Environmental & Engineering Geoscience May 2017 Vol. 23 No. 2 by Association of Environmental & Engineering Geologists (AEG)

Environmental & Engineering Geoscience MAY 2017

VOLUME XXIII, NUMBER 2

Waiting on Image

THE JOINT PUBLICATION OF THE ASSOCIATION OF ENVIRONMENTAL AND ENGINEERING GEOLOGISTS AND THE GEOLOGICAL SOCIETY OF AMERICA SERVING PROFESSIONALS IN ENGINEERING GEOLOGY, ENVIRONMENTAL GEOLOGY, AND HYDROGEOLOGY

Environmental & Engineering Geoscience (ISSN 1078-7275) is published quarterly by the Association of Environmental & Engineering Geologists (AEG) and the Geological Society of America (GSA). Periodicals postage paid at AEG, 1100 Brandywine Blvd, Suite H, Zanesville, OH 43701-7303 and additional mailing offices. EDITORIAL OFFICE: Environmental & Engineering Geoscience journal, Department of Geology, Kent State University, Kent, OH 44242, U.S.A. phone: 330-672-2968, fax: 330-672-7949, ashakoor@kent.edu. CLAIMS: Claims for damaged or not received issues will be honored for 6 months from date of publication. AEG members should contact AEG, 1100 Brandywine Blvd, Suite H, Zanesville, OH 43701-7303. Phone: 844-331-7867. GSA members who are not members of AEG should contact the GSA Member Service center. All claims must be submitted in writing. POSTMASTER: Send address changes to AEG, 1100 Brandywine Blvd, Suite H, Zanesville, OH 43701-7303. Phone: 844331-7867. Include both old and new addresses, with ZIP code. Canada agreement number PM40063731. Return undeliverable Canadian addresses to Station A P.O. Box 54, Windsor, ON N9A 6J5 Email: returnsil@imexpb.com. DISCLAIMER NOTICE: Authors alone are responsible for views expressed inarticles. Advertisers and their agencies are solely responsible for the content of all advertisements printed and also assume responsibility for any claims arising therefrom against the publisher. AEG and Environmental & Engineering Geoscience reserve the right to reject any advertising copy. SUBSCRIPTIONS: Member subscriptions: AEG members automatically receive digital access to the journal as part of their AEG membership dues. Members may order print subscriptions for $60 per year. GSA members who are not members of AEG may order for $60 per year on their annual GSA dues statement or by contacting GSA. Nonmember subscriptions are $295 and may be ordered from the subscription department of either organization. A postage differential of $10 may apply to nonmember subscribers outside the United States, Canada, and Pan America. Contact AEG at 844-331-7867; contact GSA Subscription Services, Geological Society of America, P.O. Box 9140, Boulder, CO 80301. Single copies are $75.00 each. Requests for single copies should be sent to AEG, 1100 Brandywine Blvd, Suite H, Zanesville, OH 43701-7303. © 2017 by the Association of Environmental and Engineering Geologists All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or by any information storage and retrieval system, without permission in writing from AEG. THIS PUBLICATION IS PRINTED ON ACID-FREE PAPER Abdul Shakoor Department of Geology Kent State University Kent, OH 44242 330-672-2968 ashakoor@kent.edu

EDITORS

Brian G. Katz Florida Department of Environmental Protection 2600 Blair Stone Rd. Tallahassee, FL 32399 850-245-8233 eegeditorbkatz@gmail.com

EDITORIAL BOARD Jerome V. DeGraff CSU Fresno Chester (Skip) F. Watts Radford University Thomas Oommen Michigan Technological Univ. Syed E. Hasan University of Missouri

Thomas J. Burbey Virginia Polytechnic Institute Abdul Shakoor Kent State University Brian G. Katz Florida Department of Environmental Protection

ASSOCIATE EDITORS John W. Bell Nevada Bureau of Mines and Geology Richard E. Jackson Geofirma Engineering, Ltd. Jeffrey R. Keaton AMEC Americas Paul G. Marinos National Technical University of Athens, Greece June E. Mirecki U.S. Army Corps of Engineers Peter Pehme Waterloo Geophysics, Inc Nicholas Pinter Southern Illinois University

Paul M. Santi Colorado School of Mines Robert L. Schuster U.S. Geological Survey Roy J. Shlemon R. J. Shlemon & Associates, Inc. Greg M. Stock National Park Service Resat Ulusay Hacettepe University, Turkey Chester F. “Skip” Watts Radford University Terry R. West Purdue University

SUBMISSION OF MANUSCRIPTS Environmental & Engineering Geoscience (E&EG), is a quarterly journal devoted to the publication of original papers that are of potential interest to hydrogeologists, environmental and engineering geologists, and geological engineers working in site selection, feasibility studies, investigations, design or construction of civil engineering projects or in waste management, groundwater, and related environmental fields. All papers are peer reviewed. The editors invite contributions concerning all aspects of environmental and engineering geology and related disciplines. Recent abstracts can be viewed under “Archive” at the web site, “http://eeg. geoscienceworld.org”. Articles that report on research, case histories and new methods, and book reviews are welcome. Discussion papers, which are critiques of printed articles and are technical in nature, may be published with replies from the original author(s). Discussion papers and replies should be concise. To submit a manuscript go to http://eeg.allentrack.net. If you have not used the system before, follow the link at the bottom of the page that says New users should register for an account. Choose your own login and password. Further instructions will be available upon logging into the system. Please carefully read the “Instructions for Authors”. Authors do not pay any charge for color figures that are essential to the manuscript. Manuscripts of fewer than 10 pages may be published as Technical Notes. For further information, you may contact Dr. Abdul Shakoor at the editorial office. Cover photo Construction of concrete slabs to prevent leakage from Melen Reservoir, Turkey” Photo courtesy of Ilyas Bekci (see article on page 79).

Environmental & Engineering Geoscience Volume 23, Number 2, May 2017 Table of Contents

Application of Multiple Criteria Decision Making Model for Evaluation of Levee Sustainability Stephen N. Semmens, Wendy Zhou, Bregje K. Van Wesenbeeck, and Paul M. Santi

Numerical Simulation of Uplift Pressure and Relief Drains at a Dam Reservoir Mehmet Ekmekci, Necati Erdem Kalaycioglu, Sukran Acikel, Otgonbayar Namkhai, and Salih Bilgin Akman

Comparing Rock Discontinuity Measurements Using Geological Compass, Smartphone Application, and Laser Scanning Methods Nicholas J. Farny

113

Water Level Monitoring to Assess the Effectiveness of Stormwater InďŹ ltration Trenches Laura Toran and Catherine Jedrzejczyk

125

A Durability ClassiďŹ cation of Clay-Bearing Rocks Based on Particle Size Distribution of Slaked Material Tej P. Gautam and Abdul Shakoor

137

Regional Landslide Inventory Mapping of Bashilo River Watershed, Ethiopia Elamin H. Ismail, J. David Rogers, Muhammed F. Ahmed, and Mohamed G. Abdelsalam

Application of Multiple Criteria Decision Making Model for Evaluation of Levee Sustainability STEPHEN N. SEMMENS WENDY ZHOU Department of Geology and Geological Engineering, Colorado School of Mines, 1516 Illinois Street, Golden, CO 80401, emails: ssemmens@mines.edu; wzhou@mines.edu

BREGJE K. VAN WESENBEECK1 Deltares, Marine and Coastal Systems, Ecosystem Analysis and Assessment, Rotterdamseweg 185, P.O. Box 177, 2600 MH Delft, The Netherlands, email: bregje.vanwesenbeeck@deltares.nl

PAUL M. SANTI Department of Geology and Geological Engineering, Colorado School of Mines, 1516 Illinois Street, Golden, CO 80401, email: psanti@mines.edu

Key Terms: GIS, MCDM, Logistic Regression, Logit Model, Environmental Factor, Land-use Planning ABSTRACT In this study, we seek to improve our ability to predict the sustainability of a levee by analyzing the character of the surrounding environment. Utilizing geographic information systems (GIS), approximately 140 mi (225 km) of levees within the Lower Mississippi River Valley were divided into small segments and assigned a series of environmental factors, including the configuration of Quaternary geology with respect to the levee alignment, the hydrogeological characteristics of the alluvial uifer beneath the levee, and soil physical properties. Next, a binary logistic regression was applied to evaluate the correlation between environmental factors and development of levee distress features (seepage lines and sand boils) to generate a predictive model. Results of the logistic regression were then fed into a multiple criteria decision making (MCDM) system to categorize environments into levee sustainability groups. Logistic regression results indicated significant correlation between levee distress features and four environmental characteristics: paleo-channel orientation, soil classification, normalized difference vegetation index (NDVI), and saturated hydraulic conductivity. The predictive model correctly predicted the status of distress feature development with up to 62 percent accuracy. The MCDM system identified forests of sweetgum, Nuttall oak, and willow oak as

1 Also at: Department of Hydraulic Engineering, Delft University of Technology, P.O. Box 5048, 2600 GA Delft, The Netherlands.

areas of elevated levee sustainability. Plots of sycamore, pecan, and American elm trees and water bodies were rated as decreasing levee lifetime. With additional development, future models may serve as tools to improve our ability to assess, maintain, and design levee systems in better coherence with their natural surroundings.

INTRODUCTION Levees are a vital part of civil infrastructure, providing flood protection and tidal protection. Improvement in ability to maintain, manage, and design levees prolongs the service life of the levee and reduces the potential for levee failure. For this study, the definition of a levee was adopted from the U.S. Army Corps of Engineers (USACE) Levee Owners Manual (USACE, 2006). The manual defines levees as structures, normally of earth or stone, generally built parallel to a river to protect land from flooding (USACE, 2006). In many cases, this infrastructure is built atop and surrounded by pervious foundations. During high-water events, the difference between the hydrostatic pressure in the flooding area (water side) and the protected area (land side) leads to seepage of water beneath the levee (underseepage). This resulting increase in hydraulic pressure head can initiate subsurface erosion and, potentially, failure of the levee. Two indicators of this internal erosion are the development of sand boils and seepage lines (Kolb, 1976). Sand boils are round, sometimes conical mounds of sand that develop around a seep when sand is carried to the ground surface on the landward side of a levee by seepage forces. These features indicate that active

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 65–77

Semmens, Zhou, van Wesenbeeck, and Santi

piping is occurring, and the safety of the nearby levee is threatened. Linear tensile fractures are also known to form during flood events as a result of hydraulic fracturing. As the pore pressure within saturated soils approaches the effective stress, cracks form in the soil. These cracks facilitate a sudden drop in head and increased exit velocity of water from the fractured soil, causing erosion and piping of the ground. When these linear erosional features manifest at the ground surface, they are referred to as “seepage lines.” While a limited amount of underseepage is generally accepted, excessive seepage along seepage lines is a cause for concern, as it may result in levee failure (Kolb, 1976). The development of levee distress features has previously been tied to several factors. During a critical review of levee underseepage controls by Wolff (2002) for the USACE, the presence of sand boils was highly correlated to specific local geologic features. These findings supported USACE observations concerning paleo-channel and levee interaction, which were noted as early as 1956 (Mansur and Kaufman, 1956). These paleo-channels (also referred to as channel-fill deposits) are composed of finer grains than their immediate surroundings and serve as barriers that focus and funnel groundwater flow through the subsurface (Mansur and Kaufman, 1956; Kolb, 1976; and Glynn et al., 2012). Wolff (2002) in p. 26 concluded by stating, “Although the local geology is identified as being of great importance in the development of underseepage problems, in practice it is incorporated into the analysis procedure only in a very indirect and judgmental manner and may often be overshadowed by the number-crunching aspects of the design.” In response to concerns over recurring piping events along a 16 mi (26 km) section of Mississippi River levees, Glynn et al. (2012) investigated the potential to predict piping events based on a series of geotechnical and geologic factors recorded for the areas surrounding the levee. The model was constructed using multivariate logistical regression (also referred to as a logit regression or logit model) to analyze a series of factors, including surficial geologic configuration, net head on the levee, vertical permeability of the riverside, and landside top blanket, among others. Two important conclusions were drawn as a result of this work. First, while the predictive model struggled to remove false positives, it showed promise by correctly identifying locations that would suffer from piping issues. Second, according to their model, the three most important factors for predicting piping events were unfavorable geologic configuration, thickness of the top stratum (blanket), and effective aquifer grain size (Glynn et al., 2012). Effective aquifer grain size refers to the D10 particle diameter, of which 10 percent of the material by weight is smaller (Dunn et al., 1980).

Land cover has also been connected to the performance of levees. Corcoran et al. (2010) attributed many deficiencies in levee performance to woody vegetation located adjacent to the levee toe. While some of this may be a result of vegetation impairing proper maintenance and inspection of levees, tree roots have been known to loosen soils and create seepage paths, particularly as they decay, leaving behind voids in the soil (Dise, 1996; USBR, 1989; and FEMA, 2005). Vegetation also provides a habitat for burrowing animals, which also create voids in nearby embankments and adjacent blanket material (Kolb, 1976; USBR, 1989; and Verachter et al., 2013). The purpose of the current study is to evaluate the relationship between environmental conditions surrounding a levee and the development of levee distress indicators, specifically sand boils and seepage lines. Current strategies for maintaining and designing levees do little to account for the interaction between the levee and the surrounding environment. While the immediate subsurface is addressed during design and construction, it is hypothesized that the character of the surrounding landscape, such as the presence of certain vegetation types, may also play a significant role in determining the development of features capable of degrading levee stability. Using a probabilistic statistical classification model and geographic information system (GIS) methodology, correlations between levee performance and a series of environmental factors are identified. Feedback from this initial analysis is then used to compare local ecological landforms with the observed and expected performance of an adjacent levee. These two analyses culminate in a multiple criteria decision making (MCDM) system that predicts the sustainability of an environment for levee construction and maintenance. STUDY AREA The study area chosen is composed of four counties adjacent to the Mississippi River: East Carrol and Madison Counties in Louisiana and Issaquena and Warren Counties in Mississippi (Figure 1). The region was chosen for a variety of reasons.

• The four counties contain over 140 mi (225 km) of levees running adjacent the Mississippi River. In order to perform the later statistical comparisons, it is necessary to have a signiﬁcant stretch of levees for analysis. • Most of the necessary environmental factor data are freely available to the public, including satellite imagery and soil data. • Data that are not open to the public could be obtained through other sources. This includes

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 65–77

Multiple Criteria Model Evaluation of Levee Sustainability

Figure 1. Map of Louisiana and Mississippi with the study area outlined in red.

information such as the location of levee distress features (i.e., sand boils and seepage lines) and certain maps that had to be digitized in ArcGIS (e.g., levee location, river banks, etc.). • The study area contains several paleo-channels in a variety of orientations. The relationship between paleo-channel orientation and the development of levee distress features is a key point of interest in this study. • A large ﬂood event in 2011 led to the development of many sand boils and seepage lines along the levees within this study area. The locations and extent of these features are necessary to evaluate the proposed methodology. The study area is located along the Lower Mississippi River. According to Kolb (1976), the region is categorized as a broad alluvial valley averaging 50 mi (80.5 km) in width. Along the river, the topography is very ﬂat and transitions to hills at its edges. Bedrock is located over 400 ft (122 m) beneath the ground surface and is overlain by glacial sediments deposited during the latter stages of the Wisconsin Glaciation. During this glacial period, meltwaters scoured the landscape to create the valley. Following the Last Glacial Maximum, glacial sands and gravels that remained in the valley were subsequently overlain by new sand and gravel deposits that steadily transitioned upward to sand alone over the course of the late Pleistocene and Holocene Epochs.

The initial appearance of clays and silts in the stratum did not occur until approximately 10,000 years ago, beginning at the southern end of the valley and progressing northward over time. During the same period, the Mississippi River transitioned from a shallow, braided stream to its more familiar modern, meandering counterpart. Since this time, the river has deposited several generations of meander belts, resulting in a highly variable top stratum sequence of mixed clays, silts, and sands. The meander belts range 3–9 mi (5–15 km) in width and consist of active and abandoned channels, point bar deposits, and natural levee deposits (Smith, 1996). Prior to the construction of artificial levees throughout the region, natural levees formed adjacent to the Mississippi River. These levees formed as the coarsest material carried by floodwaters fell out of suspension near the river banks during frequent flooding events. The most developed of these levees occur along the outside bends of meanders. Within the Lower Mississippi River Valley, these natural levees appear as low ridges approximately 10–15 ft (3–4.5 m) in height (Kolb, 1976). The average sedimentation rate for natural levee accretion has been calculated to be as high as 0.12 in./yr (0.3 cm/yr) (Farrell, 1987). Today, the natural levee deposits are complexly distributed across the floodplain as a result of the continued migration of the Mississippi River. It is common to find artificial levee systems or other human construction erected atop natural levee deposits due to their utility as a welldraining foundation material and topographical high (Smith, 1996). The rate of river discharge is distinctly seasonal, with the largest rates occurring in spring and late winter. Between 1799 and 1931, before the extensive modern system of levees and reservoirs was installed, the Lower Mississippi River flooded on average every 2.8 years. During this time, the average river discharge near Vicksburg, MS, at the southeastern end of the study area was estimated to be approximately 9,000,000 gal/s (34,160 m3/s) (Hudson et al., 2008). In 1717, European riparian land owners constructed the first series of levees in the Lower Mississippi Valley. These structures were typically 37 ft (12 m) in height. As more levees were constructed and the floodplain narrowed, levee heights gradually rose in order to contain the concentrated river flow (Smith and Winkley, 1996). Levee infrastructure continued to evolve as a patchwork system of independently managed levee sections through much of the eighteenth and nineteenth centuries. Comprehensive flood management of the Lower Mississippi was finally adopted following the largest historical flood in North America in 1927, when peak discharge rates exceeded 13000000 gal/s (49812 m3 /s) (Hudson et al., 2008).

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 65–77

Semmens, Zhou, van Wesenbeeck, and Santi Table 1. List of environmental factors examined during this study, their source, and their GIS features. Factor Floodplain width Paleo-channel orientation Normalized difference vegetation index (NDVI) AASHTO, percent clay, percent sand, percent silt, depth to water table, embankment material rating, saturated hydraulic conductivity, liquid limit, plastic index

Data

Source

GIS Feature

Aerial imagery Surﬁcial geology Landsat 8 imagery Soil properties, hydrogeology

ESRI ArcGIS USGS USGS EarthExplorer USDA Soil Survey

Feature class (line) Feature class (polygon) Raster Feature class (polygon)

METHODOLOGY Environmental Factor Preparation The ﬁrst step of the study was to select the levees of interest within the study area. The levee center lines (LCL) were digitized based on the ESRI World Imagery base map. The LCL were then buffered to the actual width of the levee and split so as to divide the feature into square segments approximately 500 ft (150 m) in length (Figure 2). Next, three buffer areas extending outwards from both the land and water sides of the levee were generated. The buffer distances were 250 ft (75 m), 500 ft (150 m), and 1000 ft (300 m). All three buffers were necessary for generating the paleo-

Figure 2. Map showing the LCL split into segments and buffered out to generate the yellow levee segment polygons in addition to the water- and land-side study regions (blue and green, respectively).

channel orientation ranking adopted from Glynn et al. (2012), but only the 1000 ft (300 m) buffer was used as the area of interest surrounding the levee for the remaining environmental factors. This allowed for three regions around the levees to be analyzed: the landside buffer zone, the water-side buffer zone, and the area combining both the land- and water-side buffer zones. Next, factors that may influence levee sustainability were identified for analysis (Table 1). American Association of State Highway and Transportation Officials (AASHTO) soil classification, liquid limit, plasticity index, and the percentages of clay, sand, and silt were included to investigate whether or not the soil composition and grain size were significant for the development of distress indicators. Hydrologic soil group classification and saturated hydraulic conductivity were included because they are both measures of water transmission: Materials that reduce the ability for water to move through the subsurface are generally assumed to prevent the development of sand boils and seepage lines. The embankment material rating was considered because it reflects the soil material’s ability to resist seepage, piping, and erosion in addition to having favorable compaction characteristics for the construction of fill in embankments, dikes, and levees. Depth to water table was included because it was hypothesized that regions with shallower water tables may be associated with shorter flow paths and higher seepage pressures compared to the surrounding region. The normalized vegetation difference index (NDVI) was also included in order to investigate the relationship between vegetation type and the development of levee distress features. The orientation of paleo-channels, which include channel-fill deposits and swales, was considered for two reasons. First, these paleo-channels are composed of finer grains than their immediate surroundings and serve as barriers that focus underseepage through the subsurface (Mansur and Kaufman, 1956; Kolb, 1976; Glynn et al., 2012). Second, the investigation by Glynn et al. (2012) concluded that their classification system (Table 2) for paleo-channel orientation was successful in predicting piping events. Applying their system a second time serves as a way to test their findings and

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 65–77

Multiple Criteria Model Evaluation of Levee Sustainability Table 2. Channel orientation ranking system adopted from Glynn et al. (2012). Ranking 0 0.5 0.6 0.7 0.8 0.9 1.0

Description No unfavorable conﬁguration Intersects the levee at an angle >90◦ Exists parallel to the levee between 500 and 1000 ft (152.4 and 304.8 m) on the land side of the levee toe Intersects perpendicular with the levee Exists parallel to the levee between 250 and 500 ft (76.2 and 152.4 m) on the land side of the levee toe Exists parallel to the levee within 250 ft (76.2 m) of the levee toe with no overlap Intersects the levee at an angle <90◦

further investigate the interaction between paleochannels and levees. Once the factors of interest were identified, the data were gathered and assigned to their respective levee segments. Floodplain width was calculated using GIS as the shortest distance between the levee segment and the nearest riverbank. Paleo-channel orientation was calculated according to the Glynn et al. (2012) system (Table 2 and Figure 3). The numerical values were built upon previous observations made by the USACE (1956) and Kolb (1976) concerning the orientation of these unfavorable geologic features and their influence on piping. The final rank was assigned manually to the levee segments by comparing the relative orientations of channel-fill deposits derived from surficial geologic maps of the region (Saucier, 1994) and LCL. Both the floodplain width and paleo-channel orientation factors are independent of the three buffer zones (land side, water side, and combined), so only one representative value for each factor was assigned to the respective segment. Next, NDVI was generated using Landsat 8 imagery acquired from the U.S. Geological Survey (USGS)

EarthExplorer online portal (USGS Earth Resources Observation and Science Center, 2014) (Keranen and Kolvoord, 2014). Two images taken a year apart were merged to cover the full extent of the study area. The images were both taken during the month of July when vegetation coverage was peaking and NDVI should be most effective in differentiating between land cover. The year gap was necessary to avoid signiﬁcant cloud coverage, which can generate erroneous data during the NDVI calculation process. The NDVI calculation and ﬁnal map were generated using ESRI ArcGIS (2015). A second map converting the NDVI values to categorical values representing four different land-cover types (water, barren land, light vegetation, and heavy vegetation) was also created. The mean value for both the original and categorical NDVI was then calculated for each of the levee buffer zones (1000 ft [300 m] from the land-side toe, water-side toe, and combined landand water-side) and assigned to the respective levee segment. Data from the U.S. Department of Agriculture (USDA) Soil Survey were acquired as a series of soil survey geographic (SSURGO) databases. For each

Figure 3. Schematic illustrating how the Glynn ranking system is applied to paleo-channel deposits (gray lines).

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 65–77

Semmens, Zhou, van Wesenbeeck, and Santi

factor of interest derived from these data, a single map was created displaying the factor values either at the ground surface or measured from the ground surface to the limit of the survey’s depth (approximately 6 ft [2 m]). This study assumes that the character of the top strata as described by the SSURGO data set is worthy of consideration. Attempts to use the data for calculating and incorporating the thicknesses of top strata were dismissed due to the severe limitations of the survey depth. Using ArcGIS, the median value for each factor was calculated for each levee buffer zone (1000 ft [305 m] from the land-side toe, water-side toe, and combined land- and water-side) and assigned to the respective levee segment. LEVEE DISTRESS FEATURE PREPARATION Underseepage data from the regional flood in 2011 were provided by the USACE (Dunbar et al., 2015) as a series of maps. The maps were georeferenced and the distress feature data digitized using GIS. Due to the fact that individual sand boils and seepage lines were not definitively associated with a particular levee, series of density maps were generated. These maps were used to identify the segments with which the features were most likely associated and to assign a binary (present/absent) distress feature rating to the segment (Figure 4). ENVIRONMENTAL LANDFORM PREPARATION Three environmental landforms were used in this study to construct the MCDM system: wetlands type, forest type, and ecophysiography type. Wetlands data were acquired from the U.S. Fish and Wildlife Service National Wetlands Inventory (2016). According to the metadata, this information represents the extent, approximate location, and type of wetlands and deepwater habitats as defined by Cowardin et al. (1995). Forest data were retrieved from the USDA Forest Service (2016), using the National Forest Type Dataset for the contiguous United States. The information was organized as a raster layer created using the USDA Forest Service Forest Inventory and Analysis (FIA) program and the Remote Sensing Applications Center (RSAC) in order to display the extent, distribution, and types of forest across the United States. World ecophysiographic land unit data were acquired from the ESRI ArcGIS REST Services Directory. The information was generated by a partnership between the USGS and ESRI in 2015 by combining the most accurate, up-to-date, and finest resolution data for bioclimates, landforms, lithology, and land cover. For each environ-

Figure 4. Map of a portion of the study area showing distress features (turquoise) and the corresponding halo region generated by the kernel density tool (red) used to assign distress features to levee segments.

mental type, GIS was used to identify the dominant attribute in the combined land-and water-side 1000 ft (300 m) buffer region and assign it to the respective levee segment.

APPLICATION OF PROBABILISTIC STATISTICAL CLASSIFICATION MODEL With all the data acquired and organized, the full data set was bootstrapped to generate 10 sample data sets. These sample data sets are composed of a random selection of 80 percent of the levee segments. The binary logistic regression was then applied to the sample data sets. The results were compiled, and the p-values for each factor were compared. Factors with a median p-value ≤ 0.05 calculated across all samples were considered to be significant. A principal components analysis was then performed to identify redundant variables within the significant factors. The binary logistic regression was then performed a second time to generate the final predictive model utilizing the unique significant factors.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 65–77

Multiple Criteria Model Evaluation of Levee Sustainability

DEVELOPMENT OF MCDM SYSTEM The MCDM was constructed to divide environments into three groups based on their perceived effect on levee sustainability. For this study, levee sustainability was measured by the development of levee distress events. Areas with few distress features were considered to be more sustainable because they require less attention and fewer resources to be maintained over the course of their lifetimes. Regions where sand boils and seepage lines develop at greater frequency were considered to be less sustainable due to the effort and resources necessary to maintain the levee. A multivariate cluster analysis was utilized to divide the environments into sustainability groups by comparing patterns in the observed and predicted appearance of distress events within the different environments. The cluster analysis was performed according to the complete linkage method and Euclidean distance measure so as to generate three clusters. The appropriate sustainability rating (high, moderate, and low) was then assigned to the appropriate cluster. RESULTS AND DISCUSSION Three observations were made while preparing data for the logistic regression. First, although the NDVI analysis was effective at dividing land cover within the study area into appropriate land-cover categories, it did not consistently identify standing water. The most common error was identifying sections of oxbow lakes as lightly vegetated land, probably due to dense growths of algae or other plant matter within the standing water. This error was almost nonexistent in areas of moving water, such as in the Mississippi River itself. Thus, NDVI values for segments immediately adjacent to oxbow lakes may be artificially high. Second, the NDVI classification divided cropland into multiple categories. Land with healthy, growing crops received a higher NDVI value and was categorized as heavily vegetated. Land left fallow generated lower NDVI values and was typically categorized as lightly vegetated and less often as barren land. Third, seepage lines tended to form near the land-side toe of the levee. The location of sand boils would vary from immediately adjacent to the land side of the levee, adjacent the toe, to several hundred feet from the levee structure. Additionally, the appearance of sand boils did not always coincide with the development of seepage lines nearby or vice versa. Results from the first iteration of the binary logistic regression indicated that the most important factors for predicting the formation of distress features were NDVI (water side), NDVI categorical value (land side), AASHTO (land side, water side, and combined), saturated hydraulic conductivity (combined), and paleo-

channel orientation rating. The principal component analysis indicated that the AASHTO land-side and combined land- and water-side factors were collinear and likely redundant. To determine if and which of the two collinear factors should be removed, three additional logistic regressions were performed: one with the AASHTO land-side factor removed, one with the AASHTO combined land- and water-side factor removed, and one keeping both collinear factors. These regressions also differed from the first logistic regression because they were performed using only the significant factors identified as a result of the first logistic regression. Series of confusion matrices were then generated in order to compare the performance of each model. After comparing the results from the confusion matrices, the model with AASHTO (combined) removed was selected as the final logistic regression model for three reasons. First, it produced consistently lower false negative prediction rates compared to the other two models. Second, despite having a lower false negative rate, the model maintained a maximum accuracy essentially equal to the other models. Third, the model correctly predicted the true presence and lack of distress features at greater than 60 percent accuracy and reduced both false prediction rates to below 40 percent. The accuracy of this model also indicates that the status of distress features can be predicted with moderate success. The following equations represent the final series of predictive equations.

P (1) =

1 + eY

(1)

Y’ = 0.325 + 0.877 (NDV IW) − 0.01940 (Ksat ) (2) +AW + AL + NDV IC + OGeo AW = 1.436 (AW1 ) − 0.597 (AW2 ) − 0.169 (AW3 ) −0.418 (AW4 ) − 0.647 (AW5 ) − 0.223(AW6 ) (3) AL = 2.173 (AL1 ) + 0.041 (AL2 ) − 0.046 (AL3 ) −0.483 (AL4 ) + 1.322 (AL5 ) − 0.230(AL6 ) (4)

NDV IC = 0 (NDV IC1 ) + 0.950(NDV IC2 ) −0.432(NDV IC3 ) − 0.168(NDV IC4 ) (5)

OGeo = 0 (O0 ) + 0.208 (O5 ) − 0.465 (O6 ) − 0.911 (O7 ) (6) −0.166 (O8 ) + 0.722 (O9 ) − 0.912(O10 )

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 65–77

Semmens, Zhou, van Wesenbeeck, and Santi

Figure 5. Histogram of probabilities calculated using the logistic regression predictive equations. The black line at 14 percent is the cutoff percentage.

where

• P Probability of a distress feature developing along the levee segment. • NDVIW Value of the water-side NDVI. • Ksat Representative saturated hydraulic conductivity at the ground surface for the combined water- and land-side buffer areas. • A W1–6 Categorical value for the AASHTO rating in the water-side buffer area. • NDVIC1–4 Categorical value for NDVI for the landside buffer area. • AL1–6 Categorical value for the AASHTO rating in the land-side buffer area. • O0,5–10 Categorical value for the paleo-channel orientation rating. Using Eq. 1–6, the ﬁnal probabilities for distress features along each levee segment were calculated. Figure 5 shows the distribution of probabilities as a histogram. The black line on the graph is the cutoff percentage denoting the boundary between predictions. Segments with a predicted probability exceeding the cutoff percentage are expected to develop a distress feature, while probabilities below the cutoff are not. Series of relationships between factor values and their effect on predicted probability of distress features can be pulled from the ﬁnal logistic regression model (Table 3). First, for both continuous variables (water-side NDVI and combined land- and water-side saturated hydraulic conductivity), increasing the value results in an increase in the probability of a distress feature. However, the probability is more sensitive to

increases in the saturated hydraulic conductivity value than the water-side NDVI. The effect of categorical land-side NDVI values is almost opposite of the continuous water-side NDVI effect. On the land side, both vegetation types reduce the probability of a distress feature developing, while barren land sharply increases the probability. It should be noted that heavily vegetated regions are more difficult to survey for distress features due to the visual obstruction from the plants. Distress features are more readily observed on barren and lightly vegetated land because the features are not masked by natural obstructions. In general, the presence of soils on the water side of the levee with AASHTO classifications showing more fines reduces the probability of distress features (AASHTO, 1982). On both the land and water sides, the presence of A-2-4 (silty/clayey gravel and sand) increases the probability of distress features. The effect of fine material on the land side is more mixed. A-4 (silty soil) and A-6 (low-plasticity clay) are not good predictors of distress features, and A-7-5 (high-plasticity clay) is strongly correlated with increases in the probability of a distress feature. It is unclear why the clay classified as A-7-5 has the opposite effect on the predicted probability as the other two high-plasticity clays for the land-side AASHTO classification. This phenomenon may occur because the permeability of the high-plasticity clay is low enough to allow pore pressures beneath the clay to build instead of bleeding off, as in the case of low-plasticity clay and silty topsoil, leading to the development of seepage lines and sand boils. For each case, the AASHTO rating can be traced back to over 120 levee segments, so it is unlikely that

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 65–77

Multiple Criteria Model Evaluation of Levee Sustainability Table 3. List of signiﬁcant predictors and their effect on the probability generated by the logistic regression model.

Factor

Categorical Variable

Water-side AASHTO

AW1 AW2 AW3 AW4 AW5 AW6

Land-side AASHTO

AL1 AL2 AL3 AL4 AL5 AL6

Paleo-channel Orientation

Land-side NDVI

Water-side NDVI Combined-sides Saturated Hydraulic Conductivity

O0 O5 O6 O7 O8 O9 O10 NDVIC1 NDVIC2 NDVIC3 NDVIC4 NDVIW

Categorical Description A-2-4, silty/clayey gravel and sand A-4, silty soil A-6, low-plasticity clay A-7, high-plasticity clay A-7-5, high-plasticity clay A-7-6, high-plasticity clay A-2-4, silty/clayey gravel and sand A-4, silty soil A-6, low-plasticity clay A-7, high-plasticity clay A-7-5, high-plasticity clay A-7-6, high-plasticity clay 0 0.5 0.6 0.7 0.8 0.9 1 Water Barren land Lightly vegetated Heavily vegetated

Ksat

the discrepancy can be attributed to a small sample size (Table 3). Finally, the effects of individual paleo-channel orientation ratings on the predictive model vary considerably. According to the system adopted by Glynn et al. (2012) (Table 2), an increase in the value of the orientation ranking should correlate with a higher likelihood of distress features developing. This trend does not hold true for the logistic model in this study. Only ranks 0.5 (channel intersects the levee at an angle greater than 90 degrees) and 0.9 (channel exists parallel to and within 250 ft [76 m] of the land-side levee toe) increase the probability of distress features. The remaining ratings either reduce the probability or have little effect. Some of this variability between rankings may be a result of complexities within the subsurface that were not considered within the ranking system or the logistic regression, including the inclined stratigraphy of the paleo-channel deposits and the na-

Effect on Predicted Probability Moderately increases

Sample Size (No. of Segments) 43

Sharply reduces Moderately reduces Sharply reduces

246 106 25

Sharply reduces

272

Sharply reduces

398

Sharply increases

No signiﬁcant effect No signiﬁcant effect Sharply reduces

239 506 126

Sharply increases

158

Sharply reduces

255

No signiﬁcant effect Moderately increases Sharply reduces Sharply reduces Sharply reduces Sharply increases Sharply reduces No signiﬁcant effect Sharply increases Sharply reduces Moderately reduces Increases with factor value Increases with factor value

914 193 68 82 25 12 209 8 59 175 1,261 1,503 1,503

ture of the truncation between intersecting channel deposits. To begin construction of the MCDM, the percentage of predicted and observed distress features was calculated as a function of their segment’s environment type. Some environment types contained fewer than 32 segments and were considered to have a sample size too small to provide an accurate statistical illustration of the relationship between the environment and development of levee distress features. Thus, these environments were omitted from the MCDM system. Following the logistic regression, a Minitab cluster analysis was performed on the observed and predicted distress feature event percentages. Figure 6 shows the resulting dendrogram used to group similarly performing environment types into their MCDM sustainability groups. The dendrogram shows that there was very little difference in the observed and predicted occurrence of distress features for most environment types.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 65–77

Semmens, Zhou, van Wesenbeeck, and Santi

Figure 6. Dendrogram splitting environment types into three groups based on the observed and predicted event percentage for levee segments. The dendrogram was generated using a complete linkage, Euclidean distance cluster analysis in Minitab. Environment numbers are deﬁned in Table 4.

Only Forest_6, Forest_7, and Ecophysiography_17 differed significantly from the majority. Forest_6 had significantly fewer observed and predicted distress events compared to the other environment types. Forest_7 and Ecophysiography_17 had significantly more observed and predicted distress events compared to the other environment types (Table 4). An interesting note is the number of segments with which each water body group was associated. The ecophysiography water body designation (environment number 14 in Figure 6) was calculated as the dominant condition along 58 segments, a relatively small sample size when compared to the 119 and 163 segments for freshwater ponds (number 3) and lakes (number 4), respectively. This indicates a large disparity between the two sources in their labeling of water bodies. Looking back at the GIS data for each layer, the wetlands data (numbers 1–4) have a much higher resolution than the ecophysiography data and would be expected to generate more accurate results. As a result, the low sustainability rating for regions associated with the ecophysiography water body type is likely not as representative of the relationship with distress feature development as are medium ratings derived from the wetlands type water body data (Table 4).

The low sustainability rating for sycamore, pecan, and American elm forest (Forest_7) was another interesting result. Of all the environment types, this subset of forest showed the strongest positive correlation with distress feature development. The reasoning behind this relationship is not clear, but two possibilities are worth considering. First, the forest may grow best in the conditions that often lead to seepage lines and sand boils (such as sandy soils). Second, the trees may be more directly assisting the development of seepage pathways through the subsurface (such as providing habitats for burrowing fauna). Additional research will be necessary to determine which, if either, of these hypotheses are correct. On the other end of the spectrum, sweetgum, Nuttall oak, and willow oak forest (Forest_6) were rated as the only highly sustainable environment type (Table 4). From the observed data, over 70 percent of segments surrounded by this forest type did not have seepage lines or sand boils develop in the vicinity. The forest type was also associated with a strong sample size of 134 segments. Once again, the reasoning behind this relationship is not known, but it may be due to the forest being an indicator for local conditions less suitable for the formation of sand boils and seepage lines, or due to

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 65–77

Multiple Criteria Model Evaluation of Levee Sustainability Table 4. Table showing the environment types, their respective number displayed on the dendrogram, the sustainability rating, and the number of segments with which each type is associated. Dendrogram Environment No.

Environment Type

Sustainability Rating

1 2 3 4 5 6 7 8

Wetlands_1 Wetlands_2 Wetlands_3 Wetlands_4 Forest_5 Forest_6 Forest_7 Ecophysiography_1

Medium Medium Medium Medium Medium High Low Medium

Ecophysiography_2

Medium

Ecophysiography_4

Medium

Ecophysiography_5

Medium

Ecophysiography_6

Medium

Ecophysiography_9

Medium

Ecophysiography_17

Low

a more direct relationship between the trees and their interference of seepage pathways. The remaining environment types fell into the medium sustainability rating (Table 4). These environment types did not show a particularly strong trend towards events or nonevents. The two exceptions were for freshwater emergent wetlands and lakes. However, this result occurred using the predicted probability data and did not hold up when compared to the observed event data.

CONCLUSION This research demonstrates that certain soil, geologic, and land-cover characteristics appear to correlate with the development of sand boils and seepage lines. Signiﬁcant factors for predicting the development of these distress features were:

• NDVI rating on the water- ide of the levee: Probability of distress feature increases with increasing NDVI value.

Description Freshwater emergent wetland Freshwater forested/shrub wetland Freshwater pond Lake Sugarberry/hackberry/elm/green A Sweetgum/Nuttall oak/willow oak Sycamore/pecan/American elm Hot wet flat plains unconsolidated sediment closed (>40 percent) broadleaved deciduous forest (>5 m) Hot wet flat plains unconsolidated sediment closed (>40 percent) needle-leaved evergreen forest (>5 m) Hot wet flat plains unconsolidated sediment closed to open (>15 percent) herbaceous vegetation (grassland, savannas, or lichens/mosses) Hot wet flat plains unconsolidated sediment closed to open (>15 percent) mixed broadleaved and needle-leaved forest (>5 m) Hot wet flat plains unconsolidated sediment mosaic cropland (50–70 percent)/vegetation (grassland/shrub land/forest) (20–50 percent) Hot wet flat plains unconsolidated sediment mosaic vegetation (grassland/shrub land/forest) (50–70 percent)/cropland (20–50 percent) Water body

Sample Size (No. of Segments) 34 708 119 163 413 134 48 225

569

200

142

125

• NDVI categorical rating on the land side of the levee: Probability of distress features is reduced in the presence of vegetated land, increases sharply in the presence of barren land, and has no significant effect if covered by a water feature. • AASHTO soil classification rating of the soil surrounding the levee: Probability of distress features is generally reduced with the presence of silty and clayey soil but increases with increased sand and gravel composition. • Saturated hydraulic conductivity of the soil surrounding the levee” Probability of distress feature increases with increasing saturated hydraulic conductivity. • Paleo-channel orientation rating of the levee segment: Probability of distress features is reduced for orientation rankings of 0.6, 0.7, 0.8, and 1 but is increased for rankings of 0.5 and 0.9. The effects of water-side AASHTO soil classification and combined-sides hydraulic saturated conductivity agree with current observations concerning local conditions and the development of distress features. The

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 65–77

Semmens, Zhou, van Wesenbeeck, and Santi

effects of land-side AASHTO soil classification values on predicted probability potentially indicate a more complex relationship than previously considered. The results also support the previous finding by Glynn et al. (2012), i.e., that unfavorable geologic orientation is a significant factor for identifying the location of sand boils. This result is particularly noteworthy because the study by Glynn et al. (2012) was performed in a different geologic setting (Middle Mississippi River Valley versus the Lower Mississippi River Valley). However, examining the effect of each orientation category on the predicted probability indicates that Glynn’s rankings of 0.5 and 0.9 are the only orientations that significantly increase the probability of a distress feature for this study. With only 62 percent prediction accuracy, the logistic model shows promise, but it is not a reliable, independent tool for predicting the development of sand boils and seepage lines. Given this moderate success, it is likely that a more accurate prediction method would require additional elements beyond the significant factors identified within this study. Additional factors of importance may include the location of distress features from flood events prior to those investigated in this study, the delineation of old and new growth forests, and the location of human features, such as pipes, that transect the levee foundation. Following the logistic regression, a cluster analysis (Figure 6) was performed to organize wetland, forest, and ecophysiography environments into sustainability groups. The criteria used for construction of the cluster analysis were the percentage of observed distress features and percentage of predicted distress features within each environment. The MCDM was constructed from a series of relationships among the spatial distribution of distress features, the predicted distribution of distress features generated by the logistic model, and the distribution of a variety of local environments. The final model was formed by examining the correlations between these three pieces of information. This analysis indicated that forests composed of sweetgum, Nuttall oak, and willow oak correlated with significantly fewer distress features than the other environments. Thus, this forest type was rated as being highly sustainable for levee construction and maintenance. Forest plots composed of sycamore, pecan, and American elm trees and water bodies identified by the ecophysiography data correlated with significantly more distress features than the other environments. Due to the significant correlation with the appearance of levee distress features, these two environments were rated as having low sustainability for the construction and maintenance of levees. The remaining environments were given a moderate rating. Further investigation is necessary before anything can be determined about whether environment type is ac-

tively affecting the ground conditions or is merely an indicator of settings that lead to distress feature development. Regardless, the MCDM system is the ﬁrst to correlate environment types to the development of sand boils and seepage lines. While these results are promising, the models generated over the course of this study are not considered to be robust enough to address the challenges to current levee systems. Given further development and an expansion of our knowledge of levee-environment interactions, future models may become useful tools capable of improving our ability to assess, maintain, and design levee systems. ACKNOWLEDGMENTS The authors would like to thank the National Science Foundation for funding this research (Award Number: 1243539). We would also like to thank Joe Dunbar at the U.S. Army Corps of Engineers, Engineer Research and Development Center, for providing valuable levee distress feature data and comments. Data for this research project were acquired from the Natural Resources Conservation Service (NRCS) Soil Survey, the U.S. Geological Survey EarthExplorer, U.S. Fish and Wildlife Service National Wetlands Inventory, U.S. Department of Agriculture (USDA) Forest Service National Forest Type Dataset, United States Forest Service Forest Inventory and Analysis (FIA) and Remote Sensing Applications Center (RSAC), and the ESRI ArcGIS REST Services Directory. The data are available for free and can be obtained by contacting the above agencies or institutions.

REFERENCES AMERICAN ASSOCIATION OF STATE HIGHWAY AND TRANSPORTATION OFFICIALS, 1982, AASHTO Materials, Part 1, Speciﬁcations, Washington, D.C. CORCORAN, M. K.; GRAY, D. H.; BIEDENHARN, D. S.; LITTLE, C. D.; LEECH, J. R.; PINKARD, F.; BAILEY, P.; AND LEE, L. T., 2010, Water Resources Infrastructure: Literature Review—Vegetation on Levees: U.S. Army Corps of Engineers: Engineer Research and Development Center, Washington, DC. COWARDIN, LEWIS, M. AND GOLET, Francis C., 1995, U.S. Fish and Wildlife Science 1979 Wetland Classiﬁcation: A review, Vegetatio, Vol. 118.1-2, pp. 139–152. DISE, K., 1996, Memoradum to Area Manager, Mills, WY. Tree Removal, Pilot Butte Dam, Pick-Sloan, Missouri Basin Program, Wyoming: unpublished report. DUNBAR, J. B.; ENSIGN, A. L.; AND CORCORAN, M. K., 2015, Occurrence of Seepage and Sand Boils in the Vicksburg District, Mississippi River 2011 Flood, Volume III: Photo Maps: U.S. Army Corps of Engineers, Engineer Research and Development Center, Vicksburg, MS. DUNN, I. S.; ANDERSON, L. R.; AND KIEFER, F. W., 1980, Fundamentals of Geotechnical Analysis: Wiley and Sons, New York.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 65–77

Multiple Criteria Model Evaluation of Levee Sustainability ESRI ARCGIS, 2015, World Ecophysiographic Land Units 2015: Electronic document [accessed July 3, 2016], available at http://www.arcgis.com/home/item.html?id= 140af3e5389a4afcb421ee4633d18d3a FARRELL, K. M., 1987, Sedimentology and facies architecture of overbank deposits of the Mississippi River, False River region, Louisiana. In Ethridge, F. G.; Flores, R. M. ; and Harvey, M. D. (Editors), Recent Developments in Fluvial Sedimentology: Special Publication 39, Society of Economic Paleontologists and Mineralogists (SEPM), pp. 111–120. FEDERAL EMERGENCY MANAGEMENT AGENCY (FEMA), 2005, FEMA L-263: Dam Owner’s Guide to Plant Impact on Earthen Dams: FEMA, Washington, DC. GLYNN, E.; QUINN, M.; AND KUSZMAUL, J., 2012, Predicting Piping Potential Along Middle Mississippi River Levees: ICSE6, Paris, France. HUDSON, P. F.; MIDDELKOOP, H.; AND STOUTHAMER, E., 2008, Flood management along the Lower Mississippi and Rhine Rivers (The Netherlands) and the continuum of geomorphic adjustment: Geomorphology, Vol. 101, pp. 209–236. KERANEN, K. AND KOLVOORD, R., 2014, Module 10: Normalized difference vegetation index. In Making Spatial Decisions Using GIS and Remote Sensing: A Workbook: ESRI Press, Redlands, CA, pp. 210–222. KOLB, C. R., 1976, Geologic control of sand boils along Mississippi River levees. In Coates, D. R. (Editor), Geomorphology and Engineering: Dowden, Hutchinson and Ross Inc., Vicksburg, MS, pp. 99–113. MANSUR, C. I. AND KAUFMAN, R. I., 1956, Underseepage, Mississippi River levees, St. Louis District: Journal of the Soil Mechanics and Foundations Division, Vol. 82, No. 1, pp. 385–406. SAUCIER, R. T., 1994, Geomorphology and Quaternary Geologic History of the Lower Mississippi Valley: U.S. Army Engineer Waterways Experiment Station, Mississippi River Commission, Vicksburg, MS.

SMITH, L. M., 1996, Fluvial geomorphic features of the Lower Mississippi alluvial valley: Engineering Geology, Vol. 45, pp. 139–165. SMITH, L. M. AND WINKLEY, B. R., 1996, The response of the Lower Mississippi River to river engineering: Engineering Geology, Vol. 45, pp. 433–455. U.S. ARMY CROPS OF ENGINEERS (USACE), 1956, Investigation of Underseepage and Its Control: Technical Memorandum TM 3-424. U.S. ARMY CORPS OF ENGINEERS (USACE), 2006, Levee Owner’s Manual for Non-Federal Flood Control Works: USACE Rehabilitation and Inspection Program. U.S. BUREAU OF RECLAMATION (USBR), 1989, Water Operation and Maintenance Bulletin No. 150. USBR, Facilities Engineering Branch, Engineering Division, Denver, CO. U.S. DEPARTMENT OF AGRICULTURE (USDA) FOREST SERVICE, 2016, National Forest Type Dataset: Electronic document [accessed July 3, 2016], available at http://data.fs.usda.gov/ geodata/rastergateway/forest_type/index.php U.S. FISH AND WILDLIFE SERVICE (USFWS) NATIONAL WETLANDS INVENTORY, 2016, The National Wetlands Inventory: Electronic document [accessed July 3, 2016], available at http://www.fws.gov/wetlands/NWI/index.html U.S. GEOLOGICAL SURVEY (USGS) EARTH RESOURCES OBSERVATION AND SCIENCE CENTER, 2014, Landsat 8 OLI (Operational Land Imager) and TIRS (Thermal Infrared Sensor): USGS, Sioux Falls, SD. VERACHTER, E., VAN DEN EECKHAUT, M., MART´INEZ-MURILLO, J. F., NADAL-ROMERO, E., POESEN, J., DEVOLDERE, S., WIJNAUTS, N.; AND DECKERS, J. , 2013, Impact of soil characteristics and land use on pipe erosion in a temperature humid climate: Field studies in Belgium: Geomorphology, Vol. 192, pp. 1–14. WOLFF, T. F., 2002, Performance of Levee Underseepage Controls: A Critical Review: USACE Engineer Research and Development Center, Vicksburg, MS.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 65–77

Numerical Simulation of Uplift Pressure and Relief Drains at a Dam Reservoir MEHMET EKMEKCI1 Hydrogeological Engineering Program, Hacettepe University, Beytepe 06800 Ankara, Turkey

NECATI ERDEM KALAYCIOGLU Hydrogeological Engineering Program, Hacettepe University, Beytepe 06800 Ankara, Turkey, email: necatierdemkalaycioglu@gmail.com

SUKRAN ACIKEL Hydrogeological Engineering Program, Hacettepe University, Beytepe 06800 Ankara, Turkey, email: sukransahbudak@yahoo.com

OTGONBAYAR NAMKHAI Hydrogeological Engineering Program, Hacettepe University, Beytepe 06800 Ankara, Turkey, email: obayar1098@gmail.com

SALIH BILGIN AKMAN ES Proje, Yıldızevler 4. Cadde 712. Sokak No:6/8 C ¸ ankaya Ankara, Turkey, email: sbakman@esproje.com

Key Terms: Concrete Lining, Hydrogeology, Karst, Melen Dam, Numerical Simulation, Relief Drain, Uplift Pressure ABSTRACT The Melen Dam is under construction to support the water supply to the Istanbul Metropolitan area. Karstiﬁed limestone, which forms the major aquifer in the region, crops out at the left bank of the reservoir and extends to the adjacent basin, where it is drained by two main springs. An area of 20 ha that is covered by this limestone is planned to be lined by concrete mass cover. Another spring that discharges within this lining area will be plugged by the concrete mass. As a consequence of uplift pressure, piezometric pressure is expected to build up underneath the concrete lining. Prediction and relief of the uplift pressure are of utmost importance in optimizing the concrete lining. This was achieved by performing a numerical simulation of the uplift pressure of the pre-lining and post-lining conditions at the site. This required development of a representative hydrogeological conceptual model as the very ﬁrst stage of this study. The conceptual model was then transferred to a numerical model through the software Seep/w of GeoStudio (2007 1 Corresponding

author email: ekmekci@hacettepe.edu.tr

version). The optimal design of the concrete lining was developed after performing a number of numerical simulations on the calibrated model. The model suggested that the uplift pressure would be relieved to an acceptable level by construction of drains underneath the concrete slabs at certain locations.

INTRODUCTION Statistics show that about 54 percent of the world’s population live in cities, and this figure is projected to reach 66 percent by 2050 (UNDESA, 2014). Access to sufficient and high-quality water is one of the major problems that authorities face in providing a water supply to this massive concentration of population in cities and metropolitan areas. Istanbul, a metropolitan city with a population of about 18 million in northwest Turkey, suffers from a shortage of water. This is a severe problem that threatens the near future of the city because of the ever-increasing population due to immigration and expanding hinterland. The Metropolitan Municipality, the authority responsible for supplying sufficient and healthy water to the inhabitants, has decided to tap the water of the Buyuk Melen River that discharges to the Black Sea, about 180 km to the east of the Istanbul city center. The project is known as the “Melen Dam Project” and includes

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Ekmekci, Kalaycioglu, Acikel, Namkhai, and Akman

Figure 1. Location and simpliﬁed geological map of the study area.

construction of a dam and a 105-km-long double-tube pipeline to convey the water to Istanbul at a rate of approximately 34 m3 /s. Location of the study area is depicted in the simplified geological map of the area shown in Figure 1. Since new dam sites are needed to support the existing but insufficient water sources geologically and/or hydrogeologically less favorable sites, such as karstic terrain, must be considered. This situation often causes geotechnical and/or hydrogeological challenges. A major challenge encountered during construction of dams at such sites is water seepage, either through the dam foundation or from the reservoir area (Bruce, 2003; Knight et al., 2010; Milanovic, 2011; and Mozafari and Reisi, 2015). Rendering the unfavorable feasible requires the design of efficient engineering structures, such as construction of grout curtain, compacted earthen lining, concrete-slab cover, and concrete lining, to mention just some of the examples provided in the research (Gebhart, 1973; Guifarro et al., 1996;

Milanovic, 2004; Fazeli, 2007; Bruce et al., 2010; Milanovic, 2011; and Roman et al., 2013). Uplift pressure is among the major factors contributing to failures of dams (Hanna et al., 1993; USACE, 1993; Martt et al., 2005; Ozkan et al., 2008; Utili et al., 2008; and Chen et al., 2010). The success and efﬁciency of these structures depend on representativeness of the conceptual model, the method of analysis, and the quantitative and qualitative adequacy of data needed in the analyses (Anderson and Woessner, 1992; Kresic, 2006; and Geo-Studio, 2009). Numerical analysis is becoming a common practice in the design of engineering structures to solve water leakage problems (Chen et al., 2010; Lu et al., 2013; and Youseﬁ et al., 2013). It is a preferred method because it provides a powerful tool with which to perform “numerical experiments” using a computer medium. However, such a practice strictly requires a sound and representative conceptualization of the hydrogeological setting and accurate characterization of the

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Simulation of Uplift Pressure

Figure 2. Schematic cross section illustrating the hydrogeological setting across adjacent basins.

lithostratigraphic units (Anderson and Woessner, 1992; ASTM, 1996, 1997; and Kresic, 2006). This study demonstrates the use and efﬁciency of numerical modeling applied in the effective design of a concrete lining and associated relief drains to prevent the expected water leakage from the reservoir of the Melen Dam in Turkey.

STATEMENT OF THE PROBLEM The Melen Dam is being constructed on the Buyuk Melen River, where the thalweg elevation is 25 m above sea level (a.s.l.). The long-term mean annual river flow at the dam site is 47.8 m3 /s. The height of the crest of the dam is 99 m a.s.l. from the thalweg, creating a reservoir volume of 693.54 hm3 with a water elevation of 110 m a.s.l. (ES Proje, 2011). The dam site and the major part of the reservoir area are covered by a lowpermeability sedimentary unit that is composed of an alternation of medium-thick layers of marine claystone and thin to medium-thick sandstones described as flysch. An older carbonate rock unit has been thrusted over the low-permeability flysch within the reservoir area, about 1.3 km at the left upstream side of the dam site. Cumayeri Spring, with an average flow of 26 L/s, discharges close to the contact between the carbonate rock and the flysch at an elevation of 46 m a.s.l. Cumayeri Spring will be submerged by the reservoir under a water column of 67 m after full impoundment. The carbonate rock that forms the aquifer for the spring extends to the north toward the Black Sea and under the topographic divide between the Melen River Basin and the Black Sea. Other springs that drain the same carbonate rock aquifer occur at the north side of the

topographic divide. The springs at the adjacent (Black Sea) basin issue at elevations equal to or lower than the Cumayeri Spring elevation. A schematic cross section is depicted in Figure 2 to illustrate the setting described above. The two practical questions that arose as a consequence of the springs’ occurrence were the following: 1. Will the groundwater flow direction under natural conditions be reversed after the impoundment, and could this possibly drain the reservoir? 2. If this will be the case, how could an economical but effective concrete lining be designed as a cap over the carbonate rock that will be submerged after impoundment? The answer to the first question required a detailed hydrogeological study, including development of a representative hydrogeological conceptual model, and characterization of the groundwater system. Detailed geological mapping, drilling, in situ testing, hydrochemical and isotopic analyses, and piezometric level observations were performed to achieve this objective. The answer to the second question required a detailed analysis of the impact of covering the carbonate rock area in the reservoir. First, construction of a concrete lining to block the spring flow will develop piezometric pressure underneath the lining, also called “uplift pressure” (Terzaghi et al., 1996; Verruijt, 2001; and Das and Shoban, 2014). Estimation of the magnitude and spatial variation of the uplift pressure is essential in the design of the thickness of the concrete lining. Secondly, the uplift pressure will need to be relieved if its magnitude is so high that it requires a costly

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Ekmekci, Kalaycioglu, Acikel, Namkhai, and Akman

thick concrete lining (Ekmekci and Kalaycioglu, 2015). Analysis is required to assist in deciding on a costeffective method by which to achieve the relief of uplift pressure and to determine the best location for the relief structures (Middlebrooks and Jervis, 1997; Mansur et al., 2000; and Azizi et al., 2012). These objectives were achieved by performing numerical tests using a calibrated mathematical model and a conceptual hydrogeological model. PRELIMINARY DESIGN OF THE CONCRETE LINING An area of about 20 ha of a limestone outcrop on the left side of the Melen Dam reservoir area is to be covered by a concrete lining to prevent leakage (Figure 3A). The lining area is surrounded by low-permeability flysch unit at the south and east boundaries. The rock surface beneath the cap is designed to be terraced with steps, as shown in Figure 3B. The northern boundary is the 113-m elevation contour line, which demarcates the maximum water level of the reservoir. The concrete cap will cover the entire outcrop of the limestone that extends below the maximum reservoir level. Furthermore, the concrete cap is planned to extend 10 m down the limestone slope to the flysch along the southern boundary and to extend 10 m to the flysch along the eastern boundary to assure water tightness (DSI, 2011, 2013). A grout cutoff will be constructed down to a depth at which the permeability is less than 5 Lugeon value (Lu) to create a low-permeability boundary at the southern and eastern extents of the concrete cap (ES Proje, 2011). GEOLOGICAL SETTING The four lithological units that crop out in the area of the Melen Dam project (see Figure 1), from the oldest to the youngest, are as follows: Mesozoic flysch (Mf), Cretaceous limestone (Crl), Paleocene limestone (Pal), Eocene flysch (ef), and Neogene alternation of clastics (Nk) and the Quaternary alluvial (Qal) deposits in the Melen River valley (DSI, 1997). The Mesozoic flysch is composed of an alternation of sandstone, siltstone, and claystone. This unit crops out at a small area in the southwest of the study area. Stratigraphically, the Mesozoic flysch is overlain by the Cretaceous limestone. At its base the unit consists of non-carbonate layers of basal conglomerate and sandstone at the bottom, and it grades upward to an alternation of sandstone, claystone, and mudstone with an uppermost transition to limestone. About 40 percent of the area to be lined with concrete is covered by this unit. The Cretaceous limestone unit is conformably overlain by a Paleocene limestone with a much higher carbonate content. This thin- to medium-bedded limestone covers large areas

not only at the concrete lining area but also extends to the north and northwest to the adjacent basins. About 60 percent of the area to be covered by concrete is made of Paleocene limestone. In the study area the general dip of the limestone layers ranges between 20◦ and 30◦ to the north. Eocene flysch, comprising alternating marine claystone, siltstone, sandstone, and marl, stratigraphically overlies the Paleocene limestone. The sandstone layers in the flysch are thin to medium thick, carbonaceous, and fractured. A great part of the reservoir area is covered by the Eocene flysch. Therefore, the Paleocene limestone does not crop out at the reservoir area, as it is overlain by the Eocene flysch. However, large outcrops of the Paleocene limestone are observed in uplands where the Eocene flysch has been eroded. The northern ridge that separates the Melen Basin from the northern Black Sea Basin is completely composed of Paleocene limestone. At the Black Sea coast the Paleocene limestone is overlain directly by the younger Neogene siliciclastic sediments, where the Eocene flysch is lacking. The study area is situated in the tectonically weakly active Black Sea Region, where the main features were formed mainly by old compressional tectonics. This has initiated the closure of the northern branch of the Neotethys in the Maastrichtien (Okay et al., 1994). Thrusts and reverse faults are common in the region. The contacts between the Paleocene limestone and Eocene flysch units are of tectonic character in the study area. The older units overthrust the younger Eocene flysch at the southern boundary. The eastern contact between the Paleocene limestone and the Eocene flysch is of tectonic character, with a low-angle normal fault. None of the faults in the study area are active. A seismic risk analysis has been performed for the dam site and vicinity (Bozkurt, 1990). HYDROGEOLOGICAL SETTING AND CHARACTERIZATION The limestone units, namely the Paleocene limestone, being the more permeable and karstified, are the only water-bearing formations in the region. The limestone is medium to thick bedded and jointed. No clear manifestation of karstification has been observed in the study area. An epikarstic zone with small discontinuous cavities interconnected with enlarged fractures has been observed in very shallow excavations. Small-scale solutional openings have been developed, mainly along the bedding planes. A few closed depressions such as dolines are observed only at high elevations. The flysch units and the young Neogene siliciclastic sediments act as barriers to groundwater flow (Figure 4). The highlands covered by the Paleocene limestone exhibit karstic features such as dolines and sinkholes. These

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Simulation of Uplift Pressure

Figure 3. General layout of the concrete lining: A) extent of the area and B) north-south proﬁle.

features indicate that the Paleocene limestone aquifer is karstiﬁed to a certain extent. The aquifer is drained through three spring zones, one of which is located within the reservoir area of the dam (Cumayeri Spring),

whereas the other two (Demiracma and Hizardere) discharge at the adjacent basin. The Hizardere and Demiracma spring zones occur at the contact between the Paleocene limestone aquifer and the overlying

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Ekmekci, Kalaycioglu, Acikel, Namkhai, and Akman

Figure 4. Regional hydrogeological map.

Neogene siliciclastic sediments at elevations of 40 m and 55 m, respectively (see Figure 4). The Cumayeri Spring issues with an average flow rate of about 26 L/s at the reservoir area at an elevation of 46 m a.s.l., at the contact between the Paleocene limestone aquifer and the underlying Eocene flysch. The occurrence of karstic depressions and sinkholes on top of the ridge area suggests that the springs on either side of the topographical divide may have a common recharge area (see Figure 2). If this is the case, then locating the groundwater divide, which might not coincide with the topographical divide, is of utmost importance from the standpoint of the impact of the reservoir on the natural groundwater circulation. Several studies have been conducted to elucidate the hydrogeological setting at the site. Three tracer tests have been conducted by DSI (1992) to delineate and differentiate the recharge areas of the springs and to determine the groundwater flow direction and velocity. The tracer has been injected from a sinkhole and two deep boreholes and observed at the springs. The locations of tracer injection and the results obtained from the tests are depicted on the hydrogeological map in Figure 4. In two of the tracer tests, the tracer injected

at one side of the topographic divide is observed only at those springs located in the same drainage basin. No tracer has been detected when the injection is made from the adjacent drainage basin. However, in the third test, in which the tracer was injected at the adjacent basin and observed at the Melen Basin, the two sides of the topographic divide suggest that the groundwater divide may not coincide with topographic divide. The injection point is a borehole drilled at the top of the divide. Therefore, it is not surprising to observe the tracer at both sides of the divide. Based on this observation, it can be postulated that the groundwater divide, for the most part, coincides with the topographic divide. The groundwater velocities calculated from the three tracer tests range between 18 m/d and 1,028 m/d, implying localized ﬂow along conduits. A total of 24 exploration trenches were excavated and 42 boreholes were drilled at the study area for a) precise mapping and b) hydrogeological characterization of the lithological units. Packer tests were conducted in 2-m intervals to obtain the permeability proﬁle with depth. Performance of the test and evaluation of the results were made according to the method of Houlsby (1990). The boreholes were later constructed and

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Simulation of Uplift Pressure Table 1. Basic statistics of permeability of lithological units. Limestone Statistics Minimum Maximum Geometric mean Standard deviation cv (%)

P (Lu) 0.21 267.78 10.80 39.45 365.23

K (m/s) 2.09E-08 5.75E-01 1.23E-06 2.71E-02 2.20E + 06

Flysch P (Lu) 0.00 44.64 3.39 6.48 191.16

Total K (m/s) 2.42E-10 4.49E-06 3.39E-07 6.40E-07 188.72

P (Lu) 0.00 267.78 7.16 33.36 465.70

K (m/s) 2.42E-10 5.75E-01 7.80E-07 2.18E-02 2.79E + 06

K = hydraulic conductivity; P = permeability; cv = coefﬁcient of variability.

completed to be used for hydrochemical and isotopic sampling as well as groundwater level observations. One of the boreholes was drilled at the top of the topographical divide to observe the groundwater level fluctuation at the divide. The permeability in Lu changes over a wide range, with a geometric mean of about 7 Lu. Table 1 shows the basic statistics of permeability and the corresponding hydraulic conductivity of the limestone and the flysch. The geometric mean of the hydraulic conductivity of the limestone (1.2 × 10−6 m/s) is more than one order of magnitude higher than the mean of flysch (3.4 × 10−7 m/s). It should be noted that the high hydraulic conductivity of the flysch is due to the presence of the medium-thick sandstone layers. When only the sandstone interval was tested, permeabilities as high as 10 Lu were recorded. Comparatively, test intervals corresponding to siltstone, claystone, and marl layers recorded permeabilities less than 5 Lu. From the packer tests, it was also found that the permeability decreases with depth for both limestone (Figure 5A) and flysch (Figure 5B). The groundwater level measurements were made at 21 piezometers between January 2014 and February 2015. All piezometers were installed within the limestone. The 2-m screened interval was sealed with bentonite and cement. The groundwater level was found to slightly fluctuate during the observation period. Therefore, the mean groundwater level was considered to be representative in hydrogeological conceptualization. The flow rates of the Demiracma and Cumayeri Springs were recorded by the Devlet Su Isleri (DSI) during a period between 1991 and 1995. The flow rate of Demiracma Spring was always smaller than that of the Cumayeri Spring, presumably owing to the difference in elevation of issue. The flow rate of the Cumayeri Spring was measured to be as high as 260 L/s in high– water level periods. HYDROGEOLOGICAL CONCEPTUAL MODEL Using all hydrogeological data, a hydrogeological conceptual model was constructed to obtain a better understanding of the hydrogeological system at the

site and to provide a base for the numerical modeling. The hydraulic head distribution at the study area is depicted in Figure 6. The faults also demark the contacts between different lithological units having different hydrogeological properties. Therefore, faults were considered as hydrostratigraphic boundaries in the conceptual and numerical models. Based on the hydraulic conductivity values the limestone units were assumed to form a homogeneous hydrostratigraphic unit. The flysch was considered as the less permeable hydrostratigraphic unit. Both hydrostratigraphic units were assumed to be hydrogeologically isotropic. Karst has not been shown to affect the homogeneity and isotropy of the domain, which allowed us to assume that the groundwater flow is mainly of the diffuse type. This assumption is justified by the fact that the hydraulic conductivity of the limestone unit has a value in the order of 10−6 m/s, and it varies within only one order of magnitude in the limestone unit. Furthermore, the water pressure (Lu) tests showed that in only a few cases did turbulent flow occur during injection of water under pressure. Therefore, the conceptual model assumed a heterogeneous but isotropic flow domain. It is apparent that the Cumayeri Spring captures the groundwater that occurs at the site to be overlaid by concrete lining. A hydrogeological cross section through the Cumayeri Spring shows the flow lines and the groundwater divide in Figure 7A. The hydrogeological conditions expected to occur after the concrete lining are schematized in Figure 7B. According to this conceptualization the Cumayeri Spring is to be plugged and the free drainage will move from the Cumayeri Spring location to a higher elevation (113 m a.s.l.) at the upper edge of the concrete lining. Plugging of the Cumayeri Spring by the concrete slabs will obviously create a pressure buildup beneath the lining. The distribution of the pressure head actually will not be uniform throughout the covered area, and the value and distribution of the pressure head in terms of porewater pressure need to be predicted for an optimum design of the concrete lining in terms of thickness and the characteristics of the concrete mixture to be used in lining.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Ekmekci, Kalaycioglu, Acikel, Namkhai, and Akman

Figure 5. Permeability decrease with depth: A) limestone and B) ﬂysch.

CALCULATION OF UPLIFT PRESSURE The net uplift pressure that may build up under the concrete slabs is deﬁned by the net difference between

the downward and upward pressures acting on the top and bottom of the slab, respectively (Bollaert, 2009). It may be created under both hydrostatic and dynamic conditions (Terzaghi et al., 1996; Verruijt, 2001; and

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Simulation of Uplift Pressure

Figure 6. Hydraulic head distribution at the concrete lining area.

Das and Shoban, 2014). Predictions in this study were made considering the worst-case scenario to stay on the safe side. According to this scenario, the hydraulic head is at its maximum value at the groundwater divide, while the reservoir is emptied and the hydrostatic conditions prevail. This scenario allows the maximum buildup of the pore-water pressure underneath the concrete lining. The concrete lining should be designed to resist this hydrostatic uplift. The uplift pressure expected to develop beneath the planned concrete lining was predicted numerically using a commercial software Seep/W (Geo-Studio, 2009). The modeling was performed at two stages: simulation of the pre-lining (natural) conditions (calibration) and simulation of post-lining hydrogeological conditions. Following the calibration phase, sensitivity analyses were performed to evaluate the effect of input parameters and boundary conditions on the calculated heads and ﬂows. The model was found to be less sensitive to the boundary conditions. The hydraulic conductivity was found to be the most effective parameter, but mainly on the ﬂows. On the contrary, the head distribution, and therefore the pore pressure distribution, was less sensitive to this parameter.

Simulation of Pre-Lining Conditions The hydrogeological conceptual model was transferred to the numerical model, which solves the groundwater flow equation by finite element techniques. A finite element mesh was first constructed with a resolution of 1 m. The solution used is a two-dimensional solution that simulated the flow domain in the x-z (crosssection) plane. The area was simulated on 10 cross sections. Cross section 7, passing through the Cumayeri Spring, represents the most complex setting in the area. This cross section is used in this article to demonstrate the numerical modeling performed. The flow domain at this cross section was divided into 27,165 finite elements of 1 m in size. The solution requires definition of the boundary conditions and the hydraulic characteristics of the materials. The geometric means for the hydraulic conductivity were assigned to the limestone and the flysch. The flow domain was assumed to be isotropic and heterogeneous. The groundwater divide is defined as “constant head boundary,” and the bottom and the eastern boundaries of the model are defined as the “no-flow boundary.” The top of the model is a free surface boundary represented by the water table

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Ekmekci, Kalaycioglu, Acikel, Namkhai, and Akman

Figure 7. Flow net constructed for A) pre-lining and B) post-lining conditions.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79â&#x20AC;&#x201C;95

Simulation of Uplift Pressure

Figure 8. Simulated pore-water distribution on cross section 7 under natural conditions.

(see Figure 7). The groundwater divide, which is about 1.3 km from the Cumayeri Spring, was defined as an “infinite boundary,” a useful property of the software. The model was run to simulate steady-state conditions and was calibrated using both the hydraulic head distribution and the flow rate at the Cumayeri Spring location. The model estimates the hydraulic head at piezometers within a normalized root mean square error of about 9 percent. The simulated average flow rate at the Cumayeri Spring location is 1.76 m3 /d/m. This value corresponds to a discharge rate of 25.4 L/s, which is very close to the average of the observed value, 26 L/s. The result of the simulation is given as porewater pressure distribution in Figure 8. The pore-water pressure under natural conditions ranges between 0 kPa at the water table and 600 kPa. Simulation of Post-Lining Conditions Having calibrated to the natural (pre-lining) conditions, the model was used to simulate the effects of lining on the groundwater flow and on redistribution of the hydraulic head and the pore-water distribution. This information allowed for the prediction of the uplift pressure that will act on the concrete slabs. The only change on the model needed to achieve this was assignment of a no-flow boundary at the top of the model for the lined area and another no-flow boundary in the eastern boundary simulating the grout curtain

within the flysch material. The results of the simulation are given as hydraulic head and pore-water pressure distributions in Figure 9A and B, respectively. From Figure 9A it is apparent that the spring is relocated to a point at the aquifer that is now drained along the 113-m elevation contour line. The flow rate remained the same as expected. Figure 9B, on the other hand, shows that the pore-water pressure will increase to a value of 700 kPa. The software calculates the uplift pressure using the pore-water pressure distribution. Vertical profile of the calculated uplift pressure at the 53-m step is shown in Figure 10 as an example. This calculation was repeated for each step of the terraced surface. The modeling results revealed that the uplift pressure that builds up underneath the concrete lining is beyond the acceptable limit for a feasible thickness of reinforced concrete slabs. For a feasible design, the engineering company set up a threshold for the maximum uplift pressure of 50 kPa (ES Proje, oral communication). Therefore, the uplift pressure that was predicted to develop underneath the lining should be relieved (down to values below 50 kPa) by taking appropriate measures. OPTIMIZATION OF THE DESIGN BY RELIEF DRAINS The calibrated numerical model was then used for optimization of the design by trying different

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Ekmekci, Kalaycioglu, Acikel, Namkhai, and Akman

Figure 9. Simulated A) hydraulic heads and B) pore-water pressure distribution after concrete lining.

methods of reducing or relieving the piezometric pressure. A number of alternatives were attempted in order to relieve the uplift pressure by conducting numerical tests using the numerical model. One of the possible solutions was to drill boreholes at different depths along the upper edge of the concrete lining; these boreholes would act as relief wells when the piezo-

metric head exceeds the maximum water level in the reservoir. This option was not a solution to the problem because it only affected the piezometric pressure around the well itself (Figure 11). The second option was to construct a grout curtain along the upper edge of the concrete lining to a depth of practical application of grouting, which was given as 60 m from the

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79â&#x20AC;&#x201C;95

Simulation of Uplift Pressure

Figure 10. An example for vertical proﬁle of calculated uplift pressure for step 53-m elevation.

surface (ES Proje, oral communication). Application of this option has almost no effect on relieving the uplift pressure except in the vicinity of the grout curtain (Figure 12).

Construction of drains beneath the concrete lining was tried in the model and was found to be effective in reducing the pore-water pressure to an acceptable level. After several tries, the best solution found involved providing the acceptable relief with a total of six drains. Apart from that located at the Cumayeri Spring to drain the spring itself, five of the drains were located 2 m beneath the lining at the following elevation steps: 98 m, 83 m, 68 m, 53 m, and 38 m. The drain at elevation 38 m is located at the downstream end of the Cumayeri Spring to relieve the pressure at the contact with the flysch. All drains are 40 cm in diameter. The drains are to be constructed in 1-m-wide and 2-m-deep trenches within fine gravel cushion. The optimized solution for cross section 7 is depicted in Figure 13. The distribution of the final pore-water pressure along the terraced lining from the upper to the lower edge is given in Figure 14. The calculated flow through these drains showed that for high water level conditions the drains will convey water at approximately 120 L/s. The flow will concentrate at Cumayeri drain and 53-m step drain, each of which will be carrying water at about 35 L/s. Having the drains constructed, only about 1.2 L/s is expected to spill out from the upper edge of the lining. The drains beneath the 98-m, 83-m, 68-m, and 38-m elevation steps were found to conduct water at flow rates of 8, 10, 17, and 12 L/s. The flow beneath the 20-m-deep grout curtain within the flysch was calculated to be around 1 L/s. Based on the pore-water distribution and the flow rates, this design was found to be acceptable and technically and economically

Figure 11. Effect of relief well on pore-water pressure distribution under lining.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Ekmekci, Kalaycioglu, Acikel, Namkhai, and Akman

Figure 12. Effect of grouting curtain on pore-water pressure distribution under lining.

feasible. The drains are designed to convey the groundwater by free ﬂow to an area downstream of the dam site. CONCLUSIONS This study was conducted to predict the uplift pressure that was expected to develop underneath the

concrete lining after plugging a free-ﬂowing Cumayeri Spring at the Melen Dam reservoir area and to propose an optimum design for the concrete lining needed to prevent any leakage from the reservoir that might occur by reversal of the hydraulic gradient. This was achieved by numerical modeling of the hydrogeological system in three steps: simulation of the prelining and post-lining conditions and simulation of

Figure 13. Effect of relief drains on pore-water pressure distribution under lining.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Simulation of Uplift Pressure

Figure 14. Variation of pore-water pressure along x-axis of cross section 7.

various alternatives to relieve the uplift pressure buildup to an acceptable level. The studies performed to achieve these objectives revealed the following conclusions: The outcrop of the karstified limestone at the left bank of the Melen Dam reservoir area extends well beyond the hydrologic basin of the dam site and hydraulically connects the reservoir area with the adjacent hydrological basin. This hydrogeological setting suggests an uncertainty regarding the “water tightness” of the reservoir area. Lining the limestone area with concrete cap has been suggested to prevent any water leakage from the reservoir area. The lining was planned to cover the limestone area to the maximum water level of the reservoir, where a spring drains the limestone aquifer. Lining will also plug the free-flowing spring, which, according to the hydrogeological conceptualization and characterization, will consequently cause a piezometric pressure buildup underneath the concrete slabs. Numerical simulation of post-lining conditions revealed that the uplift pressure will range between 200 kPa and 700 kPa. Drains beneath the concrete lining were found to be the most effective method of maintaining the uplift pressure to a level less than or equal to 50 kPa. The Cumayeri Spring issuing at the lining area should be drained out and not plugged. Draining the Cumayeri Spring was found insufficient to prevent a piezomet-

ric pressure buildup. More drains at different locations should be constructed to relieve the uplift pressure. The most effective number and location of the drains were determined after a number of numerical simulations of the calibrated numerical model. A total of five drains of 40-cm diameter should be constructed in 2-m-deep trenches to relieve the uplift pressure to a value below 50 kPa at every point of the lining. The maximum uplift pressure under drained conditions is predicted to occur between steps 53 m a.s.l. and Cumayeri Spring with a value of 46 kPa. The effectiveness of the drains was not considered in this study. Keeping in mind the fact that the effectiveness of the drains tends to decline over time, design and construction of the drains is a crucial issue, particularly when the drains are located within the epikarstic zone hosting fine sediments. The numerical model was found to be sensitive to hydraulic conductivity but not to boundary conditions. ACKNOWLEDGMENTS This study was carried out for Ecetur, Inc. The authors are thankful to Mr. Ibrahim Tasbası (General Director of Ecetur, Inc.), civil engineer Yunus Emre Ulukutuk, and geologist Ilyas Bekci (Ecetur, Inc.) for their help with the field studies. The authors are

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Ekmekci, Kalaycioglu, Acikel, Namkhai, and Akman

grateful to Mr. Kerry Cato and the other two anonymous reviewers of the manuscript. They have made invaluable contributions to improve the manuscript with their constructive critique and comments. REFERENCES AMERICAN SOCIETY FOR TESTING AND MATERIALS (ASTM), 1996, ASTM Standards on Analysis of Hydrologic Parameters and Ground Water Modeling, ASTM Committee D-18: American Society for Testing and Materials, Philadelphia, PA. 144 p. ASTM, 1997, Guide for conceptualization and characterization of ground-water systems, D-5979-96. In ASTM Standards Relating to Environmental Site Characterization, ASTM Committee D-18: American Society for Testing and Materials, Philadelphia, PA. 1408 p. ANDERSON, M. P. and WOESSNER, W. W., 1992, Applied Groundwater Modeling: Simulation of Flow and Advective Transport: Academic Press, London, UK. 381 p. AZIZI, S.; SALMASI, F.; ABBASPOUR, A.; and ARVANAGHI, H., 2012, Weep hole and cut-off effect in decreasing of uplift pressure, case study: Yusefkand Mahabad Diversion Dam:Journal Civil Engineering Urban, Vol. 2, No. 3, pp. 97–101. BOLLAERT, E. F. R., 2009, Dynamic uplift of concrete linings: Theory and case studies. In Managing our Water Retention Systems, 29th Annual USSD Conference: Nashville, TN, April 20–24, 2009, pp. 149–164. BOZKURT, E., 1990, Seismic Risk Analysis Report for the Great Melen Dam Site: DSI, Ankara.Bruce, D. A., 2003, Sealing of massive water inﬂows through karst by grout: Principles and practice. In BECK BF (Editor), Sinkholes and the Engineering and Environmental Impacts of Karst: Geotechnical Special Publication 122, American Society of Civil Engineers, Reston, VA, 615 p. BRUCE, D. A.; DREESE, T. L.; and HEENAN, D. M., 2010, Design, construction, and performance of seepage barriers for dams on carbonate foundations: Environmental Engineering Geoscience, Vol. 16, No. 3, pp. 183–193. CHEN, J. Y.; ZHANG, L.; CHEN, Y.; YANG, B. Q.; and DING, Z. L., 2010, Equivalent method for simulating uplift pressure in dam model test: Advances Structural Engineering, Vol. 13, No. 6, pp. 1063–1073. DAS, B. M. and SHOBAN, K., 2014, Principles of Geotechnical Engineering, 8th ed.: CENGAGE Learning, Stamford, CT 06902 USA. 770 p. DSI, 1992, unpublished source, Groundwater Tracing Tests Report: Istanbul Drinking Water Project-Great Melen Dam: General Directorate of State Hydraulic Works, Ankara. DSI, 1997, unpublished source, Engineering Geological Report of Planning Phase of Great Melen Dam, Great Istanbul Drinking Water Project-Second Phase: Directorate of 14th District of State Hydraulic Works, Istanbul. DSI, 2011, unpublished source, Construction of Melen Dam Project. Volume 3/1: Directorate of 14th District of State Hydraulic Works, Istanbul. DSI, 2013, unpublished source, Technical Report of Melen Dam Mission: General Directorate of State Hydraulic Works, Ankara, 7 p. EKMEKCI, M. and KALAYCIOGLU, N. E., 2015, Hydrogeological Assessment of Concrete Lining at Left Bank of Melen Reservoir Area to Prevent Water Leakage: Project Report, Hacettepe University, International Research Center for Karst Water Resources, Ankara, 60 p. ES PROJE, 2011, oral communication, The Melen Project: DSI-Great

Istanbul Drinking Water Project-Second Phase: Project Introduction Report, Ankara, 90 p. FAZELI, M. A., 2007, Construction of grout curtain in karstic environment case study: Salman Farsi Dam: Environmental Geology, Vol. 51, pp. 791–796. GEBHART, L. R., 1973, Foundation seepage control option for existing dams: American Society of Civil Engineers, Inspections, Maintenance and Rehabilitation of Old Dams, Proceedings of Engineering Foundation Conference in Washington, pp. 660– 674. GEO-STUDIO, 2009, Seepage Modeling with SEEP/W 2007, An Engineering Methodology, 3rd ed.: GEO-SLOPE International Ltd., Alberta, Canada, 319 p. GUIFARRO, R.; FLORES, J.; and KREUZER, H., 1996, Francisco Morozan Dam, Honduras: The successful extension of a grout curtain in karstic limestone: International Journal Hydropower Dams, Vol. 3, No. 5, pp. 38–45. HANNA, A. W.; PLEWES, H. D.; WANG, L.; and SEYERS, W. C., 1993, Investigation of high uplift pressures beneath a concrete dam: Canadian Geotechnical Journal, Vol. 30, No. 6, pp. 974–990. HOULSBY, A. C., 1990, Construction and Design of Cement Grouting—A Guide to Grouting in Rock Foundations: John Wiley & Sons, Inc., New York, NY. 445 p. KNIGHT, M. A.; HARRIS, M. C.; VAN CLEAVE, B. E.; and HOCKENBERRY, A. N., 2010, Seepage remediation and karst foundation treatment at Clearwater Dam, Piedmont, Missouri: Environmental Engineering Geoscience, Vol. 16, No. 3, pp. 195–210. KRESIC, N., 2006, Hydrogeology and Groundwater Modeling, 2nd ed: CRC Press, Taylor and Francis Group, New York. 828 p. LU, R. L.; SUN, D. P.; WEI, W.; and ZHOU, J. I., 2013, Numerical simulation of seepage ﬁeld in the tailing dam with draining seepage system. In Proceedings of the 4th International Conference on Digital Manufacturing and Automation (ICDMA), June 29–30, 2013, Qindao, Shandong. IEEE, China, pp. 858– 861. MANSUR, C.; POSTOL, G.; and SALLEY, J., 2000, Performance of relief well systems along Mississippi river levees: Journal Geotechnical Geo-Environmental Engineering, Vol. 126, No. 8, pp. 727– 738. MARTT, D. F.; SHAKOOR, A.; and GREENE, B. H., 2005, Austin Dam, Pennsylvania: The sliding failure of a concrete gravity dam: Environmental Engineering Geoscience, Vol. 11, No. 1, pp. 61–72. MIDDLEBROOKS, T. A. and JERVIS, W. H., 1997, Relief wells for dams and levees: Journal Hydraulic Engineering, Vol. 112, No. 2, pp. 1323–1336. MILANOVIC, P. T., 2004, Water Resources Engineering in Karst: CRC Press, Boca Raton, FL USA. 328 p. MILANOVIC, P. T., 2011, Dams and reservoirs in Karst. In VAN BEYNEN, P. E. (Editor), Karst Management: Springer, Dordrecht, Heidelberg Germany. pp. 47–74. MOZAFARI, M. and REISI, E., 2015, Understanding karst leakage at the Kowsar Dam, Iran, by hydrogeological analysis: Environmental Engineering Geoscience, Vol. 16, No. 4, pp. 325–339. ¨ UR ¨ , A. M. C.; and GOR ¨ , N., 1994, Kinematic OKAY, A. I.; S¸ENGOR history of the opening of the Black Sea and its effect on the surrounding regions: Geology, Vol. 22, pp. 267–270. OZKAN, S.; ADRIAN, D.; SILLS, G.; and SINGH, V., 2008, Transient head development due to ﬂood induced seepage under levees: Journal Geotechnical Geo-Environmental Engineering, Vol. 134, No. 6, pp. 781–789. ROMAN, W. M.; HOCKENBERRY, A. N.; BEREZNIAK, J. N.; WILSON, D. B.; and KNIGHT, M.A., 2013, Evaluation of grouting for hydraulic barriers in rock: Environmental Engineering Geoscience, Vol. 19, No. 4, pp. 363–375.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Simulation of Uplift Pressure TERZAGHI, K.; PECK, R. P.; and MESRI, G., 1996, Soil Mechanics in Engineering Practice, 3rd ed.: John Wiley & Sons, Inc., New York. 549 p. United Nations Department of Economic and Social Affairs (UNDESA), 2014, World Urbanization Prospects, The 2014 Revision: Highlights: United Nations, New York. 27 p. U.S. Army Corps of Engineers (USACE), 1993, Seepage Analysis and Control for Dams: U.S. Army Corps of Engineers Manual No. 1110-2-1901, Washington, DC. 392 p.

UTILI, S.; YIN, Z.; and JIANG, M., 2008, Inﬂuences of hydraulic uplift pressures on stability of Gravity Dam: Chinese Journal Rock Mechanics Engineering, Vol. 27, No. 8, pp. 1554–1568. VERRUIJT, A., 2001, Soil Mechanics: Delft University of Technology, The Netherlands. 315 p. YOUSEFI, S.; NOORZAD, A.; GHAEMIAN, M.; and KHARAGHANI, S., 2013, Seepage investigation of embankment dams using numerical modelling of temperature ﬁeld: Indian Journal Science Technology, Vol. 6, No. 8, pp. 5078–5082.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 79–95

Comparing Rock Discontinuity Measurements Using Geological Compass, Smartphone Application, and Laser Scanning Methods NICHOLAS J. FARNY nfarny@gmail.com

Key Terms: Engineering Geology, Rock Mechanics, Site Investigations, Remote Sensing ABSTRACT To determine the stability of rock slopes, discontinuity orientations must be measured accurately. Several methods exist for taking field measurements of discontinuity orientations. The objective of this study was to compare the traditional method of hand measuring discontinuity orientations using a Brunton compass to the more modern methods of measuring discontinuities using smartphone applications and ground-based light detection and ranging (LiDAR) laser scanning. The field site is a rock outcrop along Stroubles Road on Price Mountain, in Montgomery County, VA. The Price Mountain structure is a doubly plunging anticline exposed in a window in the Pulaski Thrust Sheet. The rock outcrop consists of a Mississippian sandstone mapped as the Upper Price Formation. The site features a set of bedding planes that dip steeply into the cut slope face, yielding toppling failures. In addition, two distinct sets of joints exist, creating both planar and wedge failures in the cut slope. Using window mapping, discontinuity orientations were measured along 200 ft (61 m) of outcrop using a Brunton compass, a smartphone application, and a laser scan. These measurements were compared using stereonet analyses to determine the relative accuracy of the different methods. The results show a strong agreement between measurements taken with the Brunton compass and the smartphone application. However, the laser scan shows that scanner data need calibration with field measurements and observations to yield equally good results. Remote-sensing methods using laser scanning, such as terrestrial photogrammetry, cannot be used completely independent of traditional field characterization and input from experienced professionals. INTRODUCTION Statement of Problem Rock slope stability analyses are dependent on the accurate survey of potentially unstable areas. To accurately evaluate the stability of a rock mass, the ori-

entations and properties of discontinuities need to be evaluated in the field, and the discontinuities need to be characterized relative to the existing or proposed slope orientations. Traditional engineering geology rock mass characterization methods such as the detail line and window mapping methods require precise measurements using geologic compasses. Rock mass characterization for rock slope stability analysis using these traditional methods is often a hazardous and time-consuming endeavor, and it comes with many inherent uncertainties from human error. Studying a rock slope with potential for falling or sliding blocks can place the surveyor in danger. Recent technological developments have led to more advanced tools and methods that can enhance efficiency. The remote-sensing technique of laser scanning can analyze an unstable slope at a safe distance. In addition to the remote-sensing technologies, various different software applications have been developed to utilize the accelerometer in a smartphone, allowing it to be used as a geological compass. Using these applications, a smartphone can take dip directions and dips all in one step, similar to a Clar compass. Objective of Study The purpose of this study was to compare window mapping using a Brunton compass with more modern tools such as smartphone applications and remote-sensing techniques such as laser scanning. Rock mass discontinuities along a rock outcrop were measured using three different tools, a Brunton geological compass (Brunton), an iPhone4 with the “GeoID: Smartphone Inclinometer” application software by DutchSaigon (2016; smartphone), and a ground-based laser scanner. The discontinuity orientations were then compared using stereonet analysis to determine their accuracy relative to each other, with the results from the Brunton geological compass, which were used as the scientific control. SITE LOCATION AND REGIONAL GEOLOGY Price Mountain, VA, is located near Blacksburg near the junction of the Southern and Central Appalachi-

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97–111

Farny

Figure 1. Site location and vicinity map (U.S. Geological Survey, 2013).

ans in the Appalachian Overthrust Belt, bounded by the Blue Ridge overthrust to the southeast. The Southern Appalachians trend N60–65E, while the Central Appalachians trend N30–35E (Lowry, 1979). Price Mountain is a doubly plunging anticline of Mississippian and Devonian rock in a structural window of the Pulaski Thrust Sheet, which itself is part of the Overthrust Belt of the Southern Appalachians. The speciﬁc site is a road cut on the northwestern edge of Price Mountain, along Stroubles Creek

Road (Figure 1). This 200 ft (61 m) long, 10 ft (3 m) high outcrop of Mississippian Upper Price Formation sandstone contains numerous discontinuities in the rock mass that can form plane and wedge blocks that may be kinematically subject to sliding. The discontinuous rock mass in this road-cut exposure was chosen for characterization and analysis in this study (Figure 2). The rock outcrop lies entirely within the Mississippian Upper Price Formation, which has a thickness of

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97–111

Rock Discontinuity Measurements Using Compass, Smartphone, and Laser Scanning

Figure 2. The road outcrop in Upper Price Formation along Stroubles Creek Road, used for window 1 of the window mapping survey.

350 m in the Saltville Thrust Sheet. The basal 20 to 40 m of this formation are a dark-gray to black thinly laminated mudstone with plant fossils. Within the mudstones, there are several coal seams. Above the basal mudstone and coal seams, there is a cyclic nonmarine sequence of mudstone, siltstone, and sandstones with plant fossils (Bartholomew et al., 2000). The exposure analyzed for this study consists entirely of sandstone with plant fossils. METHODS The characterization of rock masses requires extensive mapping to accurately record the orientations and properties of the discontinuities. This section covers the methods and tools utilized for this study. A combination of traditional and modern methods of rock mass characterization was utilized to present an objective comparison of discontinuity orientations only. The methods used in this study for discontinuity orientation documentation were traditional and modiﬁed window mapping and laser scanning. In addition, direct shear testing was utilized to determine the shear strength properties of discontinuities in the Upper Price Formation at the study area. Window Mapping This traditional method of rock mass characterization involves establishing “windows” on a rock slope. A window is a subset of the rock slope that is representative of the dominant rock mass conditions. Once the window size and location are established, the ge-

ologist visually assesses the rock mass within the window to select representative locations for measurement of the dominant and minor discontinuity orientations. For convenience in data management, the discontinuities are grouped first into type, and joints are further grouped into sets. Window mapping was the chosen method for obtaining Brunton and smartphone measured discontinuities. This survey was also used to obtain rock mass information other than discontinuity orientations. For each discontinuity identified, information was recorded using a data sheet. The properties recorded included rock type, discontinuity structure type, dip direction, dip, water presence, roughness, aperture, discontinuity length and persistence, location of the discontinuity, and infilling amount and type. Other pertinent information, such as the time, date, location, job title, and weather conditions were recorded as well. Two windows were utilized, each with a length of 100 ft (30 m) and a height ranging between 10 to 15 ft (3 to 5 m). In total, 45 discontinuities were measured in window 1, and 55 were measured in window 2, for a total of 100 discontinuities. Only discontinuities within reachable height of the cut slope were measured; no rappelling or rock climbing was done during the characterization. For each discontinuity, the dip direction and dip were each measured twice, once with a Brunton compass and once with the smartphone. This resulted in 200 total discontinuity orientation measurements. The difference between the two windows was the traverse trend and structure. Window 1 has a traverse trend of N50◦ E, and window 2 has a traverse trend of N56◦ E. The difference in alignment of the road cut results from the change in the direction of Stroubles Creek Road. In window 1, the dip direction and dip measurements taken by both the Brunton compass and the smartphone were taken on bare discontinuity surfaces. In window 2, a clipboard was introduced and placed over the discontinuity surface prior to Brunton and smartphone measurements. The purpose was to provide a uniform surface on which to take orientation measurements that was unaffected by the small-scale roughness and waviness of the discontinuity. The metallic clip of the clipboard was removed prior to use, in order to prevent the clip metal from disrupting magnetic readings of the Brunton compass. The smartphone application software used during the time of this study does not correct for the magnetic declination of the area. In order to account for this, the magnetic declination on the Brunton compass was set to zero. The laser scanner point clouds were orientated and georeferenced using prominent discontinuities with Brunton measurements. In calibrating the

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97–111

Farny

point cloud in this way, it is ensured that all of the data are consistently uncorrected for magnetic declination. Declination was added later before the stereonet analysis. Laser Scanning The laser scan of the test outcrop was completed using the Leica ScanStation 2. The scanner was mounted on a tripod, placed across the road from the outcrop at a distance of 21 to 25 ft (6 to 8 m). Each scan, including equipment setup, took approximately 30 minutes to complete. Tripod-mounted targets were used for scanner registration, with at least two targets being present in each scan. In total, six 360-degree scans were used to cover the entire outcrop. The six scans had a large percentage of overlap in an effort to minimize shadow zones in the laser scanner point clouds. Data Processing In order to extract the discontinuity orientations from laser scan point clouds, they must be processed and trimmed to ensure the best results possible. Leica Cyclone (Leica Geosystems, 2016) and Split-FX software (Split Engineering, 2016) were used to process the data from the laser scanner. Leica Cyclone by Leica Geosystems (2016) was used first to process the raw data from the scanner. Since the six scans from the ground-based laser scanner survey incorporated 360 degrees of horizontal rotation, they contained a large amount of unnecessary data points. Cyclone was used to eliminate these unnecessary data points, trimming the 360-degree point clouds down to the rock slope data points only. This program was also used to remove anomalous data points, such as vegetation and outliers caused by atmospheric interference or lighting problems. Cyclone was then used to convert the scans from text files in the .txt format to the .xyz file format, for use in the software package Split-FX. Split-FX by Split Engineering (2016) is software designed to analyze laser scanner point clouds. This program takes x-y-z point clouds and uses an automated process to identify discontinuity patches within the point cloud. The first step in this process is registering the point cloud based on real-world reference orientations (Figures 3 and 4). Once the point cloud is correctly orientated, a triangulated mesh surface is generated. A three-dimensional surface reconstruction is performed using Delauney triangulation, a polygonal technique (Figure 5). This creates triangular facets based on three points using interpolation. After this, the discontinuity “patches” are identified within the

100

Figure 3. Field sites of window 1 (above) and window 2 (below).

triangular mesh (Figure 6). Patches are surfaces built from adjacent triangles that have similar orientations, within specified limits. Once the normal to a flat triangular facet of the mesh is found in the mesh, the surrounding area is searched for triangles with similar properties to expand the patch (Kemeny et al., 2006). In addition to the automated discontinuity identification process, experience and judgement were used to manually insert, modify, and delete misidentified patches in order to create a more robust and accurate model. Following the identification of discontinuities and their sets, orientations were exported for stereonet analysis. STEREOGRAPHIC ANALYSIS Stereographic projection was used to analyze the orientations of the discontinuities measured by the Brunton, smartphone, and laser scanning, cluster them into sets, and examine the kinematic stability of potential planar, wedge, and toppling failures. The discontinuity measurements taken by the Brunton were used as

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97–111

Rock Discontinuity Measurements Using Compass, Smartphone, and Laser Scanning

Figure 4. Point clouds in Split-FX generated by a laser scan of sites shown in Figure 3.

the control in this study, as it is the standard and most widely used tool for measuring discontinuity orientations. Methods Stereographic projection allows for threedimensional data to be represented in two dimensions on a stereonet. This provides an effective way to quickly examine the angular relationships, but it does not take the position or size of feature into account (Wyllie and Mah, 2004). The Dips (Rocscience, 2016) and RockPack III (RockWare 2016) programs were used to perform the stereographic analysis for this study. In order to compare the relative accuracy of the different discontinuity measurement methods used in this study, contoured pole plots of the data were created. Based on this information, discontinuities poles

were assigned to sets or were excluded from sets and considered random or background orientations in the rock mass. The uncorrected, contoured pole plots of the window mapping survey are shown in Figures 7 to 12. From the identiﬁed discontinuity sets, an average dip/dip direction orientation for the sets was calculated (Tables 1 and 2). These average dip/dip direction set orientations are represented on the stereonets in Figures 7 to 12 by great circles. These great circles, representing average discontinuity set orientations, are key factors in Markland’s tests for kinematic analysis. Markland’s test involves plotting both a great circle for the orientation of slope and a friction circle that represents the friction angle of the discontinuities. If the mid-point of a discontinuity set great circle, or an intersection of discontinuity set great circles, plots between the angle of the slope and the friction angle,

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97–111

101

Farny

Figure 5. Point clouds of ďŹ eld sites in Split-FX overlaid with triangular mesh.

it lies in the critical zone for either a planar failure, in the case of a single great circle, or a wedge failure, in the case of an intersection of great circles. That is, the average dip of the discontinuities or intersection is both (1) less than the dip angle of the slope, and therefore will daylight, and (2) greater than the friction angle of the material, and therefore may be

unstable. A separate Markland test is used for toppling failure, in which the discontinuity set orientation great circle must plot inside a separate critical zone in which interlayer slip occurs (Goodman, 1980). The Markland tests performed on the discontinuity data from the various surveys are shown in Figures 13 to 16.

Table 1. Summary of average joint set orientations from window 1.

Table 2. Summary of average joint set orientations from window 2.

Window 1 Average Discontinuity Set Orientation (Dip/Dip Direction) Method Brunton Compass Smartphone Laser Scanner

102

Bedding Planes

Joint Set A

Joint Set B

69/325 68/323 77/319

41/109 42/110 45/115

37/160 37/165 45/161

Window 2 Average Discontinuity Set Orientation (Dip/Dip Direction) Method Brunton Compass Smartphone Laser Scanner

Bedding Planes

Joint Set A + B

61/335 58/333 63/337

47/155 46/150 51/156

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97â&#x20AC;&#x201C;111

Rock Discontinuity Measurements Using Compass, Smartphone, and Laser Scanning

Figure 6. Point clouds of ﬁeld sites in Split-FX with patches outlined in blue. Areas outlined in red indicate portions of a patch that lie outside the set allowable deviation for triangle orientation.

RESULTS AND DISCUSSION The window 1 stereonets with Brunton- and smartphone-measured discontinuities (Figures 7 and 8 and Table 1) both show nearly the same discontinuity sets. These include the bedding planes, and two set of joints, designated as Joints A and Joints B. Great circle mid-points for both sets of joints plot in the critical zone created by the slope great circle (dip of 90 degrees and a dip direction of 140 degrees) and the friction angle circle, which is assigned a value of 30 degrees based on the direct shear testing of Upper Price Formation sandstones. Therefore, they may be susceptible to planar failure (Figure 13). In addition, Joints A and Joints B intersect in the critical zone, creating kinematically possible rock wedges. Joints A and B also intersect the bedding planes for possible wedge fail-

ures, but these intersections are located outside of the critical zone and therefore have a low probability of movement based purely on the orientation of the discontinuities and the cut slope. In addition, the bedding planes great circle plots inside the critical zone designated for potential toppling failure. The results of these Markland tests reﬂect the failure types observed in the ﬁeld. The window 1 Brunton contoured pole plot and the laser scanner contoured pole plot (Figures 7 and 9 and Table 1) show some radical differences. The most prominent of these is the greater number of steeply dipping discontinuities included in the laser scan bedding planes set relative to the bedding planes set for the Brunton. Additionally, the laser scan produces a bedding plane set with a much higher concentration of poles relative to sets for Joints A and Joints

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97–111

103

Farny

Figure 7. Contoured pole plots of discontinuities from window 1 measured using a Brunton compass.

B. As before, Joints A and Joints B plot inside the Markland critical zone, creating kinematically possible planar failures, and they also intersect within the critical zone, creating a possible wedge failure (Figure 14). The same potential for toppling failures of the bedding planes set also exists in the laser scanner results. In window 2, changes in the discontinuity orientations, the orientation of the road alignment, and the orientation of the slope face (dip of 90 degrees and a dip direction of 146 degrees) lead to a slightly different distribution of discontinuity sets. The same friction angle of 30 degrees is used in window 2 to construct a zone of critical concern for stability. As seen in the plots of the Brunton- and smartphonemeasured discontinuities (Figures 10 and 11 and Table 2), Joints A and Joints B have merged into a single cluster. It is difďŹ cult to differentiate between the two sets, and therefore they are treated as a single set named Joints A + B. This joint set plots in the critical zone for planar failure (Figure 15). It also intersects the bedding planes for possible wedge failures, but these intersections are located outside of the Markland critical zone and therefore have

104

a low probability of sliding. As with window 1, the bedding planes present similar possible toppling hazards. As with the window 1 results, the laser scanner data from window 2 (Figures 10 and 12 and Table 2) reveal additional steeply dipping discontinuities, enlarging both the bedding planes set and the Joints A + B set. This does not change the potential failure types indicated by the Markland tests in Figure 16.

Discrepancies between Laser Scanner and Window Mapping Surveys The laser scanning results contain numerous steeply dipping discontinuities that are absent from the window mapping results. These additional discontinuities increase the dip values of certain discontinuity sets, speciďŹ cally those of the bedding planes in both windows, and Joints A + B in window 2, to values closer to vertical. There are three possible explanations for these differences.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97â&#x20AC;&#x201C;111

Rock Discontinuity Measurements Using Compass, Smartphone, and Laser Scanning

Figure 8. Contoured pole plots of discontinuities from window 1 measured using a smartphone.

The variation in the near steeply dipping discontinuities of both windows was perhaps caused, at least in part, by geologic bias, as they were not identified during the window mapping survey. Reevaluation of the slope and these discontinuities showed them to be near-vertical surfaces with small areas, less than a 1 ft2 (0.09 m2 ) in most cases. Because of the small size and near-vertical orientations, they are not as readily identifiable as the larger bedding planes or joint sets. Therefore, the window mapping survey had a bias from the geologist, where larger, more dominant discontinuities with more control over the slope were preferentially measured over other, smaller discontinuities. Another likely explanation relates to the height on the cut slope where these discontinuities were measured. In both windows, several of the discontinuities were located high on the slope, out of reach. The slope itself is not much higher than 10 ft (3 m) in most cases, but this is still high enough to create difficulties in reaching these discontinuities on the slope, which could have only been measured by rappelling or climbing, techniques that were not utilized in this study. The

laser scanner survey was able to scan the entirety of slope, picking up such discontinuities and providing more detail and a complete picture of the character of the rock mass. Idiosyncrasies in the Split-FX discontinuity identification process represent another probable source of these near-vertical discontinuities. The automated discontinuity identification algorithm faces daunting challenges; is not a perfect algorithm, and certain conditions can either create false discontinuities or leave discontinuities unidentified. Objects such as fallen rocks, soil surfaces, or flat surfaces on the rock face can be mislabeled as discontinuities. Shadow zones of missing points in the laser scanner point cloud can be created by obstructions or poor scan angles. The shadow zones lead to holes in the triangular mesh, and therefore missing discontinuities. Because of these difficulties, it is vital that the laser scanning results be closely examined after processing. Split-FX allows for the user to delete, modify, or insert discontinuities to remedy these problems, but misinterpretations from the software and user inexperience are likely to remain.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97–111

105

Farny

Figure 9. Contoured pole plots of discontinuities from window 1 measured using a laser scanner.

Aspects of the three explanations for the differences between laser scanning and window mapping results are likely to be applicable in this study. The laser scanner can capture a greater percentage of the slope in more detail than a simple window mapping survey, leading to a better understanding of the rock mass without bias from the geologist. However, errors in the laser scanner point cloud and in the software processing can create anomalous discontinuities. The majority of these anomalies were likely corrected in this study by the use of geologic judgment, which is essential in characterizing discontinuous rock masses, but it is expected that not all were found and corrected.

CONCLUSIONS Accuracy of Rock Mass Characterization From the stereonet analysis, some conclusions can be drawn about the relative accuracy of the smartphone

106

and laser scanner data in comparison with the Brunton data, which is used as the scientiďŹ c control. Overall, the discontinuity measurements from the Brunton agree very well with the measurements collected with the smartphone. From Tables 1 and 2, the difference between the average set orientations for the Brunton and smartphone data is no more than 5 degrees. It should be noted that the comparison between the Brunton and smartphone measurements was done on a one-to-one basis, as each discontinuity on the exposure was measured twice, once with the Brunton and once with the smartphone. The comparison between the laser scanner data and the Brunton data is not as straightforward. The measurements from the Brunton generally agree with the measurements collected by laser scanning. The high accuracy of discontinuity measurements from remotesensing methods is echoed in other sources, such as Hanzel (2012), Shaffner et al. (2004), and Otoo et al. (2011). However, in the case of this study, the overall characterization of the slope differs between the two methods. As mentioned in the Results and Discussion section, several small, steeply dipping discontinuities

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97â&#x20AC;&#x201C;111

Rock Discontinuity Measurements Using Compass, Smartphone, and Laser Scanning

Figure 10. Contoured pole plots of discontinuities from window 2 measured using a Brunton compass.

were included in the laser scan that were not included in the window mapping survey because the window was defined as the full height of the road cut, which included exposures that were beyond reach of the engineering geologist measuring the discontinuities. Their exclusion from the window mapping survey can be explained as geologist bias, as well as certain challenges in the use of laser scanners with Split-FX for purposes of measuring discontinuities. One issue regarding the laser scanning of rock masses is that more geologic judgment is still needed to interpret laser scanning models of certain slope failure types. Most of the rock slope presented in this study had large, well-defined clearly visible faces and discontinuities that were effectively identified by the laser scanner. However, a slope with stability that is controlled not by sets of discontinuities, but rather more complex processes like raveling will require greater judgement by the geologist in evaluating laser scanner results, because flat surfaces of the rock mass are not discriminated from flat surfaces of fallen rock blocks. An example of this is a section between windows 1 and 2 at the Price Mountain site. While windows 1 and 2 have

well-defined structural regimes with prominent discontinuities, the middle 20 ft (6 m) section of transition is controlled by raveling and requires more geologic judgment than either of the two adjacent evaluation windows. Geologic observation is still needed because many properties of discontinuities that play into the stability of a rock slope require direct observation by an experienced qualified geologist. Some of these properties include structure type, rock type and shear strength, discontinuity infilling materials, site water conditions, discontinuity persistence and length, and rock quality and hardness. Field observations by experienced geologists and engineers continue to be required. The results of any remote-sensing technique like laser scanning or photogrammetry need to be checked for validity and accuracy in order to ensure the validity of the results. At the same time, a geologist might sometimes miss important features in the slope that a remote-sensing technique will pick up. Therefore, in terms of accuracy, it is often best to use traditional and modern methods in concert with one another, as they supplement each other in various ways.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97–111

107

Farny

Figure 11. Contoured pole plots of discontinuities from window 2 measured using a smartphone.

Recommendations for Further Study While this study provides a preliminary examination of various new methods and tools used to characterize rock slopes, it is not exhaustive. There are many more ways to examine the stability of slopes than are discussed here, but they were not utilized in this study. Similarly, the analysis methods also have room for expansion and improvement. The following is a list of possible topics for further consideration and study:

• This comparison was intended to include a terrestrial digital photogrammetry survey. Due to problems with the ﬁeld survey and acquiring appropriate data processing software, the terrestrial digital photogrammetry portion of this study remains unﬁnished. A detailed comparison between this method and laser scanning would be especially useful, as both are remote-sensing methods used to accomplish the same task. Terrestrial photogrammetry software produces a three-dimensional model from stereoscopic photographs that have position 108

and scale features visible in both photos. The threedimensional model has thousands to a million points with x-y-z coordinates that can be rotated to north orientation and translated to coordinate reference, if needed. These points can be exported for use in other software, including Split-FX, which is essentially identical to the laser scanner raw data. The terrestrial photogrammetry systems include software that allows removal of points on vegetation, fallen rocks, and soil surfaces, as well as virtual mapping of geologic structures using points, lines, and polygons. Each measurement is made by the user, who uses geologic judgment to select locations for measurements and can toggle back and forth between the three-dimensional photographic model view and stereographic projection view. The user can zoom into a location to differentiate between bedding and joints, and to identify end points of aligned joints or fractures needed for deﬁning rock bridges. Measurements can be grouped using conventional computer drag-and-drop functionality. Thus, the virtual geologic mapping of discontinuities using terrestrial

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97–111

Rock Discontinuity Measurements Using Compass, Smartphone, and Laser Scanning

Figure 12. Contoured pole plots of discontinuities from window 2 measured using a laser scanner.

digital photogrammetry, albeit the result of remote sensing, is very different from automated discontinuity deﬁnition and is more similar to geologic ﬁeld observations.

• Only one traditional method of rock mass characterization was used in this study, window mapping using a Brunton compass. In addition to this, there is the detail line method. It is possible that a detail

Figure 13. Markland tests of window 1 discontinuity sets measured with a Brunton compass and a smartphone.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97–111

109

Farny

Figure 14. Markland tests of window 1 discontinuity sets measured with a Brunton compass and a laser scanner.

line survey would characterize the slope differently. Another traditional method is oriented core logging. Like the detail line, it surveys a narrow, linear portion of the rock mass. • The choice of equipment and processing software affects the accuracy of both laser scanning and photogrammetry surveys. A useful comparison could be made of the results of surveys made with different scanners/cameras and/or processing software. Similarly, “GeoID: Smartphone Inclinometer” is not the only smartphone application that can measure discontinuity orientations. A comparison of different applications could yield valuable information on their relative accuracy.

• The transition zone between the two windows had relatively poor exposures. The sandstone present in both windows may have been eroded by a channel, which allowed plants to grow in Mississippian time, resulting in the coal seams observed during the study. The details of this transition zone are not important to the topic of this paper but provide an opportunity to note that neither the traditional methods nor the modern methods of rock mass characterization seem to have an advantage. Further work can be done on developing methods of reﬁning the laser scan and/or processing steps to better capture the structure of slopes like this, for which stability is controlled by raveling or the strength of the rock material.

Figure 15. Markland tests of window 2 discontinuity sets measured with a Brunton compass and a smartphone.

110

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97–111

Rock Discontinuity Measurements Using Compass, Smartphone, and Laser Scanning

Figure 16. Markland tests of window 2 discontinuity sets measured with a Brunton compass and a laser scanner.

ACKNOWLEDGMENTS I would like to thank my thesis committee for their support and guidance; my major professor and thesis committee chair, Terry R. West, of Purdue University; Chester F. Watts of Radford University; and Thomas M. Tharp of Purdue University. Thanks are extended to Brendan Fisher, Brian Norton, Marcus Jessee, and Stephen Underwood for their assistance in this study. Thanks are also extended to the Purdue University Department of Earth, Atmospheric, and Planetary Sciences, which provided teaching assistant funding during my graduate studies. REFERENCES BARTHOLOMEW, M. J.; SCHULTZ, A. P.; LEWIS, S. E.; MCDOWELL, R. C.; AND HENIKA, W. S., 2000, Digital Geologic Map of the Radford 30 × 60 Minute Quadrangle Virginia and West Virginia: Commonwealth of Virginia Department of Mines, Mineral, and Energy, Division of Mineral Resources, Charlottesville, VA. DUTCHSAIGON, 2016, GeoID: Smartphone Inclinometer: Electronic document, available at paly.google.com: https://play.google. com/store/apps/details?id=snu.geo.id&hl=en GOODMAN, R. E., 1980, Introduction to Rock Mechanics, 2nd ed.: John Wiley and Sons, Toronto, Canada. HANZEL, J., 2012, LiDAR-Based Fracture Characterization: An Outcrop-Scale Study of the Woodford Shale, McAlister Shale Pit, Oklahoma: Master’s Thesis, Oklahoma State University, Stillwater, OK.

KEMENY, J.; TURNER, K.; AND NORTON, B., 2006, LiDAR for rock mass characterization: Hardware, software, accuracy and bestpractices. In Laser and Photogrammetric Methods for Rock Face Characterization Workshop: American Rock Mechanics Association, Golden, CO, pp. 49–62. LEICA GEOSYSTEMS, 2016, Leica Cyclone: 3D Point Cloud Processing Software: Electronic document, available at http://hds.leicageosystems.com/en/Leica-Cyclone_6515.htm LOWRY, W. D., 1979, Nature of thrusting along the Allegheny Front near Pearisburg and of overthrusting in the BlacksburgRadford area of Virginia. In 11th Annual Virginia Geotechnical Field Conference Guidebook 8: Virginia Polytechnic Institute and State University Department of Geological Sciences, Blacksburg, VA. OTOO, J. N.; MAERZ, N. H.; DUAN, Y.; AND XIAOLING, L., 2011, LiDAR and optical imaging for 3-D fracture orientations. In NSF Engineering Research and Innovation Conference: Atlanta, GA. ROCKWARE, 2016, RockPack III: Electronic document, available at http://www.rockware.com/product/overview.php?id=119 ROCSCIENCE, 2016, Dips 6.0: Graphical and Statistical Analysis of Orientation Data: Electronic document, available at https://www.rocscience.com/rocscience/products/dips SHAFFNER, P.; KROSLEY, L.; AND KOTTENSTETTE, J., 2004, U.S. Bureau of Reclamation Digital Photogrammetry Research Report: Department of the Interior, Bureau of Reclamation, Report No. DSO-04-01. SPLIT ENGINEERING, 2016, Split-FX 2.1.0 Software: Electronic document, available at http://www.spliteng.com/products/splitfx-analysis-service/ U.S. GEOLOGICAL SURVEY, 2013, U.S. Blacksburg Quadrangle, Virginia, Montgomery Co., 7.5-Minute Series: U.S. Department of the Interior, Blacksburg, VA, scale 1:24,000. WYLLIE, D. C. AND MAH, C. W., 2004, Rock Slope Engineering: Civil and Mining, 4th ed.: Spon Press, New York, NY.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 97–111

111

Water Level Monitoring to Assess the Effectiveness of Stormwater Inﬁltration Trenches LAURA TORAN1 CATHERINE JEDRZEJCZYK Temple University, Department of Earth and Environmental Science, 1901 N 13th Street, Philadelphia, PA 19122

Key Terms: Infiltration Trenches, Stormwater Control Measures, Monitoring, Urban Geology ABSTRACT Successful control measures require proper design, maintenance, and monitoring to evaluate effectiveness. To study stormwater runoff mitigation techniques, a row of infiltration trenches with different designs was monitored for 2.5 years. The three trench designs included gravel-filled, gravel-filled with a leaf filter, and sandfilled trenches. Water level loggers in monitoring wells provided low-cost monitoring of effectiveness over time and differences between trench designs. In addition, infiltration experiments were conducted to deliver a controlled volume of water to each trench. The center gravel trench with leaf filter drained more slowly during experiments. The monitoring showed that the gravel trench tended to have the highest peaks and responded to more storms. These differences were related to uneven water delivery, based on field observations during storms, controlled experiments, and similarity in response for the other two trenches. The water level recession rate did not decrease over time, indicating that clogging was not a significant factor. Maintenance that consisted of removing fines captured on the top of the sand trench did not significantly change infiltration in that trench. Monitoring with water level loggers was an effective method of determining that uneven delivery of water occurred among the three trenches, but there was no decrease in effectiveness of the trenches over the observation period, conclusions that could not be drawn by visual inspection alone. INTRODUCTION Background Increased extent of impermeable surfaces due to land development leads to stormwater management problems. Paved roads, parking lots, rooftops, and com-

1 Corresponding

author email: ltoran@temple.edu

pacted soils all inhibit rainwater from infiltrating into the ground. The increased runoff in streams leads to higher discharge, erosion, and consequent changes in stream morphology (Booth et al., 2004; Walsh et al., 2005). In addition, stream ecosystems are impaired by temperature changes, pollutants brought in by storm runoff, and alteration in biogeochemical cycles (Paul and Meyer, 2001; Kaushal and Belt, 2012). Just 10 percent coverage by impervious surfaces causes harmful increases in runoff (Booth and Jackson, 1997). The impacts of urbanization on watershed hydrology need to be managed to limit further impairment. A variety of low-impact urban stormwater Best Management Practices have been developed as non–point source pollution and stormwater control measures (USEPA, 1999b). Some experts now prefer the term “stormwater control measures” (SCMs) (National Research Council, 2008). Infiltration SCMs are relatively small-scale structures that can be installed within existing construction to facilitate infiltration of runoff from rooftops or parking lots (Holman-Dodds et al., 2003). Infiltration trenches are practical in urban areas because they can be sized to fit available space, and in some cases runoff can be delivered to the trench via underground pipe (converting the trench to a small infiltration gallery). The size of the trench primarily depends on the size of the drainage area, the percentage of impervious surfaces within the drainage area, the infiltration rate of the surrounding soil, and the climate of the region. An infiltration trench consists of an excavated hole with vertical or gently sloping sides and a level bottom. The hole is lined with a geotextile filter layer and filled with stone with higher hydraulic conductivity than the surrounding soil. Most often the fill medium is gravel. During the excavation, care must be taken to avoid further compaction of the soil within the trench. In addition, sediment control measures must be implemented so that fine sediment from the construction site is not deposited into the trench, causing early clogging (USEPA, 1999a). Typically, runoff from impermeable surfaces is piped into the infiltration trench, but sometimes swales and berms are used to direct the water overland.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 113–123

113

Toran and Jedrzejczyk

Trenches should be monitored to evaluate whether they continue to function for stormwater capture and, in some designs, for water quality filtering. A number of factors can alter the predicted lifespan of an infiltration trench, including flawed prediction of the seasonal high-water table, errors in infiltration rate calculations, too much compaction and early sediment clogging from construction, poor upkeep of the structure, and lack of sediment removal by a SCM used in conjunction with infiltration (Livingston, 2000). However, once constructed, relatively little monitoring and evaluation of the effectiveness of these systems is conducted (National Research Council, 2008), and monitoring is typically not required by permitting rules. Clogging is a concern for both stormwater capture and water quality filtering. In some cases the trenches need to filter sediment, so to some extent, clogging is part of the design. Models are frequently used to estimate the amount of water than can be filtered before clogging occurs (Siriwardene et al., 2007; Freni et al., 2009). Models are also used to assess the appropriate size of trench for the drainage basin (Browne et al., 2008; Nimmer et al., 2009; and Campisano et al., 2011). Laboratory studies have been used to evaluate how filter media and sediments flowing into media affect clogging (Siriwardene et al., 2007; Kandra et al., 2014) in addition to evaluating the potential for water quality filtration (Fischer et al., 2003; Hatt et al., 2007). Significant clogging occurs at the interface between the geotextile filter and the surrounding soil (Siriwardene et al., 2007). Temperature-dependent infiltration has been observed, which indicates seasonal variation in performance is possible (Emerson and Traver, 2008). Some studies have evaluated infiltration out of the bottom versus the sides of trenches and found slower rates out of the bottom (Barraud et al., 2014); faster infiltration rates were attributed to higher horizontal permeability and higher contributing area of side walls in narrow trenches (Winston et al., 2016). Relatively few field studies have been conducted to examine the amount of time until trench performance decays, and those that have have shown widely differing longevity of trenches. Emerson et al. (2010) found that sedimentation and change in recession began immediately, and infiltration declined an order of magnitude in the first year. Their study trench was purposely under-designed (capturing more water than design recommendations) to increase failure rate. Brown and Borst (2014) found a decline in infiltration by approximately a factor of 5 in the first year on one end of an urban stormwater trench, but less change at the other end. The rapid clogging was in part attributed to the fine-grained materials used in construction. They also related differences in infiltration rates to heterogeneity of urban soil. Bergman et al.

114

(2011) conducted a long-term study of two infiltration basins in Copenhagen and found little change in infiltration after 3 years, but a reduction in hydraulic conductivity by a factor of 3.5 in one of the trenches after 15 years. Another study in Copenhagen by Warnaars et al. (2009) found measureable declines in hydraulic conductivity after 2 to 3 years; declines varied from 30 to 70 percent. Dechesne et al. (2005) looked at four basins in France aged 10–21 years and found little difference between trench infiltration rates and surrounding soil. Additional factors that impact trench performance and infiltration rates are storm size, antecedent conditions, and temperature (Lewellyn et al., 2015). Although design criteria typically focus on large storms, the study of Lewellyn et al. evaluated smaller storms to improve predictions of trench performance and longevity. The wide variation in performance points to the need for additional monitoring to improve stormwater management success. Potter (2006) highlights that small-scale stormwater control projects are particularly subject to uncertainty because of variations in local conditions. This study compared different types of material in a field application of infiltration trenches using low-cost monitoring techniques. Site Description The infiltration trenches in this study were on the grounds of the Pennypack Preserve of the Pennypack Ecological Restoration Trust (PERT) in Southeastern Pennsylvania. The Pennypack Preserve is located within the Upper Pennypack Creek Watershed. Land development has impaired the streams located in the Pennypack Creek watershed, with 79 percent considered to be impaired (Philadelphia Water Department, 2009). About 33 percent of the watershed has impervious surface, and development is primarily residential. A row of three infiltration trenches was installed on PERT property to provide an accessible demonstration of the local community and industry as well as an experimental study site. The infiltration trenches were constructed using three different designs to compare maintenance requirements and performance. The right and center trenches were filled with 5–10 cm–sized gravel. The designations “left trench” and “right trench” are determined by looking downslope (Figure 1). The center gravel trench inflow pipe was screened with a filter to catch leaves; the right gravel trench was unscreened. The leaf filter was intended to simulate maintenance to remove organic matter that could break down and clog the trench. Both sediment and organic matter can contribute to clogging (Gonzalez-Merchan et al., 2012). The purpose of leaving one gravel trench unfiltered was

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 113–123

Monitoring Inﬁltration Trenches

Figure 1. Photograph showing layout of piping to direct water to the three inﬁltration trenches. The tan sand can be seen in the foreground and gray gravel can be seen in the two more distant trenches. Monitoring wells in trenches are shown in the photograph before they were trimmed to the level of the ground surface. For scale, the blue pipes near the surfaces of the trenches are 10 cm in diameter, and the distance across the trenches is about 10 m.

to reproduce a scenario in which no removal of organic matter was performed on the trench. The left trench was filled with medium-grained sand and was not equipped with a leaf filter. Medium-grained sand was used to test whether fines would be captured at the top rather than the bottom of the trench. Thus, the three trench designs were intended to capture fines differently (Figure 2). The left sand trench was expected to capture fines on top because it is more difficult for fines to infiltrate the sand than the gravel. The two gravel trenches were expected to accumulate fines on the bottom because the fines move through the highly porous gravel and then accumulate where the finer-grained natural soil material occurs. The center gravel trench with leaf filter was designed to collect fewer fines because the leaf filter would prevent some material from reaching the trench. Each trench was 1.2 m deep, 1.5 m wide, and 3 m long (Figure 2). The bottom and sides of each of the trenches were lined with a geotextile filter fabric.

The geotextile lining prevents fines from moving upward into the sand or gravel but allows water to flow through. The three trenches were each outfitted with a separate pipe receiving stormwater from an upslope swale and directing it to flow onto the surface of the trench. The trenches were part of a treatment train, and this study examines variations in trench design rather than runoff capture design. The drainage area was approximately 3 acres (1.2 ha), consisting of lawns, wooded areas, and pavement, with an estimated 50 percent impervious surface. During rain events, stormwater runoff was first captured by an infiltration chamber beneath a parking lot on the PERT property. The infiltration trenches were downslope from this parking area and received overflow from an infiltration chamber there, which was not quantified. Water from the parking lot flowed down a driveway and was directed to a buried 8-inch polyvinyl chloride (PVC) pipe that moved stormwater beneath a hill slope onto the center of a swale above the trenches. From the swale, separate buried pipes directed water through a berm to each of the trenches (Figure 1). The outlets of these pipes were about 10 cm above the trench. The trenches were intended to increase infiltration to groundwater, which was about 4 m below the bottom of the trenches.

METHODS Monitoring Each trench was outﬁtted with a monitoring well (Figure 1) that housed a pressure transducer to measure the water level response to rain events. HOBO water level loggers were placed at the bottom of the trench and set to log at 15-minute intervals. A barometric pressure logger onsite was used to correct the transducers from pressure to water level. Data were recorded from December 2006 through June 2009. In between storms, the trenches were typically dry. The sand trench logger’s recording depth was changed in June 2007 after the monitoring well became

Figure 2. Schematic of clogging potential for (A) sand trench with clogging on top, (B) gravel trench with leaf ﬁlter reducing clogging by organic matter on bottom, and (C) gravel trench with clogging on bottom. (D) Photograph of trench excavation shows scale of trenches.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 113–123

115

Toran and Jedrzejczyk

clogged. The bottom of the well should have been plugged but was not, so sand moved up into the well and surrounded the transducer. As a result, the HOBO logger inside the well could not be removed to download data. The well was pulled out and then a monitoring well with bottom plug and side screen was driven back into the sand trench to a level 30 cm higher than the original setup (because of infilling of the hole during removal and reinstallation). Data were collected during the plugged phase. However, the baseline increased after the new well was installed. Only storm responses higher than 30 cm were recorded after June 2007. In addition to water level monitoring, a tipping-bucket rain gauge was installed on site to record precipitation at 15-minute intervals. Water level data from the trenches were plotted and compared over the 2.5-year period to look for changes in the hydraulic behavior of the three different trench designs. To assess storm response, the number of storms that created a response in each trench was recorded. The response was calculated as a percent of the total number of storms to compare periods with similar precipitation. For the purpose of data organization during this study, the term “Winter” refers to the period from January 1 to March 31. “Spring” is designated as April 1 to June 30, “Summer” is July 1 to September 30, and “Fall” is October 1 to December 31. In this study, the term “season” always refers to these periods, as opposed to standard calendar seasons. In addition, recession rate or drainage rate was approximated by assuming a straight-line slope between the peak and the return to baseline. Assuming linear recession is a common analysis technique (Warnaars et al., 1999; Emerson and Traver, 2008; Emerson et al., 2010; and Bergman et al., 2011) and is supported by the data presented in results. The rise in water level was divided by the time to recovery. Each storm greater than 0.5 cm was evaluated for response in each of the trenches. Based on observation of water levels in the trenches, smaller storms do not typically lead to sufficient overland flow to create a water level response in the trenches or the swale infiltrated the water before reaching the trench pipes. Storms that had multiple peaks were sometimes difficult to evaluate, because small increases in rain intensity can result in multiple water level peaks with short recession. These events were included when there was a sufficient water level increase (3 cm) to identify a recession, and the rainfall associated with that peak was treated as a separate event.

trenches were not hydrologically connected and to compare the trench response under identical conditions. Delivery of water to the trenches during natural storm events may be uneven and does not necessarily provide equal recharge to each trench. A garden hose outfitted with a flow splitter was used to control the delivery of water to the trenches. The splitter had four outlets to attach additional hoses, each with a valve to control the flow rate. A graduated cylinder and a timer were used to measure the rate of flow through the splitters to each trench. The first experiment was designed to assess interaction between the trenches. Water was added only to the center trench (gravel with leaf filter trench), and the water level was observed in the other trenches to determine whether the trenches were draining into each other. The second trench experiment was designed to provide even delivery of water to all three trenches. Water was delivered at approximately 100 ml/min ( ± 10 ml) to each trench with the flow splitter for 4 hours, and pressure was recorded on the transducers at a 5minute interval. When no rise in water level was observed in the sand trench after the 4 hours (because of the higher elevation of the logger), the valves to the two gravel trenches were shut off, and the water to the sand trench thus increased to 265 ml/min. The sand trench received the higher rate of flow for 45 minutes. The water levels were logged at 5-minute intervals during the experiments. The average drainage rate can be calculated for the infilling period during these experiments. The volume of water in the trench can be calculated from the peak water level multiplied by the area of the trench (1.5 × 3 m) with an assumed porosity of 30 percent (typical of unconsolidated sediment). This volume at peak saturation was subtracted from the volume of water injected to calculate the volume drained. If a constant drainage rate is assumed (an approximation for comparison purposes based on slow rise in water level at the end of injection), the infiltration rate is the difference in water volume divided by the infiltration time. For the sand trench the rate was calculated assuming that the trench filled to just below detection during the phase of the experiment when flow was to all three trenches or 0.3 m, based on the rapid water level rise during injection in that trench alone. The steady-state drainage rate for each of the trenches provides a comparison of infiltration under uniform delivery of water.

Controlled Inﬁltration Experiments

The use of sand to fill one of the trenches was intended to test whether fines would accumulate at the surface and result in clogging. If clogging occurred at the top, the layer of finer material could be scraped off

A series of experiments with controlled delivery of water were designed with two goals: to ensure that the

116

Sand Trench Scraping

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 113–123

Monitoring Inﬁltration Trenches

Figure 3. Photograph of core taken from sand trench 18 months after construction showing a layer of ﬁnes in the top 1 to 2 cm of dark sediment. Scale bar added is 0.5 m.

to return the trench to a higher infiltration rate. The surface of the sand trench was scraped on two occasions, March 2008 (18 months after construction) and May 2009 (another 15 months later). First, a core was taken to determine the depth of the fine layer (Figure 3). Then the top layer of fines was removed with a trowel. The scraped sediment was sieved to quantify grain size distribution, and hydraulic conductivity was calculated using the Hazen method. The layer below the fines (presumably original trench material) was also sieved. A Hazen coefficient of 60 to 80 was used based on the amount of fines in the sample. RESULTS AND DISCUSSION Controlled Flow Trench Experiments The experiment with flow directed into the center gravel trench with leaf filter was conducted to determine if the trenches were draining into each other. For this experiment, the water level rose 0.41 cm in the center gravel trench with leaf filter. The water level did not rise in the other two trenches. The volume of water added to the center gravel trench with leaf filter was above the sensor in the sand trench, even though it was 0.3 m above the base of the trench. The geotextile-lined earthen barrier between the trenches seems to inhibit flow. Thus, the response of each trench to water level increase can be interpreted independently of the adjacent trench. The purpose of adding water to all three trenches simultaneously was to observe the drainage rates of the trenches under controlled conditions with approximately the same flow rate into each trench (80–100 ml/s). The water level in the right gravel trench increased more gradually than the water level in the center trench (gravel-filled with leaf filter configuration), and the water level did not rise as high (Figure 4).

Figure 4. Water level increases observed from controlled experiment injecting about 100 ml/s into each trench. The water level in the sand trench can only be observed above 0.3 m because the sensor was raised above the base of the trench during repair of the well. The sand trench sensor did not show an increase until the water in the other trenches was shut off at 15:30 and all water was directed into the sand trench. The water level then rose immediately, suggesting the prior water level was just below the 0.3-m detection limit. Recession was linear, and the left sand and right gravel trenches reached steady state.

The porosity and water input rate was approximately the same in these two trenches, so it was expected that the water level rise would be the same. Since the water level in the right gravel trench did not increase as quickly or reach the same height, water must have been exiting faster from this trench than from the center gravel trench with leaf filter. After the water was turned off, the water level in the right gravel trench decreased rapidly. The gravel with filter trench took 3.5 hours longer to empty. Thus, the gravel with filter trench had a higher peak and stored water longer. This trench was approaching steady state at the end of the experiment, while the other two trenches reached steady state. The sand trench did not show a response for the first 4 hours of the experiment. The flow rate in the sand trench was increased after the flow to the other trenches was stopped. The water level rose above the 0.3-m baseline within 5 minutes, indicating that the sand trench was close to the detectable water level at the start of the experiment. The increase in flow rate resulted in a 0.2-m increase in water level after 45 minutes at the higher rate (or 0.5-m increase from base of the trench). The sand trench both filled and drained faster than the either of the gravel trenches. Both lower porosity in the sand and heterogeneity in the surrounding soil can contribute to this difference in rates. The three trenches had distinct responses to the controlled fill experiment. In particular, they did not drain at the same rate, based on both the slopes and the

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 113–123

117

Toran and Jedrzejczyk Table 1. Inﬁltration rates during experiments for rising limb of response (initial 240 minutes up to the peak response).

Gravel Gravel with ﬁlter Sand

Water Level at Peak (m)

Volume Injected (L)*

Volume at Peak (L)†

Amount Drained Drained in 240 min (L)

Drainage Rate (L/min)

0.14 0.36 0.3 (approximate)‡

1,129 1,143 1,454

188 493 391

941 650 1,063

3.9 2.7 4.4

Injection volume based on actual flow rates measured at flow splitter for each trench. Volume at peak is water level multiplied by the area of the trench bottom (1.5 × 3 m). ‡ For the sand trench the rate was calculated assuming that the trench filled to just below detection during the phase of the experiment when flow was to all three trenches, or 0.3 m, based on the rapid response of the trench once the inflow was increased. †

volume of water added and the observed peaks. Based on the steady-state drainage assumption described in the Methods section, the drainage rates were 2.7 L/min in the gravel with filter trench, 3.9 L/min in the gravel trench, and 4.4 L/min for the sand trench (Table 1). The two trenches on the ends of the row had higher drainage rates than the middle trench. The lower drainage rate in the middle trench was also observed after storms early in the observation period (Figure 5); therefore, the slower drainage is unlikely to be due to clogging over time. Because the lower rate was observed from the beginning of the available data collection period, the effect of leaf filter on reducing maintenance was difficult to evaluate. The middle trench could have a lower infiltration rate due to restricted horizontal drainage out the sides or due to smearing during construction. Based on the linear response of geometry and permeability on steadystate infiltration rates, some simple calculations can evaluate their relative influence. The two gravel trenches were compared since they contain the same material. The right gravel trench had 1.4 × higher infiltration.

If both trenches reached the same peak water level of 0.4 m, then the right gravel trench should have infiltrated 1.08 × faster because of horizontal flow out the right side, which was not available to the center gravel trench with leaf filter. The infiltration factor 1.08 reflects the area increase for outflow from the bottom and three sides for the right gravel trench and from the bottom and only two sides for the central gravel trench with leaf filter. The calculated factor is not sufficient to explain the observed rate. Furthermore, the right gravel trench did not peak as high. The difference in peaks creates less drainage area in the right gravel trench (0.15-m peak instead of 0.4 m). If the difference in peak water levels is accounted for, the steady-state drainage rate of the right gravel trench should have been 1.2 × slower than the center gravel trench with leaf filter, not faster. Thus, it is likely that lower permeability near the center gravel trench with leaf filter was a factor since the geometry of the infiltration area did not explain the observed difference. The permeability factor could be caused by spatial variability or smearing of side walls by the backhoe during construction. Seasonal Storm Response and Effects of Sand Trench Scraping

Figure 5. Water level increases observed in an early storm, before the sensor was raised in the sand trench. Data recorded at 15-minute intervals show multiple peaks as the rain starts and stops. The highest and most rapid response is observed in the gravel trench. A more gradual recovery is observed in the sand and gravel with ﬁlter trench. Recession was linear.

118

The seasonal plots and storm hydrographs typically showed that the right gravel trench experienced the highest water level peaks in response to storms, followed by the center gravel trench with leaf filter, and then the sand trench (Figures 5 through 7). A storm hydrograph from within the first year showed these distinct responses between the trenches (Figure 5). Although the sand trench sensor was positioned higher than the other two sensors after June 2007, the sand trench typically exhibited a response during storms that resulted in a water level increase above 30 cm in the gravel with filter trench (Figure 6). There are variations to this pattern of higher response in the gravel trench: for example, the gravel with filter trench had higher responses in late Spring 2009 (Figure 7). There were also periods during which storm response was nearly absent in the sand and center gravel trench with leaf filter

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 113–123

Monitoring Inﬁltration Trenches

Figure 6. Typical storm response showed higher water levels in the gravel trench (right side) and response in the sand trench only when the gravel with ﬁlter trench was at least 0.3 m. The baseline for the sand trench was 30 cm above the other trenches for these periods and does not indicate a water-ﬁlled trench.

(early Spring 2009, Figure 7). Similar observations are apparent for other seasons (Jedrzejczyk, 2010). The height of the peak water levels during storms differed from the controlled experiment. For the controlled experiment, the gravel trench had the lowest peak, which, based on calculations, was due to faster drainage. During storms, the gravel trench typically had a higher peak than the other two trenches. Comparing the two gravel trenches, the right trench had a higher response for 85 percent of storms (Figure 8). This higher storm response in the right gravel trench suggests there were differences in the amount of water delivered to each trench rather than differences in drainage rates.

Figure 7. Typical storm response is higher in the gravel trench on the right side, but in late spring the gravel with filter trench in the center had a higher response and in early spring there was no storm response in the sand or gravel with filter trench. These changes may reflect water delivery issues. The baseline for the sand trench was 30 cm above the other trenches for these periods and does not indicate a water-filled trench. Only storm responses greater than 30 cm are recorded in this trench.

Figure 8. Comparison of water level peaks for the center gravel trench with leaf ﬁlter and right gravel trench. The line shows equal peaks. The right gravel trench was higher 85 percent of the time (points above the line). Differences in peak height are related to uneven delivery of water and lower permeability near the center gravel trench with leaf ﬁlter.

Furthermore, in site visits during storms, two factors were observed that support uneven delivery of water to the trenches. First, water was observed to bypass the swale on the right side, where erosion of the berm occurred. Second, an additional storm pipe was uncovered on the hillslope during a large storm that preferentially delivered water to the center gravel trench with leaf filter. All pipes were supposed to be removed during construction, and this pipe was not visible during most of the monthly site visits. A smoke test in the pipe to reveal the source of water was unable to trace the origin. This pipe was later covered by hillslope erosion, but for the storms with higher water levels in the center gravel trench with leaf filter or left sand trench, this additional pipe could be a source of water. Thus, variations in peak height can be attributed to a combination of permeability differences and uneven water delivery. The number of precipitation events and number of water level responses was catalogued for each trench for 10 seasons (Figure 9). In general, precipitation events measuring less than 0.5 cm did not show a response. Not all storms show a response in each well, and the number of responses per season varies over time. Spring 2007 had the greatest percent storm response in each of the trenches. The lowest number of storm responses occurred during Winter 2009, when neither the gravel with filter nor the sand trench had any storm response. The gravel trench had the highest response rate, except for during Winter 2008, when it was equal to the gravel with filter trench rate, and Summer 2008, when the gravel with filter trench response rate was slightly higher. The sand trench had the lowest response rate. Although the sensor was raised at the end of Spring

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 113–123

119

Toran and Jedrzejczyk

Figure 9. Percent storm response shown for each season. The number of storms greater than 0.5 cm is given in parentheses after the season name. Both the gravel with ﬁlter and the sand trench showed a decline in response, but these responses were likely due to water delivery issues rather than clogging. The gravel trench on the right side varied both up and down. There was not a distinct change in the sand trench response after the logger was raised (Summer 2007) or after the top layer was scraped off (Spring 2008).

2007, the Summer 2007 season had a similar response rate, so it is not clear that the sensor elevation caused changes in the response rate. Furthermore, when the sand trench had a lower response, the gravel with filter trench nearly always had a lower response as well. The similarity in response for the center gravel trench with leaf filter and left sand trenches suggests that increased delivery of water to the right gravel trench caused the change. The response rate did not vary consistently from one season to another over the 2.5 years of the study (Figure 9). Spring had the highest response rates in 2007, but not in 2008 or 2009. The lowest response rate in the gravel trench was in Summer 2009, but in the sand and gravel with filter trench the lowest response rate was in Winter 2009. The top layer of the sand trench was scraped off two times during the study. The top 1 to 2 cm was removed and grain size analysis was conducted on this layer and the layer just below the surface. The top layer of sediment from March 2008 (18 months after construction) showed a hydraulic conductivity of 2.2 × 10−4 cm/s based on the Hazen method. The 1-cm layer below that had a hydraulic conductivity of 3.4 × 10−2 cm/s. Thus, the finer material accumulated at the surface of the sand and decreased the hydraulic conductivity by 2 orders of magnitude. From the May 2009 scraping, the top 1 cm of sediment was calculated to have a hydraulic conductivity of 9.8 × 10−3 cm/s. The hydraulic conductivity of the 3-cm layer below (from 1 to 4 cm below surface) was 3.5 × 10−2 cm/s. In 13 months, fine 120

Figure 10. Recession rates for the three trenches do not show a signiﬁcant decrease over time. The gravel trench experienced the highest rates, but the rates were similar at the beginning and end of the monitoring period. Relationship of recession to heads shown in Table 2.

sediment again accumulated at the surface of the sand trench, although the grain size and conductivity difference was not as great as during the previous sampling. However, the seasonal storm data did not show any distinguishable increase in response after scraping the sand trench (Figure 9), and the hydrograph response in Spring 2009 (Figure 7) showed an increase late in the season for both the sand and the gravel with filter trench. The laboratory estimation of hydraulic conductivity does not take into account macropores that can form in the field, which can allow infiltration to bypass the layer of fines. Any effect of removing the layer of fine sediment seems obscured by water delivery inconsistencies and macropores. Comparing responses of one particular season over different years provides a sense of whether responses are changing, because each season is likely to have similar antecedent conditions. If clogging was significant, the drainage rate should decrease and there should be a greater number of storm responses recorded. However, except for the gravel trench, the response rate decreased rather than increased over time (Figure 10). The higher response rate in the gravel trench over time was attributed to increased water delivery based on the controlled experiment, observed erosion of the berm on that side, and decreased responses in the sand and gravel with filter trenches. Storm Recession The recession rate, or drainage rate, of the trenches is the slope of the recession limb of the water level peaks. A decrease in drainage rate over time can be an indication of clogging. Analyses of the recession rates

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 113–123

Monitoring Inﬁltration Trenches Table 2. Regression between water level and storm recession rate. Trench Gravel Gravel with ﬁlter Sand

Maximum Rate (cm/min)

0.73 0.73 0.71

0.86 0.42 0.39

for the different trenches reveal that the gravel trench responded to the most storms, had the highest recession rate of the three trenches, and experienced the largest range of recession rate values (Figure 10). Within each trench, some of the variation in recession rates can be explained by different-sized storm events, which contributes to increased water level response. Regression between recession and water level showed a linear relationship, with an R2 of 0.7 for all three trenches (Table 2). As the height of the water level peak increases, the recession rate increases. This relationship can be explained in part by the change in the hydraulic gradient (difference in water level inside and outside the trench, since the water level rises faster inside the trench) and in part by the increase in trench wall area through which the water can drain. This linear relationship between head and recession would be less strong if clogging occurred to reduce drainage over time. The maximum recession rates (Table 2) are higher in the right gravel trench than in the left sand trench (0.86 and 0.39 cm/min, respectively). Percolation tests were performed prior to construction, which showed a similar spatial variability. On the surface just left of the trenches, the average infiltration rate was 1.3 cm/min, and on the right side it was 0.6 cm/min (LaBrake, 2010). The rates calculated by water level recession are 1.5 × lower than the pre-construction rates, but the difference could be due to soil compaction at depth rather than clogging over time. For recession over time, the data points are highly scattered, and there does not appear to be a significant change in the recession rate over time for any of the trenches (Figure 10). Linear regression fitting showed no significant trends, and high values were observed in the final year of monitoring for the gravel trenches. For example, the gravel trench experienced recession rates greater than 0.6 cm/s in both 2007 and 2009. In addition, the trenches experienced recession rate values below 0.1 cm/min both at the beginning and the end of the recorded period. Clearly, no progressive change to indicate clogging was detectable from the calculated recession rates. It should be noted that a lack of storm response in the gravel with filter and sand trenches between October 2008 and March 2009 limits the detectability of a trend. Furthermore, the data were not separated by storm type because uneven delivery of wa-

ter to the trenches affected the input for different storm events. CONCLUSIONS This study points out the benefits of infiltration trench monitoring to evaluate the performance of SCMs. The study site contained a row of trenches with varying media and maintenance practices. However, the performance of the infiltration trenches was influenced by factors outside of the initial infrastructure design. In this study, uneven delivery of water to the row of trenches, caused by erosion of a berm and periodic exposure of an old pipe, was a stronger influence on trench performance than was either media or maintenance. Differences in infiltration were observed between the trenches, both rate of infiltration and rate of response to storm events. First, the berm directing the water to the piping system eroded over time and more water was delivered to the right side. Thus, the right gravel trench had a higher response rate to storms. Second, the center gravel trench with leaf filter had a lower infiltration rate than the two side trenches. Based on steady-state infiltration calculations, the lower infiltration rate was primarily due to permeability differences and only partially due to restricted flow in the center of the row of trenches. Temporal changes in clogging were not directly observed in these trenches over the 2.5-year study. This conclusion is based on the scatter in recession rates observed in the three trenches, with both high and low values at the beginning and end of the monitoring period. Furthermore, the lower recession rate observed in the center gravel trench with leaf filter was observed both in initial storms and in a later controlled experiment when the same amount of water was delivered to all three trenches. Finally, no change in recession rate was observed in the sand trench after fines were scraped off the top. Understanding the delivery of water and performance of SCMs over time is important to improving SCM design. In this study, a simple monitoring network of a single well in each trench showed a higher response rate in the right gravel trench, indicating uneven delivery of water to the trenches. Monitoring showed that a piping system designed to split the flow of water was unable to deliver an even flow of water across a 9-m length of trenches. The wells also provided long-term monitoring to evaluate performance with and without maintenance. In controlled experiments the wells were used to observe different responses among the trenches under a measured flow rate. The controlled experiments more clearly established differences in infiltration rates in the two gravel trench responses when they both received the same amount of water. The slower rate of

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 113–123

121

Toran and Jedrzejczyk

infiltration in the center gravel trench with leaf filter was related to lower permeability, possibly as a result of smearing during construction, since low rates were observed early in the monitoring period. Without the water level loggers installed in the trenches, it would have been difficult to determine whether each of the trenches was receiving water, what the drainage rates were over time, and how the trenches responded to controlled experiments. ACKNOWLEDGMENTS This study was funded by the William Penn Foundation under a grant to Temple University’s Center for Sustainable Communities. We are grateful to Derron LaBrake, who designed the trench system, and to the Pennypack Ecological Trust, which provided access to the monitoring site. Both the William Penn Foundation and the Pennsylvania Department of Environmental Protection Grower Greener program provided funding for the trench installation. REFERENCES BARRAUD, S.; GONZALEZ-MERCHAN, C.; NASCIMENTO, N.; MOURA, P.; AND SILVA, A., 2014, A method for evaluating the evolution of clogging: Application to the Pampulha Campus infiltration system (Brazil): Water Science Technology, Vol. 69, No. 6, pp. 1241–1248. BERGMAN, M.; HEDEGAARD, M. R.; PETERSEN, M. F.; BINNING, P.; MARK, O.; AND MIKKELSEN, P. S., 2011, Evaluation of two stormwater infiltration trenches in central Copenhagen after 15 years of operation: Water Science Technology, Vol. 63, No. 10, pp. 2279–2286. BOOTH, D. B. AND JACKSON, C. R., 1997, Urbanization of aquatic systems: Degradation thresholds, stormwater detection, and the limits of mitigation: Journal American Water Resources Association, Vol. 33, No. 5, pp. 1077–1090. BOOTH, D. B.; KARR, J. R.; SCHAUMAN, S.; KONRAD, C. P.; MORLEY, S. A.; LARSON, M. G.; AND BURGES, S. J., 2004, Reviving urban streams: Land use, hydrology, biology, and human behavior: Journal American Water Resources Association, Vol. 40, No. 5, pp. 1351–1364. BROWN, R. A. AND BORST, M., 2014, Evaluation of surface and subsurface processes in permeable pavement infiltration trenches: Journal Hydrologic Engineering, Vol. 20, No. 2, http://dx.doi.org/10.1061/(ASCE)HE.1943-5584.0001016. BROWNE, D.; DELETIC, A.; MUDD, G. M.; AND FLETCHER, T. D., 2008, A new saturated/unsaturated model for stormwater infiltration systems: Hydrological Processes, Vol, 22, No. 25, pp. 4838–4849. CAMPISANO, A.; CREACO, E.; AND MODICA, C., 2011, A simplified approach for the design of infiltration trenches: Water Science Technology, Vol. 64, No. 6, pp. 1362–1367. DECHESNE, M.; BARRAUD, S.; AND BARDIN, J. P., 2005, Experimental assessment of stormwater infiltration basin evolution: Journal Environmental Engineering, Vol. 131, No. 7, pp. 1090–1098. EMERSON, C. H. and TRAVER, R. G., 2008, Multiyear and seasonal variation of infiltration from storm-water best management

122

practices: Journal Irrigation Drainage Engineering, Vol. 134, No. 5, pp. 598–605. EMERSON, C. H.; WADZUK, B. M.; AND TRAVER, R. G., 2010, Hydraulic evolution and total suspended solids capture of an infiltration trench: Hydrological Processes, Vol. 24, No. 8, pp. 1008–1014. FISCHER, D.; CHARLES, E.; AND BAEHR, A., 2003, Effects of stormwater infiltration on quality of groundwater beneath retention and detention basins: Journal Environmental Engineering, Vol. 129, No. 5, pp. 464–471. FRENI, G.; MANNINA, G.; AND VIVIANI, G., 2009, Stormwater infiltration trenches: A conceptual modeling approach: Water Science Technology, Vol. 60, No. 1, pp. 185–199. GONZALEZ-MERCHAN, C.; BARRAUD, S.; LE COUSTUMER, S.; AND FLETCHER, T., 2012, Monitoring of clogging evolution in the stormwater infiltration system and determinant factors: European Journal Environmental Civil Engineering, Vol. 16 (supplement 1), pp. s34–s47. HATT, B. E.; FLETCHER, T. D.; AND DELETIC, A., 2007, Treatment performance of gravel filter media: Implications for design and application of stormwater infiltration systems: Water Research, Vol. 41, No. 12, pp. 2513–2524. HOLMAN-DODDS, J. K.; BRADLEY, A. A.; AND POTTER, K. W., 2003, Evaluation of hydrologic benefits of infiltration-based urban storm water management: Journal American Water Resources Association, Vol. 39, pp. 205–215. JEDRZEJCZYK, C., 2010, Monitoring the Effectiveness of Stormwater Infiltration Trenches at the Pennypack Preserve, Montgomery County, Pennsylvania: Unpublished M.S. Thesis, Temple University, 81 p. KANDRA, H. S.; DELECTIC, A.; AND MCCARTHY, D., 2014, Assessment of impact of filter design variables on clogging in stormwater filters: Water Resources Management, Vol. 28, pp. 1873–1885. KAUSHAL, S. S. and BELT, K. T., 2012, The urban watershed continuum: Evolving spatial and temporal dimensions: Urban Ecosystems, Vol. 15, No. 2, pp. 409–435. LABRAKE, D., 2010, personal communication, Wetlands and Ecology, Inc., Havertown, PA. LEWELLYN, C.; LYONS, C. E.; TRAVER, R. G.; AND WADZUK, B. M., 2015, Evaluation of seasonal and large storm runoff volume capture of an infiltration green infrastructure system: Journal Hydrologic Engineering, Vol. 21, No. 1, http://dx.doi.org/ 10.1061/(ASCE)HE.1943-5584.0001256. LIVINGSTON, E. H., 2000, Lessons learned about successfully using infiltration practices. In National Conference on Tools for Urban Water Resource Management and Protection, Proceedings of Conference, February 7–10, 2000, Chicago, IL: U.S. Environmental Protection Agency, Cincinnati, OH: EPA/625/ R-00/001, pp. 81–96. NATIONAL RESEARCH COUNCIL, 2008, Urban Stormwater Management in the United States: Committee on Reducing Stormwater Discharge Contributions to Water Pollution: National Academies Press, Washington, DC, 598 p. NIMMER, M.; THOMPSON, A.; AND MISRA, D., 2009, Water table mounding beneath stormwater infiltration basins: Environmental Engineering Geoscience, Vol. 15, No. 2, pp. 67–79. PAUL, M. J. and MEYER, J. L., 2001, Streams in the urban landscape: Annual Review Ecology Systematics, Vol. 32, pp. 333–365. PHILADELPHIA WATER DEPARTMENT, 2009, Pennypack Creek Watershed Comprehensive Characterization Report: Philadelphia Water Department, Philadelphia, PA, 329 p. POTTER, K. W., 2006, Small-scale, spatially distributed water management practices: Implications for research in the hydrologic sciences: Water Resources Research, Vol. 42, p. 2.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 113–123

Monitoring Infiltration Trenches SIRIWARDENE, N. R.; DELETIC, A.; AND FLETCHER, T. D., 2007, Clogging of stormwater gravel infiltration systems and filters: Insights from a laboratory study: Water Research, Vol. 41, pp. 1433–1440. UNITED STATES ENVIRONMENTAL PROTECTION AGENCY (USEPA), 1999a, Infiltration Trench: Storm Water Technology Fact Sheet, EPA 832-F-99-019, 7 p. USEPA, 1999b, Preliminary Data Summary of Urban Stormwater Best Management Practices: EPA-821-R-99-012, 216 p. WALSH, C. J.; FLETCHER, T. D.; AND LADSON, A. R., 2005, Stream restoration in urban catchments through redesigning stormwa-

ter systems: Looking to the catchment to save the stream: Journal North American Benthological Society, Vol. 24, No. 3, pp. 690–705. WARNAARS, E.; LARSEN, A. V.; JACOBSEN, P.; AND MIKKELSEN, P. S., 1999, Hydrologic behavior of inﬁltration trenches in a central urban area during 2 34 years of operation: Water Science Technology, Vol. 39, No. 2, pp. 217–224. WINSTON, R. J.; DORSEY, J. D.; AND HUNT, W. F., 2016, Quantifying volume reduction and peak ﬂow mitigation for three bioretention cells in clay soils in northeast Ohio: Science Total Environment, Vol. 553, pp. 83–95.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 113–123

123

A Durability Classiﬁcation of Clay-Bearing Rocks Based on Particle Size Distribution of Slaked Material TEJ P. GAUTAM Department of Petroleum Engineering and Geology, Marietta College, Marietta, OH 45750

ABDUL SHAKOOR1 Department of Geology, Kent State University, Kent, OH 44242

Key Terms: Durability Classification, Clay-Bearing Rocks, Disintegration Behavior, Slake Durability Index, Disintegration Ratio ABSTRACT Clay-bearing rocks disintegrate at varying rates, due to slaking, when exposed to moisture. This research aims to develop a durability classification of clay-bearing rocks based on particle size distribution of slaked material, quantified in terms of disintegration ratio (DR ). DR is the ratio of the area under the particle size distribution curve of slaked material for a sample, upon completion of the standardized slake durability index test, to the total area encompassing all particle size distribution curves of the samples tested. Although second-cycle slake durability index (Id2 ) is the most frequently used parameter for classifying the slaking behavior of clay-bearing rocks, it does not consider the range of particle sizes in the slaked material after the test. Compared to Id2 , DR accounts for all particle sizes present in the slaked material and provides a better measure of the degree of slaking. The slake durability index test was used to investigate the slaking behavior of samples from 20 different clay-bearing rocks, and Id2 and DR values were determined for all samples. Id2 showed a nonlinear relationship with DR . The relationship was used to develop a durability classification, based on disintegration ratio, as follows: low durability: DR = 0 to 0.20; medium durability: DR = 0.20 to 0.65; medium-high durability: DR = 0.65 to 0.85; and high durability: DR = 0.85 to 1.00. In order to use this classification, one needs to perform a sieve analysis on material left after the slake durability test and determine DR . INTRODUCTION Clay-bearing rocks, including claystones, mudstones, siltstones, and shales, are frequently encountered in engineering and environmental projects because of their widespread occurrence on the land surface (Potter 1 Corresponding

author.

et al., 1980; Franklin, 1981; Blatt, 1982; and Dick and Shakoor, 1992). These rocks disintegrate and crumble at varying rates when exposed to moisture; this process is known as “slaking.” Most of the previous research on assessment and classification of slaking behavior of clay-bearing rocks is based on the slake durability index test or some simplified/modified version of it (Franklin and Chandra, 1972; Wood and Deo, 1975; Chapman et al., 1976; Olivier, 1979; Franklin, 1981; Dick and Shakoor, 1992; Dearman, 1995; Moon and Beattie, 1995; Santi, 1998; Koncagul and Santi, 1999; and Molina et al., 2011). The slake durability test was proposed initially by Franklin and Chandra (1972) and was later standardized by both the American Society for Testing and Materials (ASTM) and the International Society for Rock Mechanics (ISRM). The test procedure (ASTM D 4644, ASTM 2013; ISRM 2007) uses 10 oven-dried pieces of rock, each weighing 40–60 g, with a total weight of 450–550 g. The oven-dried sample is placed in a test drum of 2-mm mesh (#10 sieve) and the drum is rotated in a tank of water for 10 minutes at a rate of 20 rotations per minute. At the end of the test, the sample is removed from the drum, oven-dried, and weighed to determine the slake durability index (Id), defined as follows: Id = [(oven-dried weight after the test/initial oven-dried weight) (100)]. Usually two cycles are run, and the second-cycle slake durability index (Id2 ) is used for evaluating and classifying the durability of a rock. A photograph of the slaked material, retained in the drum, is taken (Figure 1), and the slaking behavior is described qualitatively as belonging to one of three types: Type I—retained specimen remains virtually unchanged; Type II—retained specimen consists of large and small fragments; or Type III—retained specimen consists exclusively of small fragments. Previously Developed Durability Classifications A series of durability classifications for clay-bearing rocks has been developed over the past few decades. The more notable among these are as follows.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 125–135

125

Gautam and Shakoor

Figure 1. Types of material retained in drum after slaking. Type I— retained specimen remains virtually unchanged; Type II—retained specimen consists of large and small fragments; and Type III— retained specimen consists exclusively of small fragments.

Gamble (1971) proposed a durability-plasticity classiﬁcation based on Id2 and plasticity index (PI) values. The classes include very low durability— Id2 = 0–30 percent; low durability—Id2 = 30– 126

60 percent; medium durability—Id2 = 60–85 percent; medium-high durability—Id2 = 85–95 percent; high durability—Id2 = 95–98 percent; and very high durability—Id2 = 98–100 percent. The PI is divided into three categories: low PI (0–10), medium PI (10– 25), and high PI (>25). Olivier (1979) suggested a geodurability classification based on the ratio of unconfined compressive strength and swelling coefficient. The swelling coefficient is measured by saturating an oven-dried core sample (dried at 105◦ C for at least 12 hours) for a minimum period of 12 hours and then noting the increase in axial dimension. The free swelling coefficient is defined as follows: change in sample length/initial sample length. Both unconfined compressive strength and swelling coefficient reflect the lithological composition of the rock. Based on the ratio of unconfined compressive strength and swelling index, Olivier (1979) categorized durability into six classes, A through F, as follows: A—excellent, ratio >50; B—good, ratio 25–50; C—fair, ratio 12–25; D—moderately poor, ratio 6–12; E—poor, ratio 3 to 6; and F—very poor, ratio <3. Grainger (1984) classified clay-bearing rocks into durable and nondurable on the basis of compressive strength and slake durability index, with durable rocks having compressive strengths of >3.6 MN/m2 and Id2 values of >90 percent and nondurable rocks having compressive strength from 0.6 to 3.6 MN/m2 and Id2 values of <90 percent. He also emphasized the roles of particle size, mineralogical composition, microfabric, anisotropy, and tectonic deformation in influencing the durability of clay-bearing rocks. Dick et al. (1994) proposed three classes of durability for clay-bearing rocks, based on Id2 values, and related each class to the relevant engineering properties, such as slickensides, microfracture frequency index (Imf ), and absorption (Table 1). Imf was determined as the number of microfractures per unit length of a traverse across the cut surface of a rock sample, the minimum traverse length being 5 cm. According to this classification, the Id2 boundaries between the three classes are as follows: high durability—Id2 > 85 percent; mediumdurability—Id2 = 50–85 percent; and low durability— Id2 < 50 percent (Table 1). Santi (1998) modified the jar slake test, proposed by Wood and Deo (1975), to classify the durability behavior of clay-bearing rocks. The modified version of the test consists of oven-drying a 30–50-g sample at 110◦ C for 16 hours, cooling it for 20 minutes, immersing it in water, and observing the disintegration behavior. Careful observations are made during the first 30 minutes and periodically thereafter until the completion of 24 hours to monitor the nature and rate of disintegration. After 24 hours, a final observation

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 125–135

Classifying Clay-Bearing Rocks Table 1. Durability classiﬁcation of clay-bearing rocks proposed by Dick et al. (1994).

Durability Class High: Id2 > 85% Medium: 50 < Id2 < 85% Low: Id2 < 50%

Claystone Slickensided

Mudstone Imf (Microfractures/cm)

Shale Absorption (%)

Combined Siltstone-Siltshale Absorption (%)

NA NA All

<0.4 0.8–0.4 >0.8

<5.5 10–5.5 >10

<5 9.5–5 >9.5

NA = not applicable

is made. Based on the nature of slaked material after 24 hours, Santi (1998) categorized the disintegration behavior into six classes: mud, flakes, chips, fractured (chunky), slabs, and no reaction. The classification provides only a qualitative description of the disintegrated material. Sadisun et al. (2005) used a modified slake durability index test and proposed a durability classification based on visual changes observed during the slaking process and on quantitative measurements made on the samples. The modification consisted of using cubical samples, with 4- to 6-cm side dimensions, and monitoring the transition of incipient discontinuities into open cracks. Sadisun et al. (2005) calculated a modified slaking index as the ratio of the weight of loosened sample to initial weight of the oven-dried sample, expressed as a percentage. Based on their modified slaking index, they categorized clay-bearing rocks into six durability classes: very low durability—slaking index = 0 to 2 percent; low durability—slaking index = 2–10 percent; medium durability—slaking index = 10–25 percent; high durability—slaking index = 25–50 percent; very high durability—slaking index = 50–85 percent; and extremely high durability—slaking index = 85–100 percent. Bryson et al. (2012) proposed using swelling potential and weathering susceptibility for durability classification. They suggested that the slake durability index is not an adequate parameter with which to characterize the durability of clay-bearing rocks. Limitations of Previous Classifications The durability classifications described above, based on the slake durability index test or its modifications, have some serious limitations. The procedure for the slake durability index test has remained essentially unchanged since its introduction. Many researchers have demonstrated the shortcomings of the test, and some have attempted to improve the slake durability index and jar slake tests (Hopkins and Deen, 1983; Richardson, 1985; Taylor and Smith, 1986; Bell et al., 1997; Santi, 1998; Gokceoglu et al., 2000; Czerewko and Cripps, 2001; Marques et al., 2005; Sadisun et al., 2005; Singh et al., 2005; Nickmann et al., 2006; Erguler and Ulusay, 2009; Moradian et al., 2010; Youn and

Tonon, 2010; Miˇscˇ ević and Vlastelica, 2011; Molina et al., 2011; Admassu et al., 2012; and Corominas et al., 2015). These researchers emphasize the need for developing new indices for classifying the degree of slaking (disintegration). Some recommend using different number of cycles, ranging from 1 to 5, to better categorize the slaking behavior. According to Bell et al. (1997), the slake durability index test lacks a sensitivity for the upper range of durability values and overestimates the durability of clay-bearing rocks in this range. Bell et al. (1997) suggested using the slake durability test in combination with uniaxial compressive strength to characterize the durability of clay-bearing rocks, and they recommended using the third-cycle test results. Crosta (1998) developed an ultrasonic bath method for predicting the durability index. The method is useful for fragile rocks with low to medium durability but has shortcomings for rocks with high durability. Santi (1998) investigated the weaknesses of the slake durability index and jar slake tests and proposed modifications. He recommended using the first-cycle slake durability index to characterize the slaking behavior of a rock sample. Czerewko and Cripps (2001) stated that as a result of the dynamic nature of the slake durability test, it is very aggressive in breaking down the low-durability materials. An additional problem with respect to sensitivity arises in distinguishing between slaked and non-slaked material using the arbitrarily selected 2-mm mesh size. Czerewko and Cripps (2001) found the results from the third cycle of slake durability test to be more reliable than the second-cycle results specified by ASTM. Further, for greater reliability, they recommended using a combination of the jar slake and slake durability index tests to evaluate the slaking behavior of a wide variety of clay-bearing rocks. Martinez-Bofill et al. (2004) suggested using fourthor fifth-cycle slake durability index values for argillaceous rocks and stated that the disintegration rate of rocks experiencing freezing-thawing is faster than that of those exposed to warmer climatic conditions. Similarly, Yagiz et al. (2012) recommended using the fourthcycle slake durability index to predict rock properties like uniaxial compressive strength and modulus of elasticity.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 125–135

127

Gautam and Shakoor Table 2. Formation names, geologic ages, and site locations for the sampled rocks. Location No.

Sample

Formation

Age

State

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

CST-1 CST-2 CST-3 CST-4 CST-5 MST-1 MST-2 MST-3 MST-4 MST-5 SLT-1 SLT-2 SLT-3 SLT-4 SLT-5 SHL-1 SHL-2 SHL-3

Conemaugh Formation Dakota Group Mowery Shale Conemaugh Fruitland Formation Conemaugh Monogahela Red Pine Shale Dunkard Dunkard Conemaugh Monongahela Conemaugh Wasatch Green River Milboro Shale Conemaugh Tradewater

Penn Cret Cret Penn Cret Penn Penn Prot Perm Perm Penn Penn Penn Eoce Eoce Devo Penn Penn

West Virginia Kansas South Dakota Pennsylvania Colorado Ohio Ohio Utah Ohio Ohio Ohio Ohio West Virginia Colorado Wyoming Virginia Ohio Kentucky

19 20

SHL-4 SHL-5

Straight Cliffs Rome

Cret Camb

Utah Tennessee

Another important consideration in evaluating the disintegration behavior of clay-bearing rocks is that disintegration behavior can be different between laboratory and natural climatic conditions. Under natural conditions, rocks are exposed to not only wetting and drying cycles, as in the laboratory tests, but also to heating and cooling and freezing and thawing cycles. Additionally, anthropogenic activities, surface erosion, and gravity action can remove the weathered material from the slope, further complicating the assessment of disintegration behavior. To address this issue, some researchers have attempted to assess rock durability under field conditions or by simulating natural climatic conditions (Bell et al., 1997; Exadaktylos, 2006; Binal, 2009; Erguler and Ulusay, 2009; Erguler and Shakoor, 2009a, 2009b; Gautam and Shakoor, 2013, 2016; and Rincon et al., 2016). Some authors have suggested using P-wave velocities to predict slake durability index (Sharma and Singh, 2008; Moradian, 2010; and Khandelwal, 2013). STUDY OBJECTIVE Despite the extensive research conducted previously, characterizing the slaking behavior of clay-bearing rocks and developing a widely applicable durability classification system remains a challenge. The objective of this study was to develop a durability classification for clay-bearing rocks that takes into account the particle size distribution of the slaked material, expressed in

128

Road I-77 S SR-159 S I-90 Dam site SR-139 N SR-50 S I-77 N SR-191 S SR-7 S SR-260 I-77 N SR-22 E I-77 S I-70 E I-80 W SR-11 N SR-75 SR William H. Natcher Parkway SR-10 SR-381 N

Site Ripley/Fairplan Exit 700 miles S of I-70 1 mile N of Oacoma Point Marion 27 miles N of Loma Carﬁeld Mile marker 23 1 mile to Dexter City Exit 1 mile E of Flaming Gorge Dam 1 mile off Loganport Exit 3 miles from SR-7 S 2 miles N of Dexter City Exit 13 miles W of SR-7 Ramp of Eden Fork Exit Mile marker 156 Mile marker 216 2 miles from Radford 15 miles S of Loganport Mile marker 52 8 miles S of Price Town Bristol Motor Speedway

terms of disintegration ratio (DR ). We believe the proposed classification, based on the standard slake durability test (ASTM D 4644; ASTM, 2013) and particle size distribution of slaked material, alleviates the shortcomings of the previously proposed classifications. RESEARCH METHODS Sampling and Laboratory Testing Twenty clay-bearing rocks, including five claystones, five mudstones, five siltstones, and five shales, from 11 different states were sampled for the study. Table 2 provides information about site locations, rock formation names, and rock formation ages. The classification system proposed by Potter et al. (1980) and modified by Dick and Shakoor (1992) was used to classify the claybearing rocks into claystones, mudstones, siltstones, and shales. In the laboratory, the slake durability index test, involving five cycles, was performed on samples from all 20 rock types. At the end of each cycle, oven-dried weight of the sample retained in the test drum was recorded and particle size distribution of the slaked material was determined using the following sieve sizes: 50.8 mm (2 in.), 25.4 mm (1 in.), 19.0 mm (3/4 in.), 12.5 mm (1/2 in.), 4.75 mm (#4), and 2.0 mm (#10). The weight of the slaked material was used to determine slake durability index after each cycle, and particle size distribution was used to determine disintegration ratio.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 125–135

Classifying Clay-Bearing Rocks

Figure 2. Procedure for determining the disintegration ratios for various rock types from the particle size distribution curves of slaked material (after Erguler and Shakoor [2009b]).

Additionally, we determined clay content, absorption, and adsorption values for all 20 rocks. Details of these tests are available in Gautam and Shakoor (2013). Quantifying the Amount of Slaking We used the slake durability index, as determined by ASTM method D 4644 (ASTM, 2013), and disintegration ratio (DR ), as proposed by Erguler and Shakoor (2009b), to quantify the amount of slaking. DR is deďŹ ned as the ratio of the area under the particle size distribution curve of the slaked material for a given sample to the total area encompassing particle size distribution curves of all samples tested. Figure 2 illustrates the procedure for determining DR. For example, DR for claystone (5) in Figure 2 is obtained by dividing

the area under the particle size distribution curve for claystone (bceg) by the total area (abcd), giving a DR value of 0.191. Similarly, the area under the particle size distribution curve for siltstone (3) (bcdj) divided by the total area (abcd) gives a DR value of 0.990 for the siltstone. Note that the total area in Figure 2 (abcd) is taken as the area of the rectangle formed between 0 percent to 100 percent retained by weight along the vertical axis and the smallest to the largest sieve sizes (i.e., mesh sizes) along the horizontal axis. We used sieves ranging in mesh size from 2 mm (same as the mesh size of the drum in the slake durability test) to 50.8 mm (2-in. sieve) to determine particle size distribution of slaked material from all samples. However, none of the samples had any slaked material retained on 50.8-mm sieve, and all particle size distribution curves

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 125â&#x20AC;&#x201C;135

129

130

Id1

93.8 59.7 89.3 91.8 33.2 95.8 18.4 88.7 97.5 95.7 98.1 98.2 98.6 99.6 95.7 99.3 98.1 64.7 98.6 98.6

Sample No.

CST-1 CST-2 CST-3 CST-4 CST-5 MST-1 MST-2 MST-3 MST-4 MST-5 SLT-1 SLT-2 SLT-3 SLT-4 SLT-5 SHL-1 SHL-2 SHL-3 SHL-4 SHL-5

80.1 43.3 76.3 64.0 1.5 93.3 3.4 73.2 95.2 75.7 96.7 99.2 95.0 99.4 86.9 99.0 95.8 28.0 95.8 97.7

Id2 72.7 0.6 58.8 49.6 0.1 90.9 2.0 67.0 93.4 63.5 95.7 98.3 93.9 99.2 82.2 98.7 94.2 19.0 93.0 97.0

Id3 65.2 0.0 45.2 42.5 — 88.1 1.5 61.8 91.2 53.0 94.5 97.2 92.9 98.9 76.9 98.5 92.5 13.8 90.2 96.1

Id4

Slake Durability Index (Id) (%) After 1 to 5 Cycles

59.8 — 36.6 35.6 — 86.3 1.4 58.5 90.0 46.6 93.7 96.7 92.3 99.0 72.5 98.5 91.8 9.8 88.1 95.7

Id5 0.625 0.417 0.895 0.752 0.191 0.920 0.022 0.652 0.956 0.600 0.963 0.974 0.990 0.996 0.829 0.987 0.849 0.536 0.954 0.979

1 0.337 0.009 0.747 0.525 0.003 0.795 0.003 0.315 0.868 0.328 0.946 0.973 0.926 0.990 0.397 0.974 0.685 0.096 0.838 0.932

2 0.229 0.004 0.546 0.405 0.000 0.777 0.002 0.269 0.826 0.199 0.933 0.966 0.918 0.989 0.255 0.968 0.614 0.029 0.817 0.899

3 0.153 — 0.393 0.347 — 0.748 0.001 0.214 0.809 0.122 0.922 0.956 0.908 0.987 0.185 0.964 0.573 0.015 0.790 0.852

Disintegration Ratio of Lab Samples After 1 to 5 Cycles

0.031 — 0.326 0.301 — 0.727 0.001 0.190 0.796 0.092 0.913 0.937 0.840 0.986 0.031 0.960 0.538 0.010 0.729 0.773

5 41 77 38 38 45 22.2 20.3 29 27 28 17 21 17 12 18 19 16 28 25 18

<0.002 mm 53 82 52 51 57 34.3 47.2 33 35 36 22 27 22 22 22 24 19 36 33 24

<0.004 mm

Clay Content in Bulk (%)

44 62.2 45.8 14.1 66.3 5.1 31.6 10.8 7.5 36.7 7.7 3.8 3.1 1.9 8.8 4.2 5.3 24.5 14.6 3.2

Absorption (%)

Table 3. Slake durability index, disintegration ratio, clay content, absorption, and adsorption values for the 20 rocks tested.

3.9 10.0 5.2 2.5 5.6 — — 1.7 3.0 3.5 1.9 1.4 1.7 1.2 2.6 1.7 2.1 3.5 3.3 1.3

Adsorption (%)

Gautam and Shakoor

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 125–135

Classifying Clay-Bearing Rocks

fell within an area bounded by 2.0-mm and 25.4-mm sieves. Thus, the total area, abcd, in Figure 2 depends on the range of sieves used (2.0 mm to 25.4 mm) for particle size distribution and remains unchanged from sample to sample. Figure 2 shows that a DR value of 1 indicates a completely durable rock (no slaking; Id2 = 100 percent), and a DR value of 0 indicates a completely non-durable rock (no particles of >2 mm left after the slaking process; Id2 = 0 percent). Therefore, DR values of 0.191 and 0.990 for claystone (5) and siltstone (3), respectively, indicate that claystone has a very low durability and siltstone has a very high durability.

Table 4. R2 values for the relationship between slake durability index and disintegration ratio (DR ) determined after each of the ﬁve cycles of slake durability test.

DATA ANALYSIS

Table 4 shows the coefficient of determination (R2 ) values for the relationship between slake durability index and DR data for all five cycles of the slake durability test. It is clear from the table that the relationship between the two parameters is stronger for the first- and second-cycle data and becomes weaker with an increasing number of cycles. Therefore, the second-cycle data, as recommended by ASTM (ASTM, 2013), are considered to be most appropriate for developing a durability classification.

Table 3 presents the slake durability index, disintegration ratio, clay content, absorption, and adsorption data for the laboratory samples. The results in Table 3 show that clay-bearing rocks are highly variable even within the same group (claystones, mudstones, etc.). This variability of slaking behavior is due to variation in the clay content, absorption, and adsorption values of different samples (Shakoor and Gautam, 2015). Figure 3 shows that the relationship between slake durability index (Id2 ) and disintegration ratio (DR ) data, after the second-cycle test, is exponential in nature. The exponential nature of the relationship suggests that DR is very sensitive to particle size distribution of the slaked material at Id2 values exceeding 50 percent. In other words, DR is a better parameter for characterizing the degree of disintegration than is Id2 because even a highly disintegrated material can have a high Id2 value as long as the disintegrated particles are slightly larger than 2 mm in size (mesh size of the test drum).

Figure 3. Relationship between slake durability index and disintegration ratio after the second-cycle test.

Slake Durability Index (Id) Values for 1 to 5 Cycles

DR DR DR DR DR

after ﬁrst cycle after second cycle after third cycle after fourth cycle after ﬁfth cycle

Id1

Id2

Id3

Id4

Id5

0.85 0.93 0.88 0.87 0.7

0.7 0.88 0.88 0.93 0.79

0.57 0.85 0.83 0.86 0.76

0.47 0.76 0.83 0.81 0.73

0.46 0.62 0.72 0.76 0.7

DR -BASED DURABILITY CLASSIFICATION Based on the relationship between Id2 and DR (determined after the second-cycle test), presented in Figure 3, we propose the durability classification shown in Figure 4. Note that the R2 for the relationship in Figure 4 is improved to 0.92, compared to 0.88 in Figure 3, as a result of omission of the two outliers. The two outliers were omitted because they do not represent the well-defined trend noted in the other 18 samples. It is not clear why these two samples deviate from the trend exhibited by the other samples. The proposed classification has four classes: low durability—DR = 0 to 0.20; medium durability—DR = 0.20 to 0.65; medium-high durability—DR = 0.65 to 0.85; and high durability— DR = 0.85 to 1.00. A t-test was used to verify that the boundaries between different durability classes belonged to different populations (Gautam, 2012). Furthermore, clay content, absorption, and adsorption properties were considered for different classes to ascertain that the boundaries represented not only the durability characteristics of the proposed classes but also the differences in other geologic characteristics, as shown in Table 3. Table 5 relates the proposed durability classes to clay content, absorption, and adsorption values of the rocks tested as well as the samples belonging to each class. Clay content, absorption, and adsorption exhibit stronger correlations with DR than with Id2 , as indicated by multiple regression and principal component analyses (Shakoor and Gautam, 2015).

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 125–135

131

Gautam and Shakoor

Figure 4. Proposed durability classiﬁcation for clay-bearing rocks based on disintegration ratio.

and because the second-cycle slake durability index test is the most commonly performed durability test for clay-bearing rocks in engineering practice. Although the DR -based classiﬁcation proposed in this study uses the relationship between DR and Id2 , we believe it is an improvement over previous classiﬁcations because DR is a more meaningful parameter than Id2 for differentiating between different classes of durability (Erguler and Shakoor, 2009a, 2009b; Gautam and Shakoor, 2015). For example, in Figure 5, Id2 values for claystone 1 and claystone 3 samples are nearly the same (80 percent and 76 percent, respectively), but their degrees of disintegration, as indicated by the range of particle sizes, are very different.

A comparison of the samples with the proposed durability classes shows that mudstones are the most variable in durability. Mudstones exhibiting lower durability are the ones that contain microfractures (Dick and Shakoor, 1992). Table 5 also provides tentative ranges of Id2 values corresponding to the proposed durability classes. Table 6 compares the classification proposed in this study with the classifications proposed by Gamble (1971) and Dick et al. (1994). The classification shown in Figure 4 encompasses a range of DR values, representing a wide variety of clay-bearing rocks. Laboratory values of DR are used to classify durability of clay-bearing rocks because DR can be determined easily from laboratory test results

Table 5. Durability classiﬁcation of clay-bearing rocks based on disintegration ratio in relation to clay content, absorption, and adsorption. Durability Class

Second-Cycle Slake Durability Index (%)

Clay Content (<0.002 mm, %)

Absorption (%)

Adsorption (%)

Corresponding Samples from This Study

0–0.20

0–71

>40

>25

Medium

0.20–0.65

71–91

25–40

10–25

3–5

Medium-high

0.65–0.85

91–96

15–25

5–10

High

0.85–1.00

96–100

<15

CST-2, CST-5 MST-2 SHL-3 CST-1 MST- 3, MST-5 SLT-5 MST-1 SHL- 2, SHL-4 MST-4 SLT-1 to SLT-4 SHL- 1, SHL-5

Low

132

Disintegration Ratio

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 125–135

Classifying Clay-Bearing Rocks Table 6. Comparison of the proposed classification with the durability classifications of Gamble and Dick et al. Proposed DR -Based Classification

Gamble Classiﬁcation (1971)

Dick et al. Classiﬁcation (1994)

Low Medium Medium-high High

Low to medium Medium to medium-high Medium high High to very high

Low to medium Medium to high High High

DR = disintegration ratio.

However, DR values for claystone 1 and claystone 3 are 0.337 and 0.747, respectively, indicating that DR is better able to differentiate between different degrees of disintegration (i.e., different levels of durability). According to Gamble’s (1971) classification, recommended by ISRM, both samples belong to the medium durability class. According to the DR -based classification proposed herein (Table 5), claystone 1 has medium durability, and claystone 3 (an outlier) has mediumhigh durability. DR also correlates better with other engineering properties of clay-bearing rocks (Shakoor and Gautam, 2015) than does Id2 (Dick and Shakoor, 1972; Russell, 1981; Bell et al., 1997; Czerewko and Cripps, 2001; Sadisun et al., 2005; Santi, 2006; Hajdarwish et al., 2013; Corominas et al., 2015; and Heidari et al., 2015). Additionally, DR is a better indicator of the disintegration behavior of clay-bearing rocks under natural climatic conditions that include heating and cooling and freezing and thawing cycles in addition to wetting and drying cycles (Gautam and Shakoor, 2013). CONCLUSIONS The following conclusions can be drawn from this study: 1. DR is a more meaningful parameter for durability classification of clay-bearing rocks than is Id2 . 2. Based on DR , clay-bearing rocks can be classified into four durability classes: low durability (DR = 0 to 0.20), medium durability (DR = 0.20 to 0.65), medium-high durability (DR = 0.65 to 0.85), and high durability (DR = 0.85 to 1.00). REFERENCES

Figure 5. Slake durability index values (Id2 ) for claystone 1 and claystone 3 samples are nearly the same (80 percent and 76 percent, respectively), whereas their disintegration ratio (DR ) values are signiﬁcantly different (0.337 and 0.747, respectively), indicating that DR is better able to differentiate between different degrees of durability.

ADMASSU, Y.; SHAKOOR, A.; AND WELLS, N. A., 2012, Evaluating selected factors affecting the depth of undercutting in rocks subject to differential weathering: Engineering Geology, Vol. 124, No. 4, pp. 1–11. AMERICAN SOCIETY FOR TESTING AND MATERIALS (ASTM), 2013, Annual Book of ASTM Standards 04.08: American Society for Testing Materials, West Conshohocken, PA, 1824 p. BELL, F. G.; ENTWISLE, D. C.; AND CULSHAW, M. G., 1997, A geotechnical survey of some British Coal Measures mudstones, with particular emphasis on durability: Engineering Geology, Vol. 46, pp. 115–129. BINAL, A., 2009, A new laboratory rock test based on freeze-thaw using a steel chamber: Quarterly Journal Engineering Geology Hydrogeology, Vol. 42, No. 2, pp. 179–198. BLATT, P. J., 1982, Sedimentary Petrology: W. H. Freeman and Company, San Francisco, CA, 564 p. BRYSON, L. S.; GOMEZ-GUTIERREZ, I. C.; AND HOPKINS, T. C., 2012, Development of a new durability index for compacted shale: Engineering Geology, Vol. 139–140, pp. 66–75. CHAPMAN, D. R.; WOOD, L. E.; AND SISILIANO, W. J., 1976, A comparative study of shale classiﬁcation tests and systems:

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 125–135

133

Gautam and Shakoor Bulletin Association Engineering Geologists, Vol. 13, No. 4, pp. 247–266. COROMINAS, J.; MARTINEZ-BOFILL, J.; AND SOLER, A., 2015, A textural classification of argillaceous rocks and their durability: Landslides, Vol. 12, No. 4, pp. 669–687. CROSTA, G., 1998, Slake durability vs ultrasound treatment for rock durability determinations (Technical Note): International Journal Rock Mechanics Mining Science, Vol. 35, No. 6, pp. 815– 824. CZEREWKO, M. A. and CRIPPS, J. C., 2001, Assessing the durability of mudrocks using modified jar slake index test: Quarterly Journal Engineering Geology Hydrogeology, Vol. 34, pp. 153–163. DEARMAN, W. R., 1995, Description and classification of weathered rocks for engineering purposes: The background to the BS5930:1981 proposals: Quarterly Journal Engineering Geology, Vol. 28, pp. 267–276. DICK, J. C. and SHAKOOR, A., 1992, Lithologic controls of mudrock durability: Quarterly Journal Engineering Geology, Vol. 25, pp. 31–46. DICK, J. C.; SHAKOOR, A.; AND WELLS, N., 1994, A geological approach toward developing a mudrock-durability classification system: Canadian Geotechnical Journal, Vol. 31, No. 1, pp. 17– 27. ERGULER, Z. A. and SHAKOOR, A., 2009a, Relative contribution of various climatic processes in disintegration of clay-bearing rocks: Engineering Geology, Vol. 108, pp. 36–42. ERGULER, Z. A. and SHAKOOR, A., 2009b, Quantification of fragment size distribution of clay-bearing rocks after slake durability testing: Environmental Engineering Geoscience, Vol. 15, No. 2, pp. 81–89. ERGULER, Z. A. and ULUSAY, R., 2009, Assessment of physical disintegration characteristics of clay-bearing rocks: Disintegration index test and a new durability classification chart: Engineering Geology, Vol. 105, pp. 11–19. EXADAKTYLOS, G. E., 2006, Freezing-thawing model for soils and rocks: Journal Material Civil Engineering, Vol. 18, No. 2, pp. 241–249. FRANKLIN, J. A., 1981, A shale rating system and tentative applications to shale performance: Transportation Research Record 790: Transportation Research Board, pp. 2–12. FRANKLIN, J. A. and CHANDRA, R., 1972, The slake durability test: International Journal Rock Mechanics Mining Science, Vol. 9, pp. 325–341. GAMBLE, J. C., 1971, Durability-Plasticity Classification of Shales and other Argillaceous Rocks: Ph.D. Thesis, University of Illinois, Urbana, 161 p. GAUTAM, T. P., 2012, An Investigation of Disintegration Behavior of Mudrocks Based on Laboratory and Field Tests: Ph.D. Dissertation, Kent State University, 268 p. GAUTAM, T. P. and SHAKOOR, A., 2013, Slaking behavior of clay-bearing rocks during a one-year exposure to natural climatic conditions: Engineering Geology, Vol. 166, pp. 17–25. GAUTAM, T. P. and SHAKOOR, A., 2016, Comparing the slaking of clay-bearing rocks under laboratory conditions to slaking under natural climatic conditions: Rock Mechanics Rock Engineering, Vol. 49, No. 1, pp. 19–31. GOKCEOGLU, C.; ULUSAY, R.; AND SONMEZ, H., 2000, Factors affecting the durability of selected weak and clay-bearing rocks from Turkey, with particular emphasis on the influence of the number of drying and wetting cycles: Engineering Geology, Vol. 57, pp. 215–237. GRAINGER, P., 1984, The classification of mudrocks for engineering purposes: Quarterly Journal Engineering Geology, Vol. 17, pp. 381–387.

134

HAJDARWISH, A.; SHAKOOR, A.; AND WELLS, N. A., 2013, Investigating statistical relationships among clay mineralogy, index engineering properties, and shear strength parameters of mudrocks: Engineering Geology, Vol. 159, pp. 45–58. HEIDARI, M.; RAFIEI, B.; AND MOHEBBI, Y., 2015, Assessing the behavior of clay-bearing rocks using static and dynamic slaking indices: Geotechnical Geological Engineering, Vol. 33, No. 4, pp. 1017–1030. HOPKINS, T. C. and DEEN, R. C., 1983, Identification of Shales: Kentucky Transportation Center Research Report, Paper 975. International Society for Rock Mechanics (ISRM), 1979, Suggested methods for determining water content, porosity, density, absorption, and related properties and swelling and slakedurability index properties: International Society of Rock Mechanics, Commission on Standardization of Laboratory and Field Tests: International Journal Rock Mechanics Mining Science, Vol. 16, No. 2, pp. 141–156. ISRM, 2007, The complete ISRM suggested methods for rock characterization, testing and monitoring: 1974–2006: In ULUSAY, R. and HUDSON, J. A. (Editors), Suggested Methods Prepared by the Commission on Testing Methods: ISRM, Ankara, Turkey, 628 p. KHANDELWAL, M., 2013, Correlating P-wave velocity with the physico-mechanical properties of different rocks: Pure Applied Geophysics, Vol. 170, No. 4, pp. 507–514. KONCAGUL, E. C. and SANTI, P. M., 1999, Predicting the unconfined compressive strength of the Breathitt shale using slake durability, shore hardness and rock structural properties: International Journal Rock Mechanics Mining Science, Vol. 36, pp. 139–153. MARQUES, E. A. G.; VARGAS, E. D. A., JR.; AND ANTUNES, F. S., 2005, A study of the durability of some shales, mudrocks and siltstones from Brazil: Geotechnical Geological Engineering, Vol. 23, No. 3, pp. 321–348. MARTINEZ-BOFILL, J.; COROMINAS, J.; AND SOLER, A., 2004, Behavior of the weak rock cut slopes and their characterization using the results of the slake durability test: Engineering Geology Infrastructure Planning Europe, Vol. 104, pp. 405–413. ˇ C´ , P. and VLASTELICA, G., 2011, Durability characterization MISˇ CEVI of marls from the region of Dalmatia, Croatia: Geotechnical Geological Engineering, Vol. 29, No. 5, pp. 771–781. ´ , E.; ALONSO, F. J.; CARMOLINA, E.; CULTRONE, G.; SEBASTIAN RIZO, L.; GISBERT, J.; AND BUJ, O., 2011, The pore system of sedimentary rocks as a key factor in the durability of building materials: Engineering Geology, Vol. 118, pp. 110–121. MOON, V. G. and BEATTIE, A. G., 1995, Textural and microstructural influences on the durability of Waikato Coal Measures mudrocks: Quarterly Journal Engineering Geology, Vol. 28, pp. 303–312. MORADIAN, Z. A.; GHAZVINIAN, A. H.; AHMADI, M.; AND BEHNIA, M., 2010, Predicting slake durability index of soft sandstone using indirect tests: International Journal Rock Mechanics Mining Science, Vol. 47, No. 4, pp. 666–671. NICKMANN, M.; SPAUN, G.; AND THURO, K., 2006, Engineering geological classification of weak rocks. In Proceedings of the 10th International IAEG Congress 2006, Nottingham: Paper No. 492: IAEG, London, U.K. OLIVIER, H. J., 1979, A new engineering-geological rock durability classification: Engineering Geology, Vol. 14, pp. 255–279. POTTER, P. E.; MAYNARD, J. B.; AND PRYOR, W. A., 1980, Sedimentology of Shale: Springer-Verlag, New York, 306 p. RICHARDSON, D. N., 1985, Relative durability of shale—A suggested rating system, building on/with sedimentary bedrock. In WEST, T. R. (Editor), Proceedings of the 36th Annual Highway Geology Symposium: pp. 105–137.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 125–135

Classifying Clay-Bearing Rocks RINCON, O.; SHAKOOR, A.; AND OCAMPO, M., 2016, Investigating the reliability of H/V spectral ratio and image entropy for quantifying the degree of disintegration of weak rocks: Engineering Geology, Vol. 207, pp. 115–128. RUSSELL, D. J., 1981, Controls on shale durability: The response of two Ordovician shales in the slake durability test: Canadian Geotechnical Journal, Vol. 9, pp. 1–13. SADISUN, I. A.; SHIMADA, H.; ICHINOSE, M.; AND MATSUI, K., 2005, Study on the physical disintegration characteristics of Subang claystone subjected to a modiﬁed slaking index test: Geotechnical Geological Engineering, Vol. 23, pp. 199–218. SANTI, P. M., 1998, Improving the jar slake, slake index, and slake durability test for shales: Environmental Engineering Geoscience, Vol. 4, No. 3, pp. 385–396. SANTI, P. M., 2006, Field methods for characterizing weak rock for engineering: Environmental Engineering Geoscience, Vol. 12, No. 1, pp. 1–11. SHAKOOR, A. and GAUTAM, T. P., 2015, Inﬂuence of geologic and index properties on disintegration behavior of clay-bearing rocks: Environmental Engineering Geoscience, Vol. 21, No. 3, pp. 197– 209.

SHARMA, P. K. and SINGH, T. N., 2008, A correlation between Pwave velocity, impact strength index, slake durability index and uniaxial compressive strength: Bulletin Engineering Geology Environment, Vol. 67, No. 1, pp. 17–22. SINGH, T. N.; VERMA, A. K.; SINGH, V.; AND SAHU, A., 2005, Slake durability study of shaly rock and its predictions: Environmental Geology, Vol. 47, No. 2, pp. 246–253. TAYLOR, R. K. and SMITH, T. J., 1986, The engineering geology of clay minerals: Swelling, shrinking and mudrock breakdown: Clay Mineralogy, Vol. 21, pp. 235–260. WOOD, L. E. and DEO, P., 1975, A suggested system for classifying shale materials for embankments: Bulletin Association Engineering Geologists, Vol. 12, No. 1, pp. 39–55. YAGIZ, S.; SEZER, E. A.; AND GOKCEOGLU, C., 2012, Artiﬁcial neural networks and nonlinear regression techniques to assess the inﬂuence of slake durability cycles on the prediction of uniaxial compressive strength and modulus of elasticity for carbonate rocks: International Journal Numerical Analytical Methods Geomechanics, Vol. 36, pp. 1636–1650. YOUN, H. and TONON, F., 2010, Effect of air-drying duration on the engineering properties of four clay-bearing rocks in Texas: Engineering Geology, Vol. 115, pp. 58–67.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 125–135

135

Regional Landslide Inventory Mapping of Bashilo River Watershed, Ethiopia ELAMIN H. ISMAIL J. DAVID ROGERS MUHAMMED F. AHMED1 Geosciences and Geological and Petroleum Engineering, Missouri University of Science and Technology, Rolla, MO 65409, ehif22@mail.mst.edu, rogersda@mst.edu, mfanr5@mst.edu

MOHAMED G. ABDELSALAM Boone Pickens School of Geology, Oklahoma State University, Stillwater, OK 74078, mohamed.abdelsalam@okstate.edu

Key Terms: Landslide Inventory, Hillshade Map, GIS, Topography, ASTER DEM

pipelines, electrical transmission corridors, and structures.

ABSTRACT

INTRODUCTION

This study tentatively identified large bedrock landslide features across the Bashilo River, a tributary of the upper Blue Nile River of Ethiopia, using a geospatial mapping approach. The study aims to highlight the utilization of low-cost mapping techniques that might be applicable across large tracts of land (between 1000 and 500,000 km2 ). Topographic maps with 40 m contours were draped onto 30-m-spatial-resolution hillshade digital elevation models generated from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), which were utilized to produce stitched hillshade topographic maps to allow the delineation of possible bedrock landslide features. The landslide identification process was aided by employing anomalous topographic protocols, including: divergent contours, isolated topographic benches, crenulated contours, and disturbed drainage patterns as key indicators of likely landsliderelated features. The mapping exercise identified several hundred landslide features, including some landslide complexes believed to be seismically induced, composite bedrock landslides, rotational slumps, earth flows, and translational block slides generally >500 m in length. This first-pass regional landslide inventory was based on moderate-resolution data and limited resources. The resulting maps are intended to serve as a general guide for regional hazard assessment, realizing that more detailed site-specific studies should be undertaken where mapped landslide features might pose a hazard to the placement of critical infrastructure, such as highways, railroads,

The methods used for mapping and characterizing landslide features vary considerably, depending upon the researcher’s academic training and practical experience, purpose of the mapping, size and accessibility of the study area, and resolution of the supporting topographic and geological maps and remotely sensed data (Hansen, 1984; Varnes, 1984; Cruden and Varnes, 1996; and Van Den Eeckhaut et al., 2009). Conventional landslide mapping methods include examining stereo-pair aerial photographs to identify anomalous surface morphologies indicative of recent landsliding (Liang, 1952). This technique gradually evolved into a method involving detailed analysis of topographic, geologic, and geomorphic maps, as well as aerial photographs (Varnes, 1984; Rogers, 1994; and Su and Stohr, 2000). These methods are often supplemented with field observations and identification of such features, along with historic records of past landslides. This information allows a more refined characterization of past landsliding that can be used to perform more informative and reasonable hazard assessments to aid decisions relating to land management and/or mitigation (Guzzetti et al., 2006, 2012; De Graff et al., 2012). Limitations commonly associated with traditional mapping methods include: the accuracy of published maps, the meteorological conditions when aerial photographs were imaged, the scan quality, respective scales/resolution of topographic maps, the technical expertise and experience of the workforce, and the available funds to support such efforts (Booth et al., 2009; Knapen et al., 2006). Over the past 26 years, international standards and protocols have been established for the classification

1 On leave from University of Engineering and Technology, Lahore, Pakistan.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137–151

137

Ismail, Rogers, Ahmed, and Abdelsalam

of landslides, which are useful for risk assessment and hazard mapping (International Association of Engineering Geology [IAEG] Commission on Landslides, 1990). Government agencies traditionally rely on conventional methods based on field reconnaissance and evaluation of stereo-pair aerial photographs for mapping landslides (Jackson et al., 2012). There are numerous remote-sensing techniques presently used to acquire landslide data, including optical (visible and near infrared [VNIR], shortwave infrared [SWIR], thermal infrared [TIR]), radio detection and ranging (RADAR) data, as well as digital elevation models (DEMs) compiled from photogrammetry, synthetic aperture radar interferometry (InSAR), and light detection and ranging (LiDAR) (Glenn et al., 2006; Haneberg et al., 2009). These data are collected by ground-based sensors, airborne sensors (up to 24,000 m altitude), and space-based orbital platforms (space shuttles and satellites). These tools have provided the scientific community with the rapid acquisition of an unprecedented volume of information over a relatively short period of time (beginning with the American launch of Skylab in 1973). Remotely sensed data have rapidly evolved on both the global and local scale, with resolutions enhanced to sub-meter levels. Metternicht et al. (2005) classified landslide studies that utilize remotely sensed data into three basic phases: (a) detection and identification; (b) monitoring; and (c) spatial analysis and hazard prediction. The use of geographic information systems (GIS) and the development of new algorithms make it possible to refine and maximize remote-sensing data to detect ever smaller landslide features with increasing accuracy (Ayenew, 2004; Wang and Peng, 2009; Ayele, 2009; Zvelebil et al., 2010; and Akinci et al., 2011). LiDAR can provide sub-meter resolution and is capable of penetrating tree canopies and mapping the “bare earth” conditions, but LiDAR scanners are expensive and not readily available to developing countries. They also require storage and processing of large volumes of data. The use of high-resolution remote-sensing data and spectral analysis in automated landslide mapping is an emerging alternative to traditional mapping methods. However, the traditional techniques remain in use, owing to the high initial setup cost and the technical skills required to implement automated methods (Booth et al., 2009), and the superior level of detail possible with conventional field mapping of landslides. Alternatively, the increased availability of inexpensive tools, open-source software, and freely downloadable global remote-sensing data (with moderate resolution) have made regional-level landslide mapping possible. The Shuttle Radar Topography Mission (SRTM)

138

and the Advanced Space-borne Thermal Emission and Reflection Radiometer (ASTER) are currently sources of global DEM data (Tachikawa et al., 2011) for regional-level studies. This research sought to identify and delineate landslides and related features by utilization of low-cost mapping techniques, including anomalous topographic protocols, contour patterns, and moderate-resolution satellite imagery (Rogers, 1994; Su and Stohr, 2000; Doyle and Rogers, 2005; Van Den Eeckhaut et al., 2009; and Ahmed and Rogers, 2012, 2014, 2016) in a GIS environment. These data were used to compile a cost-effective first-pass reconnaissance-level landslide map (De Graff et al., 2012) seeking to identify potential landslide features, based on expert knowledge, and to delineate the areal extents of such features on a regional scale (features >500 m long). The results of this inventory show several hundred landslides, including several landslide complexes likely induced by seismic excitation, composite bedrock landslides, retrogressive slumps, earth flows, and translational block slides having sizes generally >500 m in length. Because of the resolution of the available data, a certain degree of error should be expected, even under the best conditions. The authors’ intention was simply to identify “target sites” for more detailed evaluation of potential slope stability hazards, especially for under-developed countries like Ethiopia. OVERVIEW OF STUDY AREA The Bashilo River is located on the Northwestern Ethiopian Plateau (see Figure 1). It originates west of Kutaber and initially flows northwest, where it convolves with the Tergiya tributary. Downstream of this location, it turns southwest, eventually joining the Abay River along a structurally controlled course (Ismail, 2011). The drainage area of Bashilo watershed is about 13,242 km2 , covering parts of the Semien Gondar, Semien Wollo, and Debub Wollo areas. Tectonic and Geologic Settings Being part of Northwestern Plateau of Ethiopia, the Bashilo River watershed has experienced repeated episodes of uplift since ∼150 Ma (Hofmann et al., 1997; Wolfenden et al., 2004; and Bonini et al., 2005), exposing sandstone and limestone at altitudes of ∼2200 m within the major drainage networks of the plateau (Gani et al., 2007). Shield volcanic eruptions of flood basalts impacted the study area, extending from the western escarpments of the Afar Depression and northwestern escarpment of the Main Ethiopian Rift during the late Holocene (Figure 2). The lava flow deposition took

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137–151

Landslide Inventory Mapping of Bashilo River Watershed, Ethiopia

Figure 1. Left: The study area is shown in green, within the northwestern portion of the Ethiopia Plateau shaded. Right: Elevation map of the Bashilo watershed, showing the deep incision of the plateau tablelands.

Figure 2. Geologic units identiďŹ ed in the Bashilo River watershed.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137â&#x20AC;&#x201C;151

139

Ismail, Rogers, Ahmed, and Abdelsalam

place in a series of intermittent events of early Eocene to late Miocene age (Mohr and Zanettin, 1988; Hof¨ 2001; Kieffer et al., 2004; mann et al., 1997; S¸engor, and Beyene and Abdelsalam, 2005). The basalts tend to weather deeply, forming clay materials, including expansive smectite clays and kaolinites, some of which blanket the vast expanse of low-lying plains between Ethiopia and Sudan. The basaltic units in the study area are thoroughly jointed and weathered. The block kinematics operating within the landslides are usually influenced by these discontinuities, and they are sufficiently continuous to perturb an entire slide mass. Many discontinuities also serve as “groundwater barriers,” trapping downward percolating water and thereby elevating pore water pressures, which often influence near-surface mass movements. The regional geology of the Bashilo watershed is shown in Figure 2. Slope Stability Major slope failures have been noted in the seismically active Afar Depression and the Main Ethiopian Rift near Dessie City in the southwestern portion of the Bashilo River watershed (Fubelli et al., 2013). The Bashilo watershed was selected for this study because it is located in the uplifted part of the Northwestern Plateau. Slopes in the region are known for mass wasting processes, being bordered by Mount Guna (47,20 m) in the northwest, the elevated ridges of the western escarpments along its southern margins, and the elevated plateaus lying to the northeast. Several researchers have prepared landslide hazard maps of the Dessie City and Tigray region of northern Ethiopia. A significant population has recently settled on the prehistoric landslide debris, which could pose long-term hazards to the region’s infrastructure (Nyssen et al., 2003, 2006; Woldearegay et al., 2005; and Moeyersons et al., 2008). Field visits and subsequent investigations have sought to examine the influence of underlying geologic structure on mass wasting, because these deeply eroded slopes appear to be “out of equilibrium.” Most of the anomalous topographic features exhibit long axes parallel to the slope fall line. MATERIAL AND METHODS Data The available data utilized for mapping possible landslide features in the Bashilo River watershed included freely available ASTER DEMs (GDEM v2) with 30 m spatial resolution. ASTER is an imaging instrument carried by the Terra satellite, launched in December 1999 by the U.S. National

140

Aeronautics and Space Administration (NASA) as part of its Earth Observing System (EOS) to obtain high-resolution data of land surface temperature, reflectance, and surface elevation, and to generate DEMs (www.jspacesystems.or.jp/ersdac/GDEM/E/4.html). The ASTER GDEM v2 is a preprocessed data set with increased accuracy and fewer artifacts as compared to SRTM data (Tachikawa et al., 2011). It has been found to be more reliable for constructing hillshade topographic maps, like the ones used in this study. Topographic map sheets at a scale of 1:200,000 maintained by the Topographic Department of the Military Forces (MF) of the Ministry of Defense of the Russian Federation were downloaded at a nominal cost from an online source (mapstor.com). The cost of the data collection and processing was less than US$300, excluding the cost of GIS software (ArcGIS and ENVI), which are available at any American university. The ASTER GDEM v2 data images were mosaicked in ENVI 4.8 and exported to ArcGIS 10.0, and a hillshade image was created, employing a sun elevation angle of 45 degrees with an azimuth of 315 degrees. We found that the ASTER DEM data with 30 m resolution was suitable for the regional-level studies to delineate bedrock slides larger than 500 m (Ahmed and Rogers, 2014, 2016). The topographic sheets were then mosaicked in Global Mapper13.0 software and imported into ArcGIS 10.0. Later, all of the topographic sheets of the study area were co-registered and overlapped on the hillshade image (created from ASTER DEM data) to produce a stitched hillshade topographic map. The final product was then printed out on 48 in. (∼122 cm) by 36 in. (∼91 cm) sheets to aid in visual identification of bedrock landslide features by recognition of anomalous topographic features. Concepts Underlying Mapping and Identification Methods Various forms of mass movements can often be visually identified using topographic maps and other satellite-based imagery in combination with recognition protocols for anomalous topographic features (Sharpe, 1938; Rogers, 1994; Kellogg, 2001; Doyle and Rogers, 2005; Van Den Eeckhaut et al., 2009; Moeyersons et al., 2008; and Ahmed and Rogers, 2012, 2016). A few of the key topographic features representative of past slide movements are described below. Topographic Features Topographic maps can provide a useful base map from which to map relatively recent landslides and landslide-related features (such as outbreak flood

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137–151

Landslide Inventory Mapping of Bashilo River Watershed, Ethiopia

Figure 3. Sketches illustrating cause-and-effect sequences of mass wasting into perennial river channels.

terraces, landslide evacuation scarps, bedrock slides, debris flow lobes, etc.). Hummocky topography, transverse ridges, isolated low-gradient channel thalweg profiles, isolated erosional knobs, and lobate distal depositional fans are often direct manifestations of mass movements. The mollification of such features is suggestive of the relative age of the landslide features (Gealy, 1955; McCalpin, 1984; Wieczorek, 1984; Rogers, 1995, 1997; and Kellogg, 2001). Drainage Pattern Anomalies One of the simplest techniques to detect where past landslides have impacted channels is by searching for anomalous drainage patterns within the drainage network (Rogers, 1994). Water courses are often blocked or temporarily obstructed by landslide debris, forming short-lived “landslide lakes” upstream of the blockages. These lakes gradually fill and overtop the debris dams, rapidly excavating new channels (see Figure 3). The new channels normally entrench themselves

where the outpouring waters initially overtopped the debris, often shifting the channel from its prior course (Costa and Schuster, 1988). Rogers (1997) and Doyle and Rogers (2005) included semi-circular drainage courses around topographic obstructions, rapids in curving channel reaches devoid of tributary side-entry, and meandering, low-gradient channels upstream of rapids as physical perturbations often triggered by the rapid erosion of landslide dams (Grater, 1945; Lee and Duncan, 1975). Rising waters trapped behind debris dams usually serve to elevate pore water levels in adjacent banks, if the natural reservoir persists more than a few months (Rogers, 1997). When the debris dam suddenly breaches (see Figure 3A), the elevated pore water pressure in the opposing bank often triggers a rapid drawdown-induced failure of the opposing slope, bringing debris to the channel from both sides of the canyon. Resistant blocks emanating from both canyon walls can serve to armor the channel bed and sustain partial blockage of the channel (see Figure 3B).

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137–151

141

Ismail, Rogers, Ahmed, and Abdelsalam

Figure 4. Examples of anomalous topographic expressions, descriptive of various types of mass movement: (A) translational bedrock landslide with secondary slide features, (B) typical earth/debris ﬂow and slump expressions, and (C) planar rotational slump (Rogers, 1994).

Evaluation of Contour Patterns Coincident upslope and downslope divergence of contour lines often occurs when material has been displaced along a hillslope’s fall line by mass wasting. This is because both rotational and translational movements cause signiﬁcant dilation of the transposed material, which continues sliding until a more stable inclination is achieved, or when the pore water pressures initiating movement are sufﬁciently dissipated (see Figure 4). Mass movements also create a “zone of depletion” (Sharpe, 1938; Varnes, 1958), which di-

142

verts hillside contours inward (into the slope) relative to more stable slopes along either flank of the mass movement. The contour interval of the topographic maps used in this study was 40 m, which suggests that landslides shorter than about 500 m crossing at least five contours (minimum elevation differential of 200 m) would not normally be detectable with any significant certainty on slopes inclined between 2:1 (26.7 degrees) and 3:1 (18.1 degrees) (Ahmed and Rogers, 2014, 2016). As the translated mass of debris moves downslope, it tends to divert hillside contours outward, defining

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137–151

Landslide Inventory Mapping of Bashilo River Watershed, Ethiopia

a “zone of inflation” (Sharpe, 1938; Varnes, 1958). Landslide headscarp evacuation areas are typified by arcuate contours, which are anomalous in comparison with adjacent stable slope contours. Figure 4A presents typical examples of deep-seated translational landslides, which often develop extended topographic knobs (Rogers, 1994). Earth and rock flow features are readily recognizable by their divergent contours within the hillslopes as they form wide circular-shaped headscarp evacuation zones (Figure 4B). Crenulated contours are often associated with debris fans and flow lobes, common to debris flows, earth flows, and rockfall avalanche sturzstroms. Superposition of successive flow lobes and/or sequences of movement tend to truncate one another. Older earth flows can usually be identified by crenulated contours across multiple flow lobes, which tend to be laterally restricted (Ibsen et al., 1996). Rotational slump features (see Figure 4C) were also recognized by multiple isolated topographic benches (i.e., back-rotated grabens). In plan view, these often appear as a series of small terrasets (Sharpe, 1938; Zaruba and Mencl, 1982). Hutchinson and Bischof (1983) and Cruden et al. (1991) mentioned low-gradient channels upstream of steeply inclined rapids, and closed depressions within topographic benches as useful topographic keys, as the latter are often headscarp graben features. Breaks in contour lines along a ridge can also be caused by landslides if the anomalous pattern is localized along the slope’s natural fall line and does not continue up or down the valley along the same hillside. Anomalous topographic features also develop in areas experiencing active landslippage because of localized ground settlement that often results from dilation during movement and subsequent drainage. These isolated benches often form attractive construction sites, supporting deranged drainages that promote the development of sag ponds. Van Den Eeckhaut et al. (2009) used anomalous topographic feature protocols, aerial photo interpretation, and field methods to identify landslides across a 500 km2 area around Hagere Selam in the Tigray region of Ethiopia. The key element in their recognition of slides was the identification of anomalous topographic patterns or geomorphic expressions of features commonly associated with landslippage, which were laterally restricted.

DISCUSSION ON INVENTORY MAPPING RESULTS Figure 5 shows a summary of the visually identiﬁed and delineated landslide features of Bashilo River watershed, based on the methods described in the pre-

vious section. Various types of bedrock landslides were discerned by analyzing their respective geomorphic expression on the stitched topographic hillshade map. This inventory of bedrock landslide features nearly 1,050 landslides (generally >500 m in length) within the 13,242 km2 study area. Figure 6 shows a ﬂow slide feature typical of the study area, exhibiting crenulated contours and an arcuate headscarp. Hummocky topography, lobate features, and the anomalous topographic benches (relative to the surrounding slopes) are common indicators of past landslippage in this watershed. These features tend to locally perturb runoff patterns and create parallel or slightly converging drainage swales. Figure 7 presents a common situation where isolated topographic benches served as key topographic features in the recognition of deep-seated translational landslides. Notice the crenulated contour lines as salient features in discerning surface disturbances fomented by prehistoric mass wasting in the study area. Crenulated contours are often diagnostic of a disturbed mass just downslope of the headscarp separation zone. These features are typical of older, deepseated bedrock slides, often associated with earthquakes (Keefer, 1984). Karlin et al. (2004) reported that large-scale landslides are among the most obvious manifestations of large earthquakes, and that an earthquake of magnitude 7 may trigger landslides over a 20,000 km2 area, while a magnitude 8 earthquake may affect an area of 100,000 km2 . Figure 8 shows the semi-circular drainage courses tracing the approximate areal boundary of a translational bedrock slide, which often form temporary landslide dams in channel reaches devoid of tributary side-entry, and which exhibit anomalously low gradients (meandering) immediately upstream of the former blockage. These are key drainage features indicative of old translational slides and short-lived landslide dams (Costa and Schuster, 1988). Figure 8 also presents typical examples of deepseated, primarily translational landslides with extended topographic ridges. These are common expressions of past landslippage in the study area, and they often serve as source materials for triggering secondary landslides, often in response to prolonged surﬁcial erosion. Hutchinson and Bischof (1983) and Cruden et al. (1991) included the low-gradient channels upstream of steeply inclined linear rapids, and the closed depressions within topographic benches as useful topographic keys in identifying old landslides, as the closed depressions are usually grabens formed in the headscarp-pullapart zone of the landslide. Lobate features often form when the length of the spreading material is more than two-fold its elevation, due to increasing plastic deformation associated with

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137–151

143

Ismail, Rogers, Ahmed, and Abdelsalam

Figure 5. Locations of likely landslide features in the Bashilo River watershed. The highlighted areas show the examples discussed in this study to describe the mapping procedures.

disaggregation of the parent materials as they rotate and dilate with downslope movement. Coincident upslope and downslope divergence of contours is illustrated in the insets of Figures 6, 7, and 8. This commonly occurs when the material removed or displaced from the upper portion of the hillslope (excavation and translation) occurs along the same slope with contours demarcating the zone of inﬂation (deposition and spreading). The conditions necessary for this interpretation require that the diverging contours should be more or less situated along the same fall line (an imaginary line running perpendicular to the slope’s contour lines). The upper panel of Figure 9A shows our deep-seated landslide features map in the southeastern part of the Bashilo watershed. Four subsurface proﬁles were cut through some of these features to interpret the slope morphology of the mapped anomalies (Ismail et al., 2016). Figure 9B (sub-parts a through d) presents our preliminary interpretation of the likely subsurface structure, employing expert knowledge of regional lithology, structure, and anomalous geomorphic features. Figure 9B (a) is taken through a prominent pullapart graben and toe slump, likely triggered by riverine erosion of the slope toe. Figure 9B (a) suggests that the

144

anomalous topographic benches have formed adjacent to secondary scarp features, and flat terraces associated with retrogressive translations of various blocks within a large translational block slide complex. The lower two thirds of the slide mass exhibit features typical of progressive (or retrogressive) block glides (Zaruba and Mencl, 1976). En-echelon slump blocks are a typical scenario observed above receding shorelines or along water courses where active erosion is removing materials from the lowest slump block (Ward, 1945; Skempton, 1946; and Zaruba and Mencl, 1982). Figure 9B (b and c) also shows typical examples of retrogressive translational bedrock landslides. These features are distinguished by the confined convex-upward benches in the hillslope profile, which are laterally restricted to the boundaries of the slide features. Note the asymmetry of the valley profiles in Figure 9B (b and d). Figure 9B (d) exhibits prominent pull-apart grabens and colluvial filled benches on both sides of the incised channel, testifying to a significant landslide dam that has been breached in the recent geologic past. These sorts of conditions are ideal for the formation of significant landslide dams in the future. Other physical manifestations include hanging terraces and deposition of lacustrine sediments immediately

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137–151

Landslide Inventory Mapping of Bashilo River Watershed, Ethiopia

Figure 6. Inferring flow type of landslides such as debris flows and earth flows from the divergence and localized breaks of contour lines along hillside fall lines, the arcuate nature of headscarp evacuation scars, isolated topographic benches, and crenulated contours in the study area (sketches from Rogers, 1998).

upstream. Many of these deep-seated block glide and translational landslides may have been generated by earthquakes and/or by rapid drawdown of pore water pressures, as landslide dams are suddenly breached. The approximate boundaries of the landslide features were estimated based on the resolution of the

hillshade map. We acknowledge that the ﬁnite boundaries of such features are not readily discernible; thus, a certain degree of uncertainty should be expected. With a map Reduction Factor (RF) of 1:40,000 the relative spatial uncertainty is likely 100 m (∼3 mm) or more, and the elevation uncertainty is not less than one-half

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137–151

145

Ismail, Rogers, Ahmed, and Abdelsalam

Figure 7. Enlarged section of landslide inventory map of the Bashilo watershed showing a nice example of nested secondary slides developed with a larger parent bedrock landslide.

contour interval, or ∼40 m. When a slide mass is activated, it usually represents only a portion of previous landslippage within the surrounding area, so predictions of future landslippage would be difﬁcult to estimate without site-speciﬁc information. Results of this regional reconnaissance study were compared with on-the-ground mapping studies con-

146

ducted at the Dessie, Tigray, and Hagere Selem areas of northern Ethiopia, underlain by similar geology (i.e., flood basalts exhibiting a mixture of flow and pyroclastic sequences). Most of the landslides we identified using anomalous topographic protocols were in agreement with the field observations made in the much smaller, site-specific studies using conventional field

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137–151

Landslide Inventory Mapping of Bashilo River Watershed, Ethiopia

Figure 8. Excerpt of landslide inventory map showing some of the mapped landslides in the central portion of the Bashilo watershed: (A) example of seismically induced landslides, forming temporary landslide dam, and (B) translational block glide landslide (insets from Rogers, 1998).

methods. These findings would suggest that the present study was validated to some unknown measure, though its precise accuracy cannot be assigned at this time, without field verification (which could be unduly expensive in such an inaccessible area). This result suggests that the utilization of expert knowledge of anomalous geomorphic features and

employment of low-cost satellite imagery data could be a very cost-effective method of making an initial, “first-pass reconnaissance” of large tracts of hilly terrain. This regional inventory identified more than 1,050 landslide features, including translational bedrock slides, retrogressive slump complexes, flow slides, and composite slide features (including primary

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137–151

147

Ismail, Rogers, Ahmed, and Abdelsalam

Figure 9. (A) Excerpts of landslide inventory mapping indicating the location of the mapped landslides in the southeastern portion of the Bashilo River watershed. The colored lines show the locations of profiles drawn along the identified deep-seated bedrock landslides. (B) Example landslides: (a) steeply inclined translational composite bedrock landslide with linear slope above the slide blocks, (b) typical example of retrogressive slump-block slides, (c) translational block slides identified in the hillslope, and (d) a composite landslide (translational block slides) with headscarp grabens (topographic benches) (modified after Ismail et al., 2016).

and secondary slides). Individual tallies of the various types of landslide features were beyond the scope of this study due to the moderate data resolution (40 m contour interval) and limited resources. This study was

148

intended to be used as a general guide to suggest likely sites where site-speciﬁc landslide hazard evaluations should be performed. In areas devoid of subsurface geologic information, the local authorities would be

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137–151

Landslide Inventory Mapping of Bashilo River Watershed, Ethiopia

more informed in deciding where to locate critical infrastructure, such as roads, bridges, transmission lines, pipelines, and the like. CONCLUSIONS This study highlighted the significance of utilizing low-cost methods and procedures to prepare regional landslide hazard maps at scales of ∼1:40,000 in mountainous areas with limited access and available resources. This study identified large bedrock landslide features across the Bashilo River sub-watershed, a significant tributary of the upper Blue Nile River in Ethiopia. These regional reconnaissance-level landslide inventory maps seek to identify and delineate bedrock landslide features >500 m long using anomalous topographic protocols on hillshade topographic maps. For this study, a topographic map with 40 m resolution was stitched onto a 30-m-spatial-resolution hillshade DEM generated from ASTER satellite imagery to produce a hillshade topographic map that could be used to delineate landslides and related features. Nearly 1050 bedrock landslides of various types were identified by recognition of anomalous geomorphic expressions on the stitched topographic hillshade map. The landslides included translational complexes, composite landslides, slumps, earth flows, and translational block glides. Results of this study were also compared with field studies conducted at Dessie Tigray and Hagere Selem, which possess similar geologic settings in northern Ethiopia. Mapping of landslide features is very subjective, based on “expert knowledge.” As a consequence, significant variances in interpretation are to be expected, depending on the quality and duration of professional training and experience of the interpreter. This study could serve as a general assessment of relative landslide hazard across the watershed area to identify “target sites” for more detailed evaluation of potential slope failure hazards. The availability of high-resolution satellite data, aerial photographs, and large-scale topographic maps with smaller contour intervals would certainly enhance the detection process, allowing smaller-scale landslide features to be delineated, and would reduce interpretive errors and improve the accuracy of landslide identification in the study area. ACKNOWLEDGMENTS The authors would like to thank the Ethiopian Geological Survey for the logistics and support they provided during field investigation in Ethiopia, and our special thanks and appreciations for contribution of

the “late” Dr. Lulu Tsegi, who sincerely devoted his time and effort for most of the research conducted in Ethiopia. REFERENCES AHMED, M. F. AND ROGERS, J. D., 2012, Landslide mapping and identiﬁcation of old landslide dams along the Indus River 331 in Pakistan, using GIS techniques. In Association of Environmental & Engineering Geologists (AEG) 55th Annual Meeting: AEG, Salt Lake City, UT, Vol. 55, 44p. AHMED, M. F. AND ROGERS, J. D., 2014, First-approximation landslide inventory maps for northern Pakistan, using ASTER DEM data and geomorphic indicators: Environmental & Engineering Geoscience, Vol. 20, No. 1, pp. 67–83. AHMED, M. F. AND ROGERS, J. D., 2016, Regional level landslide inventory maps of the Shyok River watershed, northern Pakistan: Bulletin of Engineering Geology and the Environment. Vol. 75, pp. 563–574. DOI:10.1007/s10064-015-0773-2. AKINCI, H.; DOGAN, S.; KILIC¸ OGLU, C.; AND TEMIZ, M., 2011, Production of landslide susceptibility map of Samsun City Center (Turkey), by using frequency ratio method: International Journal of the Physical Sciences, Vol. 6, No. 5, pp. 1015–1025. AYELE, S., 2009, Slope Instability and Hazard Zonation Mapping using Remote Sensing and GIS Technique in Abay Gorge (Gohatsion Dehen), Central Ethiopia: M.Sc. Thesis, Addis Ababa University, Addis Ababa. AYENEW, T., 2004, Environmental implications of changes in the levels of lakes in the Ethiopian Rift since 1970: Regional Environmental Change, Vol. 4, No. 1, pp. 192–204. BEYENE, A. AND ABDELSALAM, M. G., 2005, Tectonics of the Afar Depression: A review and synthesis: Journal of African Earth Sciences, Vol. 41, No. 1, pp. 41–59. BONINI, M.; CORTI, G.; INNOCENTI, F.; AND PECSKAY, Z., 2005, Evolution of the Main Ethiopian Rift in the frame of Afar and Kenya rifts propagation: Tectonics, Vol. 24, No. 1, pp. 1–21. BOOTH, A. M.; ROERING, J. J.; AND PERRON, J. T., 2009, Automated landslide mapping using spectral analysis and high-resolution topographic data: Puget Sound Lowlands, Washington, and Portland Hills, Oregon: Geomorphology, Vol. 109, No. 3, pp. 132–147. COSTA, J. E. AND SCHUSTER, R. L., 1988, The formation and failure of natural dams: Geological Society of America Bulletin, Vol. 100, No. 7, pp. 1054–1068. CRUDEN, D.; THOMSON, S.; AND HOFFMAN, B. A., 1991, Observation of graben geometry in landslides. In Chandler, R. J. (Editor), Slope Stability Engineering: Developments and Applications: Institution of Civil Engineers, London, pp. 33–35. CRUDEN, D. M. AND VARNES, D. J., 1996, Landslide types and processes. In Turner, A. K., and Schuster, R. L. (Editors), Landslides, Investigation and Mitigation: Special Report 247, Transportation Research Board, Washington, DC, pp. 36–75. DE GRAFF, J. V.; ROMESBURG, H. C.; AHMAD, R.; AND MCCALPIN, J. P., 2012, Producing landslide-susceptibility maps for regional planning in data-scarce regions: Natural Hazards, Vol. 64, No. 1, pp. 729–749. DOYLE, B. C. AND ROGERS, J. D., 2005, Seismically induced lateral spread features in the western New Madrid seismic zone: Environmental & Engineering Geoscience, Vol. 11, No. 3, pp. 251–258. FUBELLI, G.; GUIDA, D.; CESTARI, A.; AND DRAMIS, F., 2013, Landslide Hazard and Risk in the Dessie Town Area (Ethiopia): Landslide Science and Practice, Vol. 6, pp. 357–362. GANI, N. D.; GANI, M. R.; AND ABDELSALAM, M. G., 2007, Blue

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137–151

149

Ismail, Rogers, Ahmed, and Abdelsalam Nile incision on the Ethiopian Plateau: Pulsed plateau growth, Pliocene uplift, and hominin evolution: GSA Today, Vol. 17, No. 9, pp. 4–11. GEALY, B. L., 1955, Topography of the continental slope in northwest Gulf of Mexico: Geological Society of America Bulletin, Vol. 66, No. 2, pp. 203–228. GLENN, N. F.; STREUKER, D. R.; CHADWICK, D. J.; THACKRAY, G. D.; AND DORSCH, S. J., 2006, Analysis of LiDAR derived topographic information for characterizing and differentiating landslide morphology and activity: Geomorphology, Vol. 73, pp. 131–148. GRATER, R. K., 1945, Landslide in Zion Canyon, Zion National Park, Utah: Journal of Geology, Vol. 55, pp. 116–124. GUZZETTI, F.; MONDINI, A. C.; CARDINALI, M.; FIORUCCI, F.; SANTANGELO, M.; AND CHANG, K., 2012, Landslide inventory maps: New tools for an old problem: Earth-Science Reviews, Vol. 112, pp. 42–66. GUZZETTI, F.; REICHENBACH, P.; ARDIZZONE, F.; CARDINALI, M.; AND GALLI, M., 2006, Estimating the quality of landslide susceptibility models: Geomorphology, Vol. 81, pp. 166–184. HANEBERG, W. C.; COLE, W. F.; AND KASALI, G., 2009, Highresolution LiDAR-based landslide hazard mapping and modeling, UCSF Parnassus Campus, San Francisco, USA: Bulletin of Engineering Geology and the Environment, Vol. 68, pp. 263– 276. HANSEN, A., 1984, Landslide hazard analysis. In Brunsden, D. and Prior, D. B. (Editors), Slope Instability: Wiley & Sons, New York, pp. 523–602. HOFMANN, C.; COURTILLOT, V.; G. FÉRAUD, G.; AND PIK, R., 1997, Timing of the Ethiopian flood basalt event and implications for plume birth and global change: Nature, Vol. 389, No. 6653, pp. 838–841. HUTCHINSON, M. AND BISCHOF, R., 1983, A new method for estimating the spatial distribution of mean seasonal and annual rainfall applied to the Hunter Valley, New South Wales: Australian Meteorological Magazine, Vol. 31, No. 3, pp. 179–184. IBSEN, M. L.; BRUNSDEN, D.; BROMHEAD, E.; AND COLLISON, A., 1996, Flow slide, Art. 8.3. In Dikau, R.; Brunsden, D.; Schrott, L.; and Ibsen, M. L. (Editors), Landslide Recognition, Identification, Movement and Causes: Wiley, New York, pp. 202–211. INTERNATIONAL ASSOCIATION OF ENGINEERING GEOLOGY (IAEG) COMMISSION ON LANDSLIDES, 1990, Suggested nomenclature for landslides: Bulletin of the International Association of Engineering Geology, Vol. 41, No. 1, pp. 13–16. ISMAIL, E. H., 2011, Morpho-Tectonic Analysis of the Tekeze and the Blue Nile Drainage Systems of Northwestern Ethiopian Plateau, Ethiopia: M.Sc. Thesis, Missouri University of Science and Technology, Rolla, MO, 115 p. ISMAIL, E. H.; Rogers, J. D.; AHMED, M. F.; AND ABU-BAKAR, M. Z., 2016, Subsurface profile interpretation of landslides, examples from Bashilo River watershed, Ethiopia, Environmental Earth Sciences, Vol. 75, 1153 p. DOI 10.1007/s12665-016-5964-z. JACKSON, L. E., JR.; BOBROWSKY, P. T.; AND BICHLER, A., 2012, Identification, Maps and Mapping—Canadian Technical Guidelines and Best Practices Related to Landslides: A National Initiative for Loss Reduction. Geological Survey of Canada, Open File 7059, 33 p. KARLIN, R. E.; HOLMES, M.; ABELLA, S.; AND SYLWESTER, R., 2004, Holocene landslides and a 3500-year record of Pacific Northwest earthquakes from sediments in Lake Washington: Geological Society of America Bulletin, Vol. 116, No. 1–2, pp. 94–108 KEEFER, D. K., 1984, Landslides caused by earthquakes: Geological Society of America Bulletin, Vol. 95, No. 4, pp. 406–421.

150

KELLOGG, K. S., 2001, Tectonic controls on a large landslide complex: Williams Fork Mountains near Dillon, Colorado: Geomorphology, Vol. 41, No. 4, pp. 355–368. KIEFFER, B.; ARNDT, N.; LAPIERRE, H.; BASTIEN, F.; BOSCH, D.; PECHER, A.; YIRGU, G.; AYALEW, D.; WEIS, D.; JERRAM, D. A.; KELLER, F.; AND MEUGNIOT, C., 2004, Flood and shield basalts from Ethiopia: Magmas from the African Superswell: Journal of Petrology, Vol. 45, No. 4, pp. 793–834. KNAPEN, A.; KITUTU, M. G.; POESEN, J.; BREUGELMANS, W.; DECKERS, J.; AND MUWANGA, A., 2006, Landslides in a densely populated county at the footslopes of Mount Elgon (Uganda): Characteristics and causal factors: Geomorphology, Vol. 73, No. 1, pp. 149–165. LEE, K. L. AND DUNCAN, J. M., 1975, Landslide of April 25, 1974, on the Mantaro River, Peru: Report of Inspection, National Academy of Sciences, Washington, DC, 72 p. LIANG, T., 1952, Landslides: An Aerial Photographic Study: Ph.D. dissertation, Cornell University, Ithaca, NY, 274 p. MCCALPIN, J. P., 1984, Preliminary age classification of landslides for inventory mapping. In Proceedings of the 21st Annual Symposium on Engineering Geology and Soils Engineering: University Press, Moscow, ID, pp. 99–111. METTERNICHT, G.; HURNI, L.; AND GOGU, R., 2005, Remote sensing of landslides: An analysis of the potential contribution to geo-spatial systems for hazard assessment in mountainous environments: Remote Sensing of Environment, Vol. 98, No. 2, pp. 284–303. MOEYERSONS, J.; VAN DEN EECKHAUT, M.; NYSSEN, J.; TESFAMICHAEL; GEBREYOHANNES; VAN DE WAUW, J.; HOFMEISTER, J.; POESEN, J.; DECKERS, J.; AND MITIKU, H., 2008, Mass movement mapping for geomorphological understanding and sustainable development: Tigray, Ethiopia: Catena, Vol. 75, No. 1, pp. 45–54. MOHR, P. AND ZANETTIN, B., 1988, The Ethiopian flood basalt province. In: Macdougall, J. D. (editor) Continental flood basalts. Kluwer Academic Publisher, The Netherlands, pp. 63– 110. NYSSEN, J.; MOEYERSONS, J.; POESEN, J.; DECKERS, J.; AND HAIL, M., 2003, The environmental significance of the remobilization of ancient mass movements in the Atbara–Tekeze headwaters, northern Ethiopia: Geomorphology, Vol. 49, No. 3, pp. 303– 322. NYSSEN, J.; POESEN, J.; VEYRET-PICOT, M.; AND GOVERS, G., 2006, Assessment of gully erosion rates through interviews and measurements: A case study from northern Ethiopia: Earth Surface Processes and Landforms, Vol. 31, No. 2, pp. 167–185. ROGERS, J. D., 1994, Report Accompanying Map of Landslides and Other Surficial Deposits of the City of Orinda, CA: Rogers/Pacific, Inc. for the City of Orinda Public Works Department, 141 p. ROGERS, J. D., 1995, Causes and Identification of Urban Landslides: Identification of Landforms Susceptible to Landslippage and Methods of Analysis: Association of Bay Area Governments Training Center, Oakland, CA, 161 p. ROGERS, J. D., 1997, Spatial geologic hazard analysis in practice. In Frost, J. D.(Editor), Spatial Analysis in Soil Dynamics and Earthquake Engineering: ASCE Geotechnical Special Publication 67, New York, NY, pp. 15–28. ROGERS, J. D., 1998, Topographic Expression of Deep-Seated Bedrock Landslide Complexes: Short Course Notes for Evaluation and Mitigation of Seismic Hazards, University of California Extension, Los Angeles, CA, 13 p. ¨ , A. C., 2001, Elevation as indicator of mantle-plume activS¸ENGOR ity. In: Ernest, R. E., Buchan, K. L. (Editors), Mantle plumes: Their identification through times, Special Paper 352, Geological Society of America, Boulder, CO, pp. 183–225.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137–151

Landslide Inventory Mapping of Bashilo River Watershed, Ethiopia SHARPE, C.F.S., 1938, Landslides and Related Phenomena: Columbia University Press, New York, 137 p. SKEMPTON, A.W., 1946, Earth pressure and the stability of slopes. In The Principles and Applications of Soil Mechanics, Lecture II, Institution of Civil Engineers, London, pp. 31–61. SU, W. J. AND STOHR C., 2000, Aerial-photointerpretation of landslides along the Ohio and Mississippi Rivers: Environmental and Engineering Geoscience Bulletin, Vol. VI, No. 4, pp. 311– 323. TACHIKAWA, T.; KAKU, M.; IWASAKI, A.; GESCH, D.; OIMOEN, M.; ZHANG, Z.; DANIELSON, J.; KRIEGER, T.; CURTIS, B.; HAASE, J.; ABRAMS, M.; CRIPPEN, R.; AND CARABAJAL, C., 2011, ASTER Global Digital Elevation Model Version 2—Summary of Validation Results: Report to the ASTER GDEM Validation Team, pp. 15–24. VAN DEN EECKHAUT, M.; MOEYERSONS, J.; NYSSEN, J.; ABRAHA, A.; POESEN, J.; HAILE, M.; AND DECKERS, J., 2009, Spatial patterns of old, deep-seated landslides: A case-study in the northern Ethiopian highlands: Geomorphology, Vol. 105, No. 3, pp. 239–252. VARNES, D. J., 1958, Landslide types and processes. In: Eckel, E. B., ed., Landslides and Engineering Practice: Special Report 29, Highway Research Board, NAS-NRC Publication 544, Washington, D.C., pp. 20–47. VARNES, D. J., 1984, Landslide Hazard Zonation: A Review of Principles and Practice: Commission of Landslides of the IAEG, UNESCO, Paris, France, Natural Hazards No. 3, 61 p.

WANG, J. AND PENG, X. G., 2009, GIS-based landslide hazard zonation model and its application: Procedia Earth and Planetary Science, Vol. 1, No. 1, pp. 1198–1204. WARD, W. H., 1945, The stability of natural slopes: Geographical Journal, Vol. 105, pp. 170–196. WIECZOREK, G. F., 1984, Preparing a detailed landslide-inventory map for hazard evaluation and reduction: Environmental & widow. WOLDEAREGAY, K.; SCHUBERT, W.; KLIMA, K.; AND MOGESSI, A., 2005, Landslide hazards mitigation in the northern highlands of Ethiopia. In Proceedings International Conference on Landslide Risk Management: Vancouver, Canada. WOLFENDEN, E.; EBINGER, C.; YIRGU, G.; DEINO, A.; AND AYALEW, D., 2004, Evolution of the northern Main Ethiopian rift: Birth of a triple junction: Earth and Planetary Science Letters, Vol. 224, No. 1, pp. 213–228. ZARUBA, Q. AND MENCL, V., 1976, Engineering Geology: Developments in Geotechnical Engineering 10, Elsevier Scientiﬁc Publishing, Amsterdam, Netherlands, 504 p. ZARUBA, Q. AND MENCL, V., 1982, Landslides and Their Control, 2nd ed.: Developments in Geotechnical Engineering 31, Elsevier Scientiﬁc Publishing, Amsterdam, Netherlands, 324 p. ZVELEBIL, J.; SIMA, J.; AND VILIMEK, V., 2010, Geo-risk management for developing countries—Vulnerability to mass wasting in Jemma River basin, Ethiopia: Landslides, Vol. 7, No. 7, pp. 99–103.

Environmental & Engineering Geoscience, Vol. XXIII, No. 2, May 2017, pp. 137–151

151