44%), respectively (Figure 3). Owing to the

has the least (0.68%). Agricultural water consumption

disintegration of the former Soviet Union, large-scale culti-

accounted for the largest proportion of social and economic

vation was abandoned in Central Asia. Subsquently, with

water use, with the proportion being 66%, 91%, 93%, 90%,

the independence of Central Asian countries, the social

and 94% in Kazakhstan, Tajikistan, Kyrgyzstan, Uzbekistan,

economy gradually recovered, and so did the cultivated

and Turkmenistan, respectively (FAO a). The ﬁve Cen-

areas. Therefore, the cultivated area and crop yield in Cen-

tral Asian countries have numerous transboundary rivers

tral Asia showed a trend of rapid decline ﬁrst and then a

with large stream ﬂow. Notably, Kazakhstan has a net

slow rise (Kienzler et al. ).

entry water volume of 34.2 billion m3, followed by Uzbeki-

The production potential of land resources in Central

stan. The net entry water volume of Turkmenistan is much

Asia is huge, but it is constrained by water resources.

Table 2

Water resources assessment in the ﬁve Central Asia countries (FAO 2016a; United Nations 1998) Average precipitation

Overlap between Surface

Groundwater/

surfaces water and

Water resources/ Entry-exit water

Available water

Water resources

Country in depth/mm

water/108 m3

108 m3

groundwater/108 m3

108 m3

volume/108 m3

resources/108 m3

per capita/m3

KAZ

693

161

100

754

342

1,096

4,142

250

KGZ

533

441

136

112

465

259

206

7,692

TJK

691

638

668

508

160

7,488

TKM

161

233

247

243

UZB

206

163

341

504

511

1,877

449

262

2,064

149

1,117

4,015.2

Sum

Y. Zhang et al.

1000

Figure 2

Matching pattern between water and land resources in Central Asia

Hydrology Research

51.5

2020

The types of land cover in Central Asia in 2015.

Meanwhile, a series of ecological and environmental pro-

(Figure 4(b)). Moreover, the spatial distribution of water

blems in the process of land use seriously restrict the

resources and cultivated land resources in Central Asia is

sustainable use of land resources in Central Asia. Effective

inconsistent (Figure 4). Kazakhstan is the most prominent

approaches to these problems are the key to maintaining

country, whose water resources per unit area is the lowest,

sustainable development in Central Asia.

but the reclamation rate is the highest in the region. This spatial dislocation of water and land resources in Central

Spatial distribution of water and land resources

Asia severely limits the increase of regional grain production and sustainable use of resources, which in return, exacer-

The water resources per unit area can reﬂect the richness

bates the contradiction of water problems in Central Asia.

and shortage of regional water resources, and the reclamation rate can reﬂect the degree of reclamation of

Matching pattern of water and land resources in Central

regional cultivated land resources. Five countries in Central

Asia

Asia are taken as the basic regional units, then the natural breakpoint method of GIS is used to divide the Central Asian region into ﬁve grades according to the water

Matching degree of water and land resources based on the Gini coefﬁcient

resources per unit area and the reclamation rate, producing the spatial distribution pattern of water and land resources

The total renewable water resource is taken as the water

(Figure 4).

resource parameter, and the cultivated land area is used as

The spatial distribution of water resources in Central

the land resource parameter. The Gini coefﬁcient of Central

Asia is not uniform and is characterized by a larger

Asia in 1992, 1997, 2002, 2007, 2012, and 2016 is shown in

amount of water resources in Kazakhstan and a lower

Figure 5.

amount of water resources in Kyrgyzstan (Figure 4(a)). The

The Gini coefﬁcient showed little variation in 1992,

distribution of cultivated land in Central Asia is uneven

1997, 2002, 2007, 2012, and 2016, and the matching

Y. Zhang et al.

1001

Matching pattern between water and land resources in Central Asia

Hydrology Research

51.5

2020

Spatial matching pattern of water and land resources based on the matching index model Since potential errors may exist in the unit division of the Gini coefﬁcient, the matching index model can be used to study the matching situation of water and land resources. Based on the data of total renewable water resource, cultivated area, and proportion of agricultural water in ﬁve Central Asian countries in 1992, 1997, 2002, 2007, 2012, and 2016, the water and land resources matching model is Figure 3

Proportion of land types in Central Asia in 2016.

used to obtain the matching index of water and land resources in Central Asian countries, as shown in Table 3. To visually indicate the difference in the matching situ-

situation of water and land resources in Central Asia is

ation of water and land resources, the matching index of

relatively reasonable (Figure 5(a)–5(f)). The average Gini

water and land resources in 2016 is divided into ﬁve cat-

coefﬁcient of Central Asia in 1992–2016 was 0.32. Accord-

egories according to the Natural Breaks proposed by

ing to the internationally recognized Gini coefﬁcient

Jenks: Poor matching (0.001< R 0.242), Moderate match-

division standard, the degree of matching between water

ing (0.242 < R 0.922), Relatively good matching (0.922<

resources and land resources is reasonable. Therefore, the

R 1.164), Good matching (1.164 < R 1.610), and Excel-

total renewable water resource is not a decisive factor for

lent matching (R > 1.610). Moreover, the other years are

water problems in Central Asia.

analyzed based on this criterion to form a contrast. Finally,

Due to data availability, the total water withdrawal was used as the water resource parameter and the cultivated area

Figures 7 and 8 illustrate the spatial matching pattern of water and land resources in Central Asia.

as the land resource parameter. Then, the Gini coefﬁcient in

Combined with the multi-year average of Table 3, the

Central Asia in 2008 and 2014 were calculated, as shown in

matching index of Kazakhstan is the smallest and the match-

Figure 6.

ing index of Tajikistan is the largest, the latter being six times

The average Gini coefﬁcient was 0.60 (Figure 6). The total

that of the former. By taking the ﬁve countries as a whole

water withdrawal indicates that the matching situation of

and calculating the multi-year average matching index of

water and land resources in Central Asia is extremely poor.

water and land resources from 1992 to 2016, the matching

Figure 4

Spatial distribution of water and land resources in ﬁve Central Asian countries in 2016.

Y. Zhang et al.

1002

Matching pattern between water and land resources in Central Asia

Figure 5

Lorentz curve of water and land resource matching in Central Asia.

Figure 6

Gini coefﬁcient for Central Asia.

Hydrology Research

51.5

2020

Y. Zhang et al.

1003

Table 3

Matching pattern between water and land resources in Central Asia

Hydrology Research

51.5

2020

Matching index of water and land resources in ﬁve Central Asian countries from 1992 to 2016

Country

1992

1997

2002

2007

2012

2016

Mean

Kazakhstan

0.251

0.269

0.286

0.256

0.236

0.242

0.257

Kyrgyzstan

1.601

1.557

1.569

1.624

1.625

1.610

1.598

Tajikistan

2.054

2.198

2.279

2.284

2.281

2.310

2.234

Turkmenistan

1.565

1.349

1.138

1.109

1.164

1.248

Uzbekistan

0.943

0.949

0.916

0.938

0.926

0.922

0.932

Mean

1.283

1.264

1.238

1.242

1.246

1.250

1.254

index of water and land resources in Central Asia is 1.25

space and time, in combination with excessive and

(Table 3), which is Good matching. The spatial matching

often uncontrolled withdrawal for irrigation, creates

pattern of water and land resources in Central Asia varies

water scarcity, especially in the southern area.

across different periods. The matching situation of each

2. The average Gini coefﬁcient in Central Asia is 0.32, and

country is as follows: Kazakhstan changed from Poor

the matching situation of water and land resources is rela-

matching in 2007 to Moderate matching in 2012 (Figure 7(d)

tively reasonable. It shows that Central Asia is regarded

and 7(e)); Uzbekistan ﬂuctuated between Relatively good

as a whole and its water and land resources are balanced

matching and Moderate matching; Turkmenistan has

to a certain extent. The total renewable water resource is

changed from Good matching in 1997 to Relatively good

not a decisive factor contributing to the water problems

matching in 2002 (Figure 7(b) and 7(c)); Kyrgyzstan

in Central Asia. The average Gini coefﬁcient obtained

remained as Good matching; Tajikistan was Excellent

from the total water withdrawal as a parameter of water

matching from 1992 to 2016 (Figure 7(a)–7(f)). Meanwhile,

resources in Central Asia was 0.60, and the matching situ-

differences exist among the matching index of water and

ation of water and land resources is Poor matching. This

land resources in different countries in the same period.

is in stark contrast to the results obtained from the total

The overall situation of Turkmenistan, Kyrgyzstan, and Taji-

renewable water resource as a water resource parameter

kistan is better; the upstream countries (Kyrgyzstan,

(from Reasonable matching to Poor matching). The water

Tajikistan) are better than the downstream countries (Turk-

and land resources of Central Asia as a whole are in bal-

menistan, Uzbekistan, Kazakhstan) (Figure 8).

ance, and the contradiction of water issues mainly lies in the distribution and utilization of water resources among countries.

DISCUSSION

3. The matching index of water and land resources is a quantitative indicator for the equilibrium situation and

Analysis of spatial distribution and matching pattern

satisfaction degree of water resources and cultivated

results

land resources in a certain area. The matching index of water and land resources in Central Asia is 1.25,

1. A clear dislocation of water resources and cultivated land

which is Good matching. However, the matching situ-

resources has been identiﬁed in Central Asia. Kazakhstan

ation varies greatly among different countries. The

is the most prominent, exhibiting low water resource per

matching situation of water and land resources in the

unit area with the highest reclamation rate in the region.

upstream countries (Kyrgyzstan, Tajikistan) is better

The shortage of water resources has become the main

than that in the downstream countries (Turkmenistan,

factor restricting agricultural development in the region.

Uzbekistan, Kazakhstan). This is because, after the dis-

Kazakhstan is not water-scarce in terms of total water

integration of the Soviet Union, the mode of free

supply per capita. However, the uneven distribution in

water supply to the downstream countries caused

Y. Zhang et al.

1004

Figure 7

Matching pattern between water and land resources in Central Asia

Hydrology Research

51.5

2020

Matching index of water and land resources in ﬁve Central Asian countries from 1992 to 2016.

dissatisfaction in the upstream countries. The upstream

to provide agricultural products for the upstream

countries to provide water and electricity for the

countries and other management systems have not

downstream countries and the downstream countries

been adopted.

Y. Zhang et al.

1005

Figure 8

Matching pattern between water and land resources in Central Asia

Hydrology Research

51.5

2020

Variation trend categories of resource matching in ﬁve Central Asian.

Comparison of the two models

the results of the two models also show consistency. They all show the renewable water resources is not a key factor

Two models are used in the matching of water and land

affecting water issues in Central Asia (Table 4).

resources in Central Asia. Although great discrepancies in the matching pattern are identiﬁed among the ﬁve countries, the results of the overall level of matching between water

Suggestions on water resource management

and land resources in Central Asia are reasonable. Two models both show that the matching of water and land

The Central Asian states share many common problems and

resources are balanced to a certain extent in Central Asia.

unaddressed tasks in the management of water resources

For the Gini coefﬁcient model, the average of the matching

(Zhupankhan et al. ). Based on the current situation of

situation is reasonable, similar to the matching degree esti-

water and land resources matching in Central Asia, the fol-

mated from the matching index (Good matching). In terms

lowing suggestions are put forward.

of the renewable water resource, the two models are

1. Improve the water resources management and allocation

almost, in the matching of water and land resources in Cen-

system in Central Asia and formulate a scientiﬁc plan for

tral Asia, reasonable. At the same time, from the perspective

the relative matching of water and land resources. The

of each year, combining their respective evaluation criteria,

matching of water and land resources in Central Asian

Y. Zhang et al.

1006

Table 4

Matching pattern between water and land resources in Central Asia

Hydrology Research

51.5

2020

Comparison of the results of the two models

1992

1997

2002

2007

2012

2016

Gini coefﬁcient

0.24

0.33

0.30

0.31

Matching results

Consistent

Reasonable

Matching index

1.28

1.26

1.24

1.25

Matching results

Good matching

countries has ﬂuctuated with time, indicating that the

cooperation with the international community, introduce

water resources management and distribution system

advanced water resources management programs, draw

needs to be improved, and it is impossible to ensure the

on national policies with outstanding effects on water

supply of resources among countries. The present situ-

control, and ultimately achieve social-economic-ecologi-

ation of water resources and future supply capacity

cal sustainable development in Central Asia.

should be combined to formulate reasonable social development strategies (Zhang et al. ). 2. Work out reasonable measures for constructing water

Limitations of this study

conservancy projects. The future for Central Asian development lies within the intertwined use of transboundary

1. The total amount of water and land resources will change

water resources. Upstream countries use water for hydro-

due to industrialization, urbanization, water conservancy

power while downstream countries use the water for

construction, and so on. Therefore, future studies should

agriculture and industry. Thus, water use also needs to

consider the dynamic changes in the matching of water

integrate transboundary sectors. Upstream countries pro-

and land resources in the border areas of Central Asia.

duce energy that downstream countries can use for

At the same time, the matching of water and land

agriculture and exploiting mineral resources (Zhupan-

resources should be reﬁned to evaluate the micro-states

khan et al. ). Meanwhile, it is necessary to properly

of crop water demand and land use in each state,

handle the contradiction of water problems and formu-

which also needs further research.

late reasonable water distribution, water supply, and

2. This paper analyzes the matching of water and land

compensation water agreements. The water contradiction

resources in Central Asia on the time and space scales.

in Central Asia is not the total renewable water resources,

However, due to the fact that appropriate data were not

but the total water withdrawal used by various countries.

available, it is impossible to estimate historical data and

The upstream countries are rich in water resources, but

make scientiﬁc predictions for future data. Further

the matching degree of water and land resources lags

research can improve this to make a more comprehen-

behind the downstream countries. Therefore, the ﬁve

sive analysis.

Central Asian countries should consider the relevant compensation factors when formulating the transbound-

CONCLUSIONS

ary water resources allocation plan. 3. Strengthen cooperation with the international community, which requires cooperation among the countries of

This paper takes the ﬁve Central Asian countries as the

Central Asia that goes far beyond the current situation.

research unit, and uses the total renewable water resource

Major reforms are necessary and external pressures

and total water withdrawal as the water resources par-

from neighboring Russia and China are likely required

ameters, and constructs the model of the Gini coefﬁcient

to make this happen (Howard & Howard ). There-

and the matching index between water and land resources.

fore,

Then, reasonable suggestions are put forward according to

Central

Asia

needs

actively

carry

out

1007

Y. Zhang et al.

Matching pattern between water and land resources in Central Asia

the current situation of water problems in Central Asia. The matching situation of water and land resources in Central Asia is generally better. The average Gini coefﬁcient is 0.32, and the water and land resources matching index is 1.25. The main contradiction of the Central Asian water problem is the distribution and utilization among countries. Therefore, relevant departments should improve the water resources management and distribution system in Central Asia, formulate reasonable water distribution, water supply, and compensation water agreements to achieve social-economic-ecological sustainable development in Central Asia.

ACKNOWLEDGEMENTS The research is supported by the Strategic Priority Research Program

the

Chinese

Academy

Sciences

(XDA2004030201-2), the National Key R&D Program on Monitoring, Early Warning, and Prevention of Major Natural Disasters (Grant No. 2017YFC1502506).

REFERENCES Abdullaev, I., Kazbekov, J., Manthritilake, H. & Jumaboev, K.  Participatory water management at the main canal: a case from South Ferghana canal in Uzbekistan. Agricultural Water Management 96 (2), 317–329. Bernauer, T. & Siegfried, T.  Climate change and international water conflict in Central Asia. Journal of Peace Research 49 (1), 227–239. Chen, Y., Li, Z., Fang, G. & Li, W.  Large hydrological processes changes in the transboundary rivers of Central Asia. Journal of Geophysical Research: Atmospheres 123 (10), 5059–5069. Deng, M. J., Long, A. H., Zhang, Y., Li, X. & Lei, Y.  Assessment of water resources development and utilization in the five Central Asia countries. Advances in Earth Science 25 (12), 1347–1356. Dorfman, R.  A formula for the Gini coefficient. The Review of Economics and Statistics 61 (1), 146–149. FAO a AQUASTAT Main Database. Food and Agriculture Organization of the United Nations (FAO), Rome, Italy. Available from: http://www.fao.org/nr/water/aquastat/data/ query/index.html?lang=en (accessed 5 December 2019). FAO b AQUASTAT Main Database. Food and Agriculture Organization of the United Nations (FAO), Rome, Italy. Available from: http://www.fao.org/faostat/zh/#data/EL (accessed 5 December 2019).

Hydrology Research

51.5

2020

Gafurov, A., Kriegel, D., Vorogushyn, S. & Merz, B.  Evaluation of remotely sensed snow cover product in Central Asia. Hydrology Research 44 (3), 506–522. Howard, K. W. & Howard, K. K.  The new ‘Silk road economic belt’ as a threat to the sustainable management of Central Asia’s transboundary water resources. Environmental Earth Sciences 75 (11), 976. Huang, K. W., Yuan, P. & Liu, G.  Research on Water and Soil Resources Matching in Sichuan Province Based on DEA. China Rural Water and Hydropower, China. Jalilov, S. M., Keskinen, M., Varis, O., Amer, S. & Ward, F. A.  Managing the water–energy–food nexus: gains and losses from new water development in Amu Darya River Basin. Journal of Hydrology 539, 648–661. Kienzler, K. M., Lamers, J. P. A., McDonald, A., Mirzabaev, A., Ibragimov, N., Egamberdiev, O., Ruzibaev, E. & Akramkhanov, A.  Conservation agriculture in Central Asia – what do we know and where do we go from here? Field Crops Research 132, 95–105. Lee, S. O. & Jung, Y.  Efficiency of water use and its implications for a water-food nexus in the Aral Sea Basin. Agricultural Water Management 207, 80–90. Li, Z., Chen, Y., Fang, G. & Li, Y.  Multivariate assessment and attribution of droughts in Central Asia. Scientific Reports 7 (1), 1316. Liu, Y. S., Gan, H. & Zhang, F. G.  Analysis of the matching patterns of land and water resources in Northeast China. Acta Geographica Sinica 61 (8), 847–854. Liu, D., Liu, C., Fu, Q., Li, M., Faiz, M. A., Khan, M. I., Li, T. X. & Cui, S.  Construction and application of a refined index for measuring the regional matching characteristics between water and land resources. Ecological Indicators 91, 203–211. LPDAAC  The Global Land Cover Data Product. Land Process Distributed Active Archive Center (LPDAAC). Available from: https://ladsweb.modaps.eosdis.nasa.gov/ search/ (accessed 5 December 2019). Riquelme, F. J. M. & Ramos, A. B.  Land and water use management in vine growing by using geographic information systems in Castilla-La Mancha, Spain. Agricultural Water Management 77 (1–3), 82–95. Unger-Shayesteh, K., Vorogushyn, S., Merz, B. & Frede, H. G.  Water in Central Asia-perspectives under global change. Global and Planetary Change 110 (Part A), 1–152. United Nations  Strengthening Cooperation for Rational and Efficient use of Water and Energy Resources in Central Asia. Available from: http://www.unece.org/speca/welcome.html (accessed 5 December 2019). Yang, S. T., Yu, X. Y., Ding, J. L., Zhang, F., Wang, F. & Ma, Y.  A review of water issues research in Central Asia. Acta Geographica Sinica 72 (1), 79–93. Yang, Z., Song, J., Cheng, D., Xia, J., Li, Q. & Ahamad, M. I.  Comprehensive evaluation and scenario simulation for the water resources carrying capacity in Xi’an city, China. Journal of Environmental Management 230, 221–233.

1008

Y. Zhang et al.

Matching pattern between water and land resources in Central Asia

Yao, J., Hu, W., Chen, Y., Huo, W., Zhao, Y., Mao, W. & Yang, Q.  Hydro-climatic changes and their impacts on vegetation in Xinjiang, Central Asia. Science of the Total Environment 660, 724–732. Zhang, Y., Yu, J., Wang, P. & Fu, G.  Vegetation responses to integrated water management in the Ejina basin, northwest China. Hydrological Processes 25 (22), 3448–3461. Zhang, J., Chen, Y., Li, Z., Song, J., Fang, G., Li, Y. & Zhang, Q. a Study on the utilization efﬁciency of land and water resources in the Aral Sea Basin, Central Asia. Sustainable Cities and Society 51, 101693. Zhang, Y., Lei, G. P., Zhang, H. Q. & Li, J. b Spatiotemporal dynamics of land and water resources matching of cultivated

Hydrology Research

51.5

2020

land use based on micro scale in Naoli River Basin. Transactions of the Chinese Society of Agricultural Engineering 35 (8), 185–194. Zhu, B., Yu, J., Rioual, P., Gao, Y., Zhang, Y. & Xiong, H.  Climate effects on recharge and evolution of natural water resources in middle-latitude watersheds under arid climate. In: Environmental Management of River Basin Ecosystems (M. Ramkumar, K. Kumaraswamy & R. Mohanraj, eds). Springer, Cham, Switzerland, pp. 91–109. Zhupankhan, A., Tussupova, K. & Berndtsson, R.  Water in Kazakhstan, a key in Central Asian water management. Hydrological Sciences Journal 63 (5), 752–762.

First received 12 December 2019; accepted in revised form 24 April 2020. Available online 22 June 2020

1009

51.5

2020

Different runoff patterns determined by stable isotopes and multi-time runoff responses to precipitation in a seasonal frost area: a case study in the Songhua River basin, northeast China Jie Li, Wei Dai, Yang Sun, Yihui Li, Guoqiang Wang and Yuanzheng Zhai

ABSTRACT Runoff patterns are crucial to determine the hydrological response to climate change, especially in a seasonal frost area. In this study, multi-time runoff responses to meteoric precipitation for the period from July 2014 to June 2016 and the period from 1955 to 2010 were obtained to identify different runoff patterns in the Songhua River basin, northeast China, based on six stations. Two distinctly different runoff responses are exhibited: a periodic one in response to precipitation in the Nen River and a constant one in the Second Songhua River under different scales. Stable isotopes in the plain with diverse characteristics also supported these runoff patterns. What is more, gradual runoff relatively less sensitive to precipitation in the Second Songhua Rive was attributed to upstream dam constructions. Furthermore, the Second Songhua River contributes more water to the main stream

Jie Li Guoqiang Wang Yuanzheng Zhai (corresponding author) Engineering Research Center of Groundwater Pollution Control and Remediation, Ministry of Education, College of Water Sciences, Beijing Normal University, Beijing 100875, China E-mail: diszyz@163.com Wei Dai China Ordnance Industry Survey and Geotechnical Institute Co., Ltd, Beijing 100053, China

during January to March at the seasonal scale and in the 2000s at the annual scale, with low precipitation during those periods. This study could have implications for water management in the Songhua River basin. Key words

| base ﬂow, daily runoff, precipitation, Songhua River basin, stable isotopes

Yang Sun The Bureau of Hydrology (Information Center), Songliao Water Resources Commission, Changchun 130021, China Yihui Li ShanDong Harbor Construction Group Co., Ltd, Rizhao 276800, China

INTRODUCTION Climate change has been widely reported in recent years and

Precipitation is an important driver for the global hydrological

is expected to continue (IPCC ; Han et al. ). It is very

cycle (Dai et al. ; Sivapalan et al. ; Jimenez Cisneros

likely that precipitation has increased over the mid-latitude

et al. ), carbon cycle (Chapin et al. ; Fang et al. ),

land areas of the northern hemisphere since 1901. Meanwhile,

and so on, and has attracted the attention of hydrologists

increases in the number of days with precipitation and inten-

and decision-makers.

sity of heavy precipitation with some seasonal and/or

Precipitation has an important impact on runoff, and

regional variation have been observed over many regions,

runoff patterns with diverse runoff responses to precipitation

even where there had been a reduction in annual total precipi-

are vital to assess the impacts of precipitation change on the

tation (Goswami et al. ; Donat et al. ; Liu et al. ).

hydrological cycle (Vogel et al. ; Berghuijs et al. ). Since precipitation elasticity to runoff was ﬁrst deﬁned by

Schaake (), many studies have been conducted to identify the response of runoff to precipitation based on long-term hydrometeorological data (Sankarasubramanian et al. ;

1010

J. Li et al.

Runoff patterns determined by multi-time runoff responses to precipitation

Hydrology Research

51.5

2020

Chiew ; Novotny & Stefan ). In this case, a quick

ridge of the hills is the northwestern watershed and the ridge

response in which the maximum runoff peak coincided with

of the Wanda Mountain is the southeastern watershed of the

the rainfall peak was observed (Changnon & Kunkel ;

Songhua River basin (Figure 1). Within the river basin, the

Onda et al. ), whereas in other cases, the variability of

highest elevation is 2,691 m at the Baitou Peak of the Chang-

runoff was reduced due to check dams (Batalla et al. ;

bai Mountain and the lowest elevation is about 57 m at the

Abbasi et al. ), indicating less sensitivity to precipitation.

river outlet. The Songnen and Sanjiang Plains, with an

Unraveling runoff patterns and their spatial-temporal vari-

elevation of 57–128 m, are situated in the central and eastern

ations is critical for the prediction and management of

parts, respectively. The areas of mountain, hill, and plain

surface water resources (Birsan et al. ; Wu et al. ).

within the basin account for 42.7%, 29.1%, and 27.4% of the

Runoff patterns in the frost regions are sensitive in response

total drainage area, respectively.

to climate change (Woo & Winter ; Kong & Pang ),

The Songhua River has a total channel length of 2,309 km

which is affected by seasonal evolution of soil freezing and

with an average channel gradient of about 0.00042. The Nen

thawing. The Songhua River basin, located in the far northeast

River in the north and the Second Songhua River in the

of China, is a typical seasonal frost area. There is a clear decreas-

south are the two sources for the Songhua River (Figure 1).

ing trend in annual precipitation found from 1960 to 2009 in the

The Nen River, originating from the northern part of the Dax-

Songhua River basin (Li et al. ). Meanwhile, a declining

inganling (Greater Khingan) Mountain, has a channel length

trend at 0.81 mm season 1 a 1 was also observed for the

of 1,370 km and a channel gradient of 0.00066, while the

summer series (Liang et al. ). Consequently, the stream

Second Songhua River with a channel length of 958 km and

ﬂow in the main stream of the Songhua River basin shows a

a channel gradient of 0.00196 originates from the Tianchi

decreasing trend during 1955–2004 with dry years in the

Lake located in the central part of the Changbai Mountain.

1970s and after 2000 (Miao et al. ). Although there are sev-

The two rivers join the Songhua River at Sanchahe and ﬂow

eral studies focused on the runoff changes at certain

northeastward before joining the Heilongjiang River at the

hydrological gauge stations, tributaries, and even the whole

outlet of the river basin (Figure 1).

basin of the Songhua River (e.g., Xu & Ma ; Feng et al.

The Songhua River basin has a mean annual tempera-

; Meng & Mo ; Pan & Tang ; Wang et al. ),

ture of 3–5 C with signiﬁcant monthly temperature

the runoff patterns and their response to precipitation in the

difference. Mean temperatures are below 20 C in January

tributaries and their contributions to the main stream in the sea-

and 25 C in July. The mean annual precipitation in the river

sonal scale are rarely involved.

basin during the period 1955–2010 was 525.6 mm, more

The aims of this study are: (1) to display different runoff

than 75% of which occurred during the rainy season from

patterns of the Songhua River with precipitation–runoff

June to September (Liu et al. ). In comparison, precipi-

relationships in different time scales (daily, monthly, and

tation is very low from December to February and

annual) and isotopic evidence; and (2) to quantify the contri-

represents only 5% of the total precipitation. Furthermore,

butions of the two sources to the runoff of the main stream.

spatial distribution of precipitation across the river basin is

This work will provide important insights into future sus-

heterogeneous. The mean annual precipitation in the south-

tainable water resource management and planning in the

eastern part of the river basin is about 700–800 mm, while

study area.

that in the western part is only 400 mm.

STUDY AREA

DATA AND METHODS

This study was conducted in the Songhua River catchments in northeast China (119 520 –132 310 E, 41 420 –51 380 N). It has a

Data descriptions

drainage area of 5.57 × 105 km2 with the ridges of the Daxinganling Mountain and the Changbai Mountain as its

Data used in this study included daily runoff at Jiamusi, Yilan,

northwestern and southeastern boundaries, respectively. The

Tonghe, Harbin, Dalai, and Fuyu stations from 1 July 2014 to

J. Li et al.

1011

Figure 1

Runoff patterns determined by multi-time runoff responses to precipitation

Hydrology Research

51.5

2020

Location map of the Songhua River basin showing the six hydrological gauging stations.

30 June 2016 (Figure 1), which were obtained from the

for Dalai and Fuyu stations during 1956–2010 were taken

Songliao Water Resources Commission (http://www.slwr.

from Wang et al. () and Yu (). Isotopic data were

gov.cn). There are two hydrological stations located in the

obtained from Zhang et al. () and Yang ().

tributaries, which are Fuyu station in the Second Songhua

Data for the isotopic composition of 61 surface water

River, and Dalai station in the Nen River. Harbin, Tonghe,

samples were taken from previously published studies

Yilan, and Jiamusi stations are in the main stream from the

(Yang ; Zhang et al. ). These samples were collected

upstream to the lower reach (Figure 1). The lowermost

in both the headwater area and the central plain along the

station, Jiamusi station, controls a drainage area of 5.28 ×

Nen River and the Second Songhua River.

105 km2, accounting for 94.8% of the total drainage area. Daily precipitation amount data for the stations during the period of 2014–2016 are available from the National Meteor-

METHODS

ological Information Center of the China Meteorological

The double-mass curve plotted between precipitation and

Administration (http://data.cma.cn/). Annual runoff data

runoff is commonly used to reveal the streamﬂow change.

1012

J. Li et al.

Runoff patterns determined by multi-time runoff responses to precipitation

Hydrology Research

51.5

2020

The basic idea of double mass analysis is the gradual sum-

(Figure 2). When temperatures are low, the river freezes

mation of two variables according to the same time span,

over, and snow has not melted. The lowest runoff

with one variable as abscissa and the other as ordinate. If

recorded during a year is generally considered as the

the interaction between these two variables is stable, all

base ﬂow. Our data show that the base ﬂow increases

the corresponding points should lie in a straight line. The

from upstream to downstream, with a value of 472 m3/s

cumulative anomaly was used to identify the turning year

at Jiamusi, 423 m3/s at Yilan, 380 m3/s at Tonghe, and

in runoff changes during the period 1955–2010. The

358 m3/s at Harbin. The variation coefﬁcient for the

anomaly is deﬁned as annual value minus average value.

runoff during January to March was 0.14, 0.16, 0.10,

To calculate the contributions of total runoff from the

and 0.13 for Jiamusi, Yilan, Tonghe, and Harbin stations,

two different sources, a two-component mixing model was

respectively, and showed a steady trend during the dry

used. The two-component separation of a runoff hydrograph

season. Runoff began to increase after March, which

into runoff from Nen River and runoff from the Second

can be attributed to snow melt in the basin and the thaw-

Songhua River can be described as:

ing of the river network as temperatures rise, and is referred to as spring ﬂood. Owing to the increase in pre-

qN=S

QN ¼ QN þ QSS

(1)

cipitation (Figure 2), runoff remains high in the rainy season. There are two clear peaks in river runoff during the rainy season from July to September, which are

where QN and QSS are runoff from the Nen River and the

observed at all stations (Figure 2).

Second Songhua River, respectively, and qN=S is the pro-

Runoff at the main stream (Figure 2(a)–2(d)) started to

portion of the Nen River runoff in the total runoff from

increase in April, in accordance with the increase in precipi-

the Nen River and the Second Songhua River.

tation. There were two or three peaks in runoff observed

The variation coefficient of runoff was employed to analyze fluctuation of daily or annual runoff based on: vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u N u1 X σ 2 Cv ¼ , σ ¼ t (Q(t) Q) N Q t¼1

during the rainy season. Dalai station in the Nen River showed a similar trend to what was observed at the stations in the main stream (Figure 2(e)). The maximum runoff value was 2,840 m3/s on 29 June 2015. The lowest

(2)

where Q(t) is the daily or annual runoff for day/year (t), and Q is the average of Q(t). When the value of Cv approaches zero, the runoff tends to be constant throughout the period.

values are observed in the winter season with low precipitation, indicating the base ﬂow. Runoff characteristics for Fuyu station in the Second Songhua River were strikingly different, with only gradual variations throughout the year (Figure 2(f)). The only obvious increases were observed in the months of July 2015 and June 2016. Therefore, the runoff mechanism must be different for the two tributaries.

RESULTS Monthly precipitation-runoff characteristics Daily precipitation-runoff characteristics To further understand the intra-annual variation in runoff, There are four stations in the main stream of the Songhua

monthly mean runoff was calculated at six stations

River. Figure 2 shows the temporal variations in daily

(Figure 3). Similar to the daily runoff variations, except

runoff at these four stations. Seasonal changes in runoff

for Fuyu station in the Second Songhua River, all gauge

are similar at all stations with high values in the rainy

stations showed a synchronous tendency with an annual

season. The lowest runoff was observed during January

peak ﬂow in the rainy season. On the contrary, although

to March at all stations, which is in good agreement

the intra-annual distribution of precipitation is similar to

with the lowest precipitation recorded during this period

that of other stations, considerable changes in intra-annual

J. Li et al.

1013

Figure 2

Runoff patterns determined by multi-time runoff responses to precipitation

Hydrology Research

51.5

2020

Daily variations of runoff at Jiamusi, Yilan, Tonghe, Harbin, Dalai, and Fuyu stations in the the Songhua River (2014–2016). The dotted lines and the bars present runoff in left axis and precipitation in right axis, respectively.

J. Li et al.

1014

Figure 3

Runoff patterns determined by multi-time runoff responses to precipitation

Hydrology Research

51.5

2020

Relationships between monthly runoff and precipitation. The curves and the bars present runoff in left axis and precipitation in right axis, respectively.

variability of streamﬂow were not observed in Fuyu station.

ranges 478–3,054 and 877–3,826 m3/s, respectively. The

Therefore, the mainstream stations of the main stream

streamﬂow series at the two stations demonstrate similar

showed similar intra-annual distributions of discharge

ﬂuctuations. Both stations experienced a high extreme

to Dalai station of the Nen River, indicating a greater

ﬂow in 1960 and 1998.

ﬂow contribution from Nen River than the Upper Songhua River.

The annual runoff during the period 1956–2010, measured at the two tributaries’ stations, is also presented in Figure 4. Average runoff at Dalai station was 651 m3/s

Long-term annual runoff changes

with a range of 146 and 1,971 m3/s and a variation coefﬁcient of 0.56. The highest annual runoff was approximately

The annual streamﬂow series from the Harbin, Jiamusi,

13.5 times higher than the lowest value observed. While

Dalai, and Fuyu gauging stations are presented in Figure 4.

annual runoff decreased before 1980, it showed an increas-

Annual streamﬂow discharge velocities at the Harbin

ing trend in the 1980s. The annual runoff in the 2000s was

and Jiamusi stations were mostly concentrated in the

signiﬁcantly lower than that in the 1990s. Similar to the

J. Li et al.

1015

Figure 4

Runoff patterns determined by multi-time runoff responses to precipitation

Hydrology Research

51.5

2020

Change of mean annual runoff during 1955–2010 at Harbin, Jiamusi, Dalai, and Fuyu hydrological stations.

daily/monthly runoff variations (Figures 2 and 3), there is a

DISCUSSION

similar annual changing trend at Dalai station compared to Jiamusi and Harbin stations in the main stream. Fuyu station

The runoff responses to monthly precipitation

had an average runoff of 465 m3/s with a range of 170 to 888 m3/s. The variation coefﬁcient was 0.35, indicating a

To ascertain the controlling factors of runoff change, the

more steady ﬂow when compared to Dalai station, which

monthly

is in accordance with the daily runoff data presented in

monthly runoff are presented in Figure 5. There is a

Figures 2(f) and 3(f).

good relationship between cumulative precipitation and

Figure 5

cumulative

precipitation

Double-mass curve for cumulative monthly precipitation and cumulative monthly runoff at six stations in the Songhua River.

and

cumulative

J. Li et al.

1016

Runoff patterns determined by multi-time runoff responses to precipitation

Hydrology Research

51.5

2020

cumulative monthly runoff for all the stations. Especially

points observed in April. This indicates that snow melt

in the rainy season, streamﬂow in the study area corre-

water contributes a lot to the runoff increase in the

sponded well with precipitation, which indicates that

spring season.

the precipitation is the dominant factor affecting stream-

To further conﬁrm the effect of precipitation on the

ﬂow. Meanwhile, due to low precipitation in the dry

runoff change in the stations, the monthly precipitation

season (from November to April in the next year), most

amount and monthly runoff are plotted in Figure 6. At

points are concentrated in Figure 5. However, they are

Fuyu station in the Second Songhua River, the monthly

mostly located above the ﬁtted line with the furthest

runoff is constant when the monthly precipitation amount

Figure 6

Relationship between the monthly precipitation amount and monthly runoff in the Songhua River.

J. Li et al.

1017

Runoff patterns determined by multi-time runoff responses to precipitation

Hydrology Research

51.5

2020

is lower than 100 mm (Figure 3(f)). When the monthly pre-

(Figure 8), indicating the controlling effect of precipitation

cipitation amount is higher than 100 mm, the increasing

on runoff changes at these stations in the annual scale.

monthly runoff is observed responsively. In contrast, the

However, the cumulative anomaly of annual runoff at

obvious increase of monthly runoff is observed when the

Fuyu station in the Second Songhua River markedly differed

monthly precipitation amount exceeds 20 mm at other

from that of other stations both in the Nen River and the

stations. This means that precipitation plays a more impor-

main stream (Figure 9(a)), which displayed a gentle trend.

tant role in the runoff change of the Nen River and the

There is an obvious jump at Dalai, Harbin, and Jiamusi

main stream. Therefore, although the spatial distribution of

stations during the year of 1998 (Figure 9(a)). The same

monthly precipitation in one year displays similar periodical

‘turning year’ was also identiﬁed in the studies of Wang

change characteristics at these stations (Figure 3), the runoff

et al. () and Song et al. ().

variations performed in two diverse ways. The different performances were also obvious in the cumulative anomaly of

Isotopic evidences of runoff mechanisms

monthly runoff in Figure 7. The stations in the Nen River (Dalai) and in the main stream (Jiamusi, Harbin, Tonghe,

The stable isotope compositions of oxygen and hydrogen,

and Yilan) displayed a similar curve, with an increasing

expressed as δ 18O and δ 2H values, respectively, are com-

trend from April to October and a decreasing trend from

monly used to identify water sources (Clark & Fritz

October to April in the next year. This is in good agreement

). The δ 18O and δ 2H values in the mountain area

with the intra-annual variations of precipitation amount.

ranged from 14.9 to 12.0‰ and 106.1 to 84.4‰,

Meanwhile, a steady trend was observed at Fuyu station

respectively (Figure 10). The mean values for the Nen

in the Second Songhua River, which indicates that the inﬂu-

River in the mountain area were 13.5‰ and 94.0‰,

ence of precipitation on runoff is relatively weak compared

respectively. The Second Songhua River had similar

to that of other stations.

δ 18O and δ 2H values of 13.5‰ and 97.0‰, respectively. These values are more negative than the mean

The runoff responses to annual precipitation

value of summer precipitation at GNIP Qiqihaer station ( 9.6‰ for δ 18O), indicating recharge from higher

There is an obvious relationship between annual precipi-

elevation. The mean elevation of the catchments of the

tation and runoff at Dalai, Harbin, and Jiamusi stations

rivers was estimated to be 1,100 m asl above the elevation

Figure 7

River change trends of the cumulative anomaly of monthly runoff in the Songhua River.

J. Li et al.

1018

Figure 8

Runoff patterns determined by multi-time runoff responses to precipitation

Hydrology Research

51.5

2020

Relationship between annual precipitation amount and annual runoff (1955–2004) at (a) Jiamusi, (b) Harbin, and (c) Dalai stations.

of Qiqihaer (147 m asl) using the isotopic altitude effect

strengthening the evidence of recharge from mountainous

of 0.25‰/100 m (Clark & Fritz ). They are in the

areas both in the Second Songhua River basin and the

range of the Changbai or Daxinganling Mountains,

Nen River basin.

J. Li et al.

1019

Figure 9

Figure 10

Runoff patterns determined by multi-time runoff responses to precipitation

Hydrology Research

51.5

2020

Change trends of (a) cumulative yearly runoff and (b) cumulative anomaly of annual runoff.

| δ 18O–δ 2H relationships for rivers in the Songhua River basin. The summer amount-weighted mean isotopic composition for precipitation events is shown as a plus sign. NR and SSR represent the Nen River and the Second Songhua River, respectively.

Although similar isotopic values were observed in the

intersection of the GMWL with the river water line (RWL,

headwater area, the two rivers displayed different isotopic

dotted line in Figure 10 of δ 2H ¼ 4.3δ 18O 28.5) of the

characteristics in the central plain. The samples taken in

Second Songhua River has δ 18O and δ 2H values of

the Second Songhua River plot on a line of δ H ¼ 18

10.4‰ and 73.2‰, respectively, similar to those of

5.1δ O 21.2 (R ¼ 0.97) indicate obvious evaporation

summer precipitation. The isotopic data are therefore in

effects. The samples located on the global meteoric water

good agreement with the constant runoff observed in the

line (GMWL, black line in Figure 8 of δ H ¼ 8δ O þ 10)

Second Songhua River (Figures 2(f), 3(f) and 4), which indi-

are those sampled from the downstream river. The

cates that the runoff in the Second Songhua River is mostly

1020

J. Li et al.

Runoff patterns determined by multi-time runoff responses to precipitation

Hydrology Research

51.5

2020

composed of uniform water. In the downstream, it appears

are relatively fewer, in agreement with the accordant

to be affected by signiﬁcant surface water evaporation. For

runoff response to precipitation.

the Nen River, samples in the plain also plot below the GMWL, indicating the effect of evaporation. However, these samples are dispersive (hollow squares in Figure 10)

Contributions of tributaries to the main stream

with no obvious correlation. Considering that the majority of samples plot near the isotopic values of summer precipi-

If the total ﬂow of the Songhua River is composed of water

tation, we conclude that summer precipitation plays an

from the two tributary rivers, the proportion originating

important role in the runoff of the Nen River. This is in

from the Nen River during each day can be calculated

accordance with the high runoff ﬂow during the rainy

(Figure 11). During the period from January to March,

season, shown in Figures 2(e), 3(e), and 4.

these values are lower than 0.5, which indicates that

The insensitive runoff responses to precipitation and iso-

during this period, the Second Songhua River contributed

topic characteristics in the Second Songhua River may be

more water to the total runoff in the main stream. During

attributed to its upstream water conservancy facilities. A

the rest of the year, the Nen River appears to play a more

large number of dams in the Second Songhua River have

important role in providing water to the mean stream

been constructed since 1953, including the Baishan, Hon-

(Figure 11). We assumed that the base ﬂow is equal to the

gshi, and Fengman Dams, which can attenuate ﬂood

ﬂow of the catchment during the lowest discharge period,

waves (Jia ). On the contrary, dams in the Nen River

observed from January until March, the months with the

Figure 11

Temporal variations in the contribution of the two tributary rivers to the Songhua River on the daily scale. The gray columns show the proportion of runoff in the main stream originating from Nen River (q(N/S) in the y-axis). The parts between the dashed lines indicate periods during which the main stream receives more water from the Second Songhua River than from the Nen River.

Figure 12

Temporal variations of the proportion of runoff in the main stream originating from Nen River on the annual scale (q(N/S) in the y-axis). The mean annual runoff is from Dalai and Fuyu stations, located in the two tributaries of the Songhua River (1956–2010). The gray areas indicate the period during which the main stream received more water from the Second Songhua River.

1021

J. Li et al.

Runoff patterns determined by multi-time runoff responses to precipitation

Hydrology Research

51.5

2020

lowest precipitation. In other words, the base ﬂow of the

runoff patterns should be considered in the future water

Second Songhua River is higher than that of the Nen River.

resource management.

The annual proportion of runoff from the Nen River is shown in Figure 12. The data showed that the runoff in the main stream was dominated by water originating from

ACKNOWLEDGEMENTS

the Nen River before 1999, while the Second Songhua River contributed more in the 2000s (gray area in Figure 12).

This work is supported by the National Natural Science

In fact, this was supported by the fact that the precipitation

Foundation of China (Grants 41602276 and 41877174)

decreased in the 2000s. On the other hand, due to the higher

and the Fundamental Research Funds for the Central

contributions from the Second Songhua River in the 2000s,

Universities. Many thanks to Dr Fengping Li from Jilin

the cumulative anomaly of annual runoff at Harbin station is

University for her insightful advice and great help. We are

less than that of Dalai station (Figure 9(b).

grateful to Fei Wang for writing codes to process data.

These trends are in good agreement with the results from daily and monthly runoff calculations discussed in the previous section. In conclusion, our data show that when precipitation is low, either during the dry season (January to March), or during dry years (the 2000s), more runoff in the main stream was received from the Second Songhua River. Therefore, the impact of climate change, especially the precipitation change on streamﬂow, must be evaluated under different spatial and temporal scales, to obtain more information for water resources management.

CONCLUSIONS Using observed records of daily runoff at six gauging stations along the Songhua River during July 2014 to June 2016 and annual runoff (1956–2010) at four stations, we analyzed the runoff responses to precipitation change, and quantiﬁed the relative contribution of runoff from tributaries. Two different runoff patterns were determined: a periodic one in response to precipitation in the main stream and the Nen River; and a constant one in the Second Songhua River under different scales. The isotopic values in samples from the downstream of the Nen River plot near the isotopic values of summer precipitation, further indicating the inﬂuence of precipitation on runoff changes in this area. The steady runoff in the Second Songhua River with less sensitivity to precipitation may be due to upstream dams. A two-component mixing model showed that the Second Songhua River contributes more water to the main stream during January to March on the daily scale, and in the 2000s on the annual scale, both of which correspond to low precipitation periods. The different

REFERENCES Abbasi, N. A., Xu, X., Lucas-Borja, M. E., Dang, W. & Liu, B.  The use of check dams in watershed management projects: examples from around the world. Science of the Total Environment 676, 683–691. Batalla, R. J., Gomez, C. M. & Kondolf, M.  Reservoir-induced hydrological changes in the Ebro River basin (NE Spain). Journal of Hydrology 290, 117–136. Berghuijs, W. R., Larsen, J. R., van Emmerik, T. H. M. & Woods, R. A.  A global assessment of runoff sensitivity to changes in precipitation, potential evaporation, and other factors. Water Resources Research 53, 8475–8486. Birsan, M., Molnar, P., Burlando, P. & Pfaundler, M.  Stream flow trends in Switzerland. Journal of Hydrology 314, 312–329. Changnon, S. A. & Kunkel, K. E.  Climate-related fluctuations in midwestern floods during 1921–1985. Journal of Water Resources Planning and Management 121 (4), 326–334. Chapin, F. S., Matson, P. A. & Vitousek, P. M.  Principles of Terrestrial Ecosystem Ecology. Springer, New York, USA. Chiew, F. H. S.  Estimation of rainfall elasticity of streamflow in Australia. Hydrological Sciences Journal 51, 613–625. Clark, I. D. & Fritz, P.  Environmental Isotopes in Hydrogeology. Lewis, New York, USA, p. 328. Dai, A., Qian, T. T., Trenberth, K. E. & Milliman, J. D.  Changes in continental freshwater discharge from 1948 to 2004. Journal of Climate 22, 2773–2792. Donat, M. G., Alexander, L. V., Yang, H., Durre, I., Vose, R., Dunn, R. J. H., Willett, K. M., Aguilar, E., Brunet, M., Caesar, J., Hewitson, B., Jack, C., Klein Tank, A. M. G., Kruger, A. C., Marengo, J., Peterson, T. C., Renom, M., Oria Rojas, C., Rusticucci, M., Salinger, J., Elrayah, A. S., Sekele, S. S., Srivastava, A. K., Trewin, B., Villarroel, C., Vincent, L. A., Zhai, P., Zhang, X. & Kitching, S.  Updated analyses of temperature and precipitation extreme indices since the beginning of the twentieth century: the HadEX2 data set. Journal of Geophysical Research-Atmosphere 118, 2098–2118.

1022

J. Li et al.

Runoff patterns determined by multi-time runoff responses to precipitation

Fang, Q., Wang, G., Xue, B., Liu, T. & Kiem, A.  How and to what extent does precipitation on multi-temporal scales and soil moisture at different depths determine carbon ﬂux responses in a water-limited grassland ecosystem? The Science of the Total Environment 635, 1255–1266. Feng, X., Zhang, G. & Yin, X.  Hydrological responses to climate change in Nenjiang River basin, northeastern China. Water Resources Management 25 (2), 677–689. Goswami, B. N., Venugopal, V., Sengupta, D., Madhusoodanan, M. S. & Xavier, P. K.  Increasing trend of extreme rain events over India in a warming environment. Science 314 (5804), 1442–1445. Han, D., Wang, G., Liu, T., Xue, B., Kuczera, G. & Xu, X.  Hydroclimatic response of evapotranspiration partitioning to prolonged droughts in semiarid grassland. Journal of Hydrology 563, 766–777. IPCC  Summary for policymakers. In: Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (S. Solomon, D. Qin, M. Manning, Z. Chen, M. Marquis, K. B. Averyt, M. Tignor & H. L. Miller, eds). Cambridge University Press, Cambridge, UK and New York, USA. Jia, F.  The evaluation research on runoff effect of Fengman Hydropower Station. Water Conservancy Science and Technology and Economy 24 (2), 66–71 (in Chinese). Jimenez Cisneros, B. E., Oki, T., Arnell, N. W., Benito, G., Cogley, J. G., Doll, P. & Mwakalila, S. S.  Freshwater resources. In: Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (A. Revi, D. Satterthwaite, F. Aragon-Durand & J. CorfeeMorlot, eds). Cambridge University Press, Cambridge, UK, pp. 229–269. Kong, Y. & Pang, Z.  Evaluating the sensitivity of glacier rivers to climate change based on hydrograph separation of discharge. Journal of Hydrology 434–435, 121–129. Li, F., Zhang, G. & Xu, Y.  Spatiotemporal variability of climate and streamﬂow in the Songhua River Basin, Northeast China. Journal of Hydrology 514, 53–64. Liang, L., Li, L. & Liu, Q.  Precipitation variability in Northeast China from 1961 to 2008. Journal of Hydrology 404, 67–76. Liu, Z., Xia, Z., Yu, L. & Wang, J.  Temporal and spatial variation of characteristics of precipitation in Songhua River Basin during 1958–2009. Journal of Natural Resources 27 (6), 990–1000 (in Chinese). Liu, R., Liu, S., Cicerone, R. J., Shiu, C., Li, J., Wang, J. & Zhang, Y.  Trends of extreme precipitation in eastern China and their possible causes. Advances in Atmospheric Sciences 32 (8), 1027–1037. Meng, D. & Mo, X.  Assessing the effect of climate change on mean annual runoff in the Songhua River basin, China. Hydrological Processes 26 (7), 1050–1061.

Hydrology Research

51.5

2020

Miao, C., Yang, L., Liu, B., Gao, Y. & Li, S.  Streamflow changes and its influencing factors in the mainstream of the Songhua River basin, Northeast China over the past 50 years. Environmental Earth Sciences 63, 489–499. Novotny, E. V. & Stefan, H. G.  Stream flow in Minnesota: indicator of climate change. Journal of Hydrology 334 (3–4), 319–333. Onda, Y., Tsujimura, M., Fujihara, J. & Ito, J.  Runoff generation mechanisms in high-relief mountainous watersheds with different underlying geology. Journal of Hydrology 331, 659–673. Pan, J. & Tang, L.  Distributed hydrological simulation and runoff variation analysis for the upper basin of Songhua River. Journal Hydroelectric Engineering 32 (5), 58–63 (in Chinese). Sankarasubramanian, A., Vogel, R. M. & Limbrunner, J. F.  Climate elasticity of streamflow in the United States. Water Resources Research 37 (6), 1771–1781. Schaake, J. C.  From climate to flow. In: Climate Change and U.S. Water Resources (P. E. Waggoner, ed.). John Wiley, New York, USA, pp. 177–206. Sivapalan, M., Savenije, H. H. G. & Bloschl, G.  Sociohydrology: A new science of people and water. Hydrological Processes 26, 1270–1276. Song, X., Mu, X., Gao, P., Wang, F. & Wang, S.  Trends of runoff variation from 1900 to 2005 at Harbin Station of Songhua River. Journal of Natural Resources 24 (10), 1803–1809 (in Chinese). Vogel, R. M., Wilson, I. & Daly, C.  Regional regression models of annual streamflow for the United States. Journal of Irrigation and Drainage Engineering 125 (3), 148–157. Wang, S., Wang, Y., Ran, L. & Su, T.  Climatic and anthropogenic impacts on runoff changes in the Songhua River basin over the last 56 years (1955-2010), Northeastern China. Catena 127, 258–269. Woo, M. & Winter, T. C.  The role of permafrost and seasonal frost in the hydrology of northern wetlands in North America. Journal of Hydrology 141, 5–31. Wu, C. S., Yang, S. L. & Lei, Y. P.  Quantifying the anthropogenic and climatic impacts on water discharge and sediment load in the Pearl River (Zhujiang), China (1954– 2009). Journal of Hydrology 452–453, 190–204. Xu, J. & Ma, Y.  Response of the hydrological regime of the Yellow River to the changing monsoon intensity and human activity. Hydrological Sciences Journal l54 (1), 90–100. Yang, X.  Estimation of Groundwater Recharge and Renewal Rate Based on Environmental Isotopes in Songnen Plain. China University of Geosciences, Beijing, China. Yu, T.  Research on the Impact of Climate Change on the Annual Runoff of Typical Watershed in North China. North China Electric Power University, Beijing, China. Zhang, B., Song, X., Zhang, Y., Han, D., Yang, L. & Tang, C.  Relationship between surface water and groundwater in the second Songhua River basin. Advances in Water Science 25 (3), 336–347 (in Chinese).

First received 21 December 2019; accepted in revised form 24 April 2020. Available online 10 July 2020

1023

51.5

2020

Water balance changes in response to climate change in the upper Hailar River Basin, China Junfang Liu, Baolin Xue, Yinglan A, Wenchao Sun and Qingchun Guo

ABSTRACT Projected climate change will have a profound effect on the hydrological balance of river basins globally. Studying water balance modification under changing climate conditions is significant for future river basin management, especially in certain arid and semiarid areas. In this study, we evaluated water balance changes (1981–2011) in the upper Hailar River Basin on the Mongolian Plateau. To evaluate the hydrological resilience of the basin to climate change, we calculated two Budyko metrics, i.e. dynamic deviation (d ) and elasticity (e). The absolute magnitude of d reflects the ability of a basin to resist the influence of climate change and maintain its stable ecological function, whereas parameter e is used to assess whether a basin is hydrologically elastic. Results revealed modification of the hydrological balance during the study period has manifested as a decreasing

Junfang Liu Baolin Xue (corresponding author) Yinglan A Wenchao Sun College of Water Sciences, Beijing Normal University, Beijing 100875, China E-mail: xuebl@bnu.edu.cn Baolin Xue Wenchao Sun Beijing Key Laboratory of Urban Hydrological Cycle and Sponge City Technology, Beijing, China

trend of runoff and runoff-precipitation ratio. Correspondingly, basin-averaged evapotranspiration has also shown a decreasing trend, attributable mainly to precipitation. Furthermore, the calculated elasticity (e ¼ 8.03) suggests the basin has high hydrological resilience, which indicates the basin ecosystem may maintain its hydrological function to a certain extent under a changing climate.

Qingchun Guo School of Environment and Planning, Liaocheng University, Liaocheng, 252000, China

The results of this study could assist water resource management in the study area and the prediction of ecosystem response to future climate change. Key words

| Budyko curve, climate change, Hailar River Basin, resilience, resistance, water balance

HIGHLIGHTS

•

Water balance changes (1981–2011) in the upper Hailar River Basin on the Mongolian Plateau

•

The hydrological balance during the study period has manifested as a trend of decrease of runoff

•

were investigated.

and a decreased runoff–precipitation ratio. The calculated elasticity (e ¼ 8.03) suggests the basin has high hydrological resilience, which indicates the basin ecosystem might maintain its hydrological function to a certain extent under a changing climate.

INTRODUCTION The arid and semiarid areas that cover >50% of China’s

Tibetan Plateau and in northern China. Grassland is the

national mainland territory are distributed primarily on the

main vegetation type in such regions that are usually subjected to water limitations and intense effects associated

This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying,

with climate warming (Fang et al. a; Han et al. a;

adaptation and redistribution, provided the original work is properly cited

A et al. b). Therefore, arid and semiarid areas are ecolo-

(http://creativecommons.org/licenses/by/4.0/).

gically vulnerable and sensitive to climate extremes (Miao

doi: 10.2166/nh.2020.032

1024

J. Liu et al.

Climate change inﬂuences on water balance changes

Hydrology Research

51.5

2020

et al. ; Fang et al. b). Moreover, vegetation transpira-

absorb change induced by external factors and retain its eco-

tion

accounts for a large proportion of the total

logical function (Creed et al. ). In recent years, this

evapotranspiration in arid and semiarid regions and thus it

concept has been applied in hydrological sciences (Trenbath

plays an essential role in the regional hydrological balance

; Gerten et al. ). The concept of hydrological

(Han et al. a; Wang et al. a; A et al. a).

resilience is described as the capability of a basin to

Research has shown that drought in arid and semiarid

maintain stability in multiple hydrological equilibrium

areas is projected to become intensiﬁed in the future,

states. Creed et al. () found that climate warming was

which could trigger considerable change in the processes

projected to change forest runoff, so he calculated the resili-

of the hydrological balance and affect regional water

ence and resistance of 12 watersheds across North America

resources (Dai ; Wang et al. a). Therefore, it is of

and concluded that the forest type is the dominant factor

critical importance to study the modiﬁcation of the water

affecting the elasticity of a speciﬁc watershed. Helman

balance under the effects of climate change in arid and

et al. () calculated these two metrics of forests in the

semiarid regions to ensure sustainable water resources

Eastern Mediterranean and found that a drier climate may

management.

induce higher resilience compared with a more humid cli-

Projected global climate change in the 21st century

mate. Therefore, these two metrics have been applied and

could have a considerable effect on the hydrological balance

to some extent can reﬂect the characteristics of the river

of many river basins, especially in certain arid and semiarid

basin following the climate change. In this study, we used

regions, in terms of important variables such as precipi-

the Budyko theoretical curve to describe the relationship

tation, runoff, and evaporation (Cuo et al. ; Zhang

between basin resilience and climate change (Shen et al.

et al. ; Wang et al. b). Earlier studies have investi-

). The Budyko curve, which comprises a dryness index

gated changes of the hydrological balance due to climate

(DI ¼ PET/P) and an evaporative index (EI ¼ AET/P),

change in many river basins of northern China. In the

describes the relationship between potential evaporation

middle section of the Yellow River Basin, both streamﬂow

and actual evaporation (Troch et al. ). The Budyko

and precipitation have exhibited downward trends and

curve deﬁnes two basin states with evaporation being lim-

evaporation has presented an upward trend during the pre-

ited by either energy supply or water supply, which is

vious 60 years (He et al. ; Bao et al. ). Cuo et al.

determined by the calculated value of the DI (Figure 1). A

() analyzed observed streamﬂow changes in the upper

value of DI <1 indicates an energy-limited basin, whereas

Yellow River Basin using a modiﬁed VIC model and results

a value of DI >1 indicates a water-limited basin. Based on

showed that streamflow has decreased during recent decades. In the Kuye River Basin in Northwest China, Yang & Yang () found annual runoff has declined significantly during the past 60 years. Generally, streamflow has tended to diminish and evaporation has tended to increase in many basins of northern China owing to climate change (Wang & Hejazi ; Wang et al. , b; Shen et al. ). However, the extent to which these basins may be resistant to climatic perturbations should be explored further with regard to the prediction of basin responses to future climate change (Xue et al. ). In hydrology, the concepts of resistance and resilience, which are taken from the field of ecology, are two metrics used to quantify basin response to climate change (Zhang

Figure 1

The Budyko theoretical curve. The horizontal coordinate is the dryness index (DI ¼ PET/P) and the vertical coordinate is the evaporative index (EI ¼ AET/P).

et al. ; Williams et al. ). In ecological studies, a resi-

The solid line represents two different basin states, i.e. the energy and water limitations to the EI, and the dashed line represents the theoretical Budyko

lient ecosystem is deﬁned as one that has the ability to

curve.

1025

J. Liu et al.

Climate change inﬂuences on water balance changes

the Budyko curve, we calculated two metrics, i.e. dynamic

Hydrology Research

51.5

2020

MATERIALS AND METHOD

deviation (d ) and elasticity (e), which can be used to quantify the resilience and resistance of a basin to the effects of

Study area

climate change (Creed et al. ; Helman et al. ). Dynamic deviation, which is described as the vertical depar-

The study area comprised the upper reaches of the Hailar

ture of the EI from the corresponding value calculated using

River Basin that drains a watershed of 43,345 km2 (Fang

the theoretical Budyko curve, represents the resistance of a

et al. ). This area is located in northeastern Inner Mon-

basin in terms of the runoff change caused by climate

golia, China (47 350 –50 120 N, 118 450 –122 400 E), and it

change. A positive (negative) value of the dynamic deviation

usually experiences extreme weather caused by the inﬂu-

indicates that runoff generated is smaller (greater) than the

ence of the East Asian summer monsoon, which means it

value estimated using the Budyko theoretical equations,

is highly sensitive to climate change (Figure 2). The river

and its absolute magnitude reﬂects the extent of the runoff

originates in the Greater Khingan Mountains and ﬁnally

change relative to the inherent runoff calculated by the

joins the Erguna River (Duan et al. ). The length of

Budyko theoretical curve. Smaller values of the dynamic

the main river is 708.5 km. The study area has a temperate

deviation indicate higher basin resistance. Elasticity is the

continental monsoon climate, i.e. short cool summers and

second metric that can be used to reﬂect the hydrological

long cold winters (Fang et al. ). Mean annual precipi-

resilience of a basin, which represents the extent to which

tation and mean annual temperature in this river basin are

a watershed can hold this partitioning pattern after climate

347.6 mm and –1.2 C, respectively, according to observed

perturbations. A basin with high elasticity means runoff pre-

meteorological data during 1979–2018 (www.tpedatabase.

dictions within the basin responds highly consistently with

cn). The study area is 510–1,622 m above sea level and its

the Budyko curve, i.e. when a change in the DI results in

topography is predominantly mountains, hills, and wetlands.

a change in the EI, the ecological system moves along the Budyko curve. This study investigated the upstream area of the upper Hailar River Basin, which is situated in northeastern

Standardized precipitation and evapotranspiration index

Inner Mongolia, China. The upper Hailar River Basin is a primary tributary of the Erguna River and it is the

In this study, the standardized precipitation and evapotran-

main water source for the local industry and agriculture.

spiration index (SPEI) was selected as the drought

Moreover, its location belongs to the ecologically vulner-

monitoring index. This index considers the statistical distri-

able area. So it is meaningful to detect the water

bution of precipitation and the potential evapotranspiration

balance changes following the climate change and quan-

at the same time and it can reﬂect the regional drought

tify its hydrological resilience and resistance to climate

more comprehensively. According to Abbasi et al. (),

change for future water resource management. In this

the SPEI can be calculated as described in the following.

study, we employed the Mann–Kendall test to analyze

First, the water-year potential evapotranspiration (ET0)

the changing hydrological balance of the study basin

is calculated using the radiation-based formulation of Priest-

and adopted two metrics to determine the basin’s hydrolo-

ley and Taylor (Dewes et al. ), as shown below:

gical resilience, which is used as a supplement. The objective was to investigate the following: (1) the changes in hydroclimatic variables/climate in the study basin over the previous 30 years, (2) the basin response to the

PET ¼ 1:26

Δ (RN G) Δþγ

(1)

where RN is net radiation, Δ is the gradient of saturated

changes in climate and water balances variables, and (3)

vapor pressure, G is soil heat ﬂux, and γ is the psychrometric

the resilience of the basin to climate change. The ﬁndings

constant. The unit of the variable PET is mm/d. Second, the

of this study will assist water resources management in

difference between monthly precipitation and evapotran-

the basin.

spiration is calculated as Di ¼ Pi PETi , where i is the

J. Liu et al.

1026

Figure 2

Climate change inﬂuences on water balance changes

Hydrology Research

51.5

2020

Map of the study area and its position in China.

month counter. Third, the accumulation sequence of water

according to the following formulas:

proﬁt and loss over different timescales is established using Equation (2), where k is the timescale (here, k ¼ 12): β¼

Dkn ¼

k 1 X

(Pn i PETn i ), n k

(2)

i¼0

2w1 w0 (6w1 w0 6w2 )

(w0 2w1 )β α¼ 1 1 Γ 1þ Γ 1 β β

(4)

(5)

Fourth, the Dkn data series should be fitted and normalized to calculate the SPEI. Vicente-Serrano et al. () showed that the log-logistic density function is the best fitting function to fit the Dkn data series through contrasting

1 1 Γ 1 γ ¼ w0 αΓ 1 þ β β

the different types of parameter function. The expression of the log-logistic probability density function including three parameters is as follows:

f(x) ¼

x γ β 2 β x γ β 1 1þ α α α

ws ¼

n 1X l 0:35 s 1 Xl , n i¼1 n

(6)

(7)

where ws is the probability weighted moment, s is taken as 0, (3)

1, or 2, l is the sequence of accumulated water deﬁcit X in ascending order (X1 X2,…, Xn), and Γ(β) is the gamma function. Through the three-parameter log-logistic prob-

where α, β, and γ are the parameters of scale, shape, and

ability distribution function, the cumulative probability on

beginning, respectively. The linear moment method is

a given timescale can be calculated using Equation (7)

adopted to estimate the ﬁtting parameters of this function

(Polong et al. ). Then, the SPEI is calculated using

J. Liu et al.

1027

Climate change inﬂuences on water balance changes

Hydrology Research

Equation (9): "

51.5

2020

where P, ET, and ET0 are mean annual precipitation, mean

annual actual evaporation, and potential evapotranspira-

β # 1

F(x) ¼ 1 þ

α x γ

SPEI ¼ W

C0 þ C1 W þ C2 W 2 1 þ d1 W þ d2 W 2 þ d3 W 3

(8)

tion, respectively. Parameter w is a constant determined by the characteristics of the watershed, e.g. vegetation type

(9)

where C0 ¼ 2:515517, C1 ¼ 0.802853, C2 ¼ 0.010328, d1 ¼ 1.432788, d2 ¼ 0.189269, and d3 ¼ 0.001308 (Gao et al. ). Finally, W is calculated using Equation (11): P ¼ 1 F(x) ( pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 ln (P) P 0:5 W ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 ln (P 1) P > 0:5

(10) (11)

Having obtained the SPEI, a criterion was required to determine the occurrence of drought (Table 1). In this study, if the value of the SPEI was less than –0.5, we considered a drought phenomenon happened; if the value of the SPEI was positive, we considered the drought period over (Begueria et al. ).

and soil type (Qiu et al. ). In this study, we used the adjusted equation assuming w ¼ 2 to describe the theoretical ET0 relationship between the dryness index DI ¼ and the P ET evaporative index EI ¼ . P The parameters of deviation (d) and elasticity (e) were calculated for the studied basin to represent the potential departure from the Budyko theoretical curve of the DI and EI points (Creed et al. ). Deviation is described as the vertical departure of the EI from the corresponding B value calculated from the theoretical Budyko curve, which is composed of two parts: static and dynamic deviation. Static deviation is calculated as the mean annual EI minus its theoretical B value obtained from the mean annual DI, which is the inherent deviation with average normal climate conditions (s ¼ (EIAVG B), Figure 3). Dynamic deviation

Calculation of Budyko metrics Based on the Budyko hypothesis, the annual water balance can be described using the function of water (precipitation) and energy (potential evaporation). Among the various forms of equations for Budyko curves, we selected the following (Zhang et al. ): ET ¼ P

ET0 P ET0 ET0 1 þ 1þw P P

Table 1

1þw

(12)

Classiﬁcation of the standardized precipitation evapotranspiration index Figure 3

Drought categories

SPEI values

Extreme drought

–2.0

Severe drought

–2.0 to –1.0

Moderate drought

–1.0 to –0.5

Normal

–0.5 to 0.5

Moderate wet

0.5–1.0

Severe wet

1.0–2.0

Extreme wet

2.0

Graphical representation of three Budyko metrics: static deviation (s), dynamic deviation (d), and elasticity (e). Static deviation (s) is the inherent deviation with average normal climate conditions, which is calculated as the average EI minus the B value calculated from the Budyko theoretical curve according to the mean DI under normal climate conditions (s ¼ (EIN BN )). Dynamic deviation (d) is the departure of the mean annual EI from the B value with climate change after considering the inherent static deviation, which is the additional deviation induced by climate change [d ¼ (EId Bd ) s)]. Its absolute magnitude reﬂects the hydrologic resistance of the basin to climate change. Elasticity (e) is the ratio of the range of the DI to the range of the residual EI, which is the magnitude of the EI departure from its corresponding B value according to the Budyko theoretical cure for the entire period [e ¼ (DImax DImin )=(EIR, max EIR, min)]. High elasticity indicates a basin has hydrological resilience.

J. Liu et al.

1028

Climate change inﬂuences on water balance changes

Hydrology Research

51.5

2020

(d ) is the departure of the mean annual EI from the B value

where statistic S is the cumulative number of values at time i

with climate change after considering the inherent static

larger than at time j. Under the assumption of random inde-

deviation [d ¼ (EID BD ) s]. Thus, dynamic deviation is

pendence of the time series, the statistic UFk can be deﬁned

the additional deviation induced by climate change and its

by the following formula:

absolute magnitude reﬂects the hydrologic resistance of the basin to climate change. A value of d close to zero indicates the basin has high hydrological resistance (Helman

Si E(Si ) UFi ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Var(Si )

i ¼ 1, 2, , n

(16)

et al. ). Elasticity (e) is deﬁned as the ratio of the range of the DI (DImax DImin ) to the range of the EIR

In Equation (16), UF1 ¼ 0, and E(Sk) and Var(Sk)

(EIR, max EIR, min), which is the magnitude of the depar-

represent the expected value and the variance, respectively,

ture of the EI from its corresponding B value according to

of the cumulative value Sk , which can be obtained as

the Budyko theoretical cure for the entire period of climate

follows:

change

[e ¼ (DImax DImin )=(EIR, max EIR, min)].

The

points representing the DI and the EI with low departures from the Budyko curve tend to have large values of e, indi-

E(Si ) ¼

i(i 1) 4

cating that the basin has high elasticity, and vice versa. In addition, we used a threshold of e ¼ 1 to distinguish elastic

Var(Si ) ¼

i(i 1)(2i 5) 72

(17)

(18)

and inelastic basins. The actual evapotranspiration (ET0) was calculated based on the water balance (Xue et al. , ):

UFi is a standard normal distribution, which is a sequence of statistics calculated from the order of time

ET0 ¼ P Q 4S

(13)

where P is precipitation, R is streamﬂow, E is evapotran-

series X. Given signiﬁcance level α, a condition of jUFi j > Uα indicates an obvious trend change in the sequence. Using the inverse time series, we calculated the

spiration, and ΔS is the change of water storage volume

UFi again using the above calculation process, where

(Bao et al. ). We considered water storage negligible,

UBi ¼ UFi and i ¼ n, n 1, , 1 for the same test

assuming a steady state for the study period (1980–2011).

method as described above.

The Mann–Kendall test

Data sources

The nonparametric Mann–Kendall test can be used to analyze change trends and breakpoints of hydrological time series (Sung et al. ). It can accommodate many types of samples because it does not need the samples to follow any particular distribution and it is rarely disturbed by abnormal data (Liang et al. ). For an assumed data series X (x1 , x2 , xn ), n is the length of the data series. First, the cumulative statistic S should be calculated as follows: Sk ¼

k X

ri ¼

basin outlet, which operated during 1981–2011, because we only needed annual streamflow data of the entire river basin. The daily flow was summed to an annual amount, multiplied by the corresponding length of time, and then divided by the basin area to obtain the annual streamflow. Gridded precipitation and temperature data (0.025 × 0.025 ) measured during 1981–2011 were collected from GHCN-Monthly datasets (www.ncdc.noaa.gov/ghcnm/v3.

k ¼ 2, 3, n

(14)

i¼1

We used daily Q data from the hydrological station at the

1 xi > xj 0 xi xj

php) (Zhao et al. ). Net radiation data from 1981 to 2011 with 500-m spatial resolution and 30-d composite tem-

j ¼ 1, 2, i 1

(15)

poral resolution were downloaded from the Global Land Data Assimilation System.

J. Liu et al.

1029

Climate change inﬂuences on water balance changes

RESULTS

Hydrology Research

51.5

2020

of annual mean precipitation before 2000 was much smaller than after 2001. According to the P value calculated from

Trends of climatic variables in the Hailar River Basin

the linear regression (P ¼ –0.48), the trend of decrease of annual

mean

precipitation

was

insigniﬁcant

the

This study found that the Hailar River Basin has become

α ¼ 0:05 level (Figure 4(a)). Moreover, the results of the

warmer and drier in recent decades, consistent with the

Mann–Kendall test revealed an oscillation in mean annual

results of Han et al. (b). During the study period,

precipitation before 1998 and after 1998, indicating that

annual mean precipitation was 429.56 mm, with the highest

the trend of decrease of mean annual precipitation became

(lowest) annual mean precipitation of 621.41 (248.31) mm

slightly more evident (Figure 4(b)). Therefore, although the

in 1998 (2007). The time series of areal precipitation in

trend of decrease of mean annual precipitation was mild,

the Hailar Basin showed a decreasing trend during the

the scale of the trend intensiﬁed.

studied 30 years (Figure 4(a)). The multiyear annual mean

During 1981–2011, average annual temperature in the

precipitation was 483.83, 443.56, and 367.52 mm in 1981–

study area was –1.81 C, with the lowest (highest) average

1990, 1991–2000, and 2001–2011, respectively. The multi-

annual temperature of –3.53 C (–0.19 C) in 1984 (2007).

year mean annual precipitation during 1991–2000 was

There has been a trend of increase in average annual temp-

40.27 mm less than in 1981–1990 but 76.04 mm higher

erature over the 30-year study period (Figure 4(c)). The

than in 2001–2011, which indicates that the rate of decrease

multiyear annual mean temperature was –2.4, –1.56 and

Figure 4

Change trend of (a) annual mean precipitation and (b) annual mean temperature in the Hailar River Basin during 1981–2011. Mann–Kendall test results for (c) precipitation and (d) temperature (red line: calculated UF value, blue line: calculated UB value). Please refer to the online version of this paper to see this ﬁgure in colour: http://dx.doi.org/10.2166/ nh.2020.032.

J. Liu et al.

1030

Climate change inﬂuences on water balance changes

Hydrology Research

51.5

2020

–1.51 C in 1981–1990, 1991–2000, and 2001–2011, respect-

the SPEI in the Hailar River Basin has risen and fallen

ively. The multiyear mean annual temperature during 1991–

during the 30-year study period. During 1981–1991 (except

2000 was 0.84 C higher than in 1981–1990 but 0.04 C

1987), the UF value was positive, indicating that the value

lower than in 2001–2011, which indicates that the rate of

of the SPEI showed an overall trend of rise during this

increase in temperature has decreased slightly in compari-

period. During 1992–2011 (except 1993 and 1998), the UF

son with earlier years. Statistically, the P value (P ¼

value was <0 and it exceeded the conﬁdence interval of

0.4835) calculated from the linear regression indicates that

0.05 after 2007, showing that the SPEI decreased gradually

this trend of increase was not signiﬁcant and that its process

and that the change trend of the SPEI was signiﬁcant after

of change was uncertain (Figure 4(c)). In addition, according

2007. Within the conﬁdence interval, the UF and UB

to the results of the Mann–Kendall test, the trend of increase

curves intersect in 1998, indicating that 1998 was the break-

in annual mean temperature was signiﬁcant after 1993 at the

point of the SPEI and that drought has intensiﬁed in the

α ¼ 0:05 level. Generally, despite the uncertain trend of

Hailar River Basin since 1998 (Zhai et al. ).

change of temperature, it was clearer than that of precipitation (Figure 4(d)).

Water balance change in the Hailar River Basin

We also analyzed the change of humid/dry climate in the Hailar River Basin based on the SPEI. It was found

The water balance of the Hailar River Basin has changed

that drought intensiﬁed during the study period, especially

during the study period (Figure 6). During 1981–2011,

during 1998–2011 (Figure 5). The variations of the

annual mean runoff at the outlet hydrological station has

12-month SPEI and the Mann–Kendall test results of the

decreased at the rate of 1.46 mm/year (Figure 6(a)). The

SPEI for the Hailar River Basin during 1981–2011 are

change trend of annual mean runoff was similar to that of

shown in Figure 5. The general trend of decrease indicates

precipitation (Figure 4(a)). During 1981–1990, 1991–2000,

that this area has experienced increasing occurrence and

and 2001–2011, annual mean runoff was 102.26, 86.88,

intensiﬁcation of drought. The maximum (minimum) SPEI

and 52.51 mm, respectively. Annual mean runoff during

of 2.76 (–2.69) was in 2007 (1998), which was the wettest

1991–2000 was 15.38 mm lower than in 1981–1990 but

(driest) year in the study period. The UF curve shows that

34.37 mm higher than in 2001–2011. Therefore, the rate of annual mean runoff decrease was increased during the study period. As shown in Figure 6(a), 1998 represents the breakpoint of the annual mean runoff series. Prior to 1998, repeated oscillation occurred in the time series of annual mean runoff, whereas annual mean runoff showed a more obvious decreasing trend after 1998. Throughout the entire data series, the decreasing trend of annual runoff was insigniﬁcant at the α ¼ 0:05 level, according to the obtained P value (P ¼ –0.53). In addition, annual mean evaporation has decreased generally during the 30-year study period with a change trend similar to but more evident than that of annual mean runoff (Figure 6(b)). However, the rate of decrease of annual mean evaporation has been smaller than the change trend of precipitation. The proportions of precipitation allocated to runoff and evaporation have also changed during the study period, i.e. a

Figure 5

Variation of the calculated annual SPEI and the Mann–Kendall test results of the SPEI in the upper Hailar River Basin during 1981–2011 (red line: calculated UF value, blue line: calculated UB value). Please refer to the online version of this paper to see this ﬁgure in colour: http://dx.doi.org/10.2166/nh.2020.032.

smaller proportion of precipitation has generated runoff, while a greater proportion has been evaporated. As shown in Figure 7, there has been a general trend of decrease in

J. Liu et al.

1031

Climate change inﬂuences on water balance changes

Hydrology Research

Figure 6

Change trends of hydrological variables in the upper Hailar River Basin in the studied 30 years: (a) runoff and (b) evaporation.

Figure 7

Change trends of (a) ET/P and (b) R/P during 1981–2011.

51.5

2020

the runoff–precipitation ratio and a trend of increase in

during 1981–1990, 1991–2000, and 2001–2011, respectively.

the evaporation–precipitation ratio. The multiyear mean

It is evident that there was no change in the relationship

runoff–precipitation ratio was 0.206, 0.194, and 0.143

between precipitation and runoff during 1981–2000, i.e.

J. Liu et al.

1032

Climate change inﬂuences on water balance changes

Hydrology Research

51.5

2020

the runoff–precipitation ratio during 1991–2000 was 0.012

runoff predicted using the Budyko curve, although it was

higher than during 1981–1990, which indicates that the

very close to the theoretical runoff.

same percentage of runoff was generated at this time. How-

Dynamic deviation (d ) describes the vertical departure

ever, there was a signiﬁcant decrease in the runoff–

of points (DI, EI) in the ﬁve-year dry period from the

precipitation ratio after 2000, i.e. the decrease of 0.051

Budyko curve when considering the static deviation. Its

during 2001–2011 relative to 1991–2000 indicates less pre-

absolute magnitude reﬂects the hydrological resistance of

cipitation was converted into runoff. The change trend of

the basin to climate change, according to runoff change, in

the runoff–precipitation ratio was consistent with the trend

comparison with the runoff estimated using the Budyko

of annual mean runoff.

curve. In the Hailar River Basin, the calculated value of d was –0.0183, which is less than zero and indicates higher

Resistance and resilience of the watershed Based on the calculated SPEI, we calculated five-year moving averages for the period 1981–2011. Finally, we selected 1981–1985 as the wet period with maximum average values of the SPEI and 2004–2008 as the dry period with minimum average values of the SPEI (Figure 8). According to the DI and EI in the normal climate period, s was calculated using the equation described in the above. An obtained value of s < 0 indicated that runoff generated in the five-year wet period was larger than the predicted runoff based on the Budyko curve, and vice versa. Obtained points distributed near the Budyko theoretical curve (|s| < 0.05) indicated the pre-warming runoff fitted the predicted values using the Budyko curve, which reflected the inherent characteristics of the basin. The value of s calculated for the Hailar River Basin was

runoff than estimated according to the Budyko curve. Moreover, the absolute value of d was close to zero, which shows the extent of runoff change has been small under the effects of climate change, reflecting high hydrological resistance of the basin to climate change. The hydrological resilience of a basin is reflected by its elasticity (e). In a basin with high elasticity, runoff change under the effects of climate change is consistent with the Budyko curve. For example, as can be seen from Figure 8, when change in the DI results in change of the EI along the Budyko curve, the basin has hydrological resilience (Trenbath ). Conversely, change in the DI that leads to change in the EI that deviates from the Budyko curve reflects a basin without hydrological resilience. The value of elasticity we calculated was 8.03, which is much greater than 1 and reflects the high hydrological resilience of the Hailar River Basin.

0.035, i.e. a positive value whose absolute value is within the scope of 0.05. It reﬂects that pre-warming runoff produced under a normal climate was smaller than the

DISCUSSION During 1981–2011 in the Hailar River Basin, there was a trend of decrease in annual mean precipitation and a trend of increase in annual mean temperature (Figure 4). As the calculated results showed, the rate of decrease of precipitation and the rate of increase of temperature both increased gradually to some degree, indicating that the climate of the upper Hailar River Basin during the studied 30-year period has changed and that this change might become intensiﬁed in the future (Figure 4). Moreover, the observed decreased precipitation and increased temperature

Figure 8

Distribution of 10 points chosen from the 30-year study period according to the SPEI in the Budyko theoretical framework.

are projected to lead to intensiﬁcation of drought, consistent with the results of Wang et al. (). This type of

1033

J. Liu et al.

Climate change inﬂuences on water balance changes

Hydrology Research

51.5

2020

phenomenon is consistent with that observed in the region

(wet) period (Xue et al. ; Han et al. a; Sun et al.

of the Hulunbuir grassland during 1960–2017 (Wang et al.

). This ﬂexibility of water use efﬁciency in different

b). Under the background of climate change, the water balance in the upper Hailar River Basin has changed. Annual

periods increased the basin’s ability to cope with climate change and further enhanced the hydrological resilience of the basin to climate change (Ponce-Campos et al. ).

mean runoff has shown a decreasing trend during the study period (Figure 6(b)), consistent with the simulated results (Duan et al. ). Precipitation is the origin of

CONCLUSIONS

runoff generation and the decrease of precipitation induced the reduction of runoff. In addition, a greater proportion of

In the Hailar River Basin, the major features of climate

precipitation is now evaporated and less transformed into

change during the 30-year study period (1981–2011) are rep-

runoff (Figure 7(a) and 7(b)). However, annual mean evap-

resented by increased temperature, decreased precipitation,

oration also shown a decreasing trend due to the

and intensiﬁed drought. We used the Mann–Kendall test to

substantial decrease of precipitation. Owing to the increase

investigate the water balance changes in this basin based on

of temperature, a greater proportion of precipitation is now

certain hydrological variables, e.g. precipitation and runoff.

evaporated (Han et al. a). In addition, increased veg-

Based on hydrological data obtained during the study

etation coverage has led to a greater uptake of water and

period, it was determined that the water balance change

enhanced evapotranspiration (Zhao et al. ; Fang et al.

has been manifest as trends of decrease of runoff and evapor-

a; Bao et al. ). Thus, the proportion of precipitation

ation. In addition, we used two Budyko metrics to quantify

converted to runoff has been reduced and the amount of

the resistance and resilience of runoff in this basin to the

runoff has declined.

effects of climate change. The results showed the basin has

In our study, the calculated values of d and e were – 0.0183 and 8.03, respectively. The values of these two

high hydrological resilience and resistance and that it could retain its ecological function in a changing climate.

metrics reﬂect the high level of hydrological resistance and resilience of the Hailar River Basin, conﬁrming that the predicted runoff change has high consistency with the Budyko curve. The obtained value of e > 1 indicates that the change

ACKNOWLEDGEMENTS

in the DI during the entire study period was greater than the

This study was supported by the National Natural Science

change in the EI (Creed et al. ). Several of the factors

Foundation

China

(Grant

31670451)

and

the

that contribute to high elasticity were analyzed and among

Fundamental Research Funds for the Central Universities

the dominant inﬂuencing hydrological factors were those

(No. 2017NT18). We thank James Buxton MSc from

that contribute to certain changes in ET0. For example, sea-

Liwen Bianji, Edanz Group China (www.liwenbianji.cn./

sonal precipitation has an effect on the evaporative indices

ac), for editing the English text of this manuscript.

based on the Budyko framework (Gentine et al. ; Williams et al. ). Annual precipitation in the Hailar River Basin (30–350 mm) is concentrated primarily in summer (Fang et al. a). Given the high temperature and large quantity of precipitation in summer, potential evaporation in this water-limited area is maximized and the range

DATA AVAILABILITY STATEMENT All relevant data are included in the paper or its Supplementary Information.

of the DI is increased. Additionally, Klein et al. () argued that some basins can adjust their own ecophysiologi-

REFERENCES

cal factors to cope with drought induced by climate change, such as WUE (Xue et al. ; Yao et al. ). Therefore, water use efﬁciency in the basin is large (small) in the dry

A, Y., Wang, G., Liu, T., Xue, B. & Kuczera, G. a Spatial variation of correlations between vertical soil water and

1034

J. Liu et al.

Climate change inﬂuences on water balance changes

evapotranspiration and their controlling factors in a semiarid region. J. Hydrol. 574, 53–63. A, Y., Wang, G. Q., Liu, T. X., Shrestha, S., Xue, B. L. & Tan, Z. X. b Vertical variations of soil water and its controlling factors based on the structural equation model in a semi-arid grassland. Sci. Total Environ. 691, 1016–1026. Abbasi, A., Khalili, K., Behmanesh, J. & Shirzad, A.  Drought monitoring and prediction using SPEI index and gene expression programming model in the west of Urmia Lake. Theor. Appl. Climatol. 138, 553–567. Bao, Z., Zhang, J., Wang, G., Chen, Q., Guan, T., Yan, X., Liu, C., Liu, J. & Wang, J.  The impact of climate variability and land use/cover change on the water balance in the Middle Yellow River Basin, China. J. Hydrol. 577, 123942. https:// doi.org/10.1016/j.jhydrol.2019.123942. Begueria, S., Vicente-Serrano, S. M., Reig, F. & Latorre, B.  Standardized precipitation evapotranspiration index (SPEI) revisited: parameter fitting, evapotranspiration models, tools, datasets and drought monitoring. Int. J. Climatol. 34, 3001–3023. Creed, I. F., Spargo, A. T., Jones, J. A., Buttle, J. M., Adams, M. B., Beall, F. D., Booth, E. G., Campbell, J. L., Clow, D., Elder, K., Green, M. B., Grimm, N. B., Miniat, C., Ramlal, P., Saha, A., Sebestyen, S., Spittlehouse, D., Sterling, S., Williams, M. W., Winkler, R. & Yao, H.  Changing forest water yields in response to climate warming: results from long-term experimental watershed sites across North America. Global Change Biol. 20, 3191–3208. Cuo, L., Zhang, Y., Gao, Y., Hao, Z. & Cairang, L.  The impacts of climate change and land cover/use transition on the hydrology in the upper Yellow River Basin, China. J. Hydrol. 502, 37–52. Dai, A.  Drought under global warming: a review. Clim. Change 2, 45–65. Dewes, C. F., Rangwala, I., Barsugli, J. J., Hobbins, M. T. & Kumar, S.  Drought risk assessment under climate change is sensitive to methodological choices for the estimation of evaporative demand. PLoS One 12, e0174045. Duan, L., Liu, T., Wang, X., Luo, Y. & Wu, L.  Development of a regional regression model for estimating annual runoff in the Hailar River Basin of China. J. Water Resour. Prot. 2, 934–943. Fang, Q., Wang, G., Liu, T., Xue, B.-L. & Aa, Y. a Controls of carbon flux in a semi-arid grassland ecosystem experiencing wetland loss: vegetation patterns and environmental variables. Agric. For. Meteorol. 259, 196–210. Fang, Q., Wang, G., Xue, B., Liu, T. & Kiem, A. b How and to what extent does precipitation on multi-temporal scales and soil moisture at different depths determine carbon flux responses in a water-limited grassland ecosystem? Sci. Total Environ. 635, 1255–1266. Fang, Q., Wang, G., Liu, T., Xue, B., Sun, W. & Shrestha, S.  Unraveling the sensitivity and nonlinear response of water use efficiency to the water–energy balance and underlying surface condition in a semiarid basin. Sci. Total Environ. 699, 134405. https://doi.org/10.1016/j.scitotenv.2019.134405.

Hydrology Research

51.5

2020

Gao, X., Zhao, Q., Zhao, X., Wu, P., Pan, W., Gao, X. & Sun, M.  Temporal and spatial evolution of the standardized precipitation evapotranspiration index (SPEI) in the Loess Plateau under climate change from 2001 to 2050. Sci. Total Environ. 595, 191–200. Gentine, P., D’Odorico, P., Lintner, B. R., Sivandran, G. & Salvucci, G.  Interdependence of climate, soil, and vegetation as constrained by the Budyko curve. Geophysical Research Letters 39, L19404. http://doi.org/10.1029/ 2012gl053492. Gerten, D., Lucht, W., Schaphoff, S., Cramer, W., Hickler, T. & Wagner, W.  Hydrologic resilience of the terrestrial biosphere. Geophys. Res. Lett. 32, L21408. https://doi.org/10. 1029/2005GL024247. Han, D., Wang, G., Liu, T., Xue, B.-L., Kuczera, G. & Xu, X. a Hydroclimatic response of evapotranspiration partitioning to prolonged droughts in semiarid grassland. J. Hydrol. 563, 766–777. Han, D., Wang, G., Xue, B., Liu, T., Y, A. & Xu, X. b Evaluation of semiarid grassland degradation in North China from multiple perspectives. Ecol. Eng. 112, 41–50. He, B., Miao, C. Y. & Shi, W.  Trend, abrupt change, and periodicity of streamﬂow in the mainstream of Yellow River. Environ. Monit. Assess. 185, 6187–6199. Helman, D., Lensky, I. M., Yakir, D. & Osem, Y.  Forests growing under dry conditions have higher hydrological resilience to drought than do more humid forests. Glob. Chang. Biol. 23, 2801–2817. Klein, T., Di Matteo, G., Rotenberg, E., Cohen, S. & Yakir, D.  Differential ecophysiological response of a major Mediterranean pine species across a climatic gradient. Tree Physiol. 33, 26–36. Liang, L. Q., Li, L. J., Liu, C. M. & Cuo, L.  Climate change in the Tibetan Plateau Three Rivers Source Region: 1960–2009. Int. J. Climatol. 33, 2900–2916. Miao, L., Jiang, C., Xue, B., Liu, Q., He, B., Nath, R. & Cui, X.  Vegetation dynamics and factor analysis in arid and semi-arid Inner Mongolia. Environ. Earth Sci. 73, 2343–2352. Polong, F., Chen, H., Sun, S. & Ongoma, V.  Temporal and spatial evolution of the standard precipitation evapotranspiration index (SPEI) in the Tana River Basin, Kenya. Theor. Appl. Climatol. 138, 777–792. Ponce-Campos, G. E., Moran, M. S., Huete, A., Zhang, Y., Bresloff, C., Huxman, T. E., Eamus, D., Bosch, D. D., Buda, A. R., Gunter, S. A., Scalley, T. H., Kitchen, S. G., McClaran, M. P., McNab, W. H., Montoya, D. S., Morgan, J. A., Peters, D. P. C., Sadler, E. J., Seyfried, M. S. & Starks, P. J.  Ecosystem resilience despite large-scale altered hydroclimatic conditions. Nature 494, 349–352. Qiu, H., Niu, J. & Phanikumar, M. S.  Quantifying the space– time variability of water balance components in an agricultural basin using a process-based hydrologic model and the Budyko framework. Sci. Total Environ. 676, 176–189.

1035

J. Liu et al.

Climate change inﬂuences on water balance changes

Shen, Q., Cong, Z. & Lei, H.  Evaluating the impact of climate and underlying surface change on runoff within the Budyko framework: a study across 224 catchments in China. J. Hydrol. 554, 251–262. Sun, W., Jin, Y., Yu, J., Wang, G., Xue, B., Zhao, Y., Fu, Y. & Shrestha, S.  Integrating satellite observations and human water use data to estimate changes in key components of terrestrial water storage in a semi-arid region of North China. Sci. Total Environ. 698, 134171. https://doi. org/10.1016/j.scitotenv.2019.134171. Sung, J. H., Chung, E.-S., Kim, Y. & Lee, B.-R.  Meteorological hazard assessment based on trends and abrupt changes in rainfall characteristics on the Korean peninsula. Theor. Appl. Climatol. 127, 305–326. Trenbath, B. R.  Multispecies cropping systems in India – predictions of their productivity, stability, resilience and ecological sustainability. Agrofor. Syst. 45, 81–107. Troch, P. A., Carrillo, G., Sivapalan, M., Wagener, T. & Sawicz, K.  Climate-vegetation-soil interactions and long-term hydrologic partitioning: signatures of catchment coevolution. Hydrol. Earth Syst. Sci. 17, 2209–2217. Vicente-Serrano, S. M., Beguería, S. & López-Moreno, J. I.  A multiscalar drought index sensitive to global warming: the standardized precipitation evapotranspiration index. J. Clim. 23, 1696–1718. Wang, D. & Hejazi, M.  Quantifying the relative contribution of the climate and direct human impacts on mean annual streamflow in the contiguous United States. Water Resour. Res. 47, W00J12. https://doi.org/10.1029/2010WR010283. Wang, G., Yang, H., Wang, L., Xu, Z. & Xue, B.  Using the SWAT model to assess impacts of land use changes on runoff generation in headwaters. Hydrol. Process. 28, 1032–1042. Wang, G., Hu, X., Zhu, Y., Jiang, H. & Wang, H. a Historical accumulation and ecological risk assessment of heavy metals in sediments of a drinking water lake. Environ. Sci. Pollut. Res. Int. 25 (24), 882–24,894. Wang, G., Liu, S., Liu, T., Fu, Z., Yu, J. & Xue, B. b Modelling above-ground biomass based on vegetation indexes: a modified approach for biomass estimation in semi-arid grasslands. Int. J. Remote Sens. 40, 3835–3854. Wang, G., Li, J., Sun, W., Xue, B., Y, A. & Liu, T. a Non-point source pollution risks in a drinking water protection zone based on remote sensing data embedded within a nutrient budget model. Water Res. 157, 238–246. Wang, P. Z., Yao, J. P., Wang, G. Q., Hao, F. H., Shrestha, S., Xue, B. L., Xie, G. & Peng, Y. B. b Exploring the application of artificial intelligence technology for identification of water pollution characteristics and tracing the source of water quality pollutants. Sci. Total Environ. 693. UNSP 133440, https://doi.org/10.1016/j.scitotenv.2019.07.246. Williams, C. A., Reichstein, M., Buchmann, N., Baldocchi, D., Beer, C., Schwalm, C., Wohlfahrt, G., Hasler, N., Bernhofer, C., Foken, T., Papale, D., Schymanski, S. & Schaefer, K. 

Hydrology Research

51.5

2020

Climate and vegetation controls on the surface water balance: synthesis of evapotranspiration measured across a global network of flux towers. Water Resour. Res. 48, W06523. https://doi.org/10.1029/2011WR011586. Xue, B.-L., Komatsu, H., Kumagai, T. O., Kotani, A., Otsuki, K. & Ohta, T.  Interannual variation of evapotranspiration in an eastern Siberian larch forest. Hydrol. Process. 26, 2360–2368. Xue, B.-L., Wang, L., Li, X., Yang, K., Chen, D. & Sun, L.  Evaluation of evapotranspiration estimates for two river basins on the Tibetan Plateau by a water balance method. J. Hydrol. 492, 290–297. Xue, B.-L., Guo, Q., Otto, A., Xiao, J., Tao, S. & Li, L.  Global patterns, trends, and drivers of water use efficiency from 2000 to 2013. Ecosphere 6. UNSP 174, http://dx.doi.org/10.1890/ ES14-00416.1. Xue, B.-L., Guo, Q., Gong, Y., Hu, T., Liu, J. & Ohta, T.  The influence of meteorology and phenology on net ecosystem exchange in an eastern Siberian boreal larch forest. J. Plant Ecol. 9, 520–530. Xue, B.-L., Guo, Q., Hu, T., Xiao, J., Yang, Y., Wang, G., Tao, S., Su, Y., Liu, J. & Zhao, X.  Global patterns of woody residence time and its influence on model simulation of aboveground biomass. Glob. Biogeochem. Cycles 31, 821–835. Yang, H. & Yang, D.  Derivation of climate elasticity of runoff to assess the effects of climate change on annual runoff. Water Resour. Res. 47. UNSP 174, https://doi.org/10.1029/ 2010WR009287. Yao, J. P., Wang, P. Z., Wang, G. Q., Shrestha, S. G., Xue, B. L. & Sun, W. C.  Establishing a time series trend structure model to mine potential hydrological information from hydrometeorological time series data. Sci. Total Environ. 698, 134227. https://doi.org/10.1016/j.scitotenv.2019.134227. Zhai, J. Q., Liu, B., Hartmann, H., Su, B. D., Jiang, T. & Fraedrich, K.  Dryness/wetness variations in ten large river basins of China during the first 50 years of the 21st century. Quat. Int. 226, 101–111. Zhang, L., Dawes, W. R. & Walker, G. R.  Response of mean annual evapotranspiration to vegetation changes at catchment scale. Water Resour. Res. 37, 701–708. Zhang, Y., Pena-Arancibia, J. L., McVicar, T. R., Chiew, F. H., Vaze, J., Liu, C., Lu, X., Zheng, H., Wang, Y., Liu, Y. Y., Miralles, D. G. & Pan, M.  Multi-decadal trends in global terrestrial evapotranspiration and its components. Sci. Rep. 6, 19124. https://doi.org/10.1038/srep19124. Zhao, G. J., Tian, P., Mu, X. M., Jiao, J. Y., Wang, F. & Gao, P.  Quantifying the impact of climate variability and human activities on streamflow in the middle reaches of the Yellow River basin, China. J. Hydrol. 519, 387–398. Zhao, H., Huang, W., Xie, T., Wu, X., Xie, Y., Feng, S. & Chen, F.  Optimization and evaluation of a monthly air temperature and precipitation gridded dataset with a 0.025 spatial resolution in China during 1951–2011. Theor. Appl. Climatol. 138, 491–507.

First received 9 March 2020; accepted in revised form 15 May 2020. Available online 7 July 2020

1036

51.5

2020

The spatial pattern of periphytic algae communities and its corresponding mechanism to environmental variables in the Weihe River Basin, China Yixin Liu, Jiaxu Fu, Dandong Cheng, Qidong Lin, Ping Su, Xinxin Wang and Haotian Sun

ABSTRACT Periphytic algae is a useful indicator of aquatic ecological conditions. We investigated the periphytic algae on natural substrate and the environmental variables at 44 sites on three river systems in the Weihe River Basin (WRB). A total of 84 species are identified, representing 37 genera. The most common genera were Navicula, Oscillatoria, Nitzschia, Scenedesmus, Cymbell, and Fragilaria. One-way analysis of variance (ANOVA) indicated significant differences among the three river systems in environmental variables (p < 0.05). Non-metric multidimensional scaling (NMDS) analyses also showed differences in periphytic algae communities in the three river systems (p < 0.05) and identified different dominant species in each river system. Canonical correspondence analysis (CCA) and Monte Carlo permutation tests revealed that nutrient concentration, WT, and altitude were the most important variables affecting the structure and distribution of periphytic algae communities. Chemical variables were the most accounted for environmental variables (12.5%), while physical variable and geographical factors (5.8% in total) play a relevant minor role. Our results demonstrate that Navicula pupula, Navicula radiosq, Nitzschia palea, and Nitzschia denticula, exhibiting wide ecological amplitude, are tolerant of high concentrations of nutrient pollution. Variation of periphytic algae communities in WRB is due to the combination of anthropogenic and natural factors including agricultural and domestic wastes water inputting, land use patterns, geology, climatic changes, and river hydrology. Key words

| CCA, environmental variables, nutrients, periphytic algae, Weihe River Basin

HIGHLIGHTS

• • • • •

Structure and distribution of periphytic algae communities were investigated in the Weihe River Basin. The gradient of organic pollution detected in environmental variables was the main gradient, and the periphytic algae responded to this gradient. The types of substrates in river impact periphytic algae abundance, causing higher periphytic algae abundance in cobbles substrate. The variation of periphytic algae communities is attributed to the combination of anthropogenic factors and natural factors. The results support the periphytic algae as ecological indicators in the Weihe River Basin.

This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY-NC-ND 4.0), which permits copying and redistribution for non-commercial purposes with no derivatives, provided the original work is properly cited (http://creativecommons.org/ licenses/by-nc-nd/4.0/) doi: 10.2166/nh.2020.031

Yixin Liu Jiaxu Fu Dandong Cheng (corresponding author) Qidong Lin Ping Su Xinxin Wang Haotian Sun Shaanxi Key Laboratory of Earth Surface System and Environmental Carrying Capacity, College of Urban and Environmental Sciences, Northwest University, Xi’an 710127, China E-mail: chengdandong@hotmail.com

1037

Y. Liu et al.

Periphytic algal communities and responses to environmental variables

Hydrology Research

51.5

2020

INTRODUCTION Periphytic algae are found in bodies of water ranging from

With the reform and opening-up policy in the 1980s,

small ponds to the global ocean. They are a major contribu-

China began a period of rapid urbanization. The proportion

tor to primary productivity in the river, playing a vital role in

of the urban population increased from 17.9% in 1978 to

aquatic ecosystems (Kolmakov et al. ; Pandey et al.

52.6% in 2012, while urban built-up areas increased by

). Periphytic algae have been widely associated with

78.5% (Bai et al. ). At the same time, rapid urbanization

different geographical regions with speciﬁc environmental

has brought a series of complex water environmental pro-

conditions (Soininen ; Kovács et al. ; Chen et al.

blems. It greatly affects the hydrological system by channel

a). Composition and relative abundance of the periphy-

morphology, water quality, and aquatic biota (White &

tic algae community vary with their ecological afﬁnity and

Greer ). The Weihe River Basin (WRB) is the main

preferences (Potapova & Charles ; Tornés et al. ).

grain-producing area, and is an important industrial and com-

Their structure and distribution are reﬂected in spatially

mercial center in northwest China. The Weihe River is a major

and temporally based geomorphological conditions, climate,

source of domestic water, industrial water, and irrigation in the

hydrology, and water physicochemical properties (Pan et al.

central plain. There are 76 major cities with a total population

; Bere et al. ). Based on observations in the natural

of 22 million in the basin (Chang et al. ). Due to the

river and experiments, studies have shown that the algae

increase in population, industry, and farmland area, the WRB

will not be nutrient-limited when the nutrients concen-

also faces serious water environmental problems. Presently,

tration is P > 30 μg/L and N > 1 mg/L in the river

most studies have focused on the water quality and quantity

ecosystem (Westlake ). Water temperature affects algal

of the WRB (Jiake et al. ; Wei et al. ; Song et al. ).

photosynthetic metabolism through its control of enzyme

Research on the effects of urbanization on the aquatic biota

reaction rates (Stevenson et al. ). Periphytic algae abun-

in WRB has typically focused on ﬁsh communities and macro-

dance reaches a peak at a certain level with suitable bed

invertebrates (Wu et al. ; Su et al. ), with few studies

light levels (Yang & Flower ). Hydrodynamic conditions

available on algae communities. Studies on periphytic algae

also inﬂuence periphytic algae biomass, and the suitable

communities and their relationship with environmental vari-

velocity and river depth are 0.9–1.1 m/s and 0.40–0.48 m,

ables have scarcely been carried out in the WRB.

respectively (Wang et al. a). Periphytic algae are

This study focuses on the abundance, species structure,

effective environmental indicators due to their wide distri-

and distribution of periphytic algae communities in the WRB.

The speciﬁc objectives are to investigate the relationships

environmental changes (Stevenson et al. ). They

between periphytic algae communities and environmental vari-

respond rapidly to changes in environmental variables,

ables, to detect which ecological factors explain most of the

reﬂecting the overall ecological quality and the effects of

variation. Our results may provide a valuable baseline for

different stressors (Bona et al. ; Vasiljević et al. ).

future water quality assessments in the WRB in China.

bution

range,

numerous

species,

and

sensitivity

Individual algae species often show a clear preference for speciﬁc substrates and habitats (Winter & Duthie ). Substantial differences exist in species composition and

MATERIALS AND METHODS

abundance of periphytic algae communities from the same sites but different substrates such as sand, rock surface, submerged, or emergent macrophytes (Fisher & Dunbar ;

Study area

Bere & Tundisi ). Therefore, understanding the relation-

The study was carried out in the WRB (Figure 1), which is

ship between environmental factors and the distribution of

located in the northwest of China. The WRB (east longitude

periphytic algae communities is important for developing

103 50 –110 400 , north latitude 33 400 –37 250 ) has a drainage

algae-based water quality indices.

area of 1.35 × 105 km2, with annual mean temperatures

Y. Liu et al.

1038

Figure 1

Periphytic algal communities and responses to environmental variables

Hydrology Research

51.5

2020

The locations of the sampling sites in the Weihe River Basin.

between 7.8 and 13.5 C and the annual mean rainfall ran-

systems has a different number of sampling sites: 15 in

ging from 558 to 750 mm which increases from north to

WRS (W1–W15), 17 in JRS (J1–J17) and 12 in BRS (B1–

south (Wang et al. b). The climate in the WRB is charac-

B12) (Figure 1). For the cobble-type sedimentary rivers,

terized by the temperate continental monsoon. Therefore,

within the range of 100 m upstream and downstream of

precipitation and runoff exhibited strong inter-annual vari-

the sampling point, nine stones were selected (surface area

ations and similar intra-annual variations. The runoff from

of the stones <200 cm2), and scraped with a toothbrush to

July to October is approximately 65% of the mean annual

a 7.07 cm2 area (Kelly et al. ). For the silt-type sedimen-

runoff (Chang et al. ). Also, most of the WRB is covered

tary rivers, the surface layer of silt on the sampled stones

by highly erodible loess and soil erosion is serious because

were scraped with an area of 63.62 cm2 using an inverted

of sparse vegetation, uneven rainfall distribution, and high

Petri dish along the riverbed (Carpenter & Waite ).

intensity of heavy rain (Song et al. ). There are three

The samples were preserved with 4% formaldehyde and

different water systems in the WRB, including the Weihe

2% Lugol’s iodine solution. An aliquot of the periphytic

River System (WRS), the Jinghe River System (JRS), the Bei-

algae suspension was cleaned by using hydrochloric acid

luoRiver System (BRS), and their tributaries, as well as other

and hydrogen peroxide. The acid-cleaned samples were

small independent streams (Figure 1).

mounted on microscope slides applying the high refraction mountant Naphrax. Species were identiﬁed and a minimum

Sampling and analysis

of 400 valves were counted per slide at 1,000× magniﬁcation using a microscope (Olympus BX51, Olympus, Tokyo,

Periphyton samples were collected from 44 sites in the three

Japan). Algal cells identiﬁcation was carried out according

water systems of WBR in June 2017, each of the water

to the standard guides of Krammer & Lange-Bertalot

1039

Y. Liu et al.

Periphytic algal communities and responses to environmental variables

Hydrology Research

51.5

2020

(, , a, b), Zhu & Chen (), and Hu & Wei

structure. Both NMDS and ANOSIM analyses were per-

(). The relative abundance of each observed taxa was

formed using PRIMER version 5 (Clarke & Warwick ). A total of 18 rivers geographical, hydrological, physical

calculated. Basic physical and chemical water parameters were

and chemical variables were considered for data analysis,

measured directly at the sampling sites during sampling.

including Altit, WT, width, depth, SD, velocity, ﬂux, pH,

The water temperature (WT), pH, dissolved oxygen (DO),

DO, EC, TDS, NO3-N, NH4-N, Kþ, Mg2þ, Cl , TN and

electrical conductivity (EC), and total dissolved solids

TP. All of the environmental variables, except pH, were

(TDS) were measured in situ using a portable water quality

transformed as log10 (x þ 1) before analysis. One-way

meter (HACH HQ40d). River velocity was acquired by

analysis of variance (ANOVA) was used to evaluate

using a portable ﬂow meter (MGG/KL-DCB); river width

the signiﬁcance of the differences among the WRS, the

data was obtained with the help of Trupulse 200; river

JRS and the BRS sites based on the physical and chemical

depth was measured by using a terrain probe; Secchi

variables of the transformed water, completed by IBM

depth (SD) was measured using a Secchi disc, except for

SPSS Statistics for Windows (25). Principal component

some of the restored rivers that were transparent to the

analysis (PCA) was applied to explore the main environ-

bottom; latitude longitude and altitude (Altit) information

mental

was obtained with a Global Positioning System (GPS

CANOCO 5.0 (Ter Braak & Smilauer ).

gradients

among

the

sampling

sites,

using

Etrex 201X). Two parallel water samples were collected

Multivariate analyses are mathematical tools that detect

with a water sampler and poured into a 1,000 mL bottle

the relationship between periphytic algae and environ-

at each sampling point. The water sample was ﬁxed with

mental variables, indicating the main variables and

acid, stored in a 4 C incubator and transported back to

revealing the similarities among algae samples. For analysis

the laboratory to measure the concentrations of nitrate

of their relation, 28 species were retained with rare species

(NO3-N, mg·L–1), ammonium (NH4-N, mg·L–1), total nitro-

(<1%) removed. The removal of the species with a relative

–1

gen (TN, mg L ), total phosphorus (TP, mg·L ) following

abundance of less than 1% can minimize the inﬂuence of

the Chinese Government standard for Water and Waste-

rare species in the analysis. Algae abundance data were

water Monitoring and Analysis (). Major ions (K ,

transported by log10 (x þ 1) to stabilize variances and give

Mg2þ, Cl ) were measured using the Dionex Ion Chrom-

more weight to the larger species often found at low relative

atography System (ICS; ICS-1000, Dionex, Sunnyvale,

abundance in periphytic algae communities, which are

CA, USA).

important for deﬁning assemblage (Tison et al. ). Detrended correspondence analysis (DCA) was conducted to detect the gradient length of the algae abundance data.

Data analysis

Gradient lengths were then used to select the appropriate model (linear or unimodal model) for the constrained

The counts of each algae taxon were expressed as relative

ordinations. In this study, the longest gradient of four

abundance before the analysis. The structural properties of

ordination axes was 3.8, indicating that a linear or unimodal

the periphytic algae community at each site were used to

model could be applied to the ordination analysis (Ter

characterize by the Shannon–Weiner index (H ) and species

Braak & Verdonschot ). Hence, the unimodal ordina-

richness (S) (Bellinger et al. ; Wang et al. ). These

tion technique of canonical correspondence analysis

indicators are commonly used in the biological assessment

(CCA) was used to assess the relationships between environ-

of water quality (Spellerberg & Fedor ). Non-metric

mental variables and periphytic algae communities from

multidimensional scaling (NMDS) analyses were performed

different sites. A Monte Carlo permutation test (999 permu-

to visualize the periphytic algae structure distribution

tations, p 0.05) was used to reduce the environmental

characteristics of the WRB. Furthermore, the analysis of

variables to those correlating signiﬁcantly with the ﬁrst

similarity (ANOSIM, Bray-Curtis distance measure, 999 per-

two CCA axes. Both DCA and CCA were performed using

mutations) was used to test the difference in the community

CANOCO 5.0 (Ter Braak & Smilauer ).

Y. Liu et al.

1040

Periphytic algal communities and responses to environmental variables

Hydrology Research

51.5

2020

RESULTS Environmental gradients In this study, considerable fluctuations were identified in hydrological characteristics, water physical and chemical properties among sampling sites (Table 1). One-way ANOVA indicated that among all the environmental variables, width, pH, TDS, NO3-N, Mg2þ, and Cl– were significantly different, p < 0.05 (Table 1). The average width and the concentration of NO3-N in the BRS sites were lower than those in the WRS and the JRS (p < 0.05), while the concentrations of TDS and Cl– in the BRS sites were higher than other sites (p < 0.05). The average pH and the concentration of Mg2þ in the BRS sites was higher

Figure 2

Analysis of the principal components (PCA) of sampling sites based on environmental variables. WRS, JRS, and BRS are indicated by different symbols.

than in WRS and JRS (p < 0.001). Principal component analysis (PCA) accounted for 63.17% of the total variability in the environmental data in the ﬁrst two axes (Figure 2).

TDS (Figure 2). The second axis explained 21.44% of the

The ﬁrst axis of PCA explained 41.73% of the variance, indi-

variance, mainly correlating with pH, ﬂux, and river width

–

cating that variables were primarily related to Cl , K , and Table 1

(Figure 2).

Mean value (range in parentheses) from three water system sites in July 2017

WRS

JRS

BRS

Abbreviation

n ¼ 15

n ¼ 17

n ¼ 12

Altit

913 (340–2,452)

1,049 (410–1,950)

821 (340–1,360)

0.162

Water temperature ( C)

22 (6.5–28.5)

21.5 (9.0–25.5)

26 (20.3–29.5)

0.088

River width (m)

Width

73.99 (12.3–210)

40.21 (16.9–105)

33.13 (2.5–133)

0.033*

River depth (m)

Depth

0.47 (0.2–1)

0.47 (0.25–0.65)

0.42 (0.25–0.5)

0.671

Water velocity (m/s)

Velocity

0.54 (0.29–1.02)

0.55 (0.1–1.29)

0.52 (0.07–1.27)

0.901

Environment variables

Altitude (m)

Flux (m /s)

Flux

16.53 (2.87–70.40)

8.71 (1.93–19.53)

6.11 (0.47–16.46)

0.088

Secchi depth (m)

0.16 (0.07–0.52)

0.10 (0–0.35)

0.08 (0–0.3)

0.134

Dissolved oxygen (mg/L)

8.45 (5.55–10.03)

8.69 (7.24–11.92)

9.36 (3.98–13.93)

0.79

9.00 (8.16–9.59)

9.31 (8.01–10.05)

9.63 (9.05–10.14)

0.001**

Electrical conductivity (μs/cm)

837.07 (391–1,618)

789.02 (132.4–1,893)

1,274.33 (336–2,155)

0.075

Total dissolved solids (mg/L)

TDS

372.75 (125–845)

682.45 (89.6–2,115)

893.17 (312–1,944)

0.003*

Total nitrogen (mg/L)

5.55 (1.09–8.21)

4.81 (1.84–9.76)

5.86 (2.78–10.05)

0.423

Total phosphorus (mg/L)

0.11 (0.03–0.25)

0.20 (0.03–1.95)

0.13 (0.02–0.47)

0.808

Nitrate (mg/L)

NO3-N

2.31 (0.19–4.19)

1.21 (0.04–2.50)

1.65 (0.47–3.59)

0.018*

Ammonium (mg/L)

NH4-N

0.34 (0.09–0.63)

0.31 (0.10–0.82)

0.29 (0.01–0.74)

0.687

Potassium (mg/L)

Kþ

5.07 (0.49–14.68)

5.12 (0.50–11.87)

5.21 (3.82–7.25)

0.683

2þ

Magnesium (mg/L)

21.28 (7.64–59.54)

52.05 (4.87–131.69)

70.40 (20.6–135.72)

0.000**

Chloride (mg/L)

Cl-

145.36 (1.02–380.23)

265.71 (0.84–824.23)

486.26 (160.89–960.55)

0.008*

The value of n is the number of water samples in the sites. Variables signiﬁcantly different among the river systems with p < 0.05 and p < 0.001 are marked (based on the ANOVA). *p < 0.05; **p < 0.001.

Y. Liu et al.

1041

Periphytic algal communities and responses to environmental variables

Hydrology Research

51.5

2020

Spatial distribution of periphytic algae community

vulgare, Oscillatoria tenuis and Phormidium tenue. In the

structure

BRS sites, Navicula, Scenedesmus and Oscillatori were dominant species, but the most abundant species different

A total of 84 periphytic algae taxa species were identiﬁed,

from WRS and JRS were Navicula gracilis, Navicula

distributing in 37 genera. Only 28 species and 14 genera

pupula and Scenedesmus quadricauda.

were found to reach a relative abundance of more than

In the analyzed set of samples, the density of periphytic

1% (Figure 3). The most common genera were Navicula

algae ranged from 525 to 5.9 × 105 ind/cm2, species richness

(46.5% of all counted), Oscillatoria (5.8%), Nitzschia

from 2 to 29, and the Shannon–Weiner index from 0.58 to

(4.8%), Scenedesmus (4.0%), Cymbell (3.6%), Fragilaria

4.43 (Figure 4). The periphytic algae density showed a low

(3.3%), Diatoma (3.2%), Phormidium (3.1%), Synedra

level in the whole WRB. The average of periphytic algae

(2.8%), Cyclotella (2.0%), Srauroneis (1.9%), Cocooneis

density was 7.35 × 104 ind/cm2 while the highest density

(1.3%), Achnanthes (1.2%) and Melosira (1.0%). These 14

and the lowest density appeared at sites J1 and J11, respect-

genera accounted for 84.5% of all the values counted

ively (Figure 4(a)). The average species richness of the

(Figure 3). The dominant genus of each site differed in rela-

periphytic algae communities in WRS, JRS, and BRS were

tive abundance. In the WRS sites, Navicula, Fragilaria,

13, 11, and 14, respectively, with sites J2 and J9 displaying

Scenedesmus, and Melosira occupied a higher relative abun-

the highest and lowest species richness (Figure 4(b)). The

dance (20% in at least one sample). Navicula radiosq,

average Shannon–Weiner indexes in WRS, JRS, and BRS

Navicula palcentula, Fragilaria intermedia, Scenedesmus

were 2.90, 2.45, and 2.99, respectively. The peak and

acutiformis and Melosira granulata were the most dominant

valley values of the Shannon–Weiner index appeared at

species (a relative abundance of 10% in at least one sample).

sites J2 and J15 (Figure 4(c)). According to the results of

In the JRS sites, Navicula, Cymbella, Diatoma, Oscillatoria

NMDS and ANOSIM (r ¼ 0.291, p ¼ 0.001), the periphytic

and Phormidium occupied a higher relative abundance. The

algae community structures were signiﬁcantly different

most abundant species were Cymbella ventricosa, Diatoma

among WRS, JRS, and BRS (Figure 5).

Figure 3

Percentage composition of 14 genera (relative abundance of each genus >1%) at the sampling sites.

Y. Liu et al.

1042

Figure 4

Periphytic algal communities and responses to environmental variables

Hydrology Research

51.5

2020

Structure and distribution of periphytic algae community in the Weihe River Basin: (a) algae density, (b) species richness, (c) Shannon–Weiner index.

Mg2þ, Cl–, TN) that signiﬁcantly contributed to the algae species assemblages (Figure 6). The eigenvalues of the ﬁrst two CCA axes (λ1 ¼ 0.316, λ2 ¼ 0.133) explained 49.1 and 20.7% of the variation in species and environmental variables, respectively (Table 2). The species-environment correlations were high for both of the ordination axes (r1 ¼ 0.833, r2 ¼ 0.842). Axis 1 was positively correlated with TDS (r ¼ 0.74), EC (r ¼ 0.48), WT (r ¼ 0.41), TN (r ¼ 0.35), Kþ (r ¼ 0.41), Mg2þ (r ¼ 0.73), Cl– (r ¼ 0.69) and negatively with Altit (r ¼ –0.23). Axis 2 was positively correlated with Altit (r ¼ 0.61) and negatively with EC (r ¼ 0.49) and WT (r ¼ –0.4). Figure 5

Nonparametric multidimensional scaling (NMDS) ordination of sampling sites.

The environmental variables play important roles in the distribution of periphytic algae communities. Frequently

Relationship between periphytic algae communities

found species such as Cocconeis placentula, Navicula

and environment variables

radiosq, Navicula pupula, Stauroneis ancepse, Nitzschia denticule, Nitzschia palea and Synedra acus exhibited

CCA analyzed 28 species and 18 environmental variables

wide tolerance to environmental variables (Figure 6(a)).

at 44 sites (Supplementary material, Table A.1). The

These species were positively associated with Altit, WT,

Monte Carlo unrestricted permutation test (p < 0.05) ident-

EC, and TN. Most sites in WRS and JRS were distributed

iﬁed eight environmental variables (Altit, WT, TDS, EC, Kþ,

in the second and the third quadrant, where the samples

Y. Liu et al.

1043

Figure 6

Periphytic algal communities and responses to environmental variables

Hydrology Research

51.5

2020

Canonical correspondence analysis (CCA) showed the relationship of periphytic algae species (a) and sampling sites (b) with signiﬁcant environmental variables. Legend: sp1: Achnanthes dispar, sp2: Cocconeis placentula, sp3: Cyclotella meneghiniana, sp4: Cymbella ventricosa, sp5: Diatoma vulgare, sp6: Fragilaria intermedia, sp7: Melosira granulate, sp8: Navicula cari, sp9: Navicula cari var.angusta, sp10: Navicula dicephala, sp11: Navicula exigua, sp12: Navicula gracilis, sp13: Navicula graciloides, sp14: Navicula minuscula, sp15: Navicula palcentula, sp16: Navicula pupula, sp17: Navicula radiosq, sp18: Navicula simplex, sp19: Nitzschia denticule, sp20: Nitzschia palea, sp21: Oscillatoria tenuis, sp22: Phormidium tenue, sp23: Scenedesmus acutiformis, sp24: Scenedesmus bijugayus, sp25: Scenedesmus quadricauda, sp26: Stauroneis anceps, sp27: Synedra acus, sp28: Synedra ulna.

Table 2

Summary of the CCA of most dominant periphytic algae species composition in

environmental stressors, owing to urbanization and industri-

44 samples concerning the eight environmental variables

alization taking place in the entire watershed, contributing to

Axes

Eigenvalues

0.316

0.133

0.055

0.045

Species-environment correlations

0.833

0.842

0.725

0.665

Cumulative percentage variance of species data

15.9

22.7

25.4

27.7

Species–environment relation

49.1

69.8

78.3

85.3

nutrients and organic pollution continuously (Song et al. ). Wastewater from industrial and agricultural activities, as well as an urban settlement, is the main source that contributes a great number of solid organic pollution and thermal pollution (Milovanovic ; Luo et al. ; N’guessan et al. ). Algae are an essential component in maintaining the health of aquatic ecosystems (Kelly et al.

were closely related to Altit (Figure 6(b)). The parameter

). In general, periphytic algae are regarded as a good indi-

was highly positively associated with Navicula simplex,

cator of water quality. Many studies have demonstrated that

Scenedesmus acutiformis, Scenedesmus bijugayus and

the composition and structure of the periphytic algae commu-

Scenedesmus quadricauda. However, the sites in the BRS

nity are affected by various environmental variables

were mainly dispersed in the fourth quadrant. Samples

(Chessman et al. ; Soininen et al. ; Urrea & Sabater

were associated with high WT and EC (Figure 6(b)).

; Panahy Mirzahasanlou et al. ). In this study, we compared the species composition and distribution of periphytic algae communities among three

DISCUSSION

river systems. Though each river system was represented and dominated by different periphytic algae taxa, we found

Rivers, an important part of the ecosystem, not only have eco-

that the genera Navicula existed at all samples and species

logical functions but also provide various services for people.

N. radiosq, N. pupula and N. simplex were found in almost

However, river ecosystems are increasingly impacted by

every site. From the whole WRB, the species diversity and

1044

Y. Liu et al.

Periphytic algal communities and responses to environmental variables

Hydrology Research

51.5

2020

abundance of periphytic algae changed signiﬁcantly. Mean-

the concentration of various particles, pollutants and ions

while, we found that algae abundance was higher at the

(Quilbé et al. ). EC is an index that can reﬂect the total

sites of JRS (J1–J5), part sites of WRS (W11, W13) and BRS

ionic concentration in natural water. In this research, EC

(B4, B5, B10), where the forest coverage is high and the sub-

became an important determinant to distinguish the distri-

strate of the river is dominated by boulders and cobbles

bution of the periphytic algae communities in the WRB.

(Supplementary material, Table A.2). The impact of habitat

We found that the average values of EC concentration in

quality on the periphytic algae communities is more signiﬁ-

downstream sites was 1,110 μm/cm, higher than 759 μm/cm

cant, especially the type of substrate (Eloranta & Andersson

measured in upstream sites. TDS and nutrient levels play

; Bere & Tundisi ). Soil and water loss in the WRB

an important role in benthic periphytic algae community

is serious, and a large amount of sediment is dumped into

structure. Most of the abundant and common species were

the main river, causing soft-sediment substrate at many sites.

characteristic of eutrophic ecosystems (Van Dam et al.

Periphytic algae abundance in the soft-sediment substrate is

). In the WRB, high TN concentration from anthropo-

lower than the stone substrate (Townsend & Gell ).

genic sources are likely to cause genera Naviacula (46.5%)

Periphytic algae communities were associated with

and Nitzschia (4.8%), which have wide ecological amplitude

three sets of environmental variables in the WRB. One set

and pollution tolerance, dominating in samples (Goma et al.

is a geographic variable, and only one variable is Altit. The

; Delgado & Pardo ; Chen et al. ).

other two sets are physical and chemical variables of the

WT can affect the concentration of dissolved oxygen in

WRB. The physical variable is mainly WT. Chemical vari-

water and the respiratory rate of aquatic organisms, so it

ables consist of TDS, TN, and nutrient ion concentration.

also is a major factor affecting the periphytic algae community

For the geographical and physical variables, the WT chan-

(Chen et al. b). Moreover, because that altitude drastically

ged signiﬁcantly due to the large east–west span of the

drops from upstream to downstream in the WRB, WT also

WRB and drastic variation in altitude. The similarity of per-

changes evidently and becomes a signiﬁcant factor affecting

iphytic algae community is related to the altitude distance,

the periphytic algae. Thermal pollution is equally a reason

and the altitude gradient affects the biodiversity by affecting

for WT rising. Thermal energy absorbed and stored by

the local environmental factors (Teittinen et al. ).

urban impervious underlying surface and rainwater runoff

Chemical variables are signiﬁcantly correlated with periphy-

and point discharge from wastewater treatment plants

tic algae data indicating that they are a key factor for the

increase WT (Van Buren et al. ; Kinouchi et al. ).

distribution of periphytic algae communities (Potapova & Charles ; Soininen et al. ; Tan et al. ).

Recently, the impact of catchment-scale variables on river ecosystems has attracted researchers’ attention. A widely used

The three sets of environmental variables provided infor-

approach is to establish relationships between community pat-

mation on how they affected the periphytic algae community

terns and environmental variables (Liu et al. ). In this

structure (Leira & Sabater ; Blettler et al. ). The CCA

study, Altit, WT and nutrition concentration were retained in

indicated that the chemical variables (12.5%) were the most

the CCA as signiﬁcantly affecting algal distribution environ-

accounted for in the contribution to the periphytic algae com-

mental factors. The impact of increased nutritional levels

munity structure in the WRB, while physical variables and

caused by human activities is signiﬁcantly greater than natural

geographical factors (5.8% in total) played a relevant minor

factors. Our work has important implications for river bio-

role. The periphytic algae composition in the river system

monitoring and management in the study area, especially in

downstream sites was greatly inﬂuenced by high EC, Mg2þ,

other subtropical regions throughout China.

–

Cl and TN concentration. The high concentration of these variables may be due to the domestic sewage and industrial wastewater draining to the river (Ma et al. ). The other

CONCLUSIONS

main reason is due to soil erosion which not only carries sediment into the river, increasing water turbidity, but also

There are complex variables which affect the structure and

dissolves soluble nutrients and ions in the soil, increasing

distribution of periphytic algae communities, including

1045

Y. Liu et al.

Periphytic algal communities and responses to environmental variables

hydrological, geographical and physiochemical factors. Our results revealed that altitude, EC, TN, WT and major ions were identified as the main variables with a significant influence on the structure and distribution of periphytic algae communities. Three river systems were investigated, each of them with different dominant species, and corresponding to the different variables. Multivariate analyses are a good tool for interpreting species data but a large part of variation remained unexplained in most instances. This uncertainty may be related to other variables which can explain why variation and species have broad tolerance to the variables (Passy ; Centis et al. ; Porter-Goff et al. ; Tolkkinen et al. ). Consequently, the main determinants of variation of periphytic algae communities may result from a combination of the change in land-use patterns by man, natural phenomena including geology and climatic changes and river hydrology.

ACKNOWLEDGEMENTS This study was supported by the National Natural Science Foundation

China

(Grant

Nos.

51679200

and

51379175), Science and Technology Project of Shaanxi Provincial

Water

Resources

Department

(Grant

No.2018slkj-12).

DATA AVAILABILITY STATEMENT All relevant data are included in the paper or its Supplementary Information.

REFERENCES Bai, X. M., Shi, P. J. & Liu, Y. S.  Realizing China’s urban dream. Nature 509 (7499), 158–160. Bellinger, B. J., Cocquyt, C. & O’Reilly, C. M.  Benthic diatoms as indicators of eutrophication in tropical streams. Hydrobiologia 573, 75–87. Bere, T. & Tundisi, J. G.  The effects of substrate type on diatom-based multivariate water quality assessment in a tropical river (Monjolinho), São Carlos, SP, Brazil. Water Air Soil Pollut. 216 (1–4), 391–409.

Hydrology Research

51.5

2020

Bere, T., Mangadze, T. & Mwedzi, T.  Variation partitioning of diatom species data matrices: understanding the inﬂuence of multiple factors on benthic diatom communities in tropical streams. Sci. Total Environ. 566, 1604–1613. Blettler, M. C., Oberholster, P. J., Madlala, T., Eberle, E. G., Amsler, M. L., De Klerk, A. R., Truter, J. C., Marchese, M. R., Latosinski, F. G. & Szupiany, R.  Habitat characteristics, hydrology and anthropogenic pollution as important factors for distribution of biota in the middle Paraná River, Argentina. Ecohydrol. Hydrobiol. 19 (2), 296–306. Bona, F., Falasco, E., Fassina, S., Griselli, B. & Badino, G.  Characterization of diatom assemblages in mid-altitude streams of NW Italy. Hydrobiologia 583, 265–274. Carpenter, K. D. & Waite, I. R.  Relations of habitat-speciﬁc algal assemblages to land use and water chemistry in the Willamette Basin, Oregon. Environ. Monit. Assess. 64 (1), 247–257. Centis, B., Tolotti, M. & Salmaso, N.  Structure of the diatom community of the River Adige (North-Eastern Italy) along a hydrological gradient. Hydrobiologia 639 (1), 37–42. Chang, J., Wang, Y., Istanbulluoglu, E., Bai, T., Huang, Q., Yang, D. & Huang, S.  Impact of climate change and human activities on runoff in the Weihe River Basin, China. Quat. Int. 380, 169–179. Chen, X., Bu, Z., Stevenson, M. A., Cao, Y., Zeng, L. & Qin, B. a Variations in diatom communities at genus and species levels in peatlands (central China) linked to microhabitats and environmental factors. Sci. Total Environ. 568, 137–146. Chen, X., Zhou, W., Pickett, S. T., Li, W., Han, L. & Ren, Y. b Diatoms are better indicators of urban stream conditions: a case study in Beijing, China. Ecol. Indic. 60, 265–274. Chen, S., Zhang, W., Zhang, J., Jeppesen, E., Liu, Z., Kociolek, J. P., Xu, X. & Wang, L.  Local habitat heterogeneity determines the differences in benthic diatom metacommunities between different urban river types. Sci. Total Environ. 669 (1), 711–720. Chessman, B., Growns, I., Currey, J. & Plunkett-Cole, N.  Predicting diatom communities at the genus level for the rapid biological assessment of rivers. Freshwater Biol. 41 (2), 317–331. Clarke, K. R. & Warwick, R. M.  Changes in marine communities: an approach to statistical analysis and interpretation. Mt. Sinai J. Med. 40 (5), 689–692. Delgado, C. & Pardo, I.  Comparison of benthic diatoms from Mediterranean and Atlantic Spanish streams: community changes in relation to environmental factors. Aquat. Bot. 120, 304–314. Eloranta, P. & Andersson, K.  Diatom indices in water quality monitoring of some South-Finnish rivers. SIL Proc. 1922– 2010 26, 1213–1215. Fisher, J. & Dunbar, M.  Towards a representative periphytic diatom sample. Hydrol. Earth Syst. Sci. Discuss. 11 (1), 399–407. Goma, J., Rimet, F., Cambra, J., Hoffmann, L. & Ector, L.  Diatom communities and water quality assessment in

1046

Y. Liu et al.

Periphytic algal communities and responses to environmental variables

Mountain Rivers of the upper Segre basin (La Cerdanya, Oriental Pyrenees). Hydrobiologia 551, 209–225. Hu, H. J. & Wei, Y. X.  The Freshwater Algae of China: Systematics, Taxonomy and Ecology. Science Press, Beijing, China. Jiake, L., Huaien, L., Bing, S. & Yajiao, L.  Effect of non-point source pollution on water quality of the Weihe River. Int. J. Sediment Res. 26 (1), 50–61. Kelly, M., Cazaubon, A., Coring, E., Dell’Uomo, A., Ector, L., Goldsmith, B., Guasch, H., Hürlimann, J., Jarlman, A. & Kawecka, B.  Recommendations for the routine sampling of diatoms for water quality assessments in Europe. J. Appl. Phycol. 10 (2), 215–224. Kinouchi, T., Yagi, H. & Miyamoto, M.  Increase in stream temperature related to anthropogenic heat input from urban wastewater. J. Hydrol. 335 (1–2), 78–88. Kolmakov, V., Anishchenko, O., Ivanova, E., Gladyshev, M. & Sushchik, N.  Estimation of periphytic microalgae gross primary production with DCMU-fluorescence method in Yenisei River (Siberia, Russia). J. Appl. Phycol. 20 (3), 289–297. Kovács, C., Kahlert, M. & Padisák, J.  Benthic diatom communities along pH and TP gradients in Hungarian and Swedish streams. J. Appl. Phycol. 18 (2), 105–117. Krammer, K. & Lange-Bertalot, H.  Bacillariophyceae 1 Teil: Naviculaceae. Süsswasserflora von Mitteleuropa. Gustav Fischer Verlag, Jena. Krammer, K. & Lange-Bertalot, H.  Bacillariophyceae 2. Teil: Bacillariaceae, Epthemiaceae, Surirellaceae. Süsswasserflora von Mitteleuropa. Gustav Fischer Verlag, Stuttgart. Krammer, K. & Lange-Bertalot, H. a Bacillariophyceae 3. Teil: Centrales, Fragilariaceae, Eunotiaceae. Süsswasserflora von Mitteleuropa. Gustav Fischer Verlag, Stuttgart. Krammer, K. & Lange-Bertalot, H. b Bacillariophyceae 4. Teil: Achnanthaceae, Kritische Erg?nzungen zu Navicula (Lineolatae) und Gomphonema. Süsswasserflora von Mitteleuropa. Gustav Fischer Verlag, Stuttgart. Leira, M. & Sabater, S.  Diatom assemblages distribution in Catalan rivers, NE Spain, in relation to chemical and physiographical factors. Water Res. 39 (1), 73–82. Liu, S., Xie, G., Wang, L., Cottenie, K., Liu, D. & Wang, B.  Different roles of environmental variables and spatial factors in structuring stream benthic diatom and macroinvertebrate in Yangtze River Delta, China. Ecol Indic 61, 602–611. Luo, H. B., Luo, L., Huang, G., Liu, P., Li, J. X., Hu, S., Wang, F. X., Xu, R. & Huang, X. X.  Total pollution effect of urban surface runoff. J. Environ. Sci. 21 (9), 1186–1193. Ma, J., Ding, Z., Wei, G., Zhao, H. & Huang, T.  Sources of water pollution and evolution of water quality in the Wuwei basin of Shiyang river, Northwest China. J. Environ. Manage. 90 (2), 1168–1177. Milovanovic, M.  Water quality assessment and determination of pollution sources along the Axios/Vardar River, Southeastern Europe. Desalination 213 (1–3), 159–173. N’guessan, Y. M., Probst, J.-L., Bur, T. & Probst, A.  Trace elements in stream bed sediments from agricultural

Hydrology Research

51.5

2020

catchments (Gascogne region, SW France): where do they come from? Sci. Total Environ. 407 (8), 2939–2952. Pan, Y., Herlihy, A., Kaufmann, P., Wigington, J., Van Sickle, J. & Moser, T.  Linkages among land-use, water quality, physical habitat conditions and lotic diatom assemblages: a multi-spatial scale assessment. Hydrobiologia 515 (1–3), 59–73. Panahy Mirzahasanlou, J., Ramezanpour, Z., Nejadsattari, T., Imanpour Namin, J. & Asri, Y.  Temporal and spatial distribution of diatom assemblages and their relationship with environmental factors in Balikhli River (NW Iran). Int. J. Ecohydrol. Hydrobiol. 20 (1), 102–111. Pandey, L. K., Bergey, E. A., Lyu, J., Park, J., Choi, S., Lee, H., Depuydt, S., Oh, Y.-T., Lee, S.-M. & Han, T.  The use of diatoms in ecotoxicology and bioassessment: insights, advances and challenges. Water Res. 118, 39–58. Passy, S. I.  Diatom ecological guilds display distinct and predictable behavior along nutrient and disturbance gradients in running waters. Aquat. Bot. 86 (2), 171–178. Porter-Goff, E. R., Frost, P. C. & Xenopoulos, M. A.  Changes in riverine benthic diatom community structure along a chloride gradient. Ecol. Indic. 32, 97–106. Potapova, M. G. & Charles, D. F.  Benthic diatoms in USA rivers: distributions along spatial and environmental gradients. J. Biogeogr. 29 (2), 167–187. Potapova, M. & Charles, D. F.  Distribution of benthic diatoms in US rivers in relation to conductivity and ionic composition. Freshwater Biol. 48 (8), 1311–1328. Quilbé, R., Rousseau, A. N., Duchemin, M., Poulin, A., Gangbazo, G. & Villeneuve, J.-P.  Selecting a calculation method to estimate sediment and nutrient loads in streams: application to the Beaurivage River (Québec, Canada). J. Hydrol. 326 (1–4), 295–310. Soininen, J.  Responses of epilithic diatom communities to environmental gradients in some Finnish rivers. Int. Rev. Hydrobiol. 87 (1), 11–24. Soininen, J., Paavola, R. & Muotka, T.  Benthic diatom communities in boreal streams: community structure in relation to environmental and spatial gradients. Ecography 27 (3), 330–342. Song, J., Xu, Z., Hui, Y., Li, H. & Li, Q.  Instream ﬂow requirements for sediment transport in the lower Weihe River. Hydrol. Processes 24 (24), 3547–3557. Song, J., Yang, X., Zhang, J., Long, Y., Zhang, Y. & Zhang, T.  Assessing the variability of heavy metal concentrations in liquid-solid two-phase and related environmental risks in the Weihe river of shaanxi province, China. Int. J. Environ. Res. Public Health 12 (7), 8243–8262. Spellerberg, I. F. & Fedor, P. J.  A tribute to Claude Shannon (1916–2001) and a plea for more rigorous use of species richness, species diversity and the ‘Shannon–Wiener’Index. Global Ecol. Biogeogr. 12 (3), 177–179. Stevenson, R. J., Bothwell, M. L., Lowe, R. L. & Thorp, J. H.  Algal Ecology: Freshwater Benthic Ecosystem. Academic Press, USA.

1047

Y. Liu et al.

Periphytic algal communities and responses to environmental variables

Stevenson, R. J., Pan, Y., Manoylov, K. M., Parker, C. A., Larsen, D. P. & Herlihy, A. T.  Development of diatom indicators of ecological conditions for streams of the western US. J. N. Am. Benthol. Soc. 27 (4), 1000–1016. Su, P., Wang, X., Lin, Q., Peng, J., Song, J., Fu, J., Wang, S., Cheng, D., Bai, H. & Li, Q.  Variability in macroinvertebrate community structure and its response to ecological factors of the Weihe River Basin, China. Ecol. Eng. 140, 105595. Tan, X., Ma, P., Xia, X. & Zhang, Q.  Spatial pattern of benthic diatoms and water quality assessment using diatom indices in a subtropical river, China. Clean Soil Air Water 42 (1), 20–28. Teittinen, A., Kallajoki, L., Meier, S., Stigzelius, T. & Soininen, J.  The roles of elevation and local environmental factors as drivers of diatom diversity in subarctic streams. Freshwater Biol. 61 (9), 1509–1521. Ter Braak, C. J. & Šmilauer, P.  Canoco Reference Manual and User’s Guide: Software for Ordination, Version 5.0. Microcomputer Power Press, Ithaca, USA. Ter Braak, C. J. & Verdonschot, P. F.  Canonical correspondence analysis and related multivariate methods in aquatic ecology. Aquat. Sci. 57, 255–289. Tison, J., Park, Y.-S., Coste, M., Wasson, J., Ector, L., Rimet, F. & Delmas, F.  Typology of diatom communities and the inﬂuence of hydro-ecoregions: a study on the French hydrosystem scale. Water Res. 39 (14), 3177–3188. Tolkkinen, M., Mykrä, H., Virtanen, R., Tolkkinen, M., Kauppila, T., Paasivirta, L. & Muotka, T.  Land use impacts on stream community composition and concordance along a natural stress gradient. Ecol. Indic. 62, 14–21. Tornés, E., Cambra, J., Gomà, J., Leira, M., Ortiz, R. & Sabater, S.  Indicator taxa of benthic diatom communities: a case study in Mediterranean streams. Int. J. Limnol. 43 (1), 1–11. Townsend, S. A. & Gell, P. A.  The role of substrate type on benthic diatom assemblages in the Daly and Roper Rivers of the Australian wet/dry tropics. Hydrobiologia 548, 101–115. Urrea, G. & Sabater, S.  Epilithic diatom assemblages and their relationship to environmental characteristics in an agricultural watershed (Guadiana River, SW Spain). Ecol. Indic. 9 (4), 693–703. Van Buren, M., Watt, W. E., Marsalek, J. & Anderson, B.  Thermal enhancement of stormwater runoff by paved surfaces. Water Res. 34 (4), 1359–1371. Van Dam, H., Mertens, A. & Sinkeldam, J.  A coded checklist and ecological indicator values of freshwater diatoms from the Netherlands. Netherland J. Aquat. Ecol. 28 (1), 117–133.

Hydrology Research

51.5

2020

Vasiljević, B., Simić, S. B., Paunović, M., Zuliani, T., Krizmanić, J., Marković, V. & Tomović, J.  Contribution to the improvement of diatom-based assessments of the ecological status of large rivers – the Sava River case study. Sci. Total Environ. 605, 874–883. Wang, J., Meier, S., Soininen, J., Casamayor, E., Pan, F., Tang, X., Yang, X., Zhang, Y., Wu, Q., Zhou, J. & Shen, J.  Regional and global elevational patterns of microbial species richness and evenness. Ecography 40 (3), 393–402. Wang, H., Li, Y., Li, J., An, R., Zhang, L. & Chen, M. a Influences of hydrodynamic conditions on the biomass of benthic diatoms in a natural stream. Ecol. Indic. 92, 51–60. Wang, W., Song, J., Zhang, G., Liu, Q., Guo, W., Tang, B., Cheng, D. & Zhang, Y. b The influence of hyporheic upwelling fluxes on inorganic nitrogen concentrations in the pore water of the Weihe River. Ecol. Eng. 112, 105–115. Wei, F.  Water and Waste Water Monitoring and Analysis Methods, 4th edn. Water and Waste Water Monitoring and Analysis Method Committee, China Environmental Science Press, Beijing (in Chinese). Wei, S., Yang, H., Song, J., Abbaspour, K. C. & Xu, Z.  System dynamics simulation model for assessing socio-economic impacts of different levels of environmental flow allocation in the Weihe River Basin, China. Eur. J. Oper. Res. 221 (1), 248–262. Westlake  Temporal Changes in Aquatic Macrophytes and Their Environment. Dynamique de Populations et Qualite de l’Eau, pp. 109–138. White, M. D. & Greer, K. A.  The effects of watershed urbanization on the stream hydrology and riparian vegetation of Los Penasquitos Creek, California. Landsc. Urban Plan. 74 (2), 125–138. Winter, J. G. & Duthie, H. C.  Stream epilithic, epipelic and epiphytic diatoms: habitat fidelity and use in biomonitoring. Aquat. Ecol. 34 (4), 345–353. Wu, W., Xu, Z., Yin, X. & Zuo, D.  Assessment of ecosystem health based on fish assemblages in the Wei River basin, China. Environ. Monit. Assess. 186 (6), 3701–3716. Yang, H. & Flower, R. J.  Effects of light and substrate on the benthic diatoms in an oligotrophic lake: a comparison between natural and artificial substrates. J. Phycol. 48 (5), 1166–1177. Zhu, H. Z. & Chen, J. Y.  Bacillariophyta of the Xizang Plateau. Science Press, Beijing, China.

First received 15 March 2020; accepted in revised form 12 June 2020. Available online 14 August 2020

1048

51.5

2020

Spatiotemporal variation and tendency analysis on rainfall erosivity in the Loess Plateau of China Yongsheng Cui, Chengzhong Pan, Chunlei Liu, Mingjie Luo and Yahui Guo

ABSTRACT Rainfall erosivity is an important factor to be considered when predicting soil erosion. Precipitation data for 1971–2010 from 39 stations located in the Loess Plateau of China were collected to calculate the spatiotemporal variability of rainfall erosivity, and the long-term tendency of the erosivity was predicted using data from the HadGEM2-ES model. Statistical analyses were done using Mann– Kendall statistic tests and ordinary Kriging interpolation. The results showed that the annual mean rainfall erosivity in the Loess Plateau decreased from 1,286.02 MJ mm hm 2 h 1 a 1 in 1971–1990 to 1,201.46 MJ mm hm 2 h 1 a 1 in 1991–2010 and mainly occurred in July to August. The rainfall erosivity decreased from the southeast to the northwest of the Loess Plateau and was closely related to the annual precipitation amount. However, the effect of annual precipitation on rainfall erosivity weakened under climate change: the annual precipitation increased and the rainfall erosivity decreased. Climate change, however, had little inﬂuence on the spatial variation in rainfall erosivity in the Loess Plateau. The results obtained can facilitate the prediction of spatial and temporal variations in soil erosion in the Loess Plateau. Key words

| Loess Plateau of China, rainfall erosivity, Representative Concentration Pathway (RCP) scenarios, spatiotemporal variation, tendency prediction

HIGHLIGHTS

• • •

Both precipitation and rainfall erosivity showed an insigniﬁcant decreasing trend for 1971–2010. The climate model predicts an increasing precipitation but decreasing rainfall erosivity. South and southeast of the Loess Plateau are areas susceptible to rainfall erosion under climate change.

Yongsheng Cui Chengzhong Pan (corresponding author) Chunlei Liu Mingjie Luo Yahui Guo Key Laboratory of Water Sediment Sciences, College of Water Sciences, Beijing Normal University, Beijing 100875, China E-mail: pancz@bnu.edu.cn

1049

Y. Cui et al.

Spatiotemporal variation of rainfall erosivity

Hydrology Research

51.5

2020

GRAPHICAL ABSTRACT

INTRODUCTION Soil erosion is one of the most important ecological problems

of rainfall erosivity in Brazil. Some other researchers have con-

in the world. Nearly a third of arable land has been exposed to

ducted spatiotemporal analyses of rainfall erosivity in different

soil erosion, which has increased at a rate of 1.0 × 10 hm per

regions (Angulo-Martínez & Beguería ; Panagos et al.

year (Pimentel et al. ). Soil erosion removes fertile soil and

; Xie et al. ). A high-resolution distribution map of rain-

seriously threatens agricultural production and food security.

fall erosivity has been established using global data, and the

The eroded soil usually enters rivers, causing siltation and

global rainfall erosivity was estimated to be about 2,190.0

water eutrophication, and thus endangers water safety and

MJ mm hm 2 h 1 (Panagos et al. ). Due to the difﬁculty

affects normal production and life. Predicting and protecting

in obtaining rainfall data with a high temporal resolution,

against soil erosion are of great signiﬁcance to the construction

Vrieling et al. () used satellite data and combined rainfall

of an ecological civilization and to the sustainable develop-

intensity and rainfall erosivity to establish a method for fore-

ment

Consequently,

casting rainfall erosivity, thereby demonstrating a potential

mechanisms of soil erosion and measures to control it have

the

economy

and

society.

tool for soil erosion prediction in data-poor areas. Previous

been hot research topics (Lal ; Pimentel ).

studies have enhanced our understanding of spatiotemporal

Rainfall is the driving force and prerequisite for soil ero-

variations in rainfall erosivity, but these studies were con-

sion (Zhang et al. ). The Universal Soil Loss Equation

strained by the limited availability of daily rainfall data and

(USLE) devised by Wischmeier & Smith () and the

by the length of rainfall data. Studying available long-term rain-

Revised Universal Soil Loss Equation (RUSLE) proposed by

fall data and estimating rainfall erosivity could provide

Renard & Freimund () and Renard et al. () for the

information on the potential trends of soil erosion, especially

United States are widely used to predict soil erosion (Anees

in areas that are lacking data.

et al. ). The rainfall erosivity (R-factor) in these models indi-

In the Loess Plateau of China, intensive rainfall and

cates the ability of rainfall to cause soil erosion and is

sparse vegetation can lead to relatively serious soil erosion.

considered the best factor for studying the response of soil ero-

The soil erosion modulus can reach 1.0 × 107 kg km 2 a 1.

sion to rainfall changes (Nearing et al. ). Meusburger et al.

A large amount of sediment enters the Yellow River and

() collected rainfall data for 1989–2010 and used the pro-

causes many problems, including river channel siltation and

duct of the total kinetic energy of rainfall (E) and the 30-min

the deterioration of aquatic ecosystems (Liu & Liu ).

maximum rainfall intensity (I30) as the measurement index of

Many researchers have analyzed the spatial and temporal dis-

rainfall erosivity to study the temporal and spatial variation

tribution of rainfall erosion on the Loess Plateau in different

of rainfall erosivity in Switzerland. Da Silva () used rain-

periods (Xin et al. ; Abd Elbasit et al. ; Yang & Lu

fall data in conjunction with geographic information system

). Fu et al. () calculated all the RUSLE parameters

(GIS) for spatial interpolation to map the spatial distribution

using GIS based on regional data from the Yanhe watershed

1050

Y. Cui et al.

Spatiotemporal variation of rainfall erosivity

Hydrology Research

51.5

2020

and concluded that the annual soil loss was 0.5–2.0 ×

6.35 × 105 km2. Most areas of the plateau have semi-humid

107 kg km 2. Sun et al. () analyzed the effects of topogra-

or semi-arid climates, and the climates in different areas are

phy and land-use patterns on soil erosion in the Loess Plateau

quite different. The annual mean temperature is 3.6–14.3 C,

and suggested that ‘Grain-to-Green Program’ is effective for

and the annual evaporation is 1,400–2,000 mm. The main

preventing soil erosion. Xu () carried out a quantitative

crops grown are wheat, corn, soybeans, and sorghum. The

analysis of the correlations between vegetation coverage

study area is characterized by complex landforms and deep

(Cf ), annual rainfall erosivity (Re), and annual precipitation

soil layers (mainly loose and easily eroded dark loessial soil

(Pm). Xu’s results indicated that Re will increase rapidly

and loessal soil). The annual precipitation ranges from 200

when Pm > 300 mm, and that when Pm > 530 mm, the rate

to 750 mm, with large inter-annual and seasonal variations.

at which Re increases with Pm becomes higher. Current

There are frequent storms, which are the main driving force

research generally focuses on speciﬁc watersheds where

of soil erosion. The annual mean soil erosion modulus of

daily rainfall data are easily available. Consequently, there

the Loess Plateau is 0.5–1.0 × 107 kg km 2 a 1, and the thick-

are relatively few papers addressing the long-term spatiotem-

ness of the soil layer lost every year is about 1 cm.

poral variation of rainfall erosivity in the Loess Plateau. More analysis is needed for the region as a whole based on available

Data

long-term daily rainfall data. The results of the analysis could form the basis for soil loss prediction in the region. Moreover,

We used data from the 39 representative meteorological

the impact of climate change on soil erosion needs to be

stations located in the Loess Plateau. The daily rainfall

understood (Almagro et al. ). Few studies, however,

data from 1971 to 2010 were selected for investigating the

have been carried out that have made predictions and ana-

spatiotemporal variability of precipitation. The location of

lyses of variations in rainfall erosivity under projected

each meteorological station is shown in Figure 1: SRTM

climate change in the Loess Plateau of China. Consequently,

(Shuttle Radar Topography Mission) data from http://srtm.

a climate change perspective and decision basis cannot be

csi.cgiar.org/srtmdata/, with spatial resolution approxi-

provided for decision makers.

mately 30 meters on the line of the equator. Daily rainfall

The overall aim of this study was to address these knowl-

data were obtained from the China Meteorological Data

edge gaps. Precipitation data from 1971 to 2010 were

Service Center (http://data.cma.cn/). The Linfen meteorolo-

collected from 39 representative stations in the Loess Plateau

gical station does not have data for 2000, and the Yongji

of China to analyze the spatiotemporal variability of rainfall

station does not have data for 1981–1990.

erosivity. The main objectives of this study are:

In addition to the data from the above meteorological

1. to determine the spatial and temporal variability of

stations, data were also obtained using a GCMs (Hadley Center Global Environment Model version 2 (HadGEM2-

annual rainfall erosivity in the Loess Plateau; 2. to predict the change tendency of future rainfall erosivity using a typical global circulation model (GCM).

ES)) forced by two Representative Concentration Pathway (RCP) scenarios – RCP4.5 and RCP8.5 (Chou et al. ; Yan et al. ). The HadGEM2-ES is an earth system category GCM that was developed by the Hadley Center, and the resolution is about 1.875 for longitude and 1.25 for latitude (Collins et al. ). The RCP scenarios were developed

MATERIALS AND METHODS

by the research community by considering emissions, conStudy area

centrations, and land-use trajectories and are labeled according to the expected values of global radiative forcing

The Loess Plateau is located between 100 52 –114 31 E and

in 2100. RCP4.5 is considered an intermediate scenario

33 370 –41 250 N, including most or part of Shanxi, Shaanxi,

that assumes emissions reduction during the 21st century

Gansu, Qinghai, and Henan Provinces and the Inner

by employing clean technologies and stringent climate pol-

Mongolia and Ningxia Regions. The total area is about

icies. This scenario predicts a global forcing radiation of

Y. Cui et al.

1051

Figure 1

Spatiotemporal variation of rainfall erosivity

Hydrology Research

51.5

2020

Location map of meteorological stations and sample spots on the Loess Plateau.

∼4.5W m2 and concentrations of ∼650 ppm CO2-eq at

months. In this study, half-monthly rainfall erosivity esti-

stabilization after 2100. In contrast, RCP8.5 is a kind of

mates were obtained using the model proposed by Zhang

business-as-usual scenario characterized by no implemen-

et al. (). The model has been shown to be suitable for

tation of climate policies, low rates of technology

the prediction of rainfall erosivity for the Loess Plateau

development, high energy intensities, and a reliance on

(Wu et al. ). Rainfall erosivity was calculated as follows:

fossil fuels that leads to higher emissions than RCP4.5 over time. Global forcing radiation and atmospheric concentrations of CO2-eq are projected to reach >8.5 W m

and

>1,370 ppm in 2100, respectively (Almagro et al. ). Daily precipitations for 2020–2100 are available for the RCP scenarios. The boundaries of the Loess Plateau were taken to 33.75 –41.25 N and 101.25 –114.375 E. Daily rain-

Ri ¼ α

k X

(Pj )β

(1)

j¼1

where Ri is the rainfall erosivity index in the ith half-month period (MJ·mm·hm 2·h 1); k is the number of days in the half-month period; Pj is the erosive rainfall (mm) on the

fall data for 48 sample spots were obtained (Figure 1) from

jth day in the half-month period. Daily precipitation is

https://esgf-data.dkrz.de/search/cmip5-dkrz/.

required to be greater than or equal to 12 mm, otherwise calculated as 0, and the threshold of 12 mm is consistent with the Chinese standard for erosive rainfall. The parameters α, β need to be determined for the model and are calculated

METHODS

from the following formulae:

Calculation of the rainfall erosivity values

α ¼ 21:586β 7:1891

The USLE and the RUSLE models use an algorithm and

β ¼ 0:8363 þ

daily precipitation to determine rainfall erosivity in half-

18:177 24:455 þ Pd12 Py12

(2) (3)

Y. Cui et al.

1052

Spatiotemporal variation of rainfall erosivity

where Pd12 is the average daily precipitation (mm) with a

Hydrology Research

51.5

2020

RESULTS

daily rainfall of 12 mm or more; Py12 represents the annual mean precipitation (mm) with a daily rainfall of 12 mm or

Annual variation in precipitation and erosive rainfall

more. During the calculation, the half-month period is deﬁned as the 15th day of each month: the 1st to 15th day

Data from the Linfen and Yongji meteorological stations

of each month is one half month, and the rest of the

were excluded from the analysis because data from some

month is calculated as another half month. The whole

years were missing. Data of the remaining 37 meteorological

year is divided into 24 half-month periods, and the monthly,

stations were averaged to obtain the precipitation variation

annual rainfall erosivity, and annual mean rainfall erosivity

for the Loess Plateau from 1971 to 1990 and from 1991 to

are obtained.

2010 (Figure 2(a) and 2(b)). The mean precipitation from

In the present study, the average rainfall erosivity refers

1971 to 1990 and from 1991 to 2010 was 425.26 and

to the average of data from different meteorological stations

399.87 mm, respectively. The annual precipitation ranges

over a given time, while the annual mean rainfall erosivity

were 205.39 mm (1971–1990) and 251.99 mm (1991–

refers to the average of data over a period of time for a

2010), and the coefﬁcients of variation were 0.15 and 0.14,

speciﬁc location.

respectively. In general, years with a mean precipitation of >500 mm in the Loess Plateau from 1971 to 2010 accounted for only 7.5% of the years, and years with less than 400 mm accounted for 45.0% (1971–1990 accounted for 15.0%). The

DATA PROCESSING AND ANALYSIS METHODS

annual precipitation showed an overall decreasing trend from 1971 to 2010.

MATLAB R2018b (The MathWorks, Inc., USA) and

The mean erosive rainfall on the Loess Plateau from

ORIGIN 2020 (OriginLab Corporation, USA) were used

1971 to 1990 and from 1991 to 2010 was 242.59 and

for statistical analysis and mapping of data. The ordinary

229.49 mm, respectively, which accounted for about 57.0%

Kriging interpolation method and the geostatistical modules

of the mean precipitation, and showed a decreasing trend

of ArcGIS 10.2 (Esri, Inc., USA) were used for the spatial

from 1971 to 2010. There is a strong linear regression

interpolation of precipitation and rainfall erosivity. In

relationship between mean erosive rainfall and precipitation

addition, the non-parametric Mann–Kendall trend analysis

(R 2 ¼ 0.93, P < 0.05). The results of the Mann–Kendall test

method was used to test the signiﬁcance of changes in

showed that the erosive rainfall trend from 1971 to 2010

meteorological elements (Hamed & Ramachandra Rao

has increased from 1973 to 1981 and decreased from 1982

; Ahmad et al. ).

to 2010 (except for 1985 and 1996). Both trends are

Figure 2

Annual variation of erosive rainfall and precipitation on the Loess Plateau for (a) 1971–1990 and (b) 1991–2010.

Y. Cui et al.

1053

Spatiotemporal variation of rainfall erosivity

Hydrology Research

51.5

2020

insigniﬁcant and changeable, which indicates that erosive

erosivity and precipitation was not strong. Taking 1989

rainfall is not only affected by the precipitation amount

and 1990 as an example, the rainfall erosivity was basically

but also reﬂects the annual changes in precipitation.

the same, but the difference in precipitation was 17.5%. This shows that rainfall erosivity is also affected by rainfall distribution and other factors. The Mann–Kendall tests showed

Temporal variation of rainfall erosivity

that, in general, the inter-annual changes in rainfall erosivity

The intra-annual distribution of rainfall erosivity for the years from 1971 to 2010 is shown in Figure 3(a) and 3(b). The rainfall erosivity during the year formed a single-peak distribution, ﬁrst increasing and then decreasing. The annual precipitation distribution and the rainfall erosivity

increased from 1975 to 1981 and decreased from 1982 to 2010. This is similar to the trends for erosive rainfall, but, as shown in Figure 4(a) and 4(b), the annual decrease rate in rainfall erosivity during 1991–2010 was slightly slower than that during 1971–1990, and eventually stabilized.

followed a similar pattern: in January and February, the rainfall erosivity was almost zero, it then gradually increased,

Spatial distribution of rainfall erosivity

reaching a peak in July and August, and then gradually decreased to zero in November. The total rainfall erosivity

As shown in Figure 5, the annual mean rainfall erosivity

in July to August accounted for 61.5% (1971–1990) and

decreased from southeast to northwest. For the area as a

59.4% (1991–2010) of the year’s rainfall erosivity, indicating

whole, the annual mean rainfall erosivity was 1,286.02 and

that summer rainfall, especially in July and August, was the

1,201.46 MJ mm hm 2 h 1 a 1 in 1971–1990 and 1991–

main cause of soil erosion. In 1991–2010, the precipitation

2010, respectively. The maximum values obtained (for the

decreased slightly from that of 1971 to 1990, but the maxi-

Wutai Mountain and Anze station in the province of

mum still occurred in July and August, and was about

Shanxi) were 2,433.14 MJ mm hm 2 h 1 a 1 (1971–1990)

for both series.

and 2,144.56 MJ mm hm 2 h 1 a 1 (1991–2010). The mini-

The average rainfall erosivity for 1971–1990 and 1991–

mum values obtained for the two periods (for the Jingyuan

200.0 MJ mm hm

2010 was 1,278.68 and 1,200.60 MJ mm hm 2 h 1, respect-

station,

Gansu 2

province)

were

277.57

and

298.94

ively, and the coefﬁcients of variation were 0.24 and 0.19,

MJ mm hm

respectively. Rainfall erosivity was essentially stable at

fall erosivity from 1991 to 2010 was slightly lower than

around

1,200 MJ mm hm 2 h 1

a , respectively. The annual mean rain-

2010

that from 1971 to 1990 and showed a very high decrease

(Figure 4(b)). Overall, rainfall erosivity ﬂuctuated with the

in the northern part of Shanxi Province. The areas with

increases and decreases in annual precipitation, but there

the greatest rainfall erosivity were, however, concentrated

were some years where the correlation between rainfall

in the southeast of the Loess Plateau. Consequently, most

Figure 3

from

2004

Relation between half-monthly rainfall erosivity and monthly precipitation in the Loess Plateau for (a) 1971–1990 and (b) 1991–2010.

Y. Cui et al.

1054

Spatiotemporal variation of rainfall erosivity

Hydrology Research

51.5

2020

Figure 4

Annual variation in rainfall erosivity for (a) 1971–1990 and (b) 1991–2010.

Figure 5

Spatial distribution of rainfall erosivity on the Loess Plateau for (a) 1971–1990 and (b) 1991–2010. The spatial variation of annual mean rainfall erosivity from 2020 to 2100 under RCPs is shown in Supplementary material, Appendix A.

of Shanxi Province, central and southern Shaanxi were sus-

Provinces, and below 300.0 mm a 1 for the Ningxia and

ceptible to rainfall erosion from 1971 to 2010. In these

Inner Mongolia Regions, while the annual mean precipi-

areas, soil conservation measures should be strengthened

tation value distribution from 1971 to 1990 was relatively

to reduce soil erosion.

discrete in Gansu Province, which was about 170.60– 598.97 mm a 1. The annual mean precipitation from 1991

Tendency analysis of rainfall erosivity

to 2010 was 143.45–585.32 mm a 1, which was 5.5% lower than that from 1971 to 1990. The province with the

Precipitation data for representative areas of the Loess Pla-

highest mean precipitation changed from Shanxi Province

teau are shown in Table 1. (The data were obtained using

in 1971–1990 to Shaanxi Province in 1991–2010, while

ordinary Kriging interpolation. Because there are very few

Ningxia still had the lowest. The standard deviation (STD)

representative meteorological stations that could be selected

of the average precipitation in Shanxi Province decreased

in Henan and Qinghai Provinces, and the areas of these Pro-

by 32.8% from 1971–1990 to 1991–2010. The reduction in

vinces are relatively small, these Provinces are not included

the other regions was, however, much smaller.

in Tables 1 and 2). The annual mean precipitation from 1971 1

Under the RCP4.5 scenario, the annual mean precipi505.12 ±

to 1990 was 398.50 ± 114.80 mm a . The annual mean pre-

tation

cipitation was about 500.0 mm a 1 for Shanxi and Shaanxi

130.02 mm a 1, which is 26.8 and 34.1% higher than the

from

2020

2100

would

Y. Cui et al.

1055

Table 1

Spatiotemporal variation of rainfall erosivity

Hydrology Research

Annual precipitation statistics for different provinces in the Loess Plateau (mm a

51.5

2020

)

Inner Mongolia

Gansu

Ningxia

Shanxi

Shaanxi

Observed (1971–1990)

Max. Min. Mean STD

501.44 137.53 297.82 92.78

598.97 170.60 436.66 129.79

493.70 170.24 256.47 90.63

546.51 443.33 504.33 94.69

608.61 279.45 497.25 91.18

Observed (1991–2010)

Max. Min. Mean STD

430.08 143.45 296.46 86.95

549.67 186.27 402.31 113.46

459.83 171.36 249.50 85.19

532.82 403.91 463.91 63.67

585.32 298.40 471.32 82.82

RCP4.5 (2020–2100)

Max. Min. Mean STD

518.10 206.40 341.16 67.42

1,050.60 245.03 546.09 105.58

585.14 296.13 420.12 –

826.60 229.48 680.54 89.13

1,048.18 400.28 537.73 108.90

RCP8.5 (2020–2100)

Max. Min. Mean STD

519.07 219.93 350.56 65.47

1,012.28 263.41 542.25 96.70

585.44 316.37 428.91 –

861.09 238.75 699.95 101.37

1,012.40 411.85 538.33 105.46

Table 2

Statistics on rainfall erosivity for different provinces in the Loess Plateau (MJ mm hm

)

Inner Mongolia

Gansu

Ningxia

Shanxi

Shaanxi

Observed (1971–1990)

Max. Min. Mean STD

1,584.97 274.67 901.20 364.44

1,807.94 251.49 1,066.96 515.89

1,290.51 316.21 491.31 208.05

2,405.43 831.70 1,681.49 469.92

2,054.79 524.64 1,557.78 342.67

Observed (1991–2010)

Max. Min. Mean STD

1,311.57 300.76 830.76 335.34

1,678.56 281.53 984.16 455.61

1,318.07 305.11 573.06 257.84

2,090.51 877.03 1,512.01 360.49

2,026.55 787.54 1,511.63 316.75

RCP4.5 (2020–2100)

Max. Min. Mean STD

967.26 161.32 447.29 164.89

1,513.95 194.80 560.28 77.91

776.07 262.86 459.26 –

2,573.76 263.82 1,466.75 309.46

2,104.44 515.64 914.64 222.85

RCP8.5 (2020–2100)

Max. Min. Mean STD

1,031.66 135.53 476.01 143.54

1,515.07 235.42 629.24 102.97

920.24 349.35 542.92 –

3,494.06 236.67 1,705.36 429.95

2,184.66 571.43 988.91 279.86

means for 1971–1990 and 1991–2010, respectively. The

219.93–1,012.40 mm a 1, which was an increase of about

maximum

reach

35.9% compared with 1991–2010. The regional variations

1,050.60 mm a 1. The spatial distribution of precipitation

precipitation

RCP4.5

would

in annual mean precipitation were essentially the same as

similar to the spatial distribution for 1971–1990: Shanxi

those in the RCP4.5 scenario, with a maximum of

Province had the largest annual mean precipitation of

1,000.0 mm a 1 (Gansu and Shaanxi) and a minimum of

680.54 mm a 1, while the provinces with the lowest precipi-

only 219.93 mm a 1 (Inner Mongolia).

tation changed from Ningxia (1971–2010) to Inner 1

From historical data (Table 2), the annual mean rainfall

Mongolia, with a precipitation of only 341.16 mm a . The

erosivity in the Loess Plateau from 1971 to 1990 was calcu-

range of precipitation under the RCP8.5 scenario was

lated as 251.49–2,405.43 MJ mm hm 2 h 1 a 1, with an

1056

Y. Cui et al.

Spatiotemporal variation of rainfall erosivity

Hydrology Research

51.5

2020

average of 1,137.75 ± 487.57 MJ mm hm 2 h 1 a 1. Shanxi

from 1971–1980 to 1981–1990 was 604.89 to 394.74

Province had the highest rainfall erosivity, which reached

MJ mm hm 2 h 1 a 1. In the southern Gansu Province, the

2,405.43 MJ mm hm

from 1971 to 1990, followed

central part of Shaanxi Province, and the northern Shanxi

by Shaanxi Province with 1,557.78 MJ mm hm 2 h 1 a 1.

Province, rainfall erosivity from 1981 to 1990 was less

The rainfall erosivity in Ningxia was low, and the STD was

than that in 1971–1980, while the rainfall erosivity in Ning-

also low. The STD for Gansu was the largest; this indicates a

xia and some parts of Gansu Province and the southern part

large spatial difference, which is similar to the results obtained

of Shanxi Province increased the most. The spatial variation

from the analysis of data from meteorological stations for

of rainfall erosivity from 1991 to 2010 was not the same as

1991–2010. The annual mean rainfall erosivity from 1991 to

that from 1971 to 1990. After the 2000s, rainfall erosivity

2010 was 1,082.32 ± 418.69 MJ mm hm

a . Compared

increased signiﬁcantly in southern Shaanxi Province and

with 1971–1990, except for Ningxia, which showed an erosiv-

the central part of Shanxi Province, while the other areas

ity increase of 16.6% for 1991–2010, other provinces and

showed an overall decreasing trend. For example, rainfall

regions all showed decreases. For example, erosivity in

erosivity in Qinghai Province and Inner Mongolia decreased

Shanxi and Shaanxi decreased by 10.1 and 3.0%, respectively,

by about 429.88 MJ mm hm 2 h 1 a 1.

which represents small spatial differences.

Under the RCP4.5 scenario, the variation of rainfall ero-

The calculated rainfall erosivity in future climate scen-

sivity from 2020–2060 to 2061–2100 was 66.89 to

arios was lower than that obtained from the historical

595.59 MJ mm hm 2 h 1 a 1. The areas with increased rain-

data. Under the RCP4.5 and RCP8.5 scenarios, the rainfall

fall erosivity are concentrated in the southern Loess Plateau,

erosivity in 2020–2100 would be 769.64 ± 433.32 and

particularly in southern Shanxi and in the central and

868.49 ± 507.97 MJ mm hm 2 h 1 a 1, which is a 28.9 and

southern part of Shaanxi Province, while the rainfall erosiv-

19.8% decrease, respectively, compared with 1991–2010.

ity in the other regions in 2061–2100 is slightly less than that

The areas with the highest rainfall erosivity are still in

in 2020–2060. In contrast, under the RCP8.5 scenario, the

Shanxi Province in the RCP scenarios, while the lowest

rainfall erosivity increased greatly and was concentrated in

areas change from being in Ningxia (according to the his-

the southeast of the Loess Plateau (southern Shanxi and

torical data) to Inner Mongolia. These results are similar

northern Henan Provinces). In the other regions, rainfall

to the results for the distribution of precipitation. The mini-

erosivity in RCP8.5 also increased, and the overall increase

mum values for erosivity from 2020 to 2100 would be

was around 99.19–2,009.03 MJ mm hm 2 h 1 a 1. This indi-

161.32 MJ mm hm

135.53

cates that the rainfall erosivity increased during the second

MJ mm hm 2 h 1 a 1 (RCP8.5). In the RCP8.5 scenario,

half of the 21st century under the RCP8.5 scenario, and

the rainfall erosivity ranges from 135.53 to 3,494.06

hence, the degree of soil erosion in the southeast of the

MJ mm hm 2 h 1 a 1 and is 12.8% higher than the RCP4.5

Loess Plateau would also increase.

(RCP4.5)

and

values. The value distribution under RCP8.5 is, however, more discrete than under RCP4.5. The maximum erosivity in Shanxi Province increases by 35.8% from RCP4.5 to

DISCUSSIONS

RCP8.5, while the increases in the other provinces and regions are small. This arises because heavy rainfall

Temporal changes in rainfall erosivity

becomes more concentrated in Shanxi Province in the RCP8.5 scenario.

The mean precipitation on the Loess Plateau was

The rainfall erosivity in different periods was divided

412.57 mm from 1971 to 2010 and showed an insigniﬁcant

into two parts according to the length of the series, after

decreasing trend, with an annual decrease of about

the annual mean rainfall erosivity was interpolated by

0.88 mm. These results are similar to those in the study by

ordinary Kriging in ArcGIS in every part, and the rainfall

Wu et al. (). The mean precipitation in 2020–2100, how-

erosivity variation ﬁgure (Figure 6) was obtained using

ever, increased by 30.3% (RCP4.5) and 32.1% (RCP8.5)

raster calculation. The variation in the rainfall erosivity

compared with that in 1971–2010. The spatial distribution

Y. Cui et al.

1057

Figure 6

Spatiotemporal variation of rainfall erosivity

Hydrology Research

51.5

2020

Variation of annual mean rainfall erosivity for (a) 1981–1990 minus 1971–1980, (b) 2001–2010 minus 1991–2000, and (c and d) 2061–2100 minus 2020–2060.

of future precipitation indicates that precipitation will

mean rainfall erosivity from 1991 to 2010 decreased by

increase in the southeast of the study area and decrease in

5.0%. Similarly, the rainfall erosivity from 2020 to 2100 was

the northwest. Consequently, Shanxi Province may have

28.9% (RCP4.5) and 19.8% (RCP8.5) less than that in 1991–

an increased risk of soil erosion.

2010. Almagro et al. () also obtained similar results for

Precipitation distribution causes annual rainfall erosiv-

Brazil. The similarity in results may have been caused by

ity to appear as a unimodal distribution (Figure 3). The

the rainfall in the future climate scenarios being relatively dis-

maximum half-month rainfall erosivity between 1971–1990

persed, with few short-duration heavy rainfall events. An

and 1991–2010 was about 200.0 MJ mm hm

and

alternative explanation is that the HadGEM2-ES model out-

occurs mainly in July to August, which indicates that

puts are based on meteorological factors and ignore the

heavy rainfall during the summer is the main cause of soil

effect of topography on rainfall intensity (Djebou et al.

erosion in the Loess Plateau. The comprehensive utilization

). Consequently, the model outputs may lead to underes-

managements),

timations of the erosivity of future rainfall. For further

vegetation measures (e.g. afforestation), and agricultural

research, comprehensive analysis should be combined with

measures (e.g. no-tillage) are the key elements of ecological

factors such as topography and land-use data. However,

environmental development that are required to prevent soil

despite the aforementioned limitations, the results of the pre-

erosion in the Loess Plateau.

sent study can still be used as a reference for the study of soil

engineering

measures

(e.g.

slope

In terms of inter-annual changes, the annual mean rain-

erosion under climate change in the future.

1,139.75 ±

To improve the analysis of the relationship between

487.57 MJ mm hm 2 h 1 a 1, with an insigniﬁcant decreas-

rainfall erosivity and precipitation over various time scales,

ing rate of 1.83 MJ mm hm 2 h 1 a 1 (Figure 4). When

the present study made use of the Standardized Precipi-

compared with the erosivity from 1971 to 1990, the annual

tation Index (SPI) and used the SPI PROGRAM to

fall

erosivity

from

1971

1990

was

Y. Cui et al.

1058

Spatiotemporal variation of rainfall erosivity

Hydrology Research

51.5

2020

calculate the SPI value (obtained from https://drought.unl.

April and reached a maximum in May, while in normal and

edu/droughtmonitoring/SPI/SPIProgram.aspx).

time

humid years, the maxima were often reached in August and

scale of the historical data was 36 and 60 months, and 60

September, and then gradually decreased and reached zero

and 96 months for the estimated data. The results are

in November. The rainfall erosivity from July to September

shown in Supplementary material, Appendix B. Based on

in drought years accounted for about 50.0% for the whole

the SPI value and the classiﬁcation criteria of McKee et al.

year, while in normal and humid years, this period accounted

(), cases were selected for a moderate drought year

for 75.7–90.0% of the year’s total.

The

( 1.49 SPI 1.00), a normal year ( 0.99 SPI 0.99),

A discussion of the data obtained from the RCP scen-

and a moderate humid year (1.00 SPI 1.49). The selected

arios follows. When compared with historical data, it can

hydrological years and SPI values are shown in Table 3.

be seen that precipitation derived from the RCP scenarios

It can be seen from Figure 7 that the monthly rainfall

is more dispersed and is mainly concentrated in April to

erosivity has a good corresponding relationship with precipi-

September, with the peak generally occurring in July and

tation. However, the precipitation amount is not the

August (Figure 7(c) and 7(d)). Although the precipitation

determining condition for the variation in rainfall erosivity.

increased from April to June, most of the precipitation

For example, the precipitation in June 2005 decreased by

events were of long-duration and low-rainfall intensity,

11.1% compared with May 2005, but the rainfall erosivity

with daily precipitation less than 12 mm. This precipitation,

increased by 74.9%. The increase was mainly the result of

therefore, had little effect on the distribution of annual rain-

the greater rainfall intensity in summer. The threshold of

fall erosivity. The long-duration, low-intensity rainfall events

erosive rainfall adopted in this paper was a daily rainfall

were the results of climate change and other factors. Precipi-

of P 12 mm. Consequently, a lot of the precipitation was

tation from April to June accounted for about 30.0% of the

excluded and could not be used as erosive rainfall when cal-

annual precipitation on the whole, while rainfall erosivity

culating rainfall erosivity. The use of the erosive rainfall

during this period only accounted for about 25.0% of the

threshold also leads to deviations between rainfall erosivity

annual rainfall erosivity. Except for the heavy rainfall erosiv-

and precipitation (Xie et al. ).

ity that occurred in July in humid years, the monthly rainfall

The percentage of precipitation from April to June in

erosivity values for different hydrological years were essen-

moderate drought years was relatively large and accounted

tially below 500.0 MJ mm hm 2 h 1. In a similar fashion to

for 34.0% of the annual precipitation (1971–1990), while it

the results from the historical data, 83.3% of the months

only accounted for 20.0% of the annual precipitation in

had rainfall erosivity below 200.0 MJ mm hm 2 h 1 in the

normal and humid years. In moderate drought years, the pre-

RCP4.5 and RCP8.5 scenarios.

cipitation from July to August accounted for about 25.3% of the annual precipitation, while in normal and moderate

Spatial variation of rainfall erosivity and its response to

humid years, it reached more than 35.0% (1991–2010). The

precipitation

historical data show that the monthly distribution of rainfall erosivity during different hydrological years is essentially

In the 1980s, rainfall erosivity decreased in southern Gansu,

the same. In drought years, rainfall erosivity increased from

central Shaanxi, and northern Shanxi and increased slightly

Table 3

SPI values and hydrological years for the different data series

Data series

Moderate drought years

Normal years

Moderate humid years

1971–1990s

1983 (SPI36 ¼ 1.06)

1976 (SPI36 ¼ 0.50)

1978 (SPI36 ¼ 1.72)

1991–2010s

2002 (SPI36 ¼ 1.01)

2008 (SPI36 ¼ 0.08)

2005 (SPI36 ¼ 1.15)

2020–2100s (RCP4.5)

2045 (SPI60 ¼ 1.13)

2052 (SPI60 ¼ 0.38)

2057 (SPI60 ¼ 1.12)

2020–2100s (RCP8.5)

2048 (SPI60 ¼ 1.09)

2062 (SPI60 ¼ 0.44)

2071 (SPI60 ¼ 1.02)

Y. Cui et al.

1059

Figure 7

Spatiotemporal variation of rainfall erosivity

Hydrology Research

51.5

2020

in Ningxia and some parts of Gansu and southern Shanxi.

and RCP8.5 scenarios, the rainfall erosivity in 2020–2100

After 2000, it increased in central and southern Shaanxi

decreased by 28.9 and 19.8%, respectively, compared with

and central Shanxi but decreased in Inner Mongolia. How-

that in 1991–2010, respectively, but the range of variations

ever, on the whole, the spatial distribution of rainfall

is larger (Table 2). The maximum annual mean rainfall ero-

erosivity in the Loess Plateau shows a decreasing trend

sivity reaches 3,494.06 MJ mm hm 2 h 1 a 1, while the

from the southeast to the northwest. Most parts of Shanxi

minimum is only 135.53 MJ mm hm 2 h 1 a 1.

Province, and the central and southern parts of Shaanxi

To analyze the causes of increased precipitation and

remained the areas most threatened by rainfall erosion in

reduced rainfall erosivity in future climate modes and to

the Loess Plateau, and thus requires ecological management

further study the quantitative relationship between rainfall

interventions to prevent soil erosion.

erosivity and precipitation, the present study compared

In the future climate scenarios, Shanxi continues to be

annual rainfall erosivity and annual precipitation. (Missing

the province with the greatest rainfall erosivity, while the

data were recorded as 0, and the historical data comprised

area with the lowest rainfall erosivity changes Ningxia (as

39 × 40 ¼ 1,560 pairs; 26 sample spots in the Loess Plateau

determined by historical data) to Inner Mongolia. This rep-

were selected for the predicted data, and 26 × 81 ¼ 2,106

resents a relatively regular distribution of highs in the

pairs of data were generated in the RCP4.5 and RCP8.5 scen-

southeast and lows in the northwest. Areas with increased

arios, respectively.) The ﬁtting results are shown in Figure 8.

rainfall erosion after 2060 are also concentrated in the

In the historical data, the annual precipitation was

south (RCP4.5) and southeast (RCP8.5). In the RCP4.5

mainly concentrated between 200 and 600 mm. Initially,

Y. Cui et al.

1060

Figure 8

Spatiotemporal variation of rainfall erosivity

Hydrology Research

51.5

2020

rainfall erosivity increased at a small rate. The rate of

MJ mm hm 2 h 1. In contrast to the RCP4.5 scenario, the

increase then gradually increased before gradually decreas-

annual rainfall erosivity in the RCP8.5 scenario was

ing. The annual rainfall erosivity was generally below

mainly concentrated below 3,000 MJ mm hm 2 h 1. The

h . The logistic function provided a

logistic function ﬁts the RCP4.5 data well (R 2 ¼ 0.49, P <

good ﬁtting effect with the historical data (R 2 ¼ 0.77, P <

0.05), but the function does not converge with the RCP8.5

0.01). The formula is as follows:

data. The Allometric1 function in ORIGIN 2020 was used

2,500 MJ mm hm

to do the ﬁtting, and a good ﬁtting effect was also obtained Ry ¼ 7112:09

7037:33

(4)

1 þ (Py =880:61)2:25

for the RCP8.5 data (R 2 ¼ 0.55 (P < 0.05), Figure 8(c)). For the same rainfall amount, when P < 442.51 mm, the rainfall erosivity of the RCP4.5 scenario is greater than that of

where Ry is the annual rainfall erosivity (MJ·mm·hm ·h ),

RCP8.5, and vice versa. Under the future climate modes,

and Py is the annual precipitation (mm). When the annual

the annual rainfall erosivity gradually increases and

precipitation was 200–600 mm, the annual rainfall erosivity

becomes more discrete with an increase in precipitation,

was about 316.68–2,163.19 MJ mm hm 2 h 1.

which indicates that the rainfall data generated by the

When compared with the historical data, it was evident

HadGEM2-ES model are complex. However, as Figure 8

that the annual precipitation in the RCP4.5 scenario was

shows, when the annual precipitation is less than

mainly concentrated in the 300–900 mm range, and the

1,000.0 mm, the 95% conﬁdence band is narrow, which

annual rainfall erosivity was generally below 2,500

indicates that two functions are applicable for the prediction

1061

Y. Cui et al.

Spatiotemporal variation of rainfall erosivity

Hydrology Research

51.5

2020

of rainfall erosivity using data from HadGEM2-ES. It is

and 41771305) and the National Key Research and

clear, therefore, that the detailed changes of future rainfall

Development Program of China (2017YFC0504702). We

erosivity require further quantitative research.

thank Paul Seward, PhD, from Liwen Bianji, Edanz Group China (www.liwenbianji.cn/ac), for editing the English text of a draft of this manuscript.

CONCLUSIONS Rainfall data from 1971 to 2010 for 39 typical meteorologi-

DATA AVAILABILITY STATEMENT

cal stations in the Loess Plateau were collected to calculate the spatiotemporal variation of rainfall erosivity. Addition-

All relevant data are included in the paper or its Supplemen-

ally, future trends in rainfall erosivity were predicted using

tary Information.

the RCP4.5 and RCP8.5 scenarios. The main conclusions are as follows:

REFERENCES

1. From 1971 to 2010, the mean rainfall erosivity on the Loess Plateau was 1,239.64 MJ mm hm 2 h 1. Overall, annual rainfall erosivity exhibited a slightly decreasing trend. On a monthly basis, rainfall erosivity appeared as a unimodal distribution throughout the year and was mainly concentrated in July to August. There were some differences, however, in the distribution during different hydrological years. 2. The change in annual mean rainfall erosivity was different for different time series, while the spatial distribution of rainfall erosivity showed an overall decrease from the southeast to the northwest. The province with the highest rainfall erosivity was Shanxi, while Ningxia had the lowest. Southern Shanxi, and central and southern Shaanxi were areas susceptible to rainfall erosion in the Loess Plateau. 3. Rainfall erosivity obtained from future climate scenarios is lower than that obtained from historical data. While the precipitation amount increased in future scenarios, the relationship between precipitation and rainfall erosivity was relatively discrete. Future rainfall erosivity presents a relatively regular spatial distribution pattern: high in the southeast and low in the northwest. Areas with intense rainfall erosion on the Loess Plateau were concentrated in the south (RCP4.5) and southeast (RCP8.5).

ACKNOWLEDGEMENTS This research was ﬁnancially supported by the National Natural Science Foundation of China (Grants 41530858

Abd Elbasit, M. A. M., Huang, J., Ojha, C., Yasuda, H. & Adam, E. O.  Spatiotemporal changes of rainfall erosivity in Loess Plateau, China. ISRN Soil Science 2013, 1–8. http://dx. doi.org/10.1155/2013/256352. Ahmad, I., Tang, D., Wang, T., Wang, M. & Wagan, B.  Precipitation trends over time using Mann-Kendall and Spearman’s Rho tests in Swat River Basin, Pakistan. Advances in Meteorology 2015, 1–15. http://dx.doi.org/10. 1155/2015/431860. Almagro, A., Oliveira, P. T. S., Nearing, M. A. & Hagemann, S.  Projected climate change impacts in rainfall erosivity over Brazil. Scientiﬁc Reports 7, 8130. https://doi.org/10. 1038/s41598-017-08298-y. Anees, M. T., Abdullah, K., Nawawi, M. N. M., Norulaini, N. A. N., Piah, A. R. M., Fatehah, O., Syakir, M. I., Zakaria, N. A. & Omar, A. K. M.  Development of daily rainfall erosivity model for Kelantan State, Peninsular Malaysia. Hydrology Research 5 (49), 1434–1451. https://doi.org/10.2166/nh.2017.020. Angulo-Martínez, M. & Beguería, S.  Estimating rainfall erosivity from daily precipitation records: a comparison among methods using data from the Ebro Basin (NE Spain). Journal of Hydrology 379 (1–2), 111–121. https://doi.org/10. 1016/j.jhydrol.2009.09.051. Chou, S. C., Lyra, A., Mourão, C., Dereczynski, C., Pilotto, I., Gomes, J., Bustamante, J., Tavares, P., Silva, A., Rodrigues, D., Campos, D., Chagas, D., Sueiro, G., Siqueira, G. & Marengo, J.  Assessment of climate change over South America under RCP 4.5 and 8.5 downscaling scenarios. American Journal of Climate Change 03 (05), 512–527. http://dx.doi.org/10.4236/ajcc.2014.35043. Collins, W. J., Bellouin, N., Doutriaux-Boucher, M., Gedney, N., Halloran, P., Hinton, T., Hughes, J., Jones, C. D., Joshi, M., Liddicoat, S., Martin, G., Connor, F., Rae, J., Senior, C., Sitch, S., Totterdell, I., Wiltshire, A. & Woodward, S.  Development and evaluation of an earth-system model – HadGEM2. Geoscientiﬁc Model Development 4 (4), 1051–1075. https://doi.org/10.5194/gmd-4-1051-2011.

1062

Y. Cui et al.

Spatiotemporal variation of rainfall erosivity

Da Silva, A. M.  Rainfall erosivity map for Brazil. Catena 57 (3), 251–259. https://doi.org/10.1016/j.catena.2003.11.006. Djebou, D. C. S., Singh, V. P. & Frauenfeld, O. W.  Analysis of watershed topography effects on summer precipitation variability in the Southwestern United States. Journal of Hydrology 511, 838–849. https://doi.org/10.1016/j.jhydrol. 2014.02.045. Fu, B. J., Zhao, W. W., Chen, L. D., Zhang, Q. J., Lü, Y. H., Gulinck, H. & Poesen, J.  Assessment of soil erosion at large watershed scale using RUSLE and GIS: a case study in the Loess Plateau of China. Land Degradation & Development 16 (1), 73–85. https://doi.org/10.1002/ldr.646. Hamed, K. H. & Ramachandra Rao, A.  A modiﬁed MannKendall trend test for autocorrelated data. Journal of Hydrology 201 (4), 182–196. https://doi.org/10.1016/S00221694(97)00125-X. Lal, R.  Soil erosion research methods. Geographical Review 4. https://doi.org/10.2307/215130. Liu, L. & Liu, X. H.  Sensitivity analysis of soil erosion in the Northern Loess Plateau. Procedia Environmental Sciences 2, 134–148. https://doi.org/10.1016/j.proenv.2010.10.017. McKee, T. B., Doesken, N. J. & Kleist, J.  The relationship of drought frequency and duration to time scales. In: Paper Presented at the 8th Conference on Applied Climatology, Anaheim, CA. Meusburger, K., Steel, A., Panagos, P., Montanarella, L. & Alewell, C.  Spatial and temporal variability of rainfall erosivity factor for Switzerland. Hydrology and Earth System Sciences 16 (1), 167–177. https://doi.org/10.5194/hess-16167-2012. Nearing, M. A., Pruski, F. F. & O’Neal, M. R.  Expected climate change impacts on soil erosion rates: a review. Journal of Soil and Water Conservation 59 (1), 43–50. Panagos, P., Ballabio, C., Borrelli, P., Meusburger, K., Klik, A., Rousseva, S., Tadić, M. P., Michaelides, S., Hrabalíková, M., Olsen, P., Aalto, J., Lakatos, M., Rymszewicz, A., Dumitrescu, A., Beguería, S. & Alewell, C.  Rainfall erosivity in Europe. Science of the Total Environment 511, 801–814. https://doi.org/10.1016/j.scitotenv.2015.01.008. Panagos, P., Borrelli, P., Meusburger, K., Yu, B., Klik, A., Jae, L. K., Yang, J. E., Ni, J., Miao, C., Chattopadhyay, N., Sadeghi, S. H., Hazbavi, Z., Zabihi, M., Larionov, G. A., Krasnov, S. F., Gorobets, A. V., Levi, Y., Erpul, G., Birkel, C., Hoyos, N., Naipal, V., Oliveira, P., Bonilla, C. A., Meddi, M., Nel, W., Al, D. H., Boni, M., Diodato, N., Van Oost, K., Nearing, M. & Ballabio, C.  Global rainfall erosivity assessment based on high-temporal resolution rainfall records. Scientiﬁc Reports 7 (1), 4175. https://doi.org/10.1038/s41598-017-04282-8. Pimentel, D.  Soil erosion: a food and environmental threat. Environment Development and Sustainability 8 (1), 119–137. https://doi.org/10.1007/s10668-005-1262-8. Pimentel, D., Harvey, C., Resosudarmo, P., Sinclair, K., Kurz, D., McNair, M., Crist, S., Shpritz, L., Fitton, L., Saffouri, R. &

Hydrology Research

51.5

2020

Blair, R.  Environmental and economic costs of soil erosion and conservation benefits. Science 267 (5201), 1117–1123. Available from: https://www.jstor.org/stable/ 2886079. Renard, K. G. & Freimund, J. R.  Using monthly precipitation data to estimate the R-factor in the revised USLE. Journal of Hydrology 157 (1–4), 287–306. https://doi.org/10.1016/00221694(94)90110-4. Renard, K. G., Foster, G. R., Weesies, G., McCool, D. & Yoder, D.  Predicting Soil Erosion by Water: A Guide to Conservation Planning with the Revised Universal Soil Loss Equation (RUSLE). US Government Printing Office, Washington, DC. Sun, W., Shao, Q., Liu, J. & Zhai, J.  Assessing the effects of land use and topography on soil erosion on the Loess Plateau in China. Catena 121, 151–163. https://doi.org/10.1016/j. catena.2014.05.009. Vrieling, A., Sterk, G. & de Jong, S. M.  Satellite-based estimation of rainfall erosivity for Africa. Journal of Hydrology 395 (3–4), 235–241. https://doi.org/10.1016/j. jhydrol.2010.10.035. Wischmeier, W. H. & Smith, D. D.  Predicting Rainfall Erosion Losses – A Guide to Conservation Planning. USDA, Science and Education Administration, Hyattsville, MD, USA. Wu, L., Liu, X. & Ma, X.  Spatiotemporal distribution of rainfall erosivity in the Yanhe River Watershed of Hilly and Gully Region, Chinese Loess Plateau. Environmental Earth Sciences 75 (4). https://doi.org/10.1007/s12665-015-5136-6. Xie, Y., Yin, S., Liu, B., Nearing, M. A. & Zhao, Y.  Models for estimating daily rainfall erosivity in China. Journal of Hydrology 535, 547–558. https://doi.org/10.1016/j.jhydrol. 2016.02.020. Xin, Z., Yu, X., Li, Q. & Lu, X. X.  Spatiotemporal variation in rainfall erosivity on the Chinese Loess Plateau during the period 1956–2008. Regional Environmental Change 11 (1), 149–159. https://doi.org/10.1007/s10113-010-0127-3. Xu, J.  Precipitation–vegetation coupling and its influence on erosion on the Loess Plateau, China. Catena 64 (1), 103–116. https://doi.org/10.1016/j.catena.2005.07.004. Yan, D., Werners, S. E., Ludwig, F. & Huang, H. Q.  Hydrological response to climate change: the Pearl River, China under different RCP scenarios. Journal of Hydrology: Regional Studies 4, 228–245. https://doi.org/10.1016/j.ejrh. 2015.06.006. Yang, F. & Lu, C.  Spatiotemporal variation and trends in rainfall erosivity in China’s dryland region during 1961–2012. Catena 133, 362–372. https://doi.org/10.1016/j.catena.2015. 06.005. Zhang, W. B., Xie, Y. & Liu, B. Y.  Rainfall erosivity estimation using daily rainfall amounts. Scientia Geographica Sinica 22 (06), 705–711 (in Chinese with English abstract). http://dx.doi.org/10.3969/j.issn.1000-0690.2002.06.012.

First received 24 March 2020; accepted in revised form 3 July 2020. Available online 31 July 2020

1063

51.5

2020

Potential impact of water transfer policy implementation on lake eutrophication on the Shandong Peninsula: a difference-in-differences approach Jia He, Jiping Yao, Aihua Li, Zhongxin Tan, Gang Xie, Huijian Shi, Xuan Zhang, Wenchao Sun and Peng Du

ABSTRACT Traditional research on lake eutrophication has failed to consider the effect of the South-to-North Water Transfer Project (SNWTP) policy; thus, the difference-in-differences (DID) model, which is usually applied to economic factors, was innovatively introduced to evaluate the effect of such policies on lake eutrophication. Nansi Lake and Dongping Lake in the Shandong Peninsula were selected as the experimental group, and Daming Lake and Mata Lake were selected as the control group. The eutrophication indices of the experimental group and the control group were calculated by the measured chlorophyll-a, total phosphorus, total nitrogen, water transparency and chemical oxygen demand data and used as the explanatory variables of the DID model. Nine environmental and socio-economic factors, such as dissolved oxygen and rural population, were selected as the control variables of the DID model to analyze the impact of the SNWTP policy on lake eutrophication. A joint consideration of environmental and socio-economic factors showed that the eutrophication degree of the experimental lakes deteriorated by 7.10% compared with the control under the

Jia He Jiping Yao Zhongxin Tan (corresponding author) Xuan Zhang Wenchao Sun Peng Du Beijing Key Laboratory of Urban Hydrological Cycle and Sponge City Technology, College of Water Sciences, Beijing Normal University, Beijing 100875, China E-mail: tanzhongxin@bnu.edu.cn Aihua Li Gang Xie Huijian Shi Shandong Academy for Environmental Planning, Jinan 250000, China

inﬂuence of the implemented policy. Dissolved oxygen is the main factor affecting the eutrophication of the Shandong Peninsula. This study veriﬁes that the DID model has the potential for use in quantitative analyses of the effect of the SNWTP policy on lake eutrophication. Key words

| difference-in-differences approach, lake eutrophication, parallel trends, policy impact, South-to-North Water Transfer Project

HIGHLIGHTS

• • •

Using DID model to estimate the impact of policy on lake eutrophication is reasonable. Eutrophication deteriorated by 7.10% under the effect of policy and control variables. DO is the key control variables inﬂuencing the eutrophication of the Shandong Peninsula.

INTRODUCTION China’s South-to-North Water Transfer Project (SNWTP) is

improving the ecological environment of the water demand

of great strategic signiﬁcance for alleviating water shortages,

area and promoting sustainable economic and social development. China has invested approximately $20 billion and

resettled >300,000 people to construct the pipeline, and the SNWTP has become the largest and most expensive interbasin water transfer mega-project in the world

1064

J. He et al.

Using DID model to identify the impact of SNWTP policy on lake eutrophication

Hydrology Research

51.5

2020

(Resources ; Li et al. ). The ﬁnal water volume trans-

nutrients has detrimental effects in the receiving water

ferred is expected to reach 44.8 billion m3/year by 2050 when

system, such as algal blooms and decreased dissolved

the

fully

oxygen (DO) (Barrow ; Zeng et al. ). Although a

implemented (Zhuang et al. ). The ﬁrst phase of the east-

water transfer project can improve the water quality, it

ern route of the SNWTP (SNWTP-ER) was successfully

might also increase algal growth, and nutrient-rich water

completed and has been operational since late 2015 (Li

pumped from upstream causes cyanobacteria blooms in

eastern,

central

and

western

et al. ). A total of more than 3 × 10

routes

are

m of water has

the receiving reservoir (Davies et al. ; Zhang et al.

been transferred from the lower Yangtze River in Yangzhou

). Previous studies on the SNWTP-ER have not deter-

city, Jiangsu Province, and crossed the Huai River to the

mined whether it could control algal blooms or contribute

water shortage areas in the Yellow River basin (Guo et al.

to eutrophication in the water shortage period (Wu et al.

a, b). When completed, the SNWTP-ER will consist

). This research generally examines the inﬂuence of

of 1,156 km of canals and 54 pumping stations designed to

the SNWTP on environmental factors (total nitrogen and

lift water up to 65 m over the Yellow River (Wang et al.

total phosphorus) and hydrological factors (water quantity

). The pumped water will be diverted from the south

ﬂow rate) directly based on water quality status and monitor-

to the north, primarily through the existing Grand Canal,

ing data (Guo et al. b; Rogers et al. ; Xu et al. ).

and impound in a chain of natural lakes, as regulating

However, to the best of our knowledge, few studies have

reservoirs, namely Gao-Bao-Shaobo Lake (GBSL), Hongze

considered the impact of SNWTP policy implementation

Lake (HZL), Luoma Lake (LML), Nansi Lake (NSL) and

hidden behind the monitoring data. In other words, the

Dongping Lake (DPL) (Wang et al. ; Zhang ; Wu

possibility of eutrophication in the receiving reservoirs with-

et al. ; Guo et al. , a).

out

policy

implementation

should

comparatively

Any interbasin water transfer project causes complex

analyzed when considering the impact of the SNWTP

physical, chemical, hydrological and biological changes to

policy. If only the effects of potential changes of the receiv-

the receiving system (Zeng et al. ; Yang et al. ; Yao

ing reservoirs on eutrophication are compared in parallel

et al. a, b, ; Yinglan et al. a). The water and

before and after SNWTP policy implementation, then the

sediment quality of the SNWTP-ER is of particular concern

impact of policy implementation itself could be largely

due to its large contribution to the total volume of transferred

ignored and the socio-economic and environmental effects

water, and it is also one of the most potentially polluted routes

of SNWTP operation policy can be investigated (Rogers

given its proximity to urban and industrial activities (Wu et al.

et al. ). This study aims to establish an approach for

). With the vigorous development of the manufacturing

determining the inﬂuence of water transfer policy on receiv-

industry in the Yangtze River Delta and the Bohai Rim,

ing reservoir eutrophication. Nansi Lake and Dongping

large amounts of untreated industrial wastewater are directly

Lake were selected as the objects of the transferred water

discharged to the lakes and rivers along the eastern route, and

policy, and Daming Lake and Mata Lake (without water

agricultural production has led to augmented fertilizer appli-

transferred) were treated as controls. Nansi Lake and

cation, which has substantially increased the loadings of

Dongping Lake are the largest freshwater shallow lakes in

nutrients (including nitrogen and phosphorus) and organic

the Shandong Peninsula and play a vital role in absorbing

matter into river streams, thereby deteriorating water quality

pollutants and storing water (Zhuang et al. ). The differ-

(Li et al. ; Hou et al. ; Gao et al. ; Sheng &

ence-in-differences (DID) model, a common policy impact

Webber ; Kuo et al. ; Yinglan et al. b).

model used in economics, was introduced to establish a

Eutrophication has been recognized as the primary

comprehensive modeling approach for evaluating the

water quality issue for most of the lake ecosystems in the

impacts of SNWTP policy implementation on eutrophica-

world (Smith & Schindler ; Wang et al. ), and nitro-

tion in Nansi and Dongping Lakes and analyzing the key

gen and phosphorus are the primary reasons for algal

factors and processes driving the underlying mechanisms.

blooms caused by excess nutrients (Diersing ; Fang

The objectives of the study are to: (1) unravel the spatial

et al. ). Transferred water with a high content of

and temporal distribution of lake eutrophication in Nansi

J. He et al.

1065

Using DID model to identify the impact of SNWTP policy on lake eutrophication

Hydrology Research

51.5

2020

Lake, Dongping Lake, Daming Lake and Mata Lake; (2) use

from south to north and 5–25 km wide from east to west,

the SNWTP policy as a single factor or combine environ-

has an area of 1,266 km2 and a total storage volume of

mental and socio-economic factors as a composite factor

6.37 × 1010 m3 and is the ﬁrst largest freshwater lake in the

to establish the DID model for evaluating variation trends

Shandong Peninsula along the eastern route. Nansi Lake

in eutrophication on water transfer lakes; (3) apply parallel

consists of Nanyang Lake, Dushan Lake, Zhaoyang Lake

trend and robustness analyses to validate the feasibility of

and Weishan Lake without physical boundaries deﬁning

the DID model; and (4) identify the primary driving factors

each lake, and it was divided into the upper and lower sec-

that affect eutrophication and supply policy recommen-

tions by the Erji Dam Pumping Station Hinge Project in

dations to develop management strategies for water quality

1960. The upper lake lies on the northern side of the Erji

safety and pollution control.

Dam, and the lower lake is located on the southern side. Five state-controlled monitoring sections were located in Nansi Lake, namely the Qianbaikou (S1) and Nanyang

MATERIALS AND METHODS

(S2) sections in the upper lake, the Erjiba (S3) section near the Erji Dam and the Dajuan (S4) and Daodong (S5)

Study area

sections in the lower lake (Figure 1). Dongping Lake (35 300 –36 200 N, 116 000 –116 300 E) is

Nansi Lake (34 27 –35 20 N, 116 34 –117 21 E) is located

located in Dongping County, west of Shandong Peninsula,

in Weishan County, southwest Shandong Peninsula, China

China (Figure 1). The lake has an annual mean depth of

(Figure 1). The lake, which is approximately 126 km long

2–4 m and a total storage volume of 4 × 109 m3 and covers

Figure 1

Location of Nansi Lake, Dongping Lake, Mata Lake and Daming Lake and the SNWTP-ER route in the Shandong Peninsula.

1066

J. He et al.

Using DID model to identify the impact of SNWTP policy on lake eutrophication

Hydrology Research

51.5

2020

627 km2, and it is the second-largest freshwater lake in the

implementation effect of projects or public policies. The

Shandong Peninsula along the eastern route. The Hubei

notable superiority of DID estimation is derived from its

(S6), Huxin (S7) and Hunan (S8) state-controlled monitor-

simplicity as well as its potential to circumvent many of

ing sections are distributed in the north, middle and south

the endogenous problems that typically arise when making

of Dongping Lake, respectively. Relying on its special geo-

comparisons between heterogeneous individuals (Meyer

graphical position and function along the SNWDP-ER,

; Bertrand et al. ). In the assessment, the policy

Dongping Lake serves as the ﬁnal reservoir and an essential

experimental group and the control group generally do

ﬂood control project in the lower reaches of the Yellow

not have complete randomness in sample allocation. The

River (Yao et al. a, b).

experiment involving a nonrandom allocation policy exper-

Since the implementation of the SNWDP-ER policy in

imental group and control group is known as a natural trial,

2015, Nansi Lake and Dongping Lake have served as the

and its important feature is that systematic differences might

water-supplying lakes and impounded reservoirs; however,

occur between the experimental group and the control

maintaining good water quality while meeting water

group prior to the implementation of the experiment. If

demands remains a great challenge (Grant et al. ). In

the initial difference is ignored and only a horizontal com-

recent years, Nansi Lake is surrounded by the dense indus-

parison between the experimental group and the control

trial and population zones, which has resulted in a large

group is performed after implementing the experiment, the

amount of industrial and domestic sewage discharged into

estimated experimental effect is likely to be biased due to

Nansi Lake each year, thus exacerbating water pollution

the mixed effect of the initial difference. The DID model

and eutrophication (Li ; Yao et al. a, b).

was ﬁrst introduced in 1985 to solve this problem

Severe contamination and mining, chemical, electric

(Ashenfelter & Card ), and since then, increasing

power, manufacturing and many other industries within

attention has been focused on applying the model.

the main tributary river basin are the main factors responsible for the load of pollutants in Dongping Lake (Guo

Setting and veriﬁcation of the DID model in lake

et al. ; Wang et al. ; Xu et al. ).

eutrophication research

Daming Lake is located in Jinan City and Mata Lake is located in Zibo City, and both lakes were identiﬁed as key

Since the route of the SNWTP-ER was opened to water

nature reserves of Shandong Peninsula, China (Figure 1).

supply, its impact on lake eutrophication in Shandong Pro-

The Lixiating (S9) and Mata Lake (S10) state-controlled

vince might be due to the policy effect of the project and

monitoring sections are distributed in Daming Lake and

the time effect of the time trend changes in lake water eutro-

Mata Lake, respectively. The SNWTP-ER policy has not

phication. The question of how to analyze the policy effect

been implemented on Daming Lake and Mata Lake, and

for correctly evaluating the impact of the route of the

the spatial location and characteristics are adjacent to

SNWTP-ER on lake eutrophication in Shandong Province

Nansi Lake and Dongping Lake. Reducing the difference

is highly important. The DID model can effectively analyze

between the experimental groups and the control groups is

and objectively evaluate the policy effect. In summary, we

expected to improve the accuracy of the DID model. In

apply the DID model to research the change of lake eutro-

this study, considering the comprehensiveness, integrity

phication in Shandong Province before and after the route

and availability of the data, among the 13 lakes and reser-

of the SNWTP-ER.

voirs in the Shandong Peninsula, Daming Lake and Mata

Using trophic level indices (TLIs) as the explained variable, ‘treated’ ¼ 1 indicates the water inﬂow lake of the

Lake were treated as the control group.

section in the line of the SNWTP-ER and ‘treated’ ¼ 0 indiBasic principle of the DID model

cates that the lake where the section is located has no water diversion of the SNWTP-ER line. Time is used to

DID estimation has become an increasingly popular method

express the ‘time’. The value of the year when the

SNWTP-ER is open to water and the following years is 1;

estimating

the

econometric

evaluation

the

1067

J. He et al.

Using DID model to identify the impact of SNWTP policy on lake eutrophication

otherwise, the value is 0. We use ‘did’ to represent the implementation effect of the SNWTP-ER, i.e., the intersection of ‘treated’ and ‘time’. Additionally, xit is the control variable, which includes the time ﬁxed effect and regional ﬁxed effect, and θit represents the constant term and disturbance term. The basic measurement model is shown as follows: Yit ¼ β 0 þ β 1 didit þ β 2 treated þ β 3 time þ αxit þ θit

Table 1

Hydrology Research

51.5

2020

Standard values of the TLI

Grades

Meaning

TLI

COD

Level I

Oligotrophic

<30

0.2

0.02

Level II

Mesotrophic

30–50

0.5

0.1

Level III

Light eutropher

50–60

1.0

0.2

Level IV

Middle eutropher

60–70

1.5

0.3

Level V

Hyper eutropher

>70

2.0

0.4

(1)

where β 0 represents the common initial eutrophication

Lake, Dongping Lake, Daming Lake and Mata Lake were

mean value of all sections before the SNWTP-ER, β 1

calculated using chlorophyll-a (Chl-a), total phosphorus

represents the policy effect of the SNWTP-ER after control-

(TP), total nitrogen (TN), water transparency (SD) and

ling the initial eutrophication difference and common trend,

chemical oxygen demand (CODMN), which were computed

β 2 represents the difference in lake initial eutrophication

from

between the experimental group and the control group

material, Table S1. Water samples from the monitoring

before and after the SNWTP-ER, β 3 represents the trend of

stations in Shandong Province were collected from January

eutrophication of the experimental group and the control

2009 to December 2017 and analyzed for Chl-a, TP, TN, SD

group before and after the SNWTP-ER and α is the coefﬁ-

and CODMN based on the methods outlined by the Chinese

cient of the control variable xit .

national water environmental protection standard in China

Parallel trend and robustness analyses were used to validate the DID models. The parallel trend assumption

Equations

(2)–(7)

presented

Supplementary

(GB11914-89, GB11893-89, GB11894-89 and SL88-1994) (Wei ; Huo et al. ).

indicates that the experimental group and the control group follow parallel paths over time. This assumption allows the DID to account for unobserved variables, which are assumed to remain ﬁxed over time (Dimick & Ryan ; Zhou et al. ). Replacement of the interpreted variable was selected as the robustness analysis. If the regression results are consistent with the original results and the core variables are signiﬁcant, the estimation can be considered robust.

TLI

m X

Wj × TLI (j)

(2)

j¼1

where Wj is the correlative weight for the TLI of j, TLI (j) is P the TLI of j, TLI ð Þ is the comprehensive eutrophication index and the Wj values of Chl-a, TP, TN and CODMN are 0.2663, 0.1879, 0.1790 and 0.1834, respectively. The units of Chl-a, TP, TN, SD and CODMN are mg/m3, mg/L, mg/L, m and mg/L, respectively.

Selection of interpreted and control variables for the DID model in lake eutrophication research

TLI (Chl-a) ¼ 10 × [2:5 þ 1:086 × ln (Chl-a)]

(3)

TLI (TP) ¼ 10 × [9:436 þ 1:624 × ln (TP)]

(4)

TLI (TN) ¼ 10 × [5:453 þ 1:694 × ln (TN)]

(5)

TLI (SD) ¼ 10 × [5:118 þ 1:940 × ln (SD)]

(6)

TLI (CODMN ) ¼ 10 × [0:109 þ 2:661 × ln (CODMN )]

(7)

The eutrophication index treated as an interpreted variable was estimated by standard values. Due to the varied topographies, environmental background, industrial layout and human activities, the assessment methods for lake eutrophication are diverse for different regions because the comprehensive TLI is widely applied to evaluate the conditions of water quality (Liu et al. ). The standard

Environmental factors and socio-economic factors were

values of the TLIs are given in Table 1. The TLIs of Nansi

selected as the control variables, and nine control variable

1068

J. He et al.

Using DID model to identify the impact of SNWTP policy on lake eutrophication

Hydrology Research

51.5

2020

factors in total are used in this paper. The experimental

surface water reservoir section in Shandong Province from

group included ﬁve sections of Nansi Lake, and three sec-

January 2009 to December 2017 and the Shandong Statisti-

tions of Dongping Lake and Daming Lake and Mata Lake

cal Yearbook from 2010 to 2018.

each contained a section as a control. A total of 1,080 water samples from each section were collected monthly from January 2009 to December 2017. Four environmental

RESULTS AND DISCUSSION

factors, including water temperature (Tw) and DO, were measured with a portable multiparameter water quality ana-

Spatial and temporal distribution of lake eutrophication

lyzer (YSI Professional Plus, Yellow Springs, Ohio, USA).

in the Shandong Peninsula

The N:P ratios were calculated by the TN and TP values. Hours of sunshine (HS) were collected from the ‘Statistical

To analyze the spatial and temporal variation trend of eutro-

Yearbook of Shandong Province’ from 2009 to 2017 (details

phication in Nansi Lake, Dongping Lake, Daming Lake and

are listed in Supplementary material, Table S2; Lu ).

Mata Lake affected by the implementation of the SNWTP-

Five socio-economic factors, namely gross domestic pro-

ER policy, the TLIs of each section in four lakes from

duct (GDP), gross output value of agriculture (GOVA), gross

2009 to 2017 were calculated using the CODMN, TP, TN,

output value of animal husbandry (GOVAH), rural popu-

Chl-a and SD (Figure 2). The spatial and temporal variation

lation (RP) and number of inbound tourists (NIT) received

trends of three state-controlled sections in Dongping Lake

by each city, were also collected from the ‘Statistical Year-

are generally unanimous, and they all indicated signiﬁcant

book of Shandong Province’ from 2009 to 2017 (Lu ).

improvements in 2010, from the middle eutropher to meso-

Detailed data are listed in Supplementary material,

trophic levels. However, after the ofﬁcial water transfer

Table S2. GDP refers to the ﬁnal result of production activi-

began in 2015, an obvious slight upward trend is suggested

ties of all permanent units in the region in a certain period,

throughout Dongping Lake as shown in the lower left of

which is the sum of the added value of each industry and

Figure 2. The water quality of Hubei and Hunan sections

reﬂects the relationship between the economic development

increases to light eutropher, and the center of the lake

level of the region and lake eutrophication. GOVA refers to

remains at the mesotrophic level. Certain differences

the total scale and total results of agricultural production in

occur in the spatial and temporal variation trends of ﬁve

the region in a certain period and reﬂects the relationship

state-controlled sections in Nansi Lake, although signiﬁcant

between agricultural nonpoint source pollution and lake

improvements were not observed after the implementation

eutrophication. GOVAH refers to the total scale and results

of the water transfer policy in 2015. The eutrophication

of animal husbandry production in a certain period in the

degree was more severe in the northern portion than the

region and reﬂects the relationship between livestock

southern portion of Nansi Lake. An improvement of the

waste and lake eutrophication. RP refers to the total popu-

eutrophication degree at the Qianbaikou and Nanyang

lation except for the total permanent population living in

sections occurred in 2014, although it subsequently deterio-

the urban area and reﬂects the relationship between

rated to middle eutropher at the Nanyang section. The

undischarged domestic garbage and lake nutrition. NIT

analysis of the TLI values did not provide obvious results

refers to the number of inbound international tourists and

for the effect of policy implementation in the Erjiba,

domestic tourists and reﬂects the relationship between

Dajuan and Daodong sections based on the maintenance

household garbage and lake eutrophication.

of the mesotrophic level from 2013.

Data sources of the DID model in the study area

Lake decreased from hypereutropher to mesotrophic in a

The variable data of the DID model applied to lake eutrophi-

in 2015. The lowest TLI values of Daming Lake appeared in

cation research for each monitoring section in the study area

2012 and 2013, and they slightly increased to the light eutro-

are taken from the water quality monitoring data of the

pher level from 2014.

In the control lakes, the eutrophication level of Mata year-by-year manner and signiﬁcant improvements occurred

J. He et al.

1069

Figure 2

Using DID model to identify the impact of SNWTP policy on lake eutrophication

Hydrology Research

51.5

2020

Spatial and temporal variation of TLIs in 10 state-controlled sections from 2009 to 2017.

Quantitative and qualitative analysis of the impact of

the P-value passing the 5% level test, and the ﬁtting degree

SNWTP-ER policy on lake eutrophication

of the relationship has slightly improved. The β1 value of

An analysis of the temporal and spatial distribution shows

of the SNWTP-ER policy, the eutrophication degree of the

that the eutrophication degree could be affected by the

experimental lakes deteriorates by 6.20% compared with

SNWTP-ER policy. However, it is impossible to quantitat-

that of the control lakes.

Model 2 indicates that when considering only the inﬂuence

ively and qualitatively analyze the impact of the policy.

Models 3 and 4 added the environmental control vari-

Therefore, the DID model was introduced to estimate the

ables shown in Table 2, such as Tw, DO, HS and N:P

potential inﬂuence of the SNWTP-ER policy on the lake

ratio. There are no ﬁxed time and regional effects in

eutrophication degree in the Shandong Peninsula.

Model 3, and the signiﬁcance of the P-value is only 10%.

The results are shown in Table 2 and the Model 1 and

After ﬁxing the time and regional effects in Model 4, the

Model 2 regressions considered only the effect of the

relationship between the TLIs and policy effects and

SNWTP-ER policy without control variables. Model 1

environmental factors is signiﬁcant, with R 2 ¼ 0.70, and

does not include ﬁxed time and regional effects, and the

the policy effect is signiﬁcantly improved at the 1% level.

results showed that the relationship is rather poor, with

The quantitative and qualitative analysis results of Model 4

R 2 ¼ 0.30 and a P-value at only 10% signiﬁcance. Model 2

show that the β1 value is positive at 5.89, which indicates

includes ﬁxed time and regional effects, and the SNWTP-

that the eutrophication degree of the experimental lakes is

ER policy effect of lake eutrophication is signiﬁcant, with

signiﬁcantly increased by 5.89% compared with the control

1070 J. He et al.

DID regression estimations for TLIs from different effect factors

Type

Policy effect

Regression

Model 1 Model 2 Model 3 Model 4

Model 5 Model 6

Model 7

Model 8

Model 10

Model 11

β1

6.20*

6.20**

4.33*

5.89***

9.76***

10.75***

5.48**

10.46*** 7.68**

7.02***

7.10***

Time effect

Control

Regional effect No

Control

0.51

0.64

0.70

0.60

0.58

0.48

0.82

0.80

0.33

Environmental effect

Socio-economic effect

Comprehensive effect

Model 9

Coefﬁcient of control variable Tw

1.76***

1.67***

1.56***

2.50*** NA

1.37**

2.02***

2.10***

0.01**

0.01

N:P ratio

0.08*** NA

0.11***

0.10***

GDP

0.01

2.37 × 10

GOVA

6.13 × 10 06**

5.62 × 10 06***

6.46 × 10 06***

6.66 × 10 06***

6.53 × 10 06***

GOVAH

5.66 × 10 06*

5.68 × 10 06**

3.10 × 10 06

0.02

0.03**

0.05***

NIT

0.51

0.57*

0.31

Using DID model to identify the impact of SNWTP policy on lake eutrophication

Table 2

Note: *, ** and *** indicate signiﬁcant at the level of 10, 5 and 1%, respectively.

Hydrology Research

| 51.5

| 2020

1071

J. He et al.

Using DID model to identify the impact of SNWTP policy on lake eutrophication

Hydrology Research

51.5

2020

lakes after the SNWTP-ER policy was implemented. The

In summary, although Model 2, Model 4 and Model 7

value of β1 in Model 4 is lower than that in Model 2,

evaluate the impact of policy from different perspectives,

which suggests that the estimated eutrophication degree

Model 11 represents a more reasonable model for

decreased by 0.31% under the combined inﬂuence of

estimating the inﬂuence of the water transfer policy on

policy and environmental effect factors compared with

lake eutrophication.

that when only the policy effect is considered. Tw and DO

As shown in Table 2, the TLIs of Model 11 are signiﬁ-

are the principal inﬂuence factors with regression coefﬁ-

cantly negatively correlated with the DO and N:P ratio

cients of 1.76 and 2.50, respectively.

and positively correlated with the Tw, GOVA and RP. The

Models 5–7 add GDP, GOVA, GOVAH, RP and NIT as

concentration of DO in the lake water could reﬂect the bio-

socio-economic control variables. Without the ﬁxed time and

logical processes of algae through photosynthesis and

regional effects, the signiﬁcance of Model 5 is at the 1% level

respiration (Zeng et al. ). The eutrophication degree

and the relationship coefﬁcient R is 0.60. However, Model 7

usually decreases with an increase in the concentration of

includes ﬁxed time and regional effects, and the signiﬁcance

DO. According to the law of Redﬁeld, the N:P ratio of

of the policy effect decreased to the 5% level, with R below

algal cells is in the range of 16:1, and the eutrophication

0.5. This result could occur because the selected control vari-

condition of water indicates that nitrogen and phosphorus

ables are not comprehensive without the ﬁxed time and

reach the appropriate proportion, which leads to the out-

regional effects. The value of β1 in Model 7 is similar to

break of lake eutrophication (Geider & La Roche ).

that of Model 4, which indicates that the degree to which

The low concentration of phosphorus in the lakes of

eutrophication is affected by socio-economic factors and

Shandong Province is the key to inhibiting water eutrophica-

environmental factors does not signiﬁcantly differ. In con-

tion, and the TLIs might decrease with an increase in the

trast, as shown in Table 2, the regression coefﬁcients of the

N:P ratio. Tw is another essential environmental inﬂuence

socio-economic factors are notably low, suggesting that the

factor that affects the eutrophication degree. For every

impact of socio-economic variables on eutrophication might

1 C increase in annual temperature, the annual algal bio-

be lower than that of environmental variables.

mass is expected to increase by 0.145 times (Ye et al. ).

Models 8–11 combine the environment effect factors

In addition to the direct impact of environmental factors

and socio-economic factors as comprehensive control vari-

on lake eutrophication, the indirect impact of social and

ables. Model 11 includes the ﬁxed time and regional

economic factors on lake water quality is also increasingly

effects, and the SNWTP-ER policy effect of lake eutrophica-

signiﬁcant. GOVA can reﬂect the relationship between agri-

tion is signiﬁcant with a P-value reaching the 1% level, and

cultural nonpoint source pollution and lake nutrients. An

the ﬁtting degree of the relationship is obviously improved

increase in the GOVA number increases the risk of nitrogen

with R 2 ¼ 0.80. The best estimation accuracy among the

and phosphorous fertilizers and pesticides entering the lake

11 models is that of Model 11, which includes ﬁve compre-

along with surface runoff. An increase of RP directly leads to

hensive factors. Compared with the accuracy of the other 10

domestic garbage pollution in the water body, resulting in

models developed in this paper, the estimation accuracy

eutrophication of the lake.

from Model 11 is reasonable because this model considered the combined effect rather than only the environment effect

Parallel trend and robustness analyses of the DID model

or socio-economic inﬂuence. The β1 value of 7.10 indicates that under the combined inﬂuence of the SNWTP-ER

Parallel trend analysis is a necessary prerequisite for policy

policy with all of the effect factors, the eutrophication

effect evaluations using the DID model. If the parallel

degree of the experimental lakes deteriorates by 7.10% com-

trends between the experimental group and the control

pared with the control lakes, which is 0.90% higher than

group are not signiﬁcantly different, the probability of bias

that when considering only the inﬂuence of policy. Tw and

in the empirical results might be reduced. In other words,

DO are the main inﬂuence factors, with regression coefﬁ-

in this study, prior to the implementation of the SNWTP

cients of 1.56 and 2.10, respectively.

policy, the variation trend of the TLI values of the

J. He et al.

1072

Using DID model to identify the impact of SNWTP policy on lake eutrophication

experimental group (Nansi Lake and Dongping Lake) are the same as those of the control group (Mata Lake and Daming Lake) or are different but have not changed signiﬁcantly over time. The above two cases could indicate that the eutrophication degrees of the experimental group and the control group before the implementation of the policy have the same trend.

Table 3

Hydrology Research

2020

Regression

β1

Policy effect

Model 12 Model 13

0.20*** 0.20***

Environmental effect

Model 14 Model 15

0.18*** 0.19***

Socio-economic effect

Model 16 Model 17 Model 18

0.17** 0.19*** 0.20**

Comprehensive effect

Model 19 Model 20 Model 21

0.20*** 0.18*** 0.17***

allel trends between the experimental group and the control group before 2015 (left of the red line). After 2015, a signiﬁ-

Type

tion trends for all outcomes as shown in Figure 3(a). The

cant difference appears between the two groups, with a

51.5

Robustness analysis of DID estimations for TP as interpreted variables

Statistical evidence indicated comparable preintervenvariation trend of the mean TLI values did not reject the par-

Note: ** and *** indicate signiﬁcant at the level of 5 and 1%, respectively.

decreasing trend of eutrophication degree in the control group and an increasing trend in the experimental group,

Comparison of the assessment ability of the DID model

which meets the preconditions of evaluating the policy

with unconsidered policy effect models on lake

impact using the DID model. Moreover, Figure 3(b) illus-

eutrophication

trates that the coefficient fluctuates approximately 0 before the policy is implemented and is significantly positive 1

To further assess the evaluation ability of the DID models

year after the policy is implemented. This result indicates

for lake eutrophication established in this study, a Pearson

that the parallel trends assumption outlined above can be

correlation analysis and multiple linear regression were

evaluated using a regression model in this study and that

applied to evaluate eutrophication without considering the

the policy effect appears after the SNWTP policy is

policy effect for comparative analysis. First, the Pearson cor-

implemented, with apparent changes over time.

relation analysis was established between the nine control

To verify the robustness of the DID model, the inter-

variables and TLI values. The results listed in Table 4

preted variables could be redeﬁned for the DID regression.

show that the TLIs are signiﬁcantly correlated with DO,

TP was replaced in the interpreted variables, and the

HS, GOVA, GOVAH and NIT, with the strongest corre-

policy cut-off point was still 2015. The robustness analysis

lation between GOVAH and NIT, which is not consistent

results listed in Table 3 suggest that the core variables

with the conclusion of the DID model. Moreover, HS,

passed the signiﬁcance test with a P-value of 5 or 1%,

GOVAH, GOVA and NIT were negatively correlated with

which is consistent with the regression results in Table 2.

the TLIs, indicating that when the Pearson’s analysis did

Figure 3

Parallel trends of the eutrophication degree between the experimental group and the control group. Please refer to the online version of this paper to see this ﬁgure in color: http://dx.doi.org/10.2166/nh.2020.047.

J. He et al.

1073

Table 4

Using DID model to identify the impact of SNWTP policy on lake eutrophication

Hydrology Research

51.5

2020

Pearson correlations among Tw, DO, HS, N:P ratio, GDP, GOVA, GOVAH, RP, NIT and TLI Tw

N:P ratio

GDP

GOVA

GOVAH

NIT

TLI

1.000

0.005

1.000

0.024

0.030

1.000

N:P ratio

0.001***

0.196

0.005

1.000

GDP

0.212***

0.011

0.008

0.305***

1.000

GOVA

0.276***

0.011

0.017

0.041***

0.135***

1.000

GOVAH

0.197

0.000***

0.023***

0.056***

0.093***

0.846***

0.013

0.075**

0.093**

0.220

0.198***

0.201***

0.183

1.000

NIT

0.029

0.071***

0.049

0.000***

0.012*

0.134**

0.318***

0.009

1.000

TLI

0.012

0.181***

0.067**

0.007

0.003

0.220***

0.375***

0.017

0.409***

1.000

Note: *, ** and *** indicate signiﬁcant at the level of 10, 5 and 1%, respectively.

not consider the inﬂuence of policy, a certain amount of

Policy recommendations

deviation occurred in the analysis results. A multiple linear regression is established using ﬁve control variables selected

Based on the above quantitative and qualitative analysis

in Model 11 as shown below.

results of the impact of the SNWTP-ER policy on lake eutrophication, the main inﬂuencing factors are selected from Model 11, and we propose the following three suggestions.

Y ¼ 63:0 þ 1:41 × Tw 2:89 × DO þ 0:0049 × N:P

First, while developing the local agricultural economy,

0:000004 × GOVA þ 0:0058 RP

(8)

additional attention should be focused on agricultural nonpoint

source

pollution

discharge

for

control

and

governance (Wang et al. , a, b). In agricultural The results showed that the relationship between the

production, especially in crop production, fertilizers and

TLIs and ﬁve control variables was signiﬁcant, with P ¼

pesticides are heavily used every year, which might cause

0.00 < 0.05; however, the correlation coefﬁcient R ¼ 0.51 2

a large amount of nitrogen and phosphorus to enter the

was not better than that of the DID Model 11, with R ¼

lake via the circulation process and surface runoff. In

0.80. In addition, the N:P ratio and RP did not pass the

other words, an increase in agricultural output value might

t-test of signiﬁcance at P-values greater than 0.05, although

enhance the discharge of agricultural pollutants, meaning

DO is still treated as a dominant factor affecting water qual-

that the eutrophication degree of lakes might continuously

ity, which is consistent with the DID model and the Pearson

deteriorate. One suggestion might be to adjust the planting

correlation analysis. Moreover, the DID model could be

varieties of crops and reasonably control the amount of pes-

used as an alternative method to evaluate and quantify the

ticides and fertilizers used in the lake area. Another

impact of SNWTP policy on lake water quality in Shandong

suggestion is to establish a wetland isolation zone surround-

Province. The advantage of the DID model is that it could

ing the lake to decrease the risk of nitrogen and phosphorus

reveal the extent of policy impacts hidden in the experimen-

entering the lake by way of nutrient substances consumed by

tal monitoring data. In addition, the DID model offers a

the growth of aquatic plants. Additionally, the development

diversiﬁed perspective that includes policy, environment

of the economy is expected to result in advanced production

and socio-economic diversiﬁcation to analyze the impact

technology and management experience that can alleviate

of the SNWTP on lake water quality, thus making the

the pollution caused by agricultural production to a certain

entire assessment and analysis more comprehensive than

extent and improve the adverse impact of agricultural econ-

conventional models.

omic development on lake eutrophication.

1074

J. He et al.

Using DID model to identify the impact of SNWTP policy on lake eutrophication

Hydrology Research

51.5

2020

Second, the shortage of rural sewage treatment must be

correlation analysis and multiple linear regression result

addressed. Production and domestic garbage generated by

showed that the DID models are more suitable for analyzing

an increasing population are expected to have a serious

lake eutrophication in Shandong Province than convention-

impact on the aquatic ecology of the lake if it is not handled

al models. The DID models identiﬁed the environmental

properly and directly enters the water body. It is suggested

factors that should receive additional attention to prevent

that human activities should be separated from ecological

eutrophication of Nansi Lake and Dongping Lake. It is

nature reserves to a certain extent. The capacity of domestic

suggested that attention should be focused on controlling

sewage treatment should be enhanced, and upgrades and

and treating nonpoint agricultural source pollution emis-

reconstruction should be performed in sewage treatment

sions to compensate for the shortage of rural sewage

plants (Scheren et al. ). In areas where ﬂush toilets

treatment, increasing the circulation of lakes and restoring

are mainly used, latrine improvement in the countryside

the natural hydrological ﬂuctuation rhythm of a portion of

and sewage treatment should be integrated. In areas where

the lakes.

traditional dry latrines and waterless latrines are mainly used, fecal pollution-free treatment and resource utilization should be conducted and construction space should be

ACKNOWLEDGEMENTS

reserved for later sewage treatment. Third, the circulation of lakes should be improved, and

The project was ﬁnancially supported by the National Key

the natural hydrological ﬂuctuation rhythms should be

R&D Program of China (Grant No. 2016YFC0401308),

restored in a portion of the lakes. The diversion of the

the Chinese National Special Science and Technology

SNWTP-ER might be contrary to the natural ﬂow of the

Program of Water Pollution Control and Treatment (Grant

lake, which changes the natural ﬂow direction of the water

No. 2017ZX07302004), the Fundamental Research Funds

body to a certain extent and might cause disturbances in

for the Central Universities (Grant No. 2019NTST19) and

the ecosystem. From the perspective of improvement

the National Natural Science Foundation of China (Grant

measures, optimizing lake ﬁshery management, improving

Nos. 51679006, 51879006 and 41603113) and the 111

the effective interception capacity of nutrients in the lake-

Project (B18006).

side and implementing lake ecological restoration projects (Zhu et al. ) are the keys to controlling lake eutrophication and improving the regional lake ecological quality in

DATA AVAILABILITY STATEMENT

the future. Data cannot be made publicly available; readers should con-

CONCLUSIONS The DID model is feasible for the quantitative and qualitat-

tact the corresponding author for details.

REFERENCES

ive detection of the impact of SNWTP-ER policy on lake eutrophication in the Shandong Peninsula. Our results demonstrated that Model 11, which included the influence of policy and considered a composite of environmental and socio-economic factors, acquired the optimal fitting degree with R 2 ¼ 0.80, and in this model, the eutrophication degree of the transferred lake deteriorated by 7.10% compared with that of the control lake without water transfer. DO was the main factor influencing the eutrophication of the Shandong Peninsula. A comparison of the Pearson

Ashenfelter, O. & Card, D.  Using the longitudinal structure of earnings to estimate the effect of training programs. Rev. Econ. Stat. 67, 648–660. Barrow, C.  Long-distance water transfer – a Chinese case study and international experiences – Biswas, AK, Dakang, Z, Nickum, JE, Changming, L. Third World Q. 6, 237–238. Bertrand, M., Duﬂo, E. & Mullainathan, S.  How much should we trust differences-in-differences estimates? Q. J. Econ. 119, 249–275. Davies, B. R., Thoms, M. & Meador, M.  An assessment of the ecological impacts of inter-basin water transfers, and their

1075

J. He et al.

Using DID model to identify the impact of SNWTP policy on lake eutrophication

threats to river basin integrity and conservation. Aquat. Conserv. 2, 325–349. Diersing, N.  Phytoplankton Blooms: The Basics. Florida Keys National Marine Sanctuary, National Oceanic and Atmospheric Administration, Washington, DC. Dimick, J. B. & Ryan, A. M.  Methods for evaluating changes in health care policy: the difference-in-differences approach. JAMA 312, 2401–2402. Fang, Q., Wang, G., Liu, T., Xue, B., Sun, W. & Shrestha, S.  Unraveling the sensitivity and nonlinear response of water use efficiency to the water-energy balance and underlying surface condition in a semiarid basin. Sci. Total Environ. 699, 134405. Gao, X., Zhuang, W., Chen, C.-T.A. & Zhang, Y.  Sediment quality of the SW coastal Laizhou Bay, Bohai Sea, China: a comprehensive assessment based on the analysis of heavy metals. PLoS ONE 10, 1–27. Geider, R. J. & La Roche, J.  Redfield revisited: variability of C:N:P in marine microalgae and its biochemical basis. Eur. J. Phycol. 37, 1–17. Grant, E. H. C., Lynch, H. J., Muneepeerakul, R., Arunachalam, M., Rodriguez-Iturbe, I. & Fagan, W. F.  Interbasin water transfer, riverine connectivity, and spatial controls on fish biodiversity. PLoS ONE 7, 1–7. Guo, C., Chen, Y., Li, W., Xie, S., Lek, S. & Li, Z.  Food web structure and ecosystem properties of the largest impounded lake along the eastern route of China’s south-to-north water diversion project. Ecol. Inform. 43, 174–184. Guo, C., Chen, Y., Liu, H., Lu, Y., Qu, X., Yuan, H., Lek, S. & Xie, S. a Modelling fish communities in relation to water quality in the impounded lakes of China’s south-to-north water diversion project. Ecol. Model. 397, 25–35. Guo, C., Chen, Y., Xia, W., Qu, X., Yuan, H., Xie, S. & Lin, L.-S. b Eutrophication and heavy metal pollution patterns in the water supplying lakes of China’s south-to-north water diversion project. Sci. Total Environ. 711, 134543. Hou, Q., Yang, Z., Ji, J., Yu, T., Chen, G., Li, J., Xia, X., Zhang, M. & Yuan, X.  Annual net input fluxes of heavy metals of the agro-ecosystem in the Yangtze River Delta, China. J. Geochem. Explor. 139, 68–84. Huo, S., Ma, C., Xi, B., Su, J., Zan, F., Ji, D. & He, Z.  Establishing eutrophication assessment standards for four lake regions, China. J. Environ. Sci (China) 25 (10), 2014–2022. Kuo, Y., Liu, W., Zhao, E., Li, R. & Muñoz-Carpena, R.  Water quality variability in the middle and down streams of Han River under the influence of the middle route of south-north water diversion project, China. J. Hydrol. 569, 218–229. Li, W.  Research on Water Quality of Nansi Lake and the Inflowing Rivers. Tianjin University, Tianjin. Li, X., Liu, L., Wang, Y., Luo, G., Chen, X., Yang, X., Gao, B. & He, X.  Integrated assessment of heavy metal contamination in sediments from a coastal industrial basin, NE China. PLoS ONE 7, e39690. https://doi.org/10.1371/ journal.pone.0039690.

Hydrology Research

51.5

2020

Li, Q., Peng, Y., Wang, G., Wang, H., Xue, B. & Hu, X.  A combined method for estimating continuous runoff by parameter transfer and drainage area ratio method in ungauged catchments. Water 11, 1104. Li, Q., Wang, G., Wang, H., Shrestha, S., Xue, B., Sun, W. & Yu, J.  Macrozoobenthos variations in shallow connected lakes under the influence of intense hydrologic pulse changes. J. Hydrol. 584, 124755. Liu, X., Zhang, G., Sun, G., Wu, Y. & Chen, Y.  Assessment of lake water quality and eutrophication risk in an agricultural irrigation area: a case study of the Chagan Lake in Northeast China. Water 11, 2380. Lu, W.  Editorial Board and Staff. Shandong Statistical Yearbook–2018. China Statistics Press, Beijing. Meyer, B. D.  Natural and quasi-experiments in economics. J. Bus. Econ. Stat. 13, 151–161. Ministry of Water Resources  South-North Water Transfer Project Masterplan (Summary). Ministry of Water Resources, Beijing. Rogers, S., Chen, D., Jiang, H., Rutherfurd, I., Wang, M., Webber, M., Crow-Miller, B., Barnett, J., Finlayson, B., Jiang, M., Shi, C. & Zhang, W.  An integrated assessment of China’s south–north water transfer project. Geogr. Res. 58, 49–63. Scheren, P., Zanting, H. A. & Lemmens, A. M. C.  Estimation of water pollution sources in Lake Victoria, East Africa: application and elaboration of the rapid assessment methodology. J. Environ. Manage. 58, 235–248. Sheng, J. & Webber, M.  Incentive-compatible payments for watershed services along the eastern route of China’s southnorth water transfer project. Ecosyst. Serv. 25, 213–226. Smith, V. H. & Schindler, D. W.  Eutrophication science: where do we go from here? Trends Ecol. Evol. 24, 201–207. Wang, C., Wang, Y. & Wang, P.  Water quality modeling and pollution control for the eastern route of South-to-North Water Transfer Project in China. J. Hydrodyn. 18, 253–261. Wang, G., Fang, Q., Teng, Y. & Yu, J.  Determination of the factors governing soil erodibility using hyperspectral visible and near-infrared reflectance spectroscopy. Int. J. Appl. Earth Obs. Geoinf. 53, 48–63. Wang, G., Hu, X., Zhu, Y., Jiang, H. & Wang, H.  Historical accumulation and ecological risk assessment of heavy metals in sediments of a drinking water lake. Environ. Sci. Pollut. Res. 25, 24882–24894. Wang, G., Li, J., Sun, W., Xue, B., Yinglan, A. & Liu, T. a Nonpoint source pollution risks in a drinking water protection zone based on remote sensing data embedded within a nutrient budget model. Water Res. 157, 238–246. Wang, P., Yao, J., Wang, G., Hao, F., Shrestha, S., Xue, B., Xie, G. & Peng, Y. b Exploring the application of artificial intelligence technology for identification of water pollution characteristics and tracing the source of water quality pollutants. Sci. Total Environ. 693, 133440. Wang, H., Yan, H., Zhou, F., Li, B., Zhuang, W. & Shen, Y.  Changes in nutrient transport from the Yangtze River to the East China Sea linked to the Three-Gorges Dam and water transfer project. Environ. Pollut. 256, 113376.

1076

J. He et al.

Using DID model to identify the impact of SNWTP policy on lake eutrophication

Wei, F.  Methods of Monitoring and Analysis of Water and Wastewater. China Environmental Science, Beijing. Wu, B., Wang, G., Jiang, H., Wang, J. & Liu, C.  Impact of revised thermal stability on pollutant transport time in a deep reservoir. J. Hydrol. 535, 671–687. Wu, B., Wang, G., Wang, Z., Liu, C. & Ma, J.  Integrated hydrologic and hydrodynamic modeling to assess water exchange in a data-scarce reservoir. J. Hydrol. 555, 15–30. Wu, Y., Dai, R., Xu, Y., Han, J. & Li, P.  Statistical assessment of water quality issues in Hongze Lake, China, related to the operation of a water diversion project. Sustainability 10, 1885. Xu, Z., Cai, X., Yin, X., Su, M., Wu, Y. & Yang, Z.  Is water shortage risk decreased at the expense of deteriorating water quality in a large water supply reservoir? Water Res. 165, 114984. Xu, T., Yang, T. & Xiong, M.  Time scales of external loading and spatial heterogeneity in nutrients-chlorophyll a response: implication on eutrophication control in a large shallow lake. Ecol. Eng. 142, 105636. Yang, H., Wang, G., Wang, L. & Zheng, B.  Impact of land use changes on water quality in headwaters of the Three Gorges Reservoir. Environ. Sci. Pollut. Res. 23, 11448–11460. Yao, J., Wang, G., Xue, B., Wang, P., Hao, F., Xie, G. & Peng, Y. a Assessment of lake eutrophication using a novel multidimensional similarity cloud model. J. Environ. Manage. 248, 109259. Yao, J., Wang, G., Xue, W., Yao, Z. & Xue, B. b Assessing the adaptability of water resources system in Shandong Province, China, using a novel comprehensive co-evolution model. Water Resour. Manage. 33, 657–675. Yao, J., Wang, P., Wang, G., Shrestha, S., Xue, B. & Sun, W.  Establishing a time series trend structure model to mine potential hydrological information from hydrometeorological time series data. Sci. Total Environ. 698, 134227.

Hydrology Research

51.5

2020

Ye, C., Shen, Z. M., Zhang, T., Fan, M. H., Lei, Y. M. & Zhang, J. D.  Long-term joint effect of nutrients and temperature increase on algal growth in Lake Taihu, China. J. Environ. Sci (China) 23, 222–227. Yinglan, A., Wang, G., Liu, T., Shrestha, S., Xue, B. & Tan, Z. a Vertical variations of soil water and its controlling factors based on the structural equation model in a semi-arid grassland. Sci. Total Environ. 691, 1016–1026. Yinglan, A., Wang, G., Liu, T., Xue, B. & Kuczera, G. b Spatial variation of correlations between vertical soil water and evapotranspiration and their controlling factors in a semiarid region. J. Hydrol. 574, 53–63. Zeng, Q., Qin, L. & Li, X.  The potential impact of an interbasin water transfer project on nutrients (nitrogen and phosphorous) and chlorophyll a of the receiving water system. Sci. Total Environ. 536, 675–686. Zhang, Q.  The South-to-North Water Transfer Project of China: environmental implications and monitoring strategy. J. Am. Water Resour. Assoc. 45, 1238–1247. Zhang, J., Ni, W., Luo, Y., Stevenson, R. & Qi, J.  Response of freshwater algae to water quality in Qinshan Lake within Taihu Watershed, China. Phys. Chem. Earth 36, 360–365. Zhou, H., Taber, C., Arcona, S. & Li, Y.  Difference-indifferences method in comparative effectiveness research: utility with unbalanced groups. Appl. Health Econ. Health Policy 14, 419–429. Zhu, G. X. H., Zhu, M., Zou, W., Guo, C. J. P. D. W., Zhou, Y., Zhang, Y. & Qin, B.  Changing characteristics and driving factors of trophic state of lakes in the middle and lower reaches of Yangtze River in the past 30 years. J. Lake Sci. 31, 1510–1524. Zhuang, W., Ying, S. C., Frie, A. L., Wang, Q., Song, J., Liu, Y., Chen, Q. & Lai, X.  Distribution, pollution status, and source apportionment of trace metals in lake sediments under the inﬂuence of the South-to-North Water Transfer Project, China. Sci. Total Environ. 671, 108–118.

First received 16 April 2020; accepted in revised form 3 July 2020. Available online 31 July 2020

1077

51.5

2020

Succession of phytoplankton in a shallow lake under the alternating inﬂuence of runoff and reverse water transfer Qing Li, Guoqiang Wang

, Zhongxin Tan and Hongqi Wang

ABSTRACT Both runoff and water diversion can interfere with the physical and chemical environment of a lake and affect aquatic organisms. In this study, previously obtained data were used to analyze the phytoplankton community, water quality, water level, and temperature in Dongping Lake (DPH) before, during, and after the water diversion caused by the South-to-North Water Transfer Project. The results showed that the total density and diversity index of phytoplankton decreased in the water transfer period, and was related to low temperature. Temperature also affected the recovery of phytoplankton community structure when the water transfer period ended. In a water transfer

Qing Li (corresponding author) Guoqiang Wang Zhongxin Tan Hongqi Wang Beijing Key Laboratory of Urban Hydrological Cycle and Sponge City Technology, College of Water Sciences, Beijing Normal University, Beijing 100875, China E-mail: wanggq@bnu.edu.cn

cycle, changes in dominant genera were more drastic than that of a whole phytoplankton community, and dominant genera were sensitive to total phosphorus (TP) and total nitrogen (TN) changes. Water transfer alleviated the deterioration of water quality in DPH, but water transfer process increased the risk of water pollution. Runoff from Dawen River carried TN, TP, and chemical oxygen demand (COD) into DPH in the rainy season, which indirectly affected phytoplankton, while it also carried phytoplankton directly into DPH. Overall, these ﬁndings provide a clear understanding of the impact of water transfer projects on ecology in shallow lakes. Key words

| Dongping Lake, phytoplankton community, reverse water transfer, runoff, shallow lake, the South-to-North Water Transfer Project

HIGHLIGHTS

• • • •

Dominant genera change greater than genera in a whole phytoplankton community. Reverse water transfer and runoff affect phytoplankton community alternately. There is a risk of water quality deterioration during the water transfer process. Runoff transfers water, pollutants and phytoplankton to the lakes.

INTRODUCTION The inﬂuence of hydrology on the water ecosystem in shal-

change the original hydrologic conditions in lakes and dis-

low lakes is obvious. Both natural runoff and water

rupt natural, stable patterns. This affects the habitats of

diversion projects affect lakes by changing hydrology and

aquatic organisms and controls their community structure

water quality. In particular, the water transfer projects

and distribution. When hydrology changes, a series of water physical and chemical properties such as water

This is an Open Access article distributed under the terms of the Creative

level, ﬂow velocity, and water quality are changed (Tuttle

Commons Attribution Licence (CC BY-NC-ND 4.0), which permits copying

et al. ; Liu et al. ). When hydrology is the driving

and redistribution for non-commercial purposes with no derivatives, provided the original work is properly cited (http://creativecommons.org/

factor, the speciﬁc responses of the aquatic communities

licenses/by-nc-nd/4.0/).

in lakes are diverse and complex.

doi: 10.2166/nh.2020.163

1078

Q. Li et al.

Succession of phytoplankton under the inﬂuence of water transfer

Hydrology Research

51.5

2020

Most natural runoff is caused by rainfall. In areas signiﬁ-

introduced by water transfer projects and water pollutants

cantly affected by the monsoon climate, natural runoff is

in water conveyance channels affect the water quality in

concentrated in the rainy season. During this period,

receiving lakes (Zhuang ; Liu et al. ; Zhuang et al.

runoff is characterized by a big ﬂow amount and high ﬂow

). Unlike natural runoff, man-made water transfer pro-

velocity that results in ﬂooding or seasonal pulses (Rodger

jects have changed the natural pattern and broken the

et al. ; Galir Balkić et al. ; Han et al. ). However,

dynamic and stable inﬂuence of natural runoff on lakes.

rainfall-runoff during dry seasons is characterized by low

Although the effects of water diversion are complex, the

ﬂow velocity and a small water amount. Natural runoff

reverse water diversion exacerbates this complexity. Accord-

showed the same pattern in every hydrological year in

ingly, it is necessary to study the ecology in shallow lakes

areas controlled by a temperate continental climate. For a

under the inﬂuence of reverse water transfer projects; how-

long time, runoff has increased the water amount and

ever, few studies have investigated reverse water transfer.

brought pollution (Wang et al. a, b), which con-

As the obviously seasonal succession, phytoplankton

stantly affects the water ecosystem in lakes (Silva et al.

community change rapidly in a short period with changes

; Sun et al. ). Runoff affects the phytoplankton suc-

in habitat environments. Accordingly, phytoplankton com-

cession by changing temperature and nutrients (Cao et al.

munity changes are commonly used to reﬂect the inﬂuence

). Increased runoff in the rainy season has also intro-

of external disturbances (Sharov ; Snit’ko & Snit’ko

duced pollutants, which favor the massive development of

; Toporowska et al. ; Weng et al. ). Community

Cyanobacteria in lakes (Rao et al. ; Silva et al. ;

structure, diversity, total density, and biomass are all

Sun et al. ) and also reduce the abundance of diatoms

important indicators of habitat environment changes

(Da Silva et al. ) or phytoplankton biomass (Cobbaert

(Deng et al. ; Özkan et al. ; Anneville et al. ).

et al. ). Flood pulses in rainy seasons have also been

In the shallow lakes, the water depth is small and the

shown to inﬂuence community structure by affecting food

water stratiﬁcation not obvious, characteristics at different

and nutrient levels (Galir Balkić et al. ). In addition to

depths tend to be consistent, and the exchange of sedi-

the recognition that rainfall runoff increases nutrients in

ments and water is more frequent than in deep lakes

lakes, runoff also has the potential to dilute pollutants

(Scheffer ; Ongun Sevindik et al. ). Phytoplankton

(Cobbaert et al. ; Ho & Michalak ). These factors

in shallow lakes is more responsive to outside inﬂuences,

complicate the effects of runoff in lakes.

such as changes in temperature, hydrological conditions,

Increasingly, water transfer projects are being designed

and water quality. The effect of water quality on phyto-

to solve the problem of uneven distribution of water

plankton in lakes is obvious, total nitrogen (TN), total

resources. One of the main functions of these projects is to

phosphorus (TP), and ammonia nitrogen (NH4-N) are clo-

introduce exogenous water to a receiving lake in dry sea-

sely related to the phytoplankton in most lakes (Borics

sons. In areas in which natural precipitation is extremely

et al. ; Zhu et al. ; Tang et al. ). Temperature

scarce, water diversion has helped alleviate the decline of

also affects the phytoplankton succession, the increase in

water levels and deterioration of water quality, then affect-

temperature contributes to the increase in total phyto-

ing algal blooms in lakes (Amano et al. ; Li et al. ;

plankton biomass (Markensten et al. ). Also, Rao

Huang et al. ; Yao et al. ). In the dry season, water

et al. () thought that the effect of temperature on phyto-

diversion increases the water amount, raises the water

plankton was highly species-speciﬁc and temperature also

level, and changes the habitat environment, leading to a

modulated the effects of nutrients on phytoplankton. In

chain reaction in the community structure of aquatic organ-

addition, water level ﬂuctuation also has an obvious inﬂu-

isms (Guo et al. ). Under the inﬂuence of water transfer

ence on phytoplankton succession (Kivrak ; Tian

projects, the dominant species of the phytoplankton com-

et al. ; Qian et al. ; Rao et al. ). However,

munity has changed because of the changing water quality

under the inﬂuence of runoff and water transfer, the

(Amano et al. ) and hydrodynamic conditions in lakes

effect of temperature, hydrological conditions and water

(Li et al. ). What is more, both exogenous water

quality on phytoplankton should be rethought.

1079

Q. Li et al.

Succession of phytoplankton under the inﬂuence of water transfer

Hydrology Research

51.5

2020

Rainfall in northern China is mostly concentrated in

an important ﬂood detention area in the lower reaches of

summer and autumn, while the region is cold and dry

the Yellow River, but is also the last regulating and storing

during spring and winter. Dongping Lake (DPH) is a typical

lake of the eastern route of the SNWTP. Accordingly, the

shallow lake located in northern China that mainly receives

protection of the ecosystem in the DPH is very important.

river runoff from the Dawen River (DWR) in the rainy

The surface area of DPH is about 125 km2 and the lake

season. In the dry season before the South-to-North Water

basin is narrow in the north and wide in the south (Figure 1).

Transfer Project (SNWTP), the water level in DPH fre-

DPH is a shallow lake with an annual average water depth

quently dropped because of reduced inﬂow, which

of about 2 m. The eastern part of the lake is mainly hilly,

seriously threatens the lake ecosystem (He et al. ).

while there are dams to the south and west. Few rivers

Implementation of the SNWTP alleviated the water loss

ﬂow into DPH. The DWR ﬂows from east to west into the

and eutrophication in DPH (Hu et al. ). However, the

lake, and then ﬂows into the Yellow River. DPH mainly

SNWTP also introduced the possibility of adverse impacts

depends on surface runoff and lake surface precipitation

on the environment (Liu et al. ), and even caused bio-

supply. The region in which DPH is located is characterized

logical invasion (Qin et al. ). To date, the SNWTP has

by a cold and dry winter and hot and rainy summer, with

mainly transferred water into DPH in the dry season. There-

more than 70% of the annual precipitation falling during

fore, after implementation of the SNWTP, DPH mainly

the ﬂood season. Therefore, the main inﬂow of the lake is

received runoff in the rainy season and transferred water

from the DWR in the rainy season. The eastern route of

in the dry season. The alternations of runoff and transferred

the SNWTP was completed in 2013. Liuchang River

water have complicated the effects on the ecology in DPH.

serves as the route from south to north to bring water

Accordingly, water transfer projects participate in the

from the Nansihu Lake (NSH) to DPH via Denglou station

effect of runoff on the aquatic organisms in lakes. What is

(DL). The water transfer period of the SNWTP is generally

more, it is important to understand how runoff and water

concentrated in winter and spring, when there is less rainfall

transfer under the continuous inﬂuence of the SNWTP

and low water depth in DPH. The water transfer period in

affect the ecology in DPH within a water transfer cycle,

DPH in 2018 was from December 2018 to June 2019,

but this has not yet been investigated. Therefore, we (1)

during which the inﬂow was mainly composed of the

investigated and analyzed the phytoplankton and water

inﬂow of the DWR and the transferred water from the

quality in DPH and the DWR before, during, and after the

upper lakes of NSH (UNSH) via the Liuchang River. There-

water transfer period and (2) analyzed the main factors

fore, the inﬂow during the water transfer period and non-

inﬂuencing phytoplankton and the impacts of runoff

water transfer period is different.

inﬂow

and

diversion

inﬂow

the

phytoplankton

community in DPH. Additionally, (3) the phytoplankton

Data description and sample collection

community structural changes and dominant taxa replacement of phytoplankton under the alternating inﬂuence of

Water diversion in DPH always occurs in the dry season, so

runoff and water diversion were discussed.

one year is considered as a water transfer cycle, including before the water transfer period (S1) (August 2018), during the water transfer period (S2) (March 2019), and after the

MATERIALS AND METHODS

water transfer period (S3) (July 2019). Water quality samples and phytoplankton samples were collected at four sites in

Study area

DPH and one site in the DWR. Water quality samples were collected at one site in the UNSH.

DPH, located in Shandong Province, China, is a major fresh-

Quantitative phytoplankton samples were collected

water lake with diverse aquatic life. This lake has many

using a 2.5 L water sampler from a depth of 0.5 m, after

functions including ﬂood control, storage, irrigation, water

which, samples were ﬁxed with Lugol’s solution and 40%

supply, and preservation of biodiversity. DPH is not only

formalin. Qualitative phytoplankton samples were collected

Q. Li et al.

1080

Figure 1

Succession of phytoplankton under the inﬂuence of water transfer

Hydrology Research

51.5

2020

The location of Dongping Lake (DPH). The arrows indicate the directions of the water ﬂow, and the dotted arrow indicates the direction of the water transfer.

using a 25# ﬂoating net and then were kept at 4 C. All

laboratory, total nitrogen (TN), total phosphorus (TP),

samples were collected twice at each sampling site. After

ammonia nitrogen (NH4-N), and chemical oxygen demand

the samples were sent to the laboratory, the phytoplankton

(COD) were measured. Data describing the transferred

was identiﬁed and quantiﬁed. Upon arrival in the

water and water level were obtained from the local

Q. Li et al.

1081

Succession of phytoplankton under the inﬂuence of water transfer

Hydrology Research

51.5

2020

government, while rainfall and temperature data were

phytoplankton phylum. The calculation and visualization

downloaded from the China Meteorological Data Service

of RDA results are based on the R package ‘vegan’. Niche breadth models were used to analyze the change

Center.

in phytoplankton community with the environment. Shannon–Wiener index (Equation (3)) is often used to calculate

Materials and methods

niche breadth (Heino & Tolonen ; Li et al. ), and its signiﬁcance is different to Equation (1):

Calculation of phytoplankton community diversity We calculated the diversity to evaluate changes in phyto-

H0i ¼

) to evaluate the diversity of the phytoplankton community:

(Pij log2 Pij )

(3)

j¼1

plankton communities during different periods. Here, we used the Shannon–Wiener index (Equation (1)) (Shannon

R X

where H0i is the niche breadth index of the i-th genus; R is the total number of sampling sites; Pij is the ratio of the numbers of the i-th species at the j-th sampling site to the total number of the i-th genus.

H¼

N X

(Pi log2 Pi )

(1) Identiﬁcation of dominant taxa

i¼1

where H is the Shannon–Wiener index; Pi is the density

Dominant genera in the phytoplankton were screened to

ratio of the i-th genus; and N is the total number of the

analyze community changes. The dominant genera were

genera.

determined according to the dominance. The genus in the

We used the Jaccard similarity coefﬁcient (Equation (2)) (Jaccard ) to obtain the community similarity:

top 25% of dominance was selected as the dominant genus. The formula used to calculate dominance was as follows:

j C¼ aþb j

(2)

PM Yi ¼

j¼1

nij

× fi

(4)

where C is the Jaccard similarity coefﬁcient of the community; j is the number of the common genera in different

where Yi is the dominance of the i-th genus; nij is the density

communities; a and b are the number of genera in particular

of the i-th genus at the j-th site; M is the site number; N is the

communities.

total density of phytoplankton; and fi is the occurrence frequency of the i-th genus.

Analysis of key driving factors Redundancy analysis (RDA) was selected to evaluate the

RESULTS

relationship between environmental factors and phytoplankton communities. RDA is an ordination method that

Variations in the phytoplankton community in a water

combines regression analysis with principle component

transfer cycle

analysis, which is widely used to identify inﬂuencing factors. Detrended correspondence analysis (DCA) is required before RDA can be applied. If the ﬁrst axis value of the gra-

Changes in phytoplankton community structure during different periods

dient length is less than 3.0, the RDA method is appropriate. Here, environmental factors include TN, TP, NH4-N,

In the whole water transfer cycle, eight phyla and 59

COD, and temperature in water. The biomes includes

genera, including Cyanophyta, Bacillariophyta, Chrysophyta,

Q. Li et al.

1082

Succession of phytoplankton under the inﬂuence of water transfer

Hydrology Research

Table 1

51.5

2020

The composition difference of the phytoplankton in a water transfer cycle

Scenarios

Genus number

Similarity coefﬁcient

Dominant genus number

Similarity coefﬁcient

S1 ∩ S2 ∩ S3

0.11

S1 ∩ S2

0.20

S2 ∩ S3

0.33

0.04

S1 ∩ S3

0.50

0.14

DWR ∩ S2

0.39

DWR ∩ S3

0.45

similarities and differences revealed that a total of nine genera appeared in three periods at the same time with a similarity coefﬁcient of 0.11. Moreover, the community similarity coefﬁcient was 0.20 in S1 and S2, 0.33 in S2 and S3, and 0.50 in S1 and S3. Hence, the community similarity between two non-water transfer periods was higher than that between the water transfer period and the non-water Figure 2

Variations in phytoplankton community structure in a water transfer cycle. S1, S2, and S3 refer to before, during, and after the water transfer period,

transfer period.

respectively.

Pyrrophyta, Cryptophyta, Euglenophyta, Chlorophyta, and

Variations in phytoplankton diversity during different periods

Xanthophyta, were identiﬁed in DPH (Figure 2). Community structure in S1, S2, and S3 showed that Chlorophyta,

To further understand the changes in different phytoplank-

Cyanophyta, and Bacillariophyta were the dominant phyla,

ton taxa in a water transfer cycle, we analyzed the

while Cryptophyta and Euglenophyta were not identiﬁed

changes in qualitative structure of phytoplankton. As

during the water transfer period and Xanthophyta was not

shown in Figure 3, the Shannon–Wiener index of phyto-

identiﬁed during two non-water transfer periods (S1 and

plankton was signiﬁcantly different between the water

S3). Total density of phytoplankton decreased rapidly

transfer period and the non-water transfer period. During

during the water transfer period, and then recovered slowly after the water transfer period. Cyanophyta showed the same pattern as the total density. Cyanophyta dominated the phytoplankton community before the water transfer period, decreased during the water transfer period, and recovered slowly after the water transfer period. The proportion of Chlorophyta increased rapidly during the water transfer period, then decreased after the water transfer period. The proportion of Chrysophyta also increased during the water transfer period, while the proportion of Bacillariophyta was stable in different periods and did not change signiﬁcantly, indicating that this phylum had good adaptability to changing environments. Additionally, 40, 21, and 32 genera of phytoplankton were identiﬁed during the three periods, respectively (Table 1). Analysis of the

Figure 3

The diversity index of the phytoplankton in a water transfer cycle. The white boxes are Shannon–Wiener indexes, the grey boxes are richness values.

Q. Li et al.

1083

Succession of phytoplankton under the inﬂuence of water transfer

Hydrology Research

51.5

2020

the water transfer period, the Shannon–Wiener index was

with the highest dominance in S1 was Phormidium. Three

relatively low, and the richness value was also low; this indi-

dominant genus in S1 and S2 were the same, Cyclotella,

cated that low phytoplankton diversity was probably due to

Dactylococcopsis, and Oscillatoria. Additionally, the domi-

low richness. Moreover, the Shannon–Wiener index differed

nance value of Oscillatoria in S1 and S3 was different, and

greatly among sites during the water transfer period. Speciﬁ-

it presented the smallest dominance value in S3. No

cally, the Shannon–Wiener index in the central area (D04)

repeat dominant genus was found in S1 and S2.

(1.61) and the northern part of DPH (D01) (1.41) was smaller than at other sites, while the Shannon–Wiener index

Environmental factors affecting phytoplankton

near the entrance of the DWR (D05) (2.74) and at the

communities

southern part of DPH (D07) (2.95) was larger than at other sites. During two non-transfer periods, the Shannon–

Variations in environmental and hydrological factors

Wiener index of genus was relatively high and consistent at each site. Overall, the density of the phytoplankton com-

The water level in DPH ﬂuctuated throughout the water

munity in the study area was relatively uniform before and

transfer cycle (Figure 5), being generally low in S1 and

after the water transfer period. After the water transfer

increasing in S2. In 2018, Shandong Province experienced

period, phytoplankton diversity gradually recovered, but

the heaviest rainfall in years. During that year, rainfall in

was not the same as the original state, which might also

the rainy season was much heavier than that in the historical

be related to the insufﬁcient sampling interval between S2

period (570 mm), and the water storage volume in DPH was

and S3.

63% higher than usual. Therefore, the rainfall runoff replenishment of DPH resulted in the overall elevation of the

Dominant taxa during different periods

water level in the rainy season. The highest water level in S2 came as a result of the heavy rainfall. DPH was dry in

Ten, ﬁve, and nine dominant genera were screened in S1,

the dry season. The rainfall in post-ﬂood season of 2018

S2, and S3, respectively. Based on the distribution of the

and pre-ﬂood season of 2019 was lower than the post-

dominant genera in the three different periods, the domi-

ﬂood (110 mm) and pre-ﬂood (257 mm) seasons in the

nance of each genus was not signiﬁcantly different, but

same historical period, and the water level dropped accord-

none of the dominant genera were the same (Figure 4).

ingly (Figure 6). At this time, the transferred water ﬂowed

There was only one genus that was dominant in both S2

into DPH, which alleviated the water level decline. Water

and S3, namely, Chlorella, which was the dominant genus

demand around DPH changes with the seasons. The drop

with the highest dominance value in S2 and S3. The genus

of water level after March was due to an increase in agricultural activities; accordingly, the transferred water amount is adjusted to keep the water level in DPH stable. Water quality is one of the major factors affecting aquatic communities. In a whole water transfer cycle, water quality differed in different periods, and different water quality indicators also showed different changes. After a water transfer cycle, TP and NH4-N in DPH decreased, while COD and TN increased. Additionally, TN rose rapidly in S2 and fell in S3, which returned to close to the original level. COD decreased slightly in S2, but showed a higher value in S3. These changes indicated that the water quality ﬂuctuated within a water transfer cycle, but that it returned to close to its original level when the

Figure 4

The dominant genera in different periods. The codes are the same as Table 2.

water transfer was completed. However, the water quality

Q. Li et al.

1084

Figure 5

Succession of phytoplankton under the inﬂuence of water transfer

Hydrology Research

51.5

2020

The dynamics of water level and temperature in DPH during a water transfer cycle.

did not return to the original level after water diversion

niche breadth index was largest on the COD gradient and

when it was not in the rainy season. It is possible that the

smallest on the TP gradient, indicating that the change of

effects of precipitation on the improvement of water quality

COD in DPH had no signiﬁcant impact on the dominant

in the lake were not reﬂected yet in the collected samples

genera, while the change in TP had a greater impact on

during the third sampling.

the

Key factors driving phytoplankton distribution

mental gradient, indicating that they were well adapted to

dominant

genera.

Phormidium,

Merismopedia,

Chlorella, and Cyclotella ranked high on each environthe changing environment. Merismopedia and Cyclotella To better understand the environmental impact on phyto-

were present throughout the water transfer cycle, even

plankton, it is necessary to identify the main impact

though they were not dominant genera in every period.

factors. According to the RDA results (Figure 7), water qual-

Chlorella appeared in S2 and S3, while Phormidium

ity had great inﬂuence on phytoplankton community in

appeared in S1 and S3.

phylum level. Especially, the COD and water temperature (T water) in DPH greatly affected the phytoplankton. COD was related to the Bacillaroiphyta. T water was positively

DISCUSSION

correlated with both Cryptophyta and Pyrrophyta, but negatively correlated with Xanthophyta. TN was positively

Water transfer and river inﬂow jointly affect water

correlated with Xanthophyta, but negatively correlated

quality and water level during the water transfer period

with Cyanophyta, while NH4-N was closely correlated with Cyanophyta. To further analyze the dominant genera

Water transfer is always implemented in the dry season,

changes along environmental gradient, niche breadth

when there is little rainfall and limited runoff ﬂow into

values were calculated. As shown in Table 2, the mean

DPH. DPH mainly accepts transferred water and little

Q. Li et al.

1085

Table 2

Succession of phytoplankton under the inﬂuence of water transfer

Hydrology Research

51.5

2020

Niche breadth index of the dominant genera along environmental gradients

Code

Dominant genera

COD

NH4-N

RNP

Mean

Rank

Phormidium

1.4

1.05

1.87

1.04

1.282

Merismopedia

1.22

0.97

1.05

1.58

0.87

1.138

Scenedesmus

0.86

0.54

0.55

0.53

0.696

Westella

0.41

0.44

0.42

0.77

0.46

0.5

Cyclotella

0.6

1.24

0.8

0.83

0.69

0.832

Dactylococcopsis

0.51

0.73

0.43

0.61

0.48

0.552

Oscillatoria

0.49

0.79

0.45

0.51

0.39

0.526

Cryptomonas

0.38

0.24

0.18

0.45

0.07

0.264

Melosira

0.2

0.17

0.16

0.15

0.1

0.156

Fragilaria

0.3

0.15

0.22

0.27

0.07

0.202

Dinobryon

0.35

0.31

0.98

0.3

0.33

0.454

Tribonema

0.15

0.19

0.94

0.17

0.3

0.35

Ankistrodesmus

0.08

0.14

0.7

0.13

0.21

0.252

Synedra

0.06

0.1

0.43

0.09

0.51

0.238

Chlorella

0.45

1.63

1.48

0.7

1.62

1.176

Micryocystis

0.09

0.91

0.03

0.23

0.47

0.346

Chroococcus

0.06

0.65

0.03

0.14

0.34

0.244

Eudorina

0.06

0.52

0.01

0.13

0.26

0.196

Nitzschia

0.1

0.89

0.07

0.21

0.5

0.354

Pandorina

0.03

0.77

0.05

0.18

0.48

0.302

Mean

0.39

0.621

0.501

0.516

0.486

Rank

The table shows the sensitivity of each dominant genus to environmental indicators, and the bold values represent the average value. The larger the value was (Mean), the higher the ranking was (Rank), indicating that the species was less sensitive to the corresponding environmental indicators. This provided important support for the discussion of the relationship between phytoplankton and water quality indicators in this paper. Bold is for better presentation of important data.

inﬂows in the dry season. According to the section ‘Environ-

TN content in DWR. Although NH4-N content in DWR

mental factors affecting phytoplankton communities’, the

was higher than in DPH, it was not high enough to affect

phytoplankton community structure was closely related to

the NH4-N in DPH considering the DWR inﬂow amount.

water quality factors, and the inﬂuence of water diversion

More importantly, a large amount of transferred water

on water quality in DPH was obvious. Compared with the

with low NH4-N content also reduced the inﬂuence of

results before the water transfer period, TN increased,

NH4-N in DWR inﬂow on DPH. COD and TP showed

NH4-N decreased during the water transfer period. Based

little change during the water transfer period, because the

on the water quality changes in DPH (Figure 6), TN content

differences in COD and TP content between transferred

in the transferred water from UNSH was about 0.5 times

water and DPH, DWR and DPH were small. It is worth

that of DPH and in the inﬂow from DWR was almost

noting that, as the source of transferred water for DPH,

three times that of DPH. Even if the inﬂow from DWR

the content of each water quality indicator in UNSH was

was too small to be measured during the water transfer

less than that in DPH. However, water quality had already

period (Figure 5), it was still inevitable that a quite high

deteriorated at Denglou station. Increase in TN content

TN concentration entered the DPH through DWR inﬂow.

was evident in Denglou station, and the content of other

Thus, the increase in TN in DPH was mainly due to high

indicators, such as COD, NH4-N, TP, were also higher

Q. Li et al.

1086

Figure 6

Succession of phytoplankton under the inﬂuence of water transfer

Hydrology Research

51.5

2020

Variations in water quality factors in different periods. The black arrows refer to the ﬂow direction. (a) The water quality in DPH and (b) the water quality of inﬂows of DPH.

than that in DPH. Liangji Canal and the Liuchang River

Changes in phytoplankton community structure and

serve as water conveyance channels, but are surrounded

replacement of dominant taxa

by residential communities and farmland. Moreover, spring is the primary season for wheat growth in northern

Changes in habitat will lead to changes in community struc-

China, so fertilization, ﬁshing, and other agricultural activi-

ture, such as the replacement of dominant taxa. Cyanophyta

ties likely inﬂuence the TN content in DPH. Thus, human

played a dominant role in the community before and after

activities along the diversion route might have polluted the

water diversion. As the dominant phylum, Cyanophyta

transferred water, thus affecting the phytoplankton in

occupied a large percentage in the whole water transfer

DPH. Therefore, the water diversion and DWR jointly

cycle, and their distribution was closely related to NH4-N

affect the water quality in DPH. Water diversion not only

and TN (Figure 7). Cyanophyta decreased rapidly during

affected water quality, but also maintained the water level

the water transfer period, but gradually recovered after

in DPH directly. To ensure sufﬁcient water volume in the

water diversion, which was related to the decrease of

next stage, the transferred water in DPH must be large

NH4-N during the water transfer period (Figure 6). In

enough to raise the water level (Figure 5). Agricultural activi-

addition to water quality change, temperature differences

ties in the spring also greatly increase the water demand

caused by seasonal changes also inﬂuenced the dramatic

during the water transfer period, and this reduces the

changes in Cyanophyta. Increases in temperature have pre-

amount of water in the lake. These frequent water level ﬂuc-

viously been shown to be conducive to the dominance of

tuations promote material circulation in the lake and are

Cyanophyta (Markensten et al. ; Gray et al. ).

detrimental to the growth of phytoplankton (Kivrak );

Hence, Cyanophyta dominated the phytoplankton in the

therefore, they likely also affected the community structure

warm non-transfer periods, especially in S1, which had the

in DPH to some extent. Therefore, these water quality and

highest temperature among three samplings. The diversity

water level changes caused by water diversion and DWR

and the total density of phytoplankton also decreased in

inﬂow drove the phytoplankton succession in DPH.

S2. Dai et al. () also believed that the diversity of

Q. Li et al.

1087

Figure 7

Succession of phytoplankton under the inﬂuence of water transfer

Hydrology Research

51.5

2020

RDA of environmental factors and phytoplankton community during the whole water transfer cycle.

phytoplankton in the receiving water was lower during the

disappeared as the environment changed, but the taxa

water transfer period than the non-water transfer period,

with better adaptability to the environment were still at a

which was consistent with our results. The decrease in the

relatively stable level. When compared with the research

diversity and the total density might be related to low temp-

results before the SNWTP, the phytoplankton community

erature (Ho & Michalak ). Low temperature narrows

structure and Shannon–Wiener index in DPH changed

the niche breadth of genus with different niche require-

greatly during the same periods (Tian et al. ). Further

ments, leading to a decrease in the genera number in S2

analysis results of quantitative structure showed that the

and thus a decrease in the diversity index and richness

genera among three periods had a similarity coefﬁcient of

(Yang et al. ). Even though the phytoplankton commu-

0.10, while no dominant genus was the same (Table 1).

nity structure and total density showed a recovery trend

Additionally, as shown in Figure 4 and Table 1, the three

after the water transfer period, there were still some differ-

same dominant genera appeared before and after the

ences.

seasonal

water transfer period, while only one dominant genus was

changes might be one of the main reasons for the total phy-

the same during and after the water transfer period. More-

toplankton density difference between S1 and S2. Water

over, the similarity coefﬁcients of genera (0.33 and 0.50)

temperature is known to be closely related to the growth

were larger than that of dominant genera (0.04 and 0.14)

of phytoplankton (Markensten et al. ; Bergstrom et al.

during the same periods. Four in nine genera that recurred

), and the growth of some algae is known to have an

during these three periods belonged to Bacillariophyta,

obvious circadian rhythm (Straub et al. ). Before and

which maintained a stable percentage throughout the

after water diversion, the air temperature in DPH was simi-

whole water transfer cycle (Figure 2). Therefore, in a water

lar, while the water temperature in S1 was higher than in S3

transfer cycle, the change in dominant genus was more dras-

(Figure 5). Thus, the total phytoplankton density remained

tic than that of a whole phytoplankton community. The

at a low value after the water diversion.

sensitivity of each dominant genus to the environment was

Temperature

differences

caused

Compared with the results before the water transfer

further analyzed by the niche breadth model. In general,

period (S1), the phytoplankton genus number decreased

the species with narrower niche breadth are more environ-

(Table 1) and phytoplankton diversity increased at the

mentally sensitive. Results in Table 2 showed that the

genus level (Figure 3) after the water transfer period (S3).

dominant genera were sensitive to TP and the N/P ratio.

These

ﬁndings

indicated

that

some

sensitive

taxa

Most of the screened dominant genera belonged to the

1088

Q. Li et al.

Succession of phytoplankton under the inﬂuence of water transfer

Hydrology Research

51.5

2020

Cyanophyta and Chlorophyta, which were closely related to

with runoff in the rainy season, directly affecting the phyto-

the TN and TP. Thus, the TN and TP changes affected the

plankton in DPH. These ﬁndings conﬁrmed previous

dominant genera of phytoplankton.

ﬁndings that hydrological connectivity increased community similarity (Yuan et al. ).

Runoff affects phytoplankton by affecting water quality during non-transfer periods Water quality in DPH gradually recovered after the fluctuation during the water transfer period (Figure 6). Similar water quality conditions led to similar phytoplankton structures before and after water diversion, as well as high genera repetition rates. The water quality in DPH during the non-water transfer periods was mainly influenced by runoff inflow. In the rainy season in northern China, DPH mainly received the runoff supply from DWR. After the water transfer period, the COD in DPH rose rapidly, which was mainly caused by the high COD in the DWR, because of the higher COD content in DWR than in DPH (Figure 6). Even though samples collected after the water transfer were not obtained in the rainy season, the rainfall that occurred carried enough COD into DPH to increase the COD content. TN gradually decreased after the water transfer period, which might also be related to the runoff from the DWR. When compared with the water transfer period, TN content in DWR after the water diversion decreased greatly, but it was still higher than that in DPH. Rainfall-runoff caused the DWR to carry TN into DPH continuously, thus affecting water quality. NH4-N changed a little after the water transfer period because of the similar content in DPH and DWR. TP concentration in DPH was already very low and could be easily affected. Thus, the slight decrease in TP might be caused by the low TP content in DWR after the water transfer period (S3). The third sampling (S3) occurred soon after the water transfer period, so the dilution effect of the rainfall on pollutants was not reflected. This explained why the water

CONCLUSIONS Implementation of the SNWTP led to alternations between water diversion inﬂow and runoff in DPH, but phytoplankton

succession

was

still

not

clear.

this

study,

phytoplankton community before, during, and after the water transfer period in a water transfer cycle were analyzed, and various effects of water transfer and runoff on phytoplankton were discussed in combination with the corresponding changes in water quality, temperature, and water level. Low temperature caused Chlorophyta to replace Cyanophyta as the main dominant phylum, and low temperature was also related to the decrease in the total density and diversity index of phytoplankton during the water transfer period. Water temperature affected the recovery of total phytoplankton density after the water transfer period. In a water transfer cycle, changes in dominant genus were more drastic than that of a whole phytoplankton community, and dominant genera were sensitive to TP and TN changes. Transferred water and DWR inﬂow jointly affected the water quality in DPH during the water transfer period and water quality in DPH was improved. DWR was responsible for the increase in TN content. At the same time, the water transfer process increased the risk of water pollution. Runoff from DWR affected the water quality in DPH during the non-transfer periods. In addition, phytoplankton in DWR entered DPH through runoff, which affected the phytoplankton community directly.

quality indicators did not return to the level before the water transfer period. Moreover, similarity coefﬁcient of the phytoplankton community in DWR and DPH after the

ACKNOWLEDGEMENTS

water transfer period (S3) (0.45) was bigger than that during the water transfer period (S2) (0.39), and the total

This study was supported by the National Key R&D

number of phytoplankton genera in DPH and DWR in the

Program of China (Grant No. 2016YFC0401308) and the

non-transfer period were much higher than that in the

Chinese

water transfer period (Table 2). This indicated that DWR

Program of Water Pollution Control and Treatment (Grant

brought a large amount of phytoplankton into DPH along

No. 2017ZX07302004).

National

Special

Science

and

Technology

1089

Q. Li et al.

Succession of phytoplankton under the inﬂuence of water transfer

DATA AVAILABILITY STATEMENT All relevant data are included in the paper or its Supplementary Information.

REFERENCES Amano, Y., Sakai, Y., Sekiya, T., Takeya, K., Taki, K. & Machida, M.  Effect of phosphorus ﬂuctuation caused by river water dilution in eutrophic lake on competition between blue-green alga Microcystis aeruginosa and diatom Cyclotella sp. J. Environ. Sci. (China) 22, 1666–1673. Anneville, O., Chang, C. W., Dur, G., Souissi, S., Rimet, F. & Hsieh, C. h.  The paradox of re-oligotrophication: the role of bottom–up versus top–down controls on the phytoplankton community. Oikos 128, 1666–1677. Bergstrom, A. K., Faithfull, C., Karlsson, D. & Karlsson, J.  Nitrogen deposition and warming – effects on phytoplankton nutrient limitation in subarctic lakes. Global Change Biol. 19, 2557–2568. Borics, G., Nagy, L., Miron, S., Grigorszky, I., László-Nagy, Z., Lukács, B. A., G-Tóth, L. & Várbíró, G.  Which factors affect phytoplankton biomass in shallow eutrophic lakes? Hydrobiologia 714, 93–104. Cao, J., Hou, Z., Li, Z., Chu, Z., Yang, P. & Zheng, B.  Succession of phytoplankton functional groups and their driving factors in a subtropical plateau lake. Sci. Total Environ. 631–632, 1127–1137. Cobbaert, D., Wong, A. & Bayley, S. E.  Precipitation-induced alternative regime switches in shallow lakes of the Boreal Plains (Alberta, Canada). Ecosystems 17, 535–549. Dai, J., Wu, S., Wu, X., Xue, W., Yang, Q., Zhu, S., Wang, F. & Chen, D.  Effects of Water diversion from Yangtze River to Lake Taihu on the phytoplankton habitat of the Wangyu River channel. Water 10, 759–769. Da Silva, C. F. M., Torgan, L. C. & Schneck, F.  Temperature and surface runoff affect the community of periphytic diatoms and have distinct effects on functional groups: evidence of a mesocosms experiment. Hydrobiologia 839, 37–50. Deng, J., Qin, B., Sarvala, J., Salmaso, N., Zhu, G., Ventelä, A.-M., Zhang, Y., Gao, G., Nurminen, L., Kirkkala, T., Tarvainen, M. & Vuorio, K.  Phytoplankton assemblages respond differently to climate warming and eutrophication: a case study from Pyhäjärvi and Taihu. J. Great Lakes Res. 42, 386–396. Galir Balkić, A., Ternjej, I. & Špoljar, M.  Hydrology driven changes in the rotifer trophic structure and implications for food web interactions. Ecohydrology 11, e1917–e1919. Gray, E., Elliott, J. A., Mackay, E. B., Folkard, A. M., Keenan, P. O. & Jones, I. D.  Modelling lake cyanobacterial blooms Disentangling the climate-driven impacts of changing mixed depth and water temperature. Freshwat. Biol. 64, 2141–2155.

Hydrology Research

51.5

2020

Guo, C., Chen, Y., Liu, H., Lu, Y., Qu, X., Yuan, H., Lek, S. & Xie, S.  Modelling fish communities in relation to water quality in the impounded lakes of China’s South-to-North Water Diversion Project. Ecol. Model. 397, 25–35. Han, D., Wang, G., Liu, T., Xue, B.-L., Kuczera, G. & Xu, X.  Hydroclimatic response of evapotranspiration partitioning to prolonged droughts in semiarid grassland. J. Hydrol. 563, 766–777. He, S., Li, R. & Li, B.  Dongping lake water regulation scheme of the East Route of South-to-North Water Transfer Project (in Chinese). Yellow River 36, 52–54. Heino, J. & Tolonen, K. T.  Ecological niche features override biological traits and taxonomic relatedness as predictors of occupancy and abundance in lake littoral macroinvertebrates. Ecography 41, 2092–2103. Ho, J. C. & Michalak, A. M.  Exploring temperature and precipitation impacts on harmful algal blooms across continental U.S. lakes. Limnol. Oceanogr. 9999, 1–18. Hu, Z., Kang, F., Zou, A., Yu, L., Li, Y., Tian, T. & Guiling, K.  Evolution trend of the water quality in Dongping Lake after South-North Water Transfer Project in China. J. Groundw. Sci. Eng. 7, 333–339. Huang, J., Gao, J., Zhang, Y. & Xu, Y.  Modeling impacts of water transfers on alleviation of phytoplankton aggregation in Lake Taihu. J. Hydroinform. 17, 149–162. Jaccard, P.  Contribution au problème de l’immigration postglaciaire de la flore alpine: étude comparative de la flore alpine du massif de Wildhorn, du haut bassin du Trient et de la haute vallée de Bagnes (Contribution to post-glacial alpine flora migration: a comparative study of alpine flora in the Wilderhorn Range, the Upper Trent Basin and the Upper Barness Valley). Bull. Soc. Vaud. Sci. Nat. 36, 87–130. Kivrak, E.  Seasonal and long term changes of the phytoplankton in the lake Tortum in relation to environmental factors, Erzurum, Turkey. Biologia 61, 339–345. Li, Y., Tang, C., Wang, C., Anim, D., Yu, Z. & Acharya, K.  Improved Yangtze River diversions: are they helping to solve algal bloom problems in Lake Taihu, China? Ecol. Eng. 51, 104–116. Li, Q., Wang, G., Wang, H., Shrestha, S., Xue, B., Sun, W. & Yu, J.  Macrozoobenthos variations in shallow connected lakes under the influence of intense hydrologic pulse changes. J. Hydrol. 584, 124755–124766. Liu, J., Rühland, K. M., Chen, J., Xu, Y., Chen, S., Chen, Q., Huang, W., Xu, Q., Chen, F. & Smol, J. P.  Aerosol-weakened summer monsoons decrease lake fertilization on the Chinese Loess Plateau. Nat. Clim. Chang. 7, 190–194. Liu, J., Li, M., Wu, M., Luan, X., Wang, W. & Yu, Z.  Influences of the south–to-north water diversion project and virtual water flows on regional water resources considering both water quantity and quality. J. Clean Prod. 244, 118920–118927. Markensten, H., Moore, K. & Persson, I.  Simulated lake phytoplankton composition shifts toward cyanobacteria dominance in a future warmer climate. Ecol. Appl. 20, 752–767.

1090

Q. Li et al.

Succession of phytoplankton under the inﬂuence of water transfer

Ongun Sevindik, T., Tunca, H., Gönülol, A., Gürsoy, N., Küçükkaya, Ş . N. & Kinali, Z.  Phytoplankton dynamics and structure, and ecological status estimation by the Q assemblage index: a comparative analysis in two shallow Mediterranean lakes. Turk. J. Bot. 41, 25–36. Özkan, K., Jeppesen, E., Davidson, T., Bjerring, R., Johansson, L., Søndergaard, M., Lauridsen, T. & Svenning, J.-C.  Long-term trends and temporal synchrony in plankton richness, diversity and biomass driven by re-oligotrophication and climate across 17 Danish lakes. Water 8, 427–450. Qian, K., Liu, X. & Chen, Y.  Effects of water level fluctuation on phytoplankton succession in Poyang Lake, China – a five year study. Ecohydrol. Hydrobiol. 16, 175–184. Qin, J., Cheng, F., Zhang, L., Schmidt, B., Liu, J. & Xie, S.  Invasions of two estuarine gobiid species interactively induced from water diversion and saltwater intrusion. Manag. Biol. Invasion 10, 139–150. Rao, K., Zhang, X., Yi, X. J., Li, Z. S., Wang, P., Huang, G. W. & Guo, X. X.  Interactive effects of environmental factors on phytoplankton communities and benthic nutrient interactions in a shallow lake and adjoining rivers in China. Sci. Total Environ. 619–620, 1661–1672. Rodger, A. W., Mayes, K. B. & Winemiller, K. O.  Preliminary findings for a relationship between instream flow and shoal chub recruitment in the Lower Brazos River, Texas. Trans. Am. Fish. Soc. 145, 943–950. Scheffer, M.  Ecology of shallow lakes. Popul. Commun. Biol. 22, 76–121. Shannon, C. E.  A mathematical theory of communication. Bcll Syst. Tech. J. 27, 379–423. Sharov, A. N.  Phytoplankton as an indicator in estimating long-term changes in the water quality of large lakes. Water Resour. 35, 668–673. Silva, T. F. G., Vinçon-Leite, B., Lemaire, B. J., Petrucci, G., Giani, A., Figueredo, C. C. & Nascimento, N. d. O.  Impact of urban stormwater runoff on Cyanobacteria dynamics in a tropical urban lake. Water 11, 946–974. Snit’ko, L. V. & Snit’ko, V. P.  Phytoplankton as an indicator in assessing long-term variations in water quality of lakes Bolshoye Miassovo and Turgoyak, the South Urals. Water Resour. 41, 210–217. Straub, C., Quillardet, P., Vergalli, J., de Marsac, N. T. & Humbert, J. F.  A day in the life of microcystis aeruginosa strain PCC 7806 as revealed by a transcriptomic analysis. PLoS ONE 6, e16208–e16220. Sun, W., Wang, Y., Fu, Y. H., Xue, B., Wang, G., Yu, J., Zuo, D. & Xu, Z.  Spatial heterogeneity of changes in vegetation growth and their driving forces based on satellite observations of the Yarlung Zangbo River Basin in the Tibetan Plateau. J. Hydrol. 574, 324–332. Tang, C., Yi, Y., Yang, Z., Zhou, Y., Zerizghi, T., Wang, X., Cui, X. & Duan, P.  Planktonic indicators of trophic states for a

Hydrology Research

51.5

2020

shallow lake (Baiyangdian Lake, China). Limnologica 78, 125712–125723. Tian, C., Lu, X., Pei, H., Hu, W. & Xie, J.  Seasonal dynamics of phytoplankton and its relationship with the environmental factors in Dongping Lake, China. Environ. Monit. Assess. 185, 2627–2645. Tian, C., Pei, H., Hu, W., Hao, D., Doblin, M. A., Ren, Y., Wei, J. & Feng, Y.  Variation of phytoplankton functional groups modulated by hydraulic controls in Hongze Lake, China. Environ. Sci. Pollut. Res. 22, 18163–18175. Toporowska, M., Ferencz, B. & Dawidek, J.  Impact of lakecatchment processes on phytoplankton community structure in temperate shallow lakes. Ecohydrology 11, e2017–e2019. Tuttle, C. L., Zhang, L. & Mitsch, W. J.  Aquatic metabolism as an indicator of the ecological effects of hydrologic pulsing in flow-through wetlands. Ecol. Indicators 8, 795–806. Wang, G., Li, J., Sun, W., Xue, B., Yinglan, A. & Liu, T. a Nonpoint source pollution risks in a drinking water protection zone based on remote sensing data embedded within a nutrient budget model. Water Res. 157, 238–246. Wang, P., Yao, J., Wang, G., Hao, F., Shrestha, S., Xue, B. L., Xie, G. & Peng, Y. b Exploring the application of artificial intelligence technology for identification of water pollution characteristics and tracing the source of water quality pollutants. Sci. Total Environ. 693, 133440–133455. Weng, B., Yang, Y., Yan, D., Wang, J., Dong, G., Wang, K., Qin, T. & Dorjsuren, B.  Shift in plankton diversity and structure: influence of runoff composition in the Nagqu River on the Qinghai-Tibet Plateau. Ecol. Indicators 109, 105818–105825. Yang, Y., Stenger Kovács, C., Padisák, J. & Pettersson, K.  Effects of winter severity on spring phytoplankton development in a temperate lake (Lake Erken, Sweden). Hydrobiologia 780, 47–57. Yao, X., Zhang, L., Zhang, Y., Du, Y., Jiang, X. & Li, M.  Water diversion projects negatively impact lake metabolism: a case study in Lake Dazong, China. Sci. Total Environ. 613–614, 1460–1468. Yuan, Y., Jiang, M., Liu, X., Yu, H., Otte, M. L., Ma, C. & Her, Y. G.  Environmental variables influencing phytoplankton communities in hydrologically connected aquatic habitats in the Lake Xingkai basin. Ecol. Indicators 91, 1–12. Zhu, R., Wang, H., Chen, J., Shen, H. & Deng, X.  Use the predictive models to explore the key factors affecting phytoplankton succession in Lake Erhai, China. Environ. Sci. Pollut. Res. 25, 1283–1293. Zhuang, W.  Eco-environmental impact of inter-basin water transfer projects: a review. Environ. Sci. Pollut. Res. 23, 12867–12879. Zhuang, W., Ying, S. C., Frie, A. L., Wang, Q., Song, J., Liu, Y., Chen, Q. & Lai, X.  Distribution, pollution status, and source apportionment of trace metals in lake sediments under the influence of the South-to-North Water Transfer Project, China. Sci. Total Environ. 671, 108–118.

First received 12 May 2020; accepted in revised form 3 July 2020. Available online 10 August 2020

1091

51.5

2020

A framework for event-based ﬂood scaling analysis by hydrological modeling in data-scarce regions Jianzhu Li, Kun Lei, Ting Zhang, Wei Zhong, Aiqing Kang, Qiushuang Ma and Ping Feng

ABSTRACT Flood scaling theory is important for flood predictions in data-scarce regions but is often applied to quantile-based floods that have no physical mechanisms. In this study, we propose a framework for flood prediction in data-scarce regions by event-based flood scaling. After analyzing the factors controlling the flood scaling, flood events are first simulated by a hydrological model with different areally averaged rainfall events and curve number (CN) values as inputs, and the peak discharge of each subcatchment is obtained. Then, the flood scaling is analyzed according to the simulated peak discharge and subcatchment area. Accordingly, the relationship curves between the scaling exponent and the two explanatory factors (rainfall intensity and CN) can be drawn. Assuming that the flood and the corresponding rainfall event have the same frequency, the scaling exponent with a specific flood frequency can be interpolated from these curves. Key words

| data-scarce region, event-based ﬂood scaling, ﬂood prediction, hydrological modeling

HIGHLIGHTS

• • •

Hydrological modeling was used to conduct flood scaling analysis. Rainfall intensity and curve number were identified to be the two main factors affecting flood scaling. A framework of flood prediction by event-based flood scaling in data-scarce regions was proposed.

Jianzhu Li Kun Lei Ting Zhang (corresponding author) Qiushuang Ma Ping Feng State Key Laboratory of Hydraulic Engineering Simulation and Safety, Tianjin University, Tianjin, China E-mail: zhangting_hydro@tju.edu.cn Wei Zhong School of Management, Tianjin University of Technology, Tianjin, China Aiqing Kang State Key Laboratory of Simulation and Regulation of Water Cycle in River Basin, China Institute of Water Resources and Hydropower Research, Beijing 100038, China

1092

J. Li et al.

A framework of event-based ﬂood scaling analysis

Hydrology Research

51.5

2020

GRAPHICAL ABSTRACT

INTRODUCTION Flood scaling, which is extensively applied to ﬂood predic-

has been tested in many river basins of different sizes from

tion in ungauged and poorly gauged river basins of

small-scale to mesoscale basins (Ogden & Dawdy ;

different sizes, is usually expressed by an equation that

Gupta et al. ; Yue & Gan ; Kroll et al. ;

relates peak discharge to catchment descriptors such as

Ayalew et al. ). Generally, the scaling exponent is less

the drainage area, slope of the watershed, land use, and rain-

than 1 in quantile-based ﬂood scaling but might exceed 1

fall characteristics; among these descriptors, the drainage

in event-based ﬂood scaling. Moreover, ﬂood scaling ana-

area is the most widely used explanatory variable to predict

lyses are often conducted in homogeneous and nested

ﬂood peak discharge (Jothityangkoon & Sivapalan ;

watersheds, in which the ﬂood generation mechanisms are

Menabde & Sivapalan ; Galster et al. ; Al-Rawas

similar (Eash ; Ogden & Dawdy ; Furey & Gupta

& Valeo ; Ishak et al. ; Han et al. ; Farmer

, ). Accordingly, the function between the ﬂood

et al. ; Furey et al. ; Lee & Huang ). The

quantile and watershed area can be established, and the

power law function recommended worldwide is Q ¼ αA ,

scaling parameters can be obtained. However, the ﬁtted

where Q is the peak discharge for a given drainage area A,

function may not have satisfactory precision; in this situ-

α is the scaling intercept, and θ is the scaling exponent

ation, other controlling factors, such as the corresponding

(Gupta et al. ; Medhi & Tripathi ). If a reference

quantile-based rainfall intensity, should be taken into

river basin with drainage area A1 has a ﬂood peak discharge

consideration in addition to the drainage area.

Q1 and θ is determined, then the ﬂood peak Q2 in an

Although the quantile-based scaling exponent is widely

ungauged basin with drainage area A2 can be estimated by

used for future ﬂood prediction, the physical mechanism

the following equation: Q1/Q2 ¼ (A1/A2)θ.

cannot be interpreted because the peak discharges of

Most previous studies focused on the relation between

ﬂoods with the same return period in different catchments

the quantile-based peak discharge and drainage area. The

may not be generated by the same rainfall events

peak discharge shows simple scaling or multiscaling

(Liu et al. ). Therefore, Gupta et al. () proposed per-

for snowmelt-generated or rainfall-generated ﬂoods, which

forming ﬂood scaling research by using event-based ﬂoods

means that the scaling parameters (especially the scaling

for statistical analysis. As such, observed ﬂoods that occur

exponent) remain invariant or vary, respectively, with the

simultaneously in different catchments can be selected,

ﬂood return period (Gupta & Dawdy ). This hypothesis

and the spatial and temporal distributions of rainfall can

1093

J. Li et al.

A framework of event-based ﬂood scaling analysis

Hydrology Research

51.5

2020

be considered. Furey & Gupta () presented a new index

Hourly observed ﬂood data at the Zijingguan hydrologi-

to reﬂect the temporal variability of excess rainfall, and the

cal station were used for ﬂood modeling. The corresponding

results showed that the scaling parameters depended mainly

hourly rainfall data at seven rainfall gauging stations located

on the duration and amount of excess rainfall.

in the Zijingguan catchment were obtained from the Hydrol-

Together with statistical analysis, hydrological modeling

ogy and Water Resource Survey Bureau of Hebei Province.

can provide intuitive results regarding how meteorological

The hourly data range from 1956 to 2018. Since no large

and land surface factors impact ﬂood scaling parameters

ﬂoods occurred after 2000, some ﬂood events before 2000

(Yang et al. ). For simulated ﬂood events, other control-

were selected for simulation and ﬂood scaling analysis.

ling factors of the scaling parameters, such as the channel

To establish a hydrological model, a digital elevation

slope, ﬂow velocity, and rainfall distribution, can be ident-

model (DEM) with a resolution of 30 m was downloaded

iﬁed by changing the hydrological model parameters

from the Geospatial Data Cloud (http://www.gscloud.cn/).

(Mantilla ; Venkata ). Therefore, it is better to use

Remotely sensed land use data from 1980 to 2000 were

physically based distributed hydrological models that con-

collected from the Institute of Geographic Sciences and

ﬂood

Resources of the Chinese Academy of Sciences (http://

generation processes. For example, Ayalew et al. ()

www.resdc.cn/) and were classiﬁed into forestland, grass-

tain

all

the

important

parameters

reﬂecting

applied the CUENCAS rainfall–runoff model to three river

land, agricultural land, water area, urban area, and unused

basins in Iowa to investigate how the rainfall intensity, dur-

land. Soil data were downloaded from the Environmental

ation, hillslope overland ﬂow velocity, and channel ﬂow

and Ecological Science Data Center for West China

velocity inﬂuenced the scaling parameters; they found that

(http://westdc.westgis.ac.cn/).

the key factors affecting the scaling parameters were the rainfall duration and overland ﬂow velocity.

Methods

The main aims of this paper are to (1) perform a simulated ﬂood scaling analysis and evaluate the effects of the

HEC-HMS model calibration and validation

rainfall characteristics and land surface and (2) propose a framework for estimating the scaling exponent by hydrologi-

The Hydrologic Engineering Center-Hydrologic Modeling

cal modeling. Compared with previous studies, the novelty

System (HEC-HMS) model developed by the US Army

of this study is that a framework for determining the ﬂood

Corps of Engineers is widely used for watershed hydrologi-

scaling exponent is proposed by hydrological modeling in

cal simulations. This model has been veriﬁed to be

data-scarce regions, especially for multiscaling. Hence, the

applicable in semi-arid regions throughout China (Wang &

proposed method can be used in other data-scarce regions

Sun ). In addition, the Soil Conservation Service curve

on a global scale.

number (SCS-CN) method is often used to calculate the amount of generated runoff, and the SCS unit hydrograph method is used for ﬂow routing. Therefore, we selected

DATA AND METHODS

these two modules for the model. The CN value is determined based on the spatial distributions of the soil water

Data

content and vegetation types (A et al. ) and will vary considerably under wet and dry antecedent conditions

The ﬂood scaling analysis was carried out in the Zijingguan

(Han et al. ).

catchment, situated in the northeastern part of China. The

A set of default parameters is established when the data

catchment has a drainage area of 1,760 km2. The long-term

are input into the model, after which the parameters are

average annual precipitation is approximately 650 mm,

adjusted artiﬁcially to ensure that the simulated and

70–80% of which is concentrated in the ﬂood season (July,

observed ﬂood hydrographs ﬁt well. The relative error

August, and September). Flood events are mostly generated

(RE) and the Nash–Sutcliffe efﬁciency (NSE) coefﬁcient,

by rainstorms with a high intensity and short duration.

which are expressed as Equations (1) and (2), respectively,

J. Li et al.

1094

A framework of event-based ﬂood scaling analysis

Hydrology Research

51.5

2020

HEC-HMS model, which is applicable in semi-humid and

are used to evaluate the performance of the model.

semi-arid regions, is employed to simulate ﬂood events due RE ¼

^ i qi q qi

(1)

to its data availability and structure (Alfy ). The simulated peak discharges and corresponding subcatchment areas are ﬁtted with power functions, and then the scaling

^ i )2 (qi q NSE ¼ 1 P i )2 (qi q

(2)

î are the observed and simulated discharges, where qi and q i is the mean observed discharge of the respectively, and q flood event. If the value of the NSE coefficient is in the range of 0.6–1, the hydrological model is considered to perform well in simulating the flood event.

exponent is obtained for each flood event. Framework for determining the scaling exponent based on simulated flood events Based on the simulated flood events and scaling analysis, a framework that links event-based scaling exponents to quantile-based scaling exponents can be proposed. There are two main assumptions: (1) the input rainfall at each rain gauge is

Flood scaling analysis

spatially uniform and (2) the rainfall events have the same

This study aims to analyze the ﬂood scaling exponent in a

cedures are as follows:

data-scarce region. Therefore, the Zijingguan catchment

1. A typical rainfall event is selected from the observed data.

frequency as the generated ﬂood events. The idea and pro-

with a drainage area of 1,760 km (Figure 1), which has

A typical rainfall event should have a large rainfall

only one hydrological station (Zijingguan station), is

amount and high rainfall intensity. Generally, the largest

selected. Scaling analysis is conducted by using the simu-

rainfall event is selected for its rare occurrence as the

lated ﬂood peak discharges from 11 subcatchments. The

low-frequency design ﬂood.

Figure 1

Subcatchments of the Zijingguan catchment.

J. Li et al.

1095

A framework of event-based ﬂood scaling analysis

2. The typical rainfall event is ampliﬁed. The typical rainfall event is ampliﬁed by multiple constants, such as 1.5, 2, 2.5, … , to form hypothetical rainfall events with the same temporal distribution as the typical rainfall

Hydrology Research

51.5

2020

corresponding scaling exponent is calculated according to the relationship established in step 4. The framework is illustrated in Figure 2.

event. 3. The ampliﬁed rainfall events are used as inputs to drive the established HEC-HMS model, and the corresponding

RESULTS

ﬂood events at the outlets of the subcatchments and Zijingguan catchment are output with various CN

Flood event simulations

values ranging from 10 to 100. 4. The ﬂood scaling is analyzed for all the simulated ﬂood

The Zijingguan catchment was divided into 11 subcatch-

events generated by the hypothetical rainfall events.

ments, and 17 outlets were selected with drainage areas

Then, using the relationship between the scaling expo-

ranging from 3 to 1,760 km2. Fourteen observed ﬂood

nent and rainfall intensity (total amount), the CN value

events were selected for HEC-HMS calibration and vali-

is obtained, and the existence of simple scaling or multi-

dation, including large ﬂoods (>5-year return period with

scaling is concluded.

a peak discharge greater than 500 m3/s) and small ﬂoods

5. If the ﬂood shows simple scaling, then the scaling exponent can be used in quantile-based ﬂood scaling.

(<5-year return period with a peak discharge smaller than 500 m3/s). The model parameters were adjusted artiﬁcially

6. If the ﬂood shows multiscaling, then the rainfall and the

to ﬁt the observed ﬂood hydrographs, and comparisons

corresponding ﬂood events are assumed to have the same

between the simulated and observed ﬂood events are listed

frequency. The rainfall with a speciﬁc frequency is

in Table 1 to estimate the model performance. Some simu-

obtained from rainfall frequency analysis, and the

lated and observed ﬂood events are shown in Figure 3.

Figure 2

Event-based ﬂood scaling exponent estimation framework.

J. Li et al.

1096

Table 1

A framework of event-based ﬂood scaling analysis

Hydrology Research

51.5

2020

Comparison between the simulated and observed ﬂood characteristics Flood events

Fpo (m3/s)

Fps (m3/s)

RE of Fp (%)

Fdo (mm)

Fds (mm)

RE of Fd (%)

NSE

Calibration

3 Aug 1956 7 Jul 1958 31 Jul 1959 6 Aug 1963 9 Aug 1964 14 Aug 1966 13 Aug 1973 28 Jul 1974

1490 905 648 4490 752.3 242 108 309

1537.4 922.9 697.4 3600 644 262.7 92.6 282.3

3.18 1.98 7.62 19.82 14.40 8.55 14.26 8.64

136.70 12.12 91.44 229.79 67.94 12.60 4.39 13.81

126.26 10.63 103.89 233.98 62.28 11.77 4.58 11.53

7.64 12.29 13.62 1.82 8.33 6.59 4.33 16.51

0.83 0.51 0.62 0.80 0.81 0.88 0.65 0.58

Validation

6 Aug 1975 25 Aug 1978 14 Aug 1979 30 Jul 1988 10 Aug 1988 28 Jul 1996

139 428 245 175 349.8 739.4

132.2 401.1 281.5 182 282.3 810.3

4.89 6.29 14.90 4.00 19.30 9.59

3.94 26.24 19.47 8.28 27.27 97.88

4.41 27.61 20.07 7.85 22.38 82.49

11.93 5.22 3.08 5.19 17.93 15.72

0.91 0.95 0.90 0.79 0.77 0.95

Note: Fp is the flood peak, Fpo is the observed flood peak, and Fps is the simulated flood peak. Fd is the flood depth, Fdo is the observed flood depth, and Fds is the simulated flood depth.

Figure 3

Some simulated and observed ﬂood events: (a) 3 August 1956; (b) 6 August 1963; (c) 25 August 1978; and (d) 28 July 1996.

J. Li et al.

1097

A framework of event-based ﬂood scaling analysis

Hydrology Research

51.5

2020

The RE values of the peak discharges for all 14 ﬂood

which is less than 1. These results reﬂect multiscaling, and

events are less than 20%, ranging from 1.98 to 19.82%,

large ﬂood events show greater scaling exponents and coefﬁ-

and the RE values of the ﬂood depths range from 13.62

cients of determination (Figures 4 and 5). The values of θ for

to 17.93%. The NSE coefﬁcients of the 14 ﬂood events are

large ﬂoods range from 0.7621 to 0.9234 with an average of

between 0.51 and 0.95 with an average of 0.77. Therefore,

0.8634, while those of small ﬂoods range from 0.5396 to

the calibrated model is capable of capturing the peak dis-

0.9922 with an average of 0.7225. Although the values of θ

charges, and the RE and NSE coefﬁcient have acceptable

for two small ﬂood events (occurring on 13 August 1973

accuracies. The simulated ﬂood hydrographs ﬁt the obser-

(730,813) and 6 August 1975 (750,806)) are relatively high,

vations very well during both the calibration and the

the R 2 is approximately 0.66, which is lower than that of

validation periods, which implies that the performance of

large ﬂoods. These results imply that there must be other fac-

the HEC-HMS model is sufﬁcient to simulate ﬂood events

tors, such as rainfall and land surface properties, inﬂuencing

in the Zijingguan catchment and can be used in the follow-

the statistics (Fang et al. ).

ing sections to analyze the ﬂood scaling in this area. Effects of rainfall and CN parameters on the scaling Event-based ﬂood scaling

exponent

According to the simulated ﬂood events in each subcatch-

Due to the large drainage area of the Zijingguan catchment,

ment, the scaling of the peak discharge for each ﬂood

rainfall is not spatially uniform. Figure 6 shows the spatial

event was analyzed with the subcatchment area as the

distributions of the two rainfall events. Each rainfall event

only explanatory variable. The scaling parameters for each

was distributed nonuniformly over the whole catchment,

ﬂood event are listed in Table 2. The power function ﬁt

especially the small rainfall events. This may be the reason

the plots very well with acceptable coefﬁcients of determi-

for the small values of the ﬁtted R 2 in the above section.

nation except for the flood event that occurred on 30 July 1988, verifying that flood scaling exists in this catchment. The scaling exponent θ varies for different flood events in

Flood scaling with areally averaged rainfall as the model input

the range from 0.5396 to 0.9922 with an average of 0.7829, To research the effects of the spatial rainfall distribution on Table 2

the ﬂood scaling properties, the areally averaged rainfall cal-

Scaling parameters for each ﬂood event

Flood event

3 Aug 1956

1.9212

0.9045

0.9071

7 Jul 1958

0.9144

0.7621

0.5166

31 Jul 1959

1.1848

0.8663

0.9804

6 Aug 1963

4.0131

0.9015

0.9646

9 Aug 1964

1.4038

0.8225

0.8916

14 Aug 1966

0.7012

0.7853

0.7529

13 Aug 1973

0.1041

0.8577

0.6613

28 Jul 1974

1.1296

0.671

0.6007

6 Aug 1975

0.0363

0.9922

0.6601

25 Aug 1978

1.8798

0.646

0.6364

14 Aug 1979

1.2422

0.6801

0.8649

30 Jul 1988

0.6129

0.5396

0.3288

10 Aug 1988

1.1759

0.6078

0.5921

28 Jul 1996

0.9474

0.9234

0.9358

culated by the Thiessen polygon method was input at each of the seven rain gauges. Then, the simulated flood peak discharge at each subcatchment was subjected to flood scaling analysis, as shown in Figure 7 and Table 3. The peak discharge simulated by uniform rainfall shows multiscaling as well. Compared with the fitted power function determined by the observed rainfall, all coefficients of determination increase, demonstrating stronger flood scaling (Figure 8) with the lowest R 2 of 0.6149. The scaling exponent also varies with an average of 0.7829, which is the same as the value under the condition of nonuniform spatial rainfall. For small flood events, subcatchment W150 has a relatively higher peak discharge, as shown in Figure 7 because of its higher calibrated CN value (CN ¼ 65). W150 is located at the lower part of the catchment, and thus, the initial soil moisture content may be higher

J. Li et al.

1098

Figure 4

A framework of event-based ﬂood scaling analysis

Hydrology Research

51.5

2020

Variations in θ and R 2 with the flood peak discharge (the dotted line is 500 m3/s, and the red line is the average of the small and large flood peaks, respectively). Please refer to the online version of this paper to see this figure in color: https://doi.org/10.2166/nh.2020.042|0|0|2020.

Figure 5

Scaling of the peak discharge with the catchment area fitted by a power function (the flood events of 1975 and 1979 represent small flood events, while those of 1963 and 1996 represent large flood events).

than that in the other subcatchments, leading to a higher

shows obvious scaling properties. This may be because the

runoff generation capacity and higher peak discharge. How-

subcatchments are unsaturated during small ﬂood events

ever, for large ﬂood events, the peak discharge of W150

but become saturated for large ﬂood events.

J. Li et al.

1099

A framework of event-based ﬂood scaling analysis

Figure 6

Spatial distributions of rainfall events over the Zijingguan catchment: (a) 6 August 1963 and (b) 14 August 1979.

Figure 7

Flood scaling for the simulated peak discharges with the areally averaged rainfall as input.

Hydrology Research

51.5

2020

J. Li et al.

1100

Table 3

A framework of event-based ﬂood scaling analysis

Hydrology Research

51.5

2020

different temporal distributions. Therefore, the effects of

Scaling parameters with the areally averaged rainfall as input

Flood events

3 Aug 1956

13.531

0.6041

0.9345

7 Jul 1958

1.1047

0.7705

0.8777

31 Jul 1959

0.9179

0.8912

0.9898

6 Aug 1963

3.4961

0.9425

0.9986

9 Aug 1964

1.2534

0.8373

0.9675

14 Aug 1966

0.5259

0.8045

0.9340

13 Aug 1973

0.0753

0.7235

0.9264

28 Jul 1974

1.1235

0.8207

0.9713

6 Aug 1975

0.0806

0.8248

0.8246

25 Aug 1978

0.8298

0.8484

0.9719

14 Aug 1979

1.2297

0.6834

0.9139

30 Jul 1988

0.324

0.5884

0.6149

10 Aug 1988

1.0227

0.7357

0.9138

28 Jul 1996

1.2509

0.8849

0.9848

the rainfall intensity on ﬂood scaling need to be studied with the same spatial and temporal distributions.

Effects of rainfall with the same temporal distribution and CN value on the scaling exponent The rainfall event that occurred on 6 August 1963 was selected as the typical event, based on which some hypothetical rainfall events were obtained by amplifying the typical rainfall event to drive the HEC-HMS model. The ﬂood hydrographs were simulated under different initial CN values. Then, a scaling analysis was conducted, as shown in Figure 10. The larger the maximum 1 h rainfall amount and CN value are, the larger the scaling exponent. When the CN value is sufﬁciently large (CN 70), the exponent becomes constant regardless of the rainfall intensity. For CN values less than 70, the scaling exponent increases

Relationships between the rainfall characteristics and the scaling exponent

with

increasing

maximum

rainfall

and

ﬁnally

approaches a constant. From this ﬁgure, if we determine the CN value and the design maximum 1 h rainfall, we

On the basis of the above 14 scaling exponents θ simulated

can obtain the scaling exponent and make ﬂood predictions.

by the areally averaged rainfall, the relationships between the maximum 1 h rainfall (P1) and θ and between the total

Linking the framework to the quantile-based scaling

rainfall (P) and θ were analyzed (Figure 9). The scaling expo-

exponent

nent has no obvious relation with P1 or P, which is uncommon. This may be because the rainfall intensity and

Based on the above ﬂood scaling analysis, ﬂood predictions

total rainfall amount from different rainfall events have

can

Figure 8

| θ and R 2 values under the conditions of spatially uniform and nonuniform rainfall.

conducted

implementing

hydrological

J. Li et al.

1101

Figure 9

A framework of event-based ﬂood scaling analysis

Hydrology Research

51.5

2020

Relationships (a) between the maximum 1 h rainfall (P1) and θ and (b) between the total rainfall amount (P) and θ.

DISCUSSION Flood scalings in ungauged basins should be analyzed by analyzing homogeneous regions with sufficient flood data. In watersheds possessing only one or a few hydrological stations, flood scaling analysis cannot be conducted by fitting the observed flood peak and watershed attributes. Therefore, hydrological modeling is an approach for performing scaling research by simulating the flood processes in each subwatershed. Figure 10

Relationship among the maximum 1 h rainfall (P1), CN, and θ.

During the ﬂood scaling analysis, the spatial rainfall distribution was found to be a controlling factor for the scaling exponent. Spatially uniform rainfall resulted in better ﬁtting

simulation according to the following steps. (1) A hydrologi-

of the simulated peak discharge and watershed area and

cal model is established ﬁrst and then calibrated and

resulted in a different scaling exponent. Hence, the differ-

validated. (2) A typical rainfall event with a high rainfall

ence between the scaling exponent obtained from ﬁtting

intensity and a large total amount is selected, and some

the observed rainfall and that obtained from ﬁtting the

hypothetical rainfall events are formed by amplifying the

areally averaged rainfall cannot be neglected. In practice,

typical rainfall event to keep the temporal distribution of

we use the scaling exponent obtained from ﬁtting the

the hypothetical rainfall events the same. (3) The hypotheti-

observed peak discharge. However, this may lead to a sig-

cal rainfall events are areally averaged and used to drive

niﬁcant error.

the hydrological model to simulate the peak discharge for

Similarly, land surface conditions signiﬁcantly inﬂu-

each subcatchment. (4) Flood scaling analysis is conducted

enced the scaling exponent. However, for different rainfall

by using the simulated peak discharge and the area of

events, the relationship between θ and the maximum 1 h

each subcatchment, and the relations among the scaling

or total rainfall is not good. Since all 14 ﬂood events were

exponent, rainfall and CN values are built, as shown in

simulated by using the same CN value, the temporal rainfall

Figure 10. (5) According to the assumption that the rainfall

distribution must have a signiﬁcant inﬂuence on θ. Furey &

and the corresponding ﬂood have the same frequency, from

Gupta () presented a variable to represent the effective

Figure 10, the scaling exponent of a speciﬁc quantile can be

rainfall duration and analyzed the relationship with the scal-

obtained by interpolation under any rainfall intensity and the

ing exponent. Therefore, the inﬂuence of the temporal

CN value.

rainfall distribution needs further research.

1102

J. Li et al.

A framework of event-based ﬂood scaling analysis

Hydrology Research

51.5

2020

In addition, the simulated peak discharge of each sub-

and the National Natural Science Foundation of China

catchment was used for ﬂood scaling analysis. Thus, the

(No. 51779165). We extend sincere thanks to the

result must be affected by the peak discharge simulation

Hydrology and Water Resource Survey Bureau of Hebei

error. In this catchment, the peak discharge simulation

Province for providing the research data.

errors ranged from 1.98 to 19.82%, with the largest simulation error obtained for the event that occurred on 6 August 1963. However, this ﬂood event was selected to illus-

DATA AVAILABILITY STATEMENT

trate the multiscaling analysis framework. If the result is applied to a data-scarce region, the simulation accuracy

Data cannot be made publicly available; readers should con-

should be improved further.

tact the corresponding author for details.

Gupta & Dawdy () noted that ﬂoods caused by rainstorms exhibited multiscaling, and the results of our study further verify their statements. However, our results were

REFERENCES

obtained in semi-humid and semi-arid regions with multiscaling, whereas in humid regions, the initial CN value is large, and simple scaling may be inferred from the ﬁndings in this study.

CONCLUSIONS In this study, we proposed a method to conduct flood scaling analysis by hydrological modeling in data-scarce regions. The scaling exponents were very different between using the observed rainfall and the areally averaged rainfall as model inputs, and the coefficient of determination was larger for spatially uniform rainfall. Therefore, the spatial rainfall distribution has a significant impact on flood scaling. Based on how the spatial rainfall distribution and CN values influence the scaling exponent, a framework was proposed to estimate the scaling exponent in a data-scarce region. We recommend that the areally averaged rainfall of an event be used as the input to the hydrological model; accordingly, the correlation curves relating the scaling exponent, rainfall intensity, and CN values can be drawn. As long as the flood frequency and CN value are determined, the scaling exponent can be obtained from these correlation curves.

ACKNOWLEDGEMENTS This research is supported by the National Key Research and Development Program of China (2018YFC0407902)

A, Y., Wang, G., Liu, T., Xue, B. & Kuczera, G.  Spatial variation of correlations between vertical soil water and evapotranspiration and their controlling factors in a semiarid region. Journal of Hydrology 574, 53–63. Alfy, M. E.  Assessing the impact of arid area urbanization on flash floods using GIS, remote sensing, and HEC-HMS rainfall-runoff modeling. Hydrology Research 47, 1142–1160. Al-Rawas, G. A. & Valeo, C.  Relationship between wadi drainage characteristics and peak-flood flows in arid northern Oman. Hydrological Sciences Journal 55 (3), 377–393. Ayalew, T. B., Krajewski, W. F., Mantilla, R. & Small, S. J.  Exploring the effects of hillslope-channel link dynamics and excess rainfall properties on the scaling structure of peakdischarge. Advances in Water Resources 64, 9–20. Ayalew, T. B., Krajewski, W. F. & Mantilla, R.  Analyzing the effects of excess rainfall properties on the scaling structure of peak discharges: insights from a mesoscale river basin. Water Resources Research 51 (6), 3900–3921. Eash, D. A.  Techniques for estimating flood-frequency discharges for streams in Iowa. Center for Integrated Data Analytics Wisconsin Science Center. Fang, Q., Wang, G., Xue, B., Liu, T. & Kiem, A.  How and to what extent does precipitation on multi-temporal scales and soil moisture at different depths determine carbon flux responses in a water-limited grassland ecosystem? Science of the Total Environment 635, 1255–1266. Farmer, W. H., Over, T. M. & Vogel, R. M.  Multiple regression and inverse moments improve the characterization of the spatial scaling behavior of daily streamflows in the Southeast United States. Water Resources Research 51 (3), 1775–1796. Furey, P. R. & Gupta, V. K.  Effects of excess rainfall on the temporal variability of observed peak-discharge power laws. Advances in Water Resources 28 (11), 1240–1253. Furey, P. R. & Gupta, V. K.  Diagnosing peak-discharge power laws observed in rainfall–runoff events in Goodwin Creek experimental watershed. Advances in Water Resources 30 (11), 2387–2399.

1103

J. Li et al.

A framework of event-based ﬂood scaling analysis

Furey, P. R., Troutman, B. M., Gupta, V. K. & Krajewski, W. F.  Connecting event-based scaling of flood peaks to regional flood frequency relationships. Journal of Hydrologic Engineering 21 (10), 04016037. Galster, J. C., Pazzaglia, F. J., Hargreaves, B. R., Morris, D. P., Peters, S. C. & Weisman, R. N.  Land use effects on watershed hydrology: the scaling of discharge with drainage area. Geology 34 (9), 713–716. Gupta, V. K. & Dawdy, D. R.  Physical interpretations of regional variations in the scaling exponents of flood quantiles. Hydrological Processes 9 (3–4), 347–361. Gupta, V. K., Castro, S. L. & Over, T. M.  On scaling exponents of spatial peak flows from rainfall and river network geometry. Journal of Hydrology 187 (1), 81–104. Gupta, V. K., Troutman, B. M. & Dawdy, D. R.  Towards a nonlinear geophysical theory of floods in river networks: an overview of 20 years of progress. Nonlinear Dynamics in Geosciences 121–151. Gupta, V. K., Ayalew, T. B., Mantilla, R. & Krajewski, W. F.  Classical and generalized Horton laws for peak flows in rainfall-runoff events. Chaos 25 (7), 075408. Han, S. J., Wang, S. L., Xu, D. & Zhang, Q.  Scale effects of storm-runoff processes in agricultural areas in Huaibei Plain. Transactions of the Chinese Society of Agricultural Engineering 28 (8), 32–37. Han, D., Wang, G., Liu, T., Xue, B., Kuczera, G. & Xu, X.  Hydroclimatic response of evapotranspiration partitioning to prolonged droughts in semiarid grassland. Journal of Hydrology 563, 766–777. Ishak, E., Haddad, K., Zaman, M. & Rahman, A.  Scaling property of regional floods in New South Wales Australia. Natural Hazards 58 (3), 1155–1167. Jothityangkoon, C. & Sivapalan, M.  Temporal scales of rainfall–runoff processes and spatial scaling of flood peaks: space–time connection through catchment water balance. Advances in Water Resources 24 (9–10), 1015–1036.

Hydrology Research

51.5

2020

Kroll, C. N., Rapant, D. F. & Vogel, R. M.  Prediction of hydrologic statistics in nested watersheds across the United States. In: World Environmental and Water Resources Congress, ASCE, Baltimore, MD, pp. 2326–2335. Lee, K. T. & Huang, J. K.  Influence of storm magnitude and watershed size on runoff nonlinearity. Journal of Earth System Science 125 (4), 777–794. Liu, S. Y., Huang, S. Z., Xie, Y. Y., Wang, H., Leng, G. Y., Huang, Q., Wei, X. T. & Wang, L.  Identification of the nonstationarity of floods: changing patterns, causes, and implications. Water Resources Management 33 (3), 939–953. Mantilla, R.  Physical Basis of Statistical Scaling in Peak Flows and Stream Flow Hydrographs for Topologic and Spatially Embedded Random Self-Similar Channel Networks. Dissertations and Theses – Gradworks. Medhi, H. & Tripathi, S.  On identifying relationships between the flood scaling exponent and basin attributes. Chaos 25, 075405. Menabde, M. & Sivapalan, M.  Linking space–time variability of river runoff and rainfall fields: a dynamic approach. Advances in Water Resources 24 (9), 1001–1014. Ogden, F. L. & Dawdy, D. R.  Peak discharge scaling in small Hortonian watershed. Journal of Hydrologic Engineering 8 (2), 64–73. Venkata, P. M.  Role of Rainfall Variability in the Statistical Structure of Peak Flows. Dissertations and Theses – Gradworks. Wang, L. & Sun, W. J.  Research on HEC-HMS and Vflo rainfall characteristics simulation and comparative based on DEM data: a case study of Miyun District, Beijing. Acta Scientiae Circumstantiae 39, 3559–3564. Yang, W. C., Yang, H. B. & Yang, D. W.  Classifying floods by quantifying driver contributions in the Eastern Monsoon Region of China. Journal of Hydrology 585, 124767. Yue, S. & Gan, T. Y.  Scaling properties of Canadian flood flows. Hydrological Processes 23 (2), 245–258.

First received 27 March 2020; accepted in revised form 6 August 2020. Available online 11 September 2020

1104

51.5

2020

Response of redox zonation to recharge in a riverbank ﬁltration system: a case study of the Second Songhua river, NE China Xiaosi Su, Yaoxuan Chen, Hang Lyu, Yakun Shi, Yuyu Wan and Yiwu Zhang

ABSTRACT Bank filtration induced by groundwater pumping results in redox zonation along the groundwater flow path. Besides the river water, recharge from other sources can change local redox conditions; therefore, redox zonation is likely to be complex within the riverbank filtration (RBF) system. In this study, hydrodynamics, hydrogeochemistry, and environmental stable isotopes were combined together to identify the redox conditions at an RBF site. The recharge characteristics and redox processes were revealed by monitoring the variations of water level, δ 2H and δ 18O, and redox indexes along shallow and deep flow paths. The results show that local groundwater is recharged from river, regional groundwater, and precipitation. The responses of redox zonation are sensitive to 2 different sources. In the river water recharge zone near shore, O2, NO 3 , Mn(IV), Fe(III), and SO4 are

reduced in sequence, the ranges of each reaction are wider in deep groundwater because of the high-velocity deep ﬂow. In the precipitation vertical recharge zone, precipitation intermittently drives O2, NO 3 , and organic carbon to migrate through vadose zone, thereby decreasing the groundwater reducibility. In the regional groundwater lateral recharge zone in the depression cone, the reductive regional groundwater is continuously recharging local groundwater, leading to the cyclic reduction of Mn(IV) and Fe(III). Key words

| hydrodynamics, hydrogeochemistry, recharge characteristics, redox zonation, riverbank ﬁltration system, stable isotope

HIGHLIGHTS

• • •

Hydrodynamics, hydrogeochemistry, and environmental stable isotopes were combined to explore the redox zonation in an RBF system. Redox zones and sequential redox processes involved were identiﬁed and partitioned. Analysis of the redox zonation response to ﬂow regime and recharge of river, precipitation, and regional groundwater.

Xiaosi Su Yaoxuan Chen Yiwu Zhang Institute of Water Resources and Environment, Jilin University, Changchun 130026, China Xiaosi Su College of Construction Engineering, Jilin University, Changchun 130021, China Yaoxuan Chen Yiwu Zhang College of New Energy and Environment, Jilin University, Changchun 130026, China Yaoxuan Chen Hang Lyu (corresponding author) Yuyu Wan Key Laboratory of Groundwater Resources and Environment (Jilin University), Ministry of Education, Changchun 130026, China E-mail: lvhangmail@163.com Hang Lyu Yuyu Wan Jilin Provincial Key Laboratory of Water Resources and Environment, Jilin University, Changchun 130026, China Yakun Shi No.1 Institute of Geo-environment Survey of Henan, Zhengzhou 450000, China

1105

X. Su et al.

Response of redox zonation to recharge in a riverbank ﬁltration system

Hydrology Research

51.5

2020

INTRODUCTION Riverbank ﬁltration (RBF) involves water extraction by

2þ 2þ concentrations of NHþ at 0.98– 4 , Mn , and Fe

pumping wells at riverside ﬁelds to enhance water supply

3.62 N mg/L,

by stimulating river recharge to groundwater. This can

respectively,

attenuate or degrade pollutants, such as suspended solids,

0.50 N mg/L, 0.40 mg/L, and 2.00 mg/L, respectively. (WHO

inorganic or organic substances, poisonous heavy metals,

). Therefore, an appraisal of the characteristics of the

pathogenic viruses, and bacteria; hence, it is considered an

redox environment and biogeochemical behavior of the sensi-

efﬁcient and natural treatment technology for water quality

tive components during bank ﬁltration under different sources

improvement (Hiscock & Grischek ; Tufenkji et al.

of recharge should be undertaken to gain a deeper understand-

; Trauth et al. ; Muz et al. ). Rivers are rich in

ing of the complex biogeochemical processes.

4.71–6.82 mg/L,

and

8.54–12.74 mg/L,

which exceed the WHO guideline limits of

O2 and organic matter; driven by Gibbs free energy, the

This study aims to: (1) carry out survey and analysis of

organic matter supplies electrons to the lowest unoccupied

the hydrodynamics of the river and groundwater around

molecular orbital locations during river ﬁltration; therefore,

the Kaladian well ﬁeld by combining δ 2H and δ 18O

2 O2, NO are reduced in 3 , Mn(IV), Fe(III), and SO4

analyses to identify the spatial distribution characteristics

sequence (Kedziorek & Bourg ; Farnsworth & Hering

of recharge; (2) survey the spatial distribution of the redox

). However, this process can also cause incomplete

environment, sensitive components in river water and

removal of organic carbon, suspended solids, and inorganic

groundwater, and the soluble N and C components in the

pollutants, and release of heavy metals, such as Fe, Mn, and

aquifer medium, as well as analyze their relationships with

As, from sediment or the aquifer medium to groundwater

hydrodynamic conditions; and (3) partition the redox zona-

(Gandy et al. ; Su et al. ).

tion along the groundwater ﬂow path based on the spatial

Bank ﬁltration controlled by groundwater pumping results in a relatively stable redox zonation along the

distribution of recharge and redox conditions, as well as clarify the hydrogeochemical processes within each zone.

groundwater ﬂow path and previous research has mostly focused on the spatial distribution of the redox environment (Greskowiak et al. ; Kumar & Riyazuddin ). Many

MATERIALS AND METHODS

factors affecting the formation of redox zonation, such as ﬂuctuations in river stage and groundwater level, intensity

Study area

of pumping, lithology and structure of aquifer medium, and water temperature, have all been considered in previous

The Kaladian riverside well ﬁeld, located on the alluvial–

studies, yet the inﬂuence of recharge conditions of local

proluvial plain of the Second Songhua River in northeastern

groundwater, such as water from precipitation and regional

China, is characterized by ﬂat topography with elevations

groundwater conditions have been neglected (Burt et al.

varying from 124 to 129 m above sea level. The unconﬁned

; Massmann et al. ; Kohfahl et al. ; Wang

aquifer is mainly composed of sand with the thickness of

et al. ). In addition to river water, groundwater recharge

17–20 m, and a continuous and stable clay-laminated layer

from other sources can also change the local redox con-

of ∼24 m thickness forms the impervious base of the uncon-

ditions, and hence, redox zonation is likely to be more

ﬁned aquifer. The upper and lower parts of the unconﬁned

complex within the RBF system (Kedziorek & Bourg ;

aquifer consist of ﬁne sand and medium sand with a thin

Buzek et al. ).

layer of silty-clay in the middle, respectively. The long-term

The Kaladian well ﬁeld, located in northeastern China,

pumping wells (C1–C8) operating intermittently at a total

is characterized by regional groundwater rich in Fe and

pumping rate of 6,000–10,000 m3/day have formed a

Mn with Mn

2þ

and Fe

contents of over 6 and 10 mg/L,

respectively. The local groundwater in recent 3 years has

stable groundwater depression cone centered around well C5 and C6.

X. Su et al.

1106

Response of redox zonation to recharge in a riverbank ﬁltration system

Hydrology Research

51.5

2020

The well ﬁeld is located in a cold temperate zone that

NS1, NS2, NS3, NS4, NS5, and NS6 were established at

experiences a continental semi-arid monsoon climate with a

0, 2, 5, 8, 14, and 30 m from the shore, respectively. Further-

dry and windy spring, hot and rainy summer, cool autumn

more, six far-shore monitoring points FS1, FS2, FS3, FS4,

with early frost, and a cold and long winter. Monthly average

FS5, and FS6 were located at 80, 200, 420, 700, 850, and

temperatures in January and July can reach 17.5 and

1,110 m from the shore. Each monitoring point has a shal-

23.3 C, respectively, while the annual average temperature

low (7.5 m) and deep (14 m) well with a screen interval of

is 4.7 C. Annual mean precipitation is 425.7 mm, of which

6–7.5 and 12.5–14 m, respectively. Furthermore, two

more than 70% is concentrated from June to August, while

regional groundwater monitoring wells, RG1 and RG2, are

the annual mean evaporation is 1,689.8 mm, with the maxi-

located at 2,100 and 2,700 m, respectively, from the shore.

mum in May at 316.5 mm. The Second Songhua River, ﬂowing from southeast to northwest, is controlled by climate

Sampling

and upstream water conservancy projects and has an annual average runoff, river surface width, and depth of 476.0 m3/s,

River and groundwater sampling

400–450 m, and 3–7 m, respectively. Herein, 22 long-term monitoring wells with a depth of

River water and groundwater samples were collected in Sep-

6–9 m were constructed to monitor the groundwater level

tember 2018. River water was collected 0.5 m below the

and quality since December 2015. Based on the monitoring

surface. Groundwater was pumped at a rate <5 L/min from

results, a dense monitoring section from the river to the

the wells, and the samples were collected after ﬂushing the

depression cone center was constructed in August 2018, as

volume of the well pipe three times. Sampling bottles were

shown in Figure 1. The hydrodynamic and environmental

ﬁlled to the brim and then sealed. Samples for SO2 4 , NO3 ,

conditions are susceptible to change within 30 m of the

and NHþ 4 were stored in polyethylene bottles, while concen-

shore; therefore, six near-shore monitoring points NS1,

trated sulfuric acid was added to the samples for NHþ 4 to

Figure 1

Location of Kaladian riverside well ﬁeld and distribution of monitoring wells.

1107

X. Su et al.

Response of redox zonation to recharge in a riverbank ﬁltration system

generate a pH <2. Samples for Mn2þ were packed in brown

Hydrology Research

51.5

2020

In situ permeability test for riverbed sediment

glass bottles with diluted hydrochloric acid to generate a pH <2. Samples for DOC were stored in brown glass bottles and

The standpipe falling head test method, also known as the

concentrated sulfuric acid was added to generate a pH <2.

tube test, was used to determine the permeability of the riv-

Samples for testing stable isotopes were collected in

erbed sediments; four test points R1, R2, R3, and R4 were

500 mL polyethylene bottles and stored at 4 C.

arranged, respectively, at 5, 30, 150, and 300 m from point

A Hach HQ40d portable meter (Hach Company, USA)

NS1, as shown in Figure 1. The length of the standpipe

was used to test pH, Eh, and dissolved oxygen (DO) content

was 1.50 or 2.80 m for different river depths, with an inner

2þ

in water, while a Hach DR1900 was used to test Fe

and

diameter of 0.04 m. After being inserted into the depth Lv

on site. An atomic absorption analyzer (Scientiﬁc

of the riverbed sediment, the water head change process in

iCE 3300, Thermo, Germany) was used to analyze Mn2þ

the tube was recorded after adding water from the upper

content, and an ion chromatograph (881 Compact IC pro,

end, and the calculation was carried out by Equation (1)

Metrohm, Switzerland) was used to test SO2 content at 4

(Chen ),

the Institute of Water Resources and Environment, Jilin University (Changchun, China). A continuous ﬂow analyzer (San þ þ, Skalar, Netherlands) was used to measure NO 3

Lν h1 Kν ¼ , ln t2 t1 h2

(1)

and NHþ 4 contents, and a TOC analyzer (TOC-L CPH Shimadzu, Japan) was used to test DOC at the Northeast

where Kv is the vertical hydraulic conductivity of riverbed

Institute of Geography and Agroecology, Chinese Academy

sediments, m/day; Lv is the length of the riverbed sediments

of Sciences (Changchun, China). The δ 2H and δ 18O values

in the test tube, m; t1 and t2 are the instantaneous moments

were tested using a laser isotope meter (L2140-i, Picarro,

at the beginning and end of the test days; and h1 and h2

USA) at the Third Institute of Oceanography, Ministry of

correspond to the water level in the test tube, m.

Natural Resources (Xiamen, China), with accuracies of ±0.50 and ±0.20‰, respectively.

RESULTS Aquifer medium sampling River stage and groundwater level dynamics Sample sites of the vadose zone and aquifer medium were located in the vicinity of point NS5, FS1, and FS3 and

The river stage and groundwater level dynamics from Janu-

samples were collected using percussion drills in September

ary 2016 to October 2018 are shown in Figure 2. Annual

2018. The collection horizons were 0–0.5, 1.0–1.5, 2.0–2.5,

variations in the river stage ranged between 1.0 and 3.2 m,

and 6.5–7.5 m below the land surface. Samples were

and the river stage was high during July–September and rela-

stored in 500 mL glass bottles and were immediately trans-

tively low during October–March of the following year. The

ported to the laboratory at 20 C for testing total

surface of the river was frozen from December–March of the

nitrogen (TN), soil organic carbon (SOC), and ion exchange

following year with the frozen layer thickness being 1 m.

þ þ forms of NO 3 /NH4 (IEF-NO3 /NH4 ).

The trend of regional groundwater level changes was con-

TN content was tested at 120–124 C in an alkaline sub-

sistent with that of the river stage, and the annual change

strate using K2S2O8 to oxidize all forms of nitrogen in the

was generally less than 1.5 m. The local groundwater level

medium to NO 3 , then tested using a continuous ﬂow analy-

of the well ﬁeld was lower than the river level throughout

zer. The loss-on-ignition method was used to obtain the

the year, indicating year-round river ﬁltration.

organic matter content, and then by using the Van Bemme-

In point NS1–NS6 near the shore, the groundwater level

len factor of 1.724, was converted to the SOC content. The

could quickly respond to the change in the river level, show-

þ IEF-NO 3 /NH4 in the 2 mol/L KCl extraction medium was

ing an annual variation of 2.8–3.0 m. In point FS1–FS4 far

tested using a continuous ﬂow analyzer.

from shore, the annual variation in the groundwater level

X. Su et al.

1108

Figure 2

Response of redox zonation to recharge in a riverbank ﬁltration system

Hydrology Research

51.5

2020

Water level dynamics of river and groundwater.

was low, at 0.9–1.8 m. During winter when the pumping

values at the RFZ are closer to those of the river end mem-

intensity was high, variations in the water levels in point

bers, along the direction of the groundwater runoff,

FS3 and FS4, located within the depression cone, were rela-

whereas the groundwater δ 2H and δ 18O values at the GRZ

tively large, indicating that the far-shore groundwater level

and DCZ increase gradually, and are close to the values of

was less affected by the ﬂuctuations in the river level and

the regional groundwater end member. In the vertical dimen-

was instead mainly affected by pumping.

sion, the shallow and deep groundwater δ 2H and δ 18O values

According to the above-mentioned water level dynamics

show no obvious difference in the RFZ and GRZ within

of the river and groundwater, the NS1–FS4 monitoring sec-

200 m from the shore, while the shallow groundwater δ 2H

tion was divided into three hydrodynamic zones as follows:

and δ 18O values are obviously closer to those of the regional

(1) river ﬁltration zone (RFZ) (River-NS6); (2) groundwater

groundwater end member than deep groundwater in the

runoff zone (GRZ) (NS6–FS3); and (3) depression cone

DCZ, which is 420–700 m from the shore.

zone (DCZ) (FS3–FS4). Spatial distribution characteristics of redox-sensitive 2

Spatial distribution of δ H and δ O in river water and

indexes

groundwater Redox indexes in river water and shallow groundwater River water and regional groundwater are two stable lateral recharge sources, and two major end members of the ground-

Eh/DO. The Eh values and DO contents of the river water

water in the well ﬁeld. The Second Songhua River originates

were high at 152.25 mV and 8.205 mg/L, respectively, as

in the Changbai Mountains (elevation 2,750 m), and the

shown in Figure 4(a) and 4(b), indicating an oxidizing environ-

elevation difference with respect to the well ﬁeld is more

ment. During river ﬁltration, shallow groundwater Eh and DO

than 2,000 m, leading to lower δ 2H and δ 18O values in the

at the RFZ rapidly dropped to below 150.00 mV and

river water. The western part of Jilin, where the well ﬁeld fea-

1.0 mg/L, respectively, gradually making the environment in

tures low terrain and shallow groundwater depth, is easily

this zone reducing in nature. Regional groundwater Eh and

affected by evaporation; hence, the δ H and δ O values of

DO were 151.62 mV and 0.68 mg/L, respectively, indicat-

the regional groundwater are relatively high.

ing a reducing environment. Average shallow groundwater

In the NS1–FS4 monitoring section, the distribution 2

Eh and DO at the GRZ and DCZ were 126.25 mV and

ranges and average of δ H and δ O values for shallow and

1.18 mg/L, respectively, which were higher than those of

deep groundwater lie between those for the river water and

the shallow groundwater at the RFZ and regional ground-

regional groundwater, as shown in Figure 3(a) and 3(b). In

water, with a tendency of increasing along the direction of

the horizontal dimension, the groundwater δ 2H and δ 18O

ﬂow until reaching their highest values at the DCZ.

X. Su et al.

1109

Response of redox zonation to recharge in a riverbank ﬁltration system

Hydrology Research

Figure 3

Spatial distribution of δ 2H and δ 18O values in river water and groundwater along the groundwater ﬂow path in the monitoring section.

Figure 4

Spatial distribution of environmental and hydrochemical indexes in shallow groundwater along the groundwater ﬂow path in the monitoring section.

51.5

2020

1110

X. Su et al.

Response of redox zonation to recharge in a riverbank ﬁltration system

Hydrology Research

51.5

2020

pH. The pH value of the river water was 7.72 as shown in

Meanwhile, Mn2þ and Fe2þ at the DCZ were closer to

Figure 4(c). During river ﬁltration, pH continuously

that of the regional groundwater.

decreased to 7.28 in well NS1-1, then slowly increased along the direction of ﬂow, while the average shallow

2 SO2 4 . The SO4 content in the river water was 29.72 mg/L

groundwater pH in the RFZ was 7.64. Average regional

as shown in Figure 4(f). Average shallow groundwater SO2 4

groundwater pH was 7.18. Average shallow groundwater

in well NS1-1–NS4-1 that is within 14 m from shore was

pH at both the GRZ and DCZ was 7.31, which is between

29.98 mg/L without obvious change. In well NS5-1, which

the shallow groundwater at the RFZ and regional ground-

is 30 m from shore, SO2 4 began to decrease, reaching its

water values, with a tendency of decreasing along the

lowest value of 8.97 mg/L in well FS1-1, which is 200 m

direction of ﬂow; the lowest pH was recorded in the DCZ.

from shore at the GRZ. Average regional groundwater

þ þ NO 3 /NH4 . The contents of NO3 and NH4 in river water

2 SO2 4 was 58.36 mg/L. Average shallow groundwater SO4

at the DCZ was 62.29 mg/L, which is higher than the shallow groundwater at the RFZ, GRZ, and regional ground-

were 2.274 and 0.124 N mg/L as shown in Figure 4(d) and

water. Meanwhile, the content of HS was relatively low at

4(e), respectively. During river ﬁltration, the shallow ground-

all monitoring wells and was maintained below the detec-

þ water NO 3 decreased rapidly, while NH4 increased rapidly,

until in well NS1-1 at the shore, where NO 3 then dropped to 0.762 N mg/L, having been consumed up to approximately NHþ 4

tion limit. DOC. The DOC content of the river water was 23.14 C mg/L

increased to 0.613 N mg/L, a 4-fold

as shown in Figure 4(i). During river ﬁltration, at the RFZ and

increase, until in well NS4-1 that is 8 m from shore, where

GRZ within 80 m from the shore, average shallow ground-

70%. Meanwhile,

þ NO 3 dropped below the detection limit, while NH4 was

NO 3

water DOC of well NS1-1–FS1-1 gradually decreased along

was very low,

the direction of ﬂow with a value of 9.33 C mg/L on average,

mostly below the detection limit, while average NHþ 4 was

which is approximately 60% less than that of the river water.

relatively high at 1.881 N mg/L. In the shallow groundwater

The average regional groundwater DOC was 16.38 C mg/L.

0.865 N mg/L. Regional groundwater

of the GRZ and the DCZ, average

NO 3

was 0.759 N mg/L,

At the GRZ and DCZ within 80–700 m from the shore, the

which was higher than the shallow groundwater at the RFZ

average shallow groundwater DOC of well FS1-1–FS4-1

and regional groundwater, while average NHþ 4 was 1.188 N

was 17.75 C mg/L, which is higher than that of the shallow

mg/L, which was between that of the shallow groundwater

groundwater at the RFZ and regional groundwater.

at RFZ and regional groundwater. Redox indexes in deep groundwater Mn2þ/Fe2þ. The contents of Mn2þ and Fe2þ in the river

Eh/DO. The Eh values and DO contents of deep ground-

water were extremely low as shown in Figure 4(g) and

water were relatively stable as shown in Figure 5(a) and

4(h). During river ﬁltration, shallow groundwater Mn2þ and Fe2þ began to increase until reaching well NS1-1; with Mn2þ peaking at 1.12 mg/L in well NS3-1, which is 5 m from the shore. Fe2þ peaked at 3.42 mg/L in well NS6-1, which is 30 m from the shore. Both Mn2þ and Fe2þ began to decrease after peaking until they dropped to a minimum value in well FS1-1, which is 80 m from the shore. Average regional groundwater Mn2þ and Fe2þ were 5.98

5(b), respectively. Average deep groundwater Eh and DO in well NS1-2–NS6-2 that are within 30 m from shore at the RFZ were 125.32 mV and 0.63 mg/L respectively, indicating a reducing environment. Deep groundwater Eh and DO in well FS1-2–FS4-2 that are 80–700 m from shore at the GRZ and DCZ were 129.49 mV and 0.74 mg/L, respectively, and there was no signiﬁcant change when compared with the deep groundwater at the RFZ.

and 12.21 mg/L, respectively. Average shallow groundwater Mn2þ and Fe2þ at the GRZ and DCZ were 3.79 and

pH. The pH value was relatively stable in deep groundwater

6.38 mg/L, respectively, ranging between the shallow

as shown in Figure 5(c) along the direction of ﬂow. Average

groundwater at the RFZ and the regional groundwater.

deep groundwater pH in well NS1-2–FS2-2 within 420 m

X. Su et al.

1111

Figure 5

Response of redox zonation to recharge in a riverbank ﬁltration system

Hydrology Research

51.5

2020

Spatial distribution of environment and hydrochemistry indexes for deep groundwater along the groundwater ﬂow path in the monitoring section.

from shore at the RFZ and GRZ was 7.39. The average deep

mg/L, respectively, which are higher than those of deep

groundwater pH in well FS3-2–FS4-2 at the DCZ was 7.21,

groundwater at the RFZ and regional groundwater.

which was closer to that of the regional groundwater. Mn2þ/Fe2þ. As shown in Figure 5(g) and 5(h), the contents þ NO 3 /NH4 .

Compared with the shallow groundwater, the

of Mn2þ and Fe2þ in deep groundwater at the RFZ and GRZ

deep groundwater had lower NO 3 content and higher

began to increase in well NS1-2, with Mn2þ in well NS6-2,

NHþ 4

content as shown in Figure 5(d) and 5(e), respectively.

which is 30 m away from the shore, peaking at 1.18 mg/L,

þ Average deep groundwater NO 3 and NH4 in well NS1-2–

and average Fe2þ in well FS1-2 and FS2-2, which are 80–

NS6-2, which are within 30 m from shore at the RFZ,

200 m from shore peaking at 3.76 mg/L. After peaking,

were 0.394 and 2.081 N mg/L, respectively. Average NO 3

Mn2þ began to decrease in well FS1-2, which is 80 m from

NHþ 4

in well FS1-2–FS4-2, which are 80–700 m

shore, while Fe2þ did not show a signiﬁcant decline, and

from shore at the GRZ and DCZ, were 0.452 and 2.230 N

both average Mn2þ and Fe2þ gradually increased in well

and

X. Su et al.

1112

Response of redox zonation to recharge in a riverbank ﬁltration system

Hydrology Research

51.5

FS2-2–FS3-2, which are at the junction of the GRZ and DCZ

Spatial distribution characteristics of carbon and

with values of 3.79 and 6.18 mg/L, respectively. The Mn2þ

nitrogen in aquifer

2þ

and Fe

2020

contents were closer to those in the regional

groundwater than to the water near the depression cone

Vertical distributions of N and C contents in the vadose zone

center.

and aquifer medium are shown in Figure 6. The points NS5, FS2, and FS4 are 14, 200, and 700 m from the shore and are

SO2 4 .

The

SO2 4

content in deep groundwater of well NS1-

located in the RFZ, GRZ, and DCZ, respectively.

2–FS1-2, which are within 80 m from shore at the RFZ and

In the vertical proﬁle, the surface medium of the vadose

GRZ, remained relatively stable with an average value of

zone of each hole had a high content of ﬁne particles, as the

24.53 mg/L as seen in Figure 5(f). Deep groundwater

surface is mostly covered by trees and crops; the surface

SO2 4 in well FS1-2 and FS2-2, which are 80–200 m from

medium forms an important growth and metabolism area

shore, decreased for a short interval, and then continued

for plants and microorganisms where organic life and

to increase along the direction of ﬂow to reach 32.43 mg/

humus are abundant, as well as fertilizers. The average con-

L in well FS3-2 and FS4-2, which are 420 and 700 m from

tents of TN and SOC were 0.262 N and 6.273 C g/kg,

shore, respectively. Concentrations of HS were the same

respectively, and the average C:N value was 14.217, indicat-

as those for shallow groundwater, with relatively low

ing that nitrogen assimilation and mineralization might be

concentrations.

prevalent (Dhondt et al. ; Evans et al. ). The content of coarse particles in the lower part of the vadose zone and

DOC. The content of DOC in deep groundwater was higher

aquifer medium increased, while contents of TN and SOC

than that in the shallow groundwater as shown in Figure 5(i).

were lower with an average of 0.144 N g/kg and 1.495 C g/

In deep groundwater, average DOC in well NS1-2–NS6-2,

kg, respectively, and the average C:N ratio was 8.186.

which is within 30 m from shore at the RFZ, was 24.75 C

Although this C:N value indicates that ammoniﬁcation

mg/L. Average DOC in well FS1-2–FS4-2, which are 80–

could occur, the organic nitrogen content within the substrate

700 m from shore at the GRZ and DCZ, was 25.85 C mg/L,

was limited. Meanwhile, the spatial distribution character-

which is higher than the value recorded for the deep ground-

istics of IEF-NHþ 4 and IEF-NO3 in the medium were

water at the RFZ and the regional groundwater; DOC at the

similar to those of TN and SOC, with a tendency to gradually

GRZ increased signiﬁcantly.

decrease.

Figure 6

Vertical distribution of nitrogen and carbon in the vadose zone medium at point NS5, FS2, and FS4.

X. Su et al.

1113

Response of redox zonation to recharge in a riverbank ﬁltration system

Hydrology Research

51.5

2020

In the horizontal direction, the lithological composition

sand with a hydraulic conductivity (K ) of 28.47 m/day,

and structure of pores in point NS5 and FS4 were similar, as

while the lower aquifer consisted of medium sand with a

well as the nitrogen and carbon indicators. In point FS2, the

K of 40.38 m/day.

content of coarse particles in the vadose zone was signiﬁ-

The heterogeneity of the water-bearing medium affected

cantly higher than those in point NS5 and FS4, indicating

the different hydraulic exchange characteristics between the

that the permeability of the vadose zone of point FS2 is rela-

river and groundwater, showing two ﬂow paths from the

tively stronger. Meanwhile, the contents of TN, SOC,

river to the center of the depression cone, as shown in

IEF-NHþ 4 , and IEF-NO3 in the medium of each layer in

Figure 7. In shallow riverbed with lower permeability, ﬂow

point FS2 were also lower than those of point NS5 and FS4.

occurred at a shallow depth (9 m) with lower velocity and was lateral. In deeper riverbed with higher permeability,

DISCUSSION Hydrodynamic exchange between river and

ﬂow occurred at a depth greater than 9 m (reaching 17– 19 m at the bottom of the aquifer), had a higher velocity, and was vertical to the lateral ﬂow.

groundwater Recharge of local groundwater Flow path along the direction of ﬁltration and runoff To more precisely clarify the recharge conditions along the The hydraulic gradient between the river and groundwater

direction of the two ﬂows, the hydrogen and oxygen stable

acts as a driving force, while gravity ﬂow induces ground-

isotopes of the river water and groundwater were calculated

water runoff from the river to the pumping wells (Su et al.

and analyzed based on mass conservation. Relative contri-

). As Figure 7 shows, the riverbed sediments near

butions of the river water and regional groundwater could

point NS1 were relatively denser and continuous, compris-

be estimated using an end member mixing model which

ing of silty-clay mixed with ﬁne sand with a thickness of

can be determined using Equation (2),

0.05–0.3 m, while for tube test points R1 and R2 located at this district, the results of Kv varied from 4.82 to 6.39 m/ day. Riverbed sediments in the middle of the river contained

δ 1 n1 þ δ 2 n2 ¼ δ(n1 þ n2 ) n1 þ n2 ¼ 1

(2)

ﬁne sand with low horizontal continuity and stratiﬁcation, and the tube test points R3 and R4 showed that Kv varied

where δ1, δ2, and δ are the isotope values for river water,

from 46.11 to 59.37 m/day. Otherwise, the vertical differ-

regional groundwater, and local groundwater, respectively;

ence in lithology and permeability of the upper and lower

and n1 and n2 are the proportions of river and regional

aquifer was evident; the upper aquifer consisted of ﬁne

groundwater, respectively.

Figure 7

Direction of ﬂow during river ﬁltration and groundwater runoff at the monitoring section.

X. Su et al.

1114

Response of redox zonation to recharge in a riverbank ﬁltration system

Hydrology Research

51.5

2020

In this study, δ 18O was used for this calculation. The

of reduction processes indicates DOC as the electron

contribution rates of the river at the RFZ, GRZ, and DCZ

donor, as shown in Table 1. Generally, ΔG (W) values

were 84.37–93.56%, 72.80–74.82%, and 26.08–50.42%,

of aerobic respiration and denitriﬁcation are higher and

respectively, at the shallow depths; the contribution rates

closer, leading to in situ overlapping of their ranges

of the river at the three zones were 78.45–89.80%, 59.93–

(Korom ; Rivett et al. ). The reactions of Mn(IV)

64.82%, and 52.23–55.03%, respectively, at deeper depths.

and Fe(III) reduction with lower ΔG (W) values could

Thus, it can be concluded that the contribution rate of the

occur sequentially with a clear boundary. However, when

river to deep local groundwater is greater than that of the

the Eh values become low, SO2 4 reduction and methane fer-

shallow groundwater, indicating a closer relationship

mentation can occur simultaneously (Champ et al. ;

between the river and deep groundwater. Meanwhile,

Stumm & Morgan ). In this study, as the redox environ-

along the direction of groundwater ﬂow, with increasing dis-

mental indexes and sensitive components along the

tance away from the river, the contribution rate of the river

direction of ﬂow varied, O2, NO 3 , Mn(IV), Fe(III), and

to both shallow and deep groundwater continuously

SO2 were reduced sequentially, as shown in Figure 8(a) 4

decreased, especially at the DCZ, where the contribution

and 8(b), until reaching well FS1-1, which is 80 m from

of the regional groundwater was quite high.

the shore at shallow ﬂow, and well FS2-2, which is 200 m

The amplitude of the contribution rate of the river at

from the shore at deep ﬂow.

shallow ﬂow was higher than that at deep ﬂow, indicating

Based on this, regional groundwater makes the second

that the recharge condition was more complex for shallow

highest contribution toward the quantity of recharge, which

groundwater. In point FS2 at the GRZ, where the per-

is continuous and stable, and shows spatial variability. The

meability of the vadose zone is greater than that in point

variations in Mn2þ and Fe2þ contents in groundwater might

NS5 and FS4, precipitation could more easily pass through

reﬂect the inﬂuence of regional groundwater lateral recharge

the vadose zone to recharge the shallow groundwater. In

in cases when: (1) the regional groundwater has high concen-

points FS1 and FS2 at the GRZ, the δ O values of shallow

trations of Mn2þ and Fe2þ; (2) the Mn2þ and Fe2þ contents in

groundwater are lower than that of deep groundwater, as

groundwater at well FS1-1, which is 80 m away from shore,

shown in Figure 3, which implies that the shallow ground-

are lowest because of the sequential redox processes; and

water at the GRZ could be affected by precipitation.

(3) mixing of groundwater with two different reducibilities suggests a new environment wherein the concentrations of Mn2þ and Fe2þ in the far-shore groundwater are inﬂuenced

Redox zonation of RBF

by groundwater mixing and reaction. Precipitation recharge may also contribute to the quan-

Principle and identiﬁcation processes

tity of groundwater. However, variations in NHþ 4 , NO3 ,

River water is the most important recharge component of

and DOC contents in groundwater might reﬂect the inﬂu-

the RBF system. During river ﬁltration, the main sequence

ence of the vertical recharge of precipitation, in cases

Table 1

Redox processes in a closed system (modiﬁed after Stumm & Morgan 1995; Champ et al. 1979) Equation

ΔG (W) kJ/eq

Aerobic respiration

CH2 O þ O2 ¼ CO2 þ H2 O

125.1

Denitriﬁcation

þ CH2 O þ 4=5NO 3 þ 4=5H ¼ CO2 þ 2=5N2 þ 7=5H2 O

118.8

Mn(IV) reduction

CH2 O þ 2MnO2 þ 4Hþ ¼ CO2 þ 2Mn2þ þ 3H2 O

81.8

Fe(III) reduction

CH2 O þ 4Fe(OH)3 þ 8Hþ ¼ CO2 þ 4Fe2þ þ 11H2 O

28.9

Sulfate reduction

þ CH2 O þ 1=2SO2 4 þ 1=2H ¼ CO2 þ 1=2HS þ 2H2 O

25.3

Methane fermentation

CH2 O ¼ 1=2CH4 þ 1=2H2 O

23.2

Reaction

X. Su et al.

1115

Figure 8

Response of redox zonation to recharge in a riverbank ﬁltration system

Hydrology Research

51.5

2020

Variation trends of the redox environmental index and sensitive components along the directions of shallow and deep ﬂows.

when: (1) contents of TN and SOC are quite low which inhi-

partition the redox zonation, the range of the river water

bits the release of NHþ 4 and contents of TN, NHþ , 4 NO3 ,

DOC to the groundwater; (2)

recharge zone (RWZ) was inferred as 0–80 m in shallow

and SOC of the vadose zone

ﬂow and 0–200 m in deep ﬂow. The range of the precipi-

medium are abundant, especially at the upper region of

tation vertical recharge zone (PVZ) was 80–420 m in

the aquifer; and (3) increasing NHþ 4 and SOC contents are

shallow ﬂow and 200–420 m in deep ﬂow. The range of

positively correlated with the particle contents of the

the regional groundwater lateral recharge zone (RGZ) was

vadose zone medium.

420–700 m at both the shallow and deep ﬂows.

Therefore, along the directions of both shallow and deep ﬂows, the increasing rates of NHþ 4 , NO3 , and DOC contents

were greater than those of the Mn2þ and Fe2þ contents

Range of sequential redox reactions in the river water recharge zone

between points FS1 and FS3, while the trend was opposite between points FS3 and FS4, as shown in Figure 8(a) and

Precise partitions of the RWZ were carried out and the

8(b). Thus, by adjusting the hydrodynamic zonation to

results are shown below. Vertical differences in the

1116

X. Su et al.

Response of redox zonation to recharge in a riverbank ﬁltration system

Hydrology Research

51.5

2020

permeability of aquifer medium are known to lead to the

content. However, this range of this zone could be assigned

differences in ﬂow velocity, which in turn, inﬂuence water

in well NS4-1–NS6-1, 8–30 m from the shore for shallow

retention time, nutrient ﬂux, microbial community structure,

groundwater and well FS1-2–FS2-2, 80–200 m from the

acid–base condition, and redox condition between the river

shore for deep groundwater.

water and groundwater (Massmann et al. ; Su et al. ), leading to sequential reduction reactions that have

SO2 4 reduction zone

wider distribution ranges in deep groundwater. Meanwhile, this zone is the most important for producing NHþ 4 by

This zone is symbolized by an obvious decrease in SO2 4 con-

ammoniﬁcation (Dhondt et al. ; Evans et al. ).

tents, as shown in Table 1; the ΔG (W) values of Fe(III) and SO2 reduction were relatively similar, leading to a 4

Aerobic respiration and denitriﬁcation zone

non-deﬁnite boundary. The HS produced by SO2 4 reduction would precipitate upon reaction with Fe2þ, which

Microorganism-induced aerobic respiration and denitriﬁca-

is one of the reasons for the decreasing levels of Fe2þ and

tion occurred preferentially and overlapped; they occurred

low concentrations of monitored HS as mentioned above.

within point NS2, less than 2 m from shore. As denitriﬁca-

The ranges could be assigned in well NS6-1–FS2-1, 30–

tion was catalyzed by denitrifying bacteria, the facultative

200 m from the shore for the shallow groundwater, and

anaerobic bacteria could reduce NO 3 to N2 under a

well FS2-2, ∼200 m from the shore for the deep groundwater.

relatively O2-rich environment and a DO content of up to 2–4 mg/L (Rivett et al. ). However, aerobic respiration

Redox conditions within the PVZ

occurred within ∼0.1–1.5 m beneath the riverbed surface, as previously reported (Revsbech et al. ; Kumar &

The contents of Mn2þ and Fe2þ reduced, but stable values

Riyazuddin ; Su et al. ).

were observed between well NS6-1 and FS1-1 for shallow

Mn(IV) reduction zone

of precipitation and regional groundwater as shown in

groundwater during RBF with a relatively little inﬂuence Figure 8(a); therefore, shallow groundwater in well FS1-1 When the Eh values of groundwater dropped to a desired

and the regional groundwater were determined to be the

range, manganese minerals in the riverbed sediment and

two Mn2þ and Fe2þ mixing end members for far-shore

aquifer medium were released into the groundwater by

groundwater. By using the end member mixing model as

organic complexation or reductive dissolution, as symbo-

shown in Equation (2), the mixing ratio of water using

content, which increased

δ 18O was calculated; then, the Mn2þ and Fe2þ mixed lines

slowly to reach a peak value along the ﬂow direction and

for the shallow and deep ﬂows were obtained. Results are

then decreased because of oxidation and precipitation

shown in Figure 9(a) and 9(b).

lized by the variation in Mn

2þ

(Greskowiak et al. ; Kedziorek & Bourg ).

The redox background at the PVZ, which is represented

Hence, the range of this zone could be assigned as well

by point FS2, was the SO2 reduction environment. The 4

NS2-1–NS4-1, 2–8 m from shore for shallow ﬂow and well

measured Fe2þ concentration was lower than that in the

NS2-2–FS1-2, 2–80 m from shore for deep ﬂow.

Fe2þ mixed line for both the shallow and deep ﬂows, indicating the occurrence of reactions that consume Fe2þ. Possible

Fe(III) reduction zone

reasons for this might be (1) Fe2þ combined with HS produced by SO2 4 reduction and (2) oxidizing by O2 recharge

Similar to Mn(IV) reduction, under favorable Eh values of

occurred vertically through the vadose zone as Fe2þ was

groundwater, ferrous minerals in the riverbed sediment

readily oxidized, except for DOC and HS as shown in

and aquifer medium were released into the groundwater

Table 2. However, in the shallow groundwater, HS was con-

by organic complexation or reductive dissolution, as symbo-

sumed almost completely in well FS2-1 and DO content was

lized by the variation in Fe2þ content similar to that in Mn2þ

still relatively high as seen in Figure 4(a), thus O2 would act

X. Su et al.

1117

Figure 9

Table 2

Response of redox zonation to recharge in a riverbank ﬁltration system

Hydrology Research

51.5

2020

The Mn2þ and Fe2þ concentrations at mixed lines of shallow and deep ﬂows.

Redox processes in an open system (modiﬁed after Stumm & Morgan 1995; Champ et al. 1979) Equation

ΔG (W) kJ/eq

Aerobic respiration

CH2 O þ O2 ¼ CO2 þ H2 O

125.1

Sulﬁde oxidation

þ 1=2HS þ O2 ¼ 1=2SO2 4 þ 1=2H

Reaction

2þ

99.8 þ

þ O2 þ 10H2 O ¼ 4Fe(OH)3 þ 8H

96.2

Fe(II) oxidation

4Fe

Nitriﬁcation

þ 1=2NHþ 4 þ O2 ¼ 1=2NO3 þ H þ 1=2H2 O

43.3

Mn(II) oxidation

2Mn2þ þ O2 þ 2H2 O ¼ 2MnO2 þ 4Hþ

40.3

on the consumption. In the deep groundwater, SO2 4 began to reduce in well FS2-2, while DO content was maintained at low levels as seen in Figure 5(a); thus, it is inferred that HS and O2 contributed to the decreasing of Fe2þ. Unlike Fe2þ, the measured Mn2þ concentration was

Redox conditions within the regional groundwater lateral recharge zone At this zone, under conditions where NHþ 4 , NO3 , and DOC

along the ﬂow direction did not increase, (NH4)2SO4 fertili-

mixed lines of both shallow

zer used at the surface caused signiﬁcant accumulation of

and deep groundwater, indicating high Mn(IV) reduction.

SO2 4 in both shallow and deep groundwater as shown in

obviously higher than the Mn

2þ

The redox environment was still reductive despite an increase of Eh in comparison with the

SO2 4

reduction

Figures 4(f) and 5(f), indicating that vertical recharge is signiﬁcant but still less than the lateral recharge. The Eh values

zone. Therefore, it is inferred that the actual lowest Eh

and DO contents along the ﬂow direction at points FS3 and

values in shallow and deep groundwater should be located

FS4 increased, where they were located within the

between the controlling values of Mn(IV) and Fe(III)

depression cone, and except for direct vertical recharge,

reduction. Under this relatively weak reductive condition,

the intermittent state of pumping would promote ground-

water ﬂuctuation leading to O2 dissolution by air trap and

nitriﬁcation would hardly occur in the presence of HS 2þ

and Fe . Therefore,

NO 3

could be recharged and accom-

pany O2 vertically, and both would be reduced before

diffusion (Massmann et al. ; Kohfahl et al. ). The points FS3 and FS4 represent the groundwater of

This shows that the vertical recharge

RGZ, where the measured Mn2þ and Fe2þ concentrations

affected the sequential redox zonation mainly because of

were obviously higher than the mixed lines at both shallow

the recharge of electron accepters with higher Gibbs free

and deep groundwater as shown in Figure 9(a) and 9(b),

energy and electron donors.

indicating that Mn(IV) and Fe(III) reduction were both

Fe(III) and

SO2 4 .

1118

X. Su et al.

Response of redox zonation to recharge in a riverbank ﬁltration system

Hydrology Research

51.5

2020

active. The actual lowest Eh values in shallow and deep

the aquifer medium would consistently release more Mn2þ

groundwater should be located at the controlling values of

and Fe2þ into the groundwater.

Fe(III) reduction, which was lower than that of the PVZ, owing to continuous lateral recharge of the high Fe2þ reduc-

ACKNOWLEDGEMENTS

tive regional groundwater. Similar to the PVZ, under relatively reductive con2þ ditions, oxidation of HS , Fe2þ, NHþ would 4 , and Mn

The authors would like to thank the editors of Hydrology

barely occur, and small quantities of O2 and NO 3 recharged

Research and the reviewers for their thoughtful and

vertically would be rapidly reduced. In summary, at the

constructive

RGZ and PVZ, O2,

NO 3,

Mn(IV), and Fe(III) would be

comments,

which

helped

improve

the

manuscript.

reduced in sequence again, which is affected by vertical and lateral recharge including: (1) depression of groundwater reduction in different degrees; (2) recharge intensity

FUNDING INFORMATION

of electron donors; and (3) recharge intensity of electron accepters with higher Gibbs free energy.

This work was supported by the National Natural Science Foundation of China (Grant numbers: 41877178, 41372238).

CONCLUSIONS DATA AVAILABILITY STATEMENT With a background of high ferrous and manganese content in regional groundwater, river ﬁltration and groundwater

All relevant data are included in the paper or its Supplemen-

runoff toward the groundwater depression cone in the

tary Information.

RBF system have two different groundwater ﬂow paths (shallow and deep) due to the variation in permeability of riverbed sediments at different intervals of the river. Mean-

REFERENCES

while, recharge intensities of the river water, regional groundwater, and precipitation showed differences in the spatial distribution along the groundwater path from the river to the depression cone, inﬂuencing the redox conditions. At the RWZ with a high river recharge contribution, O2, 2 NO 3 , Mn(IV), Fe(III), and SO4 were reduced in sequence

meaning the DOC, as the electron donor, formed a strong reducing environment accompanied by an increase in NHþ 4 through ammoniﬁcation. However, lateral recharge from reductive regional groundwater with high Fe2þ and Mn2þ concentrations and vertical recharge from oxidative precipi2 tation with high O2, NO 3 , SO4 , and DOC concentrations

change the redox conditions of the local groundwater from RBF. At the PVZ, the inﬂuence of vertical precipitation recharge was more signiﬁcant than that of the lateral recharge, which led to higher Eh values in groundwater in the PVZ than that in the RGZ. These redox conditions provided an opportunity for Mn(IV) reduction at the PVZ and for both Mn(IV) and Fe(III) reduction at the RGZ; therefore,

Burt, T. P., Pinay, G., Matheson, F. E., Haycock, N. E., Buttruini, A., Clement, J. C., Danielescu, S., Dowrick, D. J., Hefting, M. M., Hillbrichtllkowaka, A. & Maitre, V.  Water table fluctuations in the riparian zone: comparative results from a pan-European experiment. Journal of Hydrology 265 (1), 129–148. Buzek, F., Kadlecova, R., Jackova, I. & Lnenickova, Z.  Nitrate transport in the unsaturated zone: a case study of the riverbank filtration system Karany, Czech Republic. Hydrological Processes 26 (5), 640–651. Champ, D. R., Gulens, J. & Jackson, R. E.  Oxidationreduction sequences in ground water flow systems. Canadian Journal of Earth Sciences 16 (1), 12–23. Chen, X.  Measurement of streambed hydraulic conductivity and its anisotropy. Environmental Geology 39 (12), 1317–1324. Dhondt, K., Boeckx, P., Cleemput, O. V. & Hofman, G.  Quantifying nitrate retention processes in a riparian buffer zone using the natural abundance of 15N in NO-3? Rapid Communications in Mass Spectrometry 17 (23), 2597–2604. Evans, D. M., Schoenholtz, S. H., Wigington, P. J. & Stephen, M. G.  Nitrogen mineralization in riparian soils along a river continuum within a multi-land-use basin. Soil Science Society of America Journal 75 (2), 719.

1119

X. Su et al.

Response of redox zonation to recharge in a riverbank ﬁltration system

Farnsworth, C. E. & Hering, J. G.  Inorganic geochemistry and redox dynamics in bank filtration settings. Environmental Science & Technology 45 (12), 5079–5087. Gandy, C. J., Smith, J. W. N. & Jarvis, A. P.  Attenuation of mining-derived pollutants in the hyporheic zone: a review. Science of the Total Environment 373, 435–446. Greskowiak, J., Prommer, H., Massmann, G. & Nützmann, G.  Modeling seasonal redox dynamics and the corresponding fate of the pharmaceutical residue phenazone during artificial recharge of groundwater. Environmental Science & Technology 40, 6615–6621. Hiscock, K. M. & Grischek, T.  Attenuation of groundwater pollution by bank filtration. Journal of Hydrology 266 (3), 139–144. Kedziorek, M. A. M. & Bourg, A. C. M.  Electron trapping capacity of dissolved oxygen and nitrate to evaluate Mn and Fe reductive dissolution in alluvial aquifers during riverbank filtration. Journal of Hydrology 365 (1–2), 74–78. Kohfahl, C., Massmann, G. & Pekdeger, A.  Sources of oxygen flux in groundwater during induced bank filtration at a site in Berlin, Germany. Hydrogeology Journal 17 (3), 571–578. Korom, S. F.  Natural denitrification in the saturated zone: a review. Water Resources Research 28, 1657–1668. Kumar, A. R. & Riyazuddin, P.  Seasonal variation of redox species and redox potentials in shallow groundwater: a comparison of measured and calculated redox potentials. Journal of Hydrology 444 (10), 187–198. Massmann, G., Nogeitzig, A., Taute, T. & Pekdeger, A.  Seasonal and spatial distribution of redox zones during lake bank filtration in Berlin, Germany. Environmental Geology. 54, 53–65. Muz, M., Oswald, S. E., Schäfferling, R. & Lensing, H.  Temperature-dependent redox zonation, nitrate removal and attenuation of organic micropollutants during bank filtration. Water Research 162, 225–235.

Hydrology Research

51.5

2020

Revsbech, N. P., Jacobsen, J. P. & Nielsen, L. P.  Nitrogen transformations in microenvironments of river beds and riparian zones. Ecological Engineering 24 (5), 447–455. Rivett, M. O., Buss, S. R., Morgan, P., Smith, J. W. & Bemment, C. D.  Nitrate attenuation in groundwater: a review of biogeochemical controlling processes. Water Research 42 (16), 4215–4232. Stumm, W. & Morgan, J. J.  Aquatic Chemistry: Chemical Equilibria and Rates in Natural Waters, 3rd edn. WileyInterscience, New York. Su, X., Lu, S., Gao, R., Su, D., Yuan, W., Dai, Z. & Papavasilopoulos, E. N.  Groundwater flow path determination during riverbank filtration affected by groundwater exploitation: a case study of Liao River, Northeast China. Hydrological Sciences Journal 62 (14), 2331–2347. Su, X., Lu, S., Yuan, W., Nam, C. W., Dai, Z., Dong, W., Du, S. & Zhang, X. Y.  Redox zonation for different groundwater flow paths during bank filtration: a case study at Liao River, Shenyang, northeastern China. Hydrogeology Journal 26, 1573–1589. Trauth, N., Musolff, A., Knöller, K., Kaden, U. S., Keller, T., Werban, U. & Fleckenstein, J. H.  River water infiltration enhances denitrification efficiency in riparian groundwater. Water Research 130, 185–199. Tufenkji, N., Ryan, J. N. & Elimelech, M.  The promise of bank filtration. Environmental Science & Technology 36 (21), 422–428. Wang, P., Yao, J., Wang, G., Hao, F., Shrestha, S., Xue, B., Xie, G. & Peng, Y.  Exploring the application of artificial intelligence technology for identification of water pollution characteristics and tracing the source of water quality pollutants. Science of the Total Environment 693, 133440. WHO  Guidelines for Drinking-Water Quality, 4th edn. World Health Organization, Geneva, Switzerland.

First received 30 July 2020; accepted in revised form 27 August 2020. Available online 14 September 2020

1120

51.5

2020

Coincidence probability of streamﬂow in water resources area, water receiving area and impacted area: implications for water supply risk and potential impact of water transfer Xingchen Wei, Hongbo Zhang, Vijay P. Singh, Chiheng Dang, Shuting Shao and Yanrui Wu

ABSTRACT Under changing environment, the feasibility and potential impact of an inter-basin water transfer project can be evaluated by employing the coincidence probability of runoff in water sources area (WSA), water receiving area (WRA), and the downstream impacted area (DIA). Using the Han River to Wei River Water Transfer Project (HWWTP) in China as an example, this paper computed the coincidence probability and conditional probability of runoff in WSA, WRA and DIA with the copulabased multivariate joint distribution and quantiﬁed their acceptable and unfavorable encounter probabilities for evaluating the water supply risk of the water transfer project and exploring its potential impact on DIA. Results demonstrated that the most adverse encounter probability (dry–dry– dry) was 26.09%, illustrating that this adverse situation could appear about every 4 years. The

Xingchen Wei Hongbo Zhang (corresponding author) Chiheng Dang Shuting Shao Yanrui Wu School of Water and Environment, Chang’an University, Xi’an, China E-mail: hbzhang@chd.edu.cn Hongbo Zhang Key Laboratory of Subsurface Hydrology and Ecological Effect in Arid Region, Ministry of Education, Chang’an University, Xi’an, China

acceptable and unfavorable probabilities in all encounters were 44.83 and 55.17%, respectively, that is the unfavorable situation would be dominant, implying ﬂood and drought risk management should be paid greater attention in project operation. The conditional coincidence probability (dry WRA & dry DIA if dry WSA) was close to 70%, indicating a requirement for an emergency plan and management to deal with potential drought risk. Key words

| bivariate copula, coincidence probability, conditional probability, Han River to Wei River Water Transfer Project, trivariate copula

HIGHLIGHTS

• • • • •

Coincidence probability of annual runoff in all main impacted areas is investigated. Copula-based method to capture the encounter situations of water transfer project. Acceptable and unfavorable probabilities in all encounters imply water transfer risk. Conditional coincidence probability with dry water resources area was computed. An emergency water scheduling plan is required to deal with potential drought risk.

Vijay P. Singh Department of Biological and Agricultural Engineering & Zachry Department of Civil and Environmental Engineering, Texas A&M University, College Station, TX, USA and National Water Center, UAE University, Al Ain, UAE

1121

X. Wei et al.

Coincidence probability of streamﬂow for water transfer risk, China

Hydrology Research

51.5

2020

GRAPHICAL ABSTRACT

INTRODUCTION Inter-basin water transfer (IBT) means building water trans-

project feasibility. Some multivariate approaches as main

fer projects that span two or more basins for transferring

analysis tools were applied to calculate synchronous–asyn-

water from a basin with abundant water resources to those

chronous probabilities of streamﬂow, precipitation and

in shortage and for redistributing water resources among

drought between water source area and WRA. For instance,

the basins to meet water demand in the water-deﬁcient

He et al. () employed a Bayesian network model to

area (Zhuang ). IBTs project, as an important safeguard

investigate the rich–poor rainfall encounter risk between

to join different water systems, has been widely applied in

WSA and WRA in the middle route of the South-to-

many countries and regions, e.g. Australia, Canada, China,

North Water Transfer Project (SNWTP) in China and

India and the United States, with the purpose of supporting

implemented real-time scenario simulations, reﬂecting the

economic and societal development (Manshadi et al. ;

change in risk in the operation of water transfer projects.

Yan & Chen ; Du et al. ). However, it is being

Liu et al. () investigated the concurrent drought prob-

debated that the hydrological cycle has altered in some

ability between the water source and destination regions of

basins or regions under the combined inﬂuence of climate

the central route of the SNWTP using Clayton copula and

change and human activities (Zou et al. ), with the

general circulation models (GCMs) and found that the prob-

result that the feasibility of IBT projects planned or under

ability of concurrent drought events was highly likely to

construction may be questionable. The questions include

increase during 2020–2050, representing the WSA probably

whether there will be enough water to be diverted and the

having insufﬁcient amounts of water for diversion while the

new potential inﬂuence on water use, ecological protection

WRA urgently needing the diversion of water in concurrent

and disaster control (e.g. drought and pollution) in

drought years. Du et al. () investigated the synchronous

the downstream areas of the water resources area under the

and asynchronous exceedance probabilities of wet versus

changing environment. Thus, the coincidence probability

dry conditions of precipitation in WSA and WRA of the

analysis of annual runoff in these areas, including synchro-

Shuhe to Futuan Water Transfer Project (SFWTP) using a

nous and asynchronous probabilities between or among the

bivariate copula joint distribution function, and results

water sources area (WSA), water receiving area (WRA) and

indicated that climate change had positive impacts on the

the downstream impacted area (DIA), is appealing for evalu-

exceedance probabilities, demonstrating that the project

ating the feasibility and potential inﬂuence of IBT projects.

risk was decreasing. In addition, the coincidence probability

In previous studies, the coincidence probability analysis

analysis of precipitation in WSA, WRA and DIA had been

of IBT projects played an important role in determining

reported by Yan & Chen (). This analysis quantiﬁed

1122

X. Wei et al.

Coincidence probability of streamﬂow for water transfer risk, China

Hydrology Research

51.5

2020

the synchrony and asynchrony of precipitation for the

to build the joint probability distribution between the annual

middle route of SNWTP and veriﬁed the effectiveness of tri-

runoff of three regions (WSA, WRA and DIA) and trivariate

variate Clayton copula in the study area, and obtained the

coincidence probability and conditional coincidence prob-

combination frequencies for the middle route of SNWTP,

ability were analyzed to obtain information for ﬂood and

representing that the amount of transferable water was

drought risk management for the project.

generally assured, but the possibility for water transfer was very small if extreme deﬁcit rainfall events occurred in the WRA.

STUDY AREA AND DATA

It is found from the literature that bivariate and trivariate copulas have been widely used in the coincidence

Han River to Wei River Water Transfer Project

probability analysis of hydrological variables in the relevant areas of IBT projects (Yan & Chen ; Du et al. ).

As an important project to resolve the water conﬂict in

Therein, bivariate copulas are generally used to investigate

northwest China due to increasing industrial and dom-

the hydrological combination frequencies between WSA

estic water consumption, the HWWTP connecting the

and WRA for determining the feasibility of a water transfer

upper Han River basin (WSA) and Wei River basin

project, and that between WSA and DIA for analyzing the

(WRA) in China (as shown in Figure 1) is designed to

potential inﬂuence of the project on the downstream areas

annually transfer an average of 1,000 million m3 of water

(Yan & Chen ). Trivariate copulas are employed to

in 2025 and 1,500 million m3 in 2030 from Huangjinxia

calculate the hydrological combination frequencies and con-

reservoir and Sanhekou reservoir in the upper Han

ditional frequencies among WSA, WRA and DIA by

River basin to large-scale industrial zones and urban

capturing the spatial dependence structure of hydrological

areas in the Wei River basin, with the main purpose of

variables inﬂuencing the coincidence probabilities of asyn-

alleviating watershed water conﬂict (Wu et al. ).

chrony and synchrony. Recent years have seen a growing

Because 70% of water is for industrial development, it

interest in applying copulas to hydrological frequency analy-

requires stable water transfer, implying high requirement

sis, which can be attributed to manifold advantages of

in controlling water supply risk, especially in the changing

copulas in modeling joint distributions, representing ﬂexi-

environment.

bility in selecting arbitrary margins and dependence

The WSA and the DIA of HWWTP are located in the

structure, the ability to deal with three variables or more,

upper

and the separability in analyzing marginal and dependence

95,200 km2 and have a mainstream length with approxi-

Han

River

basin,

with

drainage

area

structure (Salvadori & De Michele ; Serinaldi et al.

mately 925 km (Figure 1). The basin is inﬂuenced by the

; Zhang & Singh ).

north subtropic monsoon, representing the annual mean

In this study, trivariate combination frequencies and

temperature ranging between 12 and 16 C (Li et al. )

conditional frequencies of annual runoff series in WSA,

and average annual precipitation varying from 700 to

WRA and DIA inﬂuenced by the Han River to Wei

1,800 mm, and 80% of annual precipitation falls in the wet

River Water Transfer Project (HWWTP) in China, were

period from May to October in which the corresponding

investigated through copula-based multivariate probability

runoff is about 4.11 × 1010 m3, accounting for 70% of the

distribution, with the purpose of estimating water supply

total annual runoff. Also, the runoff of the entire basin

risk of water transfer project and exploring its potential

demonstrates a large annual and interannual variability (Li

impact on the downstream area of the water source areas.

et al. ). Therein, the WSA mainly refers to the watershed

Different from the previous relevant studies on the water

above Huangjinxia and Sanhekou reservoirs and the DIA is

transfer project, it uses annual runoff data from three

downstream of them, primarily covering the watershed

gauges located in WSA, WRA and DIA, respectively and

below Shiquan gauge.

provides a reference for the scheduling of inter-basin transfer

The WRA of HWWTP mainly refers to Guanzhong

projects. The bivariate and trivariate copulas were employed

plain region located in the central part of Shaanxi Province

X. Wei et al.

1123

Figure 1

Coincidence probability of streamﬂow for water transfer risk, China

Hydrology Research

51.5

2020

Study area and locations of three gauges potentially impacted by the Han River to Wei River Water Transfer Project (HWWTP) in China.

and belongs to the Wei River basin. The Guanzhong plain 2

DSDJK gauges were selected during 1956–2000. Consider-

region covers a total area of 55,500 km , as the most

ing that measured runoff is greatly affected by human

economically developed area in Shaanxi Province, including

activities in WRA, representing nonstationary feature (Lan

ﬁve prefecture-level cities, i.e. Xi’an city, Baoji city,

et al. ), the natural runoff data, covering 45 years from

Xianyang city, Weinan city, Tongchuan city and one agricul-

1956 to 2000, were selected from the XY gauge. The

tural high-tech industries demonstration zone as Yangling

runoff data were obtained from hydrological manuals pub-

(Tian et al. ). The Guanzhong plain region has an aver-

lished by the Hydrological Bureaus of the Yellow River

age elevation of approximately 500 m and belongs to the

Conservancy Commission and Yangtze River Conservancy

continental monsoon climate, with the annual mean temp-

Commission. Statistical features of the annual data series

erature of 6–13 C and precipitation of 500–800 mm (Deng

are shown in Table 1.

et al. ). Data

Table 1

Statistical features of annual runoff during 1956–2000

Standard

According to the water transfer route of the HWWTP, the

deviation (108 m3)

Coefﬁcient of variation

Coefﬁcient of skewness

data used in this paper included annual runoff data from

Gauges

Mean (108 m3)

Shiquan (SQ) gauge on the WSA, Danjiangkou gauge (the

Shiquan (SQ)

327.5

143.6

0.4385

0.9964

watershed above SQ is not covered or the data deducted

Xianyang (XY)

49.8

20.6

0.4125

0.7392

that from at SQ, DSDJK) on the DIA, and Xianyang (XY)

Danjiangkou (DSDJK)

368.1

136.4

0.3703

1.0753

gauge on the WRA (Figure 1) (see Supplementary Material). Therein, the observed runoff data series from SQ and

Note: Runoff at Danjiangkou does not include the runoff from the area above Shiquan gauge.

X. Wei et al.

1124

Coincidence probability of streamﬂow for water transfer risk, China

METHODOLOGIES

Hydrology Research

51.5

2020

can be easily constructed and are capable of capturing a variety of dependence structures with several desirable properties, such as symmetry and associativity, they have

Copula functions

become very popular (Nelsen ; Hofert ). Sklar () introduced the copula theorem to model the

The widely used Archimedean family copulas include

stochastic nature of multi-dimensional processes by using

the Frank copula, Clayton copula, Ali–Mikhail–Haq

univariate marginal distributions to derive multivariate dis-

copula (AMH) and Gumbel–Hougaard (G–H) copula, with

tribution functions. Nowadays, copula models play an

a parameter θ. For bivariate copula function, Kendall’s cor-

active role in the hydrological ﬁeld (Yan & Chen ;

relation coefﬁcient τ has a certain relation with parameter θ,

Zhang & Singh ), such as drought assessment

as shown in Table 2. The appropriate copulas can be deter-

(Chanda et al. ), ﬂood risk analysis (Favre et al. ;

mined by different values of Kendall’s τ from observations

Serinaldi et al. ; Chebana et al. ) and streamﬂow

(Hofert ; Xie & Wang ).

simulation (Chen et al. ). According to Sklar’s theorem,

Moreover, parameters of multi-dimensional copulas can

the copula function C can be used to describe a joint multi-

be estimated by the maximum-likelihood estimator (MLE)

variate probability distribution of n correlated variables (X1,

or inference of functions for margins (IFM) method (Joe

X2, X3,…, Xn). If variables (X1, X2, X3,…, Xn) have arbitrary

, ). In view of the well-known optimality properties

marginal distribution functions FX1 (x1 ), FX2 (x2 ), . . . , FXn (xn ),

of the MLE, it would be the preferred option for estimating

respectively, there will exist a copula function C, which

θ. Nevertheless, it is found from application that the more

can combine these marginal distribution functions to give

ﬂexible IFM method is preferable to the MLE. Although

the joint distribution function, FX1 ,X2 ,...,Xn (x1 , x2 , . . . , xn ), as

the conceptual bases of the two methods are very similar, and the efﬁciency of these two methods is almost the same

follows:

in many cases, the IFM method is easier to calculate than FX1 ,X2 ,...Xn (x1 , x2 , . . . , xn ) ¼ Cθ [FX1 (x1 ), FX2 (x2 ), . . . , FXn (xn )] ¼ Cθ (u1 , u2 , ::, un )

u1 , u2 , ::, un ∈ [0, 1]

where

are

uniformly

the MLE method. For example, the trivariate copula parameter θ is estimated in two steps in the IFM method. In the ﬁrst step,

(1)

the parameters αk (k ¼ 1, 2, 3) of each marginal distribution

distributed

are estimated separately via Xki, i ¼ 1,2,…,n, and this estima∧

tor is expressed as αk . The second step is that θ is estimated

u1 ¼ FX1 (x1 ), u2 ¼ FX2 (x2 ), . . . , un ¼ FXn (xn ), and the copula

by replacing αk for αk in the log-likelihood (as shown in

random

realizations

the

variables,

deﬁned

parameter is θ. A unique copula function C(⋅) exists, when

∧

Equation (2)) and the IFM estimate of θ is as follows:

these marginal distributions are continuous (Sraj et al. ). There are many types of copula functions, such as Plackett copula (Plackett ), Archimedean copulas and elliptical

∧

θ ¼ arg max

n X

∧

i¼1

(2)

copulas (Fang et al. ). Because the Archimedean copulas Table 2

∧

log [cθ {F1 (X1i , α1 ), F2 (X2i , α2 ), F3 (X3i , α3 ); θ}]

Types of Clayton, Frank and GH copula (Nelsen 2006) Cθ

θ0 range

Relation between θ and τ

Frank

1 (e θu 1)(e θv 1) ln 1 þ θ θ e 1

∞ < θ < ∞

τ ¼1þ

Clayton

(u θ þ v θ 1) 1=θ

θ>0

AMH

uv=[1 θ(1 u)(1 v)]

1 θ < 1

G–H

exp { [( ln u)θ þ ( ln v)θ ]

Archimedean copula

1=θ

}

θ 1

ð 4 1 θ t dt 1 t θ θ 0e 1

θ 2þθ 2 2 1 2 τ ¼ 1 1 ln (1 θ) 3θ 3 θ 1 τ ¼1 θ τ¼

1125

X. Wei et al.

Coincidence probability of streamﬂow for water transfer risk, China

Hydrology Research

Marginal distribution

expressed as follows:

The Pearson type III (P-III) distribution, Gumbel distri-

Dn ¼ max jPEi PTi j (i ¼ 1, 2, . . . , n)

bution and Lognormal distribution were employed in this

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u 2 N pffiffiffiffiffiffiffiffiffiffiffi u 1X RMSE ¼ MSE ¼ t (PEi PTi ) N i¼1

study for describing the marginal distribution of single annual runoff. The cumulative distribution function (CDF) of P-III distribution can be deﬁned as

F(x) ¼ P(x > xp ) ¼

βα Γ(α)

ð∞

pffiffiffiffiffiffiffiffiffiffiffi AIC ¼ N ln ( MSE) þ 2k

51.5

2020

(6)

(7)

(8)

where N is the sample size, k is the number of parameters

(x δ)α 1 e β(x δ)

(3)

of different distributions, PEi and PTi are the empirical frequency and theoretical frequency, respectively. Therein,

where α, β and δ are the shape, scale and location parameters of the P-III distribution, respectively.

the empirical joint probability always plays a signiﬁcant role in the selection of a copula function, which is used as a criterion for judging and selecting the best theoretical dis-

The CDF of Gumbel distribution can be deﬁned as

tribution from the Archimedean copula functions. Assuming the variables X and Y with the same length, the empirical

h i x k F(x) ¼ P(x > xp ) ¼ exp e γ

(4)

frequency of bivariate variables (X and Y ) was estimated by using the Gringorten formula (Gringorten ), as

where γ and k are the scale and the location parameters of the Gumbel distribution, respectively. The CDF of Lognormal distribution can be defined as " 1 F(x) ¼ P(x > xp ) ¼ pffiffiffiffiffiffi σ 2π

ð ∞ exp xp

( ln x μ)2 2σ 2 t

follows. H(x, y) ¼ P(X xi , Y yi ) ¼

# dt

(5)

NO: of(X xi , Y yi ) 0:44 N þ 0:12

(9)

where (xi , yi ) is the combination of the ith values in the X and Y series arranged in increasing order, i is 1: N, and N is the length of the series. For the trivariate copula function

where μ and σ are the mean and standard deviation formed by the natural logarithm of the variable x, respectively. The linear moment method (Hosking ) was employed to estimate the parameters of the above three distributions, and a ﬁtting test was used to choose the suitable marginal distribution of a single variable.

also, the above equation was used. The theoretical frequencies can be obtained by the models of marginal and copula distributions. It is noted that according to RMSE and AIC criteria, the smaller AIC and RMSE are, the better the ﬁtness of the distribution. Coincidence probability

Fitting test

The coincidence probability is usually deﬁned as the probability of two or more events that happen at the same

In this paper, Kolmogorov–Smirnov (K–S) test (Dn, Massey

time. For the IBT project, it is considered that the runoff

), root mean square error (RMSE, Zhang & Singh ,

coincidence is generally deﬁned as the simultaneous occur-

), Akaike’s information criterion (AIC, Zhang & Singh

rence of runoff in two or more basins, which represent the

, ) were employed to measure the goodness of ﬁt

frequency P of wetness, dryness or normal of one basin

of the marginal and joint distributions which can be

when the other basin is in a condition of wetness, dryness

X. Wei et al.

1126

Coincidence probability of streamﬂow for water transfer risk, China

or normal. It is reported that the 37.5 and 62.5% quantiles

Hydrology Research

51.5

2020

Wet–normal–dry periods coincidence probability, Pwnd:

are usually used as thresholds to deﬁne the condition of wetness or dryness in precipitation coincidence (Liu & Zheng ; Yan & Chen ), and this paper used them due to

Pwnd ¼ Px,w ∧ Py,n ∧ Pz,d ¼ (X Xw , Yd YYw , ZZd ) ¼ Cθ (vw , ωw ) Cθ (vd , ωd ) Cθ (uw , vw , ωw ) þ Cθ (uw , vw , ωd )

high rainfall–runoff relation and simultaneous frequency.

(14)

For the annual runoff variable X, Xw and Xd were assigned the values of Pw ¼ 62.5% and Pd ¼ 37.5%, respect-

Dry–dry–dry periods coincidence probability, Pddd:

ively, as the threshold quantiles of wetness and dryness, respectively. The degree of wet–dry runoff can be described as wetness (X Xw ¼ X62.5%), dryness (X < Xd ¼ X37.5%) and normal (Xw > X Xd). In terms of bivariate copula, there were nine encounter situations (X and Y ), and the trivariate copula (X, Y and Z ) had 27 combinations. Some of all 36 combination functions were as follows.

¼ Cθ (ud , vd , ωd )

(15)

where u, v and ω are the marginal distributions of X, Y and Z, and subscripts w, n and d mean wet, normal and dry conditions. The composite letters note the coincidence

Wet–wet periods coincidence probability, Pww:

condition. For example, ddd presents the dry–dry–dry condition (period), and Pddd is the coincidence probability

Pww ¼ Px,w ∧ Py,w ¼ P(X Xw , Y Yw ) ¼ 1 uw vw þ Cθ (uw , vw )

Pddd ¼ Px,d ∧ Py,d ∧ Pz,d ¼ (X < Xd , YYd , ZZd )

(10)

of this condition. Here, the ∧ product notation is used to express the simultaneous occurrence of events. Other combination probabilities listed in Tables 3 and 4 were

Wet–dry periods coincidence probability, Pwd:

calculated in a similar way.

Pwd ¼ Px,w ∧ Py,d ¼ P(X Xw , Y < Yw ) ¼ vw Cθ (uw , vd )

(11)

RESULTS AND DISCUSSION

Normal–wet periods coincidence probability, Pnw:

Parameter estimation and ﬁtting test of marginal distributions

Pnw ¼ Px,n ∧ Py,w ¼ P(Xd X < Xw , Y Yw ) ¼ uw vd Cθ (uw , vw ) þ Cθ (ud , vw )

(12)

Wet–wet–wet periods coincidence probability, Pwww: Pwww ¼ Px,w ∧ Py,w ∧ Pz,w ¼ (X Xw , Y Yw , Z Zw ) ¼ 1 uw vw ww þ Cθ (uw , vw ) þ Cθ (uw , ωw ) þ Cθ (vw , ωw ) Cθ (uw , vw , ωw ) (13) Table 3

The parameters of the P-III, Gumbel, and Lognormal distributions mentioned above were estimated using the linear moment method, and results are shown in Table 5. Also, the K–S test (Dn), RMSE and AIC were employed to test the feasibility of ﬁtting these three distributions to runoff data, and results are shown in Figure 2. The statistical results in Figure 2 indicate that the P-III, Gumbel and Lognormal distributions all passed the K–S test

Nine coincidence combinations of bivariate copula Y

Wet

Normal

Dry

Wet

Pww ¼ Px,w ∧ Py,w

Pwn ¼ Px,w ∧ Py,n

Pwd ¼ Px,w ∧ Py,d

Normal

Pnw ¼ Px,n ∧ Py,w

Pnn ¼ Px,n ∧ Py,n

Pnd ¼ Px,n ∧ Py,d

Dry

Pdw ¼ Px,d ∧ Py,w

Pdn ¼ Px,d ∧ Py,n

Pdd ¼ Px,d ∧ Py,d

X. Wei et al.

1127

Table 4

Coincidence probability of streamﬂow for water transfer risk, China

Hydrology Research

51.5

2020

Twenty-seven coincidence combinations of trivariate copula Z

Wet

Normal

Dry

Wet

Wet Normal Dry

Pwww ¼ Px,w ∧ Py,w ∧ Pz,w Pwnw ¼ Px,w ∧ Py,n ∧ Pz,w Pwdw ¼ Px,w ∧ Py,d ∧ Pz,w

Pwwn ¼ Px,w ∧ Py,w ∧ Pz,n Pwnw ¼ Px,w ∧ Py,n ∧ Pz,n Pwdn ¼ Px,w ∧ Py,d ∧ Pz,n

Pwwd ¼ Px,w ∧ Py,w ∧ Pz,d Pwnd ¼ Px,w ∧ Py,n ∧ Pz,d Pwdd ¼ Px,w ∧ Py,d ∧ Pz,d

Normal

Wet Normal Dry

Pnww ¼ Px,n ∧ Py,w ∧ Pz,w Pnnw ¼ Px,n ∧ Py,n ∧ Pz,w Pndw ¼ Px,n ∧ Py,d ∧ Pz,w

Pnwn ¼ Px,n ∧ Py,w ∧ Pz,n Pnnn ¼ Px,n ∧ Py,n ∧ Pz,n Pndn ¼ Px,n ∧ Py,d ∧ Pz,n

Pnwd ¼ Px,n ∧ Py,w ∧ Pz,d Pnnd ¼ Px,n ∧ Py,n ∧ Pz,d Pndd ¼ Px,n ∧ Py,d ∧ Pz,d

Dry

Wet Normal Dry

Pdww ¼ Px,d ∧ Py,w ∧ Pz,w Pdnw ¼ Px,d ∧ Py,n ∧ Pz,w Pddw ¼ Px,d ∧ Py,d ∧ Pz,w

Pdwn ¼ Px,d ∧ Py,w ∧ Pz,n Pdnn ¼ Px,d ∧ Py,n ∧ Pz,n Pddn ¼ Px,d ∧ Py,d ∧ Pz,n

Pdwd ¼ Px,d ∧ Py,w ∧ Pz,d Pdnd ¼ Px,d ∧ Py,n ∧ Pz,d Pddd ¼ Px,d ∧ Py,d ∧ Pz,d

Table 5

Estimated parameters of P-III, Gumbel and Lognormal distributions

P-III

Gumbel δ

Lognormal

Gauges

4.0371

71.4811

38.9582

110.7327

263.6216

327.5376

142.0196

8.1332

7.2092

8.7950

15.8512

40.6888

49.8382

20.3298

DSDJK

3.4689

73.1856

114.2435

105.0921

307.4579

368.1180

134.7853

1:36 with the critical value D0:05 ¼ pffiffiffiffiffiffi ¼ 0:2027. At SQ gauge, it 45 was seen that the AIC of the Gumbel distribution was the

tau (τ) are ﬁrst calculated to describe the dependence (con-

smallest of the three models, while the P-III distribution

and the values of R and τ are shown in Table 6.

had the smallest RMSE. After consideration, the P-III distri-

For this, Pearson’s correlation coefﬁcient (R) and Kendall’s cordance) of annual runoff for any two of the three gauges, In Table 6, it is seen that runoff series from any two of

bution was ﬁnally selected to ﬁt the annual runoff data from

the three gauges had positive correlation, thus the G–H

the SQ gauge. For XY and DSDJK gauges, the Gumbel dis-

copula and Clyton cupola functions were considered suit-

tribution had both the smallest AIC and RMSE, thus the

able for the data series analysis.

Gumbel distribution was employed to ﬁt at the two

Then, Kendall’s correlation coefﬁcient τ was counted

gauges. The curves in Figure 2 represent the performances

between annual runoff data from any two of the three

of three models in ﬁtting annual runoff data from the three

gauges, and the corresponding copula parameters θ can be

gauges visually.

computed in terms of the equations of the G–H and Clyton copulas listed in Table 2. The values of τ and θ are shown

Parameter estimation and ﬁtting test of joint distribution

in Table 7. Furthermore, the bivariable joint distributions from the Clayton copula and G–H copula were constructed by

Various forms of the copula function have different require-

using the estimated parameters and the equations in

ments for the correlation of variables. For example, the

Table 2, respectively, while the parameters of trivariate Clay-

G–H copula and Clayton copula functions are suitable

ton and G–H copula functions need to be estimated through

for random variables with positive correlation, the Frank

Equation (2) for the construction of trivariate distributions,

copula function applicable to random variables with posi-

which is different from the bivariable joint distribution. So,

tive and negative correlations, and the AMH function is

the ﬁtness results of all joint distributions, representing Dn,

often applied to random variables with weak correlations.

RMSE and AIC, are shown in Tables 8 and 9.

X. Wei et al.

1128

Coincidence probability of streamﬂow for water transfer risk, China

Hydrology Research

Table 6

51.5

2020

Pearson’s correlation coefﬁcient (R) and Kendall’ tau (τ) between runoff series from three gauges

Gauges

SQ–XY

0.7983

0.5993

SQ–DSDJK

0.8542

0.6721

XY–DSDJK

0.7355

0.4611

Table 7

Parameter values of bivariate copula functions

Θ Gauges

G––H

Clayton

SQ–XY

0.5993

2.494

2.988

SQ–DSDJK

0.6721

3.049

4.098

XY–DSDJK

0.4611

1.855

1.711

Tables 8 and 9 show that bivariate and trivariate copulas, based on the G–H and Clayton copula, both passed the K–S test (Dn less than 0.207). The smallest values of RMSE and AIC (0.0811, 224.12) in Table 8 demonstrate that Clayton copula was the best-ﬁtted bivariable joint distribution between runoff series from SQ and XY gauges and was also suitable for that between XY and DSDJK gauges, with RMSE and AIC of 0.0995 and 205.71. Differently, the best-ﬁtted bivariable joint distribution between SQ and DSDJK gauges was the G–H copula, and the smallest values of RMSE and AIC were 0.0700 and 237.33. Additionally, it can be seen from Table 9 that the smallest values of RMSE and AIC are 0.1646 and 212.66, respectively, implying the G–H copula was the best suitable function among trivariate copula functions and can describe the joint distribution among three gauges. Coincidence probability of bivariate copula According to the chosen and established bivariate copula functions, the bivariate joint runoff distribution between any two gauges was obtained, and the joint coincidence probability was calculated under certain conditions. The joint cumulative probability distribution and contours at any two of the three coupled gauges are shown in Figure 3. From the contours in Figure 3, one can obtain the probabilities that annual runoff of any two gauges was less than a certain value at the same time, and that possible combiFigure 2

Marginal distributions ﬁtted with empirical frequency and models.

nations of annual runoff from different gauges under a

X. Wei et al.

1129

Table 8

Coincidence probability of streamﬂow for water transfer risk, China

Hydrology Research

51.5

2020

K–S test (Dn), RMSE and AIC of bivariate copulas Dn

RMSE

AIC

Gauges

G–H

Clayton

G–H

Clayton

G–H

Clayton

SQ–XY

0.1600

0.1280

0.0854

0.0811

219.47

224.12

SQ–DSDJK

0.1279

0.1086

0.0700

0.0701

237.33

237.16

XY–DSDJK

0.1840

0.1472

0.1016

0.0995

203.82

205.71

Table 9

K–S test (Dn), RMSE and AIC of trivariate copulas

probabilities generally pointed to wet–wet periods coincidence probability, normal–normal periods coincidence

Trivariate copula functions

RMSE

AIC

G–H copula

3.49

0.0921

0.1646

212.66

Clayton copula

3.27

0.1160

0.1648

191.85

probability and dry–dry coincidence probability, while other probabilities were regarded as asynchronous coincidence probabilities. The synchronous and asynchronous coincidence probabilities at any two gauges were computed,

certain probability, such as the combination of annual runoff from the SQ gauge with a 30% probability and corresponding annual runoff from the XY gauge, or the combination of annual runoff from the XY gauge with a 30% probability and annual runoff from the SQ gauge. At this time, the joint cumulative probability in Figure 3 was able to analyze the worst situation of simultaneous dry periods occurring from any two gauges. For instance, it can be calculated that the joint probability was 10%, that the annual runoff was less than 2 × 1010 m3 at SQ gauge and less than 0.27 × 1010 m3 at XY gauge; that the joint probability was 90%, that the annual runoff was less than 0.86 × 1010 m3 at XY gauge and less than 7 × 1010 m3 at DSDJK gauge. For other examples, the possible combination of annual runoff at SQ and DSDJK gauges with a joint probability less than 50% was that the annual runoff was less 10

than 5 × 10

m at SQ gauge and less than 3.5 × 10

m at

DSDJK gauge, or that the annual runoff was less than 10

3.05 × 10

m at SQ gauge and less than 6 × 10

m at

DSDJK gauge. Through the constructed bivariate joint distributions, the nine coincidence probabilities of annual runoff data mentioned in Table 3 were computed at any two gauges. In the IBT project, the most adverse encounter situation

as shown in Table 10. Obviously, it is seen from the table that the synchronous coincidence probability was generally higher than asynchronous coincidence probability. It can be explained well due to the close geographical position among three gauges and similar climatic conditions. The larger synchronous coincidence probabilities indicated when there was enough water to be diverted in WSA, WRA or DIA was also likely to be wet, implying water demand in WRA for transferring water was small or the potential influence caused by the water transfer in DIA was low. Contrarily, when WSA was dry and cannot provide abundant water for diversion, water demand in WRA for transferring water was very high or the potential influence in DIA was greatly strong due to WRA or DIA being probably dry at that time. This was not expected in the IBT project, because it took on the project with a big risk, threatening the water security in WSA, WRA and DIA. Of course, when there was enough water to be diverted in WSA, WRA just happened to be dry and DIA was wet, and transferring water in the interbasin provided the benefit at this time and caused the low impact in the downstream river basin. Thus, it was clear that coincidence probability of bivariate copula failed to provide a valuable reference when considering WSA, WRA and DIA of the IBT project

was considered as dry–dry periods, which means that a shortage of water resources appeared both in the WSA

Coincidence probability of trivariate copula

and the WRA. The statistical results showed that Pdd was not too high, 30, 29.19 and 26.49% in SQ–XY, XY–DSDJK

The coincidence probability of trivariate copula certainly

and SQ–DSDJK, respectively. The synchronous coincidence

provided an effective tool to comprehensively analyze the

X. Wei et al.

1130

Figure 3

Coincidence probability of streamﬂow for water transfer risk, China

Hydrology Research

51.5

2020

Contours and joint cumulative probability distribution: (a) joint cumulative probability distribution of annual runoff from the SQ and XY gauges; (b) contours of the joint cumulative probability distribution of annual runoff from the SQ and XY gauges; (c) joint cumulative probability distribution of annual runoff from the SQ and DSDJK gauges; (d) contours of the joint cumulative probability distribution of annual runoff from the SQ and DSDJK gauges; (e) joint cumulative probability distribution of annual runoff from the XY and DSDJK gauges; (f) contours of the joint cumulative probability distribution of annual runoff from the XY and DSDJK gauges.

X. Wei et al.

1131

Table 10

Coincidence probability of streamﬂow for water transfer risk, China

SQ– XY

XY– DSDJK

SQ– DSDJK

Synchronous probability Wet–wet Normal–normal Dry–dry Total

26.78 10.50 30.00 67.29

30.43 12.06 29.19 71.68

23.33 8.38 26.49 58.19

Asynchronous Probability

8.86 1.86 8.86 5.64 1.86 5.64 32.72

5.85 1.22 5.85 7.09 1.22 7.09 28.32

9.89 4.28 9.89 6.73 4.28 6.73 41.8

Periods

Wet–normal Wet–dry Normal–wet Normal–dry Dry–wet Dry–normal Total

51.5

2020

safety in Guanzhong plain, China. Considering this, the

Coincidence probability of bivariate copula (%)

Probability

Hydrology Research

wet–wet–wet and dry–dry–dry encounter situations were determined as unfavorable circumstances, in which the wet–wet–wet situation implying ﬂood risk and the dry– dry–dry situation implying drought risk in this study. Other encounter situations were also regarded as acceptable conditions. These results are shown in Table 11 where the values of synchronous coincidence probabilities Pwww, Pnnn and Pddd were 29.08, 7.99 and 26.09%, respectively. Therein, the most adverse encounter situation was in dry– dry–dry periods, with the coincidence probability (Pddd) of 26.09%. Through the probability, the recurrence interval was obtained in the worst encounter situation, and the value was 3.8 years, which means the most adverse situation

inﬂuence of the IBT project. Based on the G–H copula, this

(dry–dry–dry) could appear once every 3.8 years on average.

study constructed trivariate joint distributions, by which

Also, it can be seen from the table that unfavorable prob-

27 coincidence probabilities from the trivariate copula,

ability (Pwww þ Pddd) reached 55.17%, greater than the

corresponding to the combinations shown in Table 4, were

acceptable probability, implying the hydrological condition

computed.

Similarly,

probability

Pwww,

wet–wet–wet

coincidence

was very unfavorable to water transfer in more than half

normal–normal–normal

the

coincidence

of the whole operation period. It undoubtedly posed a big

probability Pnnn, and dry–dry–dry coincidence probability

challenge to the scheduling of inter-basin projects, threaten-

Pddd were considered as synchronous coincidence probabil-

ing the safety of water delivery.

ities of trivariate copula, while other probabilities were

Furthermore, in order to implement the optimal schedul-

asynchronous coincidence probabilities. Furthermore, syn-

ing of the transfer project, it was necessary to identify the

chronous coincidence probabilities and asynchronous

possible runoff in WRA and DIA when different runoff

coincidence probabilities at three gauges (SQ–XY–DSDJK)

rhythms happened in WSA. Fortunately, the joint distribution

can be obtained, as shown in Table 11.

model established by the copula function provided an effec-

As is known, the purpose of HWWTP was to balance

tive tool for solving this problem. So three equations were

the non-uniform temporal and spatial distributions of

constructed to calculate dry–dry coincidence probabilities

water resources in WSA and WRA and increase water

of annual runoff from the XY and DSDJK gauges with the

supply for guaranteeing industrial and domestic water

condition that SQ gauge was in wet, normal and dry periods.

Table 11

Coincidence probability of trivariate copula (%)

Wet (SQ)

Normal (SQ)

Dry (SQ)

Periods

Wet (XY)

Normal (XY)

Dry (XY)

Wet (XY)

Normal (XY)

Dry (XY)

Wet (XY)

Normal (XY)

Dry (XY)

Wet (DSDJK)

29.08

2.23

0.05

2.23

2.78

0.34

0.05

0.34

0.39

Normal (DSDJK)

2.23

2.78

0.34

2.78

7.99

2.40

0.34

2.40

3.75

Dry (DSDJK)

0.05

0.34

0.39

0.34

2.40

3.75

0.39

3.75

26.09

Sum of synchronous probability (Pwww þ Pnnn þ Pddd)

63.16

Sum of asynchronous probability (Pother)

36.84

Unfavorable probability (Pwww þ Pddd)

55.17

Acceptable probability (Pother)

44.83

1132

X. Wei et al.

Coincidence probability of streamﬂow for water transfer risk, China

Hydrology Research

51.5

2020

The conditional dry–dry coincidence probabilities were

the conditional coincidence probability of a certain magni-

obtained as follows, where X notes runoff from the SQ

tude was obtained directly from the contours. For

gauge, Y is that from the XY gauge and Z refers to the

example, when SQ gauge was dry, the conditional coinci-

runoff from the DSDJK gauge.

dence probability of dry XY and dry DSDJK was 20%, in which dry XY illustrated the annual runoff at XY gauge

Wet SQ, dry XY and dry DSDJK,

was less than 0.5 × 1010 m3 and dry DSDJK demonstrated that it was less than 2.2 × 1010 m3.

F(y, zjx) ¼ P(Y < Yd , Z < Zd jX Xw ) ¼

Cθ (vd , ωd ) Cθ (uw , vd , ωd ) ¼ 0:0104 1 uw

(16)

Also, the conditional coincidence probability of the most adverse encounter situation (dry A and dry B if dry C) for the IBT project was 69.57%, representing the encounter prob-

Normal SQ, dry XY and dry DSDJK,

ability of dry XY and dry DSDJK in the case that SQ was in the dry period. Compared with the coincidence probability

F(y, zjx) ¼ P(Y < Yd , Z < Zd jXd X < Xw ) ¼

Cθ (un , vd , ωd ) Cθ (ud , vd , ωd ) ¼ 0:4483 un ud

(dry SQ and dry XY and dry DSDJK) of trivariate copula in (17)

Table 11, the conditional coincidence probability (dry XY and dry DSDJK if dry SQ) better showed the importance of drought risk, indicating the dry–dry–dry coincidence prob-

Dry SQ, dry XY and dry DSDJK,

ability in the study area was close to 70% when the drought event occurred in WSA. Thus, it should be emphasized

F(y, zjx) ¼ P(Y < Yd , Z < Zd jX < Xd ) ¼

Cθ (ud , vd , ωd ) ud

¼ 0:6957

again that the implementation effect of designed water transfer could not be guaranteed to a large extent, once WSA was (18)

in a dry period during the operation of HWWTP.

where u, v and ω are the marginal distributions of X, Y and Z, and subscript w, n and d mean the wet, normal and dry

CONCLUSIONS

conditions, respectively. The above calculation showed that the coincidence

The Han River to Wei River Water Transfer Project

probability of wet SQ, dry XY and dry DSDJK was 1.04%;

(HWWTP) is a strategic project serving the water resources

that of normal SQ, dry XY and dry DSDJK was 44.83%;

allocation in northwest China, aiming to mitigate water

that of dry SQ, dry XY and dry DSDJK was 69.57%. Therein,

shortage in the Guanzhong Plain. It is important for

the probability that dry XY and dry DSDJK happened at the

HWWTP to estimate the water supply risk of the water transfer

same time reached 69.57%, in the case that dry SQ was

project and explore its potential impact on the downstream of

occurring, and that of dry XY and dry DSDJK arrived at

the water sources areas. To discuss this issue, this study ana-

44.83% even though normal runoff occurred at SQ gauge.

lyzed the coincidence probability of annual runoff series

They implied that water supply deﬁcit in WRA due to the

between or among the water source area (WSA), WRA and

insufﬁcient transfer of water caused by low watershed

the DIA. In this analysis, the Archimedean copula-based

yield in WSA, and water supply deﬁcit in DIA due to low

method was used to capture the encounter situations of var-

ﬂow from the upstream watershed caused by normal water

ious water yields in different areas by constructing bivariate

transfer, could disturb the operation of the HWWTP,

and trivariate joint distributions and calculated various combi-

when there was not abundant water yield in WSA.

nation frequencies, implying the water supply risk of the

To visualize the conditional coincidence probability among three gauges mentioned above, the distribution and

HWWTP in China. The main conclusions of the study can be summarized as follows.

contours of different magnitudes of runoff from XY and

The coincidence probability of annual runoff series from

DSDJK gauges in the case that SQ gauge was in dry

WSA (SQ gauge), WRA (XY gauge) and DIA (DSDJK gauge)

period were drawn, as shown in Figure 4. From the ﬁgure,

was obtained for different encounter situations by using

X. Wei et al.

1133

Figure 4

Coincidence probability of streamﬂow for water transfer risk, China

Hydrology Research

51.5

2020

Conditional coincidence probability of XY–DSDJK given the runoff of SQ gauge.

multivariate copula functions, which provided a valuable refer-

According to the coincidence probability of annual runoff

ence for water resources scheduling of the IBT project because

series from the three gauges, it was found that the acceptable

they covered all primary areas impacted by the project.

and unfavorable probabilities were 44.83 and 55.17%,

The most adverse encounter in all situations was the

respectively. Obviously, the unfavorable probability was

case that three hydrological gauges (WSA, WRA and DIA)

greater than the acceptable probability and should be paid

were in the dry period at the same time, i.e. dry–dry–dry,

greater attention to ﬂood and drought risk management

and the coincidence probability was 26.09% obtained by

during the operation of the project.

the trivariate copula. Accordingly, the recurrence interval

Considering the most adverse encounter situation in

was calculated with a value of 3.8 years, illustrating that

WRA (XY gauge) and DIA (DSDJK gauge) under the con-

this adverse situation would appear about every 4 years.

dition that WSA (SQ gauge) was in the dry period, i.e. dry

1134

X. Wei et al.

Coincidence probability of streamﬂow for water transfer risk, China

XY and dry DSDJK if dry SQ, the conditional coincidence

Hydrology Research

51.5

2020

REFERENCES

probability was computed with a value of 69.57%. Different from the coincidence probability (dry–dry–dry) of trivariate copula, it better emphasized the risk of drought for the water transfer project, representing the probability of drought event simultaneously occurring in WRA and DIA was close to 70% when a drought event occurred in WSA. Therefore, it was inferred that the volume of designed transferable water could not be guaranteed to a large extent once WSA was in a dry period during the operation of HWWTP, which could lead to water shortages caused by drought in WRA and DIA could not be alleviated effectively and could further threaten regional water security. Thus, it is suggested that an emergency water resources scheduling plan and management programs should be drawn up, with the purpose of dealing with potential ﬂood and drought risk of the IBT project.

ACKNOWLEDGEMENTS The authors gratefully acknowledge the ﬁnancial support from the National Natural Science Foundation of China (No. 51979005 and 51809005), the Natural Science Basic Research Program of Shaanxi (No. 2020JM-250) and State Key Laboratory of Eco-hydraulics in Northwest Arid Region, Xi’an University of Technology (No. 2018KFKT-4).

AUTHOR CONTRIBUTIONS All authors contributed to this study. H.Z. provided the writing idea; X.W. carried out data analyses and wrote the first manuscript draft; C. D. and S. S. contributed to the analysis of results; Y.W. drew the figures; H.Z. and V.S. revised the manuscript. All authors read and approved the final manuscript.

DATA AVAILABILITY STATEMENT All relevant data are included in the paper or its Supplementary Information.

Chanda, K., Maity, R., Sharma, A. & Mehrotra, R.  Spatiotemporal variation of long-term drought propensity through reliability-resilience-vulnerability based drought management index. Water Resources Research 50 (10), 7662–7676. doi:10.1002/ 2014WR015703. Chebana, F., Dabo-Niang, S. & Ouarda, T. B. M. J.  Exploratory functional flood frequency analysis and outlier detection. Water Resources Research 48 (4), 4514. doi:10. 1029/2011WR011040. Chen, L., Singh, V. P., Guo, S., Zhou, J. & Zhang, J.  Copulabased method for multisite monthly and daily streamflow simulation. Journal of Hydrology 528 (528), 369–384. doi:10. 1016/j.jhydrol.2015.05.018. Deng, W. B., Liu, W. J., Li, X. X. & Yang, Y. G.  Source apportionment of and potential health risks posed by trace elements in agricultural soils: a case study of the Guanzhong Plain, Northwest China. Chemosphere. doi:10.1016/j. chemosphere.2020.127317. Du, H., Wang, Y., Liu, K. & Cheng, L.  Exceedance probability of precipitation for the Shuhe to Futuan Water Transfer Project in China. Environmental Earth Sciences 78 (7), 1–12. doi:10.1007/s12665-019-8207-2. Fang, H. B., Fang, K. T. & Kotz, S.  The meta-elliptical distributions with given marginals. Journal of Multivariate Analysis 82 (1), 1–16. doi:10.1006/jmva.2001.2017. Favre, A. C., Adlouni, S. E., Perreault, L., Thiemonge, N. & Bobee, B.  Multivariate hydrological frequency analysis using copulas. Water Resources Research 40 (1), 1–12. doi:10.1029/ 2003WR002456. Gringorten, I. I.  A plotting rule for extreme probability paper. Journal of Geophysical Research 68 (3), 813–814. doi:10. 1029/JZ068i003p00813. He, X. C., Kang, L., Cheng, X. J. & Ding, Y.  Flood risk analysis in the middle route of South-to-North Water Diversion Project of China based on Bayesian network. South-to-North Water Diversion and Water Science & Technology 10 (4), 10–13. doi:10.3724/sp.j.1201.2012.04010. Hofert, M.  Sampling Archimedean copulas. Computational Statistics & Data Analysis 52 (12), 5163–5174. doi:10.1016/j. csda.2008.05.019. Hofert, M.  Efficiently sampling nested Archimedean copulas. Computational Statistics & Data Analysis 55 (1), 57–70. doi:10.1016/j.csda.2010.04.025. Hosking, J. R.  L-moments: analysis and estimation of distributions using linear combinations of order statistics. Journal of the Royal Statistical Society Series B – Methodological 52 (1), 105–124. doi:10.1111/j.2517-6161. 1990.tb01775.x. Joe, H.  Multivariate models and dependence concepts. Journal of the American Statistical Association 93 (443). doi:10.2307/2669872.

1135

X. Wei et al.

Coincidence probability of streamﬂow for water transfer risk, China

Joe, H.  Asymptotic efficiency of the two-stage estimation method for copula-based models. Journal of Multivariate Analysis 94 (2), 401–419. doi:10.1016/j.jmva.2004.06.003. Lan, T., Zhang, H., Xu, C. Y., Singh, V. P. & Lin, K.  Detection and attribution of abrupt shift in minor periods in humanimpacted streamflow. Journal of Hydrology 584, 124637. doi:10.1016/j.jhydrol.2020.124637. Li, S., Gu, S., Liu, W., Han, H. & Zhang, Q.  Water quality in relation to the land use and land cover in the Upper Han River basin, China. Catena 75 (2), 216–222. doi:10.1016/j. catena.2008.06.005. Li, S., Liu, W., Gu, S., Cheng, X., Xu, Z. & Zhang, Q.  Spatiotemporal dynamics of nutrients in the upper Han River basin, China. Journal of Hazardous Materials 162 (2), 1340–1346. doi:10.1016/j.jhazmat.2008.06.059. Liu, C. & Zheng, H.  South-to-north water transfer schemes for China. International Journal of Water Resources Development 18 (3), 453–471. doi:10.1080/ 0790062022000006934. Liu, X., Luo, Y., Yang, T., Liang, K., Zhang, M. & Liu, C.  Investigation of the probability of concurrent drought events between the water source and destination regions of China’s water diversion project. Geophysical Research Letters 42 (20), 8424–8431. doi:10.1002/2015GL065904. Manshadi, H. R. D., Niksokhan, M. H. & Ardestani, M.  Water allocation in inter-basin water transfer with the virtual water approach. In: World Environmental & Water Resources Congress: Showcasing the future, American Society of Civil Engineers, Reston, VA, pp. 2510–2521. doi:10.1061/ 9780784412947.247. Massey, F. J.  The Kolmogorov-Smirnov test for goodness of fit. Journal of the American Statistical Association 46 (253), 68–78. doi:10.2307/2280095. Nelsen, R. B.  An Introduction to Copulas. Springer, New York. doi:10.2307/1271100. Plackett, R. L.  A class of bivariate distributions. Journal of the American Statistical Association 60 (310), 516. doi:10.1080/ 01621459.1965.10480807. Salvadori, G. & De Michele, C.  On the use of copulas in hydrology: theory and practice. Journal of Hydrologic Engineering 12 (4), 369–380. doi:10.1061/(ASCE) 10840699(2007)12:4(369). Serinaldi, F., Bonaccorso, B., Cancelliere, A. & Grimaldi, S.  Probabilistic characterization of drought properties through

Hydrology Research

51.5

2020

copulas. Physics & Chemistry of the Earth Parts A/B/C 34 (10–12), 596–605. doi:10.1016/j.pce.2008.09.004. Sklar, C. A.  Fonctions de répartition à n dimensions et leurs marges. Publication de l’Institut de Statistique de l’Université de Pair 8, 229–231. Sraj, M., Bezak, N. & Brilly, M.  Bivariate flood frequency analysis using the copula function: a case study of the Litija gauge on the Sava River. Hydrological Process 29 (2), 225–238. doi:10.1002/hyp.10145. Tian, H., Wang, P., Tansey, K., Zhang, S. & Li, H.  An IPSOBP neural network for estimating wheat yield using two remotely sensed variables in the Guanzhong Plain, PR China. Computers and Electronics in Agriculture 169, 105180. doi:10.1016/j.compag.2019.105180. Wu, L., Bai, T., Huang, Q., Wei, J. & Liu, X.  Multi-objective optimal operations based on improved NSGA-II for Hanjiang to Wei River Water Diversion Project, China. Water 11 (6). doi:10.3390/w11061159. Xie, H. & Wang, K.  Joint-probability methods for precipitation and flood frequencies analysis. Third International Conference on Intelligent System Design & Engineering Applications, Los Alamitos, CA. doi:10.1109/ ISDEA.2012.217. Yan, B. & Chen, L.  Coincidence probability of precipitation for the middle route of South-to-North water transfer project in China. Journal of Hydrology 499, 19–26. doi: 10.1016/j. jhydrol.2013.06.040. Zhang, L. & Singh, V. P.  Bivariate flood frequency analysis using the copula method. Journal of Hydrologic Engineering 11 (2), 150–164. doi:10.1061/(ASCE)1084-0699(2006)11:2(150). Zhang, L. & Singh, V. P.  Gumbel-Hougaard copula for trivariate rainfall frequency analysis. Journal of Hydrology 12 (4), 409–419. doi:10.1061/(ASCE)1084-0699(2007)12:4(409). Zhang, L. & Singh, V. P.  Copulas and Their Applications in Water Resources Engineering. Cambridge University Press, Cambridge, UK. Zhuang, W.  Eco-environmental impact of inter-basin water transfer projects: a review. Environmental Science & Pollution Research. 23 (13), 12867–12879. doi:10.1007/ s11356-016-6854-3. Zou, L., Xia, J. & She, D.  Analysis of impacts of climate change and human activities on hydrological drought: a case study in the Wei River Basin, China. Water Resources Management 32 (4), 1421–1438. doi:10.1007/s11269-017-1877-1.

First received 27 July 2020; accepted in revised form 17 August 2020. Available online 14 September 2020

1136

51.5

2020

The exploration of a Temporal Convolutional Network combined with Encoder-Decoder framework for runoff forecasting Kangling Lin, Sheng Sheng, Yanlai Zhou, Feng Liu, Zhiyu Li, Hua Chen, Chong-Yu Xu

, Jie Chen and Shenglian Guo

ABSTRACT The Temporal Convolutional Network (TCN) and TCN combined with the Encoder-Decoder architecture (TCN-ED) are proposed to forecast runoff in this study. Both models are trained and tested using the hourly data in the Jianxi basin, China. The results indicate that the forecast horizon has a great impact on the forecast ability, and the concentration time of the basin is a critical threshold to the effective forecast horizon for both models. Both models perform poorly in the low flow and well in the medium and high flow at most forecast horizons, while it is subject to the forecast horizon in forecasting peak flow. TCN-ED has better performance than TCN in runoff forecasting, with higher accuracy, better stability, and insensitivity to fluctuations in the rainfall process. Therefore, TCN-ED is an effective deep learning solution in runoff forecasting within an appropriate forecast horizon. Key words

| artiﬁcial neural network, deep learning, Encoder-Decoder architecture, runoff forecasting, Temporal Convolutional Network

Kangling Lin Sheng Sheng Hua Chen (corresponding author) Jie Chen Shenglian Guo State Key Laboratory of Water Resources and Hydropower Engineering Science, Wuhan University, Wuhan 430072, China and Hubei Provincial Key Lab of Water System Science for Sponge City Construction, Wuhan University, Wuhan 430072, China E-mail: chua@whu.edu.cn Yanlai Zhou Chong-Yu Xu Department of Geosciences, University of Oslo, P O Box 1047, Blindern, N-0316, Oslo, Norway

HIGHLIGHTS

• • • • •

For the ﬁrst time, TCN and TCN-ED models are proposed to forecast runoff. TCN-ED has better performance than TCN in runoff forecast in this study.

Feng Liu School of Computer Science, Wuhan University, Wuhan 430072, China

The concentration time is a critical threshold to the effective forecast horizon. Both models perform better in median and high flow than in low flow. It is subject to the forecast horizon for both models to forecast peak flow.

Zhiyu Li Hua Chen Department of Geography and Geographic Information Science, University of Illinois at Urbana-Champaign, Champaign, IL 61801, USA

INTRODUCTION Runoff forecasting is of considerable signiﬁcance to water

operation, ﬂood control, drought relief, and navigation.

resources management. Accurate runoff forecasting can

According to the extent of physical principles, models for

guide the hydraulic engineering construction, reservoir

runoff forecasting can be divided into two categories: process-driven models and data-driven models (Yuan

This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying,

et al. ). Process-driven models represent a speciﬁc

adaptation and redistribution, provided the original work is properly cited

physical process employing experimental formulas before

(http://creativecommons.org/licenses/by/4.0/).

inputting data. Because of the valuable interpretability of

doi: 10.2166/nh.2020.100

1137

K. Lin et al.

TCN combined with Encoder-Decoder framework for runoff forecasting

Hydrology Research

51.5

2020

process-driven models, they have been widely used by

and ant lion optimizer model in the prediction of monthly

hydrologists (Beven et al. ; Ren-Jun ; Douglas-

runoff. These studies prove that the ANNs, especially the

Mankin et al. ; Wang et al. ). However, given the

recurrent neural network (RNN), can capture the non-

uncertainty of the hydrological process and the limitations

linear and non-stationary dynamics of river ﬂow effectively

of artiﬁcially constructed process-driven models, the par-

and stand out among multiple data-driven models.

ameters

and

simulation

process

are

challenging

Nevertheless, the ANN research on runoff forecasting

represent hydrological phenomena fully. Due to process-

mainly focused on the RNN (Chang et al. ), including

driven models’ physical meaning, researchers without a pro-

the LSTM (Hu et al. ) and further improvement based

fessional background cannot improve the models, resulting

on the LSTM (Yuan et al. ). The reason is conjectured

in slow model iteration and difﬁculty in introducing new

to be that the RNN, particularly the LSTM, has a promising

technologies. With the improvement of data availability

ability to process time-series data. However, the develop-

and quality, data-driven models can substitute or sup-

ment of ANNs is changing soon with each passing day,

plement to process-driven models for runoff forecasting

and there are oncoming ANNs worthy of applying for

(Yuan et al. ). Data-driven models focus on the optimal

hydrological modeling and runoff forecasting. The Temporal

mathematical relationships between a forecast object and a

Convolutional Network (TCN) is one of the novel ANNs

predictor, without considering the physical mechanism

designed for sequence modeling and prediction (Bai et al.

(Adamowski & Sun ). Therefore, data-driven models

). TCN is essentially a combination of the one dimen-

are highly transferable.

sion fully convolutional network (1D FCN) and causal

The artiﬁcial neural network (ANN), a popular data-

convolutions (Bai et al. ). The 1D FCN guarantees the

driven model, was applied in runoff modeling at the end

length of the input and output of TCN can be kept the

of the last century already (Hsu et al. ; Smith & Eli

same. The causal convolutions ensure that the future infor-

; Shamseldin ; Dawson & Wilby ). However,

mation will not be used during convolutions. The dilated

due to the limitation of theoretical research and computing

convolutions and residual modules, which dramatically

power at that time, the structure of ANNs was relatively

increase the receptive ﬁeld of the convolutional neural net-

simple. Therefore, ANNs were tough to obtain high

work and make it easier to train, respectively, are also

accuracy results and dwarfed by another representative

used in TCN. With the empirical study demonstrated by

data-driven

(SVM).

Bai et al. (), TCN has been proved superior for sequence

Sivapragasam et al. () used the prediction technique

modeling tasks to LSTM. Recently, more and more studies

based on singular spectrum analysis coupled with SVM to

have conﬁrmed the ability of TCN in time-series data proces-

predict the Tryggevælde catchment runoff data. Further-

sing such as the stock trend prediction, anomaly detection,

more, they used the ﬂood data at Dhaka to demonstrate

and recognition of sepsis (Deng et al. ; He & Zhao

the forecast ability of SVM superior to that of the

; Moor et al. ). Therefore, TCN is introduced for

ANN-based model, particularly at longer lead days (Liong

the ﬁrst time in hydrology by this research to provide

& Sivapragasam ).

more options for runoff forecasting based on ANNs.

model,

support

vector

machine

With the rise of deep learning, hydrologists have turned

Besides, most studies mentioned above considered all

to study the application of advanced deep ANNs in hydrolo-

hydrological information contained in input as equal factors

gical modeling and forecasting (Kisi ; Awchi &

for forecasting the runoff sequences (global relationship)

Srivastava ; Trajkovic ; Abdellatif et al. ; Baba

without further reﬁning inﬂuence factors, therefore missing

et al. ; Shiri et al. ). For instance, Chang et al.

the interactions of the dependent variables (local relation-

() presented an ANN termed real-time recurrent learn-

ship) (Park et al. ). Recently, Kao et al. ()

ing for streamﬂow forecasting in the Da-Chia River. Hu

introduced Encoder-Decoder architecture in ﬂood forecast-

et al. () applied a long short-term memory (LSTM) net-

ing by applying the LSTM based Encoder-Decoder model,

work to the Fen River basin and obtained good prediction

whose output sequence of models stepped into 1- up to

results. Yuan et al. () used the hybrid LSTM network

6-h-ahead. This is the ﬁrst journal article on the application

1138

K. Lin et al.

TCN combined with Encoder-Decoder framework for runoff forecasting

Hydrology Research

51.5

2020

of the Encoder-Decoder architecture in hydrology. The

As can be seen from Figure 1, there are 16 precipitation

Encoder-Decoder architecture is helpful to overcome some

stations, three evaporation stations, and seven runoff stations

obstacles in the application of ANNs on runoff forecasting,

in the study area. The hourly data of these monitoring sites

which has been successfully applied to many other similar

from 2009 to 2015 are obtained from the local Hydrology

themes such as the Shihmen Reservoir ﬂood forecast (Kao

Bureau. The hourly runoff of the Qilijie station, the arithmetic

et al. ), Dadu River runoff forecast (Xu et al. ),

mean of the hourly precipitation, and hourly evaporation

and Yangtze River streamﬂow prediction (Liu et al. ).

among all corresponding stations in the basin, abbreviated

The Encoder-Decoder architecture is comprised of two

hereafter as the ‘runoff’, ‘precipitation’, and ‘evaporation’,

sub-models that work in a symbiotic manner (Loyola et al.

respectively, are used in this study. The runoff concentration

). An encoder captures the local relationship of inﬂu-

time is about 12 h in this basin ( Jie et al. ).

ence factors by encoding the sequence of inputs. Then, a

In order to ease the negative effect of the different scales

decoder extracts the encoded vectors for the runoff forecast-

of data on models’ learning ability, the normal standardiz-

ing by jointly modeling the global and local relationships.

ation is applied to data preprocessing, which is deﬁned as

Therefore, sequential regularities contained in a time series

follows:

can be learned in a more comprehensive, ﬁne-grained way than using a single network (Bian et al. ). The objective of this study is to explore the ability and

X (t) ¼

X(t) X σ

(1)

stability of TCN and the integration of TCN with Encodermulti-step ahead times. To achieve this goal, the remainder

where X (t) is the normal standardization for input data in and σ are the tth time. X(t) is the input sequence, and X

of this study is organized as follows. The study area, hydro-

the mean and standard deviation of the input sequence,

logical data, and the data preprocessing method are given

respectively.

Decoder architecture (TCN-ED) for runoff forecasting with

ﬁrst. TCN, the Encoder-Decoder architecture, the proposed data-driven runoff forecasting model, the comparison model, and metrics for the evaluation are then introduced.

METHODS

Next, the experimental results and thorough discussion are presented. Finally, the conclusions of this study are

Temporal Convolutional Network

summarized. The TCN, based on a 1D convolutional network, is a generic network structure for sequence modeling (Liu et al. ).

STUDY AREA AND DATA

With the empirical study demonstrated by Bai et al. (), TCN has been proved superior performance for sequence

The Jianxi River is the primary up-stream tributary of

modeling tasks to LSTM.

the Minjiang River in southeast China. The Qilijie station

In order to meet the requirements of sequence modeling

(118 180 16″E, 27 010 21″N) is located in the Jianxi River.

tasks, TCN utilizes causal convolutions. Therefore, outputs

The Qilijie station with a drainage area of 14,787 km2 is

are only inﬂuenced by present and past inputs in each

selected for this study (Jie et al. ). The primary soils in

layer (Moor et al. ). In addition, TCN uses a 1D FCN

this area are red, yellow, and paddy soils. The regional climate

structure, whose convolution layers have the same size as

is dominated by southeast Paciﬁc Ocean and southwest

RNNs by adding zero paddings (He & Zhao ).

Indian Ocean subtropical monsoons and partly inﬂuenced

In order to increase the receptive ﬁeld with less compu-

by regional landforms (Tang et al. ). The catchment is

tation cost, dilated convolutions have been introduced to

moist and rainy, with the mean annual rainfall from 1,800

TCN. The ﬁlter of a dilated convolution is applied over a

to 2,200 mm, most of which occurs from March to Septem-

region larger than its size by skipping input values with a

ber. The map of the study catchment is shown in Figure 1.

given

step.

The

dilated

factor

commonly

increases

K. Lin et al.

1139

Figure 1

TCN combined with Encoder-Decoder framework for runoff forecasting

Hydrology Research

51.5

2020

Map of the study catchment.

exponentially with the depth of the network, which ensures

~ tþk is the target. ~ tþ1 , . . . , X X

the receptive ﬁeld covering each input in the history (He & Zhao ). Moreover, TCN applies residual connections that combine the previous input and the result of the convolution with an addition to ease deterioration, which has been

~ tþk ¼ arg max ~ tþ1 , . . . , X X

^ tþk ^ tþ1 ,...,X X

^ tþ1 , . . . , X ^ tþk jXt jþ1 , Xt jþ2 , . . . , Xt ) p(X

(2)

proved very beneficial for deep networks (He et al. ). The encoder is to refine the information contained in the input and fix it into a context value. Encoder-Decoder architecture The Encoder-Decoder architecture was proposed by Cho et al. () who used RNNs both as the encoder and decoder. This architecture is comprised of two ANNs: the first one is to take the information as input and encode it into a context value. The second one is used for decoding the context value into the expected output sequence. The purpose of this architecture is to compress various

~ tþk ≈ arg max ~ tþ1 , . . . , X X

^ tþk ^ tþ1 ,...,X X

^ tþ1 , . . . , X ^ tþk jfencoder (Xt jþ1 , Xt jþ2 , . . . , Xt )) p(X

(3)

Then, the decoder decodes the context value and outputs the ﬁnal prediction. ~ tþk ≈ gdecoder ( fencoder (Xt jþ1 , Xt jþ2 , . . . , Xt )) ~ tþ1 , . . . , X X (4)

information contained in the whole input sequences into a ﬁxed-length vector to make tensor ﬂow more stable (Xu et al. ).

TCN combined with Encoder-Decoder model

If the observation at the time step t is represented by Xt , the sequence to sequence forecasting problem is to predict ^ tþ1 , . . . , X ^ tþk whose length is k, accordthe next sequence X

TCN-ED is proposed in this study, which is compared with the TCN model to verify their ability in runoff forecasting.

ing to the previous j observation Xt jþ1 , . . . , Xt (Shi et al.

The precipitation, evaporation, and runoff for the ﬁrst 48 h

). Among them, the most likely next sequence

are used as input to both models to forecast the runoff for

K. Lin et al.

1140

TCN combined with Encoder-Decoder framework for runoff forecasting

Hydrology Research

51.5

2020

the next 24 h. The framework of the TCN-ED model

context value is copied 24 times, and then enters the

(Figure 2) is explained below.

second TCN (TCN decoder). Since the time step of the

As shown in Figure 2, the sample enters the temporal

second TCN’s input is half of the ﬁrst TCN, the number of

block through 1 × 1 convolution. Each temporal block con-

temporal blocks is correspondingly reduced by one. In this

tains two layers of the causal convolutions. The zero

way, overﬁtting is eased on the premise that the receptive

paddings are added to make the output length between the

ﬁeld of the TCN decoder can still cover the context value

layers the same, and the rectiﬁed linear unit is used to activate

sequence. The output of the TCN decoder at each time step

the output. The input of each temporal block is added to the

performs dimension reduction through two fully connected

output after 1 × 1 convolution with reference to Bai et al.

layers, and ﬁnally outputs the forecast runoff sequence.

(), to increase the nonlinearity of the residual link when

In this study, the output dimensions at each time step of

transferring information across layers. The dilation factor gen-

the TCN encoder and the TCN decoder are 256 and 128,

erally increases exponentially with the depth of the network.

respectively. The output dimensions at each time step of

Regarding the settings of Moor et al. (), for the ﬁrst TCN

the ﬁrst fully connected layer and the second fully con-

(TCN encoder), dilation factors correspond to temporal

nected layer are 64 and 1, respectively. With such a

blocks are 1, 2, 4, 8, and 16, respectively, to ensure that the

setting, the tensor ﬂow among the network layers is hoped

receptive ﬁeld can cover the input sequence. In order to

to be smoother under the premise of giving the models

meet the requirements of the output sequence length, the

enough parameter space.

Figure 2

The framework of the proposed TCN-ED. P, E, and Q represent the precipitation, evaporation, and runoff, respectively. f, k, and d represent the number of ﬁlters, ﬁlter size, and dilation factor, respectively.

1141

K. Lin et al.

TCN combined with Encoder-Decoder framework for runoff forecasting

Flood forecasting model based on the proposed method

Hydrology Research

51.5

2020

128, respectively. The output dimensions of the ﬁrst fully connected layer are 64. In order to output the runoff

To build a dataset to train and test the model, the prepro-

sequence with 24 forecast horizons, the output dimensions

cessed data from 2009 to 2015 are sorted in chronological

of the last fully connected layer are 24.

order, which are further divided into training samples with

In this study, Nash–Sutcliffe efﬁciency (NSE), volu-

a 5-year period and testing samples with a 2-year period.

metric efﬁciency (VE), and coefﬁcient of determination

Therefore, the training and testing samples are provided

(R 2) (Dodge ) are employed as hydrological metrics.

with temporal continuity. Although the runoff concentration

NSE is a popular indicator in runoff forecasting deﬁned by

time is about 12 h in the study basin, the forecast horizons

Nash & Sutcliffe (), but it is oversensitive to the extre-

are set to range from 1 to 24 h to detect the forecast ability

mal ﬂood (Legates & McCabe ; Jiang et al. ). VE

of TCN and TCN-ED during different forecast horizons.

is an alternative indicator focusing on the measurement of

Each model continually trains and optimizes the performance

overall performance that addresses certain limitations of

by reducing the errors in the training set. The Adam optimizer

NSE (Criss & Winston ). The formulae of NSE and

(Kingma & Ba ) and mean square error (MSE) objective

VE are expressed as follows:

function (Allen ) are used in the training stage, and the number of epochs is set to 100. The test set is used to evaluate the performance of the models. The numerical calculations in this paper are accomplished on the supercomputing system in the Supercomputing Center of Wuhan University. Calculating the ﬂow structure of each model is as follows: First, the data

P ^ 2 (Q Q) NSE ¼ 1 P Q)2 (Q VE ¼ 1

P ^ jQ Qj P Q

(5)

(6)

are preprocessed and divided into training data and testing tinuous 48 time step samples and 24 time step targets, input

^ is the predicted runoff, Q is the observed runoff, and where Q Q is the average of the observed runoff. Both indicators range

into the constructed TCN and TCN-ED, with MSE as the

from negative inﬁnity to 1, where 1 represents a perfect ﬁt

objective function, and Adam optimizer for model training

(Jiang et al. ). Four classes are chosen for each criterion:

in 100 epochs. Finally, evaluate the model at each forecast

‘very good’ when >0.66, ‘good’ when 0.33–0.66, ‘average’

horizon using the testing data.

when 0–0.33, and ‘poor’ when <0 (Ecrepont et al. ).

data. Second, the training data are further divided into con-

Since ANNs have strong randomness, all models are Evaluation of the model performance

recalculated 40 rounds (with different random initial weights) (Kao et al. ). With the repeated calculation

In order to objectively illustrate the improvement of the

results, the accuracy, stability, and robustness of each

Encoder-Decoder models compared with the ANN models

model are analyzed. The best model is used to forecast the

without the Encoder-Decoder architecture, TCN is set as

largest ﬂood event in the testing stage, which is determined

the comparison model for TCN-ED with the same input

as the model that produces the highest NSE value averaging

and output. In order to make TCN as similar as possible

over 24 forecast horizons in the testing stages.

to TCN-ED, the comparison model stacks two TCN layers as well. The time steps of the ﬁrst and second TCN layer are the same as the sample’s time steps, so the number of

RESULTS AND DISCUSSION

temporal blocks for both TCN layers is 5. The second TCN connects its output at the last time step to the two

Evaluation and comparison of models performances

fully connected layers for dimension reduction, and ﬁnally outputs the runoff sequence with forecast horizons from

In order to compare the performance of TCN and TCN-ED

t þ 1 to t þ 24. The output dimensions at each time step of

in the learning and forecasting phase, the minimum, mean,

the ﬁrst TCN layer and the second TCN layer are 256 and

and maximum NSE and VE values averaging over 24

K. Lin et al.

1142

TCN combined with Encoder-Decoder framework for runoff forecasting

Hydrology Research

51.5

2020

forecast horizons of each model in the training and testing

indicating that the Encoder-Decoder architecture can

stages with 40 rounds are shown in Table 1.

improve the stability with higher minimum accuracy.

As can be seen from Table 1, TCN and TCN-ED show

For evaluating the performance of each model intuitively

good performance, while TCN-ED maintains higher NSE

and holistically, Figure 3 shows Gaussian kernel density esti-

and VE values than TCN in the training and testing stages.

mation (GKDE) for all NSE and VE values of each model in

Especially for the minimum accuracy, TCN-ED has a

the training and testing stages, respectively. For the GKDE

4.48% improvement rate for NSE and a 4.65% improvement

curve, the accuracy when the density peak appears is the

rate for VE compared with TCN in the testing stages,

mode accuracy of the model, the sharpness of the curve

Table 1

The minimum, mean, and maximum NSE and VE values averaging over 24 forecast horizons of each model in the training and testing stages with 40 rounds

NSE (%)

VE (%)

TCN

TCN-ED

TCN

TCN-ED

Min

Mean

Max

Min

Mean

Max

Min

Mean

Max

Min

Mean

Max

Training

87.47

90.12

93.70

89.37

91.64

94.42

71.05

75.68

80.31

74.79

78.27

81.76

Testing

82.67

87.40

89.15

87.15

88.84

90.22

71.41

77.14

79.47

76.06

79.07

81.14

Figure 3

GKDE for NSE and VE values of TCN and TCN-ED in the (a) training and (b) testing stages.

K. Lin et al.

1143

TCN combined with Encoder-Decoder framework for runoff forecasting

Hydrology Research

51.5

2020

represents the concentration of the accuracy. If the GKDE

The above phenomenon shows that TCN may overﬁt in the

curve is on the right of the X-axis and the shape is sharp, it

training stage, resulting in weaker transferability than TCN-

means that the model has stable and high-precision results.

ED whose GKDE curves change less between the training

As shown in Figure 3, distributions of NSE and VE

stage and the testing stage.

values of the two models are both left-skewed, i.e., the left tails of the GKDE curves are longer, the mass of distri-

Models performance with multi-step ahead times

butions are concentrated on the right, showing both models have a few relatively low values. For each subgraph,

The effective forecast horizons of ANNs have always been

the GKDE curve’s density peak of TCN is higher than that of

the bottleneck and difﬁculty of the artiﬁcial intelligence

TCN-ED. However, the value at the GKDE curve’s density

model research (Zhu et al. ; Cardenas-Barrera et al.

peak of TCN-ED is higher than that of TCN-ED, and the

; Claveria et al. ). In order to compare the accuracy

GKDE curve of TCN-ED has a longer tail in the high end

of TCN and TCN-ED models during different forecast hor-

than that of TCN, causing the GKDE curve of TCN-ED is

izons, boxplots for NSE and VE values in the testing stage

more rightward than that of TCN. Therefore, TCN-ED has

are shown in Figure 4.

a higher mode accuracy and more high-precision results

It can be seen from Figure 4 that both models perform

than TCN. Both models perform better under the VE indi-

very well as their NSE and VE values are very high and

cator from the training stage to the testing stage. However,

stable within t þ 12 forecast horizons, while exceeding the

the NSE GKDE curve’s density peak and the NSE value

t þ 12 forecast horizon, the forecast accuracy of both

at the density peak of TCN drop more than those of TCN-

models decreases gradually with the forecast horizon

ED from the training stage to the testing stage, respectively.

increasing. This shows that the forecast ability of the two

Figure 4

Boxplots for NSE and VE values of TCN and TCN-ED during forecast horizons from t þ 1 to t þ 24 in the testing stage.

K. Lin et al.

1144

TCN combined with Encoder-Decoder framework for runoff forecasting

Hydrology Research

51.5

2020

models during different forecast horizons is closely related

seriously during the short forecast horizons. In comparison

to the concentration time of the basin, which needs to be

to TCN, TCN-ED’s context value which has been reﬁned

further veriﬁed by more cases in more basins. It is evident

already ease the problem caused by redundant information

that TCN-ED’s forecast accuracy and stability are higher

and make TCN-ED more advantageous during the short fore-

than those of TCN at almost each forecast horizon, respect-

cast horizons. In addition, the context value is the output of

ively, especially for short forecast horizons up to t þ 12 and

the last time step of the encoder, so the learning memory is

long horizons close to t þ 24. In the process of the Encoder-

time-sensitive, and TCN-ED is superior obviously to TCN

Decoder, the encoder simulates the process of reading and

during long horizons close to t þ 24.

preprocessing in the brain. The context value with a speciﬁc

In ﬂood control forecasting, more attention is paid to

length symbolizes the formed memory. The decoder rep-

the model’s ability to forecast large-scale ﬂoods. Therefore,

resents the phase when combining known memory and

the best model results of TCN and TCN-ED are used to

new information to react by the brain. Compared with

evaluate the largest ﬂood event, whose peak is maximum

ANNs whose network layers are directly stacked, the Enco-

among 19 ﬂood events in the testing stage at forecast hor-

der-Decoder ANNs are not only more conducive to our

izons t þ 6, t þ 12, t þ 18, and t þ 24, as shown in Figure 5.

understanding of the learning process but also make the

The largest ﬂood event, which is considered moderately

tensor transmission between network layers more efﬁcient

hazardous, has a maximal ﬂow peak reaching 7,043 m3/s in

and stable. Therefore, TCN-ED has higher forecast accuracy

the study period, and the accumulated precipitation in the

and stability than TCN at almost every forecast horizon. For

basin during the ﬂood rising period achieves 134 mm. It

TCN and TCN-ED, in order to keep the output length consist-

can be found from Figure 5 that because there are multiple

ent, the zero paddings are added, resulting in the redundant

rain peaks, the forecast of rising limb and ﬂood peaks by

information which interferes with causal convolutions more

both models are unstable, causing the hydrograph jagged.

Figure 5

The observed and predicted ﬂoods with the maximum peak in the testing stage using the best model results of TCN and TCN-ED during horizons t þ 6, t þ 12, t þ 18, and t þ 24.

1145

K. Lin et al.

TCN combined with Encoder-Decoder framework for runoff forecasting

Hydrology Research

51.5

2020

This situation becomes more and more evident as the

of the observations, respectively, which are drawn on the

forecast horizon increases. In terms of peak time, within

top of each subplot. The percentages of the predictions

t þ 12 forecast horizons, the forecast peak time of each

and their mean R 2 values over the 40 rounds in the ﬁve sec-

model is basically the same as the observed one, but at the

tions are shown on the right side of each subplot. R 2 values

t þ 24 forecast horizon, the models’ forecast peaks appear

of the whole observed and 40 predicted series in the testing

signiﬁcantly later.

stage are calculated, and the maximum, median, and mini-

At the t þ 6 forecast horizon, the forecast runoff curves

mum R 2 values among 40 rounds are shown in each subplot.

of TCN and TCN-ED both ﬁt the observed runoff curve

As Figure 6 shows, both TCN and TCN-ED have high R 2

well, which reﬂects the excellent forecast ability of TCN

values (>0.9) for the whole series at the t þ 6 and t þ 12 fore-

and TCN-ED at the short forecast horizon. However,

cast horizons in the testing stage, while R 2 values decrease a

TCN’s forecast peak ﬂow is later than the observed peak

great deal at the t þ 18 and t þ 24 forecast horizons, especially

ﬂow, which will put more pressure on ﬂood control. In con-

at the t þ 24 horizon. The scatters are closer to the 45-degree

trast, the forecast results of TCN-ED are more practical. At

line at the t þ 6 and t þ 12 forecast horizons, while they are

the t þ 12 forecast horizon, it is obvious that TCN-ED has

farther away from the 45-degree line at the t þ 18 and t þ 24

higher accuracy in the peak ﬂow forecast than TCN, as

forecast horizons. These indicate that the observation and pre-

TCN underestimates the peak ﬂow compared to the

dication scatters have good ﬁtness at the t þ 6 and t þ 12

observed. The forecast runoff of TCN near the peak ﬂow

forecast horizons, and poor ﬁtness at the t þ 18 and t þ 24

has a quick drop, which may be due to TCN being sensitive

forecast horizons, which is consistent with the results above.

to the ﬂuctuation of precipitation. The context value used by

The percentage distribution and R 2 value of each ﬂow

TCN-ED at each time step of decoding is uniformly the

section can further reﬂect the forecast ability of the

output of the last time step of the encoder, so the drastic

models for different magnitude ﬂow. It can be seen from

changes in the rain sequence will not be immediately

each subplot in Figure 6 that the difference of percentage

reﬂected in the output sequence, which is more in line

between the two models in different sections is relatively

with the smooth change process of the runoff. At the fore-

small, and the percentage of TCN-ED is closer to that of

cast horizons t þ 18 and t þ 24, due to the lack of

the observation than TCN, especially in the low and peak

hydrological information, the forecast runoff hydrograph

ﬂow sections. It can be seen from Figure 6 that R 2 values

lags far behind the observed with increasing ahead hours.

of both models in the low ﬂow section are less than 0.3 at

This indicates that both TCN and TCN-ED models may

four forecast horizons, reﬂecting poor ﬁtting between the

gradually lose its forecast ability when the forecast horizons

observation and prediction in the low ﬂow section, while

are beyond the concentration time of the basin.

in the medium and high ﬂow sections, R 2 values are greater than 0.5, indicating that both models have better forecast

Robustness of the models based on regression analysis

ability for median and high ﬂow. It also can be found from Figure 6 that forecast horizons have a great inﬂuence on

In order to explore the robustness of the models, the

the prediction of peak and extreme peak ﬂow. In Figure 6(a)

relationships between the observation and predictions of

and 6(b), R 2 values of both models in the peak ﬂow and

the 40 rounds are drawn at the t þ 6, t þ 12, t þ 18, and

extreme peak ﬂow sections are greater than 0.5, indicating

t þ 24 forecast horizons in the testing stage, which are dis-

that both models have good forecast ability for peak and

played as scatter plots with the KDE curves in Figure 6.

extreme peak ﬂow at the t þ 6 and t þ 12 forecast horizons.

The runoff range is divided into the low ﬂow section

However, R 2 values drop to about 0.3 in Figure 6(c), and

(178 m3/s), medium section (178–550 m3/s), high section

drop below 0.1 in Figure 6(d), indicating that both models

(550–1,374 m /s), peak ﬂow section (1,374–2,885 m /s),

gradually lose the forecast ability to peak and extreme

and extreme peak ﬂow section (>2,885 m3/s) by the 25th,

peak ﬂow with forecast horizons increasing. In terms of

75th, 95th, and 99th percentiles of the observations ( Jiang

the R 2 value of each section in each subplot in Figure 6,

et al. ). The ﬁve sections contain 25, 50, 20, 4, and 1%

TCN-ED performs better than TCN as its higher R 2 value.

K. Lin et al.

1146

Figure 6

TCN combined with Encoder-Decoder framework for runoff forecasting

Hydrology Research

51.5

2020

Relationships between the observed and predicted runoff in the testing stage using the 40 rounds results of TCN and TCN-ED at horizons (a) t þ 6, (b) t þ 12, (c) t þ 18, and (d) t þ 24.

1147

K. Lin et al.

TCN combined with Encoder-Decoder framework for runoff forecasting

Hydrology Research

51.5

2020

According to the above analysis, within an appropriate

further veriﬁed in more basins, which will also help to

forecast horizon, both TCN and TCN-ED have good forecast

improve the forecast ability of ANNs in runoff

ability. For example, the appropriate forecast horizon in this

forecasting.

study is the concentration time (12 h) of the basin. Both

3. In general, TCN-ED has better performance than TCN in

models perform poorly in the low ﬂow section, and good

runoff forecasting in this study. TCN-ED shows better

in the medium and high ﬂow sections, while the forecasting

transferability from the training stage to the testing

for peak ﬂow and extreme peak ﬂow is greatly affected by

stage. NSE and VE values of TCN-ED for most forecast

the variation of the forecast horizons.

horizons are higher, and R 2 values of TCN-ED for most ﬂow sections are higher. In addition, with the unique con-

CONCLUSIONS This study constructs TCN and TCN-ED and demonstrates their applicability in the Jianxi basin, China. In addition, with the comparison of TCN and TCN-ED, the ability of the Encoder-Decoder architecture is further illustrated for the runoff forecasting. First, with the repeated calculation

text value which maintains the continuity of hydrological information, TCN-ED shows better stability and is insensitive to ﬂuctuations in the rainfall process. Liu et al. () proved that the LSTM combined with EncoderDecoder architecture has better performance in streamﬂow forecast than the LSTM. More follow-up studies are needed to verify the advantages of Encoder-Decoder architecture integrated with more ANN models.

results of each model in the training and testing stages, the accuracy and transferability of each model are analyzed. Second, the performance of each model during various forecast horizons in the testing stage is comprehensively

ACKNOWLEDGEMENTS

compared. Third, the best model results at the forecast horizons t þ 6, t þ 12, t þ 18, and t þ 24 are used to analyze the model’s forecast ability for the maximum ﬂoods. Finally, the robustness of the model is explored by the relationships

The study is ﬁnancially supported by the National Key Research and Development Program (2018YFC0407904) and the Research Council of Norway (FRINATEK Project 274310).

between the observation and predictions of the 40 rounds at the t þ 6, t þ 12, t þ 18, and t þ 24 forecast horizons. The major ﬁndings of this study are summarized as follows. 1. The forecast horizon has a signiﬁcant impact on the forecast ability of TCN and TCN-ED. In this study, NSE and VE values of both models are high and stable within t þ 12 forecast horizons. As the forecast horizon increases

DATA AVAILABILITY STATEMENT Data cannot be made publicly available; readers should contact the corresponding author for details.

after the t þ 12 forecast horizon, NSE and VE values decrease rapidly, indicating that the forecast ability of the models becomes poor. Since the concentration time

REFERENCES

of the study basin is about 12 h, it can be inferred that the concentration time is a critical threshold to the effective forecast horizon for both models, which needs to be further demonstrated in more basins. 2. Both models perform poorly in the low flow section, and good in the medium and high flow sections during most forecast horizons, while it is conditional to the forecast horizon in forecasting peak flow for both models. Whether this rule is universal or not also needs to be

Abdellatif, M., Atherton, W. & Alkhaddar, R.  A hybrid generalised linear and Levenberg-Marquardt artiﬁcial neural network approach for downscaling future rainfall in North Western England. Hydrology Research 44 (6), 1084–1101. Adamowski, J. & Sun, K. R.  Development of a coupled wavelet transform and neural network method for ﬂow forecasting of non-perennial rivers in semi-arid watersheds. Journal of Hydrology 390 (1–2), 85–91. Allen, D. M.  Mean square error of prediction as a criterion for selecting variables. Technometrics 13 (3), 469–475.

1148

K. Lin et al.

TCN combined with Encoder-Decoder framework for runoff forecasting

Awchi, T. A. & Srivastava, D. K.  Analysis of drought and storage for Mula project using ANN and stochastic generation models. Hydrology Research 40 (1), 79–91. Baba, A. P. A., Shiri, J., Kisi, O., Fard, A. F., Kim, S. & Amini, R.  Estimating daily reference evapotranspiration using available and estimated climatic data by adaptive neuro-fuzzy inference system (ANFIS) and artificial neural network (ANN). Hydrology Research 44 (1), 131–146. Bai, S., Kolter, J. Z. & Koltun, V.  An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv:1803.01271. Beven, K., Kirkby, M., Schofield, N. & Tagg, A.  Testing a physically-based flood forecasting model (TOPMODEL) for three UK catchments. Journal of Hydrology 69 (1–4), 119–143. Bian, C., He, H., Yang, S. & Huang, T.  State-of-charge sequence estimation of lithium-ion battery based on bidirectional long short-term memory encoder-decoder architecture. Journal of Power Sources 449, 227558. Cardenas-Barrera, J. L., Meng, J. L., Castillo-Guerra, E. & Chang, L. C.  A neural network approach to multi-step-ahead, short-term wind speed forecasting. In: 12th International Conference on Machine Learning and Applications, Miami. Vol. 2, pp. 243–248. Chang, F.-J., Chang, L.-C. & Huang, H.-L.  Real-time recurrent learning neural network for stream-flow forecasting. Hydrological Processes 16 (13), 2577–2588. Cho, K., Van Merrienboer, B., Gulcehre, C., Bougares, F., Schwenk, H. & Bengio, Y.  Learning phrase representations using RNN encoder-decoder for statistical machine translation. arXiv:1406.1078. Claveria, O., Torra, S. & Monte, E.  Modelling tourism demand to Spain with machine learning techniques. The impact of forecast horizon on model selection. Revista De Economia Aplicada 24 (72), 109–132. Criss, R. E. & Winston, W. E.  Do Nash values have value? Discussion and alternate proposals. Hydrological Processes 22 (14), 2723–2725. Dawson, C. W. & Wilby, R.  An artificial neural network approach to rainfall-runoff modelling. Hydrological Sciences Journal 43 (1), 47–66. Deng, S., Zhang, N., Zhang, W., Chen, J., Pan, J. & Chen, H.  Knowledge-driven stock trend prediction and explanation via temporal convolutional network. In: Companion Proceedings of the 2019 World Wide Web Conference, San Franscisco, CA, pp. 678–685. Dodge, Y.  Coefficient of determination. Alphascript Publishing 31 (1), 63–64. Douglas-Mankin, K., Srinivasan, R. & Arnold, J.  Soil and water assessment tool (SWAT) model: current developments and applications. Transactions of the ASABE 53 (5), 1423–1431. Ecrepont, S., Cudennec, C., Anctil, F. & Jaffrézic, A.  PUB in Québec: a robust geomorphology-based deconvolution-

Hydrology Research

51.5

2020

reconvolution framework for the spatial transposition of hydrographs. Journal of Hydrology 570, 378–392. He, Y. & Zhao, J.  Temporal convolutional networks for anomaly detection in time series. Journal of Physics: Conference Series 1213 (4), 042050. He, K., Zhang, X., Ren, S. & Jian, S.  Deep residual learning for image recognition. In: IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, CA, pp. 770–778. Hsu, K. l., Gupta, H. V. & Sorooshian, S.  Artificial neural network modeling of the rainfall-runoff process. Water Resources Research 31 (10), 2517–2530. Hu, C. H., Wu, Q., Li, H., Jian, S. Q., Li, N. & Lou, Z. Z.  Deep learning with a long short-term memory networks approach for rainfall-runoff simulation. Water 10 (11), 1543. Jiang, S., Zheng, Y., Babovic, V., Tian, Y. & Han, F.  A computer vision-based approach to fusing spatiotemporal data for hydrological modeling. Journal of Hydrology 567, 25–40. Jie, M.-X., Chen, H., Xu, C.-Y., Zeng, Q., Chen, J., Kim, J.-S., Guo, S.-l. & Guo, F.-Q.  Transferability of conceptual hydrological models across temporal resolutions: approach and application. Water Resources Management 32 (4), 1367–1381. Kao, I. F., Zhou, Y. L., Chang, L. C. & Chang, F. J.  Exploring a long short-term memory based encoder-decoder framework for multi-step-ahead flood forecasting. Journal of Hydrology 583, 12. Kingma, D. P. & Ba, J.  Adam: a method for stochastic optimization. arXiv:1412.6980. Kisi, O.  River flow forecasting and estimation using different artificial neural network techniques. Hydrology Research 39 (1), 27–40. Legates, D. R. & McCabe Jr, G. J.  Evaluating the use of ‘goodness-of-fit’ measures in hydrologic and hydroclimatic model validation. Water Resources Research 35 (1), 233–241. Liong, S. Y. & Sivapragasam, C.  Flood stage forecasting with support vector machines. Journal of the American Water Resources Association 38 (1), 173–186. Liu, J., Zhu, H., Liu, Y., Wu, H., Lan, Y. & Zhang, X.  Anomaly detection for time series using temporal convolutional networks and Gaussian mixture model. Journal of Physics: Conference Series 1187, 042111. Liu, D., Jiang, W., Mu, L. & Wang, S.  Streamflow prediction using deep learning neural network: case study of Yangtze River. IEEE Access 8, 90069–90086. Loyola, P., Liu, C. & Hirate, Y.  Modeling user session and intent with an attention-based encoder-decoder architecture. In: Proceedings of the Eleventh ACM Conference on Recommender Systems, Como, Italy, pp. 147–151. Moor, M., Horn, M., Rieck, B., Roqueiro, D. & Borgwardt, K.  Early recognition of sepsis with Gaussian process temporal convolutional networks and dynamic time warping. Machine Learning for Healthcare Ann Arbor, USA, 106, 2–26.

1149

K. Lin et al.

TCN combined with Encoder-Decoder framework for runoff forecasting

Nash, J. E. & Sutcliffe, J. V.  River ﬂow forecasting through conceptual models part I – a discussion of principles. Journal of Hydrology 10 (3), 0–290. Park, C., Kim, D. & Yu, H.  An encoder–decoder switch network for purchase prediction. Knowledge-Based Systems 185, 104932. Ren-Jun, Z.  The Xinanjiang model applied in China. Journal of Hydrology 135 (1–4), 371–381. Shamseldin, A. Y.  Application of a neural network technique to rainfall-runoff modelling. Journal of Hydrology 199 (3–4), 272–294. Shi, X. J., Chen, Z. R., Wang, H., Yeung, D. Y., Wong, W. K. & Woo, W. C.  Convolutional LSTM network: a machine learning approach for precipitation nowcasting. Advances in Neural Information Processing Systems 28, 802–810. Shiri, J., Marti, P., Nazemi, A. H., Sadraddini, A. A., Kisi, O., Landeras, G. & Fard, A. F.  Local vs. external training of neuro-fuzzy and neural networks models for estimating reference evapotranspiration assessed through k-fold testing. Hydrology Research 46 (1), 72–88. Sivapragasam, C., Liong, S.-Y. & Pasha, M.  Rainfall and runoff forecasting with SSA–SVM approach. Journal of Hydroinformatics 3 (3), 141–152. Smith, J. & Eli, R. N.  Neural-network models of rainfallrunoff process. Journal of Water Resources Planning Management 121 (6), 499–508.

Hydrology Research

51.5

2020

Tang, Q., He, X., Bao, Y., Zhang, X., Guo, F., Zhu, H., Tang, Q., He, X., Bao, Y. & Zhang, X.  Determining the relative contributions of climate change and multiple human activities to variations of sediment regime in the Minjiang River, China. Hydrological Processes 27 (25), 3547–3559. Trajkovic, S.  Testing hourly reference evapotranspiration approaches using lysimeter measurements in a semiarid climate. Hydrology Research 41 (1), 38–49. Wang, G., Zhang, J., Jin, J., Pagano, T., Calow, R., Bao, Z., Liu, C., Liu, Y. & Yan, X.  Assessing water resources in China using PRECIS projections and a VIC model. Hydrology and Earth System Sciences 16 (1), 231–240. Xu, J., Luo, W. & Huang, Y.  Dadu river runoff forecasting via Seq2Seq. In: Proceedings of the 2019 International Conference on Artiﬁcial Intelligence and Computer Science, Hangzhou, China, pp. 494–498. Yuan, X. H., Chen, C., Lei, X. H., Yuan, Y. B. & Adnan, R. M.  Monthly runoff forecasting based on LSTM-ALO model. Stochastic Environmental Research and Risk Assessment 32 (8), 2199–2212. Zhu, X. T., Wang, H., Xu, L. & Li, H. Z.  Predicting stock index increments by neural networks: the role of trading volume under different horizons. Expert Systems with Applications 34 (4), 3043–3054.

First received 16 July 2020; accepted in revised form 26 August 2020. Available online 21 September 2020

1150

51.5

2020

Impact of urbanization on variability of annual and ﬂood season precipitation in a typical city of North China Peijun Li, Depeng Zuo, Zongxue Xu, Xiaoxi Gao, Dingzhi Peng, Guangyuan Kan, Wenchao Sun, Bo Pang and Hong Yang

ABSTRACT Urbanization plays an important role in a global change, but there are few studies that combine land use with topography and precipitation. The urbanization in Jinan, a typical city of North China, between the 1980s and 2005 was analyzed by transition matrix analysis, and the topographic effects on land use changes were explored considering altitude, slope, and aspect. The temporal trends and abrupt changes of annual and ﬂood season precipitation for the last 60 years were detected by multiple nonparametric detection methods, and the precipitation indices were adopted to characterize the frequency and intensity of precipitation. The relationship between urbanization and precipitation was ﬁnally investigated by grey correlation analysis. The results showed that land use

Peijun Li Depeng Zuo (corresponding author) Zongxue Xu Xiaoxi Gao Dingzhi Peng Wenchao Sun Bo Pang College of Water Sciences, Beijing Key Laboratory of Urban Hydrological Cycle and Sponge City Technology, Beijing Normal University, Beijing 100875, China E-mail: dpzuo@bnu.edu.cn

structure in Jinan experienced a dramatic change during recent decades due to the urbanization process, the conversion of cropland to forest, and the protection of spring in Jinan. The land use types that closely related to mankind’s activities and living behaviors were more concentrated in areas with lower altitude and slope, and detected more significant changes. The significant abrupt changes of precipitation generally occurred in 1989 and concentrated in urban areas and southern mountainous areas with high increment. The increase of flood season precipitation was more significant in the central urban region with a maximum increasing rate of 39.52%, and the increase in the number of storm and precipitation days was also obvious. The precipitation indices in the flood season were more closely related to changes in farmland and settlements affected by urbanization with a maximum correlation coefficient of 0.75. These findings illustrate the impact of urbanization on the variability of precipitation and support that the changes in the urban underlying surface have a certain effect on the surface energy balance. Key words

| abrupt change, land use change, North China, precipitation, topographic effects, urbanization

HIGHLIGHTS

• • •

A significant urbanization process has undergone in Jinan during recent decades. Flood season precipitation showed a more significant upward trend during the last 60 years. The rainstorm days, precipitation amount, and the probability of heavy rain in urban areas have significantly increased.

Guangyuan Kan Research Center on Flood & Drought Disaster Reduction of the Ministry of Water Resources, China Institute of Water Resources and Hydropower Research, Beijing 100038, China Hong Yang Eawag, Swiss Federal Institute of Aquatic Science and Technology, P.O. Box 611, 8600 Dübendorf, Switzerland

1151

•

P. Li et al.

Impact of urbanization on variability of precipitation in a city of North China

Hydrology Research

51.5

2020

The precipitation indices in the ﬂood season were more closely related to changes in farmland and settlements affected by urbanization.

INTRODUCTION The urbanization process is generally accompanied by the

(UCM) to assess the impact of extensive urbanization on

rapid succession of underlying surface (Mahmood et al.

regional precipitation across the Beijing–Tianjin–Hebei

), resulting in changes of topography, thermal dynamic

region of China, which showed that extensive urbanization

conductivity, and hydraulic permeability (Oke ), which

considerably decreased precipitation over and downwind

in turn have an impact on the urban hydrological cycle,

of Beijing city.

such as an increase in regional precipitation and local

Other studies focused on the separation of climate

extreme precipitation, particularly in downwind suburban

change and urbanization on precipitation at different spatial

areas (Pathirana et al. ; Yu et al. ). Urbanization

scales. Gu et al. () separated potential contributions of

not only exacerbates the ﬂood response, but also increases

urbanization and climate change to precipitation trends in

the frequency of heavy rains (Zhang et al. ). The

China at national, regional, and local scales using a ‘trajec-

Metropolitan Meteorological Experiment (METROMEX)

tory’-based method, which concluded that climate change

conﬁrmed that the urban heat island circulation can trigger

was the principle factor for variations of precipitation,

and enhance convective weather, such as thunderstorms,

while urbanization generally has a greater effect on total pre-

heavy precipitation, and strong storms, which especially

cipitation than precipitation extremes. Whether the effect of

has a signiﬁcant impact on urban downwind areas

urbanization on precipitation is increasing or decreasing is

(Changnon ). Furthermore, studies in Netherland

still controversial, and some studies have even reached the

showed that extreme precipitation in urban areas is more

opposite conclusion. The precipitation in the urban down-

intense compared with rural areas, which have a vital connec-

wind direction was suppressed in the study of major urban

tion with the urbanization process (Golroudbary et al. ).

areas and industrial facilities areas, because the addition of

There are kinds of methods for investigating the impact

ice nuclei and cloud condensation nuclei reduced the con-

of urbanization on regional precipitation. For statistical

version efﬁciency of cloud water to rainwater (Rosenfeld

methods, Lu et al. () applied four nonstationary general-

; Givati & Rosenfeld ). Moreover, a decrease in sur-

ized extreme value models to evaluate the impact of

face water storage and in water vapor supply to the upper

urbanization on extreme precipitation in the Yangtze River

atmosphere may also lead to a decrease in precipitation in

Delta metropolitan region, which found that urban

urban areas. The impacts of heat island effect and aerosol

expansion could increase the magnitudes of extreme precipi-

emissions on precipitation may be insufﬁcient, compared

tation and its recurrence levels under different return

with the impact of changes in the hydrological character-

periods. Zhu et al. () analyzed hourly precipitation

istics of the underlying surface (Kaufmann et al. ;

data in Beijing Municipality during the period of 2011–

Wang et al. ).

2015 using the circular statistical analysis and the grange

However, most of the research focused on the ﬁve pre-

causality test technique, which indicated that impacts of

cipitation gauging stations located in the Xiaoqing River

urbanization on precipitation varied with different types of

basin, occupying only a quarter of the total area of the

urbanization. In addition, the climate model and the land

study area, which is not enough to comprehensively reﬂect

surface model are the alternative methods for exploring

the spatial heterogeneity of precipitation in the whole of

the impact of urbanization on regional precipitation. Wang

Jinan. Therefore, more detailed information on the relation-

et al. () coupled the Weather Research and Forecasting

ship between land use change and precipitation in the study

(WRF) model with a single-layer urban canopy model

area should be further investigated. In this study, a variety of

P. Li et al.

1152

Impact of urbanization on variability of precipitation in a city of North China

Hydrology Research

51.5

2020

methods were used to analyze the spatial pattern and tem-

North China Plain (116 110 –117 440 E, 36 010 –37 320 N)

poral trend of precipitation and the changes of land use

(Figure 1). The area under the jurisdiction of the Jinan

types in Jinan during the last 60 years, and the relationship

municipal administrative area is 8,151 km2, including

between land use change and precipitation variation in the

seven districts and two counties. The terrain is high in the

study area was explored. This study tried to reveal the evol-

south and low in the north. The terrain is complex and

ution law between the two environmental variables and

can be divided into three zones: the north is near the

point out the issues that need to be faced in the development,

Yellow River area, the middle is the piedmont plain area,

utilization, and protection of water and soil resources, which

and the south is the hilly mountain area. There are many

is of great signiﬁcance for further urban development and the

rivers in Jinan, mainly the Yellow River and the Xiaoqing

coordination of the water and environment.

River. Jinan belongs to the sub-humid continental monsoon climate and has obvious monsoon characteristics, with an annual average temperature of 14.3

STUDY AREA AND DATA DESCRIPTION

C and an annual

average precipitation of 648.0 mm (Zuo et al. ). The precipitation is extremely uneven, and the inter- and intra-

Study area

annual variability is strong (Chang et al. ). In summer, the average precipitation in various regions is more than

Jinan, the capital of Shandong Province, is located in the

400 mm, which concentrates 60% of the annual precipi-

Midwest of Shandong Province, the downstream region of

tation. The average number of precipitation days in July is

the Yellow River Basin and on the eastern edge of the

about 15 days, and the number of rainstorm days with

Figure 1

Digital elevation model (DEM) and locations of precipitation gauging stations in Jinan.

1153

P. Li et al.

Impact of urbanization on variability of precipitation in a city of North China

daily precipitation above 50 mm is concentrated in July and

Hydrology Research

51.5

2020

METHODOLOGY DESCRIPTION

August, accounting for 70% of the number of rainstorm days throughout the year.

To analyze the relationship between urbanization and

According to statistics, the urban built-up area of Jinan

precipitation in Jinan, urban areas (including municipal

is expanded by 57.94 km2 during the period of 1986–2000

districts of Licheng, Shizhong, Tianqiao, Lixia, and

(Jiang et al. ). With the rapid economic and social

Huaiyin), suburban areas (Zhangqiu and Changqing dis-

development, human activities in Jinan have signiﬁcantly

tricts), and rural areas (including Jiyang, Shanghe, and

intensiﬁed. According to the results of the sixth national

Pingyin) were divided based on GDP levels and distance

census in 2010 (http://www.stats.gov.cn), the permanent

from the urban core. In exploring the topographic effects

resident population of Jinan is 6.184 million, which is an

on the characteristics of spatial distribution of land use

increase of 892,300 compared with 10 years ago, with a

types, a sampling point was set at each 0.01 km2 in the

growth rate of 15.07%. With the processes of urbanization,

study area, and a scatter plot was drawn based on the attri-

population growth and economic development, land use

butes of sampling points. The areas of each land use type

patterns have also undergone tremendous change. Under

were calculated at 50 m increments of elevation and 1

the inﬂuence of the drastic changes in underlying surface

increments of slope.

conditions during the recent decades, large- and mediumsized cities in China have suffered from rainstorm ﬂood and waterlogging, in which Jinan is a typical case (Hammond et al. ; Cheng et al. ).

Land use transition matrix Land use transition matrix is derived from the quantitative analysis of system state and state transition in system

Data description The land use type data of the study area in 1980s and 2005 were provided by the National Science and Technology Foundation Platform Project ‘Data-Sharing Network of Earth System Science’ (www.geodatda.cn). The daily precipitation data at 47 gauging stations during the period of 1950–2016 were collected for further analysis, which was provided by the Jinan Hydrology Bureau. The locations of the precipitation gauging stations are shown in Figure 1, in which 23 ﬂood season gauging stations were mainly used to study the spatiotemporal

analysis. The matrix reﬂects the land use dynamic process information of the mutual transformation between the beginning and end of a certain period at a certain area. It can be used to characterize the structural characteristics of land use change and show the trend and direction of various land-based transformations in a concrete and comprehensive way (Lu et al. ). The general formula is shown in Table 1, in which P11–Pnn is the transfer area between different land use types A1–An during the period from T1 to T2. Sn1 and Sn2 indicate the n-type area in the land use period T1 and the period T2, respectively.

variability of precipitation in the ﬂood season; the remainder of the gauging stations for annual precipitation monitoring was adopted for abrupt change detection and

Table 1

trend analysis of precipitation. The missing data of the precipitation series were filled by neighboring stations through the linear regressive method, in which all the correlation coefficients R 2 > 0.8, which indicates that the filled series can satisfy the quality requirements for this study. Furthermore, the economic and social statisticsrelated administrative regions in the Statistical Yearbook were also used in this study.

The description of land use transition matrix T2

…

Total

P11

P12

…

P1n

S11

P21

P22

…

P2n

S21

…

Pn1

Pn2

…

Pnn

Sn1

Total

S12

S22

…

Sn2

Total area

P. Li et al.

1154

Impact of urbanization on variability of precipitation in a city of North China

Temporal trend and abrupt change detection methods

Hydrology Research

51.5

2020

decreasing trend. When the value of UFk or UBk exceeds a critical straight line, it indicates that the upward or downward trend is signiﬁcant. If two curves appear at the

Nonparametric Mann–Kendall test method

intersection point, and the point appears between the critiThe Mann–Kendall method is a nonparametric statistical

cal lines, then the moment corresponding to the point is

test method used to calculate the changing characteristics

the start time of the abrupt changes (Zuo et al. ).

of hydro-meteorological factors (Mann ; Kendall ). Its advantage is that the sample does not need to follow a certain regular distribution and will not be disturbed by a small number of abnormal values. It is simple to calculate and suitable for type variables and order variables. For a time series x with n sample sizes, construct an ordered sequence:

sk ¼

k X

k ¼ 2, 3, , n

(1)

Nonparametric Pettitt test method The Pettitt method is also a kind of nonparametric test method, which directly uses the order column to detect mutation points (Petitt ). For a time series x with n sample sizes, construct an ordered sequence: 8 < þ1 xi > xj ri ¼ 0 xi ¼ xj : 1 xi < xj

(j ¼ 1, 2, , i)

(5)

i¼1

If in which ri ¼

þ1 xi >xj 0 xi xj

kt0 ¼ Maxjsk j (j ¼ 1, 2, , i)

(2)

Assuming the time series are random and independent, the defined statistics are as follows: [sk E(sk )] ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi UFk ¼ p Var(sk )

(k ¼ 2, 3, , n)

(6)

The time corresponding to t0 is the start time of the abrupt changes. 6k2t0 P ¼ 2 exp 3 n þ n2

! (7)

If P 0.5, the detected abrupt changes are considered to (k ¼ 1, 2, , n)

(3)

Moving T-test method

where UF1 ¼ 0, E(sk) is the mean of the cumulative number sk, and Var(sk) is the variance of the cumulative number sk. n(n 1) E(sk ) ¼ 4 n(n 1)(2n þ 5) Var(sk ) ¼ 72

be statistically signiﬁcant.

The basic test idea of the moving T-test is to deal with the problem of whether there is a signiﬁcant difference between the mean values of the two subsequences in the climate series. When the difference between the mean values of

(4)

Given the signiﬁcance level α is 0.05, check the normal distribution table to determine the critical value u0.05 ¼ ± 1.96. Repeat the above process in reverse order of time series, while making UBk ¼ UFk, k ¼ n, n 1, …, 1, UB1 ¼ 0, and draw a sequence curve diagram of UFk and UBk. If the value of UFk or UBk is greater than 0, it indicates that the sequence is increasing, and conversely, it indicates a

two subsequences exceeds a certain level of signiﬁcance, it is considered that the mean value has changed qualitatively at this moment. For a time series x with n samples, a moment is selected as the reference point, and the samples of x1 and x2 in the two subsequences before and after the reference point deﬁne the statistics t for n1 and n2, respectively: t¼

x1 x2 rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 1 s þ n1 n2

(8)

P. Li et al.

1155

Impact of urbanization on variability of precipitation in a city of North China

Hydrology Research

51.5

2020

in which

which shows that the development and change of the two

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi n1 s21 þ n2 s22 s¼ n1 n2 2

factors are closer. (9)

where s21 and s22 are the variances of the two subsequences. To improve the reliability of the calculation results and avoid the shift of abrupt changes caused by the length of the subsequence, the experimental comparison was performed by changing the length of the subsequence for several times. Finally, the value of n1 and n2 was selected as 7 and the conﬁdence level α was set as 0.05. Therefore, the critical value u0.05 ¼ ±2.18 is determined.

RESULTS AND ANALYSIS Changes in land use types from 1980s to 2005 Spatiotemporal variation of land use types From the point of view of spatial distribution, the maps of land use type in Jinan for the periods of the 1980s and 2005 are shown in Figure 2. Farmland dominates the land use types in Jinan, reaching about 65% of the total

Grey relational analysis

area, mainly distributed in the northern and central plains affected by the terrain. The southern mountainous

Grey relational analysis (GRA) is a dynamic correlation

area is characterized as higher altitudes, the main land

analysis method that quantitatively describes the strength,

use types of which are forests and grasslands. The main

size, and order of the relationship between factors (Tan &

stream of the Yellow River runs through Jinan ﬂowing

Deng ). During the development of the system, if the

from the central region to the northeast. The main

trends of the two factors are consistent, the correlation

urban area of Jinan City is located in the middle. The

between the two factors is higher; otherwise, it is lower.

evolution of construction land was characterized as an

When performing data analysis, x0 is set to be the reference

increase of urban area from east to west and an expansion

sequence and xi is set to be the comparison sequence.

of suburban area in the northwest and the southern

Dimensionless is calculated by means of the mean

mountainous areas.

method, which is transformed into a pure number sequence x00

and

xi0 .

The changes in various land use types between the two

The absolute value of the corresponding point

periods are calculated in Figure 3. The land use structure of

difference constitutes the difference sequence, and the maxi-

Jinan experienced a great change from 1980s to 2005, in

mum difference and the minimum difference between the

which settlements and farmland saw the most obviously

two sequences are identiﬁed. The grey correlation coefﬁ-

change. The area of settlements increased 173.26 km2 with

cient formula is as follows:

a growth rate of 18.5%, while the area of farmland decreased

ξ0i(k) ¼

190 km2. The period from the 1980s to 2005 was the period

min jx0(k) xi(k) j þ ξ max jx0(k) xi(k) j i k i k 0 0 Δ0i(k) þ ξ max jx0(k) xi(k) j i k

(10)

values according to different background requirements.

1 ξ N k¼1 0i(k)

urbanization is advancing rapidly, which reﬂects the impact of urbanization and population growth on land resources. It can be seen from the Annual Statistical Year-

A correlation degree is calculated as follows: γ 0i ¼

of Jinan City’ were promulgated and implemented. The sharp increase in the settlement area indicates that Jinan’s

in which, ξ is the resolution coefﬁcient, 1 > ξ > 0. ξ takes

N X

when the ‘Administrative Measures for the Urban Planning

book of Jinan in 2005 that Jinan’s per capita GDP was 1,263 yuan in 1985, and dramatically reached to 31,606 (11)

yuan in 2005, with an increase of 25 times. The economy of Jinan rapidly developed during this period, and the level

in which γ0i is the correlation degree. The greater the grey

of urbanization gradually increased, which is in line with

correlation degree, the closer the geometric curve shape is,

the trend of land use change.

P. Li et al.

1156

Impact of urbanization on variability of precipitation in a city of North China

Figure 2

The spatial distributions of land use types in Jinan during the periods 1980s and 2005.

Figure 3

Proportions of land use types in Jinan during the periods 1980s and 2005.

Forests are the second most dominant type of land use following farmland, which showed an overall growth

Hydrology Research

51.5

2020

hydrology, but also social development and human activities.

trend during the study period, from 952.56 km2 in the 1980s to 954.12 km2 in 2005, increasing 0.16%. The area

Transition matrix of land use types

of wetlands and water showed an upward trend during the study period, and the growth rate was second only

The land use transition matrix of Jinan from 1980s to 2005

to settlements, reaching 10.12% with a total increase of

in Table 2 explains how the land use types changed in

21.92 km . Grassland and desert areas showed a slight

detail. The transferred area of settlement accounted for

decline with little change overall. The changes in land

17.36% of the total transferred area of all land types in the

use types during the study period were not only affected

study area; the increased area accounted for 32.83% of the

by natural factors such as topography, landforms, and

total increased area of all land use types. The increased

P. Li et al.

1157

Table 2

Impact of urbanization on variability of precipitation in a city of North China

Hydrology Research

51.5

2020

Transition matrix of land use changes in Jinan from the 1980s to 2005 (km2)

2005 1980s

Forest

Grassland

Farmland

Settlements

Wetland water

Desert

Total

Forest

806.96

58.22

79.22

9.62

2.04

0.29

956.34

Grassland

71.81

441.41

74.10

8.67

1.57

0.14

597.70

Farmland

75.22

87.82

4,887.42

344.98

57.47

3.74

5,456.65

Settlements

2.75

3.36

181.63

741.70

6.40

0.46

936.30

Wetland water

0.84

1.11

39.29

3.89

168.11

0.05

213.30

Desert

0.38

0.26

4.51

0.55

0.11

11.11

16.92

Total

957.95

592.19

5,266.15

1,109.41

235.72

15.79

8,177.21

area of settlement was 1.89 times the transferred area, and

Topographic effects on land use changes

the growth rate of which was signiﬁcantly faster than other land use types. More than half of the increased settle-

The scatter plot was used to explore the topographic effects

ment came from the conversion of farmland. The settlement

on the characteristics of spatial distribution of land use types

area converted from farmland reached 345.04 km2, account-

(Figure 4). At the altitude of 400–600 m, the distribution of

ing for 60.62% of the total transferred area of farmland,

forests on each aspect and slope was relatively uniform,

which reﬂects the rapid development of urbanization in

while that was concentrated in the areas with the west

Jinan and the impact of population growth on land

aspect at the altitude of 150–250 m. The grassland was

resources. The transferred area of farmland was 1.5 times

generally distributed around 200 m, which had a good

the increased area of farmland.

adaptability to slope and aspect. The distribution of

The area of wetland and water showed an upward trend

cultivated land characterized as that where most was

during this period, and the increasing rate of 10% ranked

located in the areas with a slope of 0–4 and an altitude of

only second to settlements. During the research period, Jinan

about 0 m. Water bodies were mostly distributed in low-

invested billions of dollars in the treatment and restoration

lying areas with a slope of 0–5 and an altitude of about

of rivers and lakes under the jurisdiction of the municipal gov-

50 m, and most of the settlements were located in the

ernment, which not only improved the water quality, but also

areas of 0–10 and 50–100 m. The distribution of desert

expanded the water area, reﬂecting that the ‘spring city’ has

was relatively loose, which was less affected by topographic

put considerable effort into ecological protection and restor-

factors.

ation. With the issuance and promotion of the ‘Jinan City

The area statistics of land use types according to

Ecological Restoration and Urban Remediation Pilot’, the

aspects are shown in Figure 5. The distributions of farm-

government implemented the ecological restoration and pro-

land and water body were less affected by the aspect,

tection of water and paid more attention to the surrounding

showing a strong adaptability of aspect. Compared with

ecological greening of river banks and reservoir areas. It is

1980s, the area of farmland with easterly aspect had

expected that the wetland and water area will maintain

decreased signiﬁcantly, while the area of water body had

steady growth in the future. The transferred area of forest is

an overall increase, especially for SW aspect. In the

not much different from the increased area, with an overall

1980s, settlements were mainly oriented toward westerly

increase less than 0.1%. Among the newly added areas, the

aspect, while that tended to expand to easterly aspect in

land use types were shifted mostly from farmland, accounting

2005. Most of the forest land was located on W and NW

for 50% of the total new increased area of forest, reﬂecting the

aspect, which had decreased from the 1980s to 2005. The

impact of environmental protection policies such as returning

desert on westerly and southerly aspects showed a signiﬁ-

farmland to forests on land use structure in Jinan City.

cant decrease trend, which experienced an obvious shift

P. Li et al.

1158

Figure 4

Impact of urbanization on variability of precipitation in a city of North China

The distributions of land use types on various aspects and elevations in Jinan during the periods 1980s and 2005.

Hydrology Research

51.5

2020

P. Li et al.

1159

Figure 5

Impact of urbanization on variability of precipitation in a city of North China

Hydrology Research

51.5

2020

Land use changes on different aspects in Jinan during the periods 1980s and 2005.

from westerly aspect to easterly aspect. The area of grass-

implemented the Regulations of Forest Resources Protec-

land in westerly aspect signiﬁcantly decreased during the

tion and Management since 2000, which severely

study period, while that in easterly aspect signiﬁcantly

cracked down on deforestation and encouraged the con-

increased.

version of cropland to forest. Compared with 2000, the

With the rise of elevation, the area of each land

forest coverage increased from 19.7 to 24.1% in 2005,

use type increased ﬁrst and then decreased (Figure 6).

which is consistent with the results reﬂected in the

Most areas of farmland, settlements, wetland and

ﬁgure. The area of settlements signiﬁcantly increased

waters, and deserts were concentrated at 0 m, which

in the elevation belt of 0–150 m, which was also

are the land use types closely related to human activities.

transformed from farmland, indicating an obvious urban-

Forests were mostly distributed at the elevation of 250–

ization process. Wetland and waters and deserts were

350 m, while the peak value of the grasslands area

generally distributed in low-lying areas, which was

appeared at the elevation of about 200 m. The two

strongly constrained by elevation. Compared with the

types of land use were mostly distributed in mountainous

1980s, the area of wetland and waters greatly increased

and hilly areas, and their topographic effect was less

at the elevation of 0 m, while that of deserts signiﬁcantly

than that on the other four types. Compared with the

decreased at the same elevation, indicating a consider-

1980s, the area of forest increased by nearly 30% at

able improvement of the eco-environment in the study

the elevation of 0 m, while the area of farmland

area during recent decades.

decreased by 108.17 km , showing the effect of the

As for the slope effect on land use changes, the dis-

Grain for Green Project in the study area. Jinan has

tributions of forest and grassland were still different

P. Li et al.

1160

Figure 6

Impact of urbanization on variability of precipitation in a city of North China

Land use changes at different elevations and slopes in Jinan during the periods 1980s and 2005.

Hydrology Research

51.5

2020

P. Li et al.

1161

Impact of urbanization on variability of precipitation in a city of North China

Hydrology Research

51.5

2020

from the other four land use types, similar to the

Spatiotemporal variability of ﬂood season and nonﬂood

elevation effect on land use changes. As the slope

season precipitation

increased, the areas of forest and grassland ﬁrst increased and then decreased, and the peak values of

area appeared in the regions with slope of 10 and 5 ,

Spatial patterns of ﬂood season and nonﬂood season precipitation

respectively, which showed a strong suitability of the two land use types in the regions with relatively large

Based on the daily precipitation data at 24 gauging stations

slopes. Compared with the 1980s, the areas of forest

in Jinan during the period of 1966–2016, the spatial distri-

and grassland below 10 signiﬁcantly decreased in 2005.

butions of average annual and ﬂood season precipitation

Farmland, settlements, and wetland and waters were

are shown in Figure 7. Both spatial patterns are character-

mainly concentrated in the ﬂat areas with slopes of 1–2 ,

ized as decreasing from southeast to northwest and

due to the fact that areas with slopes exceeding 5 are unsui-

southwest. The maximum average annual precipitation is

table for construction and farming activities. The area of

759.54 mm at the Zaolin Station in the southern mountai-

desert was small and scattered with stochastic distribution.

nous area, while the minimum value is 557.39 mm at the

Generally speaking, as the slope increases, the areas of var-

Baiyunhu Station in the east. The maximum value of the

ious land use types that are closely related to mankind’s

average precipitation during the ﬂood season appeared at

producing activities and living behaviors will gradually

the Wupu Station, also located in the southern mountainous

decrease, while the distribution of land use types that are

area, reaching 579.58 mm, while the minimum value

less disturbed by humans gradually has the advantage,

appeared at the Baiyunhu Station with 408.48 mm.

showing a transition process from artiﬁcial ecosystems to natural ecosystems.

Figure 7

The spatial distributions of average AP and ﬂood season precipitation in Jinan.

The annual and ﬂood season precipitation in mountainous areas is generally higher than that in the plain areas.

P. Li et al.

1162

Impact of urbanization on variability of precipitation in a city of North China

Hydrology Research

51.5

2020

The settlements in concentrated areas are mostly in areas

Changes in precipitation between pre-change and post-

with relatively less precipitation. Compared with the

change point periods

characteristics of the terrain in the study area, the spatial distribution of average annual and ﬂood season precipitation

According to the results of the abrupt change detection of

in Jinan has a signiﬁcant relationship with the topography.

precipitation, the study period was divided into two subperiods: Period I: 1966–1989 (before the change point)

Abrupt changes of the ﬂood season and nonﬂood season precipitation

and Period II: 1990–2016 (after the change point). The spatial variabilities in the differences between the prechange and post-change points for annual precipitation in

To comprehensively analyze and validate the abrupt

the study area are calculated and shown in Figure 9(a).

changes of ﬂood season and nonﬂood season precipitation

For the spatial variability of ﬂood season precipitation, the

in Jinan during the period of 1966–2016, the widely used

recorded data from the 23 ﬂood season stations were incor-

nonparametric Mann–Kendall test method, the Pettitt test

porated to improve the spatial heterogeneity, which is

method, and the moving T-test method were adopted to

shown in Figure 9(b). Since the rainfall stations in the cen-

detect the abrupt changes of ﬂood season and nonﬂood

tral city are mostly ﬂood season stations, the ﬂood season

season precipitation at 47 gauging stations in the study

data can well reﬂect the actual precipitation distribution.

area. The comparison between the abrupt change detection

The annual and ﬂood season precipitation in the study

results obtained by the three different methods showed that

area have signiﬁcantly increased during the post-change

the annual precipitation at all stations exhibited an upward

point period with different extents. The increase of annual

trend. The abrupt change points generally occurred in the

precipitation and ﬂood season precipitation were similar

1960s, 1989, and 2002, among which most abrupt change

in spatial distributions. The increasing rates of the annual

points in 1989 were detected at the signiﬁcance level 0.05

and ﬂood season precipitation in the eastern and southwes-

(Figure 8). Most sites of the detected mutations are concen-

tern plain regions were relatively low with less than 10%,

trated in the urbanized areas of central Jinan and the

while the increasing rate of that in the central urban

southern mountainous areas, corresponding to the strong

region

change of land use structure during the rapid development

Especially for the Liujiazhuang Station close to the central

of Jinan.

city, the average ﬂood season precipitation increased from

Figure 8

The abrupt change points of annual and ﬂood season precipitation in Jinan.

and southern

mountain

was

relatively high.

P. Li et al.

1163

Figure 9

Impact of urbanization on variability of precipitation in a city of North China

Hydrology Research

51.5

2020

The amplitudes of annual (a) and ﬂood season (b) precipitation during the pre-change and post-change periods in Jinan.

370.4 to 516.8 mm, which indicates that Jinan faced con-

were concentrated in the urban areas of central Jinan. The

siderable risk from ﬂood disasters during the ﬂood season

Xiaolipu Station, located near the central urban area,

in the central urban region over recent years.

increased from 22 to 33.5 days. The increase in the

To comprehensively study the characteristics of the

number of storm days throughout the year is obvious, and

variability of precipitation, two more indices, number of

the annual precipitation generally increased, which is

precipitation days and precipitation intensity, were adopted

shown in Figure 11. The proportion of rainstorm days at

for further analysis. The daily rainfall above 50 mm is

most stations has increased at different rates during the

counted as an indicator of storm intensity. To avoid the

ﬂood season, and the stations with larger increases are con-

impact of the initial year difference, the precipitation data

centrated in central Jinan. The largest increase of rainstorm

of each site from 1980 to 2016 were selected for further

days was at the Guanying Station in the southern mountai-

analysis.

nous region. The proportion of heavy rain in the ﬂood

As can be seen from Figure 10, the number of precipi-

season increased from 3.1 to 7%, and the proportion of

tation days throughout the year and ﬂood season in each

heavy rain in the Xinglong Station, located in the central

rainfall station generally showed an increasing trend. The

urban area of Jinan, also increased by nearly 4%. The

spatial distribution of average annual precipitation days is

increase in the number of precipitation days at stations

similar to that of precipitation days in the ﬂood season.

close to the urban area was signiﬁcantly higher than the

The stations with a larger increase of annual precipitation

ﬂood season, and the increase in precipitation indices

days were mostly concentrated in the southern mountainous

during the ﬂood season was also signiﬁcantly higher than

areas, and the Nanergao Station increased from 55.8 to 77.4

that of the whole year. It can be inferred that the threat of

days. In the ﬂood season, the stations with a larger increase

ﬂood disasters to the urban area of Jinan continues to

1164

Figure 10

P. Li et al.

Impact of urbanization on variability of precipitation in a city of North China

The changes of annual and ﬂood season PDs during the pre-change and post-change periods in Jinan.

Hydrology Research

51.5

2020

1165

Figure 11

P. Li et al.

Impact of urbanization on variability of precipitation in a city of North China

Hydrology Research

51.5

2020

The amplitudes of rainstorm proportion during the pre-change and post-change periods in Jinan.

increase after the abrupt change year. The analysis of the precipitation days and precipitation intensity showed a significant increasing trend for annual and flood season scales, which indicates that beside the total amount, the frequency and intensity of precipitation in the study area also increased to some extent. Impact of urbanization on precipitation The arithmetic average of each index in the region is calculated to obtain the variabilities of land use types and precipitation indices between the pre-change point and post-change point periods (Figure 12). Six precipitation indices, including annual precipitation (AP), precipitation days (PDs), rainstorm days (RDs), AP in the flood season (AP_F), PDs in the flood season (PD_F), and RDs in the

Figure 12

The relationship between precipitation indices and land use changes in different regions of Jinan.

ﬂood season (RD_F), were adopted in this study to describe the characteristics of precipitation. AP describes the

variabilities from frequency, and RD describes the variabil-

variabilities from rainfall amount, PD describes the

ities from intensity.

1166

P. Li et al.

Impact of urbanization on variability of precipitation in a city of North China

Hydrology Research

51.5

2020

As can be seen from Figure 12, the increase of settlement in the urban area is the largest with an increasing rate of 33.01%, followed by wetland water area with an increase of 25.94%, while the farmland in the urban area showed a signiﬁcant decrease of 9.13%. The increase of urbanization is highest in the urban area, followed by suburban areas. According to the statistics of the Statistical Yearbook of Jinan City in 2005, the GDP per capita in the urban area is 44,805.6 yuan, Zhangqiu district is 22,483.6 yuan, Changqing district is 22,117.6 yuan, and Pingyin, Jiyang, and Shanghe counties are 20,992.6, 15,991.7, and 12,483.3 yuan, respectively. The level of per capita GDP in each region is consistent with the increase in the land use

Figure 13

Grey correlation analysis of precipitation indices and land use changes in Jinan.

type of settlements, indicating that the socioeconomic development is generally corresponds to the land use change.

and desert had the least impact on AP_F, with a correlation

Precipitation indicators show that the growth rates of

coefﬁcient of 0.5. The ﬂuctuation of RD_F was directly related

various precipitation indices in the urban area were generally

to the changes of cultivated land and settlements, with corre-

the most signiﬁcant. Among them, the number of rainy days

lation coefﬁcients of 0.73 and 0.7, respectively, while the

in the ﬂood season increased by 58.95%, and the amount of

lowest correlation occurred in the grassland area. Wetland

precipitation in the ﬂood season increased by 18.96%. Com-

and water, and settlements had a greater impact on PD than

pared with the average value of other areas in Jinan, the

other land use types, and the changes in desert were less

number of rainy days in the central district increased by

synchronized with that in PD. Generally speaking, the precipi-

15.38%, and the probability of heavy rains increased by

tation indices associated with the ﬂood season in the study

4.73%, showing typical characteristics of the ‘rain island

area were more closely related to land use changes. Changes

effect’. The number of rainy days increased by 14.48% in

in farmland and settlements affected by urbanization were

Changqing District, while the number of heavy rains

more closely related to precipitation indices.

increased by 57.61% in Zhangqiu District. Although the

The above analysis shows that in the context of further

settlement area of Pingyin has increased signiﬁcantly, the

development of the urban economy and society and drastic

actual urbanized area is relatively small, resulting in little

changes in land use types, the precipitation in urban areas

impact on precipitation. As for some stations in the south

during the ﬂood season is more signiﬁcantly affected com-

of Jiyang, precipitation indicators are generally affected

pared with other regions. This phenomenon may be the

because they are closer to the central urban area.

result of a combination of the heat island effect, the blocking

In order to further explore the relationship between land

effect, and the condensation effect (Mohammad et al. ).

use types and precipitation factors, GRA was used to calculate the correlation coefﬁcient in each area. The indicators were ﬁrstly listed in order of the economic level of each district

DISCUSSION

from the largest to the smallest, and dimensionless was then carried out, taking the indicator of urban area as the initial

Different statistical test methods have their own advantages

value. The calculation results after dimensionless processing

and disadvantages, and the cross-veriﬁcation of the results

are shown in Figure 13. The changes in land use and precipi-

obtained by the three abrupt change test methods in this

tation indices in different administrative areas were well

study could minimize the instability and maximize the

synchronized, with a range of correlation coefﬁcients between

rationality of the determination of the abrupt change

0.5 and 0.75. Changes in the settlement had the greatest impact

point. In previous studies, it was found that the change in

on AP_F, with a correlation coefﬁcient of 0.75, while grassland

reﬂection or absorption of solar radiation on the urban

1167

P. Li et al.

Impact of urbanization on variability of precipitation in a city of North China

Hydrology Research

51.5

2020

underlying surface has a certain effect on the surface energy

the spatial pattern and temporal trends of precipitation

balance, which is one of the physical mechanisms by which

and the changes of land use types in Jinan during the last

land use changes affect regional precipitation (Oke ;

60 years. Conclusions can be summarized as follows:

Arnfield ). Moreover, studies in Beijing and Tokyo showed that the increase of surface impermeable area

1. The land use types in Jinan experienced great changes

ratio makes the local evaporation decrease, the sensible

during the period 1980–2005, in which settlements and

heat ﬂux increases, and the water vapor mixing in the

farmland were the most obvious changing types. The

boundary layer becomes more uniform, which affects the

area of settlements converted from farmland accounted

development of the local weather system (Zhang et al.

for 60.6% of the total transfer area of farmland, reﬂecting

; Souma et al. ). There are some previous studies

the rapid development of urbanization in Jinan and the

on simulating the impact of land use change on rainfall.

impact of growing population on land resources. It is

For example, the WRF model and UCM are recently used

worth mentioning that the growth rate of wetland water

in the ﬁeld; however, it requires large amounts of computing

in Jinan ranked only second to the settlements, which

resources. More sophisticated models with long-term simu-

reﬂects the great efforts for ecological protection and

lations are required in the future to simulate the speciﬁc

restoration.

spatial correlation between land use change and precipitation (Miao et al. ; Wang et al. ).

2. The distribution of land use was characterized as a certain terrain gradient under the inﬂuence of topographic

Through the research, it could be found that the distri-

factors. In lower altitude and slope areas, settlements,

bution of cultivated land had a signiﬁcant relation with the

farmland, and wetland and water were more concen-

requirements of crop growth and the convenience of cultiva-

trated,

tion, which was obviously controlled by the topographic

producing activities and living behaviors. The forest and

factors. In addition to urbanization, the transferred area of

grassland show a strong adaptability of higher altitudes

farmland is also affected by environmental protection pol-

and steeper slopes, accounting for a larger proportion

icies such as returning farmland to forest and returning

of land use types in the areas. The inﬂuences of aspect

farmland to lake. It is important to note that our study focused

on the distribution of various land use types were not

on the local recycled precipitation instead of large-scale

obvious.

which

are

closely

mankind’s

advected precipitation which is beyond the scope of this

3. The spatial pattern of precipitation in Jinan is generally

study but deserves further research. For example, Daniels

characterized as decreasing from southeast to northwest,

et al. () indicated that the impact of land use change on

which is mainly affected by monsoon and topography.

summer precipitation in the Netherlands was smaller than

The temporal variability of precipitation is obvious, and

that of climate change, but it cannot be ignored. Similar ﬁnd-

the intra-annual distribution is uneven. The average AP

ings were found in the Beijing–Tianjin–Hebei region, about

and ﬂood season precipitation in Jinan showed an overall

11% of the reduction of the total precipitation was caused

increasing trend. The signiﬁcant abrupt change point of

by the decrease of local recycled precipitation induced by

precipitation generally occurred in 1989 detected by

urbanization, while the remainder 89% was caused by the

three methods. The growth rate of precipitation in eastern

reduction of large-scale advected precipitation (Li et al.

and western regions was relatively low, while that in the

). It remains a great challenge to clearly elucidate the

central urban region was relatively high. The frequency

independent effects of land use change on climate.

and intensity of precipitation in Jinan also increased signiﬁcantly. 4. Compared with other regions, the urban region with the

CONCLUSIONS

largest area of settlements produced the largest increase of precipitation especially for the post-change period,

In this study, the impact of urbanization on annual and

and the water and wetland area have also increased sig-

ﬂood season precipitation was qualitatively analyzed from

niﬁcantly. The increase of the number of RDs and the

1168

P. Li et al.

Impact of urbanization on variability of precipitation in a city of North China

amount of precipitation during the flood season in the urban area were much higher than that in other areas. The changes in land use and precipitation indices in different administrative areas were well synchronized. The precipitation indices in the flood season were more closely related to changes in farmland and settlements affected by urbanization with a maximum correlation coefficient of 0.75. The results obtained in this study are of great significance for the further urban development and the coordination of the water and environment.

ACKNOWLEDGEMENTS This study is jointly supported by the National Key Research and

Development

Program

China

(Grant

No.

2017YFC1502703), the Beijing Natural Science Foundation (Grant No. 8202030), the National

Natural Science

Foundation of China (Grant No. 91647202), Major Science and Technology Program for Water Pollution Control and Treatment (Grant No. 2017ZX07302-04), and 111 project (B18006).

CONFLICT OF INTEREST The authors declare that they have no conﬂict of interest.

DATA AVAILABILITY STATEMENT All relevant data are available from an online repository or repositories (National Science and Technology Foundation Platform Project ‘Data-Sharing Network of Earth System Science’ (www.geodatda.cn)).

REFERENCES Arnﬁeld, A. J.  Two decades of urban climate research: a review of turbulence, exchanges of energy and water, and the urban heat island. International Journal of Climatology 23 (1), 1–26. Chang, X., Xu, Z., Zhao, G., Cheng, T. & Song, S.  Spatial and temporal variations of precipitation during 1979–2015 in

Hydrology Research

51.5

2020

Jinan City, China. Journal of Water and Climate Change 9 (3), 540–554. Changnon, S. A.  Rainfall changes in summer caused by St. Louis. Science 205 (4404), 402–404. Cheng, T., Xu, Z., Hong, S. & Song, S.  Flood risk zoning by using 2D hydrodynamic modeling: a case study in Jinan city. Mathematical Problems in Engineering 5659197. Daniels, E., Lenderink, G., Hutjes, R. & Holtslag, A.  Relative impacts of land use and climate change on summer precipitation in the Netherlands. Hydrology and Earth System Sciences 20 (10), 4129–4142. Givati, A. & Rosenfeld, D.  Quantifying precipitation suppression due to air pollution. Journal of Applied Meteorology 43 (7), 1038–1056. Golroudbary, V. R., Zeng, Y., Mannaerts, C. M. & Su, Z. B.  Detecting the effect of urban land use on extreme precipitation in the Netherlands. Weather and Climate Extremes 17, 36–46. Gu, X., Zhang, Q., Singh, V. P., Song, C., Sun, P. & Li, J.  Potential contributions of climate change and urbanization to precipitation trends across China at national, regional and local scales. International Journal of Climatology 39 (6), 2998–3012. Hammond, M. J., Chen, A. S., Djordjevic, S., Butler, D. & Mark, O.  Urban flood impact assessment: a state-of-the-art review. Urban Water Journal 12 (1 SI), 14–29. Jiang, L. G., Nie, X. H. & Liu, E. F.  Analysis on the spatial structure of urban land use in Jinan city. Economic Geography 23 (1), 70–73 (in Chinese). Kaufmann, R. K., Seto, K. C., Schneider, A., Liu, Z., Zhou, L. & Wang, W.  Climate response to rapid urban growth: evidence of a human-induced precipitation deficit. Journal of Climate 20 (10), 2299–2306. Kendall, M. G.  Rank Correlation Methods, 4th edn. Charles Griffin, London, UK. Li, X., Zhang, X., Zhang, L. & Zheng, J.  The varied impacts of land use/cover change on summer precipitation over eastern China. Geographical Research 36, 1233–1244. Lu, M., Xu, Y., Shan, N., Wang, Q., Yuan, J. & Wang, J.  Effect of urbanization on extreme precipitation based on nonstationary models in the Yangtze River Delta metropolitan region. Science of the Total Environment 673, 64–73. Lu, D., Gao, G., Lu, Y., Xiao, F. & Fu, B.  Detailed land use transition quantification matters for smart land management in drylands: an in-depth analysis in northwest China. Land Use Policy 90, 104356. Mahmood, R., Pielke, R. A. S., Hubbard, K. G., Niyogi, D., Dirmeyer, P. A., Mcalpine, C., Carleton, A. M., Hale, R., Gameda, S., Beltran-Przekurat, A., Baker, B., Mcnider, R., Legates, D. R., Shepherd, M., Du, J., Blanken, P. D., Frauenfeld, O. W., Nair, U. S. & Fall, S.  Land cover changes and their biogeophysical effects on climate. International Journal of Climatology 34 (4), 929–953. Mann, H. B.  Non-parametric test against trend. Econometrical 13, 245–259.

1169

P. Li et al.

Impact of urbanization on variability of precipitation in a city of North China

Miao, S., Chen, F., Li, Q. & Fan, S.  Impacts of urban processes and urbanization on summer precipitation: a case study of heavy rainfall in Beijing on 1 August 2006. Journal of Applied Meteorology and Climatology 50 (4), 806–825. Mohammad, P., Goswami, A. & Bonafoni, S.  The impact of the land cover dynamics on surface urban heat island variations in semi-arid cities: a case study in Ahmedabad city, India, using multi-sensor/source data. Sensors 19 (17), 3701. Oke, T. R.  The energetic basis of the urban heat-island. Quarterly Journal of the Royal Meteorological Society 108 (455), 1–24. Pathirana, A., Denekew, H. B., Veerbeek, W., Zevenbergen, C. & Banda, A. T.  Impact of urban growth-driven land use change on microclimate and extreme precipitation – a sensitivity study. Atmospheric Research 138, 59–72. Petitt, A. N.  A non-parametric approach to the change point problem. Applied Statistics 28, 125–135. Rosenfeld, D.  Suppression of rain and snow by urban and industrial air pollution. Science 287 (5459), 1793–1796. Souma, K., Tanaka, K., Suetsugi, T., Sunada, K., Tsuboki, K., Shinoda, T., Wang, Y., Sakakibara, A., Hasegawa, K., Moteki, Q. & Nakakita, E.  A comparison between the effects of artiﬁcial land cover and anthropogenic heat on a localized heavy rain event in 2008 in Zoshigaya, Tokyo, Japan. Journal of Geophysical Research-Atmospheres 118 (20), 11600–11610. Tan, X. & Deng, J.  Grey relational analysis: a new method for multivariate statistical analysis. Statistical Research 11 (3), 46–48 (in Chinese).

Hydrology Research

51.5

2020

Wang, J., Feng, J. & Yan, Z.  Impact of extensive urbanization on summertime rainfall in the Beijing region and the role of local precipitation recycling. Journal of Geophysical Research-Atmospheres 123 (7), 3323–3340. Yu, M., Miao, S. & Zhang, H.  Uncertainties in the impact of urbanization on heavy rainfall: case study of a rainfall event in Beijing on 7 August 2015. Journal of Geophysical Research-Atmospheres 123 (11), 6005–6021. Zhang, C. L., Chen, F., Miao, S. G., Li, Q. C., Xia, X. A. & Xuan, C. Y.  Impacts of urban expansion and future green planting on summer precipitation in the Beijing metropolitan area. Journal of Geophysical Research-Atmospheres 114, D02116. Zhang, W., Villarini, G., Vecchi, G. A. & Smith, J. A.  Urbanization exacerbated the rainfall and ﬂooding caused by hurricane Harvey in Houston. Nature 563 (7731), 384. Zhu, X., Zhang, Q., Sun, P., Singh, V. P., Shi, P. & Song, C.  Impact of urbanization on hourly precipitation in Beijing, China: spatiotemporal patterns and causes. Global and Planetary Change 172, 307–324. Zuo, D., Xu, Z., Yao, W., Jin, S., Xiao, P. & Ran, D.  Assessing the effects of changes in land use and climate on runoff and sediment yields from a watershed in the Loess Plateau of China. Science of the Total Environment 544, 238–250. Zuo, D., Cai, S., Xu, Z., Li, F., Sun, W., Yang, X., Kan, G. & Liu, P.  Spatiotemporal patterns of drought at various time scales in Shandong Province of Eastern China. Theoretical Applied Climatology 131 (1–2), 271–284.

First received 11 June 2020; accepted in revised form 28 August 2020. Available online 1 October 2020

1170

51.5

2020

Effect of initial plant density on modeling accuracy of the revised sparse Gash model: a case study of Pinus tabuliformis plantations in northern China Yiran Li, Xiaohua Liu, Chuanjie Zhang, Zedong Li, Ye Zhao and Yong Niu

ABSTRACT An accurate quantitative description of interception is necessary to understand regional water circulation. The revised sparse Gash model (RSGM) is currently used to estimate interception loss. Previous studies have proven that changes in initial plant density, which are caused by thinning, affect the accuracy of RSGM; however, the direct effect of initial density on modeling accuracy remains poorly understood because few studies have collected field data of the same species with various initial densities under similar site conditions. Therefore, six Pinus tabuliformis Carr. plantations with various initial densities were assessed from May to October 2016 in northern China. In summary, RSGM performs better with higher initial densities, and it cannot be suitably applied for plantations with lower initial densities, with the relative error ranging from 18.38 to 53.03%. Sensitivity analysis indicated that the predicted interception is highest sensitive to canopy structure, irrespective of initial density. The influence of climate parameters on simulated results decreased, as initial density increased. These support the notion that amending the representation of the canopy structure in the model and improving the estimation methods for determining the evaporation rate in open canopies can improve accuracy, and that the use of RSGM must first involve the consideration of initial density. Key words

| initial plant density, interception loss, revised sparse Gash model, sensitivity analysis, simulation accuracy

Yiran Li† Zedong Li Yong Niu (corresponding author) College of Forestry, Shandong Agricultural University, Shandong 271018, China E-mail: niuyong1988@126.com Yiran Li† Xiaohua Liu† School of Soil and Water Conservation, Beijing Forestry University, Beijing 100083, China Ye Zhao College of Biological Sciences and Technology, National Engineering Laboratory for Tree Breeding, Beijing Forestry University, Beijing 100083, China Chuanjie Zhang† College of Water Conservancy and Civil Engineering, Shandong Agricultural University, Shandong 271018, China †

HIGHLIGHTS

• • • • •

Application of the revised sparse Gash model (RSGM) in six Chinese pine plantations of different initial densities. RSGM still does not perform well at the sparse initial densities. Changes in initial plant density can affect the simulation accuracy of RSGM. Discuss the possibility of perfecting RSGM to improve its accuracy. Explore the changes in sensitivity of the parameters in RSGM under different initial densities.

Y. R. Li, X. H. Liu and C. J. Zhang should be considered co-ﬁrst authors

1171

Y. Li et al.

Initial plant density effects modeling precision

Hydrology Research

51.5

2020

GRAPHICAL ABSTRACT

INTRODUCTION The development of plantations has continued to increase in

(RSGM) (Valente et al. ; Limousin et al. ), and the

recent years in China (Gao et al. ; Ma et al. ), with

so-called RS-Gash model (Cui & Jia ). Prior studies

the area under plantation increasing to 693,338 km2

have mainly focused on the applicability of either the

(National Forestry and Grassland Administration ).

original Gash model or its revised versions in plantations,

Nevertheless, several underlying issues, such as poor

natural forests, or shrub forests (Sadeghi et al. ;

quality of plantations, have emerged because of the rapid

Fernandes et al. ; Ghimire et al. ; Fathizadeh et al.

increase in green areas. With abundant forests in northern

). Similarly, some previous studies have evaluated the

China, water resources are considered a crucial restrictive

effect of rainfall characteristics or model parameters on

factor for plantation growth. In these areas with limited

the prediction accuracy of the model using only a single-

water resources, a good understanding of ecohydrological

tree species in a particular area (Sadeghi et al. ; Su

processes plays a critical role in the essential components

et al. ). In addition, despite some research on the

of water resources management and stand structure

response of model accuracy to tree density, available

optimization.

information is limited.

A substantial amount of precipitation can be intercepted

Tree densities, tree species, and canopy spatial arrange-

by and can subsequently evaporate from forest canopies

ments are signiﬁcant in improving the prediction of

(van Dijk & Bruijnzeel a; Hassan et al. ). This

interception loss (Owens et al. ; Bui et al. ; Chen

intercepted precipitation plays a crucial role in both regional

et al. ). Managing initial density could be more effective

water balance in forested areas and spatial rainfall distri-

than altering species composition in decreasing the impacts

bution (Zeng et al. ; Llorens & Domingo ); in

on water resources (Licata et al. ). Furthermore, in com-

addition, it contributes to the growth of woods. Models of

parison with changing forest structure variables, it could be

interception loss (I ), such as the original Gash model

simpler and cruder in inﬂuencing the moisture cycle of

(Gash ), are important tools for researching ecohydrolo-

forest watersheds. Therefore, studies on hydrological model-

gical characteristics; these models have garnered substantial

ing that have focused on initial density are considered more

attention from scholars, and several modiﬁed or improved

signiﬁcant, particularly for poor-quality plantations, because

models have now been developed, such as the sparse Gash

of unsuitable initial density or water resources acting as

model (Gash et al. ), the revised sparse Gash model

restraining

factors.

Additionally,

some

studies

have

1172

Y. Li et al.

Initial plant density effects modeling precision

Hydrology Research

51.5

2020

investigated the relationship between initial density changes

study were to explore the effect of initial plant densities on

caused by thinning and the accuracy of canopy interception

the predicted accuracy of RSGM and to evaluate the

models. Limousin et al. () reported that RSGM provides

impact of different initial densities on model parameters as

good estimates for the canopy interception of stands with

well as examine the possibility of perfecting the model to

different initial densities caused by thinning. Shinohara

improve its accuracy.

et al. () compared the performance of three canopy interception models (Mulder, sparse Gash, and WiMo) in forests both pre- and post-thinning and concluded that the

MATERIALS AND METHODS

canopy interception model might be unsuitable for Japanese cedar (Cryptomeria japonica) forests after intensive thin-

Site description and experimental design

ning. Ma et al. () explored the differences between the application of the sparse Gash model and the WiMo

The research site (Figure 1) is located 8 km northeast

model in pre- and post-thinning plantations, respectively,

of Mount Tai (Northern China) in the Yaoxiang National

and demonstrated that both models had better performance

Forest

in the un-thinned stands than in the post-thinned stands.

36 20 30″). The research site elevation ranges from 400 to

However, these observations only focused on the effects of

956 m (average: 710 m). This region is located in the warm

thinning-related initial density changes on canopy intercep-

temperate zone and has a semi-humid and monsoon cli-

tion prediction, and studies regarding the direct impact of

mate, with a mean annual temperature of 18.5 C and a

initial plant density variations on the modeling accuracy of

mean annual frost-free period of 198 days. The multi-year

RSGM have rarely been reported. We expected a novel ﬁnd-

average precipitation is 727.9 mm, with 75% of the annual

ing when using RSGM in plantations with different initial

precipitation being concentrated from June to September.

plant densities.

The soil type is brown soil. The underlying rock is ancient

Park

(E11 050 39″–117 090 26″,

N36 170 58″–

Initial density is the most common index of forest man-

gneiss of a high metamorphic grade, with a thickness

agement in China. Since the beginning of the last century,

ranging from 10 to 90 cm; in most areas, thickness ranges

China has afforested large swathes of land to compensate

from 30 to 50 cm. Vegetation types belong to coniferous

for the insufﬁcient green area. Although the area under

and deciduous broad-leaved forests in the warm temperate

afforestation is rapidly increasing, the initial plant density

zone, and the main tree species include Pinus tabuliformis,

of the same tree species in the same area is different. More-

Quercus acutissima Carr., Pinus densiﬂora Sieb., Robinia

over, as several familiar forest management measures to

pseudoacacia, and Castanea mollissima (Sun et al. ).

improve forest quality, such as thinning and replanting, are

In the study area, six plots (each with a projection area

infrequently used in most regions of China, these together

of 100 m2, length of 20 m, and width of 5 m) were set up

result in complications such as poor plantation quality.

according to the initial densities of P. tabuliformis planta-

Therefore, numerous studies focusing on ecohydrological

tion as follows: (1) 720 trees ha 1; (2) 1,010 trees ha 1; (3)

and biogeochemical ﬂuxes as well as on either growth

1,334 trees ha 1; (4) 1,695 trees ha 1; (5) 1,720 trees ha 1;

characteristics or economic beneﬁts of stands with different

and (6) 2,746 trees ha 1. We conducted a forestland survey

initial densities have provided a scientiﬁc basis for further

in March 2016 (Table 1) and installed the HOBO automatic

forest management in China. Furthermore, the accuracy

weather station (Onset, USA) surrounding the open area of

with which the model of interception loss estimates

the sample plots to collect necessary meteorological data

canopy interception is considered a signiﬁcant part of

such as rainfall, wind speed, and radiation. The weather

such forest management, particularly for the response of

station collected data at 15-min intervals, and we down-

model simulation efﬁciency to initial plant density, which

loaded the data once a month. Because the six plots are

is a gap that cannot be ignored for the development of

relatively close in terms of distance and there is little

interception models as well as for the ﬁne reﬁnement man-

change in meteorological factors, only one meteorological

agement of forests. Therefore, the objectives of the present

station was installed.

Y. Li et al.

1173

Figure 1

Table 1

Initial plant density effects modeling precision

Hydrology Research

51.5

2020

Study plot location.

Investigation results of stand plots 1

Plot

Sp.

Sl. ( )

Id (trees ha

P. tabuliformis

2 3

TDP (trees)

Ah (m)

ADBH (cm)

760

10.12 ± 0.41

17.87 ± 0.79

0.74

1,010

8.98 ± 0.48

16.88 ± 1.03

0.75

1,334

7.44 ± 0.46

15.04 ± 0.72

0.78

1,695

9.04 ± 0.52

13.98 ± 0.81

0.80

1,720

7.01 ± 0.34

12.25 ± 0.75

0.80

2,746

12.18 ± 0.63

15.19 ± 0.58

0.70

)

ADBH, average diameter at breast height (mm); Sp., species; Sl., slope ( ); Id, initial density (trees ha ); TDP, tree density per plot (trees); Ah, average height (m); c, canopy cover.

Field measurements

collection, the collectors were arranged parallel to the slope along with the sample sites and the height from the

Total precipitation (TP) was determined according to the

collector to the ground was 50 cm, connecting the collect-

data from the automatic weather station. Due to the proxi-

ing barrel at the bottom of the pipe (Sheng et al. ; Ji &

mity of the measured sample plots to each other, only one

Cai ). We measured the amount of water in the collect-

station was placed in the central position relative to each

ing barrel after each rainfall event. TF depth (mm) was

plot.

calculated using the following equation:

Three PVC tubes (DN90) with a length of 8 m and an inner diameter of 9 cm were cut open to collect throughfall (TF) in each plot. To avoid the inﬂuence of herbs on TF

TF ¼

3 × V × 1000 3 × L × W × cos α

(1)

1174

Y. Li et al.

Initial plant density effects modeling precision

Hydrology Research

51.5

2020

where V is the amount of water collected in the collecting

by Valente et al. () and simpliﬁed by Limousin et al.

barrel after each rainfall event (L), L is the length of the

():

PVC tube (m), W is the inner diameter of the PVC tube (m), α is the slope of the sample plots ( ), and L × W × cos α

nX þm

is the cross-sectional area of the PVC tube (m ).

Ij ¼ c

j¼1

breast height, and three trees from each class were selected

PG,j þ

j¼1

Based on the survey results of the sample plots, the trees were categorized into three classes based on diameter at

m X

þ cptc

n q X j¼1

n n X X cE 0 P ) þ c P0 G þ qcStc (P G,j G R j¼1

j¼1

E 1 (PG,j P0 G ) R (3)

as samples. To measure stemﬂow (SF), a rubber pipe cut in the middle was stapled onto the trunks in a spiral shape. The gap between the rubber pipe and trunk was sealed

where n is the number of rainfall events during which the

with silicone, the bottom of the rubber pipe was connected

canopy reached saturation, m is the number of rainfall

to a collecting bucket, and the amount of water collected in

events that were insufﬁciently large to saturate the canopy,

this bucket was measured after each rainfall event. SF depth

c is the fraction of canopy cover, PG is the total rainfall is the wet canopy evaporation rate per amount (mm), E

(mm) was calculated using the following equation:

unit area of canopy cover (mm; calculated by the 1 Xn Gn SFj ¼ Mn i¼1 K M n

(2)

Penman–Monteith equation) (Gash et al. ; Valente is the mean rainfall intenet al. ; Limousin et al. ), R sity (mm/h), P0G is the minimum amount of precipitation

where M is the mean number of trunk plants in the pro-

required to saturate the canopy (mm), ptc ¼ pt =c is the pro-

jected area of the plot, n is the number of the diameter

portion of rain diverted into SF per unit area of cover,

class, Gn is the amount of SF in each diameter class (mm),

Stc ¼ St =c is the stem storage capacity per unit cover area,

Kn is the projected canopy area in each diameter class

and q is the number of saturated trunks.

(cm2), and Mn is the number of trees in each diameter class. Derivation of the parameters used in the model

REVISED SPARSE GASH MODEL It was evident that the canopy storage capacity (S) is an Formula

important parameter in the model; in the present study, S was estimated using the original regression method based

The original Gash model was proposed by Gash (). For

on the relationship between TP and TF. This method was

simplicity, some assumptions have been made: (1) a series of

proposed by Leyton et al. (); owing to its effectiveness

discrete storms are used to represent the real rainfall pat-

because of its simplicity (Liu ), this method has been

tern, and there is a sufﬁciently long interval between these

used more often than the method described by Klaassen

storms to allow the canopy and trunks to dry; (2) the con-

et al. () as well as has been used by Gash & Morton

ditions of rainfall and evaporation remain sufﬁciently

() and Gash (). André et al. () reported that

similar throughout all storms during this period; and (3)

the estimated performance of Leyton’s approach is better

there is no water drip from the canopy during wetting-up,

than a spraying laboratory experiment and mechanistic

and large amount of water on the canopy rapidly evaporates

modeling exercise.

at the end of a storm. The original Gash model was also

The minimum amount of water required to saturate the

revised based on these assumptions (Gash et al. ;

canopy was determined using the equation provided by

Valente et al. ; Limousin et al. ). In the present study, the following equation of RSGM was used, which is used for calculating canopy interception that was proposed

Limousin et al. (): E)S c ln (1 (E= R)) P0G ¼ (R=

(4)

1175

Y. Li et al.

Initial plant density effects modeling precision

Hydrology Research

where Sc ¼ S=c is the canopy storage capacity per unit area

51.5

2020

The rainfall distribution components of P. tabuliformis plantation of different densities varied during the study

of cover (in mm). The minimum amount of water that can ﬁll the trunk

period (Table 2). With the same total rainfall, TF decreased

storage capacity is calculated using the following equation

and I increased with increasing initial density, but they

and is denoted as

P00G

(Valente et al. ; Limousin et al.

increased and decreased when the initial density increased from 1,720 to 2,746 trees ha 1, respectively. Typically, TF

):

and I tended to increase with increasing initial density P00G

0 R E))(S ¼ (R=( tc = ptc ) þ PG

(5)

during the study period. By contrast, SF initially increased with increasing initial density, following which it decreased

In addition, TF and SF can be simulated using the fol-

and increased further when initial density increased from

lowing equations provided by the model: q X

SFj ¼ cptc

j¼1 nþm X

q X

0 R))(P (1 (E= G,j P G,j ) qcStc

when initial density increased from 1,334 to 1,720 trees ha 1 1,720 to 2,746 trees ha 1. In general, SF decreased with increasing density.

(6)

j¼1

TFj ¼

j¼1

nþm X

PG,j

j¼1

nþm X j¼1

q X

Model parameters SFj

(7)

j¼1

We calculated the parameters of RSGM using the previously described methods (Table 3). S, estimated from the regression between TF and TP (Figure 2), increased with

Sensitivity analysis In the present study, we referred to separate sensitivity analyses in previous studies for each density (cf. Valente et al. ; Fan et al. ; Fathizadeh et al. ; Liu et al. ), R, St, and pt were tested in the range of an and S, c, E, increase and decrease by up to 50% of their original values and in terms of the results of models compared with ﬁeld values.

increasing initial density. The value of S for 2,746 trees ha 1 (3.973 mm) was approximately four times greater than that for 720 trees ha 1 (0.894 mm). Values generated for Sc and P0G

ranged

from

1.207 mm

(for

720 trees ha 1)

5.706 mm (for 2,746 trees ha ) and from 1.229 mm (for 720 trees ha 1) to 5.807 mm (for 2,746 trees ha 1), respectively. Furthermore, the variation trends of Sc and P0G were the same as that of S, and the first two parameters for 2,746 trees ha 1 were approximately five times greater than those for 720 trees ha 1. Similar findings were obtained for St, pt, and P00G (St and pt were estimated from the regression

RESULTS Rainfall partitioning

between SF and TP, as shown in Figure 3); they typically increased with increasing initial density, and their values

Table 2

Rainfall redistribution composition of P. tabuliformis forests 1

In the present study, 8 h was considered the standard for

Plot

Sd. (trees ha

dividing rainfall events. Based on rainfall data from May

720

2016 to October 2016, a total of 28 rainfall events with a

1,010

total amount of 430.3 mm and a mean of 15.4 mm were

recorded in the study area. Based on individual rainfall

events, the maximum amount, duration, and intensity of rainfall in the plots were 73.7 mm, 37 h, and 26.32 mm/h, respectively, and the corresponding minimum values were 1.1 mm, 0.75 h, and 0.08 mm/h, respectively.

TP (mm)

TF (mm)

SF (mm)

I (mm)

430.3

357.9

1.6

70.8

316.5

20.3

93.6

1,334

296.8

23.6

109.9

1,695

286.2

19.7

124.4

1,720

277.2

22.2

131.0

2,746

280.1

25.3

124.9

)

TP, total precipitation (mm); TF, throughfall (mm); SF, stemﬂow (mm); I, interception loss (mm).

Y. Li et al.

1176

Table 3

Initial plant density effects modeling precision

Hydrology Research

51.5

2020

Parameters of RSGM for P. tabuliformis plantation 1

S (mm)

Sc (mm)

St (mm)

P 0G (mm)

P 00G (mm)

720

0.894

1.207

0.035

0.006

0.74

1.229

7.244

1,010

1.281

1.708

0.394

0.073

0.75

1.738

7.394

1,334

2.361

3.027

0.404

0.081

0.78

3.027

8.234

1,695

2.774

3.467

0.414

0.073

0.80

3.529

9.422

1,720

3.076

3.797

0.363

0.075

0.81

3.865

8.865

2,746

3.937

5.706

0.545

0.094

0.69

5.807

11.795

Plot

Id (trees ha

)

Id, initial density (trees ha ).

Figure 2

Relationship between TF and TP for rainfall events. In the upper left of each ﬁgure, lowercase letters represent the sample plots with different initial densities (a ¼ 720

trees ha 1, b ¼ 1,010 trees ha rainstorm event.

, c ¼ 1,334 trees ha

, d ¼ 1,695 trees ha

, e ¼ 1,720 trees ha

, and f ¼ 2,746 trees ha

); the same is true in Figure 5. Each circle denotes a

ranged from 0.035 mm (for 720 trees ha 1) to 0.545 mm (for

calculations ranging from 36.0 mm (for 720 trees ha 1) to

2,746 trees ha 1) for St, from 0.006 (for 720 trees ha 1) to

104.4 mm (for 2,746 trees ha 1) and from 1.6 mm (for 720

0.096 (for 2,746 trees ha ) for pt, and from 7.244 mm (for

trees ha 1) to 19.6 mm (for 2,746 trees ha 1), respectively.

720 trees ha 1) to 11.795 mm (for 2,746 trees ha 1) for P00G .

Moreover, the simulated TF value decreased with an increasing initial density as a whole, differing from the trend of

Estimated results

measured values, and ranged from 306.3 mm (for 2,746 trees ha 1) to 392.7 mm (for 720 trees ha 1). In addition,

The simulated TF value was overestimated, whereas the

the simulation results indicated that irrespective of the

simulated I and SF values were less than the measured

initial density, there was substantially high I resulting from

values (Table 4). The simulated I and SF values showed a

evaporation from saturation until rainfall stopped (6.75–

uniform variation trend during the study period, with

28.33%) and after rainfall stopped (62.00–67.85%), and the

Y. Li et al.

1177

Figure 3

Initial plant density effects modeling precision

Hydrology Research

51.5

2020

Relationship between SF and TP for rainfall events. Each circle denotes a rainstorm event. In the upper left corner of each ﬁgure, lowercase letters represent sample plots with different initial densities (speciﬁc densities are the same as those in Figure 2).

Table 4

Simulation results of RSGM

trunks, the other model components increased with increasing initial density.

Id (trees ha

)

We used the mean of the absolute error and relative error

Components of the model

720

1,010

1,334

1,695

1,720

2,746

to describe the simulation accuracy of RSGM for I, TF, and

For m small storms, insufﬁcient to saturate the canopy (mm)

2.34

3.55

5.17

5.31

5.37

18.09

SF (Figure 4). Regarding SF, an increase in initial density decreased the absolute error; in general, there was a tendency

Wetting-up the canopy, for 0.40 n storms > P0G that saturate the canopy (mm)

0.55

0.88

1.41

1.26

toward the lower absolute error of predicted TF and SF with increasing initial density, whereas the opposite was observed for I. Although there was a corresponding slight upward and

Evaporation from saturation until rainfall ceases (mm)

10.21 10.02 9.58

9.53

9.44

7.05

Evaporation after rainfall stops (mm)

22.34 30.74 49.58 63.79 70.74 70.86

Evaporation from trunks (mm)

0.75

downward trend in the absolute errors of TF and I when the initial density varied from 720 to 1,750 trees ha 1, the above-

8.22

7.77

7.31

6.49

7.18

mentioned conclusion was obtained irrespective of this trend. These relationships are presented in Figure 4(a). Similarly, as shown in Figure 4(b), for I, there was a clear negative correlation between initial density and relative error,

Simulated interception loss 36.0 (mm)

53.1

73.0

87.1

93.3

104.4

which ranged from 18.38% (for 2,746 trees ha 1) to 53.03% (for 720 trees ha 1). However, for SF, there was a general

Simulated TF (mm)

392.7 358.6 337.5 326.5 319.2 306.3

positive correlation between initial density and relative

Simulated SF (mm)

1.6

error, which ranged from 2.01% (in 720 trees ha 1) to

18.7

19.9

16.8

17.8

19.6

21.55% (in 2,746 trees ha 1). Regarding TF, although the lowest I occurred under saturated canopy conditions (1.10–

trend of relative error with initial density was as variable as

1.35%). With the exception of the amount of evaporation

the trend of absolute error, the difference was that the vari-

from canopy saturation until rainfall stopped and from

ation of the former was more similar to the convex type.

Y. Li et al.

1178

Figure 4

Initial plant density effects modeling precision

Hydrology Research

51.5

2020

The error between the simulated and measured values of interception varied with the initial density, including (a) absolute error and (b) relative error.

The relative error of TF in high-density stands was only

St, and pt – showed a positive correlation with the prec, E,

slightly smaller than that in low-density stands (10.14 and

dicted I. Moreover, with a 50% increase or decrease in all

10.50%, respectively), and the highest relative error was

the parameters tested, the representation indicates that S

noted in medium-density stands (15.85% in 1,720 trees ha 1).

was the chief parameter for the change in predictions regardless of the initial density, which can result in a maxi-

Sensitivity analysis Figure 5 illustrates the sensitivity analysis results for the vari all remaining parameters – S, ation of predicted I. Besides R,

Figure 5

mum change of ±30% in calculation, and the minimum inﬂuence on the prediction of interception was provided on the prediction results by pt. Subsequently, the effect of E when the parameters tended to be higher than that of R

, mean rainfall intensity; c, canopy cover; E, wet canopy evaporation rate per unit area of Sensitivity analysis of RSGM for change in predicted I. S, canopy storage capacity; R canopy cover; St, stem storage capacity; and pt, drainage partitioning coefﬁcient. The lowercase letters on the right side of the longitudinal axis represent the samples of different initial densities in each ﬁgure, and the exact meaning is the same as in Figure 2.

1179

Y. Li et al.

Initial plant density effects modeling precision

changed in the range of 0 to þ50%. However, when the par ameters tested changed from 50 to 0%, the inﬂuence of R

Hydrology Research

51.5

2020

forest, and mixed White Oak forest). Therefore, we compared our estimated S with the variation range (0.1–3.0 mm) of S in

for all plots. Furthermore, higher was higher than that of E and E. initial density corresponded to lower inﬂuence of R

coniferous forests under different initial densities, as summar-

and c, the inﬂuence of R on the With a 50% decrease in R

P. tabuliformis forests remained higher than that in coniferous

prediction results was higher than that of c, and there was

forests of similar density such as Pinus sylvestris, Picea sitch-

a change in the corresponding average simulated cumulative

ensis, and Pinus elliottii. This may be due to the distinctive

I of 29.38 and 20.12% within 760 trees ha 1, and 19.69 and

broad-ovate crown proﬁle of P. tabuliformis. The developed

ized by Llorens & Gallart (). We found that the S value in

17.01% within 1,010 trees ha . Further, the effect of c on when the initial density simulation was higher than that of R

sclerenchyma on the epidermis of needles and sunken pores

exceeded 1,010 trees ha 1.

carry more water than the coniferous forests evaluated in

can provide appropriate conditions for P. tabuliformis to other studies (China Agricultural Encyclopedia Forestry Volume Editorial Committee ). Remarkably, the variation

DISCUSSION

trend of the S value of these species exhibited an approximate wavy line with an increasing initial density, rather than follow-

During the study period, the average I measured in

ing a linear trend as observed in the present study. We propose

P. tabuliformis plantation plots ranged from 16.46 to

that the S value of the same coniferous species exhibits certain

30.44% of the total rainfall. Typically, I in temperate conifer-

threshold values with increasing initial density and that the S

ous forests ranges from 10 to 60% (Teklehaimanot et al. ;

value will continue to increase and decrease. Differences in

Loustau et al. ; Licata et al. ), with I measured in our

results between this and other studies may be due to the fact

study falling in the upper half of this range. The I range in our

that the maximum initial density of the P. tabuliformis planta-

study is consistent with those for the main plantation forest in

tion in this study (2,746 trees ha 1) was lower than the initial

China (14.7–31.8%) and coniferous forests (21.0–48.0%; Car-

density corresponding to the theoretical ﬁrst threshold of S

lyle-Moses ; Wei et al. ), and it is within the range of

of P. tabuliformis artiﬁcial forests. Similarly, we hypothesize

the interception of P. tabuliformis in various study areas in

that if the initial density continues to increase, the S value of

China (15.7–36.9%; Fang et al. ). Valente et al. ()

P. tabuliformis artiﬁcial forests will show the same trend

obtained similar results there was a lower interception in

with increasing initial density as previously reported.

less dense forests. Results of the measurement of rainfall par-

In our study, ﬁxed growing season parameters were used

titioning at different initial densities indicated that more

in RSGM to predict total interception, which was the same

rainfall penetrated the canopy and entered the next layer,

as those reported by Fan et al. () and Junqueira Junior

thereby affecting the hydrology of the forest ecosystem and

et al. (). Indeed, the rigor of the derivation process

contributing to plant growth, and appropriate thinning may

may be inadequate; however, it is possible to use ﬁxed par-

improve the rainwater use efﬁciency of forests.

ameters to obtain satisfactory estimates of growing season

In the present study, the estimated S ranged from 0.894 to

interception (Wallace & McJannet ; Ghimire et al.

3.937 mm and was lower in low-density stands than in high-

; Fan et al. ). Because the study was conducted

density stands. According to Deguchi et al. (), the esti-

during the growing season, the canopy and climatic par-

mated S in different regions worldwide ranges from 0.25 to

ameters changed slightly. Moreover, the variation of the

1.69 mm, and this range is lower in medium-density (1,334,

canopy and climatic parameters will compensate for each

1,695, and 1,720 trees ha ) and high-density stands (2,746

other; therefore, the error of the ﬁnal simulation of intercep-

trees ha 1). The reason for this might be that P. tabuliformis

tion may be considered minimal, as discussed by Wallace &

plantation in this study was an artiﬁcial coniferous pure

McJannet () and Fan et al. (). The type of variations

forest with a vegetation type that is different from that in the

for the ﬁnal simulation results produced within a certain

above-mentioned literature (such as secondary broad-leaved

period when ﬁxed parameters are used at different time

deciduous forest, mature mixed deciduous forest, hardwood

scales in RSGM to predict annual interception remains

1180

Y. Li et al.

Initial plant density effects modeling precision

Hydrology Research

51.5

2020

unclear and requires further investigation. Recently, a study

analytical Gash model to improve model accuracy. Ginebra-

used models with ﬁxed parameters for the dormant and

Solanellas et al. () suggested that raindrop characteristics

growing seasons (Ma et al. ), respectively. The model

and leaf biomechanical properties may inﬂuence the modeling

showed good performance for both periods; however,

of the dynamic processes of forest canopy interception. Fur-

some parameters seemed to considerably vary between the

thermore, reconstructing the method of estimating S may be

two periods, implying that it is still questionable whether

the correct direction to reduce model simulation errors.

ﬁxed canopy structures and climatic parameters can be used

Valente et al. () proposed a new procedure to estimate S

for simulating annual canopy interception or for a period span-

by inverse extrapolation. This method has been validated at

ning the dormant and growing seasons. Moreover, Zhang et al.

the single-tree scale but has not been widely used yet because

() used variable parameters to run RSGM to estimate and E repcanopy interception in a deciduous shrub; i.e., R

it may not be effective for regions without historical monitoring information.

resent the average values during all hours in each rainfall

Differences in initial plant density may affect forest

event. It remains to be demonstrated whether this approach

hydrological processes and the nutrient cycle, subsequently

can be used to obtain a good result in forests.

inﬂuencing vegetation growth. This will be accompanied

In the present study, although we used RSGM, it demon-

by differences in forest structure variables. Therefore, we

strated better performance in P. tabuliformis forests with

also considered the implication of the effects of changes in

higher-density and could not be applied well in lower-

forest structure variables on the modeling accuracy of the

density forests; moreover, the relative error decreased with

revised Gash model, although it is not detailed enough.

increasing initial density and the revised model tended to

Moreover, Fathizadeh et al. () showed that S and I

overestimate TF and underestimate SF and I, which is simi-

have strong relationships with wood area index, except

lar to the ﬁndings obtained by Limousin et al. () and Ma

LAI and c, and supported the idea that the studies for

et al. (). Although c was added to the revised version of

canopy structure variables can considerably enhance the

the model to further describe the characteristics of canopy

performance of the hydrological models. Indeed, we plan

structure and vegetation density along with S (Gash et al.

to continue the work of Fathizadeh et al. () to conduct

), it was not adequate. The indirect and complete esti-

future detailed studies on this subject; however, for now,

mation of the complex canopy structure is a signiﬁcant

this study only discussed the inﬂuence of initial plant density

channel to improve the simulation accuracy of the model.

on the modeling accuracy of the revised Gash model owing

However, c can only interpret the horizontal structure of the

to the limitation of observation equipment and measures.

canopy and cannot adequately express canopy thickness,

Sensitivity analysis showed that S demonstrated absol-

vegetation density, and other structural and physiological par-

ute dominance in medium- and higher-density forests,

ameters, which may affect simulation accuracy, particularly in

which was similar to the results reported by Dykes (),

forests with lower density. van Dijk & Bruijnzeel (a) pro-

Liang (), and Su et al. (); however, S, the chief par-

posed correcting the revised model by supplementing the leaf area index (LAI) and expounding its relationship with S

ameter for simulated results, is not dependent on the and E gradually decreased as density. The inﬂuence of R

and c; they considered that LAI was linearly related to

the initial density increases. This may be due to the particu-

canopy capacity rather than to canopy cover in ‘constant phy-

lar characteristics of precipitation during the study period

siognomy and conﬁguration’ and accordingly presented an

(relatively long rainfall duration, intermittent rainfall pat-

application of this version (van Dijk & Bruijnzeel b).

terns, and a relatively large number of moderate and

Nevertheless, this version of the model is based on the hypoth-

extreme rainfall events). Most rainfall occurred at night

esis that canopy capacity is not affected by initial density, and

during the study period, which may be the reason for the

because this hypothesis was evidently contrary to the results of

lesser contribution of the evaporation rate in the present

the present study, it may not be a highly effective solution to

study compared with that in other studies (Loustau et al.

improve the accuracy of the model from the root. The

; Deguchi et al. ; Fathizadeh et al. ). Linhoss

canopy structure must be effectively integrated into the

& Siegert () reported that storm duration, canopy

1181

Y. Li et al.

Initial plant density effects modeling precision

Hydrology Research

51.5

2020

storage, and solar radiation were extremely important for

appropriate to describe the evaporation rate for the open

canopy interception model sensitivity. Lower-density forests

canopy is an effective approach to improve the operational

exhibit larger forest gaps, larger crown width, and thinner

performance of RSGM in sparse forests. Moreover, it is

crown thickness than higher-density forests, which may

imperative to further demonstrate the inﬂuence of estimated

explain the reason for the climatic parameters being more

mean values of climatic parameters at different time scales

sensitive at lower densities. In addition, Linhoss & Siegert

on simulation accuracy. Most importantly, the impact of initial

() have reported that the estimation of the evaporation

planting density on the simulation results cannot be ignored

rate using the Penman–Monteith approach appears preferable

when using RSGM to quantify canopy interception. The appli-

for canopies that are not completely ventilated rather than for

cability of RSGM in sparse forest remains to be investigated

sparse forests and the methods for estimating the evaporation

until the model can be further improved or validated in the

rate in sparse forests should be investigated. Therefore, there

application area.

have been studies that have attempted to use the wet-bulb method instead of the Penman–Monteith approach to derive the evaporation rate at the single-tree or sparse forest level

ACKNOWLEDGEMENTS

when using the Gash-type models (e.g., Ma et al. ; Valente et al. ), and model performance using the former is

This work was funded by the Natural Science Foundation of

superior to the latter. This implies that improving the methods

Shandong Province of China project (No. ZR2016DB12)

used for estimating meteorological parameters in models could

and a project of the Key Laboratory of Water Saving

be a valid direction for sparse forests.

Irrigation Project of Ministry of Agriculture (Farmland Irrigation Research Institute of Chinese Academy of Agricultural Sciences, No. FIRI2018-04).

CONCLUSIONS The average cumulative measured I ranged from 16.46 to

DATA AVAILABILITY STATEMENT

30.44% of gross rainfall in P. tabuliformis plantations with different initial densities, and the interception was positively associated with initial plant density. The results obtained using the revised Gash model suggested that a relatively

Data cannot be made publicly available; readers should contact the corresponding author for details.

good consistency exists between the modeled and estimated interceptions in higher-density forests and that simulation

REFERENCES

accuracy enhances with an increase in the density. This may be attributable to the revised model being inadequate with respect to certain biological characteristics of P. tabuliformis plantations and the expression of the canopy structure of the sparse forest, and these weaknesses are magniﬁed during the simulation of P. tabuliformis plantation. The predicted interception was higher sensitive to the canopy structure, irrespective of the initial density, and the close stands were less susceptible to the effects of the climate parameters on accuracy than the sparse stands. Therefore, it is crucial to effectively optimize and modify parameters describing the canopy structure in RSGM for improving simulative accuracy while ensuring the simplicity of the model and identifying and developing direct observation or estimation methods that are more

André, F., Jonard, M. & Ponette, Q.  Precipitation water storage capacity in a temperate mixed oak-beech canopy. Hydrological Processes 22 (20), 4130–4141. doi:10.1002/hyp.7013. Bui, X. D., Shusuke, M. & Takashi, G.  Effect of forest thinning on overland flow generation on hillslopes covered by Japanese cypress. Ecohydrology 4 (3), 367–378. doi:10.1002/ eco.135. Carlyle-Moses, D. E.  Throughfall, stemflow, and canopy interception loss fluxes in a semi-arid Sierra Madre Oriental matorral community. Journal of Arid Environments 58 (2), 181–202. doi:10.1016/S0140-1963(03)00125-3. Chen, S., Chen, C., Zou, C. B., Stebler, E., Zhang, S., Hou, L. & Wang, D.  Application of Gash analytical model and parameterized Fan model to estimate canopy interception of a Chinese red pine forest. Journal of Forest Research 18 (4), 335–344. doi:10.1007/s10310-012-0364-z.

1182

Y. Li et al.

Initial plant density effects modeling precision

China Agricultural Encyclopedia Forestry Volume Editorial Committee  China Agricultural Encyclopedia Forestry, Volume (II). Agricultural Press, Beijing. Cui, Y. & Jia, L.  A modified Gash model for estimating rainfall interception loss of forest using remote sensing observations at regional scale. Water 6 (4), 993–1012. doi:10. 3390/w6040993. Deguchi, A., Hattori, S. & Park, H.  The influence of seasonal changes in canopy structure on interception loss: application of the revised Gash model. Journal of Hydrology 318 (1–4), 80–102. doi:10.1016/j.jhydrol.2005.06.005. Dykes, A. P.  Rainfall interception from a lowland tropical rainforest in Brunei. Journal of Hydrology 200 (1–4), 260–279. doi:10.1016/S0022-1694(97)00023-1. Fan, J., Oestergaard, K. T., Guyot, A. & Lockington, D. A.  Measuring and modeling rainfall interception losses by a native Banksia woodland and an exotic pine plantation in subtropical coastal Australia. Journal of Hydrology 515, 156–165. doi:10.1016/j.jhydrol.2014.04.066. Fang, S., Zhao, C., Jian, S. & Yu, K.  Canopy interception of Pinus tabulaeformis plantation on Longzhong Loess Plateau, North-west China: characteristics and simulation. Chinese Journal of Applied Ecology 24 (06), 1509–1516. doi:10. 13287/j.1001-9332.2013.0326. Fathizadeh, O., Hosseini, S. M., Zimmermann, A., Keim, R. F. & Darvishi Boloorani, A.  Estimating linkages between forest structural variables and rainfall interception parameters in semi-arid deciduous oak forest stands. Science of the Total Environment 601–602, 1824–1837. doi:10.1016/ j.scitotenv.2017.05.233. Fathizadeh, O., Hosseini, S. M., Keim, R. F. & Boloorani, A. D.  A seasonal evaluation of the reformulated Gash interception model for semi-arid deciduous oak forest stands. Forest Ecology and Management 409, 601–613. doi:10.1016/ j.foreco.2017.11.058. Fernandes, R. P., Silva, R. W. D. C., Salemi, L. F., Andrade, T. M. B. D., Moraes, J. M. D., Dijk, A. I. J. M. & Martinelli, L. A.  The influence of sugarcane crop development on rainfall interception losses. Journal of Hydrology 551, 532–539. doi:10.1016/j.jhydrol.2017.06.027. Gao, Y., Zhu, X., Yu, G., He, N., Wang, Q. & Tian, J.  Water use efficiency threshold for terrestrial ecosystem carbon sequestration in China under afforestation. Agricultural and Forest Meteorology 195–196, 32–37. doi:10.1016/j.agrformet. 2014.04.010. Gash, J. H. C.  An analytical model of rainfall interception by forests. Quarterly Journal of the Royal Meteorological Society 105, 43–55. doi:10.1002/qj.49710544304. Gash, J. H. C. & Morton, A. J.  An application of the Rutter model to the estimation of the interception loss from Thetford Forest. Journal of Hydrology 38, 49–58. doi:10. 1016/0022-1694(78)90131-2. Gash, J. H. C., Lloyd, C. R. & Lachaud, G.  Estimating sparse forest rainfall interception with an analytical model. Journal of Hydrology 170 (1), 79–86. doi:10.1016/0022-1694(95)02697-N.

Hydrology Research

51.5

2020

Ghimire, C. P., Bruijnzeel, L. A., Lubczynski, M. W. & Bonell, M.  Rainfall interception by natural and planted forests in the Middle Mountains of Central Nepal. Journal of Hydrology 475, 270–280. doi:10.1016/j.jhydrol.2012.09.051. Ghimire, C. P., Bruijnzeel, L. A., Lubczynski, M. W., Ravelona, M., Zwartendijk, B. W. & van Meerveld, H. J. I.  Measurement and modeling of rainfall interception by two differently aged secondary forests in upland eastern Madagascar. Journal of Hydrology 545, 212–225. doi:10. 1016/j.jhydrol.2016.10.032. Ginebra-Solanellas, R. M., Holder, C. D., Lauderbaugh, L. K. & Webb, R.  The inﬂuence of changes in leaf inclination angle and leaf traits during the rainfall interception process. Agricultural and Forest Meteorology 285–286, 107924. doi:10.1016/j.agrformet.2020.107924. Hassan, S. M. T., Ghimire, C. P. & Lubczynski, M. W.  Remote sensing upscaling of interception loss from isolated oaks: Sardon catchment case study, Spain. Journal of Hydrology 555, 489–505. doi:10.1016/j.jhydrol.2017.08.016. Ji, Y. & Cai, T. J.  Canopy interception in original Korean pine forest: measurement and dividual simulation in Xiaoxing’an Mountains, northeastern China. Journal of Beijing Forestry University 37 (10), 41–49. doi:10.13332/j.1000-1522.20150084. Junqueira Junior, J. A., de Mello, C. R., de Mello, J. M., Scolforo, H. F., Beskow, S. & McCarter, J.  Rainfall partitioning measurement and rainfall interception modelling in a tropical semi-deciduous Atlantic forest remnant. Agricultural and Forest Meteorology 275, 170–183. doi:10.1016/j. agrformet.2019.05.016. Klaassen, W., Bosveld, F. & Water, E. D.  Water storage and evaporation as constituents of rainfall interception. Journal of Hydrology 212 (1–4), 36–50. doi:10.1016/S0022-1694(98) 00200-5. Leyton, L., Reynolds, E. R. C. & Thompson, F. B.  Rainfall interception in forest and moorland. In: International Symposium on Forest Hydrology (W. E. Sopper & H. W. Lull, eds.). Pergamon, Oxford, pp. 163–178. Liang, W.  Simulation of Gash model to rainfall interception of Pinus tabulaeformis. Forest Systems 23 (2), 300–303. doi:10.5424/fs/2014232-03410. Licata, J. A., Pypker, T. G., Weigandt, M., Unsworth, M. H., Gyenge, J. E., Fernández, M. E., Schlichter, T. M. & Bond, B. J.  Decreased rainfall interception balances increased transpiration in exotic ponderosa pine plantations compared with native cypress stands in Patagonia, Argentina. Ecohydrology 4 (1), 83–93. doi:10.1002/eco.125. Limousin, J., Rambal, S., Ourcival, J. & Joffre, R.  Modelling rainfall interception in a mediterranean Quercus ilex ecosystem: lesson from a throughfall exclusion experiment. Journal of Hydrology 357 (1–2), 57–66. doi:10.1016/j.jhydrol.2008. 05.001. Linhoss, A. C. & Siegert, C. M.  A comparison of ﬁve forest interception models using global sensitivity and uncertainty analysis. Journal of Hydrology 538, 109–116. doi:10.1016/j. jhydrol.2016.04.011.

1183

Y. Li et al.

Initial plant density effects modeling precision

Liu, S.  Estimation of rainfall storage capacity in the canopies of cypress wetlands and slash pine uplands in North-Central Florida. Journal of Hydrology 207 (1), 32–41. doi:10.1016/ S0022-1694(98)00115-2. Liu, Z., Wang, Y., Tian, A., Liu, Y., Webb, A. A., Wang, Y., Zuo, H., Yu, P., Xiong, W. & Xu, L.  Characteristics of canopy interception and its simulation with a revised Gash model for a larch plantation in the Liupan Mountains, China. Journal of Forestry Research 29 (1), 187–198. doi:10.1007/s11676-017-0407-6. Llorens, P. & Domingo, F.  Rainfall partitioning by vegetation under Mediterranean conditions. A review of studies in Europe. Journal of Hydrology 335 (1–2), 37–54. doi:10.1016/ j.jhydrol.2006.10.032. Llorens, P. & Gallart, F.  A simplified method for forest water storage capacity measurement. Journal of Hydrology 240 (1–2), 131–144. doi:10.1016/s0022-1694(00)00339-5. Loustau, D., Berbigier, P., Granier, A. & Moussa, F. E. H.  Interception loss, throughfall and stemflow in a maritime pine stand. I. Variability of throughfall and stemflow beneath the pine canopy. Journal of Hydrology 138 (3–4), 449–467. doi:10.1016/0022-1694(92)90130-n. Ma, J., Zha, T., Jia, X., Tian, Y., Bourque, C. P. A., Liu, P., Bai, Y., Wu, Y., Ren, C., Yu, H., Zhang, F., Zhou, C. & Chen, W.  Energy and water vapor exchange over a young plantation in northern China. Agricultural and Forest Meteorology 263, 334–345. doi:10.1016/j.agrformet.2018.09.004. Ma, C., Li, X., Luo, Y., Shao, M. & Jia, X.  The modelling of rainfall interception in growing and dormant seasons for a pine plantation and a black locust plantation in semi-arid Northwest China. Journal of Hydrology 577, 123849. doi:10. 1016/j.jhydrol.2019.06.021. Ma, C., Luo, Y. & Shao, M.  Comparative modeling of the effect of thinning on canopy interception loss in a semiarid black locust (Robinia pseudoacacia) plantation in Northwest China. Journal of Hydrology 590, 125234. doi:10.1016/j. jhydrol.2020.125234. National Forestry and Grassland Administration  China Forestry Statistical Yearbook. Available from: http://www. forestry.gov.cn/data.html/ (accessed 7 April 2019). Owens, M. K., Lyons, R. K. & Alejandro, C. L.  Rainfall partitioning within semiarid juniper communities: effects of event size and canopy cover. Hydrological Processes 20 (15), 3179–3189. doi:10.1002/hyp.6326. Sadeghi, S. M. M., Attarod, P., Van Stan, J. T., Pypker, T. G. & Dunkerley, D.  Efficiency of the reformulated Gash’s interception model in semiarid afforestations. Agricultural and Forest Meteorology 201, 76–85. doi:10.1016/j.agrformet. 2014.10.006. Sheng, H. C., Cai, T. J., Li, Y. & Liu, Y. J.  Rainfall redistribution in Larix gmelinii forest on northern of Daxing’an mountains, north-east of China. Journal of Soil and Water Conservation 28 (06), 101–105. doi:10.13870/j. cnki.stbcxb.2014.06.019.

Hydrology Research

51.5

2020

Shinohara, Y., Levia, D. F., Komatsu, H., Nogata, M. & Otsuki, K.  Comparative modeling of the effects of intensive thinning on canopy interception loss in a Japanese cedar (Cryptomeria japonica D. Don) forest of western Japan. Agricultural and Forest Meteorology 214–215, 148–156. doi:10.1016/j.agrformet.2015.08.257. Su, L., Zhao, C., Xu, W. & Xie, Z.  Modelling interception loss using the revised Gash model: a case study in a mixed evergreen and deciduous broadleaved forest in China. Ecohydrology 9 (8), 1580–1589. doi:10.1002/eco.1749. Sun, J., Gao, P., Li, C., Wang, R., Niu, X. & Wang, B.  Ecological stoichiometry characteristics of the leaf–litter–soil continuum of Quercus acutissima Carr. and Pinus densiﬂora Sieb. in Northern China. Environmental Earth Sciences 78 (1). doi:10.1007/s12665-018-8012-3. Teklehaimanot, Z., Jarvis, P. G. & Ledger, D. C.  Rainfall interception and boundary layer conductance in relation to tree spacing. Journal of Hydrology 123 (3–4), 261–278. doi:10. 1016/0022-1694(91)90094-x. Valente, F., David, J. S. & Gash, J. H. C.  Modelling interception loss for two sparse eucalypt and pine forests in central Portugal using reformulated Rutter and Gash analytical models. Journal of Hydrology 190 (1), 141–162. doi:10.1016/S0022-1694(96)03066-1. Valente, F., Gash, J. H., Nóbrega, C., David, J. S. & Pereira, F. L.  Modelling rainfall interception by an olive-grove/ pasture system with a sparse tree canopy. Journal of Hydrology 581, 124417. doi:10.1016/j.jhydrol.2019.124417. van Dijk, A. I. J. M. & Bruijnzeel, L. A. a Modelling rainfall interception by vegetation of variable density using an adapted analytical model. Part 1. Model description. Journal of Hydrology 247 (3), 230–238. doi:10.1016/S0022-1694(01) 00392-4. van Dijk, A. I. J. M. & Bruijnzeel, L. A. b Modelling rainfall interception by vegetation of variable density using an adapted analytical model. Part 2. Model validation for a tropical upland mixed cropping system. Journal of Hydrology 247 (3), 239–262. doi:10.1016/S0022-1694(01)00393-6. Wallace, J. & McJannet, D.  Modelling interception in coastal and montane rainforests in northern Queensland, Australia. Journal of Hydrology 348 (3–4), 480–495. doi:10.1016/j. jhydrol.2007.10.019. Wei, X., Liu, S., Zhou, G. & Wang, C.  Hydrological processes in major types of Chinese forest. Hydrological Processes 19 (1), 63–75. doi:10.1002/hyp.5777. Zeng, N., Shuttleworth, J. W. & Gash, J. H. C.  Inﬂuence of temporal variability of rainfall on interception loss. Part I. Point analysis. Journal of Hydrology 228 (3), 228–241. doi:10. 1016/S0022-1694(00)00140-2. Zhang, S., Li, X., Jiang, Z., Li, D. & Lin, H.  Modelling of rainfall partitioning by a deciduous shrub using a variable parameters Gash model. Ecohydrology 11 (7), e2011. doi:10. 1002/eco.2011.

First received 15 December 2019; accepted in revised form 18 August 2020. Available online 14 September 2020

1184

51.5

2020

Entropy weight method coupled with an improved DRASTIC model to evaluate the special vulnerability of groundwater in Songnen Plain, Northeastern China Bin Wang, Yanguo Teng, Huiqun Wang, Rui Zuo, Yuanzheng Zhai, Weifeng Yue and Jie Yang

ABSTRACT The Songnen Plain in Northeast China is the only remaining black soil agricultural area in the world and is an important food base for China. The groundwater resources in this area are abundant, but human activities have caused them polluted. This paper established a groundwater vulnerability assessment to characterize the influence of human activities which used an entropy weight method. The index was tested using the nitrate pollution distribution in the groundwater to verify the effectiveness of this method. The results showed that areas with high specific vulnerability were distributed in the northern and eastern parts of the Songnen Plain and were consistent with areas that showed serious nitrate pollution of the groundwater. The correlation coefficient between these areas was 0.2536, which greatly improved the vulnerability assessment without superimposing human activities in the model. The results clearly showed that human activities increased groundwater vulnerability on the Songnen Plain. The evaluation method provided a reference for similar evaluations and a basis for the protection and management of groundwater resources in this region. Key words

| DRASTIC model, entropy weight method, groundwater resource management, groundwater vulnerability, nitrate pollution, Songnen Plain

HIGHLIGHTS

• •

Establishing a groundwater vulnerability assessment method to characterize the inﬂuence of human activities using an entropy weight method. Entropy weight method coupled with an improved DRASTIC model to evaluate the special vulnerability of groundwater for water management.

Bin Wang Yanguo Teng (corresponding author) Huiqun Wang Engineering Research Center of Groundwater Pollution Control and Remediation of Ministry of Education; College of Water Sciences, Beijing Normal University, Beijing 100875, China E-mail: teng1974@163.com Bin Wang Chinese Academy for Environmental Planning, Beijing 100012, China Rui Zuo Yuanzheng Zhai Weifeng Yue Jie Yang College of Water Sciences, Beijing Normal University, Beijing 100875, China

1185

B. Wang et al.

Evaluation method of the special vulnerability of groundwater

Hydrology Research

51.5

2020

GRAPHICAL ABSTRACT

INTRODUCTION Groundwater is an important source of water for human

et al. ; Jia et al. ; Hao et al. ). Because of its sub-

beings (Cassardo & Jones ; Salman et al. ). Histori-

terranean nature, groundwater pollution is typically a long-

cal experience shows that water shortages do not only

term and difﬁcult-to-control issue (Ahmad & Al-Ghouti

seriously affect normal life but can also lead to disease out-

). Therefore, rational planning and utilization of ground-

breaks, wars and other unexpected hazards (Falkenmark

water resources to avoid groundwater pollution are far more

; Eliasson ; Mancosu et al. ; Egbueri ).

important than post-contamination treatment (Clemens

Given the unsustainable population and economic growth,

et al. ; Erostate et al. ; Thomann et al. ).

coupled with the failure to prevent groundwater pollution

Groundwater management encompasses a broad range

from industrial wastewater discharges and from heavy use

of activities including prevention of groundwater contami-

of pesticides and fertilizers (Bosch et al. ; Khosravi

nation. Vulnerability and pollution risk assessments to

et al. ; Li et al. ), more groundwater gets seriously

identify risk zones are the very ﬁrst important steps to gen-

polluted, especially shallow groundwater resources (Gejl

erate useful information for devising strategies aimed at

et al. ; He et al. ). In many parts of China, shallow

groundwater protection to contamination. Delineating vul-

groundwater is an important source of drinking water (Lü

nerable zones helps water resource managers to divert

1186

B. Wang et al.

Evaluation method of the special vulnerability of groundwater

Hydrology Research

51.5

2020

groundwater development activities to other safer areas and,

urgent need to carry out an evaluation of the groundwater

hence, can minimize the cost of water treatment (Shrestha

vulnerability to nitrate on the Songnen Plain, in order to

et al. ). This article carried on the groundwater vulner-

guide protection and management of the regional ground-

ability research, which is of great importance to the

water resources. In the past, vulnerability assessments

allocation and protection of groundwater resources (Foster

were mostly focused on the water-resource scale or the

et al. ; Machiwal et al. ); it is largely deﬁned by

basin scale. To evaluate a large-scale area, such as the Song-

the propensity and possibility of pollutants reaching a cer-

nen Plain, it requires not only the correct evaluation

tain location above the uppermost boundary of an aquifer.

parameters but also an improved model, in which the

Research on Chinese groundwater vulnerability began

weight of each evaluation index is determined for this

in the mid-1990s (Margat ; Gogu & Dassargues ).

area. In order to evaluate the vulnerability of groundwater

It has progressed rapidly since then (Ibe et al. ; Gogu

scientiﬁcally and reasonably, improve the reliability of the

et al. ; Polemio et al. ; Shirazi et al. ; Kumar

evaluation results, and provide guidance for the manage-

et al. ; Wachniew et al. ; Iván & Mádl-Szőnyi

ment of groundwater resources in the study area, using the

). The most common and classic evaluation model

hydrogeological characteristics of the Songnen Plain, this

used to evaluate groundwater vulnerability is the DRASTIC

article reconstructed evaluation model indicators and used

method, proposed by the US Environmental Protection

the entropy weight method and the cusp catastrophe

Agency in 1987 (Rahman ; Saidi et al. ; Barzegar

model to determine the weights of each index. The results

et al. ). The evaluation index of the DRASTIC model

of the modiﬁed DRASTIC model evaluation were compared

is based on the variables selected by US hydrogeological

with the nitrate distribution generated by human activities in

experts (Shirazi et al. ). Many efforts have been taken

the study area to verify the accuracy of the model.

to improve the DRASTIC model to make it applicable to different hydrogeological conditions (Nobre et al. ; Saidi et al. ; Wu et al. ; Zhang et al. ), although

STUDY AREA AND DATA

these improvements are often limited to adjusting the parameters of the model. In practice, it is dubious to extend

The Songnen Plain is one of three major plains in Northeast

the weightings of this method to other regions because of

China. It is located in the central part of the Songliao Basin,

the variable geographic setting of each evaluation area

bound by the Xing’an Mountains and Changbai Mountains

(Mendoza & Barmen ). Clearly, the geological and

and Songliao Watershed. It is traversed by the Songhua

hydrogeological conditions also vary a great deal (Denny

and the Nenjiang Rivers. The geographical coordinates of

et al. ). So, the evaluation indicators should be selected

the study area are 43 360 –49 260 N and 121 210 –128 180 E,

according to the speciﬁc conditions of the evaluation area,

deﬁning a total area of 180,500 km2. Its location is shown

considering the natural attributes of the evaluation area,

in Figure 1.

and assigning a relative importance to each index. Thus,

The Songnen Plain is a semi-enclosed, asymmetrical

the correct weighting of variables directly determines the

basin, with a height difference of 400 m and a low gently

accuracy of the evaluation results (Singh et al. ).

inclined central region. Inﬂuenced by the difference in tec-

The Songnen Plain hosts both large-scale commercial

tonic activity among different regions, the geomorphology

grain and oil production. The region of the Second Songhua

of the region can be divided into three types: erosional land-

River has 210 million hectares of arable land, with an

forms, denuded landforms, and stacked landforms.

annual grain output of 16.7 billion kg, accounting for 61%

The Songnen Plain has a typical East Asian continental

of Jilin Province. Groundwater is an important source of

monsoon climate. It is cold and dry in winter, hot and rainy

water for crop growth in this region. However, recent

in summer. With an average annual temperature of 3.8 C,

survey data showed that the shallow groundwater of the

average annual precipitation of 484.57 mm, and an average

Songnen Plain had been variably contaminated by nitrate

annual evaporation of 1,498.1 mm. The main surface water

(Zhu et al. ; Bian et al. ). Therefore, there is an

systems are the Songhua River, the Nenjiang River, the

B. Wang et al.

1187

Figure 1

Evaluation method of the special vulnerability of groundwater

Hydrology Research

51.5

2020

Location of the study area showing the main rivers and cities on the Songnen Plain.

Second Songhua River, and their tributaries. According to the

wells typically have no anti-seepage measures, such as hard-

latest ‘Songliao Basin Water Resources Bulletin’, the volume

ening of the ground around the well. Typically, the wells are

of the water resources of the Songnen Plain is 9.6 billion m .

simple structures, with no buried gravel, and depths of less

A total of 1,409 groundwater sampling sites were inves-

than 20 m. The ground surface comprises mostly clay and

tigated on the Songnen Plain as part of this study. The

silt, while the aquifer is mainly ﬁne sand; the groundwater

sampling was conducted from 2013 to 2015. The sampling

depth ranges from 1 to 10 m below the surface.

points are evenly distributed within the study area, as

Samples were collected in strict accordance with the

shown in Figure 2. Most of the sampling points were situated

technical speciﬁcations for water sample collection. The

in residential areas like villages. This is because most of the

site location was ﬁxed by a GPS three-parameter calibration,

sampling sites were water wells used by villagers. These

water level measurement error was less than 1 cm, and the

B. Wang et al.

1188

Figure 2

Evaluation method of the special vulnerability of groundwater

Hydrology Research

51.5

2020

Distribution of the 1,409 groundwater sampling points within the study area.

well water was sampled using pre-treated sample bottle.

NO2 , As, Hg, Se, Cl , SO4 2 , F , NO3 , I , HCO3 ,

Reagents, such as volatile organics, were neutralized with

CO3 2 , CODCr, total hardness, TDS, Cd, Cu, Pb, Zn) of

two drops of concentrated hydrochloric acid. The collected

the groundwater was completed by the Experimental Test-

samples were stored at 4 C and transferred to laboratory

ing Center of the Shenyang Geological Survey. The

storage within 5 days.

detection and analysis methods were based on full-spectrum

Two sets of parallel samples were collected for quality

direct reading plasma spectroscopy (ICP-OES; Optima

control. In order to identify the groundwater quality and

5300DV Spectrometer, USA), ultraviolet spectrophotometry

to analyze the hydrochemical characteristics of the study

(COL; PerkinElmer Lambda 950, USA), and ion chromato-

area, analytical testing of 31 inorganic components (Na ,

graphy (IC; Thermo Fisher ICS2000, USA), supplemented

Kþ, Ca2þ, TFe, Mn, Al, NH4 þ , PO4 3 , H2SiO3, Cr6þ, Br,

atomic

ﬂuorescence

spectrometry

(AFS;

Titan

1189

B. Wang et al.

Evaluation method of the special vulnerability of groundwater

Hydrology Research

51.5

2020

AFS-9120, China), plasma mass spectrometry (ICP-MS;

weights solely according to the preference or judgments of

Thermo Fisher X Series, USA), volumetric (VOL), and gravi-

decision-makers, and the potential uncertainty in this

metric methods (GR). To ensure the accuracy and reliability

method is its main disadvantage. The objective weighting

of all analytical methods, each sample was analyzed 12

method calculates weights based on actual observations,

times, yielding a relative standard deviation (RSD%) of

without any consideration of the decision-makers’ prefer-

15%. The self-test rate and mutual inspection rate were

ences, and does not beneﬁt from the expert knowledge or

both 100%, giving a total of 1,409 sets of valid data for

experience of the decision-makers. Thus, this paper uses

modeling.

weights determined using the entropy weight method. The entropy method is a measure of the uncertainty formulated in terms of probability theory (Clausius ). It is used

METHODS

describe

irreversible

phenomenon

involving

motion or a process and was introduced into information The main evaluation methods used in this paper were a modiﬁed DRASTIC model and an entropy weight method. The DRASTIC model is widely used to evaluate groundwater vulnerability to a wide range of contaminants. The acronym DRASTIC stands for the seven parameters used

theory by Shannon (). Nowadays, this method has been widely applied to various research ﬁelds to determine factor weights (Brunsell et al. ; Singh ; Ruddell et al. ; Islamoglu et al. ; Xu et al. ; Işık & Adalı ). Information entropy is the measurement of the

in the model, namely depth of the water table (D), net

disorder degree of a system (Harmancioglu ). It

recharge (R), aquifer media (A), soil media (S), topography

measures the amount of useful information based on the

(T), impact of the vadose zone media (I), and conductivity

data provided (Gao et al. ). When the difference in

of the aquifer hydraulic (C). These seven parameters are

value among evaluated objects of a given indicator is big,

further sub-divided into ranges or zones, representing var-

and the entropy is small, this illustrates that this indicator

ious hydrological settings and are assigned different ratings

provides useful information, and its weighting should be

from 1 (minimum impact) to 10 (maximum impact) based

high. In contrast, if the difference is small and the entropy

on a rating chart. In this method, a linear addition of vari-

is big, the relative weight will be low. Hence, entropy

ables is usually combined with the rating and weighting to

theory is an objective way of determining weights (Zou

determine the DRASTIC index value. The greater the rating and weight given to a parameter, the greater its relative importance in contributing to the aquifer vulnerability.

et al. ). According to Zou et al., the evaluation process is generally divided into three steps (Qiu ; Zou et al. ; Wu et al. ).

The DRASTIC formula is formulated as follows: Step1: normalization of the original matrix Ds ¼

7 X

(Wi × Ri ),

(1)

i¼1

Suppose m evaluation indicators and n evaluation objects form the original matrix of indicators, X ¼ (xij)m×n, as

where Ds represents the score of the DRASTIC index; Wi

follows:

represents the weight of factor i; and Ri represents the

score of factor i. Single-parameter sensitivity analyses explore the contribution of individual variables and use the same weights as the DRASTIC method. Therefore, it is particularly important to use an appropriate method to calculate the weight

X ¼ [xij ]m×n

x11 6 x21 6 ¼6 . 4 .. xm1

x12 x22 .. . xm2

... ... .. . ...

3 x1n x2n 7 7 .. 7 . 5 xmn

(2)

(i ¼ 1, 2 . . . , m; j ¼ 1, 2, . . . , n),

of each parameter. At present, the methods for weight calculation are divided into two types: subjective and objective

where xij presents the performance value of the ith alterna-

methods. The subjective weighting method determines

tive on the jth criterion.

B. Wang et al.

1190

Evaluation method of the special vulnerability of groundwater

51.5

2020

RESULTS AND DISCUSSION

Normalization of this matrix gives Equation (3): R ¼ (rij )m×n ,

Hydrology Research

(3)

Based on the survey results and the hydrogeological charac-

where rij is the data for the jth evaluation object under the

teristics of the study area, this paper obtained six

ith indicator and rij ∈ [0,1].

representative indicators of the DRASTIC model. The data

Among these indicators, when bigger is better, this

level at the sampling points. R was derived from meteorolo-

yields rij ¼

for D were from actual measurements of the groundwater gical data released by the government department. S was

xij min (xij ) max (xij ) min (xij )

(4)

(i ¼ 1, 2 . . . , m and j ¼ 1, 2, . . . , n),

derived from actual survey data. T was determined from the ofﬁcial digital elevation model, having 30 m × 30 m accuracy. I and C were derived from geological survey results. This article did not use the index A because attributes of A

but when smaller is better, this yields

were closely related to I and C; furthermore, if the index A

max (xij ) xij rij ¼ max (xij ) min (xij )

(5)

(i ¼ 1, 2 . . . , m and j ¼ 1, 2, . . . , n):

is used for evaluation, it will reduce the impact of the other indicators, making it difﬁcult to accurately assess the vulnerability of the regional groundwater. According to the characteristics of each index, each attribute of each index has a score. Generally, the attribute of the index is

Step 2: calculation of entropy

more likely to cause groundwater pollution, and it will get For m indicators and n evaluation objects in the evaluation

a higher score. Limited by the availability of survey data,

problem, the entropy of the ith indicator is deﬁned as

some indicators in this assessment could only use interp-

follows:

olation method with available data to obtain the situation

m P

without survey data area, which might bring certain uncerfij ln fij

ei ¼ i¼1

tainty to the assessment results. The spatial distribution of (i ¼ 1, 2 . . . , m and j ¼ 1, 2, . . . , n),

ln m rij and 0 < ej < 1: where fij ¼ m P rij

each indicator was plotted using the spatial analysis module in Arc GIS 10.3 software (Esri, Redlands, CA, (6)

USA). These results are shown in Figure 3. According to the characteristics of each index and the

i¼1

information it contained, the weights of each index were calIf fij is all the same value, then the entropy value of each

culated using the entropy weight method. The weight

criterion is the maximum value (ej ¼ 1). If fij is all 0, then fij

calculation results are shown in Table 1. According to the

ln fij is also 0 in value.

scores and weights of each indicator, this new ‘DRSTIC’ model was used to calculate the inherent vulnerability of the regional groundwater (Figure 4).

Step 3: calculation of the weight of entropy

To assess whether human activities have an impact on The weight of the entropy of the ith indicator can be formu-

the regional groundwater quality, this paper conducted a

lated as follows:

hydrogeological analysis of the shallow groundwater aquifer, the distributions of characteristic pollutants related to

wi ¼

1 ei m P n ei

where

n X

wi ¼ 1,

j¼1

i¼1

human activities, and hydrochemical characteristics. (7)

The Songnen Plain evolved from Mesozoic and Cenozoic subsidence basins, in which continental clastic sediments were deposited to a thickness of more than 8,000 m. Thus,

where the weight has a value 0 wi l.

the Songnen Plain is a large-scale groundwater aquifer

B. Wang et al.

1191

Figure 3

Evaluation method of the special vulnerability of groundwater

Hydrology Research

51.5

2020

Indicator scores of the DRSTIC model. (a)–(f) Results for depth of the water table (D), net recharge (R), soil media (S), topography (T), impact of the vadose zone media (I), and conductivity of the aquifer hydraulic (C), respectively.

B. Wang et al.

1192

Table 1

Evaluation method of the special vulnerability of groundwater

Weights of the DRSTIC model indicators (D, depth of the water table; R, net recharge; S, soil media; T, topography; I, impact of the vadose zone media; and C, conductivity of the aquifer hydraulic)

Hydrology Research

51.5

2020

mountainous areas to the west, north, and east. It ﬂows through the western piedmont plain, northern high plain, and the eastern undulating high plain, collecting into three

Indicators

streams ﬂowing to the south. The shallow groundwater

Weight

5.0

2.3

4.9

2.4

1.0

0.3

system interacts with the surface water system, comprising the Nenjiang, the Second Songhua River, and the Songhua

system comprising Quaternary pore water, Neogene ﬁssure

Rivers. Rain moves downwards by means of subsurface

water, Paleogene ﬁssure water, and Cretaceous pore and frac-

ﬂow and drains to the rivers. During runoff, evaporation

ture water. The regional shallow groundwater resource

and human usage consume a large amount of water, and

receives input from precipitation and recharge from adjacent

only a small amount ﬂows out of the study area.

Figure 4

Groundwater vulnerability within the study area based on the DRSTIC model.

B. Wang et al.

1193

Evaluation method of the special vulnerability of groundwater

Hydrology Research

51.5

2020

Runoff, circulation, and the regional aquifer character-

shallow groundwater of the study area is shown in Figure 5.

istics are all conducive to self-cleaning of the groundwater

Clearly, high nitrate concentration areas are mainly in the

system, maintaining good water quality under natural con-

northeast and southeast of the plain. Among them, values

ditions. However, according to our test results, the shallow

in the northeast are highest, reﬂecting pollution from agri-

groundwater in the study area is locally contaminated by

cultural areas and some industrial agglomeration areas;

nitrate. The proportion of groundwater samples with

high values in the southeast reﬂect industrial and densely

nitrate concentration more than 0.5 mg/L (China ground-

populated urban areas.

water quality standard class III) was 77.4%. The average

Based on sample test results, the groundwater samples

concentration of nitrate was 67.2 mg/L, and the highest con-

were classiﬁed using the Shukalev classiﬁcation, yielding a

centration is 1,000.0 mg/L. The distribution of nitrate in the

map reﬂecting groundwater chemistry. The Shukalev

Figure 5

Distribution of nitrate within the shallow groundwater of the study area.

B. Wang et al.

1194

Evaluation method of the special vulnerability of groundwater

Hydrology Research

51.5

2020

classiﬁcation is based on the concentrations of six major

in unconsolidated Quaternary deposits, the chemical

ions (Naþ, Ca2þ, Mg2þ, HCO3 , SO4 2 , Cl , Kþ incorpor-

composition

ated into Na ) and groundwater salinity.

the

groundwater

complex

and

varied, suggesting that it is greatly affected by environ-

Using groundwater index values, a Piper three-line

mental conditions and human activities. From the

diagram for shallow and deep groundwater hydrochemi-

piedmont to the plains, especially in populated areas,

cal types in the evaluation area was constructed from

the chemical signature of the groundwater varies.

the groundwater survey data (Figure 6). The main water

This reﬂects ﬂow velocities of downstream runoff,

chemistry type of the shallow groundwater in the

residence times of water–rock interactions, and inputs

study area was the HCO3–Ca type, accounting for 24.83% of

from

human

activities,

especially

artiﬁcial

mixing

all samples. This was followed by HCO3–Ca·Mg, HCO3–

such as groundwater exploitation, etc. (Zhang et al.

Na·Ca, HCO3–Na·Mg·Ca, HCO3·Cl–Ca, and HCO3·Cl–Na·Ca

; Li et al. , ). Human activities have led to

types. Although the shallow groundwater mainly resides

changes in groundwater chemistry in this region, and

Figure 6

Piper three-line diagram of shallow water chemical types within the study area (n ¼ 1,409).

B. Wang et al.

1195

Evaluation method of the special vulnerability of groundwater

clearly do have an impact on the regional groundwater

Hydrology Research

51.5

2020

Typically, areas with high special vulnerability of groundwater are mainly distributed in the north and east

quality. Because the regional groundwater has been affected by

of the Songnen Plain. These high vulnerability areas are con-

human activities, this paper selected additional indicators

sistent with those having serious nitrate pollution (Figure 5).

that reﬂect groundwater quality in the region, based on the

Human activities can cause the increase of nitrate concen-

distribution of characteristic pollutants and the driving

tration in groundwater (Jakobczyk-Karpierz et al. ;

forces of water pollution, mainly groundwater extraction

Teng et al. ). Under the same conditions, the regional

(E) and land-use (L). Water extraction data are available

nitrate concentration with high vulnerability of groundwater

from the Water Resources Bulletin of each city, while

should be higher.

land-use information comes from China’s 2015 satellite

To verify the accuracy of the evaluation results, they

remote sensing image data. In order to couple the above

were ﬁtted with the distribution of nitrate in the shallow

indexes into the model and draw a quantitative evaluation

groundwater of the study area. A correlation analysis

conclusion, the E and l indexes were assigned by the same

showed that the correlation coefﬁcient between these

principle as other indexes in the DRSTIC model, and the

two distributions was 0.2536. Comparing the vulner-

principle is that the attribute of the index is more likely to

ability of groundwater and the distribution of nitrate in

cause groundwater pollution, then it will get a higher

the study area, the correlation coefﬁcient increased

score. The speciﬁc scores of E and I in this paper came

from 0.0752 to 0.2536. This shows that human activities

from the Delphi method. The spatial distribution of each

do affect the groundwater in the study area, making it

indicator was plotted using Arc GIS 10.3 (Figure 7).

more vulnerable to pollution. Comparing Figures 4 and

The entropy weight method was used to calculate the

5, it is clear that the regions with serious nitrate pollution

weights of variables E and L. The weight of E was 3.3,

and those with high intrinsic groundwater vulnerability

while the weight of L was 5.0. This gave rise to a new

do not coincide, and that the correlation coefﬁcient

model, DRSTIC-EL. This article superimposed each indi-

between them also does not indicate an obvious relation-

cator to obtain groundwater vulnerability, as shown in

ship. This indicates that the shallow groundwater of the

Figure 8.

Songnen Plain has good self-cleaning potential. Although

Figure 7

Indicator scores for (a) groundwater exploitation (E) and (b) land-use (L) within the study area.

B. Wang et al.

1196

Figure 8

Evaluation method of the special vulnerability of groundwater

Hydrology Research

51.5

2020

Groundwater vulnerability within the study area based on the DRSTIC-EL model.

the correlation coefﬁcient is not statistically signiﬁcant,

conclusions of this study (Ahmed ; Albuquerque

given the large evaluation area and dataset, correlation

et al. ; Douglas et al. ; Modibo ). Despite

analysis at each grid point is only based on the software

all this, the quantitative impact of the above activities

platform, and the relationship between the vulnerability

on groundwater vulnerability is still far from clear, and

of groundwater and nitrate pollution cannot be compared

whether there are other inﬂuencing factors is not clear.

reliably at the regional scale. However, many studies

The research results of this paper are more inclined to

have shown that human activities, such as groundwater

determine

extraction, agricultural, and mining activities, lead to

from a macro perspective to provide guidance for water

increases in groundwater vulnerability, supporting the

resource management.

groundwater

vulnerability

characteristics

1197

B. Wang et al.

Evaluation method of the special vulnerability of groundwater

Hydrology Research

51.5

2020

CONCLUSIONS

DATA AVAILABILITY STATEMENT

A vulnerability assessment of the shallow groundwater

All relevant data are included in the paper or its Supplemen-

aquifer of the Songnen Plain was conducted using an

tary Information.

improved DRSTIC-LE model coupled with an entropy weight method to calculate the weighting of its indices. The evaluation results showed that the areas with high specific vulnerability of groundwater on the Songnen Plain are distributed to the north and east, consistent with areas affected by serious nitrate pollution of the groundwater. The correlation coefficient between them was 0.2536, an improvement over the result obtained from the direct calculation of the correlation coefficient between vulnerability and nitrate pollution. Chemical analysis of groundwater types also suggested that human activities have increased groundwater vulnerability on the Songnen Plain and made groundwater more vulnerable to pollution. This paper outlines a new way to evaluate the vulnerability of groundwater and validates this method using a case study. Our validation shows the applicability of the evaluation method to groundwater protection and management. Although the refined groundwater vulnerability assessment of each zone is difficult to achieve, in the area of regional groundwater resource management,

groundwater

vulnerability

assessment

undoubtedly provides a reliable reference in the macro aspect, which can relatively accurately measure the degree of regional groundwater vulnerability to pollution. Our assessment of groundwater vulnerability at such a large regional scale was limited by data availability. In the future or similar studies, more data should be obtained to improve the accuracy and reliability of the assessment.

ACKNOWLEDGEMENTS This work was supported by the National Natural Science Foundation of China (Nos. 41877355 and U19A20107) and Beijing Advanced Innovation Program for Land Surface Science. We thank Dr Trudi Semeniuk from Liwen Bianji, Edanz Editing China (www.liwenbianji.cn/ac), for editing the English text of a draft of this manuscript.

REFERENCES Ahmed, A. A.  Using generic and pesticide DRASTIC GISbased models for vulnerability assessment of the Quaternary aquifer at Sohag, Egypt. Hydrogeology Journal 17 (5), 1203–1217. https://doi.org/10.1007/s10040-009-0433-3. Ahmad, A. Y. & Al-Ghouti, M. A.  Approaches to achieve sustainable use and management of groundwater resources in Qatar: a review. Groundwater for Sustainable Development 11, 100367. https://doi.org/10.1016/j.gsd.2020. 100367. Albuquerque, M. T. D., Sanz, G., Oliveira, S. F., Martinez-Alegria, R. & Antunes, I. M. H. R.  Spatio-temporal groundwater vulnerability assessment – a coupled remote sensing and GIS approach for historical land cover reconstruction. Water Resources Management 27 (13), 4509–4526. https://doi.org/ 10.1007/s11269-013-0422-0. Barzegar, R., Moghaddam, A. A., Norallahi, S., Inam, A., Adamowski, J., Alizadeh, M. R. & Nassar, J. B.  Modiﬁcation of the DRASTIC framework for mapping groundwater vulnerability zones. Groundwater 58 (3), 441–452. https://doi.org/10.1111/gwat.12919. Bian, J., Zhang, Z. & Han, Y.  Spatial variability and health risk assessment of nitrogen pollution in groundwater in Songnen Plain. Journal of Chongqing University 38 (4), 104–111. https://doi.org/10.11835/j.issn.1000-582X.2015.04. 015. Bosch, A. P., Navarrete, F., Molina, L. & Martinezvidal, J. L.  Quantity and quality of groundwater in the Campo de Dalias (Almeria, SE Spain). Water Science and Technology 24 (11), 87–96. Brunsell, N. A., Schymanski, S. J. & Kleidon, A.  Quantifying the thermodynamic entropy budget of the land surface: is this useful? Earth System Dynamics 2 (1), 87–103. https://doi. org/10.5194/esd-2-87-2011. Cassardo, C. & Jones, J. A. A.  Managing water in a changing world. Water 3 (2), 618–628. https://doi.org/10.3390/ w3020618. Clausius, R.  The Mechanical Theory of Heat: With Its Applications to the Steam-Engine and to the Physical Properties of Bodies. John van Voorst, London. Clemens, M., Khurelbaatar, G., Merz, R., Siebert, C., van Afferden, M. & Roediger, T.  Groundwater protection under water scarcity; from regional risk assessment to local wastewater treatment solutions in Jordan. Science of the Total Environment 706, 136066. https://doi.org/10.1016/ j.scitotenv.2019.136066.

1198

B. Wang et al.

Evaluation method of the special vulnerability of groundwater

Denny, S. C., Allen, D. M. & Journeay, J. M.  DRASTIC-Fm: a modified vulnerability mapping method for structurally controlled aquifers in the southern Gulf Islands, British Columbia, Canada. Hydrogeology Journal 15 (3), 483–493. https://doi.org/10.1007/s10040-006-0102-8. Douglas, S. H., Dixon, B. & Griffin, D.  Assessing the abilities of intrinsic and specific vulnerability models to indicate groundwater vulnerability to groups of similar pesticides: a comparative study. Physical Geography 39 (6), 487–505. https://doi.org/10.1080/02723646.2017.1406300. Egbueri, J. C.  Groundwater quality assessment using pollution index of groundwater (PIG), ecological risk index (ERI) and hierarchical cluster analysis (HCA): a case study. Groundwater for Sustainable Development 10, 100292. https://doi.org/10.1016/j.gsd.2019.100292. Eliasson, J.  The rising pressure of global water shortages. Nature 517 (7532), 6. https://doi.org/10.1038/517006a. Erostate, M., Huneau, F., Garel, E., Ghiotti, S., Vystavna, Y., Garrido, M. & Pasqualini, V.  Groundwater dependent ecosystems in coastal Mediterranean regions: characterization, challenges and management for their protection. Water Research 172, 115461. https://doi.org/10. 1016/j.watres.2019.115461. Falkenmark, M.  Growing water scarcity in agriculture: future challenge to global water security. Philosophical Transactions of the Royal Society A: Mathematical Physical and Engineering Sciences 371 (2002). https://doi.org/10. 1098/rsta.2012.0410. Foster, S., Hirata, R. & Andreo, B.  The aquifer pollution vulnerability concept: aid or impediment in promoting groundwater protection? Hydrogeology Journal 21 (7), 1389–1392. https://doi.org/10.1007/s10040-013-1019-7. Gao, Y., Qian, H., Ren, W., Wang, H., Liu, F. & Yang, F.  Hydrogeochemical characterization and quality assessment of groundwater based on integrated-weight water quality index in a concentrated urban area. Journal of Cleaner Production 260, 121006. https://doi.org/10.1016/j.jclepro. 2020.121006. Gejl, R. N., Rygaard, M., Henriksen, H. J., Rasmussen, J. & Bjerg, P. L.  Understanding the impacts of groundwater abstraction through long-term trends in water quality. Water Research 156, 241–251. https://doi.org/10.1016/j.watres. 2019.02.026. Gogu, R. C. & Dassargues, A.  Current trends and future challenges in groundwater vulnerability assessment using overlay and index methods. Environmental Geology 39 (6), 549–559. https://doi.org/10.1007/s002540050466. Gogu, R. C., Hallet, V. & Dassargues, A.  Comparison of aquifer vulnerability assessment techniques. Application to the Neblon river basin (Belgium). Environmental Geology 44 (8), 881–892. https://doi.org/10.1007/s00254-003-0842-x. Hao, C., Zhang, W. & Gui, H.  Hydrogeochemistry characteristic contrasts between low- and high-antimony in shallow drinkable groundwater at the largest antimony mine

Hydrology Research

51.5

2020

in Hunan province, China. Applied Geochemistry 117, 104584. https://doi.org/10.1016/j.apgeochem.2020.104584. Harmancioglu, N.  Measuring the Information Content of Hydrological Processes by the Entropy Concept (in Turkish). PhD Thesis, Ege University, Izmir, Turkey. He, B., He, J., Wang, L., Zhang, X. & Bi, E.  Effect of hydrogeological conditions and surface loads on shallow groundwater nitrate pollution in the Shaying River Basin: based on least squares surface fitting model. Water Research 163, 114880. https://doi.org/10.1016/j.watres.2019.114880. Ibe, K. M., Nwankwor, G. I. & Onyekuru, S. O.  Assessment of ground water vulnerability and its application to the development of protection strategy for the water supply aquifer in Owerri, Southeastern Nigeria. Environmental Monitoring and Assessment 67 (3), 323–360. https://doi.org/ 10.1023/a:1006358030562. Işık, A. T. & Adalı, E. A.  The decision-making approach based on the combination of entropy and ROV methods for the apple selection problem. European Journal of Interdisciplinary Studies 3 (3). http://doi.org/10.26417/ejis.v3i3.p81-6. Islamoglu, M., Apan, M. & Oztel, A.  An evaluation of the financial performance of REITs in Borsa Istanbul: a case study using the entropy-based TOPSIS method. International Journal of Financial Research 6 (2), 124–138. http://doi.org/ 10.5430/ijfr.v6n2p124. Iván, V. & Mádl-Szőnyi, J.  State of the art of karst vulnerability assessment: overview, evaluation and outlook. Environmental Earth Sciences 76 (3), 1–25. https://doi.org/ 10.1007/s12665-017-6422-2. Jakobczyk-Karpierz, S., Sitek, S., Jakobsen, R. & Kowalczyk, A.  Geochemical and isotopic study to determine sources and processes affecting nitrate and sulphate in groundwater influenced by intensive human activity – carbonate aquifer Gliwice (southern Poland). Applied Geochemistry 76, 168–181. https://doi.org/10.1016/j.apgeochem.2016.12.005. Jia, Z., Bian, J. & Wang, Y.  Impacts of urban land use on the spatial distribution of groundwater pollution, Harbin City, Northeast China. Journal of Contaminant Hydrology 215, 29–38. https://doi.org/10.1016/j.jconhyd.2018.06.005. Khosravi, K., Sartaj, M., Tsai, F. T. C., Singh, V. P., Kazakis, N., Melesse, A. M., Prakash, I., Bui, D. T. & Binh Thai, P.  A comparison study of DRASTIC methods with various objective methods for groundwater vulnerability assessment. Science of the Total Environment 642, 1032–1049. https:// doi.org/10.1016/j.scitotenv.2018.06.130. Kumar, P., Bansod, B. K. S., Debnath, S. K., Thakur, P. K. & Ghanshyam, C.  Index-based groundwater vulnerability mapping models using hydrogeological settings: a critical evaluation. Environmental Impact Assessment Review 51, 38–49. https://doi.org/10.1016/j.eiar.2015.02.001. Li, X., Lin, X., Du, J. & Cui, J.  Analysis of hydrochemical evolution of phreatic water in Qiqihar City. Journal of Hydraulic Engineering 45 (7), 815–827. https://doi.org/10. 13243/j.cnki.slxb.2014.07.008.

1199

B. Wang et al.

Evaluation method of the special vulnerability of groundwater

Li, X., Wang, R. & Li, J.  Study on hydrochemical characteristics and formation mechanism of shallow groundwater in eastern Songnen Plain. Journal of Groundwater Science and Engineering 6 (3), 161–170. https://doi.org/10.19637/j.cnki.2305-7068.2018.03.001. Li, X., Tang, C., Cao, Y. & Li, D.  A multiple isotope (H, O, N, C and S) approach to elucidate the hydrochemical evolution of shallow groundwater in a rapidly urbanized area of the Pearl River Delta, China. Science of the Total Environment 724, 137930. https://doi.org/10.1016/j.scitotenv.2020. 137930. Lü, J., Qiu, H., Lin, H., Yuan, Y., Chen, Z. & Zhao, R.  Source apportionment of ﬂuorine pollution in regional shallow groundwater at You’xi County southeast China. Chemosphere 158, 50–55. https://doi.org/10.1016/j.chemosphere.2016.05. 057. Machiwal, D., Jha, M. K., Singh, V. P. & Mohan, C.  Assessment and mapping of groundwater vulnerability to pollution: current status and challenges. Earth-Science Reviews 185, 901–927. https://doi.org/10.1016/j.earscirev. 2018.08.009. Mancosu, N., Snyder, R. L., Kyriakakis, G. & Spano, D.  Water scarcity and future challenges for food production. Water 7 (3), 975–992. https://doi.org/10.3390/w7030975. Margat, J.  Groundwater Vulnerability to Contamination. Bureau de Recherches Géologiques et Minières (BRGM). Mendoza, J. A. & Barmen, G.  Assessment of groundwater vulnerability in the Rio Artiguas basin, Nicaragua. Environmental Geology 50 (4), 569–580. https://doi.org/10. 1007/s00254-006-0233-1. Modibo, S. A.  Groundwater Quality Origin Mechanism, and Exploitation in the Niger Basin Underclimate Change and Pollution Impact. PhD Thesis, Jilin University, China. Nobre, R. C. M., Rotunno Filho, O. C., Mansur, W. J., Nobre, M. M. M. & Cosenza, C. A. N.  Groundwater vulnerability and risk mapping using GIS, modeling and a fuzzy logic tool. Journal of Contaminant Hydrology 94 (3–4), 277–292. https://doi.org/10.1016/j.jconhyd.2007.07.008. Polemio, M., Casarano, D. & Limoni, P. P.  Karstic aquifer vulnerability assessment methods and results at a test site (Apulia, southern Italy). Natural Hazards and Earth System Sciences 9 (4), 1461–1470. https://doi.org/10.5194/nhess-91461-2009. Qiu, W.  Management Decision and Applied Entropy. China Machine Press, Beijing, China. Rahman, A.  A GIS based DRASTIC model for assessing groundwater vulnerability in shallow aquifer in Aligarh, India. Applied Geography 28 (1), 32–53. https://doi.org/10. 1016/j.apgeog.2007.07.008. Ruddell, B. L., Brunsell, N. A. & Stoy, P.  Applying information theory in the geosciences to quantify process uncertainty, feedback, scale. Eos, Transactions American Geophysical Union 94 (5), 56. http://doi.org/10.1002/ 2013EO050007.

Hydrology Research

51.5

2020

Saidi, S., Bouri, S. & Ben Dhia, H.  Groundwater vulnerability and risk mapping of the Hajeb-jelma aquifer (Central Tunisia) using a GIS-based DRASTIC model. Environmental Earth Sciences 59 (7), 1579–1588. https://doi.org/10.1007/ s12665-009-0143-0. Saidi, S., Bouri, S., Ben Dhia, H. & Anselme, B.  Assessment of groundwater risk using intrinsic vulnerability and hazard mapping: application to Souassi aquifer, Tunisian Sahel. Agricultural Water Management 98 (10), 1671–1682. https:// doi.org/10.1016/j.agwat.2011.06.005. Salman, S. A., Arauzo, M. & Elnazer, A. A.  Groundwater quality and vulnerability assessment in west Luxor Governorate, Egypt. Groundwater for Sustainable Development 8, 271–280. https://doi.org/10.1016/j.gsd.2018. 11.009. Shannon, C. E.  A mathematical theory of communications. Bell System Technical Journal 27 (3), 379–423. https://doi. org/10.1002/j.1538-7305.1948.tb01338.x. Shirazi, S. M., Imran, H. M. & Akib, S.  GIS-based DRASTIC method for groundwater vulnerability assessment: a review. Journal of Risk Research 15 (8), 991–1011. https://doi.org/ 10.1080/13669877.2012.686053. Shrestha, S., Kafle, R. & Pandey, V. P.  Evaluation of indexoverlay methods for groundwater vulnerability and risk assessment in Kathmandu Valley, Nepal. Science of the Total Environment 575, 779–790. https://doi.org/10.1016/j. scitotenv.2016.09.141. Singh, V. P.  Hydrologic synthesis using entropy theory: review. Journal of Hydrologic Engineering 16 (5), 421–433. https://doi.org/10.1061/(asce)he.1943-5584.0000332. Singh, A., Srivastav, S. K., Kumar, S. & Chakrapani, G. J.  A modified-DRASTIC model (DRASTICA) for assessment of groundwater vulnerability to pollution in an urbanized environment in Lucknow, India. Environmental Earth Sciences 74 (7), 5475–5490. https://doi.org/10.1007/s12665015-4558-5. Teng, Y., Zuo, R., Xiong, Y., Wu, J., Zhai, Y. & Su, J.  Risk assessment framework for nitrate contamination in groundwater for regional management. Science of the Total Environment 697, 134102. https://doi.org/10.1016/j. scitotenv.2019.134102. Thomann, J. A., Werner, A. D., Irvine, D. J. & Currell, M. J.  Adaptive management in groundwater planning and development: a review of theory and applications. Journal of Hydrology 586, 124871. https://doi.org/10.1016/j.jhydrol. 2020.124871. Wachniew, P., Zurek, A. J., Stumpp, C., Gemitzi, A., Gargini, A., Filippini, M., Rozanski, K., Meeks, J., Kvaerner, J. & Witczak, S.  Toward operational methods for the assessment of intrinsic groundwater vulnerability: a review. Critical Reviews in Environmental Science and Technology 46 (9), 827–884. https://doi.org/10.1080/10643389.2016.1160816. Wu, J., Sun, J., Liang, L. & Zha, Y.  Determination of weights for ultimate cross efficiency using Shannon entropy. Expert

1200

B. Wang et al.

Evaluation method of the special vulnerability of groundwater

Systems with Applications 38 (5), 5162–5165. https://doi. org/10.1016/j.eswa.2010.10.046. Wu, W., Yin, S., Liu, H. & Chen, H.  Groundwater vulnerability assessment and feasibility mapping under reclaimed water irrigation by a modiﬁed DRASTIC model. Water Resources Management 28 (5), 1219–1234. https://doi. org/10.1007/s11269-014-0536-z. Xu, J., Feng, P. & Yang, P.  Research of development strategy on China’s rural drinking water supply based on SWOT-TOPSIS method combined with AHPEntropy: a case in Hebei Province. Environmental Earth Sciences 75 (1), 58. https://doi.org/10.1007/s12665-0154885-6. Zhang, G., Deng, W., He, Y. & Ramsis, S.  Hydrochemical characteristics and evolution laws of groundwater in Songnen

Hydrology Research

51.5

2020

Plain, Northeast China. Advances in Water Science 17 (1), 20–28. https://doi.org/10.14042/j.cnki.32.1309.2006.01.004. Zhang, B., Li, G., Cheng, P., Yeh, T.-C. J. & Hong, M.  Landﬁll risk assessment on groundwater based on vulnerability and pollution index. Water Resources Management 30 (4), 1465–1480. https://doi.org/10.1007/s11269-016-1233-x. Zhu, W., Zhao, Y., Tang, W. & Du, J.  Study on shallow groundwater water quality status and development trends in Songnen Plain. Journal of Anhui Agriculture Science 41 (4), 1664–1669 þ 1744. https://doi.org/10.13989/j.cnki.05176611.2013.04.024. Zou, Z., Yun, Y. & Sun, J.  Entropy method for determination of weight of evaluating indicators in fuzzy synthetic evaluation for water quality assessment. Journal of Environmental Sciences 18 (5), 1020–1023. https://doi.org/10.1016/s1001-0742(06)60032-6.

First received 29 April 2020; accepted in revised form 15 July 2020. Available online 1 October 2020

1201

51.5

2020

A field investigation on rill development and flow hydrodynamics under different upslope inflow and slope gradient conditions Pei Tian, Chengzhong Pan, Xinyi Xu, Tieniu Wu, Tiantian Yang and Lujun Zhang

ABSTRACT Few studies focus on the quantitative impact of upslope inflow rate and slope gradient on rill development and erosion processes. Field plot experiments under varying inflow rates (6–36 L min 1 m 1) and slope gradients (26, 42 and 57%) were conducted to address this issue. The results showed soil loss rates significantly demonstrated temporal variability in relevance to the

Pei Tian Tieniu Wu Key Laboratory for Geographical Process Analysis & Simulation of Hubei Province, Central China Normal University, Wuhan 430079, China

rill developing process. Rill erosion and its contribution to soil loss increased with increasing inflow rates and slope gradients by power functions. There was a threshold inflow discharge (12–24 L min 1 m 1), under which, rill erosion became the dominant erosion pattern. At the initial stage, downcutting of rill bottom and headward erosion were obvious, whereas rill broadening was significant at the actively rill developing period. Rill density increased with slope gradient increasing from 26% to 42%, and then decreased. For the 57% slope under high inflow rates (24– 1

36 L min

m ), gravity caused an increase in the collapse of rills. Mean rill width increased with

increasing inflow rates but decreased as slope gradients increased, while mean rill depth increased with increasing inflow rates and slope gradients. Stream power and rill flow velocity were the best hydrodynamic parameter to simulate rill erosion and rill morphology, respectively. Key words

| ﬂow hydrodynamics, rill development, rill morphology, slope gradient, upslope inﬂow rate

HIGHLIGHTS

• • • •

Rill erosion and its contribution to soil loss increased with inﬂow rate and slope gradient by a power function. The threshold inﬂow rate that made rill erosion dominated decreased with increasing slope gradient. There was a threshold slope gradient between 42.3% and 57.4% where the rill network development began to weaken. Stream power was the best parameter simulating rill erosion with a good linear relationship.

Chengzhong Pan (corresponding author) Xinyi Xu College of Water Sciences, Beijing Normal University, Beijing 100875, China E-mail: pancz@bnu.edu.cn Tiantian Yang Lujun Zhang School of Civil Engineering and Environmental Science, The University of Oklahoma, Oklahoma 73019, USA

1202

P. Tian et al.

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Hydrology Research

51.5

2020

SYMBOLS AND ABBREVIATIONS S

Slope gradient (sine of the slope angle), %

erosivity of ﬂowing water exceeds a certain threshold of

Inﬂow discharge per unit width of the plot,

soil resistance; as a result, the micro-terrain of hillslope sur-

L min

face changes and gradually forms a rill (Govers et al. ;

Rill ﬂow depth, m

Knapen et al. ; Wang et al. ). The main factors

Surface rill ﬂow velocity, m s 1

affecting soil detachment and sediment transport on hill-

Mean rill ﬂow velocity, m s 1

slopes include slope gradients (Koulouri & Giourga ;

Correction factor in determining mean rill ﬂow

Fu et al. ), overland ﬂow velocity, and ﬂow depth

velocity

(Gimenez & Govers ). During rill erosion processes,

Acceleration due to gravity, m s 2

the evolution of rill networks greatly affects the conﬂuence

Reynolds number

path of runoff, soil loss, and the micromorphology of the

Froude number

slope surface (Consuelo et al. ; Auerswald et al. ;

Darcy–Weisbach resistance coefﬁcient 2

Fang et al. ; Tian et al. ). Especially on steep loess

Shear stress, N m

Stream power, N m 1 s 1

(Han et al. ; Yinglan et al. ), rills develop rapidly,

causing high soil loss (Lei et al. ; Zhang et al. ),

hillslopes with sparse vegetation cover in northern China

Unit stream power, m s

SLR

Soil loss rate, g m 2 min 1 2

RER

Rill erosion rate, g m

TRL

Total rill length, m Rill density, m m

and there is a lack of effective prevention methods to control 1

min

rill erosion on bare steep slopes in this region (Kou et al. ). Therefore, a further investigation of rill erosion processes on steep hillslopes is urgently needed.

MRW Mean rill width, cm

Rill erosion may be affected by hydrological parameters,

MRD Mean rill depth, cm

such as rill ﬂow stream power (Mahmoodabadi et al. a),

RWD

The rill width–depth ratio

runoff rate in rills (Wirtz et al. ), hydraulic slope, and rainfall intensity (Shen et al. ), as well as topographic factors (Kinnell ; Mahmoodabadi et al. b). The rill

INTRODUCTION

ﬂow has a signiﬁcant effect on the transportation of eroded soils, which will accordingly affect the soil erosion

Soil erosion is a critical concern for the sustainable develop-

rate (Di Stefano et al. ). For rills formed at the hillslope,

ment of agricultural regions worldwide (Bennett et al. ),

they quickly adapted their topographic characteristics in

and soil erosion could cause non-point source pollution

response to the changes of upslope inﬂow discharge and

(Wang et al. a, b), soil organic carbon loss (Fang

slope gradient (Nearing ; Nord & Esteves ; Stefano

et al. ), and biodiversity reduction (Li et al. ).

et al. ). Moreover, the erosivity of ﬂowing water is clo-

Thus, soil erosion is a serious threat to the environment

sely related to upslope inﬂow discharge and slope

and agricultural security (Jiang et al. ). Rill erosion is

gradient, thus, upslope inﬂow discharge and slope gradient

one of the main patterns of soil erosion by water on sloping

are two important factors for rill erosion (Berger et al.

croplands and rangelands (Bewket & Sterk ; Kimaro

; Sun et al. ; Shen et al. ).

et al. ; Wu et al. ).

The slope gradient is an important topographic factor

Rill erosion primarily results from concentrated surface

which has direct effects on ﬂow erosivity and inﬁltration,

ﬂow (Kimaro et al. ; He et al. ), which is the main

thus indirectly inﬂuences rill erosion processes (Shen

source of sediment yield in hillslope erosion processes

et al. ; Zhang et al. ). Giménez et al. () indi-

(Cerdan et al. ; Wagenbrenner et al. ; Mirzaee &

cated that slope roughness is enhanced with the slope

Ghorbani ). Generally, rill erosion occurs when the

gradient

becoming

steeper,

whereas

ﬂow

velocity

1203

P. Tian et al.

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Hydrology Research

51.5

2020

decreases until a critical hydraulic condition (Froude

formation and development process of rill morphology

number is 1) of the ﬂow is reached. It has also been recog-

timely responds to the rill erosion process. After the emer-

nized that the critical ﬂow hydraulic condition for rill

gence of rills, with the continuous ﬂowing water, rills

initiation is closely correlated to the slope gradient

develop and show the phenomenon of bifurcation, merging

(Ziadat & Taimeh ; Ran et al. ). In addition,

and connecting on the slope, and ﬁnally evolve into rill net-

many researchers found an increase in the slope gradient

works (He et al. ; Shen et al. ). The indicators

can lead to an increase in rill erosion amount (Berger

describing rill morphology are usually rill length, rill

et al. ; Fang et al. ). Ziadat & Taimeh ()

width, rill depth, and rill networks (Bewket & Sterk ;

reported that soil erosion in uncultivated land was

Raff et al. ). After the formation of rill networks, over-

mostly inﬂuenced by slope gradient; thus, the determi-

land ﬂows concentrated into rills can promote the

nation of slope gradient factor was an essential part of

development of rills, broadening the rill width and deepen-

rill erosion prediction (Sun et al. ). However, Liu

ing the rill length (Tian et al. ). Furthermore, the

et al. () observed that the relationship between slope

formation and development of rills contribute to the runoff

gradient and soil erosion varied when slope gradient

connectivity and concentration along the rill networks

reached certain thresholds. For instance, Yair & Klein

(Romero et al. ; Heras et al. ). However, the inﬂu-

() found the amount of soil erosion decreased with

ence mechanisms of inﬂow discharge and slope gradient

the increase of slope gradients. Koulouri & Giourga

on rill morphological characteristics and rill network for-

() stated that when the slope reached 40%, the domi-

mation need to be further investigated.

nant factor for soil loss was no longer land use, but slope

It is important to clarify the hydrodynamic mechanism

gradient. In addition, the study in a vineyard area by Pijl

of rill erosion since sediment detachment and transport by

et al. () indicated that rittochino (vertical cultivation)

overland

with a slope of 39.5% had the most severe soil loss.

initiation is determined by the established threshold ﬂow

ﬂow

are

energy-consuming

processes.

Rill

Typically, power functions are commonly used to ﬁt the

hydraulic conditions, and further rill development is con-

correlation between the slope gradient and the amount of

trolled by different thresholds (Wirtz et al. ). Rill

soil erosion (Di Stefano et al. ). In some experiments

erosion usually increases with increasing upslope inﬂow dis-

conducted on gentle slopes, signiﬁcant relationships were

charge due to the increase of ﬂow hydrodynamics (Di

found between rill erosion and slope gradient (Liu et al.

Stefano et al. ), such as ﬂow pattern (Foster et al.

; Sirjani & Mahmoodabadi ), whereas in some

), ﬂow velocity (Govers et al. ), depth (An et al.

other studies, no signiﬁcant relationship was reported (e.g.,

), shear stress (Nearing et al. ), and stream power

Chaplot & Le Bissonnais ). Zhang et al. () found

(Berger et al. ). Hillslopes’ soil loss signiﬁcantly

the soil detachment rate could be well predicted by a

increases after rill formation due to increases in rill ﬂow

power function of upslope inﬂow discharge and slope gradi-

depth, rill ﬂow velocity, and shear stress compared to

ent, and detachment rates were more sensitive to upslope

sheet ﬂow (Liu et al. ). Numerous studies have veriﬁed

inﬂow discharge than to slope gradients. Therefore, there

that the rill ﬂow hydraulic parameters are related to upslope

is an interaction effect between slope gradient and upslope

inﬂow discharge and slope gradient, whereas the relation-

inﬂow discharges on the amount of erosion and the critical

ships are different in the studies (Mirzaee & Ghorbani-

hydraulic conditions of rill initiation. Thus, it is necessary

Dashtaki ). For instance, Giménez & Govers () indi-

that more attention should be focused on the effects of

cated that rill ﬂow velocity is independent of the slope

upslope inﬂow discharges and slope gradients on rill ﬂow

gradient on the mobile bed, while it becomes slope-depen-

hydraulics and its dynamic mechanism at steep hillslopes.

dent on the ﬁxed bed. The ﬂow in rills has higher

Rill network and its morphological characteristics play

velocities, depth and sediment transport capacity than the

an important role in determining the ﬂow characteristics

ﬂow in the inter-rill areas (Gatto ; Mirzaee & Ghor-

of surface runoff and sediment delivery in hillslope erosion

bani-Dashtaki ). Furthermore, the rill bed form varies

processes (Zhang et al. ; Wu et al. ). In fact, the

with erosion processes, which in turn, changes the

1204

P. Tian et al.

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

hydraulics and erosion capacity of rill ﬂow (Zhang et al.

Hydrology Research

51.5

2020

Experimental plots

; Jiang et al. ). In summary, it is still not thoroughly understood what

A fallow loess hillslope (35 m length and 25 m width) was

mechanisms are dominating the inﬂow discharge and

selected as the experimental ﬁeld (Figure 1). The particle

slope gradient on the rill erosion process and its hydrodyn-

size distribution and basic physical and chemical properties

amic mechanisms, and the rill development and its

of the soils in the 0–20 cm surface layer are presented in

morphological characteristics, especially for steep hillslopes.

Tables 1 and 2, respectively.

Therefore, in this study, we performed ﬁeld plot experiments

The runoff plot experiments were conducted on the

under combinations of different upslope inﬂow discharges

selected hillslope. Lei et al. () found sediment transport

and slope gradients at steep loess hillslopes. The objectives

capacity would not increase beyond certain values of rill

of this study are to: (1) quantify the effects of slope gradient

length and slope. Furthermore, considering the character-

and inﬂow discharge on rill erosion processes and rill mor-

istics of small erosion plots, we designed the runoff plot

phological characteristics; (2) evaluate rill ﬂow hydraulic

of 5 m in length and 2 m in width, which is bounded by

characteristics and the hydrodynamic mechanism of rill

concrete walls or aluminum plastic plates. A speciﬁcally

erosion.

designed PVC pipe with a row of equally spaced (1.0 cm) grooves (0.5 cm width, 1.5 cm arc length) was ﬁxed at the top of the runoff plot and set along the longitudinal direction

MATERIALS AND METHODS

of the PVC pipe to make sure the upslope inﬂow water ﬂowed uniformly over the whole runoff plot. Meanwhile, a

Study site

PVC pipe was set at the bottom of the runoff plot to collect runoff sediment samples during the experiment. The sche-

The study site is located in the hilly loess area of northern

matic diagram of the runoff plot and its afﬁliated facilities

China, where the Guanting Reservoir is located (Figure 1).

is shown in Figure 2.

This region belongs to a transition zone between the northwest plateau of Beijing and the North China Plain,

Experimental design

which is one of the key areas implementing soil and water conservation in China. The Guanting Reservoir was built

In this study, we monitor the comparison results by setting the

in the early 1950s and long served as an important drinking

slope gradient at three levels, and inﬂow discharge at six levels.

water source for Beijing City. The hilly areas account for

For each slope gradient, two runoff plots with the same size

approximately 37% of the total area of Guanting Reservoir

and treatment were constructed at the selected hillslope. In

watershed (38 150 –41 14.20 N, 112 8.30 –116 20.60 E), and

total, six runoff plots with the same size of three different

most of the hilly areas are highly erodible loess, covering a

slope gradients were used in our experiments. All the exper-

total area of 500 km2 (Hu & Wang ). The main causes

iments were repeated twice to ensure the reproducibility of

leading to severe soil erosion here include erodible loessial

results. Given the fact that this experiment was performed in

soils, relatively sparse vegetation cover, and short-duration

the same area as that by Tian et al. (), six values of the

rainstorms (Tian et al. ). The climate is semi-arid conti-

inﬂow discharge per unit width (1 m) of the plot ranging

nental, with a mean annual temperature of 9 C (Li & Pan

from 6 to 36 L min 1 m 1 were used. Each inﬂow discharge

). Mean annual precipitation, most of which occurs in

was introduced at the head of the plot. Considering the

the summer months as short-duration and high-intensity

characteristics of hillslope erosion driven by short-duration

rainstorms, varies from 370 to 480 mm. Field runoff plot

rainstorms in the local area, the simulated scouring test lasts

experiments were performed at the Huailai Field Exper-

60 minutes for each experimental treatment. Rill erosion was

imental Station (40 150 32″ N, 115 370 01″ E), Beijing

the most signiﬁcant on bare loess slopes with slope gradients

Normal University, which is located along the east side of

of 17.6–57.7% (Shen et al. ). The rill is prone to intersect

the Guanting Reservoir (Figure 1).

on loess hillslopes with a slope gradient over 15 (26%),

P. Tian et al.

1205

Figure 1

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Hydrology Research

51.5

2020

(a) The study area and (b) the location of the experimental hillslopes with ﬁeld plots.

which is generally considered as the lower limit of steep slopes

driven by upslope inﬂow on a wide range of steep hillslopes,

in the hilly loess area of China. Although 25 (42%) is the upper

slope gradient (sine of the slope angle) of 26, 42, and 57%

limit of the slope gradient for cultivation in China set by the

were selected. According to the characteristics of short-term

government, the phenomenon of cultivation in steeper slopes

rainfall–runoff erosion in this area, the duration of each trial

is still common in the hilly loess area (Li & Pan ). There-

lasted 60 minutes after the initiation of surface runoff at the

fore, to investigate rill development and erosion processes

plot outlet.

P. Tian et al.

1206

Table 1

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Hydrology Research

51.5

2020

Particle size distribution for the topsoil soil (0–20 cm) of the sloping ﬁeld

Particle size distribution (%) 2–1 mm

1–0.5 mm

0.5–0.25 mm

0.25–0.1 mm

0.1–0.05 mm

0.05–0.02 mm

0.02–0.002 mm

< 0.002 mm

1.29 ± 0.07

1.21 ± 0.19

3.53 ± 0.66

7.42 ± 0.15

25.34 ± 0.35

30.40 ± 0.62

11.52 ± 0.24

19.50 ± 0.59

As an example, 1.29 ± 0.07 refers to the mean with standard deviations, the same in other cells.

Table 2

Basic physical and chemical properties for the topsoil (0–20 cm) of the sloping ﬁeld

the usage of backﬁll for the surface soil treatment, experiments with inﬂow discharges ranging from low to high were conducted.

Soil bulk

Soil water

Soil

Surface treatment

density (g cm 3)

content (%)

porosity (%)

Total N (mg g 1)

Organic C (mg g 1)

Bare slope

1.35 ± 0.50

13.0 ± 2.0

44.57 ± 2.32

0.56 ± 0.18

422 ± 5.49

As an example, 1.35 ± 0.50 refers to the mean with standard deviations, the same in other cells.

Runoff sediment samples were collected from the outlet of the plot at intervals of 4 minutes, and the duration of collecting each sample was recorded by a stopwatch. These samples were used to calculate soil loss rates and surface runoff rates during the experimental process. Surface ﬂow velocities in rills (vs) were measured by the dye (KMnO4) tracing method with the slope segment of 0.5–1.5 m, 1.5–2.5 m, 2.5–3.5 m, and 3.5–4.5 m, respectively. The water depth in rills (hr) was measured perpendicularly to the rill bed using a thin steel ruler with 0.1 mm precision and the measurement sections were the same as the velocity measurements. The measurements of rill morphology (length, width, depth) were performed along each rill at a 5 cm interval to compute the rills’ volume. Then, the rill erosion amount can be obtained by multiplying the rills’ volume with soil bulk density. In addition, photographs were taken during the experimental process. The maps of rill networks at the end of the experiment were plotted, which included three steps. First, as soon as the scouring test was over, we used a highdeﬁnition digital camera to take photos of the entire plot. Second, according to the distribution of rills formed on the

Figure 2

Field runoff plot system at the study site.

plot, we drew a sketch of the rill network on white paper according to a certain scale. Third, for each rill, we measured

Experimental procedures and measurements

its length and the width every 3–5 cm along the length of the rill. Last, the maps of rill networks were made by Adobe

Before the experiment, upslope inﬂow discharge was cali-

Photoshop, combining the photos from the above three steps.

brated to the designed values manually. In addition, the topsoils (0–20 cm) were sampled by cutting rings on the

Data calculation and analysis

top, middle, and bottom of the slope surface to measure the soil water content and dry bulk density. For each treatment, the antecedent soil water content and soil bulk density were controlled as 13 ± 2% and 1.3 ± 0.1 g cm 3, respectively, before the start of the experiment. To reduce

The slope gradient (S, %) in this study is the sine of the slope angle (θ, ) for ﬁeld plots: S ¼ sin θ × 100%

(1)

1207

P. Tian et al.

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

The surface runoff rate at the plot outlet (qout) is calcu-

51.5

2020

Shear stress (Foster et al. ), stream power (Bagnold ), and unit stream power (Yang ) are used to reﬂect

lated as follows: qout ¼ (M m)=(1000ρ bt)

Hydrology Research

(2)

where qout is surface runoff rate at the plot outlet, L min 1 m 1; M is the total mass of a runoff sediment sample, g; m is the dry weight of sediment in a runoff sediment sample, g; ρ is the density of water used in the experiment, g L 1; t is the duration of collecting a runoff sediment sample, min; b is the width of the plot, m; 1,000 is the dimensional adjustment coefﬁcient.

the hydrodynamic parameters of rill ﬂows, which are calculated as: 8 < τ ¼ ρghr J W ¼ τvr : φ ¼ vr J

(7)

where τ is ﬂow shear stress (N m 2); W is ﬂow stream power (N m 1 s 1); φ is unit stream power (m s 1); ρ is the density of water (kg m 3).

The runoff coefﬁcient (RC) corresponding to each runoff sample is obtained by the following equation (Tian et al. ):

Rill density (RD) is the total rill lengths (TRL) divided by the whole hillslope area (m m 2) (Bewket & Sterk ). The RD value is suitable for roughly describing the morphological characteristics of rills developed at hillslopes. The

RC ¼ qout =qin

(3)

where qin is upslope inﬂow discharge (L min 1 m 1).

formulas of TRL and RD are: RD ¼ TRL=A0

Mean rill ﬂow velocity (vr) is estimated by the measured surface rill ﬂow velocity (vs) multiplied by a correlation factor α as the equation: vr ¼ αvs

(4)

and 0.8 for turbulent flow (Horton et al. ; Li et al. ). Reynolds number (Re) and Froude number (Fr) can be used to reflect the rill flow regime, which are calculated as: Re ¼ vr hp r =v m ffiffiffiffiffiffiffi Fr ¼ vr = ghr

n X

TRLi

(9)

i¼1

where a is 0.67 for laminar ﬂow, 0.7 for transitional ﬂow,

TRL ¼

(8)

where TRL is the total lengths of all rills on the studied hillslope, m; RD is the rill density, m m 2; TRLi is the total length of a rill and its bifurcations (m); and i ¼ 1, …, n represents the number of the rills; A0 is the plane area of the studied hillslope (m2). In addition, for a certain cross section of a rill, the rill width–depth ratio (RWD) can reﬂect the shape of the rill cross section. For a certain cross section of a rill, RWD is

(5)

where hr is the mean rill flow depth (m); νm is kinematical viscosity (m2 s 1); and g is acceleration of gravity (m s 2). The resistance to rill flow is estimated by Darcy–Weisbach friction coefficient ( f), which is calculated as:

calculated as follows: RWD ¼ MRW=MRD

(10)

where MRW is the mean rill width, cm; MRD is the mean rill depth, cm; RWD is the rill width–depth ratio. The amount of rill erosion was estimated by volumetric measurements of rills on each soil erosion plot. In our exper-

f ¼ 8 g hr J =v2r

(6)

iments, hillslope soil loss is the sum of the amount of sheet erosion and rill erosion. Thus, sheet erosion was obtained by

where g is gravitational acceleration (m s 2); J is surface 1

subtracting the rill erosion from the hillslope soil loss. The

slope (m m ), which is equal to S in value; hr is the depth

statistical tests include variance analysis, correlation analy-

of rill ﬂow (mm).

sis, equation simulation, and were conducted using the

P. Tian et al.

1208

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Hydrology Research

51.5

2020

IBM SPSS Statistics for Windows Version 19.0. The data

conditions. This indicated the temporal variation increased

graphs were drawn using ORIGIN 9.0.

with the increase of inﬂow discharges. With the same inﬂow discharge, the surface runoff rates were higher under higher slope gradient conditions. This indicated

RESULTS AND DISCUSSION

that, to a certain extent, soil inﬁltration rates decreased with increasing slope gradients. Overall, under the same

Temporal changes in surface runoff and soil loss

inﬂow discharge condition, there were no signiﬁcant differences in surface runoff processes among different

Figure 3 shows the temporal variations of surface runoff rates at the runoff plot outlet during the experimental process under different slope gradients and inﬂow discharges. According to Figure 3, there were no signiﬁcant differences in the trend of surface runoff rate over time under

treatments. To quantify the overall effect of slope gradients and inﬂow discharges on surface runoff production, the mean runoff coefﬁcients during the whole experimental process for different treatments are presented in Figure 4.

different treatments. For each treatment, surface runoff

Under the same slope gradient condition, the mean

rates increased with time in the initial stage of the exper-

runoff coefﬁcients slightly increased with the increasing

iment, and then gradually became steady. With the same

inﬂow discharges (Figure 4), indicating soil inﬁltration

slope gradient, the increasing speed of surface runoff rate

rates decreased as inﬂow discharges increased. Under the

was

same inﬂow discharge condition, the mean runoff

higher

Figure 3

under

higher

upslope

inﬂow

discharge

Temporal changes in surface runoff rates under three slope gradients and six inﬂow discharges. S stands for the slope gradient. (a) S ¼ 26%, (b) S ¼ 42% (c) S ¼ 57%.

P. Tian et al.

1209

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Hydrology Research

51.5

2020

as inflow discharges increased. This phenomenon implied that the decreasing effect of slope gradients on soil infiltration rates became weakened with the increases of inflow discharges. To explore the temporal variations of soil loss during the experiment, the changes in soil loss rates with duration for different inflow discharges and slope gradients are illustrated in Figure 5. According to Figure 5, the soil loss rates between different treatments were of obvious differences, especially for different inflow discharges. Under different conditions, the starting time for rill initiation varied. Specifically, the starting time of rill formation became earlier with the Figure 4

Mean runoff coefﬁcients for different slope gradients and inﬂow discharges.

increasing slope gradient and inﬂow discharge, which indicated that rills formed more easily under steeper slopes

coefﬁcients increased with the increase of slope gradients,

and high inﬂow discharges. In addition, the duration of rill

suggesting the soil inﬁltration rates decreased as slope gra-

initiation reduced as the slope gradient increased. For

dients increased. However, the rate of increment in mean

most treatments, the soil loss rate curves were high in the

runoff coefﬁcients with slope gradients gradually decreased

early minutes during the experimental process, which was

Figure 5

Temporal changes in soil loss rates under different slope gradients and inﬂow discharges. (a) Slope gradient ¼ 26%, (b) slope gradient ¼ 42%, (c) slope gradient ¼ 57%.

P. Tian et al.

1210

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Hydrology Research

51.5

2020

mainly because the loose soil particles were relatively

Under the lowest inﬂow discharge (6 L min 1 m 1), the

abundant on the hillslope surface at the beginning of the

curves of soil loss rate generally declined with slight ﬂuctu-

experiment, leading to numerous sediments being detached

ation during the entire experiment. For relatively high inﬂow

and transported by surface sheet ﬂow, especially under high

discharges, the curves of soil loss rate sharply decreased to

inﬂow discharges. This phenomenon corresponds to the

the valley at approximately 10 to 20 minutes, which was

experimental ﬁndings of Parsons & Stone ().

mainly due to decreases in the detachment rate of soil par-

Overall, for most experimental treatments, two peak

ticles and the number of loose particles on the slope

values of soil loss rates existed during the whole exper-

surface. Then, the soil loss rate curves reached another

imental process. At the initial stage of the experiment,

crest at approximately 25 to 45 minutes after the run,

there was much inﬁltration on the 57% slope and the

which was mainly due to the active development of rills,

erosion capacity of the surface runoff is limited. In

contributing to a sharp increment in the soil loss rate. There-

addition, the soil loss rate was relatively low at the begin-

after, the soil loss rate curve descended and tended to be an

ning of the experiment, as shown in Figure 5(c). After 8

equilibrium since the development of rills gradually entered

minutes, there appeared a ﬁrst peak value of soil loss

into a relatively steady stage.

rate, which was because the runoff increased and there were abundant erodible soil particles on the hillslope sur-

Contributions of slope gradient and inﬂow discharge to

face. Thus, the soil loss rate also increased rapidly

rill erosion

afterwards. As the experiment continued, the soil loss rate decreased with the reduction of erodible soil par-

For each slope gradient treatment, rill erosion amount, hill-

ticles. During 20 to 35 minutes, with the active

slope soil loss, and the contribution of rill erosion to

evolution of rill erosion, the soil loss rate increased to

hillslope soil loss gradually increased with the increase of

the second peak value. After 35 minutes, the rill develop-

inﬂow discharges (Table 3). This phenomenon indicated

ment entered a relatively stable stage, and the soil loss

that the erosion capacity of ﬂowing water greatly increased

rate slowly decreased.

as inﬂow discharges increased. For the lowest slope gradient

Table 3

Rill erosion, sheet erosion, hillslope soil loss, and the contribution of rill erosion to hillslope soil loss for different slope gradients and inﬂow discharges

Rill erosion (kg)

Sheet erosion (kg)

Hillslope soil loss (kg)

Contribution of rill erosion to hillslope soil loss (%)

6 12 18 24 30 36

1.08 3.16 9.58 18.36 33.97 43.34

3.83 6.42 13.78 13.85 18.29 19.47

4.92 9.58 23.36 32.21 52.27 62.81

22.0 33.0 41.0 57.0 65.0 69.0

6 12 18 24 30 36

3.28 12.40 27.73 39.86 64.89 81.47

11.76 19.39 17.07 18.30 21.35 17.87

15.04 31.78 44.80 58.16 86.24 99.33

21.8 39.0 61.9 68.5 75.2 82.0

6 12 18 24 30 36

5.19 14.61 37.56 56.75 73.10 90.46

8.83 12.96 16.87 18.92 19.43 15.96

14.01 27.57 54.43 75.66 92.53 106.43

37.0 53.0 69.0 75.0 79.0 85.0

Slope gradient (%)

Inﬂow discharge (L min

)

1211

P. Tian et al.

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Hydrology Research

51.5

2020

condition (26%), sheet erosion increased with increasing

changed very slightly (15.04–99.33 kg and 14.01–106.43 kg

inﬂow discharges. However, with the slope becoming stee-

at slope gradients of 42 and 57%, respectively).

per, sheet erosion increased as inﬂow discharges increased

Soil loss rate (SLR) and rill erosion rate (RER) can be

up to 30 L min 1 m 1, and then decreased as inﬂow dis-

expressed by inﬂow discharges (q) and slope gradients (S)

charges continued to increase.

as dual power functions, respectively, as follows:

For the lowest slope gradient condition, the contribution of rill erosion to hillslope soil loss was 41.0% and 57.0%,

SLR ¼ 0:0199q1:239 S1:075

(R2 ¼ 0:954, p < 0:01, n ¼ 54)

respectively, under inﬂow discharges of 18 and 24

(11)

L min 1 m 1. This result showed that rill erosion became the dominant soil erosion pattern instead of sheet erosion

RER ¼ 0:0004q1:868 S1:524 (R2 ¼ 0:977, p < 0:01, n ¼ 54)

when inﬂow discharge increased to 24 L min 1 m 1. Therefore, we can conclude that there was a threshold of inﬂow discharge ranging from 18 and 24 L min 1 m 1, under which, the dominant soil erosion pattern started to change from sheet erosion to rill erosion. With the slope gradient increasing to 42%, rill erosion became the dominant erosion 1

(12) where SLR is soil loss rate, g m 2 min 1; RER is rill erosion, g m 2 min 1; q is inﬂow discharge per unit width, L min 1 m 1; and S is slope gradient, %. According to Equations (11) and (12), both soil loss and

m , which

rill erosion rates increased by a power function with an

indicated that the critical inﬂow discharge controlling the

increase in either inﬂow discharge or slope gradient. Com-

main soil erosion pattern was between 12 and 18

pared with the exponents of inﬂow discharge, the slope

pattern when inﬂow discharge was 18 L min

L min 1 m 1. However, with slope gradient increasing to

gradients were lower, which indicated that the inﬂow dis-

57%, rill erosion became the dominant soil erosion pattern

charge played a more important role in both soil loss and

when inﬂow discharge was 12 L min 1 m 1, which demon-

rill erosion rates than the slope gradient. In addition, both

strated that the critical inﬂow discharge controlling the

the exponents of inﬂow discharge and slope gradient in

main soil erosion pattern was between 6 and 12

Equation (11) were lower than those in Equation (12).

L min

m . In summary, the threshold inﬂow discharge

that controlled the dominant soil erosion pattern was related

Rill networks and morphology

to slope gradients. Overall, the critical inﬂow discharge that determined the dominant soil erosion pattern decreased as

Rill networks

slope gradients increased. For the same inﬂow discharge, with increasing slope

To explore the characteristics of rill networks developed at

gradients, both rill erosion and the contribution of rill ero-

different slopes, the rill networks at the end of the exper-

sion to hillslope soil loss generally increased, whereas the

iment for different inﬂow discharges and slope gradients

variations of sheet erosion showed no deﬁnite trend with

are illustrated in Figure 6

slope gradients. However, hillslope soil loss signiﬁcantly

Based on the rills generated on the slope surface at the

increased with slope gradients varying from 26% to 42%,

end of the experiment (Figure 6), for each slope gradient

and then slightly varied as slope gradients ranged from

treatment, both the uniformity and density of rill network

42% to 57%. The variations in the contribution of rill ero-

distribution were enhanced with the increases of inﬂow

sion to hillslope soil loss and hillslope soil loss with slope

discharges.. In addition, the cross-merging phenomenon

gradients showed that the increase of slope gradients

between the rills became more obvious under higher

enhanced rill erosion, although it implied that there was

inﬂow discharges, which indicated that the rill network

probably a threshold of slope gradient where soil erosion

became more complex. Under the same inﬂow discharge

began to weaken. For example, rill erosion increased from

condition, the density of rill network increased with slope

3.28–81.47 kg to 5.19–90.46 kg with the increase of slope

gradient increasing from 26% to 42%, and then decreased

gradient from 42% to 57%, whereas hillslope soil loss

with the slope becoming steeper. This phenomenon was

P. Tian et al.

1212

Figure 6

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Hydrology Research

51.5

2020

Rill networks at the end of the experiment for different slope gradients and inﬂow discharges. (a) Slope gradient ¼ 26%, (b) slope gradient ¼ 42%, (c) slope gradient ¼ 57%.

more signiﬁcant under higher inﬂow discharges. As the

depth of rills increased. This phenomenon was probably

slope gradient increased from 42% to 57%, the total length

different from previous studies, and was mainly related to

of rills decreased by 8.6–25.1% under inﬂow rates of 12–36

our experimental condition. In this study, rills were

L min

m , while the average rill depth increased by

formed only by the upslope inﬂow without raindrop

17.5–35.7%. Although the rill density decreased, the average

impacts. Under this circumstance, when the slope gradient

1213

P. Tian et al.

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Hydrology Research

51.5

2020

increased to a certain level (57%), the undercutting erosivity

the rills. In contrast, under other slope gradients and

of runoff increased signiﬁcantly with the inﬂuence of grav-

inﬂow rate conditions, the main driving force of soil erosion

ity, making the rills continue to deepen. As a result, more

and rill development is the upslope inﬂow. For these treat-

runoff ﬂowed into the deep rills and the number of rills

ments, the role of gravity on erosion and rill evolution was

formed was reduced compared to the 42% slope condition.

not obvious.

Therefore, the total length of rills formed under the 57% slope condition was less than that under the 42% slope,

Rill morphological characteristics

which was reﬂected in the decrease of rill density. This may also imply that there was a critical slope gradient

Different statistical indicators of rill morphology for differ-

between 42 and 57%, under which, the development of rill

ent slope gradients and inﬂow discharges are presented in

networks began to weaken. For the largest slope (57%) con-

Table 4.

dition, rills on the upper slope were signiﬁcantly denser than

Based on the dynamic changes of total rill length, mean

those on the lower part of the slope, especially under

rill width, and the rill width–depth ratio with increasing

relatively low inﬂow discharges. This phenomenon was

slope gradients (Table 4), the vertical erosion in rills was

probably due to the strong erosion capacity of ﬂowing

exaggerated with the slope becoming steeper since the grav-

water with low sediment concentration in the upper part

itational erosion occurred. This result is similar to He et al.

of the slope on the steep hillslopes.

(), in which the authors investigated the dynamic

As reported in a previous study (Zhang et al. ), the

changes of total rill length under different simulated rainfall

formation and evolution of drop pits on the slope surface

conditions. For each slope gradient treatment, total rill

was an essential procedure during rill development pro-

length, rill density, and mean rill depth gradually increased

cesses. Their ﬁnding has agreed with our experiments, in

as inﬂow discharges increased, while the increment rate

which the drop pit phenomenon was clearly observed at

declined with the increases in inﬂow discharges. However,

the initial stage of rill formation during the experimental process for all treatments. For the 26 and 42% slope, all

Table 4

rills were generally continuous at the end of the experiment. However, on the steeper slope (57%), there were some drop

Rill morphological characteristics for different slope gradients and inﬂow discharges

Slope

pits and discontinuous rills at the lower-middle part of the

gradient

Inﬂow discharge

runoff plot under relatively low inﬂow discharges at the

(S, %)

(q, L min

end of the experiment. A possible explanation for this

TRL

RD (m

MRW

MRD

(m)

(cm)

RWD

6 12 18 24 30 36

12.24 17.85 22.52 27.74 32.68 36.36

1.22 1.79 2.25 2.77 3.27 3.64

1.8 2.9 4.7 5.5 6.2 6.5

0.8 1.1 1.7 2.4 3.1 3.4

2.31 2.59 2.86 2.27 2.01 1.91

6 12 18 24 30 36

15.30 22.04 26.64 33.72 38.46 44.28

1.53 2.20 2.66 3.37 3.85 4.43

1.6 2.5 3.9 4.7 5.6 6.0

1.0 1.4 2.2 2.9 3.7 4.0

1.58 1.75 1.78 1.63 1.50 1.50

6 12 18 24 30 36

17.24 20.10 24.64 28.36 32.22 33.15

1.72 2.01 2.46 2.84 3.22 3.32

1.4 2.0 3.6 4.5 5.3 5.8

1.4 1.9 2.6 3.3 4.4 4.7

1.05 1.08 1.37 1.38 1.20 1.23

phenomenon was related to the deep rills developed at the upper part of the slope. In this case, most of the surface runoff converged into the deep rills, and intensively scoured the deep soil layer. During this process, some of the surface runoff might turn into the interflow, which decreased the surface upslope inflow discharge over the lower part of the runoff plot. Thus, the drop pits and discontinuous rills at the lower part of the slope were unable to develop into continuous rills due to the weakened energy of flowing water. Furthermore, based on our observation, the gravitational erosion obviously occurred within the plot of 57% slope under upslope inflow rates of 24, 30, and 36 L min 1 m 1. Under these circumstances, we can conclude that gravity plays an important role in slope erosion and rill evolution, which causes an increase in the collapse of

)

TRL: total rill lengths of the whole plot, m; RD: rill density, m m 2; MRW: mean rill width, cm; MRD: mean rill depth, cm; RWD: the rill width-depth ratio.

1214

P. Tian et al.

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Hydrology Research

51.5

2020

under the same inﬂow discharge condition, mean rill width

important sediment preparing processes frequently occurred

generally showed a decreasing trend with the increases of

at steep hillslopes, such as headcut retreat, step-pool effects,

slope gradients. Furthermore, the rill width–depth ratio

and bank collapse, which were not adequately considered

varied slightly with the increase of inﬂow discharges, demon-

(Dong et al. ; Zegeye et al. ). Through the obser-

strating no obvious variation trend. This result is similar to

vations during the experiment, sediment yield at the runoff

the laboratory rill experiments conducted by Shen et al.

plot outlet primarily resulted from knickpoints, chutes,

() and He et al. (). According to a least signiﬁcant

meanders, and sidewall sloughing. This was consistent

difference (LSD) test, mean rill depth under the lowest

with the ﬁndings by Stefanovic & Bryan (), who inves-

slope gradient was signiﬁcantly different from that under

tigated rill development processes on loamy sand and

the highest slope gradient condition, whereas mean rill

sandy loam under laboratory conditions.

width and total rill length among three different slope gradient treatments were not signiﬁcantly different.

According to the analysis above, the rill morphology was inﬂuenced by the interaction effects of inﬂow dis-

Under the same inﬂow discharge condition, rill density

charges and slope gradients. At the same time, the

ﬁrst increased with the increase of slope gradients up to

regression analysis also indicated that rill morphological

42% or 57%, and then decreased as the slope gradient con-

indicators could not be well expressed by only inﬂow dis-

tinued to increase. For each inﬂow discharge treatment,

charges or slope gradients. The best ﬁtting functions for

although both mean rill width and rill depth increased

rill morphological indicators, inﬂow discharges (q), and

with increasing slope gradients, the rill width–depth ratio

slope gradients (S) are presented in Table 5.

showed a decreasing trend as slope gradients increased.

Rill density, mean rill depth, and the rill width–depth

This phenomenon indicated that with the increase of slope

ratio can be described by the power function of inﬂow dis-

gradients, deepening rills were more signiﬁcant than the

charges and slope gradients, whereas mean rill width was

widening rills because of the higher shear stress of concen-

well expressed by the linear function relationship of inﬂow

trated rill ﬂow. The ﬁndings were consistent with the

discharges and slope gradients (Table 5). All the ﬁtted

observations in rill experiments under ﬁeld conditions by

equations were highly signiﬁcant (p < 0.01). In addition,

Mirzaee & Ghorbani-Dashtaki (). For the same inﬂow

the coefﬁcient of the inﬂow discharge was greater than

discharge condition, mean rill width and rill depth under

that of the slope gradient for all equations, which indicated

different slope gradients were not signiﬁcantly different

that the inﬂow discharge factor played a more important

according to an LSD test. However, the rill density was sig-

role in the rill morphological indicator than the slope gradi-

niﬁcantly different among six different inﬂow discharges

ent factor. The function form of mean rill width, inﬂow

according to an LSD test. For the highest slope gradient,

discharge, and slope gradient was consistent with the

the side wall collapse of rills was more signiﬁcant than

result by Lei & Nearing (), who conducted ﬂume exper-

that of lower slope gradients, mainly due to the obvious

iments under upslope inﬂow conditions.

phenomenon of gravity erosion under steep slopes. During the experimental process, one of the main sediment delivery processes was observed to be headwall retreat erosion at small knickpoints or steps, which contributed to the formation of continuous rills. Furthermore, for

Table 5

The ﬁtted equation of rill morphological indicators, inﬂow discharges and slope gradients

steeper slopes (57%), bank collapse processes of rills due

Regression equation

Sig.(p)

to the interactions of ﬂowing water and gravity were another

RD ¼ 0.346S 0.125q 0.533

0.92

<0.01

signiﬁcant erosion pattern resulting in soil loss. These

MRW ¼ 0.027S þ 0.157q þ 1.947

0.96

<0.01

0.96

<0.01

0.85

<0.01

phenomena were consistent with the observations for ﬁeld

MRD ¼ 0.033S

rill experiments under inﬂow conditions by Wirtz et al.

RWD ¼ 32.734 S 0.798q 0.016

() and Tian et al. (). Therefore, the limitation of soil detachment or sediment transport equations was that

0.519 0.798

RD: rill density, (m m 2); MRW: mean rill width (cm); MRD: mean rill depth (cm); RWD: the rill width-depth ratio; S: slope gradient (%); q: inﬂow discharge (L min 1 m 1).

P. Tian et al.

1215

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Hydrology Research

51.5

2020

Hydrodynamic mechanisms of rill erosion and

to 57%, respectively. For all slope gradients, the Re value

morphology

began to exceed 2,000 when the inﬂow discharge increased to 24 L min 1 m 1, which implied that rill ﬂows became tur-

Rill ﬂow hydraulics

bulent ﬂow according to the criterion of open channel ﬂow (Tian et al. ). Fr was greater than 1.0 for all treatments,

Rill ﬂow hydraulic parameters over the whole course of the

indicating that rill ﬂows belonged to supercritical ﬂow con-

experiment for different inﬂow discharges and slope gradi-

ditions. This observation is similar to the experimental

ents are presented in Table 6.

results for steep calcareous soils (Mirzaee & Ghorbani

Overall, mean rill ﬂow velocity (vr) increased with

). In addition, Darcy–Weisbach resistance coefﬁcient

increases in either slope gradient or inﬂow discharge

( f ) decreased with the increases of inﬂow discharges but

(Table 6); however, there were signiﬁcant differences in vr

increased as the slope became steeper.

among six different inﬂow discharges, but slight differences among three different slope gradients. This result indicated

Correlations of rill erosion and rill ﬂow hydrodynamics

that ﬂow velocities in rills were mainly dependent on the magnitude of inﬂow discharges, which is similar to the

The variations of rill erosion rates with three hydrodynamic

observations by Nearing (), and Li et al. (), who

parameters of the rill ﬂow, respectively, under different

found that ﬂow velocities were largely independent of

slope gradients are illustrated in Figure 7. Average rill ﬂow shear stress (τ), stream power (W ), and

slope gradients. Mean rill ﬂow depth (hr) increased with increasing

unit stream power (φ) varied from 0.645 to 4.345 N m 2,

gradients

0.183 to 3.252 N m 1 s 1, and 0.061 to 0.419 m s 1, respect-

increased. Reynolds number (Re) increased as inﬂow dis-

ively, and the values were increasing with the increases of

charges

either upslope inﬂow discharge or slope gradient (Figure 7).

inﬂow discharges increased,

but

decreased

whereas

slope

increased

and

then

decreased with increases of slope gradients, in which the

Rill erosion rates signiﬁcantly increased with an increase of

slope gradients increased from 26% to 42% and from 42%

τ, W, and φ. The relationships between rill erosion rates and

Table 6

Mean rill ﬂow hydraulic parameters for different slope gradients and inﬂow discharges

Slope gradient (%)

Inﬂow discharge (L min 1 m 1)

Rill ﬂow velocity (m s 1)

Rill ﬂow depth (×10 3 m)

Reynolds number

Froude number

Friction coefﬁcient

6 12 18 24 30 36

0.284 0.368 0.455 0.527 0.607 0.659

2.41 3.50 4.57 5.97 7.21 8.86

544 1,021 1,653 2,498 3,477 4,640

1.9 2.0 2.2 2.2 2.3 2.2

1.28 1.11 0.94 0.92 0.84 0.87

6 12 18 24 30 36

0.329 0.448 0.564 0.640 0.699 0.733

2.13 3.28 4.30 5.51 6.64 8.12

556 1,166 1,927 2,798 3,690 4,728

2.3 2.5 2.7 2.8 2.7 2.6

1.46 1.22 1.01 1.00 1.01 1.13

6 12 18 24 30 36

0.362 0.464 0.574 0.629 0.711 0.748

1.99 3.13 3.88 4.63 5.40 6.21

571 1,152 1,768 2,311 3,050 3,689

2.6 2.6 2.9 3.0 3.1 3.0

1.70 1.63 1.32 1.31 1.20 1.24

P. Tian et al.

1216

Figure 7

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Hydrology Research

51.5

2020

Rill erosion rates versus rill ﬂow shear stress, stream power, and unit stream power, respectively, for all treatments.

the above three hydrodynamic parameters were described

The critical hydrodynamic parameter could be obtained

well by the linear functions, which are listed as follows:

by the above functions. When no rill erosion occurred, i.e.,

8 < RER ¼ 48:57(τ 0:756) RER ¼ 53:98(τ 1:016) : RER ¼ 56:73(τ 1:433)

RER ¼ 0, then the critical value was determined. All the

(S ¼ 25:9%, R2 ¼ 0:96) (S ¼ 42:3%, R2 ¼ 0:98) (S ¼ 57:4%, R2 ¼ 0:97)

where RER is rill erosion rate (g m (N m 2), 8 < RER ¼ 60:33(W 0:218) RER ¼ 60:52(W 0:283) : RER ¼ 59:67(W 0:406)

hydrodynamic parameters, and further enhance the detach-

min ), τ is shear stress

). As a result, rill erosion rates increased with the

ment power and transport capacity of rill ﬂows (Wang et al. increases of slope gradients. Moreover, as the slope becomes

(S ¼ 25:9%, R2 ¼ 0:97) (S ¼ 42:3%, R2 ¼ 0:98) (S ¼ 57:4%, R2 ¼ 0:98)

steeper, the gravity inﬂuenced the ﬂow geometry in rills and increased the rill bed roughness, which led to an increase in the critical shear stress and the critical stream power for rill erosion (Nearing ). At the same time, for all treatments, the functional relations between rill erosion rates and those

W is stream power (N m 1 s 1),

three hydrodynamic parameters are analyzed, which are (S ¼ 25:9%, R2 ¼ 0:92) (S ¼ 42:3%, R2 ¼ 0:89) (S ¼ 57:4%, R2 ¼ 0:96) (15)

φ is unit stream power (m s 1).

increase in slope gradient will result in higher values of

(13)

(14)

8 < RER ¼ 997:04(φ 0:070) RER ¼ 912:31(φ 0:135) : RER ¼ 732:87(φ 0:209)

critical values increased as slope gradient increased. The

expressed as: 8 < RER ¼ 47:887(τ 0:916) RER ¼ 57:507(W 0:255) : RER ¼ 39:800(φ 0:051)

(R2 ¼ 0:89, p < 0:01, n ¼ 54) (R2 ¼ 0:64, p < 0:01, n ¼ 54) (R2 ¼ 0:92, p < 0:01, n ¼ 54) (16)

P. Tian et al.

1217

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Hydrology Research

51.5

2020

where RER is rill erosion rate (g m 2 min 1), τ is rill ﬂow

According to Equation (17), when no rills formed, i.e.,

shear stress (N m 2), W is stream power (N m 1 s 1), φ is

MRD ¼ 0, then the critical rill ﬂow velocity was determined.

unit stream power (m s 1). The critical shear stress, the criti-

Thus, in our experiments, the critical ﬂow velocity for rill

cal stream power, and the critical unit stream power

occurrence could be estimated as 0.224 m s 1. Furthermore,

computed by Equation (16) were 0.916 N m , 0.255

this critical value was close to the measured ﬂow velocity at

N m 1 s 1, and 0.051 m s 1, respectively. In contrast, the

the initial stage of rill development during the experiment

best hydrodynamic parameter to ﬁt the rill erosion rate

(He et al. ).

was unit stream power with a good linear relationship.

CONCLUSIONS

Correlations of rill morphology and rill ﬂow hydrodynamics

Plot experiments under six upslope inﬂow discharges (6–36 The Pearson correlation coefﬁcients between rill morpho-

L min 1 m 1) and three slope gradients (26–57%) were con-

logical indicators and rill ﬂow hydrodynamic parameters

ducted to explore the quantitative effects of inﬂow discharge

for all treatments are shown in Table 7.

and slope gradient on rill erosion and rill morphology, as

There was a negative correlation between the ﬂow resist-

well as the hydrodynamic mechanism of rill erosion. The

ance coefﬁcient and all the rill morphological indicators

results showed soil loss rates signiﬁcantly related to the ﬂuc-

(Table 7). However, RD, MRW, and MRD were positively

tuations during the process of rill development. Both rill

correlated with the other hydrodynamic parameters. Speciﬁ-

erosion and its contribution to hillslope soil loss increased

cally, RWD was positively related to mean rill ﬂow depth and

with the increases in inﬂow discharge and slope gradient,

Reynolds number but negatively correlated with other

which can be described by a power function. The inﬂow dis-

parameters. The most sensitive rill ﬂow hydrodynamic par-

charge played a more important role than the slope gradient.

ameter to RD, MRW, MRD, and RWD was Reynolds

There was a threshold of inﬂow discharge (12–24

number, mean rill ﬂow depth, mean rill ﬂow velocity, and

L min 1 m 1), under which, rill erosion instead of sheet ero-

Froude number, respectively. Overall, the common sensitive

sion became the dominant erosion pattern, and the critical

parameter to various rill morphological indicators was shear

inﬂow discharge reduced as slope gradient increased. The

stress and stream power.

uniformity of rill network distribution enhanced with

In addition, there was a good linear function relation-

increasing inﬂow discharges. Rill density increased (17.7–

ship between mean rill depth (MRD) and rill ﬂow velocity

25.4%) with slope gradient increasing from 26% to 42%,

(vr), which was expressed as:

and then decreased (7.5–25.1%) with the slope becoming steeper. Mean rill width increased with the increases of inﬂow discharge but decreased as the slope gradient (R2 ¼ 0:88, p < 0:01, n ¼ 54)

MRD ¼ 7:972(vr 0:224)

(17)

Table 7

increased. However, mean rill depth increased with increases in inﬂow discharge and slope gradient. The rill

Correlation coefﬁcients between rill morphological indicators and rill ﬂow hydrodynamic parameters for all treatments vr

0.939**

0.919**

0.965**

0.438

0.548*

0.785**

0.818**

0.478*

MRW

0.859**

0.954**

0.934**

0.233

0.764**

0.633*

0.689*

0.298

MRD

0.971**

0.782**

0.880**

0.661*

0.326

0.944**

0.963**

0.753**

RWD

-0.376

0.871

0.579

0.785**

0.611*

0.609*

0.533*

0.780**

*: significant at p ¼ 0.05; **: significant at p ¼ 0.01; vr: rill flow velocity; hr: rill flow depth; Re: Reynolds number; Fr: Froude number; f : Darcy–Weisbach friction coefficient; τ: shear stress; W: stream power; φ: unit stream power; MRD: mean rill depth; MRW: mean rill width; RD: rill density; RWD: the rill width-depth ratio.

P. Tian et al.

1218

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

width–depth ratio decreased as slope gradients increased,

Hydrology Research

51.5

2020

REFERENCES

which indicated that deepening rills were more significant than the widening rills with the slope becoming steeper. In addition, stream power was the best hydrodynamic parameter correlated with rill erosion with a good linear relationship, and the critical stream power of rill erosion was 0.255 N m 1 s 1. The most sensitive hydrodynamic parameter to rill density, mean rill width, rill depth, and the rill width–depth ratio was Reynolds number, rill flow depth, rill flow velocity, and Froude number, respectively. Although the most sensitive hydrodynamic parameters to various rill morphological indicators were different, the common sensitive parameters were rill flow stream power and shear stress. The findings will help to understand the impacts of upslope inflow rate on morphological variations of rills, rill network development, and hydrodynamic mechanism of rill erosion on steep hillslopes. Weakening the erosion power of upslope runoff, such as taking measures to cover the soil surface, is useful for preventing rill formation and development, as well as the downslope soil loss.

ACKNOWLEDGEMENTS The authors wish to thank all of the technicians involved in ﬁeld and laboratory work. This research is jointly funded by the

Natural

Science

Foundation

China

Project

(41907061), the Research Center on Mountain Torrent & Geologic Disaster Prevention of the Ministry of Water Resources, Changjiang River Scientiﬁc Research Institute (CKWV2019761/KY), and the Research Program from the State Key Laboratory of Soil Erosion and Dryland Farming on the Loess Plateau (A314021402-2005). We declare

that

have

ﬁnancial

and

personal

relationships with other people or organizations that can inappropriately inﬂuence our work. The authors declare no conﬂicts of interest.

DATA AVAILABILITY STATEMENT All relevant data are included in the paper or its Supplementary Information.

An, J., Zheng, F., Lu, J. & Li, G.  Investigating the role of raindrop impact on hydrodynamic mechanism of soil erosion under simulated rainfall conditions. Soil Science 177 (8), 517–526. Auerswald, K., Fiener, P. & Dikau, R.  Rates of sheet and rill erosion in Germany – a meta-analysis. Geomorphology 111 (3), 182–193. Bagnold, R. A.  An Approach to the Sediment Transport Problem From General Physics. US Geological Survey Professional Paper 442-I. USGS, Reston, VA, USA. Bennett, S. J., Gordon, L. M., Neroni, V. & Wells, R. R.  Emergence, persistence, and organization of rill network. Natural Hazards 79, S7–S24. doi: 10.1007/s11069-0151599-8. Berger, C., Schulze, M., Rieke-Zapp, D. & Schlunegger, F.  Rill development and soil erosion: a laboratory study of slope and rainfall intensity. Earth Surface Processes and Landforms 35 (12), 1456–1467. Bewket, W. & Sterk, G.  Assessment of soil erosion in cultivated fields using a survey methodology for rills in the Chemoga watershed, Ethiopia. Agriculture, Ecosystems and Environment 97, 81–93. Cerdan, O., Le Bissonnais, Y., Couturier, A., Bourennane, H. & Souchère, V.  Rill erosion on cultivated hillslopes during two extreme rainfall events in Normandy, France. Soil and Tillage Research 67 (1), 99–108. Chaplot, V. A. M. & Le Bissonnais, Y.  Runoff features for sheet erosion at different rainfall intensities, slope lengths, and gradients in an agricultural loessial hillslope. Soil Science Society of America Journal 67, 844–851. Consuelo, C. R., Stroosnijder, L. & Guillermo, A.  Sheet and rill erodibility in the northern Andean Highlands. Catena 70 (2), 105–113. Di Stefano, C., Ferro, V., Palmeri, V. & Pampalone, V.  Flow resistance equation for rills. Hydrological Processes 31, 2793–2801. Di Stefano, C., Ferro, V., Palmeri, V. & Pampalone, V.  Testing slope effect on flow resistance equation for mobile bed rills. Hydrological Processes 32 (5), 664–671. Dong, Y., Xiong, D., Su, Z., Duan, X., Lu, X., Zhang, S. & Yuan, Y.  The influences of mass failure on the erosion and hydraulic processes of gully headcuts based on an in situ scouring experiment in dry-hot valley of China. Catena 176, 14–25. Fang, H., Sun, L. & Tang, Z.  Effects of rainfall and slope on runoff, soil erosion and rill development: an experimental study using two loess soils. Hydrological Processes 29 (11), 2649–2658. Fang, Q., Wang, G., Liu, T., Xue, B. & Y, A.  Controls of carbon flux in a semi-arid grassland ecosystem experiencing wetland loss: vegetation patterns and environmental variables. Agricultural and Forest Meteorology 259, 196–210.

1219

P. Tian et al.

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

Foster, G. R., Huggins, L. F. & Meyer, L. D.  A laboratory study of rill hydraulics: II: Shear stress relationships. Transactions of the ASAE 27, 797–804. Fu, S., Liu, B., Liu, H. & Xu, L.  The effect of slope on sheet erosion at short slopes. Catena 84, 29–34. Gatto, L. W.  Soil freeze-thaw-induced changes to a simulated rill: potential impacts on soil erosion. Geomorphology 32, 147–160. Giménez, R. & Govers, G.  Interaction between bed roughness and flow hydraulics in eroding rills. Water Resources Research 37 (3), 791–799. Gimenez, R. & Govers, G.  Flow detachment by concentrated flow on smooth and irregular beds. Soil Sci Soc Am J 66 (5), 1475–1483. Giménez, R., Planchon, O., Silvera, N. & Govers, G.  Longitudinal velocity patterns and bed morphology interaction in a rill. Earth Surface Processes and Landforms 29 (1), 105–114. Govers, G., Giménez, R. & Van, O. K.  Rill erosion: exploring the relationship between experiments, modelling and field observations. Earth-Science Reviews 84 (3–4), 87–102. Han, D., Wang, G., Liu, T., Xue, B., Kuczera, G. & Xu, X.  Hydroclimatic response of evapotranspiration partitioning to prolonged droughts in semiarid grassland. Journal of Hydrology 563, 766–777. He, J., Li, X., Jia, L., Gong, H. & Cai, Q.  Experimental study of rill evolution processes and relationships between runoff and erosion on clay loam and loess. Soil Science Society America Journal 78, 1716–1725. He, J., Sun, L., Gong, H., Cai, Q. & Jia, L.  The characteristics of rill development and their effects on runoff and sediment yield under different slope gradients. Journal of Mountain Science 13 (3), 397–404. He, J., Sun, L., Gong, H. & Cai, Q.  Laboratory studies on the influence of rainfall pattern on rill erosion and its runoff and sediment characteristics. Land Degradation & Development 28 (5), 1615–1625. He, J., Sun, L., Gong, H. & Cai, Q.  Comparison of rill flow velocity regimes between developing and stationary rills. Catena 167, 13–17. Heras, M. M.-D., Espigares, T., Merino-Martín, L. & Nicolau, J. M.  Water-related ecological impacts of rill erosion processes in Mediterranean-dry reclaimed slopes. Catena 84, 114–124. Horton, R. E., Leach, H. R. & Vliet, R. V.  Laminar sheet-flow. Transactions American Geophysical Union 15 (2), 393–404. Hu, C. & Wang, Y.  Study on water-sediment optimum allocation in upstream basin and comprehensive measures of sediment control in Guanting reservoir. Reservoir sedimentation and general plan of water and sediment control. Journal of Sediment Research 2, 19–26 (in Chinese with English abstract). Jiang, F., Zhan, Z., Chen, J., Lin, J., Wang, M., Ge, H. & Huang, Y.  Rill erosion processes on a steep colluvial deposit slope under heavy rainfall in flume experiments with artificial rain. Catena 169, 46–58.

Hydrology Research

51.5

2020

Jiang, Y., Shi, H., Wen, Z., Guo, M., Zhao, J., Cao, X., Fan, Y. & Zheng, C.  The dynamic process of slope rill erosion analyzed with a digital close range photogrammetry observation system under laboratory conditions. Geomorphology 350, 106893. Kimaro, D. N., Poesen, J., Msanya, B. M. & Deckers, J. A.  Magnitude of soil erosion on the northern slope of the Uluguru Mountains, Tanzania: sheet and rill erosion. Catena 75, 38–44. Kinnell, P. I. A.  The impact of slope length on the discharge of sediment by rain impact induced saltation and suspension. Earth Surface Process and Landforms 34, 1393–1407. Knapen, A., Poesen, J., Govers, G., Gyssels, G. & Nachtergaele, J.  Resistance of soils to concentrated flow erosion: a review. Earth Science Reviews 80 (1–2), 75–109. Kou, P., Xu, Q., Yunus, A., Dong, X., Pu, C., Zhang, X. & Jin, Z.  Micro-topographic assessment of rill morphology highlights the shortcomings of current protective measures in loess landscapes. Science of the Total Environment 737, 139721. Koulouri, M. & Giourga, C.  Land abandonment and slope gradient as key factors of soil erosion in Mediterranean terraced lands. Catena 69 (3), 274–281. Lei, T. & Nearing, M. A.  Flume experiments for determining rill hydraulic characteristic erosion and rill patterns. Journal of Hydraulic Engineering l31 (11), 49–54 (in Chinese with English abstract). Lei, T., Zhang, Q., Zhao, J. & Tang, Z.  A laboratory study of sediment transport capacity in the dynamic process of rill erosion. Transaction of American Society and Agriculture Engineers 44, 1537–1542 (in Chinese with English abstract). Li, C. & Pan, C.  The relative importance of different grass components in controlling runoff and erosion on a hillslope under simulated rainfall. Journal of Hydrology 558, 90–103. Li, G., Abrahams, A. D. & Atkinson, J. F.  Correction factors in the determination of mean velocity of overland flow. Earth Surface Processes and Landforms 21 (6), 509–515. Li, G., Zheng, F., Lu, J., Xu, X., Hu, W. & Han, Y.  Inflow discharge impact on hillslope erosion processes and flow hydrodynamics. Soil Science Society of America Journal 80 (3), 711–719. Li, Q., Wang, G., Wang, H., Shrestha, S., Xue, B., Sun, W. & Yu, J.  Macrozoobenthos variations in shallow connected lakes under the influence of intense hydrologic pulse changes. Journal of Hydrology 584, 124755. Liu, B., Nearing, M. & Risse, L.  Slope gradient effects on soil loss for steep slopes. Transactions of the ASAE 37 (6), 1835–1840. Liu, G., Yang, M., Warrington, D., Liu, P. & Tian, J.  Using beryllium-7 to monitor the relative proportions of sheet and rill erosion from loessal soil slopes in a single rainfall event. Earth Surface Processes and Landforms 36 (4), 439–448. Mahmoodabadi, M., Ghadiri, H., Yu, B. & Rose, C. a Morphodynamic quantification of flow-driven rill erosion parameters based on physical principles. Journal of Hydrology 514, 328–336. Mahmoodabadi, M., Ghadiri, H., Rose, C., Yu, B., Rafahi, H. & Rouhipour, H. b Evaluation of GUEST and WEPP with a

1220

P. Tian et al.

Upslope inﬂow and slope gradient on rill developing and ﬂow hydrodynamics

new approach for the determination of sediment transport capacity. Journal of Hydrology 513, 413–421. Mirzaee, S. & Ghorbani-Dashtaki, S.  Deriving and evaluating hydraulics and detachment models of rill erosion for some calcareous soils. Catena 164, 107–115. Nearing, M.  Hydraulics and erosion in eroding rills. Water Resources Research 33 (4), 865–876. Nearing, M., Bradford, M. & Parker, C.  Soil detachment by shallow flow at low slope. Soil Science Society of America Journal 55 (2), 339–344. Nord, G. & Esteves, M.  The effect of soil type, meteorological forcing and slope gradient on the simulation of internal erosion processes at the local scale. Hydrological Processes 24 (13), 1766–1780. Parsons, A. J. & Stone, P. M.  Effects of intra-storm variations in rainfall intensity on sheet runoff and erosion. Catena 67 (1), 68–78. Pijl, A., Reuter, L. E. H., Quarella, E., Vogel, T. A. & Tarolli, P.  GIS-based soil erosion modelling under various steepslope vineyard practices. Catena 193, 104604. Raff, D. A., Ramírez, J. A. & Smith, J. L.  Hillslope drainage development with time: a physical experiment. Geomorphology 62, 169–180. Ran, H., Deng, Q., Zhang, B., Liu, H., Wang, L., Luo, M. & Qin, F.  Morphology and influencing factors of rills in the steep slope in Yuanmou Dry-Hot Valley (SW China). Catena 165, 54–62. Romero, C. C., Stroosnijder, L. & Baigorria, G. A.  Sheet and rill erodibility in the northern Andean Highlands. Catena 70, 105–113. Shen, H. O., Zheng, F. L., Wen, L. L., Lu, J. & Jiang, Y. L.  An experimental study of rill erosion and morphology. Geomorphology 231, 193–201. Shen, H. O., Zheng, F. L., Wen, L. L., Han, Y. & Hu, W.  Impacts of rainfall intensity and slope gradient on rill erosion processes at loessial hillslope. Soil & Tillage Research 155, 429–436. Sirjani, E. & Mahmoodabadi, M.  Effects of sheet flow rate and slope gradient on sediment load. Arabian Journal of Geosciences 7, 203–210. Stefano, C. D., Ferro, V., Pampalone, V. & Sanzone, F.  Field investigation of rill and ephemeral gully erosion in the Sparacia experimental area, South Italy. Catena 101, 226–234. Stefanovic, J. R. & Bryan, R. B.  Flow energy and channel adjustments in rills developed in loamy and sandy loam soils. Earth Surface Processes and Landforms 34, 133–144. Sun, L., Fang, H., Qi, D., Li, J. & Cai, Q.  A review on rill erosion process and its influencing factors. Chinese Geographical Science 23 (4), 389–402. Tian, P., Xu, X., Pan, C., Hsu, K. & Yang, T.  Impacts of rainfall and inflow on rill formation and erosion processes on steep hillslopes. Journal of Hydrology 548 (5), 24–39. Wagenbrenner, J. W., Robichaud, P. R. & Elliot, W. J.  Rill erosion in natural and disturbed forests: 2. Modeling approaches. Water Resources Research 46, 1–12.

Hydrology Research

51.5

2020

Wang, Z., Yang, X., Liu, J. & Yuan, Y.  Sediment transport capacity and its response to hydraulic parameters in experimental rill flow on steep slope. Journal of Soil and Water Conservation 70 (1), 36–44. Wang, D., Wang, Z., Shen, N. & Chen, H.  Modeling soil detachment capacity by rill flow using hydraulic parameters. Journal of Hydrology 535, 473–479. Wang, G., Li, J., Sun, W., Xue, B., Y, A. & Liu, T. a Non-point source pollution risks in a drinking water protection zone based on remote sensing data embedded within a nutrient budget model. Water Research 157, 238–246. Wang, P., Yao, J., Wang, G., Hao, F., Shrestha, S., Xue, B., Xie, G. & Peng, Y. b Exploring the application of artificial intelligence technology for identification of water pollution characteristics and tracing the source of water quality pollutants. Science of the Total Environment 693, 133440. Wirtz, S., Seeger, M. & Ries, J. B.  Field experiments for understanding and quantification of rill erosion processes. Catena 91, 21–34. Wu, S., Chen, L., Wang, N., Yu, M. & Assouline, S.  Modeling rainfall-runoff and soil erosion processes on hillslopes with complex rill network planform. Water Resources Research 54 (12), 10117–10133. Wu, S., Chen, L., Wang, N., Li, J. & Li, J.  Two-dimensional rainfall-runoff and soil erosion model on an irregularly rilled hillslope. Journal of Hydrology 580, 124346. Yair, A. & Klein, M.  The influence of surface properties on flow and erosion processes on debris covered slopes in an arid area. Catena 1 (73), 1–18. Yang, C. T.  Sediment Transport: Theory and Practice. McGraw Hill, New York, USA. Yinglan, A., Wang, G., Liu, T., Xue, B. & Kuczera, G.  Spatial variation of correlations between vertical soil water and evapotranspiration and their controlling factors in a semi-arid region. Journal of Hydrology 574, 53–63. Zegeye, A. D., Langendoen, E. J., Steenhuis, T. S., Wolde, M. & Tilahun Seifu, A.  Bank stability and toe erosion model as a decision tool for gully bank stabilization in sub humid Ethiopian highlands. Ecohydrology and Hydrobiology 20, 301–311. Zhang, G. H., Liu, B. Y., Nearing, M. A., Huang, C. H. & Zhang, K. L.  Soil detachment by shallow flow. Transactions of the ASAE 45, 351–357. Zhang, P., Tang, H., Yao, W., Zhang, N. B. & Xi, Z.  Experimental investigation of morphological characteristics of rill evolution on loess slope. Catena 137, 536–544. Zhang, P., Yao, W., Tang, H., Wei, G. & Wang, L.  Laboratory investigations of rill dynamics on soils of the loess plateau of China. Geomorphology 293, 201–210. Ziadat, F. M. & Taimeh, A. Y.  Effect of rainfall intensity, slope and land use and antecedent soil moisture on soil erosion in an arid environment. Land Degradation Development 24 (6), 582–590.

First received 24 May 2020; accepted in revised form 10 August 2020. Available online 24 September 2020