E&EG - November 2017 Volume XXIII, Number 4

Environmental & Engineering Geoscience NOVEMBER 2017

VOLUME XXIII, NUMBER 4

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

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.

EDITORS

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

Cover photo Damage in Manitou Springs, Colorado from a 2013 debris flow resulting from the 2012 Waldo Canyon Fire. The debris flow swept a vehicle off a highway and killed its occupant approximately one mile upstream of this location. Photo by Paul Santi. – see article on page 298.

Environmental & Engineering Geoscience Volume 23, Number 4, November 2017 Table of Contents

243

Hydrological and Geophysical Investigation of Streamow Losses and Restoration Strategies in an Abandoned Mine Lands Setting Charles A. Cravotta III, Laura Sherrod, Daniel G. Galeone, Wayne G. Lehman, Terry E. Ackman and Alexa Kramer

275

A Design Method for Landslide Surface Water Drainage Control Benjamin D. Haugen

291

Consideration of the Validity of Debris-Flow Bulking Factors Holly Brunkal and Paul Santi

299

Thermal Remote Sensing for Moisture Content Monitoring of Mine Tailings: Laboratory Study Bonnie Zwissler, Thomas Oommen, Stan Vitton and Eric A. Seagren

314

Deterministic Three-Dimensional Rock Mass Fracture Modeling from Geo-Radar Survey: A Case Study in a Sandstone Quarry in Italy Mohamed Elkarmoty, Camilla Colla, Elena Gabrielli, Stefano Bondua and Roberto Bruno

333

Analysis and Prediction of Gas Recovery from Abandoned Underground Coal Mines in China Wei Li, Er-Lei Su, Yuanping Cheng, Rong Zhang, Zhengdong Liu, Paul L. Younger and Dongming Pan

345

Optimizing Digital Elevation Model Resolution Inputs and Number of Stream Gauges in Geographic Information System Predictions of Flood Inundation: A Case Study along the Illinois River, USA Anas Rabie, Eric Peterson, John Kostelnick and Rex Rowley

Hydrological and Geophysical Investigation of Streamflow Losses and Restoration Strategies in an Abandoned Mine Lands Setting CHARLES A. CRAVOTTA III U.S. Geological Survey, Pennsylvania Water Science Center, 215 Limekiln Road, New Cumberland, PA 17070

LAURA SHERROD Department of Physical Sciences, Kutztown University, P.O. Box 730, Kutztown, PA 19530

DANIEL G. GALEONE U.S. Geological Survey, Pennsylvania Water Science Center, 215 Limekiln Road, New Cumberland, PA 17070

WAYNE G. LEHMAN Schuylkill Conservation District, 1206 AG Center Drive, Pottsville, PA 17901

TERRY E. ACKMAN M T Water Management, Inc., 438 Old Clairton Road, Jefferson Hills, PA 15025

ALEXA KRAMER Schuylkill Headwaters Association, Inc., 1206 AG Center Drive, Pottsville, PA 17901

Key Terms: Legacy Coal Mining, Surface Geophysics, Surface-Water Hydrology, Infiltration, Abandoned Mine Drainage ABSTRACT Longitudinal discharge and water-quality campaigns (seepage runs) were combined with surface geophysical surveys, hyporheic zone temperature profiling, and watershed-scale hydrological monitoring to evaluate the locations, magnitude, and impact of stream-water losses from the West Creek sub-basin of the West West Branch Schuylkill River into the underground Oak Hill Mine complex that extends beneath the watershed divide.

Abandoned mine drainage, containing iron and other contaminants, from the Oak Hill Boreholes to the West Branch Schuylkill River was sustained during low-flow conditions and correlated to streamflow lost through the West Creek streambed. During high-flow conditions, streamflow was transmitted throughout West Creek; however, during low-flow conditions, all streamflow from the perennial headwaters was lost within the 300 to 600 m “upper reach,” where an 1889 mine map indicated steeply dipping coalbeds underlie the channel. During low-flow conditions, the channel within the “intermediate reach,”

700 to 1,650 m downstream, gained groundwater seepage with higher pH and specific conductance than upstream; however, all streamflow 1,650 to 2,050 m downstream was lost to underlying mines. Electrical resistivity and electromagnetic conductivity surveys indicated conductive zones beneath the upper reach, where flow loss occurred, and through the intermediate reach, where gains and losses occurred. Temperature probes at 0.06 to 0.10 m depth within the hyporheic zone of the intermediate reach indicated potential downward fluxes as high as 2.1 × 10−5 m/s. Cumulative streamflow lost from West Creek during seepage runs averaged 53.4 L/s, which equates to 19.3 percent of the daily average discharge of abandoned mine drainage from the Oak Hill Boreholes and a downward flux of 1.70 × 10−5 m/s across the 2.1 km by 1.5 m West Creek stream-channel area.

INTRODUCTION Problem The availability and quality of water are of global importance (WWAP, 2012) and can be severely limited in mined landscapes, where the natural hydrology and aquatic ecology can be extensively disturbed

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

243

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer

(Bernhardt et al., 2012; Feng et al., 2014). Dynamic hydrological variations commonly result in abandoned mine land (AML) watersheds because of rapid transfers of runoff, recharge, and discharge through the disturbed surface and subsurface combined with chemical interactions between the water and rock (Cifelli and Rauch, 1986; Dixon and Rauch, 1990; Carver and Rauch, 1994; and Cravotta et al., 2014). The formation and release of abandoned mine drainage (AMD), which is characterized by elevated concentrations of dissolved sulfate, iron, and other metals derived from the oxidation of pyrite and other sulfide minerals, can negatively impact aquatic environments and water supplies for decades after coal or metal mines have closed (Younger, 1997; Lambert et al., 2004; Cravotta, 2008; Mack et al., 2010; Nordstrom, 2011; and Burrows et al., 2015). Likewise, the loss of stream water to underlying mines can decrease the supply of clean, potable water while adding to the AMD volume and pollution loads further downstream in a watershed (Younger and Wolkersdorfer, 2004; Goode et al., 2011; and Cravotta et al., 2014).

An understanding of the inter-relations and aquatic impacts of streamflow losses and the associated AMD is needed to identify remediation priorities and watershed restoration strategies in AML areas. Specific information on plausible recharge and discharge locations and, particularly, the locations of stream leakage to underground mines can be used to design and implement stream restoration to abate streamflow losses (Ackman and Jones, 1991). If the streamflow can be transmitted from headwaters downstream, bypassing the mines, the stream habitat can be improved, the total volume of water that flows through the mines and emerges as AMD can be decreased, and pollutant transport within the watershed may be decreased. However, the locations and quantities of stream-water losses can be difficult to identify because streamflow can be gained or lost at numerous locations within a stream channel, or it can vary from gains to losses at a given location, depending on streamflow volume and other factors.

This paper reports on the use of hydrological and geophysical methods of investigation in an AML area (1) to identify the primary locations, quantities, and effects of streambed leakage on the streamflow and aquatic habitat and (2) to document quantitative and qualitative relations between the streamflow losses to inter-basin transfer as groundwater and the subsequent discharge of AMD in an adjacent watershed.

Background Instantaneous stream discharge and water-quality measurements during stable flow conditions along a defined length of a stream channel can be used to indicate

244

spatial variations in stream characteristics. These longitudinal discharge surveys are commonly referred to as seepage runs and are typically used to document the magnitude and locations of streamflow loss or gain due to groundwater seepage (e.g., Risser, 2006). The stream discharge at each station along the stream is the product of the measured cross-sectional area and the mean water velocity at that station (Rantz et al., 1982). Nonuniform geometry of the stream channel, irregularities and obstructions on the streambed, coarse or porous substrate that transmits a large fraction of flow through the hyporheic zone, and/or slow flow velocity or shallow water depth are common factors that contribute to errors in the area-velocity measurement of discharge in small streams. Although alternative methods, such as weirs and flumes, generally can improve the accuracy of discharge measurement (e.g., Rantz et al., 1982), these methods may be impractical for short-duration investigations or where channel or flow conditions are unstable or difficult to contain.

Chemical (tracer) dilution and natural water chemistry can be used to quantify inflows by groundwater or small tributaries to streams with complex channel characteristics (Schemel et al., 2006; Kimball et al., 2007; and Runkel et al., 2012). Groundwater entering a stream channel commonly has higher or lower specific conductance attributed to concentrations of specific dissolved ions, such as bicarbonate or sulfate, and/or may have different temperature than the stream water flowing down a channel from upstream locations.

Thus, groundwater inflows can dilute artificial tracers (if added) from upstream inputs while also affecting the temperature, specific conductance, and concentrations of certain solutes. However, where streamflow gains and losses take place along the flow path, spatial variations in chemical tracers may be difficult to interpret because a decrease in tracer concentration could result from dilution (inflow) as well as a loss of mass (outflow from the channel). Measurement of all the inflows and outflows along the flow path generally is not feasible. Thus, changes in the natural chemistry of surface water compared to groundwater at various locations and times can be instrumental in determining the amount of exchange among the stream, hyporheic zone, and groundwater (Cardenas, 2009; Bianchin et al., 2011), particularly when longitudinal streamflow measurements are conducted to quantify the gains and losses.

Surface geophysical techniques, such as electrical resistivity and electromagnetic surveys, can be used to characterize the subsurface stratigraphy, saturation state, and grain-size variations within hydrologic systems. Electrical resistivity methods are particularly useful for monitoring changes in saturation state because there is a direct correlation between the percent saturation and the bulk resistivity (Archie, 1942). Thus,

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

Investigation of Streamflow Loss Near Abandoned Mines

resistivity surveys have been successfully used to locate the water table (Reed et al., 1983), to record temporal and spatial variations in saturation and flow through the hyporheic zone and deeper subsurface (Nyquist et al., 2008; Clifford and Binley, 2010; Sherrod et al., 2012; Toran et al., 2012, 2013; and Ward et al., 2012), to identify groundwater seepage zones in streambed and lakebed sediments (Ackman and Jones 1991; Nyquist et al., 2008; and Toran et al., 2015), and to delineate preferential subsurface flow paths (Hagrey and Michaelsen, 1999; Yang et al., 2000). Likewise, electromagnetic conductivity measurements have proven useful to delineate zones of streambed leakage and shallow subsurface flow paths (Ackman and Jones, 1991; Grgich et al., 2004). Although surface geophysical methods can be used to estimate quantitative hydrologic properties (Binley et al., 2002a, 2002b; Vanderborght et al., 2005; Linde et al., 2006; and Johnson et al., 2009), it is often possible to glean useful qualitative information from results (Sherrod et al., 2012).

Time-series measurements of temperature within the hyporheic zone can indicate when the streambed becomes dry or water saturated and can be used to estimate the potential flux of water through streambed sediments (Constantz et al., 2001; Burkholder et al., 2008; Hyun et al., 2011; Briggs et al., 2012a, 2012b; and Daniluk et al., 2013). The vertical temperature gradient can be interpreted to quantify upward or downward fluxes corresponding to streamflow gains or losses, respectively (Gordon et al., 2012). Thus, a combined approach of seepage runs, surface geophysical surveys, and streambed temperature logging could be used to locate predominant zones of streambed leakage and to quantify the spatial and temporal variations in the flux through the streambed. Such information can then be used by watershed managers and hydraulic engineers to develop stream restoration strategies.

STUDY AREA The Schuylkill River originates in uplands of Schuylkill and Carbon Counties in eastern Pennsylvania and flows more than 200 km southeastward to the Delaware River at Philadelphia (Figure 1). Along its course, the Schuylkill River is an important resource used for recreational fishing and boating, cooling water at thermoelectric generation facilities, and drinking water to more than 1.5 million people (Schuylkill Action Network, 2008). Nevertheless, the headwaters area of the Schuylkill River, referred to as the upper Schuylkill River, is underlain by the extensively mined Southern Anthracite Coalfield (Biesecker et al., 1968; Growitz et al., 1985). Environmental degradation associated with the legacy mining affects local and downstream aquatic resources.

Structurally, the Southern Anthracite Coalfield is a downwarped synclinorium (canoe-shaped fold), the axes of which generally parallel the northeastsouthwest–trending ridges and valleys in the region (Wood et al., 1968). The coal-bearing Pennsylvanianage rocks are commonly exposed on the valley sides and underlie the valleys. More than a dozen commercial coalbeds in the study area range in thickness from 2 to 3 m (Pennsylvania Geological Survey, 1889; Wood et al., 1968). Most of the anthracite mines in the area were developed during the mid-1800s to mid-1900s by the underground “room-and-pillar” method, which employed a complex network of shafts and tunnels within coal and across intervening strata to connect multiple coalbeds. Groundwater was typically pumped to the surface or to drainage tunnels during active mining, and, in various locations, streams crossing over the mines were rerouted to reduce infiltration and associated pumping costs (Ash et al., 1953). Unmined walls of coal, or “barrier pillars,” usually were left intact at the mine boundaries that later acted as underground dams, restricting the flow of groundwater between adjacent mine complexes (Ash et al., 1949; Ash and Kynor, 1953) (Figure 1).

The abandoned underground mines in the upper Schuylkill River basin now are extensively flooded by groundwater. The flooded mine workings, referred to as mine pools, are drained by gravity through tunnels, boreholes, and fractures to numerous AMD outfalls on the Schuylkill River and its tributaries (Biesecker et al., 1968; Growitz et al., 1985). Elevated concentrations of sulfate, iron, and other metals in the AMD continue to degrade the quality of receiving streams (Cravotta et al., 2014). Consequently, 200 km of stream segments in the upper Schuylkill River basin are designated “impaired by AMD” and listed under the Clean Water Act section 303(d) (Pennsylvania Department of Environmental Protection, 2014).

Spatial and temporal hydrological variability can be extreme in AML areas. For example, surface water and groundwater within the upper Schuylkill River basin are intimately connected through karst-like features that facilitate dynamic changes in groundwater recharge, storage, and discharge and that permit the inter-basin transfer of groundwater. Perennial streams crossing the abandoned underground mines beneath the West West Branch Schuylkill River and West Branch Schuylkill River (Figure 1) lose water and may stop flowing because of streambed leakage to the mines, while streamflow in downstream segments of the West Branch Schuylkill River is sustained by contaminated AMD outfalls from the Pine Knot Tunnel (PKN) and the Oak Hill Boreholes (OAK) (Cravotta et al., 2014). Because of the inter-basin transfer of groundwater through the underground mines, the streamflow

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

245

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer

Figure 1. Topographic drainage basins (colored polygons), flooded extent of underground mines (dark gray shade), and hydrological monitoring sites in the upper Schuylkill River basin, Southern Anthracite Coalfield, Pennsylvania. Approximate areal extent of flooded underground mines, including barrier pillar locations and flow directions, was reported by Biesecker et al. (1968). Descriptions of monitoring sites are given in Table A1.

yield (discharge divided by topographic drainage area) for the West Branch Schuylkill River is larger than expected, and that of the West West Branch Schuylkill River is smaller than expected compared to nearby, unmined areas (Cravotta et al., 2014), as explained below.

West Creek is a perennial tributary, upstream from Muddy Branch, in the headwaters of the West West Branch Schuylkill River that drains mostly forested land (82 percent of the 22.4 km2 watershed) and is unaffected by AMD (Figure 1). However, during low-flow conditions, segments of West Creek that cross underground mines downstream from the perennial headwaters (Figure 1) can lose all flow through the stream channel, resulting in intermittently dry segments (Figure 2).

Streamflow of West Creek in the segment extending southeastward (downstream) from Forestville (WC4) to Phoenix Park (WC9) has been hypothesized to leak

246

from the stream channel to the underlying Oak Hill Mine complex (composed of the Lytle and Oak Hill Collieries, plus part of the Phoenix Park Colliery) and then flow as groundwater westward beneath the topographic watershed divide until it exits as AMD from the Oak Hill Boreholes outfall to the adjacent West Branch Schuylkill River (Figures 1 and 3) (Cravotta et al., 2014).

MATERIALS AND METHODS This study employed multiple, complementary methods of investigation: (1) to identify the primary locations, quantities, and effects of streambed leakage on the streamflow and aquatic habitat of West Creek, and (2) to document potential quantitative and qualitative relations among the streamflow losses on West Creek, the discharge of AMD from OAK, and the effects on

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

Investigation of Streamflow Loss Near Abandoned Mines

Figure 2. Photographs of West Creek at WC4 above mined area (A, B) and at WC8 below mined area (C, D). Streamflow at WC4 is perennial; (A) high-flow conditions, (B) low-flow conditions. Streamflow at WC8 is intermittent; (C) high-flow conditions, (D) low-flow conditions.

downstream waters within the West Branch Schuylkill River and West West Branch Schuylkill River subbasins (Figure 1).

Discharge Instantaneous and continuous streamflow measurements were taken at numerous locations throughout the study area. To investigate the magnitude and approximate locations of streamflow lost or gained along West Creek, standard area-velocity techniques were used to measure instantaneous discharge on a given date at multiple locations along the 3.1 km segment from WC4

to WC9 (Figure 1 and Table A1). These synoptic seepage runs were conducted for a range of flow conditions on nine dates during April 2012 to July 2015. On a given date, all the measurements were conducted using a top set wading rod equipped with either a Price pygmy current meter, SonTek FlowTracker2 (FT2) handheld acoustic Doppler velocimeter (ADV), or Marsh McBirney Model 2000-11 Flo-Mate portable velocity meter.

Replicate measurements at a single cross section using different meters indicated consistent results for a given flow condition.

Streamflow-gauging stations for continuous monitoring of stream stage and discharge were established

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

247

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer

Figure 3. Detailed map views of West Creek flow-loss study area showing hydrological and geophysical survey site locations on: (A) 1:24,000 Minersville, PA, topographic map overlying flooded underground mine pools (Biesecker et al., 1968); and (B) historical underground mine map (Pennsylvania Geological Survey, 1889). Polygonal symbols on mine map represent underground mine workings. Current stream route (as mapped in 1955) is shown as blue trace on A and B. Point symbols indicate locations of streamflow gauges, seepage run sites, and the start and end of electrical resistivity survey segments (numbered 0–13). Graduated symbols for flow data indicate median normalized value compared to WC4 for seepage runs during May 2014 through July 2015.

248

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

Investigation of Streamflow Loss Near Abandoned Mines

by the U.S. Geological Survey (USGS) in 2014 at four locations on West Creek, near the top (WC4), bottom (WC9), and intermediate locations (WC5 and WC8) along the 3.1 km segment that crossed the Oak Hill Mine complex (Table A1). Other gauging stations were previously established in 2005–2006 at the nearby AMD outfall from PKN and at sites upstream (WB1) and downstream (WB3) from PKN on the West Branch Schuylkill River and the West West Branch (WWB) Schuylkill River and in 2014 at OAK (Figure 1 and Table A1). The four West Creek stations are topographically upstream from the streamflow gauge at WWB and are inferred to be hydrologically linked to the PKN and OAK outfalls and the associated downstream gauges on the West Branch Schuylkill River (WB2, WB3) by the underground mine pools that extend beneath the topographic divide between the two watersheds (Figure 1). At each streamflow gauging station, a vertical staff gauge and a submersible, vented pressure transducer were installed to measure stream stage (water depth). The transducers recorded the stage to a precision of 0.006 m or better at 15 minute intervals. The transducer data were downloaded, and streamflow was measured periodically from January 2012 through September 2015 for the current study. Discharge data are available in the USGS National Water Information System (NWIS; http://dx.doi.org/10.5066/F7P55KJN).

During the period of operation, instantaneous discharge at each gauging station was measured over a range of low-to-moderate flow conditions to develop stage-discharge ratings for each site (Rantz et al., 1982).

Extrapolation of stage-discharge ratings for high-flow conditions was based on established ratings for the nearby long-term streamflow gauging station on the Schuylkill River at Landingville (SRL), which is 15 km downstream from WWB and WB3 (Table A1). The daily average streamflow values at each station were used with the PART computer program (Rutledge, 1998; Risser et al., 2005) to estimate the annual hydrologic budget for the contributing area above the station, including the percentages of total streamflow that were base flow and runoff. For the purposes of comparison with the limited flow record of the newer gauges on West Creek, the hydrograph analysis was completed for the 1 year period of July 2014 to June 2015 and also for the period October 2005 through September 2015, if available. By dividing the flow rate by the topographic drainage area or contributing area, the normalized flow rates (yields) could be compared to precipitation data for the period of interest and to those for other stations. Precipitation data for 2005–2015 were available from nearby USGS streamflow gauging stations (01469500, 01470500, 01468500) and a USGS weather station (403628076134201) within the upper Schuylkill

River basin (USGS NWIS, http://dx.doi.org/ 10.5066/F7P55KJN); the average of daily totals at those stations were used for analysis.

Water Quality Water-quality monitoring was conducted repeatedly at streamflow gauges and other selected sites to document the spatial and temporal variations in stream characteristics. Data on water-quality constituents at each synoptic streamflow site or streamflow gauging station were collected when discharge was measured or stage data were downloaded from pressure transducers. A YSI 556 multiparameter sonde was used to measure temperature, pH, specific conductance (SC), dissolved oxygen (DO), and oxidation-reduction potential (ORP) where flow was concentrated (typically near the staff gauge). The pH/ORP electrode was calibrated in pH 4.0, 7.0, and 10.0 buffer solutions and in ZoBell’s solution. Values of ORP were corrected to 25◦ C relative to the standard hydrogen electrode (Eh) and used to compute the activities of Fe (II) and Fe (III) species from dissolved iron in accordance with the Nernst equation (Nordstrom, 1977).

Water-quality grab samples for chemical analysis were collected at least quarterly at the streamflow gauges on West Creek during 2014–2015 and the associated gauges on the West West Branch and West Branch Schuylkill River during 2012–2015 (Table A1; data available in USGS NWIS, http://dx.doi.org/ 10.5066/F7P55KJN). The alkalinity of the unfiltered water samples was titrated in the field using 0.16 N H2 SO4 to a fixed end point pH of 4.5 (American Public Health Association, 1998). Concentrations of major anions (SO4 , Cl) in 0.45-␮m-filtered, unpreserved subsamples were analyzed by ion chromatography (IC), and concentrations of major cations (Ca, Mg, Na, K) and selected trace metals (Fe, Mn, Al, Ni, Zn) in unfiltered, acidified subsamples and in corresponding 0.45-␮m-filtered, acidified subsamples were analyzed by inductively coupled plasma–atomic emission spectroscopy (ICP-AES) or inductively coupled plasma– mass spectrometry (ICP-MS) using standard laboratory methods (Fishman and Friedman, 1989). Anion and cation analyses were conducted at the USGS National Water Quality Laboratory in Denver, CO, or the Actlabs Laboratory in Toronto, Ontario. USGS Standard Reference Water Samples (SRWS) submitted to each of the laboratories indicated comparable results for the major and trace constituents. The net-acidity concentration was calculated using selected data (pH, Fe, Mn, Al, and alkalinity) following methods of Kirby and Cravotta (2005). The pH and concentrations of selected trace metals (Zn and Ni) were compared to criteria for freshwater aquatic life, after adjusting for

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

249

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer

sample hardness (U.S. Environmental Protection Agency, 2013).

Electrical Resistivity Surface geophysical surveys were conducted repeatedly along a 3.1 km segment of West Creek spanning the Oak Hill Mine complex. Electrical resistivity surveys along the stream channel were performed using an MPT DAS-1 Electrical Impedance Tomography System with 32 electrodes spaced at 5 m intervals (http://mpt3d.com/das1.html). Initial surveys were conducted along the “upper reach” from 230 to 530 m downstream from WC4 to WC5, in the area of Forestville (Figure 1, between points labeled 0 to 2), where complete flow loss had been observed.

These surveys were conducted over a range of low- to moderate-flow conditions at approximately the same locations on April 27, September 13, and September 27, 2012. Additional resistivity surveys were conducted during moderately high-flow conditions May 12–15, 2014, along the 2.1 km segment that begins in the upper segment at approximately 400 m downstream from the streamflow gauge at WC4 and ends in the lower segment approximately 2540 m downstream from WC4 and 270 m downstream from WC8 (Figure 1). Eleven surveys of 155 m length and a twelfth of 110 m length were numbered 1 to 12 in downstream sequence (Figure 1). The surveys were generally conducted along straight segments and skipped segments where sharp bends were present. Thus, the ending and starting locations of sequential surveys did not always coincide (e.g., survey 1 was located between resistivity points 1 and 2; survey 12 was located between resistivity points 12 and 13). The electrical resistivity data are available at http://faculty.kutztown.edu/sherrod/ PublicationMaterials.

A dipole-dipole array, which is useful for resolving spatially confined objects in the near subsurface (Oldenburg and Li, 1999; Schrott and Sass, 2008), was chosen for collection and processing of the resistivity data.

A base electrode ‘a’ spacing of 5 m was used with a maximum n value of 6. The a spacing was expanded to values of 10 m, 15 m, and 20 m as possible with the 32 electrodes to achieve greater depth of penetration. The resistivity data were inverted with ERTLab (Geostudi Astier srl and Multi-Phase Technologies LLC, 2006).

The data for each survey were filtered to remove readings from electrodes with poor contact (the pebbleto boulder-sized sediment of the streambed sometimes inhibited the passage of current into the subsurface, especially when the stream channel was dry) or with a voltage of less than 0.1 mV, projected onto a mesh having one quarter of the electrode spacing, and inverted with the numerical core set to 10 maximum outer inver-

250

sion iterations, 15 maximum internal iterations, and a tolerance of 0.001 for the internal iterations. The initial roughness factor was set to 10, and the data percent error for noise was set to 3 percent, with a constant error term of 0.001 mV. These parameters constrain the inversion process and are described in detail in the ERTLab documentation (Geostudi Astier srl and Multi-Phase Technologies LLC, 2006). All surveys were allowed to converge to an inversion solution with a data residual of less than the chi-square value, indicating a good fit of the inversion model to the raw data and forward model (LaBrecque et al., 1996).

Electromagnetic Conductivity Surface electromagnetic (EM) survey data were collected by using an EM-31 apparent conductivity device manufactured by Geonics along most of the same 2.1 km segment of West Creek that had been surveyed in 2014 using resistivity (Figure 1; stations 0 to 11). The EM survey was completed by wading the stream during moderate-flow conditions on December 7 and 8, 2015, and recording conductivity readings at 10 m spacing along the approximate midpoint in the stream channel. Both horizontal dipole (HD) and vertical dipole (VD) measurements were obtained by holding the instrument in a horizontal position 1 m above the streambed and then rotating the instrument 90 degrees (along the horizontal axis) to take the second reading. The orientation of the pole-like instrument, when taking readings, was parallel to the stream channel (with the transmitter pointing downstream and the receiver pointing upstream).

The effective depths of exploration for the horizontal and vertical dipoles are approximately 6 m and 3 m, respectively, beneath the stream channel. Ackman and Jones (1991) and Grgich et al. (2004) previously demonstrated that such measurements could indicate zones of relatively high conductivity associated with flow loss through vertical water-filled fractures and water-filled subsidence-related voids within the first 6 m beneath the stream channel. The EM survey data are available at http://faculty.kutztown.edu/sherrod/ PublicationMaterials.

Streambed Temperature Profiles To document potential temporal variations in streambed leakage, temperature probes were installed into the streambed along the intermediate segment of West Creek within the boundaries of resistivity surveys 6, 7, 9, 10, and 11 (Figure 1). Each probe was constructed of metal conduit (pipe) with cutouts at fixed distances to accommodate datalogging temperature sensors (Thermocron iButtons

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

Investigation of Streamflow Loss Near Abandoned Mines

model #DS1922L, Embedded Data Systems, www. embeddeddatasystems.com). The uppermost sensor was installed at the top of streambed, such that the other three sensors were at depths of approximately 0.03 m, 0.06 m, and 0.10 m within the streambed. The temperature sensors were programmed to record data at 20 minute intervals using a resolution of ± 0.0625◦ C at an accuracy of ± 0.5◦ C. Two sets of probes were installed, each for a 2 month period: the first set during September 12–November 6, 2014, and the second during November 11, 2014–January 5, 2015. During the first period, stream-water levels were low. Although water was pooled in some areas, the stream was not flowing at the time of installation nor at the time of extraction. The second set of temperature probes was installed in approximately the same location as the first set. The stream was not flowing at the time of installation, but it was flowing at a moderate stage at the time of extraction of the second set of probes.

The Matlab program VFLUX (Gordon et al., 2012) was used to calculate the vertical flux through the streambed at high temporal resolution. The flux was calculated for the intervals between the sensors, with the deepest (0.085 m) corresponding to the depth interval 0.06 to 0.10 m. Soil parameters in the program were set to the default values (porosity, N = 0.28; thermal dispersivity, Beta = 0.001 m; baseline thermal conductivity, kCal = 0.0045 cal/[s·cm·◦ C]; volumetric heat capacity of sediment, CS Cal = 0.5 cal/[cm3 ·◦ C]; volumetric heat capacity of water, CW Cal = 1.0 cal/[cm3 ·◦ C]), which are within the range of expected values for site conditions with large rocks and boulders overlying sandy sediment with minor silts and clays (Lapham, 1989). The temperature data are available at http://faculty.kutztown.edu/sherrod/ PublicationMaterials.

RESULTS AND DISCUSSION The results from longitudinal streamflow and waterquality surveys, surface geophysical surveys, streambed temperature profiles, continuous streamflow gauging, and quarterly water-quality monitoring at selected stations within West Creek and the surrounding watersheds were obtained independently. Thus, the results can generally be presented and discussed independently, but herein they are integrated to provide context for discussion and interpretation of the information provided by the different methods.

Streamflow and Water Quality Vary Spatially and Temporally Synoptic measurements of streamflow and water quality along the 3.1 km study segment of West Creek

(between WC4 and WC9) documented multiple locations and variations in the quantities of losses and gains in streamflow. Repeated measurements on different dates at the seepage survey stations, which generally coincided with the end points of the resistivity survey segments (Figure 3A), indicated substantial spatial and temporal variability. Thus, to reveal typical downstream changes, the streamflow values at downstream points were normalized to (divided by) the streamflow value at the most upstream station on that date, and the median normalized streamflow and the median water quality at each station along the main channel were emphasized (Figures 3B and 4).

Measurable streamflow was observed along the entire length of West Creek during typical base-flow (median) to high-flow conditions (Figures 3B and 4) and was recorded continuously at the upstream station, WC4, and downstream station, WC9 (Figure 5). However, during low-flow conditions, complete losses of streamflow in the 300 to 600 m “upper reach” between stations WC4 and WC5, between resistivity stations 0 and 2, and dry conditions at WC5 and various intermediate points to WC8 were observed, specifically during low-flow seepage runs in April and September 2012 and August and November 2014 (Figures 4 and 5A).

Streamflow was added to West Creek by intermittent contributions from three small, unnamed tributaries and diffuse groundwater inflows within the intermediate segment from 700 to 2,350 m downstream from WC4. For example, despite intermittent streamflow (intermittent dry channel) upstream and downstream, continuous groundwater inflows resulted in perennial streamflow in a short segment 1,400 to 1,500 m downstream from WC4, between resistivity stations 6 and 7.

Nevertheless, streamflow 1,650 to 2,050 m downstream from WC4, between stations 7 and 10, was intermittent and completely lost from the channel during low-flow conditions.

The locations where streamflow frequently disappeared, between resistivity stations 0 and 2 (upstream from WC5) or stations 7 and 10 (upstream from WC8), extended further upstream within these segments as the amount of streamflow transmitted from the headwaters decreased. Likewise, at moderate- to high-flow conditions, only partial flow losses or minor gains were recorded in these “losing” segments (Figures 3B, 4, and 5B). The net change in streamflow volume could be positive or negative, depending on whether losses within a measurement segment exceeded the combined volumes of the inflows and any water transmitted from upstream. Other investigations of streams underlain by longwall mines in the northern Appalachian Coalfield (e.g., Cifelli and Rauch, 1986; Dixon and Rauch, 1990; and Carver and Rauch, 1994) showed some segments of stream reaches that dried up during

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

251

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer

Figure 4. Summary results of seepage surveys along the 3.1 km study length of West Creek from WC4 (0 m) to WC9 (3130 m) during May 2014 to July 2015. Streamflow was normalized by dividing the measurement at each point by the streamflow value at WC4 on the date of the survey. Point symbols indicate individual streamflow measurements, solid symbols indicate values within the channel, and open symbols indicate measurements at tributaries. Electrical resistivity survey locations, illustrated at the top, provide spatial context; detailed results of the resistivity surveys are given in Figures 7, 8, and 10.

low-flow conditions could show gains during high-flow conditions while still losing some water to the underground mine(s). Because both gains and losses could have taken place between flow measurement stations on West Creek on a given date, the water-quality variations along the flow path augmented the streamflow measurements.

The pH and SC of West Creek increased progressively downstream from WC5 to WC9 (Figure 4), primarily because of volumetrically small inflows of highionic-strength, net-alkaline groundwater seepage that were of increasing importance during low-flow conditions. The high SC and pH of the groundwater and other inflows in the intermediate segment contrasted with the low SC and pH of water transmitted downstream from the headwaters. Thus, the chemical data revealed gains in streamflow downstream or within the same segments that experienced flow losses. Because of such gains combined with downstream losses along the streamflow path, the total streamflow lost from West Creek along its flow path was greater than the simple difference between streamflow measured at the most upstream point at WC4 and the downstream points at WC8 or WC9 (Figure 4 and Table A2). (The total flow entering West Creek upstream from WC8 or WC9 is the sum of that at WC4 plus any gains from tributaries and groundwater inflows documented during seepage surveys.)

252

Despite a high-quality physical aquatic habitat at the most upstream site on West Creek, WC4 (Figure 2; see Supplemental Material), the water chemistry was net acidic (median 3.9 mg/L as CaCO3 ) with relatively low pH (4.5), low SC (70 â?ŽS/cm), low concentrations of sulfate (13.5 mg/L), iron (0.045 mg/L), and hardness (9.9 mg/L as CaCO3 ), and elevated concentrations of dissolved aluminum (0.40 mg/L) and zinc (0.034 mg/L) compared to the water sampled in the intermediate to lower segments of West Creek (WC8, WC9; Table A3). Because of its acidity and moderately elevated concentrations of aluminum and zinc, the water at WC4 did not meet criterion continuous concentration (CCC) thresholds for protection of freshwater aquatic life (Table A3) and did not support fish (as reported in the Supplemental Material). Water quality similar to that at WC4 was observed at WC5, when streamflow was transmitted downstream through the upper segment to WC5. However, during low base-flow conditions, streamflow disappeared along the stream channel before reaching station WC5. Nevertheless, further downstream at points along the flow path to WC8 and WC9, streamflow with progressively higher pH and SC appeared within the channel as evidence of streamflow gains (Figure 4).

Although streamflow was intermittent at WC5 and WC8, perennial streamflow was maintained within a segment between resistivity end-points 6 and 7 in the

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

Investigation of Streamflow Loss Near Abandoned Mines

Figure 5. Daily discharge hydrographs for West Branch (WB3), West West Branch (WWB), Pine Knot Tunnel (PKN), Oak Hill Boreholes (OAK), and three streamflow gauges on West Creek (WC4, WC5, and WC9, in downstream order) during 2012–2015 study period: (A) daily discharge, with symbols indicating the measured flows at WC4 and WC9 plus the estimated cumulative inflow (WC9_gains) from WC4 to WC9 (flow at WC4 plus total of inflows to WC9) during synoptic seepage surveys, with vertical dashed lines for resistivity survey dates, and (B) comparison of daily discharge at OAK, WC4, WC5, and WC9, plus daily precipitation data for June 2014–September 2015.

intermediate segment between WC5 and WC8 (Figures 3 and 4) and in a longer segment in the downstream segment below end-point 12 to WC9. As explained in more detail later, these perennial stream

segments coincided with the original, natural channel route, whereas the adjoining intermittent segments coincided with zones of historical coal outcrops or shallow underground mining (Figure 3B). Water-quality

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

253

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer

samples at WC9 were net alkaline (median net acidity <0) with near-neutral pH (6.6), elevated SC (315 ␮S/cm), elevated concentrations of sulfate (99 mg/L) and hardness (108 mg/L as CaCO3 ), moderate concentration of iron (0.075 mg/L), and low concentration of aluminum (<0.1 mg/L) compared to the median values for upstream samples (Table A3). However, during high-flow conditions, a larger proportion of streamflow was transmitted from the headwaters to downstream points on West Creek. Under such conditions, the downstream water quality retained much of its upstream character (low pH, low SC, low hardness) because proportionally smaller inflows of groundwater had less influence on the overall downstream water quality. During high flows, the streamflow and water-quality measurements indicated net gains from WC4 to WC9, despite streamflow leakage through the streambed to the mine(s). Because of the streambed leakage, the rate of streamflow gain (e.g., L/s/km2 of drainage area), or yield, was less than expected if no mining had occurred (Tables 1 and A2, see “Annual Water Budget for Study Period” section).

The stream water near the mouth of the West West Branch (WWB), downstream from the confluence of West Creek and Muddy Branch, was consistently net alkaline with near-neutral pH (median 7.5) and relatively low concentrations of dissolved metals compared to upstream sites and CCC thresholds (Table A3). Excess acidity, if present, from West Creek was offset by alkalinity contributions downstream from WC9, including additional groundwater seepage and a large net-alkaline inflow from the Muddy Branch (Figure 1).

Likewise, the downstream water on the West Branch (WB3) near the confluence with WWB was consistently net alkaline with near-neutral pH; however, WB3 exhibited elevated concentrations of dissolved metals and other contaminants from AMD discharged by OAK and PKN (Table A3).

During the 13 seepage runs, an average of 13.7 L/s of “clean” stream water was lost from West Creek over the 300 m segment between stations WC4 and WC5, and an average total of 53.4 L/s was lost over the entire 2.1 km segment that crosses the Oak Hill Mine complex between stations WC4 and WC9 (Table A3). These losses equate to 6.9 percent and 19.3 percent, respectively, of the daily average discharge of 221 L/s from the Oak Hill Boreholes during the dates of the seepage runs (Table A2). Assuming the 300-m-long losing segment between WC4 and WC5 is 1.5 m wide, an average downward flux of 3.05 × 10−5 m/s through the streambed is computed (by dividing the lost streamflow of 13.7 L/s, or 0.0137 m3 /s, by the 450 m2 area of the channel). Likewise, assuming the entire 2.1 km segment between WC4 and WC9 is 1.5 m wide, the cumulative loss from West Creek equates to an average

254

Figure 6. Comparison of daily discharge from Oak Hill Boreholes (OAK) with cumulative streamflow lost from West Creek between WC4 above mined area and WC9 below mined area, during March 2012–September 2015. Note that during low-flow conditions, the discharge of OAK is sustained at approximately 140–150 L/s (intercept), and it increases with the streamflow lost from West Creek.

downward flux of 1.70 × 10−5 m/s across the total 3,150 m2 stream-channel area.

During low- to moderate-flow conditions, the estimated quantities of streamflow lost from West Creek during seepage runs were closely correlated with the corresponding daily discharge from OAK (Figure 6); however, for the two seepage runs conducted during high-flow conditions (January 18, 2013, and April 23, 2014), the cumulative losses from West Creek deviated from this correlation and approached the magnitude of the discharge from OAK on those dates (Figure 6).

The daily discharge from OAK was generally correlated with streamflow at nearby stream gauges (Figure 5A); however, the peaks and troughs in the hydrograph for OAK were subdued with extended recessions compared to those for West Creek (Figure 5B). This difference in the hydrographs implies that recharge by streambed leakage from West Creek may be stored in the Oak Hill Mine pool and released gradually as discharge from OAK. Similar surface-water/groundwater interactions had been previously reported for the West Branch and Pine Knot Mine pool (Cravotta et al., 2014).

Annual Water Budget for Study Period Wide variations in the annual streamflow yields and in the proportions of streamflow that may be attributed to base flow or runoff were exhibited at gauging stations on West Creek and nearby locations within the upper Schuylkill River basin during July 1, 2014, through

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

Investigation of Streamflow Loss Near Abandoned Mines Table 1. Hydrograph-separation analysis and components of the annual hydrologic budget for continuous streamflow gauging stations in the West Creek and associated watersheds of upper Schuylkill River basin, Schuylkill and Berks Counties, PA, July 1, 2014–June 30, 2015 (site descriptions are in Figure 1).a

Map ID

Gauge Location

Station

Drainage Area

Mean Streamflowb

Number

(km2 )

(L/s) (cm/yr)

West West Branch Schuylkill River WC4 West Cr above 01467830 Forestville WC5 West Cr at 01467832 Forestville WC8 West Cr at Main 01467835 St Phoenix Pk WC9 West Cr at 01467837 Ramtown Rd Phoenix Pk WWB West Cr West 01467861 Branch Schuylkill R West Branch Schuylkill River WB1 WB Schuylkill 01467688 River above Pine Knot PKN Pine Knot 01467689 Disch 500 m below Tunnel PKNWB PKN + WB1 OAK Oak Hill Disch 01467691 200 m below WB3 WB Schuylkill 01467752 River above WWB Schuylkill River SRL Schuylkill River at 01468500 Landingville

Stream-flow Base-flow Runoff Indexc Mean Baseflowd Index Mean Runoffe Index (%)

(L/s)

(cm/yr)

(%)

(L/s) (cm/yr)

(%)

13.3

69

16.4

16.1

48

11.4

69.5

21

5.0

30.5

13.7

63

14.5

14.2

34

7.8

53.8

29

6.7

46.2

18.9

82

13.7

13.4

46

7.7

56.2

36

6.0

43.8

22.3

84

11.9

11.7

50

7.1

59.7

34

4.8

40.3

48.5

752

48.9

47.9

639

41.6

85.1

113

7.3

14.9

49.8

365

23.1

22.6

281

17.8

77.1

84

5.3

22.9

49.1

484

31.1

30.5

472

30.3

97.4

12

0.8

2.6

49.8 21.9

849 216

53.8 31.1

52.7 30.5

768 212

48.7 30.5

90.5 98.1

81 4

5.1 0.6

9.5 1.9

61.7

1368

70.0

68.6

1218

62.3

89.0

150

7.7

11.0

340.5

6103

56.6

55.4

4933

45.7

80.7

1170

10.9

19.3

a Hydrograph separation was conducted using the “PART” computer program (Rutledge, 1998) to divide annual streamflow into base-flow and runoff contributions on the basis of daily average flow values during July 1, 2014, through June 30, 2015. b Streamflow (yield) expressed as centimeters per year by dividing streamflow in liters per second by the drainage area in square kilometers and then multiplying by the factor 3.156. c Streamflow index was computed as the ratio, expressed as percent, of total annual streamflow yield to average total annual rainfall of 102.1 cm/yr based on daily rainfall at local USGS streamflow gauging stations (01469500, 01470500, 01468500) and weather station 403628076134201. d Base flow is expressed as liters per second, centimeters per year, and percent of total annual streamflow (base-flow index). e Runoff, computed by subtracting the base flow from total streamflow, is expressed as liters per second, centimeters per year, and percent of total annual streamflow (runoff index).

June 30, 2015 (Table 1). Instead of relatively constant yields for nested or neighboring stations based on mean streamflow, which may be expected for the same rainfall under unaltered conditions, large differences were computed, ranging from 11.9 cm/yr for West Creek (WC9) to 70.0 cm/yr for West Branch (WB3) (Table 1).

The low discharge yields of the West Creek sub-basin (WC9) and the West West Branch Schuylkill River basin (WWB) are inferred to result from streamflow losses to the Oak Hill Mine complex plus losses to the Pine Knot Mine complex, which extends beneath the headwaters area (Figure 1), plus additional diversions.

In contrast, the high discharge yield of the West Branch

Schuylkill River at WB3 is hypothesized to result from the AMD outfalls at PKN and OAK that discharge water from the respective mines to the West Branch Schuylkill River.

Although the drainage area of West Creek at WC9 (22.3 km2 ) was approximately half of that downstream on the West West Branch Schuylkill River at WWB (48.5 km2 ), the mean daily streamflow at WC9 was only 11 percent of that at WWB. The corresponding annual streamflow yields were 48.9, 11.9, and 16.4 cm/yr for WWB, WC9, and WC4, respectively. Although a low yield at WC9 was anticipated, considering the previously documented streamflow losses, the low yield at

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

255

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer

WC4 was not. Approximately 1.5 km upstream from WC4, water is diverted from the Crystal Reservoir on West Creek for public supply (Figure 1). Furthermore, the drainage area upstream from the Crystal Reservoir is underlain by the Pine Knot Mine complex, which also intercepts runoff and recharge in this area (Figure 1).

The annual streamflow yield of 23.1 cm/yr at West Branch above the Pine Knot Tunnel (WB1) was less than half of the 56.6 cm/yr reference value for the downstream gauge on the Schuylkill River at Landingville (SRL), which drains the 340.5 km2 area including the West West Branch and West Branch Schuylkill River basins (Table 1). The missing streamflow from the 49.8 km2 watershed contributing to WB1 was discharged from PKN, which enters the West Branch about 70 m downstream from WB1. The combined flows of PKN and WB1 above the Pine Knot Tunnel (PKN + WB1), expressed as a yield of 53.8 cm/yr, were comparable to that for the SRL reference gauge outside the mined area, indicating surface water lost to the Pine Knot Mine complex was restored as AMD from PKN. However, downstream from OAK, the West Branch (WB3) had a much larger yield (70.0 cm/yr) than the SRL reference value or the adjacent West West Branch Schuylkill River (WWB, 48.9 cm/yr) at their confluence (Table 1). The excessive yield at WB3 can be attributed to the gain in flow from OAK. If the Oak Hill Mine complex did not capture the flow loss from the upper portions of the West Creek basin, this water would have discharged at WWB. At SRL, outside the mined area, the annual streamflow yield, base-flow yield, and base-flow index (baseflow/streamflow) were 56.6 cm/yr, 45.7 cm/yr, and 80.7 percent, respectively, from July 1, 2014, to June 30, 2015 (Table 1). The streamflow yield of 56.6 cm/yr at SRL equates to 55.4 percent of the 102.1 cm/yr total annual precipitation during the budget period. Assuming this reference yield of 56.6 cm/yr would apply to the upstream sub-basins, the corresponding streamflows based on the topographic drainage areas at WB3 and WWB are estimated to be 1107 L/s and 870 L/s, respectively. The difference between the measured and estimated flows for WB3 is 261 L/s (=1368 − 1107), which is comparable to 216 L/s discharge from the Oak Hill Boreholes (Table 1). Somewhat greater flow gained at WB3 than discharged by OAK is consistent with additional transfer from West Creek to the Pine Knot Mine complex (Figure 1). Likewise, the difference between the measured flow for WWB and the estimated flow using the SRL yield is −118 L/s (=752 − 870). This value indicates greater streamflow lost from West Creek than that measured during the seepage runs (−53.4 L/s; Table A2) and is consistent with additional losses upstream from WC4 to the Pine Knot

256

Mine complex plus diversions to the Crystal Reservoir water supply.

Surface Geophysical Surveys Indicate Variably Conductive Zones beneath Streambed The resistivity surveys demonstrated spatial and temporal variability in streambed conductivity in the upper segment of West Creek, above WC5, during 2012, and spatial variability within the upper, intermediate, and lower segments above WC9 during 2014. The resistivity data collected in 2012 were acquired on different dates for the same general locations between end-points 0 and 2 within the upper segment of West Creek (Figures 3A and 3B). The results for the 2012 surveys are arranged in downstream (left to right) and chronological (top to bottom) sequence, with data for the date of the lowest measured streamflow at the top (April 27, 2012) and data for the date of the highest measured streamflow at the bottom (September 27, 2012) (Figure 7A).

The latter survey was extended upstream to capture the full extent of the low-resistivity anomaly (potential flow loss zone) indicated during the two preceding surveys. Each profile indicates resistivity values from the streambed surface to a depth of approximately 30 m.

All document a high-resistivity upper layer of greater than 1,000 ohm-meters ( ·m) from the streambed surface to a depth of 5 to 10 m in the upper segment of West Creek. This zone transitions to an intermediateresistivity zone of between 300 ·m and 1,000 ·m, with anomalous low-resistivity (high-conductivity) values of less than 300 ·m in local zones at depths from about 5 m to greater than 30 m.

The resistivity data collected in 2014 for surveys 1 to 12 (from resistivity end-points 1 to 13) span a 2.1 km segment where West Creek crosses the underground Oak Hill Mine complex (Figure 1). The starting survey segment 1 (resistivity end-points 1 to 2) of the 2014 surveys (Figure 7B) coincided with the downstream segment of the 2012 surveys (Figure 7A). The profiles for 2014 are displayed with a different color scale than that used for 2012. As indicated by previous surveys in the upper reach, the 2014 profiles generally indicate a longitudinally extensive 5- to 10-m-thick high-resistivity layer near the surface and decreasing resistivity with depth beneath the streambed. The high-resistivity layer at the surface was disrupted locally, primarily within surveys 6, 8, 9, and 10, where low-resistivity anomalies also extended to depths of 10 to 30 m. These lowresistivity anomalies are interpreted to indicate relatively conductive water-saturated zones that could be locations of streamflow loss, as previously described for the “leaky streambed” between resistivity end-points 7 to 11, or could be locations of groundwater discharge,

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

Investigation of Streamflow Loss Near Abandoned Mines

Figure 7. Electrical resistivity survey profiles for: (A) 0.3 km length of West Creek upstream from Forestville, April 27, September 27, and November 9, 2012; and (B) 2.1 km length of West Creek downstream from Forestville to Phoenix Park, May 12–15, 2014. Locations of stream gauges (WC5 and WC8), hyporheic temperature probes (T06, T07, T09, T10, and T11), and various structures are noted. Locations of resistivity stations and stream gauges are shown in Figures 1 and 3.

as described for survey 6 (between resistivity end-points 6 and 7) (Figures 3 and 4).

The location and orientation of the low-resistivity anomaly in the uppermost survey segment, between 260 and 350 m downstream from WC4 (Figure 7A), are coincident with the mapped outcrop of the Mammoth Bed, which was indicated to dip 25 to 30 degrees southeastward approximately perpendicular to the survey segment (Figure 3B). Underground mine workings on the Mammoth Bed (Bottom Split) underlie this segment. Likewise, the location of the low-resistivity anomaly in the following survey segment, between 480 and 530 m downstream from WC4, is coincident with the mapped outcrop of the Primrose Bed (Figure 3B).

However, this survey segment was aligned at an oblique angle across the coalbed (Figure 3B), and the shape of the low-resistivity anomaly is broad compared to the thickness of the coalbed.

EM-31 survey data were collected in December 2015 during moderate-flow conditions along most of the same segment of West Creek as the 2014 resistivity surveys (between resistivity end-points 0 to 11) (Figure

8). The EM-31 conductivity readings for the HD and VD orientations at a given location were generally correlated; however, greater conductivity was indicated for 6-m-depth (HD) readings than 3-m-depth (VD) readings at most points along the stream channel. These results are consistent with the resistivity survey profiles that indicated a persistent high-resistivity layer near the streambed surface overlying more conductive material at depth (Figure 7). Furthermore, the locations of anomalously high or low EM-31 conductivity generally correlated with electrical resistivity anomalies.

Specifically, high-EM conductivity and low-resistivity readings were consistently observed toward the upstream end of survey 0 (250–320 m), the downstream end of survey 1 (500 to 600 m), the upper end of survey 5 (1,100–1,250 m), most of survey 6 (1,390–1,500 m), and most of surveys 8, 9, and 10 (1,780–2,350 m).

Note that in the upper 250–320 m zone, the HD conductivity peaks (at 6 m) were offset downstream from the VD peaks (at 3 m). This offset is consistent with the low-resistivity anomaly that angles approximately 30 degrees downward from the surface (Figure 7A)

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

257

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer

Figure 8. Results of electromagnetic conductivity (EM31) survey along 2.1 km length of West Creek from Forestville (230 m downstream from WC4) to Phoenix Park (2,360 m downstream to WC8), December 7–8, 2015. Approximate locations of resistivity surveys shown in Figure 7 are shown in compressed form along top of graph.

following the same orientation as the mapped coalbed in this location (Figure 3B).

Streambed Temperature Profiles Indicate Temporal Variations in Vertical Flux A trial investigation of streambed flux using temperature probes was conducted within segment 1 (not reported herein), which was followed by more extensive investigations within the less accessible stream segment along resistivity survey segments 6 and 11 (locations of T06, T07, T09, T10, and T11 indicated in Figure 7B). For the latter effort, temperature probes logged the vertical temperature gradients from the top of the streambed to a maximum depth of 0.085 m over two 2 month intervals. Temperature-derived flux graphs show results at the greatest depth for each location (Figure 9). No-flow to low-flow conditions dominated during the first period, from September 12 to November 6, 2014 (Figure 9A), whereas moderate- to highflow conditions dominated during the second period, from November 11, 2014, to January 5, 2015 (Figure 9B). The data for the first period at T09 were lost due to equipment failure.

The magnitude and direction of temperature-derived flux through the stream channel varied depending on the flow conditions and the location of each probe. Generally, during both the first (September to November) and second (November to January) periods, the flux estimates for probes at different locations were temporally correlated to one another and

258

were also, to varying degrees, responsive to changes in the flow recorded at WC5 (Figures 9A and 9B). Because the thermal properties, heat transport, and flow through variably saturated media violate the assumptions within VFLUX (Gordon et al., 2012), the results when flow was indicated at WC5 are emphasized. When the stream channel experienced intermittent flow due to precipitation, such as that on October 15–16, 2015, downward flux (positive value) as high as 2.1 × 10−5 m/s was indicated (Figure 9A); however, these conditions were short lived; upstream flow ceased October 18 (Figure 9A). During the second period, rain events and sustained streamflow toward the end of November and into December resulted in persistent, downward flux of 0.5 to 1.2 × 10−5 m/s at T06, T07, T10, and T11, and smaller, less persistent downward flux at T09 (Figure 9B).

During the first period, the temperature probe at the most upstream location (T06) indicated a consistent downward flux and the least variation in magnitude compared to the other probes; T06 was downstream from the perennial upstream segment between resistivity end-points 5 and 6 (Figure 3). Likewise, during the second period, the flux estimates for T06 were consistently downward and, thus, could indicate that streamflow was consistently lost through the streambed in that general location. Also, during the second period, the two most upstream probes (T06 and T07) and the most downstream probe (T11) exhibited less variability and more sustained downward flux compared to the two intermediate sites (T09 and T10), possibly

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

Investigation of Streamflow Loss Near Abandoned Mines

Figure 9. Temporal variations in streambed flux calculated from vertical temperature gradients at locations (T06, T07, T09, T10, and T11) within intermediate segment of West Creek, downstream from WC5 and upstream from WC8: (A) September 12, 2014–November 6, 2014; (B) November 11, 2014–January 5, 2015. Positive values are downward flux; negative values are upward flux. Locations of temperature probes in relation to seepage run and resistivity survey sites are shown in Figure 7B. Corresponding daily streamflow at WC5 and daily precipitation for the study area (excerpted from Figure 5) are shown for reference. Shading indicates periods of no-flow conditions, when the streambed may be variably saturated and VFLUX computations may be invalid.

reflecting a continuous streamflow past the former locations (note that a tributary adds flow to West Creek upstream from T11, just upstream from WC8) during

this period, but not past the locations of T09 and T10 until later in the period. Changes from upward (negative values) to downward fluxes during late November

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

259

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer

and greater variations indicated for T09 and T10 could indicate these segments gradually changed to flowing and, hence, losing conditions. Last, by comparing the two time periods, the maximum downward fluxes were indicated for the first large runoff event in October 14–16, 2015, which followed an extended dry period without upstream flow (at WC5) (Figure 9A). Generally smaller downward fluxes, but of comparable magnitude, were indicated during the second period, when greater, sustained upstream flow was occurring (Figure 9B). The initial extreme downward flux during the first period may have exceeded those estimated during the second period, despite larger flows, because of possible decreases in the unsaturated storage capacity and decreases in the vertical hydraulic gradient after streamflow had returned to the segment.

Despite potential uncertainties and local variations in the temperature-derived flux data, the range of values was in good agreement with estimates calculated on the basis of streamflow loss during seepage runs (Table A2). For the seepage runs, an average downward flux estimate of 1.70 × 10−5 m/s was indicated for the entire 2.1 km segment between WC4 and WC9 (computed by dividing 53.4 L/s or 0.0534 m3 /s lost by a total 3,150 m2 stream-channel area), and a corresponding estimate of 3.05 × 10−5 m/s was indicated for the leaky streambed in the upper 300 m segment between WC4 and WC5 (computed by dividing 13.7 L/s or 0.0137 m3 /s lost by the 450 m2 area of the channel).

Historical Mining Map Provides Context for Current Hydrological Observations The general location of streamflow losses within the upper reach, approximately 300 to 600 m downstream from WC4, between resistivity survey end-points 0 to 2 (Figures 3B and 4), coincided with the historically mapped outcrop locations of the Mammoth and Primrose coalbeds (Pennsylvania Geological Survey, 1889). The historic mine map indicated these coalbeds were approximately perpendicular to the channel and dipped 25 to 30 degrees southeastward (downstream).

Although the streambanks along the channel in this segment exhibited signs of excavation, the channel appeared to be intact and stable, with gravel- to bouldersized bed material along the entire stream course.

Approximately 1,000 m downstream from WC5, between resistivity end-points 6 and 7, the channel consistently gained groundwater seepage with higher pH and SC than upstream (Figures 3 and 4). This short perennial segment of the stream appeared to align with the original stream channel location compared to the 1889 mine map (Figure 3B), which shows the identical stream path as the USGS 1892 Pine Grove 15 minute topographic map. However, downstream from

260

this reach, between resistivity end-points 7 and 11, all streamflow was lost during low-flow conditions. This 600-m-long “leaky” channel segment (1,700 to 2,300 m downstream from WC4) apparently had been relocated eastward from its original path shown on the 1889 mine map (Figure 3B). Along this reach, the stream crosses over the axis of the Phoenix Park or North Delaware Anticline, where the underground mine workings on the Diamond and Primrose Beds were at relatively shallow depth (Pennsylvania Geological Survey, 1889; Wood et al., 1968, cross sections C-C and D-D ). A gangway on the Diamond Bed has a mapped elevation of 682 ft (206 m) where it crosses beneath the stream channel, between resistivity survey end-points 9 and 10, and where the surface elevation is approximately 780 ft (236 m) (Figures 3A and 3B). Again, except for a distance of about 150 m downstream from WC8, between resistivity end-points 12 and 13, and another 300-mlong segment immediately upstream from WC9, much of the present stream path appears to deviate from its original route to WC9.

Persistent, sometimes visible, groundwater inflows restored perennial streamflow and aquatic habitat (see Appendix) to West Creek upstream from WC9, below resistivity end-point 12 (Figures 3 and 4). In particular, immediately upstream from the streamflow gauge at WC9, a perennial spring (identified as USGS station 014678369; USGS NWIS, http://dx.doi.org/ 10.5066/F7P55KJN) discharged groundwater with near-neutral pH (6.4 to 7.0) and elevated SC (790 to 850 ␮S/cm) during 2012–2015. This spring was the predominant source of streamflow at WC9 during lowflow conditions. The spring emanates about 5 m east of the streambank from the southern base of an overgrown pile of mined rock that is 15–20 m higher than the streambed; the historic 1889 mine map and the 1892 topographic map show an unnamed tributary (now buried) crossing this area from the northeast and entering the stream at the same point as the current spring (Figure 3B).

SUMMARY AND CONCLUSIONS The Oak Hill Mine complex extends beneath the West West Branch and West Branch Schuylkill River sub-basins and facilitates the eastward transfer of water from the West West Branch to the West Branch.

Streamflow lost from West Creek to the underlying Oak Hill Mine complex, in the headwaters of the West West Branch Schuylkill River, eventually discharges from the Oak Hill Boreholes. Additionally, recharge in the northeastern part of West Creek is inferred to enter the Pine Knot Mine complex and discharge to PKN. The contributions of AMD to the West Branch result in greater streamflow than would be expected

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

Investigation of Streamflow Loss Near Abandoned Mines

based on its topographic drainage area and account for a large fraction of the AMD contaminant load that degrades the Schuylkill River. Likewise, because of the streamflow losses from the West Creek sub-basin, the annual streamflow of the West West Branch Schuylkill River is less than would be expected based on its topographic drainage area. The streamflow losses from West Creek account for approximately one fifth of the annual discharge from the Oak Hill Boreholes. Thus, if the streamflow losses from West Creek can be prevented, the discharge volume and contaminant loading from the Oak Hill Boreholes can be reduced, while at the same time the aquatic habitat of West Creek can be restored.

Repeated longitudinal streamflow measurements along the West Creek channel combined with continuous streamflow gauging and continuous temperature logging within the streambed at selected points indicated spatial and temporal variations in the magnitude and duration of the loss of streamflow from the stream channel to the subsurface. During low flow, complete loss of streamflow was observed between stations WC4 and WC5, whereas during high flow, streamflow was gained (net increase) along this segment, even though some streamflow was still lost to the underlying mines.

Surface geophysical surveys indicated variably conductive zones beneath the streambed of West Creek along the 2.1 km segment over the Oak Hill Mine complex. The electrical resistivity and electromagnetic conductivity surveys identified a relatively extensive low-conductivity (high resistivity) zone beneath the streambed to 5 to 10 m depth, which was underlain by moderately conductive material to a depth of 30 m.

Both layers were interrupted locally by anomalously high conductivity (low resistivity) zones, which are interpreted to indicate water saturation and potential locations of streamflow loss or groundwater inflow to the stream. The low-resistivity anomalies in the upper part of West Creek were consistently located by repeated surveys over a range of flow conditions, but they increased in size as the streamflow increased. These changes in the apparent volume of the conductive zones imply an increase in water saturation and potential for greater recharge during high-flow conditions. The locations of apparent streamflow loss and low-resistivity anomalies in the 300-m-long “leaky” upper segment of West Creek near Forestville coincide with the historically mapped locations of coalbed outcroppings and associated underground mine workings. The locations of streamflow loss along the 600-m-long leaky intermediate segment coincide with shallow (30-m-deep) underground workings where the stream crosses the axis of an anticline. This channel segment apparently has been relocated eastward from its original path as shown on historical maps.

Continuously logged temperature profiles through the streambed corroborate the interpretation of temporal variations in streambed leakage. Temperature probes at 0.085 m depth within the hyporheic zone of the intermediate segment downstream from Forestville indicated variable fluxes ranging to a maximum of 2.1 × 10−5 m/s (downward) during flowing conditions.

The downward flux estimates are comparable to the loss estimates during seepage runs, normalized by the streamflow area. Cumulative streamflow lost from West Creek during seepage runs averaged 53.4 L/s, including 13.7 L/s lost in the upper segment above Forestville.

Assuming the 2.1 km segment that crosses the Oak Hill Mine complex is 1.5 m wide, the cumulative streamflow loss during seepage surveys equates to an average downward flux of 1.70 × 10−5 m/s across the total 3,150 m2 stream-channel area, which is within the range of flux values indicated by the temperature profiles. Likewise, assuming the 300-m-long losing segment in the upper segment is 1.5 m wide, the loss of 13.7 L/s equates to an average downward flux of 3.05 × 10−5 m/s across the 450 m2 area of the channel.

Downstream from Forestville, both gains and losses were apparent over the range of hydrologic conditions.

Generally, the water from upstream was slightly net acidic (3.9 mg/L as CaCO3 ) with low pH (4.5), low SC (70 ␮S/cm), and elevated concentrations of dissolved aluminum (0.40 mg/L) and zinc (0.034 mg/L).

In contrast, inflows from tributaries or groundwater seepage downstream from Forestville were net alkaline with higher pH and SC. Consequently, the SC and pH of West Creek increased downstream from Forestville, even where downward leakage through the streambed resulted in net losses in streamflow. Streamflow gains from groundwater seepage to the stream and perennial streamflow were observed within a relatively short intermediate segment at 1,400 to 1,700 m downstream from WC4 (between resistivity survey end-points 6 and 7), where the channel appears to follow its historical route. This brief perennial segment is bracketed by losing, intermittent segments upstream and downstream, where the straightened stream channel appears to have been relocated from its natural route.

The combination of surface geophysical surveys plus instantaneous streamflow and water-quality surveys along the entire segment of West Creek on multiple dates effectively indicated the locations and the magnitude of streamflow loss and, thus, can be useful for watershed and resource managers to identify stream segments for restoration priority (Figure 10). The identified segments 1, 2, and 3 (Figure 10), in order of decreasing priority, correspond to those reaches with consistent downward flux through the streambed and anomalous subsurface conductivity. The geophysical surveys, alone, were insufficient to establish if stream-

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

261

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer

Figure 10. Summary of data and stream restoration priorities for West Creek: (A) median normalized streamflow value compared to WC4 for seepage runs during May 2014 through July 2015; (B) EM-31 vertical dipole conductivity values measured at approximately 3 m depth during December 7â&#x20AC;&#x201C;8, 2015; and (C, D) stream restoration priorities 1, 2, and 3 on (C) Minersville topographic quadrangle base (USGS, 1955) and (D) historical underground mine map (Pennsylvania Geological Survey, 1889). Symbols in A indicate locations of seepage run flow and water-quality sites and the start and end of electrical resistivity survey segments (numbered 0â&#x20AC;&#x201C;13).

flow was losing or gaining in a zone indicated to be anomalous (high conductivity) or if some other feature such as a metal object caused a geophysical anomaly.

For that matter, streamflow data alone were insufficient to isolate segments of flow loss because, in some segments, both gains and losses occurred. Corresponding water-quality data were helpful to indicate where such gains took place (based on higher pH and SC of inflowing groundwater compared to the stream water already in the channel). Additionally, the water budget evaluation plus evaluation of maps showing historical mining and hydrography were helpful to provide geographic context, to validate the interpretations of surface-water losses to underground mines, and to indicate potential strategies and limitations of future restoration.

262

The methods used in this investigation would be generally applicable for the assessment of hydrological inter-connections and interactions in other extensively mined settings. The specific information collected for the study will be useful to watershed and resource managers to develop stream restoration strategies for West Creek, which could also reduce the discharge of associated AMD from the Oak Hill Boreholes. For example, rather than replacing the stream with an artificial rip-rap or concrete-lined channel, other more environmentally appealing approaches could be applied to create aquatic habitat in the extensively mined setting. Where the streambed appears to be intact and the vertical losses take place in a relatively discrete zone, grout injection directly into the streambed may

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243â&#x20AC;&#x201C;273

Investigation of Streamflow Loss Near Abandoned Mines

be appropriate to decrease the conductivity beneath the streambed (e.g., Ackman and Jones, 1991). Grout injection would typically be conducted when water is present in the channel to transmit the grout into the high-conductivity zones. Alternatively, stream lining or stream channel relocation may be appropriate, especially where the existing channel is artificially straight, lacks integrity, and/or is devoid of desirable habitat characteristics. If a lining is used, instead of sealing the length of channel to keep water within, overlapping segments of the lining, such as shingles on a roof, may be helpful to permit diffuse groundwater seepage to enter the channel and join the water from upstream points. Concurrent streamflow and geophysical measurements could be conducted during grout injection or stream channel reconstruction to determine if the grout or channel lining is effective and guide the reconstruction work. Streamflow measurements and additional aquatic biological surveys would generally be appropriate to determine the effectiveness of any such restoration and associated recovery of aquatic life after water returns to the channel.

SUPPLEMENTAL MATERIAL Supplemental Material associated with this article can be found online at http://faculty.kutztown.edu/ sherrod/PublicationMaterials.

ACKNOWLEDGMENTS This study was conducted by the USGS Pennsylvania Water Science Center and Kutztown University (KU) in cooperation with the Schuylkill Conservation District, the Schuylkill Headwaters Association, Inc. (SHA), and the Pennsylvania Department of Environmental Protection (PaDEP). Heather Eggleston and Robin Brightbill of USGS provided expertise during aquatic ecological surveys. Several KU students assisted with data collection, most notably Jarred Swiontek, Rick Jayne, Jeff Kadegis, Dea Musa, Alex Spielman, and Leanne Hillegas. Funding for L. Sherrod and student assistants was provided by a Kutztown Research Committee Grant. Daniel Koury of PaDEP assisted with fieldwork during aquatic ecological surveys and seepage runs. Daniel Cook and Christopher Bentz of Alfred Benesch & Company Engineering offered preliminary consultation on the project design and conducted stream cross-segmental surveys and hydraulic modeling in support of potential future restoration strategies. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. government.

REFERENCES ACKMAN, T. E. AND JONES, J. R., 1991, Methods to identify and reduce potential surface streamwater losses into abandoned underground mines: Environmental Geology and Water Science, Vol. 17, pp. 227–232.

AMERICAN PUBLIC HEALTH ASSOCIATION, 1998, Alkalinity (2320)/titration method. In Clesceri, L. S., Greenberg, A. E., and Eaton, A. D. (Editors), Standard Methods for the Examination of Water and Wastewater (20th): American Public Health Association, Washington, DC, pp. 2.26–2.29.

ARCHIE, G. E., 1942, The electrical resistivity log as an aid in determining some reservoir characteristics: Transactions of the American Institute of Mining, Metallurgical, and Petroleum Engineers, Vol. 146, pp. 54–62.

ASH, S. H. AND KYNOR, H. D., 1953, Barrier Pillars in the Southern Field, Anthracite Region of Pennsylvania: U.S. Bureau of Mines Bulletin 534, 30 p.

ASH, S. H.; EATON, W. L.; HUGHES, K.; ROMISCHER, W. M.; AND WESTFIELD, J., 1949, Water Pools in Pennsylvania Anthracite Mines: U.S. Bureau of Mines Technical Paper 727, 78 p.

ASH, S. H.; HOWER, C. S.; KENNEDY, D. O.; AND LESSER, W. H., 1953, Mine Pumping Plants, Anthracite Region of Pennsylvania: U.S. Bureau of Mines Bulletin 531, 151 p.

BERNHARDT, E. S.; LUTZ, B. D.; KING, R. S.; FAY, J. P.; CARTER, C. E.; HELTON, A. M.; CAMPAGNA, D.; AND AMOS, J., 2012, How many mountains can we mine? Assessing the regional degradation of Central Appalachian rivers by surface coal mining: Environmental Science & Technology, Vol. 46, pp. 8115–8122.

BIANCHIN, M.; SMITH, L.; AND BECKIE, R., 2011, Defining the hyporheic zone in a large tidally influenced river: Journal of Hydrology, Vol. 406, 1–2, pp. 16–29.

BIESECKER, J. E.; LESCINSKY, J. B.; AND WOOD, C. R., 1968, Water Resources of the Schuylkill River Basin: Pennsylvania Department of Forests and Waters Water Resources Bulletin 3, 198 p.

BINLEY, A.; CASSIANI, G.; MIDDLETON, R.; AND WINSHIP, P., 2002a, Vadose zone flow model parameterisation using cross-borehole radar and resistivity imaging: Journal of Hydrology, Vol. 267, pp. 147–159.

BINLEY, A.; WINSHIP, P.; WEST, L. J.; POKAR, M.; AND MIDDLETON, R., 2002b, Seasonal variation of moisture content in unsaturated sandstone inferred from borehole radar and resistivity profiles: Journal of Hydrology, Vol. 267, pp. 160– 172.

BRIGGS, M. A.; LAUTZ, L. K.; AND MCKENZIE, J. M., 2012a, A comparison of fibre-optic distributed temperature sensing to traditional methods of evaluating groundwater inflow to streams: Hydrological Processes, Vol. 26, pp. 1277–1290.

BRIGGS, M. A.; LAUTZ, L. K.; MCKENZIE, J. M.; GORDON, R. P.; AND HARE, D. K., 2012b, Using high-resolution distributed temperature sensing to quantify spatial and temporal variability in vertical hyporheic flux: Water Resources Research, Vol. 48, W02527, 16 p.

BURKHOLDER, B. K.; GRANT, G. E.; HAGGERTY, R.; KHANGAONKAR, T.; AND WAMPLER, P. J., 2008, Influence of hyporheic flow and geomorphology on temperature of a large, gravel-bed river, Clackamas River, Oregon, USA: Hydrological Processes, Vol. 22, pp. 941–953.

BURROWS, J. E.; PETERS, S. C.; AND CRAVOTTA, C. A., III, 2015, Temporal geochemical variations in above- and below-drainage coal mine discharge: Applied Geochemistry, Vol. 62, pp. 84– 95. (http://dx.doi.org/10.1016/j.apgeochem.2015.02.010).

CARDENAS, M., 2009, Stream-aquifer interactions and hyporheic exchange in gaining and losing sinuous streams: Water Resources Research, Vol. 45, W06429, 13 p.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

263

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer CARVER, L. AND RAUCH, H. W., 1994, Hydrogeologic effects of subsidence at a longwall mine in the Pittsburgh Coal seam.

In Peng, S. S. (Editor), Proceedings of the 13th Conference on Ground Control in Mining: West Virginia University, Morgantown, W.V., pp. 298–307.

CIFELLI, R. C. AND RAUCH, H. W., 1986, Dewatering effects from selected underground coal mines in north-central West Virginia.

In Peng, S. S. (Editor), Proceedings of the 2nd Workshop on Surface Subsidence due to Underground Mining: West Virginia University, Morgantown, W.V., pp. 249–263.

CLIFFORD, J. AND BINLEY, A., 2010, Geophysical characterization of riverbed hydrostratigraphy using electrical resistance tomography: Near Surface Geophysics, Vol. 8, pp. 493–501.

CONSTANTZ, J.; STONESTROM, D.; STEWARD, A. E.; NISWONGER, R. N.; AND SMITH, T. R., 2001, Analysis of streambed temperatures in ephemeral channels to determine streamflow frequency and duration: Water Resources Research, Vol. 37, No. 2, pp.

317–328.

CRAVOTTA, C. A., III, 2008, Dissolved metals and associated constituents in abandoned coal-mine discharges, Pennsylvania, USA: 1. Constituent concentrations and correlations: Applied Geochemistry, Vol. 23, pp. 166–202. (http://dx.doi.org/ 10.1016/j.apgeochem.2007.10.011).

CRAVOTTA, C. A., III; GOODE, D. J.; BARTLES, M. D.; RISSER, D. W.; AND GALEONE, D. G., 2014, Surface-water and groundwater interactions in an extensively mined watershed, upper Schuylkill River, Pennsylvania, USA: Hydrological Processes, Vol. 28, pp.

3574–3601. (http://dx.doi.org/10.1002/hyp.9885).

DANILUK, T. L.; LAUTZ, L. K.; GORDON, R. P.; AND ENDRENY, T. A., 2013, Surface water-groundwater interaction at restored streams and associated reference segments: Hydrological Processes, Vol. 27, pp. 3730–3746. (http://dx.doi.org/ 10.1002/hyp.9501).

DIXON, D. Y. AND RAUCH, H. W., 1990, The impact of three longwall coal mines on streamflow in the Appalachian Coalfield. In Peng, S. S. (Editor), Proceedings of the 9th International Conference on Ground Control in Mining: West Virginia University, Morgantown, W.V., pp. 169–182.

FENG, Q.; LI, T.; QIAN, B.; ZHOU, L.; GAO, B.; AND YUAN, T., 2014, Chemical characteristics and utilization of coal mine drainage in China: Mine Water and the Environment, Vol. 33, pp. 276– 286. (http://dx.doi.org/10.1007/s10230-014-0271-y).

FISHMAN, M. J. AND FRIEDMAN, L. C. (Editors), 1989, Methods for Determination of Inorganic Substances in Water and Fluvial Sediments: U.S. Geological Survey Techniques of WaterResources Investigations Book 5, Chapter A1, 545 p.

GEOSTUDI ASTIER SRL AND MULTI-PHASE TECHNOLOGIES LLC, 2006, ERTLab 3D Electrical Resistivity Tomography Inversion Software User Manual: Geostudi Astier srl and Multi-Phase Technologies LLC, Livorno, Italy, 89 p.

GOODE, D. J.; CRAVOTTA, C. A., III; HORNBERGER, R. J.; HEWITT, M. A.; HUGHES, R. E.; KOURY, D. J.; AND EICHOLTZ, L. W., 2011, Water Budgets and Groundwater Volumes for Abandoned Underground Mines in the Western Middle Anthracite Coalfield, Schuylkill, Columbia, and Northumberland Counties, Pennsylvania—Preliminary Estimates with Identification of Data Needs: U.S. Geological Survey Scientific Investigations Report 2010–5261, 54 p. (http://pubs.usgs.gov/ sir/2010/5261/).

GORDON, R. P.; LAUTZ, L. K.; BRIGGS, M. A.; AND MCKENZIE, J. M., 2012, Automated calculation of vertical pore-water flux from field temperature time series using the VFLUX method and computer program: Journal of Hydrogeology, Vol. 420– 421, pp. 142–158.

GRGICH, P.; HAMMACK, R.; HARBERT, W.; SAMS, J.; VELOSKI, G.; AND ACKMAN, T., 2004, Delineating groundwater flow paths

264

with surface geophysics: Journal of Environmental Hydrology, Vol. 12, paper 12, 10 p.

GROWITZ, D. J.; REED, L. A.; AND BEARD, M. M., 1985, Reconnaissance of Mine Drainage in the Coal Fields of Eastern Pennsylvania: U.S. Geological Survey Water-Resources Investigations Report 83–4274, 54 p.

HAGREY, S. AND MICHAELSEN, J., 1999, Resistivity and percolation study of preferential flow in vadose zone at Bokhorst, Germany: Geophysics, Vol. 64, No. 3, pp. 746–753.

HYUN, Y.; KIM, H.; LEE, S.; AND LEE, K., 2011, Characterizing streambed water fluxes using temperature and head data on multiple spatial scales in Musan stream, South Korea: Journal of Hydrology, Vol. 402, pp. 377–387.

JOHNSON, T.; VERSTEEG, R.; HUANG, H.; AND ROUTH, P., 2009, Data-domain correlation approach for joint hydrogeologic inversion of time-lapse hydrogeologic and geophysical data: Geophysics, Vol. 74, pp. F127–F140.

KIMBALL, B. A.; WALTON-DAY, K. H.; AND RUNKEL, R. L., 2007, Quantification of metal loading by tracer injection and synoptic sampling, 1996–2000. In Church, S. E.; von Guerard, Paul; and Finger, S. E. (Editors), Integrated Investigations of Environmental Effects of Historical Mining in the Animas River Watershed, San Juan County, Colorado: U.S. Geological Survey Professional Paper 1651, pp. 417–495.

KIRBY, C. S. AND CRAVOTTA, C. A., III, 2005, Net alkalinity and net acidity 2: Practical considerations: Applied Geochemistry, Vol. 20, pp. 1941–1964.

LABRECQUE, D.; MILETTO, M.; DAILY, W.; RAMIREZ, A.; AND OWEN, E., 1996, The effects of noise on Occam’s inversion of resistivity tomography data: Geophysics, Vol. 61, pp. 538–548.

LAMBERT, D. C.; MCDONOUGH, K. M.; AND DZOMBAK, D. A., 2004, Long-term changes in quality of discharge water from abandoned underground coal mines in Uniontown Syncline, Fayette County, PA, USA: Water Research, Vol. 38, pp. 277– 288.

LAPHAM, W., 1989, Use of Temperature Profiles Beneath Streams to Determine Rates of Vertical Ground-Water Flow and Vertical Hydraulic Conductivity: U.S. Geological Survey Water-Supply Paper 2337, 35 p.

LINDE, N.; BINLEY, A.; TRYGGVASON, A.; PEDERSEN, L. B.; AND REVIL, A., 2006, Improved hydrogeophysical characterization using joint inversion of cross-hole electrical resistance and ground-penetrating radar traveltime data: Water Resources Research, Vol. 42, W12404, 16 p.

MACK, B.; MCDONALD, L. M.; AND SKOUSEN, J., 2010, Acidity decay of above-drainage underground mines in West Virginia: Journal of Environmental Quality, Vol. 39, pp. 1–8.

NORDSTROM, D. K., 1977, Thermochemical redox equilibria of Zobell’s solution: Geochimica et Cosmochimica Acta, Vol. 41, pp.

1835–1841.

NORDSTROM, D. K., 2011, Hydrogeochemical processes governing the origin, transport and fate of major and trace elements from mine wastes and mineralized rock to surface waters: Applied Geochemistry, Vol. 26, pp. 1777–1791.

NYQUIST, J.; FREYER, P.; AND TORAN, L., 2008, Stream bottom resistivity tomography to map ground water discharge: Ground Water, Vol. 46, No. 4, pp. 1–9.

OLDENBURG, D. W. AND LI, Y., 1999, Estimating depth of investigation in DC resistivity and IP surveys: Geophysics, Vol. 64, No. 2, pp. 403–416.

PENNSYLVANIA DEPARTMENT OF ENVIRONMENTAL PROTECTION, 2014, 2014 Pennsylvania Integrated Water Quality Monitoring and Assessment Report—Clean Water Act Segment 305(b) Report and 303(d) List: Pennsylvania Department of Environmental Protection, Harrisburg, PA, 786 p.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

Investigation of Streamflow Loss Near Abandoned Mines PENNSYLVANIA GEOLOGICAL SURVEY, 1889, Heckscherville Valley, Peaked Mountain, Buck Run, Phoenix Park and Muddy Branch Basins: Pennsylvania Geological Survey, 2nd Series, Southern Coal Field Mine Sheet, No. XII, January 1889, scale 1:1,200.

RANTZ, S. E. AND others, 1982, Measurement and Computation of Streamflow: U.S. Geological Survey, Water-Supply Paper 2175, 2 Volumes.

REED, P.; DUMONTELLE, P.; SARGENT, M.; AND KILLEY, M., 1983, Nuclear logging and electrical earth resistivity techniques in the vadose zone in glaciated earth materials. In NWWA/U.S. EPA Conference on Characterization and Monitoring of the Vadose (Unsaturated) Zone, Las Vegas, Nevada, National Water Well Association, Dublin, Ohio, pp. 580–601.

RISSER, D. W., 2006, Simulated Water Budgets and Ground-Water/ Surface-Water Interactions in Bushkill and Parts of Monocacy Creek Watersheds, Northampton County, Pennsylvania—A Preliminary Study with Identification of Data Needs: U.S. Geological Survey Open-File Report 2006-1143, 31 p.

RISSER, D. W.; CONGER, R. W.; ULRICH, J. E.; AND ASMUSSEN, M. P., 2005, Estimates of Groundwater Recharge Based on Streamflow-Hydrograph Methods, Pennsylvania: U.S. Geological Survey Open-File Report 2005-1333, 30 p.

RUNKEL, R. L.; KIMBALL, B. A.; WALTON-DAY, K.; VERPLANCK, P. L.; AND BROSHEARS, R. E., 2012, Evaluating remedial alternatives for an acid mine drainage stream: A model post audit: Environmental Science & Technology, Vol. 46, pp. 340–347.

RUTLEDGE, A. T., 1998, Computer Programs for Describing the Recession of Groundwater Discharge and for Estimating Mean Groundwater Recharge and Discharge from Streamflow Data— Update: U.S. Geological Survey Water-Resources Investigations Report 98-4148, 43 p.

SCHEMEL, L. E.; COX, M. H.; RUNKEL, R. L.; AND KIMBALL, B.A. 2006, Multiple injected and natural conservative tracers quantify mixing in a stream confluence affected by acid mine drainage near Silverton, Colorado: Hydrological Processes, Vol. 20, pp. 2727–2743.

SCHROTT, L. AND SASS, O., 2008, Application of field geophysics in geomorphology: Advances and limitations exemplified by case studies: Geomorphology, Vol. 93, No. 1–2, pp. 55–73.

SCHUYLKILL ACTION NETWORK, 2008, Schuylkill River Facts— Schuylkill Action Network, from Assessment to Protection: Electronic document, available at http://www. schuylkillactionnetwork.org/ SHERROD, L.; SAUCK, W.; AND WERKEMA, D., 2012, A low-cost, in situ resistivity and temperature monitoring system: Ground Water Monitoring & Remediation, Vol. 32, No. 2, pp. 31–39.

TORAN, L.; HUGHES, B.; NYQUIST, J.; AND RYAN, R., 2012, Using hydrogeophysics to monitor change in hyporheic flow around stream restoration structures: Environmental & Engineering Geoscience, Vol. 18, No. 1, pp. 83–97.

TORAN, L.; NYQUIST, J. E.; FANG, A. C.; RYAN, R. J.; AND ROSENBERRY, D. O., 2013, Observing lingering hyporheic storage using electrical resistivity; variations around stream restoration structures, Crabby Creek, PA: Hydrological Processes, Vol. 27, No. 10, pp. 1411–1425. (http://dx.doi.org/10.1002/hyp.9269).

TORAN, L.; NYQUIST, J.; ROSENBERRY, D.; GAGLIANO, M.; MITCHELL, N.; AND MIKOCHIK, J., 2015, Geophysical and hydrologic studies of lake seepage variability: Ground Water, Vol. 5, No. 6, pp. 841–850. (http://dx.doi.org/ 10.1111/gwat.12309).

U.S. ENVIRONMENTAL PROTECTION AGENCY, 2013, National Recommended Water Quality Criteria—Aquatic Life Criteria Table: U.S. Environmental Protection Agency, last updated August 22, 2013 (https://www.epa.gov/wqe/national-recommendedwater-quality-criteria-aquatic-life-criteria-table).

VANDERBORGHT, J.; KEMNA, A.; HARDELAUF, H.; AND VEREECKEN, H., 2005, Potential of electrical resistivity tomography to infer aquifer transport characteristics from tracer studies: A synthetic case study: Water Resources Research, Vol. 41, W06013, 23 p. (http://dx.doi.org/10.1029/2004WR003774).

WARD, A.; FITZGERALD, M.; GOOSEFF, M.; VOLTZ, T.; BINLEY, A.; AND SINGHA, K., 2012, Hydrologic and geomorphic controls on hyporheic exchange during base flow recession in a headwater mountain stream: Water Resources Research, Vol. 48, No. 4, W04513, 20 p. (http://dx.doi.org/10.1029/2011WR011461).

WOOD, G. H., Jr.; TREXLER, J. P.; AND KEHN, T. M., 1968, Geologic Maps of Anthracite-Bearing Rocks in the West-Central Part of the Southern Anthracite Field Pennsylvania, Eastern Area: U.S. Geological Survey Miscellaneous Geologic Investigations Map I-528, 6 plates.

WORLD WATER ASSESSMENT PROGRAMME (WWAP), 2012, The United Nations World Water Development Report 4—Managing Water under Uncertainty and Risk (Vol. 1), Knowledge Base (Vol. 2), and Facing the Challenges (Vol. 3): Paris, UNESCO, 904 p.

YANG, X.; LABRECQUE, D.; AND PAPROCKI, L.2000, Estimation of 3-D moisture content using ERT data at the Socorro-Tech Vadose Zone Facility. In Proceedings of 13th EEGS Symposium on the Application of Geophysics to Engineering and Environmental Problems, pp. 915–925.

YOUNGER, P. L., 1997, The longevity of minewater pollution: A basis for decision making: Science of the Total Environment, Vol. 194–195, pp. 457–466.

YOUNGER, P. L. AND WOLKERSDORFER, C., 2004, Mining impacts on the fresh water environment: Technical and managerial guidelines for catchment scale management: Mine Water and the Environment, Vol. 23, pp. s2–s80.

APPENDIX Continuous Discharge Record, Aquatic Biological Surveys, and Limestone Neutralization Test Results in Support of Stream Restoration Strategies in an Abandoned Mine Lands Setting, Schuylkill River Watershed, Pennsylvania, USA, 2012–2015 Continuous Discharge at Gauges Streamflow at points along the West Creek flow path was highly variable (Figure 4), but it was correlated temporally with that recorded at downstream gauging stations on the West West Branch (WWB) and West Branch (WB3) and at AMD outfalls from the Pine Knot Tunnel (PKN) and the Oak Hill Boreholes (OAK) during 2012–2015 (Figure 5A and Table A1).

The daily discharge values at WWB and WB3 generally exhibited coincident peaks and troughs; however, discharge at WB3 was typically about twice that at WWB and also was less variable. Apparently, the streamflow at WB3 was moderated by the upstream contributions of AMD from PKN and OAK. The AMD discharge volumes were relatively constant with subdued peak flows and prolonged recessions compared to the surface waters (Figure 5). During the 12 month period from July 1, 2014, through June 30, 2015 (considered

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

265

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer Table A1. Streamflow gauging and water-quality monitoring sites, West West Branch, West Branch, and upper Schuylkill River, Schuylkill County, PA (site locations [except SRL] are shown in Figure 1).

USGS Station Number

Map ID

Site Name

West West Branch Schuylkill River 01467830 WC4 West Creek above Forestville 01467832 WC5 West Creek at Forestville 01467835 WC8 West Creek at Main St Phoenix Pk 01467837 WC9 West Creek at Ramtown Rd Phoenix Pk 01467861 WWB West West Branch above West Branch West Branch Schuylkill River 01467688 WB1 West Branch Schuylkill River above Pine Knot Tunnel 01467689 PKN Pine Knot Mine discharge 500 m below tunnel 01467691 OAK Oak Hill Mine discharge 200 m below boreholes 01467692 WB2 West Branch below Oak Hill Boreholes outfall 01467752 WB3 West Branch Schuylkill River above West West Branch Schuylkill River 01468500 SRL Schuylkill River at Landingville a

Year Established

Latitudea

Longitudea

2014 2014 2014 2014 2006

40.66877 40.69361 40.69131 40.68197 40.67986

−76.23796 −76.30820 −76.30200 −76.29408 −76.28815

2005 2005 2012 2006 2006

40.70413 40.70409 40.70203 40.70169 40.67067

−76.24969 −76.24989 −76.25158 −76.25199 −76.23267

1947

40.62917

−76.12500

Coordinates referenced to the North American Datum of 1983 (NAD 83). Values are decimal degrees.

for water budget estimates; Table 1), the range of daily discharge values varied by a factor of 5 at PKN and OAK compared to factors of 15 at WB3 and 63 at WWB. In contrast, for the same period, the range of daily discharge values varied by more than three orders of magnitude at the gauges on West Creek (WC4, WC5, and WC9). The greatest variability was exhibited by the intermediate gauge (WC5), where the streamflow was intermittent.

Perennial streamflow was transmitted from the mostly forested area upstream from Forestville (WC4) to the segment overlying the Oak Hill Mine complex (Figures 3, 4, and 5). The streamflow at WC4 was frequently greater than that downstream at WC5 or WC9 during 2012–2015 (Figures 5B, A1A, and A1B).

During periods of negligible recharge or runoff, persistent decreases in flow downstream from WC4 (losing conditions) were evident, such as low rainfall conditions in September 2014 or May 2015, or subfreezing conditions in February 2015, whereas increases in flow downstream were evident during periods of higher than normal rainfall, such as June 2015 (Figures 5B and A1). Sustained streamflow losses approaching 10 L/s from the 300 to 600 m upper reach, downstream from WC4 and immediately upstream from WC5, were recorded during low-flow periods (Figure A1A); however, greater losses, exceeding 50 L/s, were indicated during wet conditions, such as those experienced during spring of 2015 (Figure A1A). During the same time, the downstream segment of West Creek between WC4 and WC9 exhibited net gains in streamflow (Figure A1B).

Considering the difference between the daily average discharge at WC4 and WC5 during July 1, 2014–June 30, 2015 (Table 1), an average loss of only 6.3 L/s was

266

computed, which is about half of the average loss indicated during the seepage runs (Table A3). Gains were measured in this upper segment during the seepage run April 23, 2015, and also for brief periods during “wet” hydrologic conditions (Table A2 and Figures 3, 5B, and 6A). During the wet conditions in April 2015, a net increase in streamflow was indicated by the difference between the recorded streamflow at upstream and downstream gauges (WC5 − WC4); however, at the same time, water was likely to have been lost through the streambed to the underlying mines. Likewise, the difference between the continuously measured discharge at WC4 and WC9 during July 1, 2014–June 30, 2015 (Table 1) indicated an average net gain of 14.5 L/s over the year; however, this indicated net difference between WC4 and WC9 does not accurately account for the total gains and cumulative losses along the segment as measured at intermediate locations during seepage runs (Figures 3 and 5B). Because of gains and losses within some stream segments, the cumulative losses (indicated by point symbols for seepage run measurements in Figures 5A and A1B) would generally be larger than those indicated by the difference in recorded streamflow at fixed location gauging stations.

Aquatic Biological Surveys Streamflow restoration combined with water-quality and habitat improvement could lead to the recovery of fish and associated aquatic organisms in the West West Branch and West Branch Schuylkill River watersheds.

Aquatic biological surveys in 2014 demonstrated the presence of fish, including brook trout and pollutionsensitive macroinvertebrate taxa, in the downstream

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

Investigation of Streamflow Loss Near Abandoned Mines

Figure A1. Daily precipitation and downstream changes in daily streamflow of West Creek at station WC4 upstream from mined area and two downstream stations, WC5 and WC9, during June 2014â&#x20AC;&#x201C;September 2015: (A) Difference between WC4 and WC5; and (B) difference between WC4 and WC9. Values greater than 0 indicate net gains in streamflow from WC4 to downstream sites. Because of gains and losses within the stream segment(s), cumulative losses (indicated by point symbols for seepage run measurements) are larger than those indicated by the difference in recorded streamflow at fixed location gauging stations.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243â&#x20AC;&#x201C;273

267

268

72.73 21.58 398.46 3.35

Average Median Maximum Minimum

59.02 11.36 457.66 0.00

3.71 0.00 0.00 0.00 23.82 96.40 97.68 61.74 0.00 0.00 457.66 11.36 14.84

5.19 1.43 0.03 2.52 21.33 133.47 187.26 101.31 4.30 2.35 537.40 64.14 28.04 83.75 21.33 537.40 0.03

−14.22 −6.40 −8.04 −3.35 −6.15 −1.33 −144.29 −19.85 −6.29 −3.40 59.20 −10.22 −14.02 −13.72 −6.40 59.20 −144.29 137.11 32.46 804.68 5.75

25.90 7.83 8.04 6.00 32.46 155.64 383.43 160.72 16.71 5.75 804.68 101.64 73.66 −53.36 −20.71 −3.40 −267.29

−20.71 −6.40 −8.01 −3.48 −11.13 −22.17 −196.17 −59.40 −12.40 −3.40 −267.29 −37.50 −45.62 55.5 58.0 99.6 14.2

80.0 81.7 99.6 58.0 34.3 14.2 51.2 37.0 74.2 59.1 33.2 36.9 61.9

221.20 222.79 381.19 123.73

234.29 171.56 169.72 123.73 133.39 222.79 235.28 325.37 252.36 134.21 381.19 214.55 277.17

6.9 4.6 61.3 −15.5

6.1 3.7 4.7 2.7 4.6 0.6 61.3 6.1 2.5 2.5 −15.5 4.8 5.1

19.3 8.8 83.4 2.5

8.8 3.7 4.7 2.8 8.3 10.0 83.4 18.3 4.9 2.5 70.1 17.5 16.5

WC9 − WC4, Cumulative Total Losses as Percentage of OAK (%)

b

Cumulative gains is the sum of flow at WC4 plus all inflows indicated by increased streamflow between measurement points from WC4 to WC9.

Cumulative losses is the sum of all outflows indicated by decreased streamflow between measurement points from WC4 to WC9. c The average discharge from OAK during seepage runs was comparable to the average daily discharge of 224 L/s during January 1, 2012–September 30, 2015 (Figure 5) and 216 L/s during the 12 month period of July 1, 2014–June 30, 2015 (Table 1).

a

17.93 6.40 8.04 3.35 29.96 97.73 241.97 81.59 6.29 3.40 398.46 21.58 28.86

20120313 20120418 20120427 20120913 20120930 20121109 20130118 20140512 20140807 20141107 20150423 20150612 20150724

Date of Seepage WC9 − WC4, WC4, WC5, WC5 − WC4, WC9, WC9 − WC4, Run Measured Measured Change in Measured Cumulative Total Cumulative Total (YYYYMMDD) Lossesb (L/s) (L/s) (L/s) Streamflow (L/s) (L/s) Gainsa (L/s)

WC9 − WC4, WC5 − WC4, Cumulative Total Upstream Losses as Losses as OAK, Daily Percentage of Percentage of Average Total Gains (%) Flowc (L/s) OAK (%)

Table A2. Measured streamflow at selected stations on West Creek during seepage runs in 2012–2015, including the estimated total gains and total losses between stations WC4 and WC5 or WC9 and the comparison of losses from West Creek with discharge from Oak Hill Boreholes (site locations are shown in Figures 1 and 3).

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

Investigation of Streamflow Loss Near Abandoned Mines Table A3. Water-quality data for West West Branch and West Branch Schuylkill River, Schuylkill County, PA, 2012–2015.a West West Branch Schuylkill River Parameter Flow (L/s) Temp. (◦ C) DO Eh (mV) SC (mS/cm) pH (units) Alkalinity Net acidity Hardness SO4 , diss.

Cl, diss.

Ca, diss.

Mg, diss.

K, diss.

Na, diss.

Al, diss.

Fe, diss.

Mn, diss.

Ni, diss.

Zn, diss.

WC4b

WC5c

WC8

WC9

WWB

33.8 (3.57/683) 13.2 (1.5/17.6) 9.85 (8.0/15.2) 425 (320/610) 70 (60/270) 4.5 (4.3/5.5) 0.5 (0.3/1.1) 3.9 (1.3/5.1) 9.9 (9.2/17.0) 13.5 (12/18) 8.4 (4.6/73.7) 1.7 (1.6/3.7) 1.3 (1.2/1.9) 0.8 (0.7/1.3) 5.2 (3.2/40.6) 0.40 (0.20/0.50) 0.045 (0.03/0.07) 0.265 (0.26/0.33) 0.007 (<0.005/0.009) 0.034 (0.025/0.049)

45.3 12.7 10 510 77 4.4 0.7 4.1 9.6 12 10.9 1.7 1.3 0.9 6.4 0.40 0.04 0.26 0.008 0.052

34.6 (0/606) 15.2 (1.8/17.7) 9.4 (5.7/13.8) 313 (230/470) 150 (120/240) 5.8 (4.6/6.2) 2.8 (1.6/6.6) −1.4 (−5.7/1.1) 28.6 (18.3/50.6) 30 (15/48) 17.7 (11.9/62.7) 6.7 (3.7/12.2) 2.9 (2.2/4.9) 1.3 (0.7/2.0) 9.3 (7.4/34) 0.10 (<0.10/0.40) 0.02 (<0.01/0.04) 0.19 (0.11/0.23) 0.008 (0.006/0.010) 0.033 (0.025/0.039)

39.6 (1.76/651) 14.3 (1.9/17.8) 9.0 (6.9/14.5) 315 (230/410) 315 (160/790) 6.6 (5.6/7.0) 16.7 (2.0/78) −15.6 (−76.4/0.1) 108 (23.8/401) 99 (21/339) 13.0 (7.9/58.9) 30.0 (5.4/117) 7.9 (2.5/26.4) 2.0 (0.9/3.0) 9.9 (8.0/32.5) <0.10 (<0.10/0.30) 0.075 (0.02/0.27) 0.22 (0.20/0.42) 0.006 (<0.005/0.010) 0.029 (0.006/0.043)

470 (166/1940) 11.4 (0.1/18.6) 9.9 (8.3/12.5) 300 (210/500) 340 (210/550) 7.5 (6.5/7.9) 42 (20/99) −38.4 (−93.2/−19.0) 142 (64.4/241) 109 (45/147) 8.4 (5.1/45.2) 28.9 (13.3/44.7) 17.2 (7.6/26.8) 1.7 (1.2/2.0) 10.6 (7.0/32.4) <0.10 (<0.10/0.12) 0.04 (0.017/0.25) 0.22 (0.047/0.66) 0.010 (<0.005/0.018) 0.021 (0.005/0.076)

West Branch Schuylkill River Flow (L/s) Temp. (◦ C) DO Eh (mV) SC (mS/cm) pH (units) Alkalinity Net acidity Hardness SO4 , diss.

Cl, diss.

Ca, diss.

Mg, diss.

K, diss.

Na, diss.

Al, diss.

Fe, diss.

Mn, diss.

Ni, diss.

Zn, diss.

WB1 177 (0/937) 13.3 (0.1/20.2) 9.9 (5.6/13.1) 400 (330/600) 170 (100/1950) 5.0 (4.0/5.5) 1.3 (0/3.6) 5.72 (0.57/182) 52.6 (15.1/1140) 52 (15/1330) 15.7 (6.16/42.8) 9.7 (3.1/199) 6.81 (1.8/161) 0.8 (0.58/3.6) 11.8 (7.0/23.7) 0.96 (0.42/26.8) 0.097 (0.045/1.05) 0.469 (0.190/14.1) 0.020 (0.005/0.538) 0.054 (0.023/0.987)

PKN 561 (267/1170) 11 (10.4/11.6) 10.3 (6.7/10.6) 270 (210/370) 560 (500/620) 6.5 (6.0/7.0) 40 (30/47) −26.6 (−31.8/−21.5) 247 (219/282) 219 (186/267) 26 (18.4/40.4) 36.8 (32.3/44.6) 36.9 (33.7/45.1) 1.45 (1.18/1.7) 14.4 (11.6/19.6) <0.10 (<0.10/0.20) 4.7 (2.77/5.74) 2.10 (1.69/2.59) 0.045 (0.036/0.057) 0.098 (0.070/0.388)

OAK 222 (119/411) 14.8 (14.3/15.1) 1.9 (0.4/3.3) 210 (150/250) 970 (830/1010) 6.3 (5.9/6.8) 150 (120/170) −115 (−133/−84.3) 441 (398/516) 372 (351/495) 9.29 (8.17/13.6) 93.6 (86.1/114) 52.6 (44.6/58.2) 2.08 (1.6/2.4) 32 (30.4/34.5) 0.10 (<0.10/0.26) 17.0 (8.72/19.8) 3.49 (2.80/4.07) 0.036 (0.031/0.041) 0.034 (0.019/0.058)

WB2 1080 (413/2310) 12.1 (7.2/13.8) 9 (7.7/10.8) 250 (210/320) 560 (400/730) 6.4 (5.7/6.8) 50 (28/82) −35.4 (−61.2/−19.1) 240 (133/323) 210 (114/315) 21.9 (15.9/36.3) 39.9 (23/58.7) 34.3 (17.7/43) 1.3 (0.85/1.8) 17.4 (13.2/77.7) <0.10 (<0.10/0.13) 5.43 (3.18/8.60) 1.95 (1.25/2.98) 0.036 (0.023/0.049) 0.073 (0.052/0.099)

WB3 1250 (530/2970) 11.8 (7.1/15) 9.65 (8.7/11.2) 240 (170/290) 550 (420/680) 7.0 (6.3/7.2) 50.5 (34/69) −42.9 (−61.6/−28.2) 233 (133/309) 189 (105/279) 25.6 (17.3/68) 41 (25/57.7) 30.4 (17.3/40.2) 1.75 (1.28/2.8) 21 (15.4/37.5) <0.10 (<0.10/0.10) 2.11 (1.39/2.78) 1.78 (1.09/2.58) 0.032 (0.018/0.044) 0.052 (0.044/0.075)

Values are median (minimum/maximum); 20 quarterly samples per site, except one sample for WC5 and eight samples for WC4, WC8, and WC9 only during 2014–2015; tot. = total; diss. = dissolved (<0.45 mm); values in italics do not meet criterion continuous concentration (CCC), and values in bold italics exceed criterion maximum concentration (CMC), for freshwater aquatic life (U.S. Environmental Protection Agency, 2013). a Site descriptions are given in Table A1; site locations are shown in Figure 1. b The median values indicate typical base-flow characteristics, whereas the minimum and maximum values represent low-flow to high-flow characteristics. c Samples for laboratory chemical analysis were collected once at WC5. Those data plus repeated field measurements of pH and SC demonstrated the water quality at WC5 was indistinguishable from that at WC4.

sites on West Creek (WC9) and West West Branch (WWB) plus West Branch (WB3); however, fish were absent from the two upstream sites surveyed on West Creek (WC4) and West Branch (WB1). The two “fish-

less” upstream sites had low pH (≤5.5) and marginally elevated concentrations of dissolved aluminum (≥0.2 mg/L) and zinc compared to hardness-adjusted CCC thresholds (Table A3).

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

269

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer Table A4. Summary of fish taxa collected September 11, 2014, at sample sites on West Creek (WC4 and WC9), West West Branch (WWB), and West Branch (WB1 and WB3) of upper Schuylkill River basin, Schuylkill County, PA.a Fish Identified

West West Branch

West Branch

Common Name

Scientific Name

WC4

WC9

WWB

WB1

WB3

Blacknose dace Creek chub Fallfish White sucker Northern hog sucker Green sunfish Pumpkinseed Bluegill Bluegill × Green sunfish Largemouth bass Brown trout Brook trout

Rhinichthys atratulus Semotilus atromaculatus Semotilus corporalis Catostomus commersoni Hypentelium nigricans Lepomis cyanellus Lepomis gibbosus Lepomis macrochirus Lepomis sp.

Micropterus salmoides Salmo trutta Salvelinus fontinalis

0 0 0 0 0 0 0 0 0 0 0 0

25 21 0 1 0 2 5 10 1 0 0 23

25 7 0 4 7 5 1 14 1 1 5 22

0 0 0 0 0 0 0 0 0 0 0 0

0 0 2 10 0 0 0 0 0 0 2 6

0 0

88 8

92 11

0 0

20 4

Total fish Total species a

Fish were collected by electrofishing over a 150 m segment consisting of mixed riffle, run, and pool habitats. Captured fish were held for identification and measurement by Heather Eggleston and Robin Brightbill of USGS, checked for anomalies, and then released in accordance with methods described by U.S. Environmental Protection Agency (1993).

Despite their absence from upstream segments, fish of several species were collected in the perennial downstream segments of West Creek, West West Branch, and West Branch (Table A4), where the pH was near neutral and dissolved aluminum and zinc concentrations were less than aquatic CCC thresholds (Table A3).

Specifically, in the perennial segment of West Creek at Phoenix Park (WC9), eight fish species including 88 total individuals were recovered, including 23 brook trout (Salvelinus fontinalis). Creek chub (Semotilus atromaculatus) and Blacknose dace (Rhinichthys atratulus), which are tolerant of low pH (Butler et al., 1973), were also abundant. Further downstream on the West West Branch above West Branch (WWB), 11 species were identified from 92 total fish recovered, including 22 brook trout and five brown trout (Salmo trutta). On the adjacent West Branch (WB3), a total of four species were identified from 20 individual fish recovered, including six brook trout and two brown trout. White sucker (Catostomus commersoni), which is a pollution-tolerant species (Barbour et al., 1999), also was found at each of the three sites where fish were captured.

Benthic macroinvertebrate samples were collected on October 1, 2014, at the sites of fish surveys using a D-frame kick net (500 ␮m mesh) to capture debris and organisms dislodged from the streambed in accordance with methods of Barbour et al. (1999). Two shallow riffle areas of approximately 0.5 m2 (1 m2 total) were “kicked” upstream from the net for a total of 30 seconds each. Macroinvertebrate metrics were computed following guidelines of the Pennsylva-

270

nia Department of Environmental Protection (2013), including the total number of individuals (organisms), percentage of pollution-sensitive taxa (e.g., Ephemeroptera or mayflies), percentage of pollutiontolerant taxa (e.g., Chironomidae or midges), and various diversity indices that consider pollution tolerances including the Hilsenhoff Biotic Index (HBI; Hilsenoff, 1988) and Shannon Diversity Index (SDI; Pielou, 1966). The basic metrics were combined to compute “multimetric” composite scores for the Index of Biotic Integrity (IBI; Pennsylvania Department of Environmental Protection, 2013), the Macroinvertebrate Aggregated Index for Streams (MAIS; Smith and Voshell, 1997), and the Macroinvertebrate Biotic Integrity Index (MBII; Klemm et al., 2003).

On the basis of macroinvertebrate data (Table A5), the benthic macroinvertebrate communities in West Creek (WC4 and WC9), West West Branch (WWB), and West Branch (WB3) Schuylkill River reflect waterquality conditions ranging from unimpaired (very good to excellent) to impaired (fair to very poor). Although no aquatic macroinvertebrates were recovered at WB1, which was previously described as exhibiting poor habitat and water quality, a diverse variety of aquatic organisms was collected at the other four sites sampled (WC4, WC9, WWB, WB3).

The macroinvertebrate IBI scores for the West Creek and West West Branch sites indicated water quality “attaining” biological usage criteria and potentially qualifying for exceptional value (EV) anti-degradation protection, whereas the IBI scores for the West Branch site with macroinvertebrates (WB3) indicated impaired

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

Investigation of Streamflow Loss Near Abandoned Mines Table A5. Summary of metrics for macroinvertebrate taxa collected October 1, 2014, at sites on West Creek (WC4 and WC9), West West Branch (WWB), and West Branch (WB1 and WB3) of the upper Schuylkill River basin, Schuylkill County, PA.a West West Branch Metricsb Total taxa richness (count of taxa) Total number (sum of individuals) Total number/m2 Percent dominant taxa (single) Percent dominant taxa (5 dominant) Total EPT (sum of all blue rows) Percent EPT Total EPT (PTV 0–4) Percent EPT (PTV 0–4) Percent Ephemeroptera Number Ephemeroptera taxa Number Plecoptera taxa Number Trichoptera taxa Beck’s Index version 3 Intolerant taxa PTV ≤5 Percent sensitive individuals (PTV 0–3) Number Chironomidae taxa Percent Chironomidae Percent non-insect taxa Number collector-filterers (CF) % scrapers (SC) % haptobenthos (CL + SP, SC + PR) Shannon Diversity Index (for IBI) Simpson Diversity Index (for MAIS) Hilsenhoff Biotic Index (HBI) HBI water quality designation PADEP IBI score PADEP IBI designation MAIS score MAIS classification MBII score MBII classification

West Branch

WC4

WC9

WWB

WB1

WB3

14 214 214 49.1 93.5 149 69.6 139 65.0 0 0 121 28 296 141 65.4 55 25.7 2.3 26 0.5 0.5 1.46 0.68 2.64 Excellent 92.3 Attaining EV 11 Fair 48.7 Fair

20 204 204 38.2 82.4 63 30.9 60 29.4 0.5 1 46 16 148 110 49.0 78 38.2 2.5 10 0.5 0 2.03 0.79 3.87 Very Good 93 Attaining EV 12 Fair 46.7 Fair

22 297 297 34.3 76.4 205 69.0 42 14.1 24.6 73 3 129 63 158 14.5 55 18.5 6.7 124 12.1 0 2.19 0.82 5.17 Fair 84 Attaining EV 16 Good 42.7 Fair

0 0 0 100.0 100.0 0 0.0 0 0.0 0 0 0 0 0 0 0.0 0 0 0 0 0 0 0.00 1.00 10 Very Poor 0 Impaired 2 Poor 0 Poor

7 200 200 95.5 99.0 2 1.0 2 1.0 0 0 2 0 4 6 1.5 191 95.5 1.5 0 1 0 0.26 0.09 5.97 Fairly Poor 20.6 Impaired 3 Poor 9.2 Poor

EPT = Ephemeroptera (mayflies), Plecoptera (stoneflies), or Trichoptera (caddisflies); PTV = pollution tolerance value; CL = clingers; SP = sprawlers; PR = predators; IBI = Index of Biotic Integrity; MAIS = Macroinvertebrate Aggregated Index for Streams; MBII = Macroinvertebrate Biotic Integrity Index. a Macroinvertebrate samples collected and identified by Heather Eggleston of USGS using D-frame kick net (500 ␮m mesh) over 1 m2 riffle kick area in accordance with Barbour et al. (1999). b Metric computations by Charles Cravotta follow guidelines of Pennsylvania Department of Environmental Protection (PADEP) (2013).

quality. Two other aquatic quality multimetric indices, the MAIS and MBII, were correlated with the IBI scores; these metrics indicated “fair” to “good” aquatic quality at WC4, WC9, and WWB and “poor” aquatic quality at WB3. Chironomids, which are tolerant of pollution, were the dominant taxa at WB3, comprising more than 95 percent of the 200 organisms counted; only 1 percent of the organisms were identified as pollution-intolerant “EPT” taxa, i.e., Ephemeroptera (mayflies), Plecoptera (stoneflies), or Trichoptera (caddisflies). At the three sites within the West West Branch sub-basin (WC4, WC9, and WWB), EPT taxa ranged from 30.9 to 69.6 percent of the total organisms counted, compared to chironomids, which ranged from 18.5 to 38.2 percent of the total taxa. The percentage of

pollution-sensitive individuals increased from 1.5 percent at WB3 to 14.5, 49.0, and 65.4 percent at WWB, WC9, and WC4, respectively. It is notable that stoneflies and caddisflies were the dominant EPT taxa at WC4 and WC9, whereas caddisflies and mayflies were dominant at WWB. In the perennial headwaters site on West Creek (WC4), a high-quality macroinvertebrate assemblage, including pollution-sensitive stoneflies (Plecoptera) and caddisflies (Trichoptera), was collected, despite the absence of fish (Table A5). The habitat at WC4 appeared relatively pristine, with sustained streamflow across a clean gravel substrate in pools formed by sandstone boulders (Figures 2A and 2B). Nevertheless, the water quality at WC4 had moderately low pH (4.3

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

271

Cravotta, Sherrod, Galeone, Lehman, Ackman, and Kramer

Figure A2. Summary changes in the pH, acid-neutralizing capacity (ANC), and specific conductance (SC) during limestone sand kinetic experiment for neutralization of water from West Creek at WC4, September 18, 2015. A total of 0.80 kg of <4 mm washed limestone “sand” was added at once to 16 L of water from West Creek at WC4.

to 5.5) and moderately elevated concentrations of dissolved aluminum (0.20 to 0.50 mg/L) and zinc (0.025 to 0.049 mg/L), none of which meets CCC thresholds (Table A3).

Limestone Sand Could Neutralize Acidity A simple kinetic experiment was performed on September 18, 2015, to evaluate the potential to neutralize the net-acidic water from the headwaters of West Creek through limestone sand dosing (e.g., Cravotta, 2010; Cravotta et al., 2010). A total of 0.80 kg of <4mm-diameter washed limestone “sand” was added at once to 16 L of water collected from West Creek at WC4. The water was gently stirred, and at 1 minute elapsed time, the pH increased from 4.4 to 5.2, while the alkalinity (acid-neutralizing capacity) increased from <2 to 4 mg/L as CaCO3 . After 5 minutes, the pH increased to 6.2, and the alkalinity to increased 5.9 mg/L as CaCO3 (Figure A2). Thus, relatively short contact time with limestone sand would be needed for the stream water to be neutralized. Given an average daily flow of 69 L/s and an average net acidity of 3.9 mg/L as CaCO3 , the total annual acidity load is 8.54 tonnes/yr at WC4. An equivalent amount of limestone would be needed to neutralize this acidity. Assuming a CaCO3 purity of 95 percent for the limestone, approximately 200 tonnes would provide a 20 year supply. Thus, end-dumping a large amount of limestone sand into West Creek at an upstream location(s) could increase the pH and alkalinity (marginally) and, presumably, facilitate removal of the dissolved Al and Zn, which are near aquatic toxicity thresholds.

272

APPENDIX REFERENCES BARBOUR, M. T.; GERRITSEN, J.; SNYDER, B. D.; AND STRIBLING, J. B., 1999, Rapid Bioassessment Protocols for Use in Streams and Wadeable Rivers—Periphyton, Benthic Macroinvertebrates, and Fish: U.S. Environmental Protection Agency Report EPA 841-B-99-002, 11 Chapters, 4 Appendixes.

BUTLER, R. L.; COOPER, E. L.; CRAWFORD, J. K.; HALES, D. C.; KIMMEL, W. G.; AND WAGNER, C. C., 1973, Fish and Food Organisms in Acid Mine Waters of Pennsylvania: U.S. Environmental Protection Agency Report EPA-R3-73-032, 158 p.

CRAVOTTA, C. A., III, 2010, Abandoned mine drainage in the Swatara Creek Basin, Southern Anthracite Coalfield, Pennsylvania, USA—2. Performance of passive-treatment systems: Mine Water and the Environment, Vol. 29, pp. 200–216. (http://dx.doi.org/10.1007/s10230-010-0113-5) CRAVOTTA, C. A., III; BRIGHTBILL, R. A.; AND LANGLAND, M. J., 2010, Abandoned mine drainage in the Swatara Creek Basin, Southern Anthracite Coalfield, Pennsylvania, USA— 1. Streamwater-quality trends coinciding with the return of fish: Mine Water and the Environment, Vol. 29, pp. 176–199. (http://dx.doi.org/10.1007/s10230-010-0112-6).

HILSENHOFF, W. L., 1988, Rapid field assessment of organic pollution with family-level biotic index: Journal of North American Benthological Society, Vol. 7, pp. 65–68.

KLEMM, D. J.; BLOCKSOM, K. A.; FULK, F. A.; HERLIHY, A. T.; HUGHES, R. M.; KAUFMANN, P. R., PECK, D. V.; STODDARD, J. L.; THOENY, W. T.; GRIFFITH, M. B.; AND DAVIS, W. S., 2003, Development and evaluation of a macroinvertebrate biotic integrity index (MBII) for regionally assessing mid-Atlantic highlands streams: Environmental Management, Vol. 31, No. 5, pp.

656–669.

PENNSYLVANIA DEPARTMENT OF ENVIRONMENTAL PROTECTION., 2013, A Benthic Macroinvertebrate Index of Biotic Integrity for Wadeable Freestone Riffle-Run Streams in Pennsylvania: Pennsylvania Department of Environmental Protection, Bureau of Point and Non-Point Source Management, Division of Water Quality Standards, Harrisburg, PA, 145 p.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

Investigation of Streamflow Loss Near Abandoned Mines PIELOU, E. C., 1966, Shannon’s formulae as a measure of specific diversity: Its use and misuse: American Naturalist, Vol. 100, pp.

463–465.

SMITH, E. P. AND VOSHELL, J. R., Jr., 1997, Studies of Benthic Macroinvertebrates and Fish in Streams within EPA Region 3 for Development of Biological Indicators of Ecological Condition. Part 1. Benthic Macroinvertebrates: U.S. Environmental Protection Agency, Washington, DC, USA, Final Report for Cooperative Agreement CF821462010.

U.S. ENVIRONMENTAL PROTECTION AGENCY, 1993, Fish Field and Laboratory Methods for Evaluating the Biological Integrity of Surface Waters: U.S. Environmental Protection Agency Report EPA 600/R-92/111, 348 p.

U.S. ENVIRONMENTAL PROTECTION AGENCY, 2013, National Recommended Water Quality Criteria—Aquatic Life Criteria Table: U.S. Environmental Protection Agency, https://www.epa.gov/ wqc/national-recommended-water-quality-criteria-aquaticlife-criteria-table. (last updated August 22, 2013).

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 243–273

273

A Design Method for Landslide Surface Water Drainage Control BENJAMIN D. HAUGEN Maptek, North America, 14143 Denver West Parkway, Suite 200, Golden, CO 80401

Key Terms: Drainage, Engineering Geology, Erosion and Sedimentation, Landslides, Slope Stability, Surface Water ABSTRACT Infiltration of surface water increases pore water pressures in slopes and reduces their stability. Common landslide features such as tension cracks and sag ponds can act as preferential pathways for surface drainage and may increase infiltration and exacerbate pore pressure– induced instability. Surface water drainage control is likewise recommended by numerous authors as an effective and inexpensive landslide mitigation method and has been shown to reduce the risk of landslides. While robust design procedures for other geotechnical applications exist (e.g., slope reduction, subsurface drains), similar procedures for landslide surface water drainage control have remained largely ad hoc and vary among practitioners. The objective of this article is to summarize technical literature related to surface water drainage control and provide a coherent design procedure for landslides.

INTRODUCTION Surface water drainage is a major concern in urban and semi-urban areas where structures and roadways impede infiltration (FAO, 1998; Brown et al., 2001).

Thus, a major portion of the criteria and procedures for surface water drainage control have been developed by governmental agencies for use in populated areas and transportation corridors (Brown et al., 2001; CDOT, 2004; King County, 2009; and GEO, 2011).

Drainage design for the purposes of irrigation and erosion control is also a major concern (FAO, 1998). Many publications address landslide surface drainage control (Schuster, 1978; Cedergren, 1989; Rogers, 1992; Chatwin et al., 1994; Turner and Schuster, 1996; Anderson et al., 2008; Highland and Bobrowsky, 2008; and GEO, 2011). Unfortunately, none provide a cogent “start-to-finish” design method, and most are focused on the design of subsurface drains (e.g., horizontal drains, French drains).

Several studies have shown the effectiveness of surface water drainage for landslide mitigation (Yilmazer et al., 2003; Mikos et al., 2005; and Mizal-Azzmi et al., 2011). Others specifically mention surface drainage as a necessary component of landslide mitigation and risk reduction plans (Chatwin et al., 1994; Anderson et al., 2008; Holcombe et al., 2012; and Sajinkumar et al., 2014). Numerous studies present photographs (Figures 1 and 2) and schematics of completed surface drainage control systems on landslides (Rogers, 1992; Mikos et al., 2005; Anderson et al., 2008; Highland and Bobrowsky, 2008; and Mizal-Azzmi et al., 2011), while comparatively few include landslide-specific design criteria and recommendations (e.g., FAO, 1988; Highland and Bobrowsky, 2008). While not exhaustive, this literature review demonstrates that in many cases, landslide practitioners have had to rely on disparate and often incomplete surface drainage design recommendations.

Transportation-specific runoff and landslide designs require similar data, and many of the standard design calculations used in transportation projects are necessary for landslide-specific surface water control (e.g., peak discharge, flow velocity, etc.). Yet the variability of location, risk to life and property, and geologic and geomorphic settings of landslides necessitate unique design criteria (e.g., rainfall event magnitudes, drain types). Landslide characteristics such as active slope movement and tension cracks further limit design options and require that they deviate from conventional urban and transportation-specific guidance. This article is intended to help practitioners identify (1) the key pre-design considerations, (2) the basic steps in the design process, (3) the key design criteria, and (4) how the criteria are estimated when designing a landslide surface water control system.

PRE-DESIGN CONSIDERATIONS Landslides have characteristics that require a unique approach to surface water drainage control. The following is a brief list of pre-design considerations for active landslides and areas with high landslide susceptibility. Each consideration is posed in the form of a question to assist practitioners in determining if surface water drainage control is a viable mitigation method. If

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275–289

275

Haugen

the answer to all the questions is “yes,” surface drainage is a reasonable mitigation method. If not, alternative landslide mitigation or risk reduction plans may be more appropriate.

For active landslides, where controlling surface water drainage is intended to slow or prevent slope movement,

Figure 1. Examples of concrete surface water drainage channels used to stabilize slopes in urban areas from Mizal-Azzmi et al.

(2011). Note the “interceptor” drain at the top of the slope in (a).

a) Will surface water drainage control have an appreciable effect on landslide motion? In cases in which infiltration rates are extremely low or pore pressures at the slip surface are elevated by confined head pressures, surface drainage may not reduce landslide activity or risk. b) Is failure of the drainage system due to landslide motion avoidable? Drainage structures may be breached if they cross shear boundaries or straddle landslide blocks. The geometry and composition of the drainage system are crucial to preventing system damage if motion is not halted by its installation. c) Is construction time sufficiently short? Landslide motion during construction may cause failure to achieve design specifications and/or complete failure of the drainage system. Related to (a), this is of particular concern where the drainage control system is expected to halt landslide motion. d) Is the probability of a system failure acceptable? The uncontrolled release of concentrated flows in ditches and other drainage structures may trigger new landslide events and/or increase landslide risk. Uncontrolled releases downslope may change the stability of slopes outside the project area.

In areas with high landslide susceptibility, where drainage is designed to reduce landslide hazard and/or risk, ask the following: Is the landslide hazard controlled by surface water infiltration or by other factors? While the majority of landslide events are triggered by heavy precipitation events (Lu and Godt, 2013), landslide susceptibility also depends on other factors (e.g., slope, earthquake shaking, volcanic activity, etc.). Drainage control may not be the best method for reducing risk where water is not the primary concern.

DESIGN PROCESS

Figure 2. Example of a geotextile-lined surface water drainage channel on an engineered slope. Photograph by Department of Transport, Energy and Infrastructure, South Australia (modified from Highland and Bobrowsky [2008]).

276

The method for designing landslide surface water drainage systems outlined here was developed using available literature and may differ from other practitioners’ own methods. It may not be practical or possible to undertake every detail of the design process as a result of budgetary or logistical constraints, but it can nonetheless serve as a design guide. The design process comprises six general steps, as summarized in Figure 3.

For further clarification, an example design is included at the end of this article.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275–289

Landslide Surface Water Drainage Control

Figure 4. Example of landslide features that may affect landslide drainage that are identifiable in a DEM. The image is a DEM created using 2013 LiDAR data from the area where the SR530 Landslide occurred on March 22, 2014 (a.k.a. Oso Landslide). Lighter shading represents shallower slopes. The shaded relief was adapted from the red relief image created by Asia Air Survey Co., Ltd, available at https://www.ajiko.co.jp/en/recent_activations02.html.

Figure 3. Summary schematic of the landslide surface water control method in this article. The details of each design step are described in the paragraphs that follow.

Step 1—Gather Topographic and Structure Data Catchment areas, slopes, and natural drainage patterns (FAO, 1998; GEO, 2011) as well as the presence of engineered structures and existing drainage networks (Brown et al., 2001; CDOT, 2004) are important design factors. They should be considered for both the landslide in question and its upslope drainage areas.

Topographic data can be used to calculate catchment areas, mean slopes, and maximum drainage distances that are useful in later design steps.

A variety of topographic data sources exist. The U.S. Geological Survey (USGS) provides topographic maps and Digital Elevation Models (DEMs) at many scales and accuracies via the online National Map Viewer: http://viewer.nationalmap.gov/viewer/. Similar data are available for other nations, but it is recommended that local sources be consulted to ensure the most accurate and up-to-date data are obtained. In remote areas, topographic data may be difficult to obtain, may have insufficient spatial resolution, and/or may be out-

dated (e.g., because of recent slide movement). In these cases, it may be necessary to conduct a topographic survey or collect LiDAR data. While LiDAR data collection can be cost-prohibitive, accurate topographic data is extremely helpful for surface water drainage design. Regardless of its source or quality, field checks of the topographic data are recommended; they may reveal differences important to the design (e.g., existing drainage systems). Wherever possible, it is also recommended that as-built plans and design specifications for existing storm water and other surface water drainage systems in the project area be obtained. Figure 4 is an example of an instance in which landslide features that affect surface water drainage are identifiable in LiDAR DEM data.

Step 2—Identify Infiltration Short Paths and Plan Mitigation Common landslide features such as scarps, tension cracks, and sag ponds can reduce flow path lengths between surface and groundwater, increasing infiltration and/or impeding surface water drainage (FAO, 1988; Cedergren, 1989; and Turner and Schuster, 1996).

Wherever possible, it is recommended that the number and size of these features be minimized using appropriate grading, excavation, and backfill techniques. In this step, the infiltration short paths are identified and appropriate mitigation methods are planned. The results of this step are included in the final design resulting from Step 6. Mitigation work is typically concurrent with system construction.

Infiltration short paths can be identified from field reconnaissance and topographic data (Step 1). It is

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275–289

277

Haugen

recommended that field reconnaissance include mapping and measurement of short path length, width, depth, and exact position. Topographic data can be used to guide field reconnaissance and used to measure short path dimensions and positions.

Short path mitigation may require significant backfill, and excavation is typically performed during the construction of the drainage system (following Step 6).

Where landslide or earth work activity may preclude a safe work environment, alternative landslide risk reduction methods are recommended. The definition of what precludes a safe work environment is outside to scope of this article and is entirely at the discretion of qualified person(s) involved in the project.

It is recommended that fine-grained, lowpermeability backfill materials be collected from on-site excavations (although any source for such materials will do). In cases in which berms or hummocks obstruct surface water drainage, they can be excavated and/or re-graded. Because of their geometry, it can be difficult to confirm fissures and whether sag ponds have been backfilled and compacted properly.

Specialized compaction techniques and/or additional settlement time may be required. The specifics of those techniques are outside the scope of this article.

Required backfill volumes are calculated using the short path measurements made in this step. Tension crack and fissure volumes can be modeled as triangular prisms (Figure 5a, Eq. 1). While scarps can be significant short paths, surface expressions of fissures forming “gaps” between the scarp and the landslide mass may not be apparent or measureable. If separation is measurable, scarp backfill volumes can be modeled and mitigated, as is the case for tension cracks.

Sag ponds can be modeled as half-ellipses where the dimensions of the pond are measured as the maximum length, width, and depth (Figure 5b, Eq. 2).

Tension crack and/or fissure volume: Vtc =

1 dwl; 2 (1)

2 ␲dwl, (2) 3 where V = short path volume; d = feature depth; w = feature width; and l = feature length.

Removal of buttressing slope materials and/or addition of backfill mass may initiate or exacerbate slope failures (Turner and Schuster, 1996; Highland and Bobrowsky, 2008). Likewise, a thorough assessment of the in situ stability of the project area prior to earth work activities is recommended. To avoid removal of buttressing mass, excavations at the sides and toes of active slides are not recommended. Similarly, a net increase in slide mass is not recommended. Headward excavaSag pong volume:

278

Vsp =

Figure 5. Schematics of (a) fissure and (b) sag pond modelling methods for excavation and backfill volume calculations using Eqs. 1 and 2.

tion is a common mitigation technique (Chatwin et al., 1994; Turner and Schuster, 1996; Cornforth, 2005; and GEO, 2011) and can provide backfill materials.

The planning of short path mitigation efforts should overlap partly with drainage network design (Step 3), as grading, excavation, and backfill activities may change surface drainage conditions. Mitigation of infiltration short paths is most easily accomplished during construction of the surface water drainage control system (following Step 6). Post-construction, it is recommended that short path cutoff measures be checked during annual (or more frequent) inspections.

Step 3—Design Preliminary Drainage Network The short path mitigation measures planned in Step 2 may not reduce infiltration rates enough to slow or stop landslide movement. Surface water drains are necessary to provide maximum infiltration reduction. It is recommended that landslide drainage designs utilize existing natural drainages and the infrastructure identified in Step 1. Existing downstream water conveyance systems may present design limitations later and likewise may need to be considered in this preliminary design phase. In this step, the general location, number, and orientation of drains are planned.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275–289

Landslide Surface Water Drainage Control

The information gathered in Steps 1 and 2 can provide useful guidance for selection of the number and distribution of surface drains in the preliminary design.

A “toe” drain for gathering water not intercepted by the chevron drains is also recommended, except where existing drains, such as roadside ditches, are proximal to the toe of the landslide zone.

A minimum drain gradient of 2 percent is recommended (Highland and Bobrowsky, 2008). Gradients greater than 10 percent may result in severe erosion and/or very high flow rates (Brown et al., 2001). To minimize excavation and re-grading requirements, it is recommended that the position and orientation of drains closely follow the natural topography; sinuous, complex drainage networks increase the uncertainty of flow estimates (Step 6) and may increase erosion rates, reducing system reliability (FAO, 1998). Creating multiple preliminary designs may increase design flexibility.

The preliminary drainage network(s) developed in this step will be refined later in the design process (Steps 4 through 6).

Damage to the drainage network will increase maintenance costs and may initiate or exacerbate slope failures. Likewise, the drainage network should be designed to minimize the number of locations in which drains cross active shear zones, tension cracks, and scarps. Construction and maintenance costs can also be reduced by designing the system with as few drains as possible.

Step 4—Estimate Peak Surface Water Discharge Figure 6. Examples of chevron surface water drainage patterns from (a) Rogers (1992) and (b) GEO (2011).

Interceptor drains that convey and discharge surface water downslope of unstable terrain have been shown to reduce landslide susceptibility (Anderson et al., 2008; Holcombe et al., 2012). It is recommended that interceptor drains be installed immediately upslope of active or potentially active landslide areas (Figure 1).

The most “secure” interceptor design will include a network of drains that encircles the entire landslide zone.

Unfortunately, this type of design is often financially or logistically infeasible. In such cases, shorter and/or smaller drains should be designed, with careful consideration of the guidance provided below.

Within the margins of active and unstable slopes, surface water drains are typically constructed in a chevron pattern (Figure 6; Rogers, 1992; Cornforth, 2005; and GEO, 2011). In areas with relatively consistent topography, a single chevron (a “trunk” drain with splaying “branches”) is recommended. Large areas and/or complex topography may require multiple chevrons.

Peak surface water flow controls the size and type of drains required and may influence the feasibility of the preliminary drainage design(s) created in Step 3.

Peak flows also control the impacts on downstream infrastructure. Several methods have been developed to calculate peak surface water runoff, but the “rational” method is most widely used and is considered appropriate for landslide areas of less than 200 acres (80 hectares) (FAO, 1988, 1998; Brown et al., 2001; CDOT, 2004; and GEO, 2011). In this step, individual and total peak flows of the drains designed in Step 3 are estimated.

The Rational Method Using the rational method, peak runoff flow, Qr , for a given drain is dependent on rainfall intensity, i, drainage area, A, and an empirical runoff coefficient, C, and is calculated using Eq. 3: Peak runoff flow: Qr = Ci A,

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275–289

(3)

279

Haugen

where Qr = peak runoff flow (ft3 /s); C = runoff coefficient (dimensionless); i = peak rainfall intensity (in./hr); and A = drainage area (acres).

The rational formula requires a number of assumptions (FAO, 1998): (1) The rate of runoff must equal the rate of supply (rainfall excess) if the rainfall duration is greater than or equal to the time of concentration, tc ; (2) The maximum discharge occurs when the entire area is contributing runoff simultaneously; (3) At equilibrium, the duration of rainfall at intensity i is t = tc ; (4) Rainfall is uniformly distributed over the basin; (5) The recurrence interval of Qr is the same as the frequency of occurrence of rainfall intensity i; and (6) The runoff coefficient is constant between storms and during a given storm and is determined solely by basin surface conditions.

Determining the Runoff Coefficient Runoff coefficients primarily depend on ground surface properties. Coefficient values for a variety of land use types are provided in Table 1. If a given drainage area has a single land use type, values of C can be selected directly from Table 1. Longer return periods and steeper slopes will lead to higher C values (FAO, 1998; CDOT, 2004). The maximum value for a given land use type in Table 1 (e.g., C = 0.30 for unimproved areas) is likewise recommended as a conservative overestimate of the runoff coefficient.

In areas with multiple land use types, a weighted runoff coefficient should be used (Brown et al., 2001).

To do so, the areas comprising each land use type are estimated, and the maximum C for each is selected from Table 1. The weighted runoff coefficient, Cw , is calculated using Eq. 4: Weighted runoff coefficient: n i =1 (Ci Ai ) , Cw = n i =1 (Ai )

(4)

where Cw = weighted runoff coefficient (dimensionless); Ci = runoff coefficient for a given land use (dimensionless); and Ai = area of a given land use type (acres).

Table 1. Runoff coefficients for use in the rational method. Adapted from Brown et al. (2001).

Type of Drainage Area Business Downtown areas Neighborhood areas Residential Single-family areas Multi-units, detached Multi-units, attached Suburban Apartment dwelling areas Industrial Light areas Heavy areas Parks, cemeteries Playgrounds Railroad yard areas Unimproved areas Lawns Sandy soil, flat, 2% Sandy soil, average, 2–7% Sandy soil, steep Heavy soil, flat, 2% Heavy soil, average, 2–7% Heavy soil, steep, 7% Streets Asphaltic Concrete Brick Drives and walks Roofs

Runoff Coefficient, C* 0.70–0.95 0.50–0.70 0.30–0.50 0.40–0.60 0.60–0.75 0.25–0.40 0.50–0.70 0.50–0.80 0.60–0.90 0.10–0.25 0.20–0.40 0.20–0.40 0.10–0.30 0.05–0.10 0.10–0.15 0.15–0.20 0.13–0.17 0.18–0.22 0.25–0.35 0.70–0.95 0.80–0.95 0.70–0.85 0.75–0.85 0.75–0.95

*

Higher values are usually appropriate for steeply sloped areas and longer return periods because infiltration and other losses have a proportionally smaller effect on runoff in these cases.

in the peak runoff flow equation (Eq. 3), drainage area, A, will vary the most at a given site and will thus be the primary control on the final design specifications of the drainage system.

No specific methodologies for measuring drainage area are discussed in the texts reviewed for this article. However, GIS-based hydrologic tools that utilize DEMs are accurate, time-efficient, and consistent (Figure 7). Where DEMs are unavailable (see Step 1), topographic maps can be used to measure drainage areas by hand.

Measuring Drainage Area

Estimating Peak Rainfall Intensity

DEMs can be used to calculate the area a drain will empty and to identify the location of streams and other natural drainage features. Both are essential to developing an appropriate design and must be considered alongside the effects of planned construction and/or mitigation activities (GEO, 2011). Of the three terms

The rainfall intensity, i, used to calculate peak flows is proportional to the chosen rainfall event return period which depends on the desired reliability of the drainage system (FAO, 1998; CDOT, 2004; and GEO, 2011). An appropriate rainfall event return period can be estimated using the desired design life and

280

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275–289

Landslide Surface Water Drainage Control

water to travel from the farthest point in a drainage area to its drainage exit (FAO, 1998; Brown et al., 2001; CDOT, 2004; and GEO, 2011). Time of concentration includes the time required for three types of flow: sheet flow, shallow concentrated flow in rills, and open-channel flow (Brown et al., 2001). According to CDOT (2004), in small basins channel flow is negligible, and time of concentration can be estimated using Eq. 6: Time of concentration: tc =

where tc = time of concentration in small basins (minutes); C = runoff coefficient used in rational equation (dimensionless); D = flow path length (ft); and S = flow path slope (percent).

Flow path length, D, and mean flow path slope, S, are calculated using the data gathered in Step 1. Flow path length for overland/rill-type flow is defined as the maximum distance between the edge of a drainage area and the exit point of its primary drain. Straight-line distances can be used, but it is generally more conservative to use the length of the longest natural channel in the drainage area. Note that “natural channels” do not have to be streams or gullies. Rather, D is the longest anticipated flow path, based on variations in topography and existing infrastructure. Flow path slope, S, is calculated using the change in elevation along the flow path length D. Slope is the change in elevation divided by the path length and is expressed as a percent.

Once the time of concentration has been calculated for each drainage area, depth-duration-frequency (DDF) rainfall curves (Figure 8) can be used to estimate the design peak rainfall intensity, i, for each drain.

The National Weather Service (NWS) maintains DDF curves for most areas of the United States on their Precipitation Frequency Data Server (PFDS), available online: http://hdsc.nws.noaa.gov/hdsc/pfds/. For a given return period and event duration (i.e., time

Figure 7. Example of delineation of drainage areas in GIS. Colored areas are distinct drainages. Light blue lines are natural flow accumulation zones and/or existing stream channels. Background image is hillshaded DEM.

acceptable probability of failure, as in Eq. 5: Rainfall event return period:

T=

1 1 − (1 − P)1/n

1.8 (1.1 − C) D1/2 , (6) S1/3

,

(5) where T = rainfall event return period (years); n = design life (years); and P = acceptable probability of failure (percent).

Design life and acceptable probability of failure are at the discretion of the designer, but risk analyses or local practitioners’ standards may be helpful in determining appropriate n and P values. As a reference, Table 2 lists return periods for various combinations of design lives and failure probabilities.

Peak runoff flows will occur at a time equal to the time of concentration, tc , for a given return period. The time of concentration is the time it takes for a unit of

Table 2. Return periods (T) for given design life (n) and probability of failure (P). Adapted from FAO (1998).

Probability of Failure During Design Life Design Life (yr) 5 10 20 30 40 50 75 100

1%

5%

498 995 1,990 2,985 3,980 4,975 7,463 9,950

98 195 390 585 780 975 1,463 1,950

10%

20%

30%

40%

50%

Rainfall Event Return Period (yr) 48 23 15 95 45 29 190 90 57 285 135 85 380 180 113 475 225 141 712 337 211 950 449 281

10 20 40 59 79 98 147 196

8 15 29 44 58 73 109 145

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275–289

281

Haugen

Figure 8. Example of DDF curves from the NWS Web site: http://hdsc.nws.noaa.gov/hdsc/pfds/. This type of curve is used to determine precipitation depth, given a certain return period and time of concentration, for any location in the United States. These data are also available in tabular format.

of concentration), DDF data curves yield a rainfall depth, dr , which is then used to calculate peak rainfall intenstity via Eq. 7: Peak rainfall intensity: i = 60

dr , tc

(7)

where i = peak rainfall intensity (in./hr); dr = DDF rainfall depth (in.); and tc = time of concentration or event duration (minutes).

Wherever possible, it is recommended that local rainfall data be used to select design rainfall intensities (FAO, 1998; Brown et al., 2001; CDOT, 2004; and GEO, 2011). In regions for which data are unavailable, regression equations developed by the USGS can be used to estimate peak discharge rates, as described

by Brown et al. (2001). Similarly, and as a result of large local variations in rainfall intensity in mountainous terrain, it is recommended that design values be carefully estimated and reviewed.

to damage from vegetation growth and burrowing animals (Rogers, 1992; GEO, 2011). Damaged or blocked French drains likewise have the potential to act as infiltration short paths that may increase pore pressures.

Because of these risk factors, in most cases open channels are preferable to French drains and other closed drainage structures, such as pipes and culverts, for landslide drainage control (GEO, 2011).

Open channels have the following advantages over French and closed drains: (1) They are relatively inexpensive to excavate.

(2) A variety of construction materials and designs can be used in their construction.

(3) They have low susceptibility to siltation and blockage.

(4) Their repair and maintenance are relatively simple and inexpensive.

(5) It is relatively easy to identify and correct drainage problems.

(6) They are adaptable to complex topographies.

Step 5—Design Drainage Structures A variety of drainage structures can be used for landslide surface water control. The most cost-effective and commonly used are ditches and channels (Figures 1 and 2; Cedergren, 1989; Cornforth, 2005; Highland and Bobrowsky, 2008; and GEO, 2011). Open gravelbackfill trenches—or “French” drains—are also common (Schuster, 1978; Cedergren, 1989; Rogers, 1992; Chatwin et al., 1994; Yilmazer et al., 2003; Cornforth, 2005; Mikos et al., 2005; and Mizal-Azzmi et al., 2011). French drains have the advantage of providing both surface and subsurface drainage, but they tend to clog with silt and debris and are susceptible

282

Channel Type GEO (2011) and FAO (1988) recommend using Ushaped channels lined with concrete, asphalt, or pipe sections. While these materials are less susceptible to erosion, landslide movement may damage their rigid structures, increasing the frequency and cost of repairs. Earthen ditches are recommended for landslide drainage control because they can deform while remaining functional, are common in transportation projects (FAO, 1998; Brown et al., 2001; and CDOT, 2004), are relatively inexpensive, and can be repaired with simple equipment at low cost (Anderson et al.,

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275–289

Landslide Surface Water Drainage Control

uous channels will complicate hydraulic analyses and increase the potential for erosion and bank overflow.

Channel Dimensions Peak runoff discharge for a given channel, Qr , calculated in Step 4 must be less than or equal to its expected discharge capacity, Qc (Eq. 8): Channel discharge capacity: 1.486 AR2/3 S1/2 , Qr ≤ Qc = n Figure 9. Schematic of recommended channel shapes modified from FAO (1998).

2008; Holcombe et al., 2012). Rigid channels are only recommended where slope movement is small and high flow velocities are expected.

To prevent erosion, vegetation overgrowth, and infiltration, it is recommended that all earthen channels be lined with an impermeable or semi-impermeable geotextile membrane. The geotextile can be covered with coarse gravel or rip rap to prevent movement of the geotextile, reduce flow velocities, and/or improve aesthetics. Covering the geotextile also helps prevent degradation from ultraviolet light, but may simultaneously increase siltation and debris buildup. The specifics of geotextile selection and design are beyond the scope of this article, but it is recommended that installation, inspection, and maintenance of the channel liner be performed per the manufacturer’s specifications.

Channel Shape It is recommended that drainage channels be either trapezoidal or triangular (Figure 9). These channel shapes are relatively simple to construct and are commonly used in transportation and erosion control practice (FAO, 1998; CDOT, 2004). Triangular channels are generally simpler to design and construct. Trapezoidal channels require wider right of way but have larger conveyance capacity for a given depth, d. Right-of-way requirements and excavation costs are the most important determinants of channel shape. Channel sides sloped between 1H:1V and 2H:1V are recommended, depending on material type. Side slopes steeper than 1H:1V may be prone to collapse, especially if channel incision has occurred, and should generally be avoided.

Side slopes shallower than 2H:1V may not provide enough conveyance capacity. It is also recommended that drainage channels be as linear as possible. Sin-

(8)

where Qr = peak runoff flow (ft3 /s); Qc = channel discharge capacity (ft3 /s); n = Manning’s roughness coefficient (dimensionless); A = cross-sectional flow area (ft2 ); R = hydraulic radius (ft); and S = channel slope (percent).

Manning’s roughness coefficient is an empirical value that depends on characteristics of the channel materials (FAO, 1998; CDOT, 2004). Values for n can be selected from tables provided in various publications (Cedergren, 1989; FAO, 1998; Brown et al., 2001; CDOT, 2004; and GEO, 2011). Kilgore and Cotton (2005) found that geotextile-lined channels have n values ranging from 0.030 to 0.040. A Manning’s n of 0.035 is likewise recommended for lined surface water drainage channels on landslides. It should be noted, however, that n will depend on the type of geotextile and channel lining (e.g., rip rap, gravel, etc.). The crosssectional flow area, A, and hydraulic radius, R, depend on the peak flow requirements and are geometrically related. The hydraulic radius is defined as the ratio of the cross-sectional flow area to the wetted channel perimeter (i.e., the distance along channel bottom from water line to water line), both measured perpendicular to the flow direction.

Drainage channel flow area, A, and hydraulic radius, R, can be optimized while constraining discharges and flow velocities to acceptable levels (see Step 6). A minimum freeboard of 0.5 ft (0.2 m) is often used in transportation ditch design (Brown et al., 2001) and is also recommended for landslide drainage. For symmetric triangular and trapezoidal channels with given freeboard, f, channel width, w, and depth, d, the following equations for A and R were developed using the geometric properties shown in Figure 9: Cross-sectional flow area for symmetric triangular channels:

1 wf ; Atri = (d − f ) w − 2 d

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275–289

(9)

283

Haugen

Hydraulic radius for symmetric triangular channels:

wf 2d 1 Rtri = wâ&#x2C6;&#x2019; sin tanâ&#x2C6;&#x2019;1 ; (10) 4 d w Cross-sectional flow area for symmetric trapezoidal channels: 1 Atrap = (11) (d â&#x2C6;&#x2019; f )2 (w â&#x2C6;&#x2019; b) ; 2d Hydraulic radius for symmetric trapezoidal channels:

2d

2 1 (d â&#x2C6;&#x2019; f ) (w + b) sin tanâ&#x2C6;&#x2019;1 wâ&#x2C6;&#x2019;b 2d

, Rtrap = 4d d â&#x2C6;&#x2019; f + 1/8b sin tanâ&#x2C6;&#x2019;1 wâ&#x2C6;&#x2019;b (12) where A = cross-sectional flow area (ft2 ); R = hydraulic radius (ft); d = channel depth (ft); f = freeboard (ft); w = channel width (ft); and b = channel bottom width (ft).

Eq. 8 can be combined with equations for A and R (Eqs. 9 through 12) to form â&#x20AC;&#x153;unifiedâ&#x20AC;? peak discharge capacity equations for each channel type that depend only on values of d, b, and w for a given freeboard, f, and Manningâ&#x20AC;&#x2122;s roughness, n (Eqs. 13 and 14): Unified peak discharge capacity in symmetric triangular channels: 0.295 Qc,tri =

n

2/3 1/2 w f 5/3 â&#x2C6;&#x2019;1 S ; sin tan (2d/w) Ă&#x2014; (d â&#x2C6;&#x2019; f ) wâ&#x2C6;&#x2019; d (13)

Unified peak discharge capacity in symmetric trapezoidal channels:

Qc,trap

1H:1V side slope channel depth for symmetric triangular channels: 1 (15) d = w; 2 2H:1V side slope channel depth for symmetric triangular channels: 1 (16) d = w; 4 1H:1V side slope channel bottom width for symmetric trapezoidal channels: b = w â&#x2C6;&#x2019; 2d;

(17)

2H:1V side slope channel bottom width for symmetric trapezoidal channels: b = w â&#x2C6;&#x2019; 4d.

(18)

Using Eqs. 15 through 18, the unified peak discharge capacities calculated using Eqs. 13 and 14 can be used to produce four simplified equations for calculating peak discharge capacities in landslide drainage channels (Eqs. 19 through 22):

â&#x17D;§ â&#x2C6;&#x2019;1 2d 2/3 â&#x17D;Ť 10/3 2/3 â&#x2C6;&#x2019;5/3 â&#x17D;Ź â&#x17D;¨ (d â&#x2C6;&#x2019; f ) (w + b) (w â&#x2C6;&#x2019; b) sin tan 0.295 d wâ&#x2C6;&#x2019;b S1/2 , =

2/3 2d â&#x17D;­ n â&#x17D;Š d â&#x2C6;&#x2019; f + 1/8b sin tanâ&#x2C6;&#x2019;1 wâ&#x2C6;&#x2019;b

where Qc = peak channel discharge capacity (ft3 /s); n = Manningâ&#x20AC;&#x2122;s roughness coefficient (dimensionless); S = channel slope (percent); d = channel depth (ft); f = freeboard (ft); w = channel width (ft); and b = channel bottom width (ft).

The depth and width, d and w, of symmetric channels are related by side slope angle. For triangular channels, d depends only on w. Given a channel width, the required channel depth can be calculated for 1H:1V and 2H:1V side slopes using Eqs. 15 and 16, respectively. For symmetric trapezoidal channels, the channel bottom width, b, depends on both channel

284

width and depth. The channel bottom width for 1H:1V and 2H:1V side slopes can be calculated using Eqs. 17 and 18, respectively. It should be noted that at a certain maximum depth, channel bottom width goes to zero and a symmetric trapezoidal channel becomes a symmetric triangular channel. That depth can be determined using Eq. 15 or Eq. 16 and is the upper bound on trapezoidal channel depths. Thus, for a given channel width and side slope angle, a trapezoidal channel will always be shallower than a triangular channel and will have a smaller cross-sectional flow area and flow capacity.

(14)

Standard peak discharge capacity in symmetric triangular 1H:1V side slope channels: 0.117 (19) (w â&#x2C6;&#x2019; 2 f )8/3 S1/2 ; n Standard peak discharge capacity in symmetric triangular 2H:1V side slope channels: Qc,tri,1H:1V =

0.043 (20) (w â&#x2C6;&#x2019; 4 f )8/3 S1/2 ; n Standard peak discharge capacity in symmetric trapezoidal 1H:1V side slope channels: Qc,tri,2H:1V =

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275â&#x20AC;&#x201C;289

Landslide Surface Water Drainage Control

Qc,trap,1H:1V 2/3

0.743 d â&#x2C6;&#x2019;1 (d â&#x2C6;&#x2019; f )5/3 (w â&#x2C6;&#x2019; d) = S1/2 ; n d â&#x2C6;&#x2019; f + 0.088 (w â&#x2C6;&#x2019; 2d)

(21)

Standard peak discharge capacity in symmetric trapezoidal 2H:1V side slope channels: Qc,trap,2H:1V

2/3

5/3 1 wâ&#x2C6;&#x2019;d 1.093 d â&#x2C6;&#x2019;1 (d â&#x2C6;&#x2019; f ) 2 = S1/2 , n d â&#x2C6;&#x2019; f + 0.056 (w â&#x2C6;&#x2019; 4d)

(22)

where Qc = channel discharge capacity (ft3 /s); n = Manningâ&#x20AC;&#x2122;s roughness coefficient (dimensionless); S = channel slope (percent); f = freeboard (ft); and w = channel width (ft).

The calculated channel discharge capacity, Qc , must be checked against estimated peak runoff flow, Qr , for each channel in the drainage network. Qr must be less than or equal to Qc . Minimum channel dimensions are determined by setting Qr equal to Qc . Once channel dimensions have been determined, design flows are calculated for each channel and appropriately summed to estimate total peak discharge from the drainage control system.

in the channel (FAO, 1998) but can be estimated using maximum allowable shear stress on geotextile linings (Kilgore and Cotton, 2005). Based on example calculations in Kilgore and Cotton (2005), a maximum channel velocity of 15 ft/s (4.6 m/s) is recommended.

Although some channel erosion or system damage may still occur at this velocity, resulting repairs are considered acceptable given that the design life of earthen channels is often less than about 30 years. Mean channel velocity can be calculated using Eq. 23: Mean flow velocity:

(23) Vmean = Qc A, where Qc = channel discharge capacity and A = crosssectional flow area.

If calculated channel velocities or total peak flows do not meet the above criteria, the slope and dimensions of drainage channels should be adjusted iteratively by repeating Steps 4 and 5 until a final design that satisfies criteria (1) and (2) above is achieved. If these criteria are not achievable with slope or dimension adjustments, Step 3 can be repeated to change the number and spatial distribution of channels and to reduce total flows and flow rates in the channels.

Step 6â&#x20AC;&#x201D;Refine and Finalize Design

CONCLUSIONS

The preliminary drainage control design prepared in Steps 3 through 5 must be optimized before drafting a final design. The number, spatial orientation, slope, and dimensions of surface water channels should be adjusted to achieve an optimal design. At a minimum, compliance with the following criteria must be verified:

Surface water drainage control can be used to reduce landslide susceptibility, slow landslide movement, and prevent landslide activation. In this work, a coherent â&#x20AC;&#x153;start-to-finishâ&#x20AC;? approach to landslide surface water drainage control design was presented. The approach presented is based on widely accepted standards of engineering and geoscience practice. It considers the peculiarities of landslide problems, including the existence of infiltration short paths and ongoing surface deformation. While no so such coherent approach is known to have been published, it is important to note that many practitioners have developed their own alternatives, and those alternatives are by no means invalidated or diminished by the work presented herein. Professional judgment should always be exercised before selecting a method for designing surface water drainage control for landslides, and the most appropriate method may, in fact, include components of multiple methods.

(1) The capacity of drainage systems immediately downstream must be greater than or equal to total peak capacity of the landslide drainage system; and (2) Channel velocities must be high enough to prevent siltation, but low enough to prevent system damage.

If total peak discharge exceeds the design capacity of downstream systems, drainage must be rerouted or stored in holding ponds to reduce downstream discharge. If necessary, plans and specifications of rerouting and holding structures should be incorporated into the final design. Design details for these structures are beyond the scope of this article, but many design resources exist for such structures (FAO, 1998; Brown et al., 2001; CDOT, 2004; and GEO, 2011).

Estimated channel velocities must also fall within a certain range. To permit sediment transport and prevent siltation, a minimum channel velocity of 3 ft/s (0.9 m/s) is recommended (FAO, 1998). Maximum permissible channel velocities depend on the materials used

EXAMPLE DESIGN The following is an example design scenario provided to demonstrate the landslide surface water drainage control design approach described herein. The scenario is entirely fictional and is provided only as a guide to the application of this approach to design.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275â&#x20AC;&#x201C;289

285

Haugen

mensions are as follows (where d indicates depth, w indicates width, and l indicates length): Sag pond: d = 5 ft w = 20 ft l = 50 ft Tension crack 1: d = 10 ft w = 1 ft l = 20 ft Tension crack 2: d = 10 ft w = 1 ft l = 20 ft Drainage path length, D, mean drainage channel slope, S, and drainage area, A, have been determined for each of the four drainage areas contributing to channel flow: Area 1: Area 2: Area 3: Area 4: Figure 10. Schematic of the example design scenario. Note the location of the tension cracks, sag pond, and preliminary design location of the surface water drainage channels.

Scenario A landslide in central Colorado has been identified as a candidate for surface water drainage control. Subsurface and slope stability investigations have revealed that it is safe for personnel to work within the landslide area, but important infrastructure remains threatened by landslide movement. Topographic data and DEMs have been gathered and a preliminary drainage network of one interceptor drain and one chevron drain with four branches has been designed (Steps 1 through 3, Figure 10). The landslide is in a rural area with grasscovered slopes. The required design life of drainage structures is 50 years. Permissible probability of failure for all structures is 25 percent. Local regulations require lined triangular channels with 2H:1V slopes.

Mean DDF rainfall curves for Colorado are applicable. The maximum excess discharge in roadway drains downstream is 300 ft3 /s (90 m3 /s). Drainage areas 1, 2, 3, and 4 are 200, 25, 50, and 50 acres (80, 10, 20, and 20 ha), respectively. They have maximum flow path lengths of 500, 100, 200, and 300 ft (150, 30, 60, and 90 m), respectively. Mean flow path slopes are all 15 percent. Initial channel slopes are 4 percent grade. Two tension cracks measuring approximately 1 ft wide, 20 ft long, and 10 ft deep and one sag pond measuring 20 ft wide, 50 ft long, and 5 ft deep must be filled. Total fill volumes for short path mitigation, peak channel discharge capacities, and required channel dimension, must be determined.

Step 1 Three landslide drainage short path features and requisite fill volumes must be accounted for. The dimensions of these features have been measured. Their di-

286

A1 = 200 acres A2 = 25 acres A3 = 50 acres A4 = 50 acres

D1 = 500 ft D2 = 100 ft D3 = 200 ft D4 = 300 ft

S = 15 percent S = 15 percent S = 15 percent S = 15 percent

Step 2 Volumes of short path features that need to be filled must be calculated. Volumes of the three short paths were calculated as follows: Sag pond (Eq. 1): Vsp = 23 ␲dwl = 23 (␲)(10)(1)(20) = 10,742 ft3 Tension crack 1 (Eq. 2): Vf 1 = 12 dwl = 12 (10)(1)(20) = 100 ft3 Tension crack 2 (Eq. 2): Vf 2 = 12 dwl = 12 (10)(1)(20) = 100 ft3 Following volume calculations, mass balance changes due to addition of fill material were determined via slope stability analyses to not reduce the overall slope stability.

Step 3 A preliminary drainage network design has been completed using data from Steps 1 and 2 (Figure 10).

Right-of-way is not a concern, the dimensions of the drains are unrestricted, and their drainage areas are known. The areas downslope of the lower chevron drains shown in Figure 10 are drained by an existing roadside ditch and thus do not need to be accounted for in the design. The drainage areas of the channel segments are as follows: Interceptor drain: Drains area 1, A1 = 200 acres Chevron segment 1: Drains area 2, A2 = 25 acres Chevron segment 2: Drains area 3, A3 = 50 acres Chevron segment 3: Drains area 4, A4 = 50 acres Trunk drain: Drains areas 2 through 4, A5 = 125 acres

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275–289

Landslide Surface Water Drainage Control

Step 4 Peak runoff flow, Qr , is calculated using data gathered in Steps 1 and 3 (Eq. 3). The maximum value of the runoff coefficient, C, for the environment is selected from Table 1. Design life, n, and acceptable probability of failure, P, are given in the scenario and used to calculate the event return period, T (Eq. 5), for use in estimating peak rainfall depth, dr , via data available on the NWS Web site. Time of concentration, tc , is calculated using drainage path lengths and slopes from Step 1 (Eq. 6). Peak rainfall intensity, i, is then calculated using dr and tc values (Eq. 7). Finally, peak runoff flows, Qr , are calculated for each channel using Eq. 3 and are summed to estimate total peak runoff flow for comparison with downstream systems.

Peak rainfall intensity (Eq. 7): Peak depths from NWS DDF curves: dr 1 = 2.44 in. dr 3 = 2.10 in. dr 2 = 2.10 in. dr 4 = 2.44 in.

Therefore:

Runoff coefficient: Unimproved areas (Table 1) C = 0.30 Design life: Given in scenario n = 50 years Probability of failure: Given in scenario P = 25 percent Design return period (Eq. 5): T= T=

1 1 − (1 − P)1/n 1

1 − (1 − 0.25)1/50

= 174 years

i = 60 dtcr

i 1 = 60

dr 1 2.44 = 2.42 in./hr = 60 tc1 61

i 2 = 60

dr 2 2.10 = 4.65 in./hr = 60 tc2 27

i 4 = 60

dr 3 2.10 = 3.29 in./hr = 60 tc3 38

i 5 = 60

dr 4 2.44 = 3.12 in./hr = 60 tc4 47

Peak runoff flows (Eq. 3):

Qr = Ci A

Qr 1 =Ci 1 A1 =(0.30)(2.42 in./hr)(200 acres)=145 ft 3/5

Time of concentration (Eq. 6): Qr 2 =Ci 2 A2 =(0.30)(4.65 in./hr)(25 acres)=34.9 ft 3/5 tc = tc1 = =

1.8(1.1 − C)D S1/3

1/2

Qr 3 =Ci 3 A3 =(0.30)(3.29 in./hr)(50 acres)=49.3 ft 3/5

1.8 (1.1 − C) D1 S1/3 1.8 (1.1 − 0.30) 5001/2 1/2

0.151/3

Qr 4 =Ci 4 A4 =(0.30)(3.12 in./hr)(50 acres)=46.8 ft 3/5 = 61 minutes

1.8 (1.1 − C) D2 1/2 tc2 = S1/3 1.8 (1.1 − 0.30) 1001/2 = = 27 minutes 0.151/3 tc3 = = tc4 = =

1.8 (1.1 − C) D3 1/2 S1/3 1.8 (1.1 − 0.30) 2001/2 0.151/3 1.8 (1.1 − C) D4 1/2 S1/3 1.8 (1.1 − 0.30) 3001/2 0.151/3

Qr 5 =Qr 2 +Qr 3 +Qr 4 =34.9 + 49.3 + 46.8 = 131 ft 3/5 Qr,total = Qr 1 + Qr 5 = 276 ft 3/5 Step 5

= 38 minutes

= 47 minutes

Local regulations require lined triangular channels with 2H:1V side slopes. If they did not, at this point the designer would pick a channel type that is most applicable, considering available equipment, right of way, etc. If that channel type is found to be less than sufficient, alternative channel designs would be tested.

Flow capacities will thus be calculated using Eq. 20 (for symmetric, 2H:1V triangular channels). Because peak runoff flows for each channel are known (Step 4), if

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275–289

287

Haugen

channel slope, S, freeboard, f, and Manning’s roughness, n, are known, peak channel discharge, Qc , can be set equal to Qr to find minimum channel width. In keeping with local practice, widths will be rounded up to the nearest half-foot. Channel slopes are given in the scenario. The recommended freeboard and Manning’s roughness values included herein can be used (note that f and S can sometimes be adjusted to account for channel width or depth restrictions). The minimum channel width is calculated by manipulating Eq. 20 using the Peak Channel flow, or, calculated in step 4. The channel depth is calculated using Eq. 16. The final design flow capacity, Qc is then re-calculated to account for conservative channel dimensions.

Manning’s roughness: Recommended value, n = 0.035 Freeboard: Recommended value, f = 0.5 ft Channel slopes: Given in scenario, S = 4 percent Channel width (Eq. 20): (w − 4 f )8/3 S1/2 Qr = Qc = 0.043 n Therefore: w = (Qr

n )3/8 + 4 f 0.043 S1/2 3/8

n +4f w1 = Qr 1 0.043 S1/2 3/8

0.035 = 145 + 4 (0.5) = 13.0 ft 0.043 (0.04)1/2

3/8 n +4f w2 = Qr 2 0.043 S1/2 3/8

0.035 + 4 (0.5) = 8.5 ft = 34.9 0.043 (0.04)1/2

3/8 n +4f w3 = Qr 3 0.043 S1/2 3/8

0.035 = 49.3 + 4 (0.5) = 9.5 ft 0.043 (0.04)1/2

3/8 n +4f w4 = Qr 4 0.043 S1/2 3/8

0.035 = 46.8 + 4 (0.5) = 9.5 ft 0.043 (0.04)1/2

3/8 n +4f w5 = Qr 5 0.043 S1/2 3/8

0.035 = 131 + 4 (0.5) = 13.0 ft 0.043 (0.04)1/2

288

Channel depth (Eq. 16):

d1 =

d = 14 w

1 1 w = (13.0) = 3.25 ft 4 4

d2 =

1 1 w = (8.5) = 2.13 ft 4 4

d3 =

1 1 w = (9.5) = 2.38 ft 4 4

d4 =

1 1 w = (9.5) = 2.38 ft 4 4

1 1 w = (13.0) = 3.25 ft 4 4 Design flow capacities (Eq. 20): d5 =

Qc =

0.043 (w − 4 f )8/3 S1/2 n

Qc1 =

0.043 (13.0 − 4(0.5))8/3 S(0.04)1/2 = 147 ft 3/5 n

Qc2 =

0.043 (8.5 − 4(0.5))8/3 S(0.04)1/2 = 8.5 ft 3/5 n

Qc3 =

0.043 (9.5 − 4(0.5))8/3 S(0.04)1/2 = 53 ft 3/5 n

Qc4 =

0.043 (9.5 − 4(0.5))8/3 S(0.04)1/2 = 53 ft 3/5 n

Qc5 =

0.043 (13.0 − 4(0.5))8/3 S(0.04)1/2 = 147 ft 3/5 n

Qc, total = Qc1 + Qc5 = 294 ft 3/5 Step 6 The design developed in Steps 1 through 5 must be checked against certain minimum criteria before being finalized. First, total design flow capacity must be less than the excess capacity downstream. In this case, excess capacity is Qe = 300 ft 3/5. Second, the mean velocity of channel waters at full capacity must be between 3 and 15 ft/s. Note that the area used in the velocity check is the cross-sectional flow area, not the drainage area, Ax , listed in Step 3. If the design meets these criteria, the design is finished. If not, the designer must return to Step 1 and try other alternatives. These checks are as follows:

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275–289

Landslide Surface Water Drainage Control

Check excess flow capacity: Qctotal ≤ Qe

?

294 cfs ≤ 300 ft 3/5. This criterion is met.

Check velocities (Eq. 23): 3 fts < Vmean1 < 15 fts 147 cfs ft Qc1 ft < Vmean1 = = = 9.7 < 15 s A 15.1 ft2 s

Yes

3

36 cfs Qc1 ft ft < Vmean1 = = = 6.8 < 15 2 s A 5.3 ft s

Yes

3

53 cfs Qc1 ft ft < Vmean1 = = = 7.5 < 15 2 s A 7.0 ft s

Yes

53 cfs Qc1 ft ft < Vmean1 = = = 7. < 15 2 s A 7.0 ft s

Yes

3

3

147 cfs Qc1 ft ft < Vmean1 = = = 9.7 < 15 2 s A 15.1 ft s This criterion is met.

Therefore, the design is finished.

3

Yes

ACKNOWLEDGMENTS The author is honored to have received the Association of Environmental and Engineering Geologists’ (AEG’s) 2015 Outstanding Student Paper Award for this manuscript. Without the encouragement and shared knowledge of many AEG members, this article would not have been possible. Deepest thanks to Professor Paul Santi at the Colorado School of Mines for his diligent and thoughtful comments on early versions of this manuscript and to the author’s graduate advisor, Associate Professor Wendy Zhou, for her support and encouragement.

REFERENCES ANDERSON, M.; HOLCOMBE, L.; FLORY, R.; AND RENAUD, J. P., 2008, Implementing low-cost landslide risk reduction: A pilot study in unplanned housing areas of the Caribbean: Natural Hazards, Vol. 47, pp. 297–315.

BROWN, S. A.; STEIN, S. M.; AND WARNER, J. C., 2001, Urban Drainage Design Manual: U.S. Department of Transportation Federal Highway Administration Hydraulic Engineering Circular No. 22, 2nd ed., 478 p.

CEDERGREN, H. R., 1989, Seepage, Drainage, and Flow Nets, 3rd ed.: John Wiley and Sons, New York. 496 p.

CHATWIN, S. C.; HOWES, D. E.; SCHWAB, J. W.; AND SWANSTON, D. N., 1994, A Guide for Management of Landslide-Prone Terrain in the Pacific Northwest, 2nd ed.: British Columbia Ministry of Forests Land Management Handbook No. 18, 220 p.

COLORADO DEPARTMENT OF TRANSPORTATION (CDOT), 2004, Drainage Design Manual: Colorado Department of Transportation, Denver, CO, 562 p.

CORNFORTH, D. H., 2005, Landslides in Practice: Investigation, Analysis, and Remedial/Preventative Options in Soils: John Wiley and Sons, Hoboken, NJ. 624 p.

FOOD AND AGRICULTURE ORGANIZATION OF THE UNITED NATIONS (FAO), 1988, Watershed Management Field Manual: Landslide Prevention Measures: Food and Agricultural Organization of the United Nations Conservation Guide 13/4, 156 p.

FAO, 1998, Watershed Management Field Manual: Road Design and Construction in Sensitive Watershed: Food and Agriculture Organization of the United Nations Conservation Guide 13/5, 196 p.

GEOTECHNICAL ENGINEERING OFFICE (GEO), 2011, Geotechnical Manual for Slopes: Government of Hong Kong Civil Engineering and Development Department Geotechnical Engineering Office, 302 p.

HIGHLAND, L. M. AND BOBROWSKY, P., 2008, The Landslide Handbook—A Guide to Understanding Landslides: U.S. Geological Survey Circular 1325, 129 p.

HOLCOMBE, E.; SMITH, S.; WRIGHT, E.; AND ANDERSON, M. G., 2012, An integrated approach for evaluating the effectiveness of landslide risk reduction in unplanned communities in the Caribbean: Natural Hazards, Vol. 31, pp. 351–385.

KILGORE, R. T. AND COTTON, G. K., 2005, Design of Roadside Channels with Flexible Linings: U.S. Department of Transportation Federal Highway Administration Hydraulic Engineering Circular No. 15, 3rd ed., 153 p.

KING COUNTY, 2009, Surface Water Design Manual: Department of Natural Resources and Parks, King County, WA. LU, N. AND GODT, J. W., 2013, Hillslope Hydrology and Stability: Cambridge University Press, Cambridge, U.K. 458 p.

MIKOS, M.; FAZARINC, R.; PULKO, B.; PETKOVSEK, A.; AND MAJES, B., 2005, Stepwise mitigation of the Macesnik landslide, N Slovenia: Natural Hazards Earth System Science, Vol. 5, pp.

947–958.

MIZAL-AZZMI, N.; MOHD-NOOR, N.; AND JAMALUDIN, N., 2011, Geotechnical approaches for slope stabilization in residential areas: The 2nd International Building Control Conference: Procedia Engineering, Vol. 20, pp. 474–482.

ROGERS, J. D., 1992, Recent developments in landslide mitigation techniques. In Slosson, J. E.; Keene, A. G.; and Johnson, J. A. (Editors), Landslides/Landslide Mitigation: Geological Society of America Reviews in Engineering Geology, Vol. 9, Boulder, CO, pp. 95–118.

SAJINKUMAR, K. S.; ANBAZHAGAN, S.; RANI, V. R.; AND MURALEEDHARAN, C., 2014, A paradigm quantitative approach for a regional risk assessment and management in a few landslide prone hamlets along the windward slope of Western Ghats, India: International Journal Disaster Risk Reduction, Vol. 7, pp. 142–153.

SCHUSTER, R. L., 1978, Landslides Analysis and Control: Transportation Research Board of the National Academies, Washington, DC. TURNER, A. K. AND SCHUSTER, R. L. (Editors), 1996, Landslides: Investigation and Mitigation: The National Academies of Science, Engineering and Medicine Transportation Research Board Special Report 247, 230 p.

YILMAZER, I.; YILMAZER, O.; AND SARAC, C., 2003, Case history of controlling a major landslide in Karandu, Turkey: Engineering Geology, Vol. 70, pp. 47–53.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 275–289

289

Consideration of the Validity of Debris-Flow Bulking Factors HOLLY BRUNKAL1 Department of Natural Sciences, Western State Colorado University, 600 North Adams Street, Gunnison, CO 81231

PAUL SANTI Department of Geology and Geological Engineering, Colorado School of Mines, 1516 Illinois Street, Golden, CO 80401

Key Terms: Debris Flow, Peak Discharge, Bulking Factor, Magnitude, Estimation, Post-Wildfire ABSTRACT Compilation of a database of debris-flow peak discharges (Q) allowed for a comparison with the expected basin discharge, as computed using the rational equation, Q = CIA. The observed values of Q for debris flows in unburned and burned areas were divided by the computed Q values of runoff using the rational method. This ratio is the bulking factor for that debris-flow event. Unburned and burned basins constitute two distinct populations; analysis shows that the bulking factors for burned areas are consistently higher than those for unburned basins.

Previously published bulking factors for unburned areas fit the data set in about 50 percent of the observed cases in our compiled data set. Bulking factors for burned areas that were found in the published literature were well below the observed increases in peak discharge in over 50 percent of the cases investigated. If used for design purposes, these bulking factors would result in a significant underestimation of the peak discharge from a burned basin for the given rainfall intensity. Peak discharge bulking rates were found to be inversely related to basin area.

INTRODUCTION Debris flows are dangerous and fast-moving natural hazards that can severely affect infrastructure that is not designed to accommodate the increased flow volumes, velocities, and peak discharges associated with these events. Peak discharge of a debris flow can be many times that of a water flood (VanDine, 1985; Ikeya, 1989; Shuirman and Slosson, 1992; Hungr, 2000; and Wilford et al., 2004). Estimation of peak discharge is of vital importance in a practical debris-flow analysis, as 1 Corresponding

author email: hbrunkal@western.edu.

it determines the maximum velocity, flow depth, momentum, impact forces, and ability to overrun channel walls and barriers, as well as factoring into run-out distances and potential inundation area (VanDine, 1996; Hungr, 2000).

Debris-flow risk factors are increasing with a documented rise of communities located at the wildland urban interface (WUI). Homes are commonly built on alluvial fans, at the mouths of the canyons, and close to the outlets of both ephemeral and perennial basins, where fire, flooding, and debris-flow hazards are amplified (Theobald and Romme, 2007). As communities expand into these marginal areas it follows that design considerations for culverts, bridges, roads, and surface drainage, as well as mitigation design for predicted floods and debris flows, must have some estimation of the peak discharge expected at the outlet of the drainage system. Peak discharge from both water floods and debris flows can be dangerous and can cause costly damage to infrastructure where these increased flows are not accommodated.

The estimation of peak discharge coming from a given basin is often calculated using the “rational method,” also known as the “rational equation” (hereinafter used interchangeably). The rational method was developed by sewage engineers, primarily to calculate the expected runoff from urban areas, and it is first mentioned in the American literature in 1889 (Chow, 1962). It has since seen application outside of its original civil engineering usage in areas such as soil conservation and wildland runoff estimation (Chow, 1962).

There are approaches to the peak discharge calculation that begin with the rational method and then multiply the results by a debris-flow bulking factor to better estimate the potential peak discharges. The accuracy and breadth of application of these techniques are widely variable and have not been systematically analyzed. This project was initiated to evaluate debrisflow bulking factors in cases of runoff-generated debris flows in both unburned and post-wildfire settings. The

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 291–298

291

Brunkal and Santi

applicability and accuracy of using a bulking factor will be assessed using a debris-flow database that compares the calculation of the peak discharges using the rational method with the measured or observed debris-flow peak discharges.

THE RATIONAL METHOD The rational equation is an estimation of the peak runoff rate from a watershed or basin based on the area (A), rainfall intensity (I), and a runoff factor (C). Often written as Q = CIA, with the variables measured in U.S. customary units (Q in ft3 /s, A in acres, and I in in./hr). For use with metric measurements, resulting Q in m3 /s (A in km2 and I in mm/hr), the equation is Q = 0.278 CIA. The rational method is common in hydrology textbooks and civil engineering manuals and is often applied in the design of road culverts and other structures emplaced for draining small areas of a few square kilometers (Brutsaert, 2005). It is applied in the design of civil engineering structures such as bridges and culverts, often with a 25-year lifespan or less (Izzard, 1953). The rational method is the most widely used uncalibrated method by which to calculate peak discharge of a defined watershed (Fern´andez et al., 2003).

There is common agreement among practitioners that the rational equation should only be applied to basins of limited area because of the difficulty associated with estimating the changes in infiltration and rainfall intensities across larger areas. Dooge (1957) states that the rational equation should only be used for basins measuring 15 km2 or less in area. Brutsaert (2005) says the rational equation should be used for small areas of a few square kilometers at most, and Fetter (2001) states that the rational method has its greatest value when applied to small drainage basins of 100 ha (1 km2 ) or less. The U.S. Department of Transportation Federal Highway Administration (2008) states that the rational method can be applied for basins measuring less than 200 acres (0.8 km2 ).

The rational method uses a runoff coefficient, C, to reflect the runoff potential of a watershed. C is a dimensionless empirical coefficient related to the infiltration, storage, evapotranspiration, and interception properties of the basin (Fetter, 2001). The C value will be a decimal value between 0 and 1.0, where a value of 0 indicates that none of the rain falling in the basin generates runoff and a value of 1.0 indicates that all of the rain falling in the basin becomes runoff. A few of the assumptions associated with the rational method, specifically the C value, are that the runoff coefficient does not vary with storm intensity or antecedent moisture, runoff is dominated by overland flow, and that the basin storage effects are negligible (Chow, 1962).

292

The rainfall intensity value, I, is often the most difficult variable to measure accurately or to estimate.

Precipitation records are not consistent, gauge types differ, precision of observation varies, distances from gauges to site of initiation varies, adjacent basins sometimes exist in different micro-climatic zones, geologic and topographic environments vary, and standard rain gauges cannot record the variability of antecedent moisture conditions in the basin (Caine, 1980).

In using the rational equation to calculate the peak discharge from a basin, rainfall intensity for the design storm is needed and is often estimated from a very short rainfall record from regional sources (Jakob and Jordan, 2001). Depending on economic justification, typical values of rainfall intensity are chosen based on the life of the infrastructure in question; 5year–recurrence events for sewers in residential areas, 20-year–recurrence storms for commercial high-value districts, and 50-year–recurrence events or more for flood-protection works (Dooge, 1957). The first approximation of Q is done by rainfall frequencyduration analysis, regional analysis of stream flow data, or by extrapolation of short records on gauged streams (Jakob and Jordan, 2001).

The rational equation is used mainly for the routine design of low-cost structures and is used because it is simple and because large amounts of input data and/or other resources are generally deemed unwarranted. It is by no means the only method available for computing peak discharge, although it is the simplest and least time consuming for agencies to apply.

DEBRIS-FLOW PEAK DISCHARGE High peak discharges are one of the most distinctive characteristics of debris flows, separating them from even the elevated peak discharges of hyperconcentrated flows and water floods. The high peak discharge and increased sediment loads are the chief causes of the particular destructiveness of debris-flow surges (Hungr, 2000). The peak discharges of debris flows can be as high as 40 times greater than those of even an extreme flood (VanDine, 1985). It is important to include some estimation of peak debris-flow discharge when designing infrastructure, especially at the expanding WUI. This requires some indication of the frequency of these peak discharge events, which has been explored by various research methods (van Steijn, 1996; Bovis and Jakob, 1999; Jakob et al., 2005; and Riley et al., 2013).

Jakob and Jordan (2001) assert that flood discharge estimated by traditional means has proven inadequate for design and risk management on alluvial fans that experience debris flows. Small streams (those that occupy small watershed areas, ∼1 km2 ) tend to be more at risk from small, frequently occurring events such as

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 291–298

Validity of Debris-Flow Bulking Factors

debris flows (Jakob and Jordan, 2001). Hungr et al.

(1984) suggest that the design event for a debris torrent (a channel that historically produces debris flows) should be the largest and most mobile debris flow that could reasonably occur during the life of the structure under consideration.

The peak discharge of a debris flow is generally associated with the front of the flow, which usually has the greatest depth, highest viscosity, and carries the largest particles (VanDine, 1996). Coe et al. (2008) describe an observed debris flow from the Chalk Cliffs in Colorado, where each surge had a steep front composed of large diameter (5–53-cm) rocks with no visible fluid at the surface. These coarse-grained, fluidpoor fronts had lengths that ranged from 4 to 8 m, velocities from 0.9 to 1.3 m/s, and peak flow depths of 0.2 to 0.35 m, with water-rich tails composed of finegrained material following the coarse-grained surge fronts.

BULKING FACTORS FOR DEBRIS FLOWS The published literature on debris flows and their countermeasures include several instances in which a debris-flow bulking factor is used to estimate the peak discharges expected in a debris-flow event. Most of these bulking factors are empirically based backcalculations for a region that has experienced and recorded a number of debris-flow events. This requires that previous debris flows were documented and that field work was done to measure the peak discharge. In other cases, workers may have to use published bulking factors from manuals to estimate debris-flow peak discharge.

Debris-flow bulking factors, presented by Costa (1984), range from 1.38 for flows with 50 percent solids by weight to 4.40 for flows with 90 percent solids by weight, with the “average” debris flow falling between 1.5 and 2.0. Using the Melton ratio, in which watershed relief is divided by the square root of the watershed area, Wilford et al. (2004) concluded that debris flows can have 5 to 40 times the discharge of floods.

For the consideration of debris-flow counter measures in Japan, the peak sediment discharge in field surveys was found to relate to peak water discharge with the multiplier of 4.6 (Ikeya, 1989).

Shuirman and Slosson (1992) calculate a ratio of 3.2 when comparing discharge from a burned basin to that of an unburned basin using a 25-year return frequency storm and the rational equation. The criteria put forth

by Caltrans and used by Los Angeles County allows for designers to add a bulking factor to peak discharge calculations, but only up to a factor of 2 (Caltrans, 2014). Matsumoto (2007) suggests a bulked discharge

for a burned area that would follow the equation Qbulked = (Cba )(I50 yr )A(B f ), where Cba is the burned area runoff coefficient value and Bf is the bulking factor, selected from the 2006 Los Angeles Country Sedimentation Manual and the 2006 Hydrology Manual (LA County Department of Public Works, 2006a, 2006b). This site-specific discharge calculation was found in a post-wildfire debris-flow hazard assessment for Catalina Island, California (Matsumoto, 2007), and for this basin scenario a burncorrected value of Cba = of 0.89, the equivalent of 89 percent runoff, was used in the calculation of peak discharge.

The considerations of a debris-flow bulking factor presented above show that there are a wide range of strategies regarding how to derive a predicted peak discharge for a debris flow and the resulting multipliers that are applied; some are general for a region, and some are specific to a particular basin of interest. It is of note that each strategy relies on the basic components of the rational method to derive the subsequent equations and bulking factors.

The purpose of this study is to assemble a database to compare the peak discharge values for basins that have had runoff-initiated debris flows (in both burned and unburned areas) with the Q values calculated using the rational method. We expect to develop more consistent bulking factors than have been previously published, and we further hypothesize that there will be a notable difference in the peak discharges from unburned and burned areas.

DATABASE DEVELOPMENT Observations of peak discharges of debris-flow events from locations around the world were compiled into a catalog (available upon request from the authors). Observations include real-time monitored channels and peak discharges that were back-calculated using field measurements of channel geometry and debris-flow velocity. Extensive literature review garnered 78 individual debris-flow events from mostly small basins in alpine areas and areas burned by wildfire.

The database includes debris flows from mostly small basins that adhere to the size limitations set forth for the use of the rational equation, measuring <1 km2 to 15 km2 . Some slightly larger basins were included in the study to provide a more robust data set and for comparison of bulking factors to basin area. A brief summary of the database is provided in Table 1.

Rainfall intensities were recorded in the data table as peak intensities in millimeters per hour. Most reports recorded a maximum intensity over an interval shorter

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 291–298

293

Brunkal and Santi Table 1. A summary of the data set used in this study. Full data set is available upon request from the authors.

Data Set Summary No. of debris flows non–fire related

post-wildfire Size of basins under 1 km2 1–5 km2 over 5 km2 Range of triggering rainfall intensities non-fire debris flows

53 non-fire 2

post-fire 30

2 1

18 5

6–60 mm/hr

post-wildfire Range of peak discharges non-fire debris flows

post-wildfire

Notes Events had to be left out of the data set because of lack of clear correlation to rainfall or lack of peak discharge values; 73 out of 97 debris flows originally found could be used in this study.

20

2–58 mm/hr 0.15–125 m3 /s

Two events report the recurrence interval (RI) of the triggering storm; 6 mm/hr = 5-year RI, 60 mm/hr = 100-year RI. Twenty-five debris flows were triggered by the 3 mm/hr, 2-year RI storm in burned areas in Montana.

Only three authors report antecedent rainfall amounts.

Unreported attributes of peak discharge include timing with respect to rainfall and location along channel length from which measurement was taken.

2–226 m3 /s

than 1 hour or gave the rainfall for the 24-hour period, or some other variation. For those areas that reported a shorter duration peak intensity rainfall, that is the value used in the rational method calculation. For example, if the rainfall peak intensity was reported at 9.1 mm for 10 minutes, in the rational equation calculation the value of 9.1 mm/hr was used to provide a conservative, or maximum, runoff calculation for the rainfall event. This simplification is supported by other technical studies: Moody and Martin (2001) found that in mountainous terrain that 79 percent of the 1-hour rainfall occurs in 30 minutes, and often storms have only a 10-minute intense burst of rainfall. Because of the difficulties associated with finding accurate rainfall data, this approach was determined to be the most straightforward with the least amount of extrapolation.

One attribute of the rainfall data that was not available, but that would have been helpful for analysis, is the recurrence interval of the storms that initiated the debris flow.

There are many ways to choose a C (runoff coefficient) value for use in the rational equation. For this database a conservative, or maximum, runoff coefficient was chosen based on values found in the published literature. In practice, a C value may be chosen from an available table or calculated as an average value of many C’s representing different percentages of the basin area. For this comparison, the C value of 0.8 was selected for alpine areas with a percentage of steep talus slopes and smaller vegetation coverage (as an example, this value is used in unburned alpine areas in the Swiss Alps by Berti et al. [1999]). The C value of 0.89 was used for burned areas, based on Matsumoto

294

Non–fire related debris flows are represented by multiple events from five basins.

Fire-related debris flows are represented by 53 unique basins.

(2007), as a maximum runoff coefficient because of the changed character of a burned basin, including loss of canopy, development of hydrophobic soils, and loss of infiltration capacity. Although there is a high likelihood that the appropriate C value for each basin lies slightly below these chosen values, this conservative approach seemed more realistic than assigning different C values for each basin without specific knowledge of each location.

Some of the difficulties in compiling the data set stem from the fact that most reports of debris flows do not include the peak discharge value, and many reports that have peak discharge values only include vague data on the intensity of rainfall preceding the debris flow. For example, in an instrumented channel in the Italian Alps with recorded debris-flow events, there were 10 peak discharges recorded that were not correlated to the rainfall intensity preceding them, so they could not be used as part of the data set (Marchi et al., 2002).

DATABASE ANALYSIS AND RESULTS Measured Q values for debris flows were plotted in a log-log space against the Q values calculated by the rational equation (Qcalc ) (Figure 1). Discrete populations of event differences were not immediately apparent.

The plotted points were next divided into unburned areas and burned areas. Some grouping of basins by burned or unburned status was apparent. A leastsquares regression line was fit for both populations so that the best-fit line for the two groups could be compared. The regression lines in this case are not meant

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 291–298

Validity of Debris-Flow Bulking Factors

Figure 1. Log-log plot of calculated Q value using the rational equation versus the actual measured peak discharges. The least-squares regression fit lines for burned and unburned areas were used to assess if these are separate populations. The two populations were found to be different, with a P value of 0.00616. The overall model has an adjusted R2 value of 0.484.

to be predictive but rather to show that the two groups are indeed different. These two populations of debrisflow types were found to be different with a significance value of P = 0.00616 at the 5 percent level. The adjusted R2 value for the entire model was 0.484.

The second relationship of interest was the bulking factor compared to the basin area. The bulking factor in this case is the value of the measured (or observed) debris-flow peak discharge divided by the calculated (rational equation) value. The bulking factor is plotted against basin area in log-log space and populations of burned and unburned debris flows were compared (Figure 2). In this case the two populations are not significantly different, with a P value of 0.936 and R2 value of 0.568.

A generalized comparison of basin area and bulking factor between burned and unburned areas results in an observable difference in the range of bulking factors for each (Table 2). In basins measuring less than 1 km2 in area, the average bulking factor for unburned basins (n = 14) was 10x, and for burned areas (n = 30) the average was 33x. For this very simple analysis the largest outliers for burned and unburned areas were removed from the averages. These values were 72x for an unburned area measuring greater than 5 km2 and 299x for a burned area measuring less than 1 km2 . Similar differences between unburned and burned basins were shown for larger basins, with bulking factors in burned

areas approximately double those in unburned areas.

Because of the small sample size and the disparity in the numbers of samples in each group this result should only be taken as a generalized range of values that were observed overall.

Figure 3 is a simple graphical representation of the results of the calculated bulking factors for burned and unburned basins in this data set. In unburned basins the bulking factors were equal to or less than 4.0Qcalc 56.6 percent of the time, in burned basins 55.4 percent of the bulking factors were equal to or greater than 10.0Qcalc .

DISCUSSION This research shows that debris flows in unburned basins and post-wildfire debris flows have a significant difference in peak discharge from those calculated using the rational equation for water floods. Peak discharges recorded in post-wildfire settings are typically two to three times higher than those of debris flows in unburned areas. Furthermore, we show that bulking factors suggested in the published literature are often not large enough to represent true debris-flow peak discharges.

The bulking factors proposed by Ikeya (1989) of 4.6x would be appropriate in 50 percent of the cases for debris flows in unburned basins included in this study.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 291â&#x20AC;&#x201C;298

295

Brunkal and Santi

Figure 2. The comparison between calculated bulking factor (Qobs /Qcalc ) and basin area in log-log space. The two populations, unburned and burned basins, show the same relationship: as basin area increases, bulking factor decreases. The least-squares regression fit for the two populations were not significantly different. The model has a P value of 0.936 and R2 value of 0.568.

Shuirman and Slosson (1992) suggest a bulking rate of 3.2 for a burned area experiencing a 25-year return interval storm. In this study, 3.2x would be appropriate for 26 percentof the burned basins that experienced debris flows; with a note that the rainfall return periods in the data set were not necessarily the 25-year return interval.

The bulking factors reviewed in the available literature would underestimate the peak discharges of debris flows from burned areas more than half the time. In this data set only 18 percent of post-wildfire debris flows fell below the 2.0Q bulking factor that is chosen, in most cases, as the maximum multiplier by Caltrans Highway Design Manual (Caltrans, 2014) for conveyance

design in San Bernardino County, Los Angeles County, and Riverside County, areas that have a fire-flood cycle that is well documented. The Design Debris Event outlined by Caltrans (Caltrans, 2014) is associated with the 50-year, 24-hour duration storm, although it has been shown that the intensity and duration of a debrisflow triggering event in a post-wildfire setting can be significantly lower (Cannon et al., 2008).

The bulking factor for unburned and burned basins versus basin area shows an interesting result in that the least-squares regression line has a negative slope, indicating that the bulking factor is reduced with increasing basin area for both populations of debris flows. The observation that the bulking factor is greatest in small

Table 2. Summary of calculated peak discharge bulking factors for burned and unburned basins based on basin size. Smaller basins and burned basins consistently have higher bulking factors for debris flows.

Bulking Factors for Q Unburned

Burned

Basin Area

Min

Max

Average

Min

Max

Average

<1 km2 1â&#x20AC;&#x201C;5 km2 >5 km2

0.9x 0.2x 0.14x

31x 22x 4.4x

10x (n = 14) 6.5x (n = 11) 1.4x (n = 4)

3.5x 0.2x 0.3x

96x 62x 9.1x

33x (n = 30) 13.9x (n = 21) 2.8x (n = 7)

296

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 291â&#x20AC;&#x201C;298

Validity of Debris-Flow Bulking Factors

Figure 3. A comparison of the range of debris-flow bulking factors calculated for unburned and burned areas. Percentages are expressed for each category, unburned basins and burned basins. More than 50 percent of the unburned basins have less than a 4x bulking factor, whereas more than 50 percent of the burned basins had bulking factors of greater than 10x. The disparity may be a factor of the overall sample size, but the two populations were shown to be statistically different.

basins is important because it is more likely that the discharge from these basins will be estimated using the rational method. We suggest two hypotheses to explain why the bulking factor decreases with increasing basin area: 1) larger basin areas would have longer channel lengths, with more opportunities for deceleration and deflation, where debris flows slow down and deposit part of their volume as levees along the side of the channel and 2) larger burned basins have a lower likelihood that the whole basin burned at the same severity; there may be larger areas of the basin that are not contributing the same runoff, and therefore a uniform C value would tend to overestimate runoff for larger basin areas.

Use of channel-length bulking data is proposed as the next step in this research. Channel length was an attribute that was not available for all the data in this study; therefore, it was not possible to compare the bulking rates per channel length, as proposed

by Williams and Lowe (1990) and Mulvey and Lowe (1992) for burned areas. Better understanding of bulking along the length of the channel could be instrumental in placement of debris barriers or other mitigation design considerations.

As a result of the small number of data points from a large range of geographical areas, coupled with the lack of other attributes that may explain the bulking of debris-flow Q values (e.g., geology, topography, climate, land use, channel length, and location along the channel where Q values were measured), this study can only be considered a preliminary step in characteriz-

ing bulking rates for debris flows worldwide. Given the limitations of the small size of the data set, the statistical analysis presented should not be considered an applicable regression equation for prediction of bulking factors in future hazard analysis. Rather, this data set should be considered a first step in cataloging debrisflow peak discharges and their relationship to the common calculations of runoff using the rational equation.

CONCLUSIONS This analysis of peak discharges of post-wildfire debris flows provides a database that has not been assembled previously, and statistical analyses suggest that the peak discharges of debris flows in burned and unburned areas are statistically distinct populations when compared to peak discharges calculated using the rational equation. This represents a first step toward a better understanding of the relationship between flood-water flow discharge and debris-flow peak discharge.

The rational equation significantly underestimates peak discharge of debris flows. Many agencies, including departments of transportation and flood control districts, use the rational equation for deriving peak flows for drainages. Current bulking factors found in the published literature and used for design purposes are not large enough to accurately describe debris-flow Q values. The results of this study show that the bulking factors range from 1.4 to 10X for unburned areas and from 2.8 to 33X for burned areas, with the factors decreasing with increasing size of drainage basin. The

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 291â&#x20AC;&#x201C;298

297

Brunkal and Santi

authors caution against using these values for prediction, as the size of the database was limited.

REFERENCES BERTI, M.; GENEVOIS, R.; SIMONI, A.; AND TECCA, P. R., 1999, Field observations of a debris flow event in the Dolomites: Geomorphology, Vol. 29, no. 3–4, pp. 265–274.

BOVIS, M. J. AND JAKOB, M., 1999, The role of debris supply conditions in predicting debris flow activity: Earth Surface Processes Landforms, Vol. 24, No.11, pp. 1039–1054.

BRUTSAERT, W., 2005, Hydrology, An Introduction: Cambridge University Press, Cambridge, U.K. 605 p.

CAINE, N., 1980, The rainfall intensity: Duration control of shallow landslides and debris flows: Geografiska Annaler. Series A. Physical Geography, Vol. 62, No. 1/2, pp. 23–27.

CALIFORNIA DEPARTMENT OF TRANSPORTATION (CALTRANS), 2014, Highway Design Manual, Chapter 810- Hydrology: Electronic document, available at http://www.dot.ca.gov/hq/oppd/ hdm/hdmtoc.htm#hdm CANNON, S. H.; GARTNER, J. E.; WILSON, R.; BOWERS, J. C.; AND LABER, J. L., 2008, Storm rainfall conditions for floods and debris flows for recently burned areas in southwestern Colorado and Southern California: Geomorphology, Vol. 96, pp.

250–269.

CHOW, V. T., 1962, Hydrologic determination of waterway areas for the design of drainage structures in small drainage basins: Engineering Experiment Station Bulletin, Vol. 59, No. 65, Bulletin No. 462. 104 p.

COE, J.; KINNER, D.; AND GODT, J., 2008, Initiation conditions for debris flows generated by runoff at Chalk Cliffs, central Colorado: Geomorphology, Vol. 96, pp. 270–297.

COSTA, J. E., 1984, Physical geomorphology of debris flows. In Costa, J. E. and Fleisher, P. J. (Editors), Developments and Applications of Geomorphology: Springer-Verlag, New York. pp. 268–317.

DOOGE, J. C., 1957, The rational method for estimating flood peaks: Engineering, Vol. 184, No. 1, pp. 311–313. ´ , F. J.; MENE´ NDEZ-DUARTE, R.; AND VALDE´ S-RIERA, FERNANDEZ R., 2003, Digital model of corrected accumulated flow for peak discharge data acquisition and drainage system design: Engineering Geology, Vol. 69, No. 3, pp. 345–358.

FETTER, C. W., 2001, Applied Hydrogeology, 4th ed.: Prentice Hall, Upper Saddle River, NJ. 598 p.

HUNGR, O., 2000, Analysis of debris flow surges using the theory of uniformly progressive flow: Earth Surface Processes Landforms, Vol. 25, pp. 1–13.

HUNGR, O.; MORGAN, G. C.; AND KELLERHALS, R., 1984, Quantitative analysis of debris torrent hazards for design of remedial measures: Canadian Geotechnical Journal, Vol. 21, No. 4, pp.

663–677.

IKEYA, H., 1989, Debris flow and its countermeasures in Japan: Bulletin International Association Engineering Geology, Vol. 40, No. 1, pp. 15–33.

IZZARD, C. F., 1953, Peak discharge for highway drainage design.

Proceedings American Society Civil Engineers, Vol. 79, No. 10, pp. 1–10.

JAKOB, M.; HUNGR, O.; AND JAKOB, D. M., 2005, DebrisFlow Hazards and Related Phenomena: Springer, Berlin, Germany.

298

JAKOB, M. AND JORDAN, P., 2001, Design flood estimates in mountain streams: The need for a geomorphic approach: Canadian Journal Civil Engineering, Vol. 28, No. 3, pp. 425– 439.

LOS ANGELES COUNTY DEPARTMENT OF PUBLIC WORKS, 2006a, Hydrology Manual, Modified Rational Method Hydrology Support Files: Electronic document, available at http://dpw. lacounty.gov/wrd/Publication/index.cfm LOS ANGELES COUNTY DEPARTMENT OF PUBLIC WORKS, 2006b, Sedimentation Manual, 2nd ed. Water Resources Division: Electronic document, available at https://dpw.lacounty.gov/ wrd/publication/engineering/2006_sedimentation_manual/ Sedimentation%20Manual-Second%20Edition.pdf MARCHI, L.; ARATTANO, M.; AND DEGANUTTI, A. M., 2002, Ten years of debris-flow monitoring in the Moscardo Torrent (Italian Alps): Geomorphology, Vol. 46, No. 1, pp. 1–17.

MATSUMOTO, C., 2007, Debris Flow Hazard Assessment, Avalon School, MISSION File Number 07-665: Mission Geoscience, Inc..

MOODY, J. A. AND MARTIN, D. A., 2001, Initial hydrologic and geomorphic response following a wildfire in the Colorado Front Range: Earth Surface Processes Landforms, Vol. 26, pp. 1049– 1070.

MULVEY, W. E. AND LOWE, M., 1992, Cameron Cove subdivision debris flow: Survey Notes, Utah Geological Survey, Vol. 25, No. 2, 25 p.

RILEY, K. L.; BENDICK, R.; HYDE, K. D.; AND GABET, E. J., 2013, Frequency–magnitude distribution of debris flows compiled from global data, and comparison with post-fire debris flows in the western US: Geomorphology, Vol. 191, pp. 118– 128.

SHUIRMAN, G. AND SLOSSON, J. E., 1992, Forensic Engineering: Academic Press, San Diego, CA. 296 p.

THEOBALD, D. M. AND ROMME, W. H., 2007, Expansion of the US wildland–urban interface: Landscape Urban Planning, Vol. 83, No. 4, pp. 340–354.

U.S. DEPARTMENT OF TRANSPORTATION FEDERAL HIGHWAY ADMINISTRATION, 2008, Hydraulic Design Series No. 4, Introduction of Highway Hydraulics; National Highway Institute, Publication FHWA-NHI-08-090. Chapter 2, Estimating storm runoff from small areas: Electronic document, available at http://www.fhwa.dot.gov/engineering/hydraulics/pubs/ 08090/HDS4_608.pdf VANDINE, D. F., 1985, Debris flows and debris torrents in the southern Canadian Cordillera: Canadian Geotechnical Journal, Vol. 22 No. 1, pp. 44–68.

VANDINE, D. F., 1996, Debris Flow Control Structures for Forest Engineering: British Columbia Ministry of Forests Research Program, Vancouver, B.C.: Electronic document, available at www.for.gov.bc.ca/hfd/pubs/Docs/Wp/Wp22.htm VAN STEIJN, H., 1996, Debris-flow magnitude—Frequency relationships for mountainous regions of Central and Northwest Europe: Geomorphology, Vol. 15, No. 3, pp. 259–273.

WILFORD, D. J.; SAKALS, M. E.; INNES, J. L.; SIDLE, R. C.; AND BERGERUD, W. A., 2004, Recognition of debris flow, debris flood and flood hazard through watershed morphometrics: Landslides, Vol. 1, No. 1, pp. 61–66.

WILLIAMS, S. R. AND LOWE, M., 1990, Process-based debrisflow prediction method. In French, R. H. (Editor), Hydraulics/Hydrology of Arid Lands (H2 AL): American Society of Civil Engineers, New York. pp. 66–71.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 291–298

Thermal Remote Sensing for Moisture Content Monitoring of Mine Tailings: Laboratory Study BONNIE ZWISSLER1 Former graduate student, Department of Civil & Environmental Engineering, Michigan Technological University, 1400 Townsend Drive, Houghton, MI 49931

THOMAS OOMMEN Department of Geological and Mining Engineering and Sciences, Michigan Technological University, 1400 Townsend Drive, Houghton, MI 49931, email: toommen@mtu.edu

STAN VITTON ERIC A. SEAGREN Department of Civil & Environmental Engineering, Michigan Technological University, 1400 Townsend Drive, Houghton, MI 49931, emails: Vitton—vitton@mtu.edu; Seagren—eseagren@mtu.edu.

Key Terms: Fugitive Dust Emissions, Mine Tailings, Thermal Remote Sensing, Moisture Content, Site Characterization ABSTRACT Mining produces massive volumes of mine tailings that are deposited into large-scale mine tailings impoundments. A key environmental objective of managing these large impoundments is mitigating fugitive dust emissions by monitoring and controlling moisture, because moisture directly affects the tailings’ strength and the ability to apply dust control measures using motorized equipment. Therefore, understanding the spatial and temporal variations in moisture content for surface tailings is critical for characterizing dust susceptibility and trafficability. Remote sensing has been proven to be a useful tool for similar applications. This study utilized laboratory testing conducted on iron mine tailings to verify that: (1) a relationship exists between moisture content and strength for the surface of mine tailings, and (2) thermal remote sensing can be used to infer spatial variations in moisture content for surface tailings. Multivariate regressions were developed to identify the critical remote sensing and climatic variables and evaluate their influence in remotely measured moisture content. For tailings samples collected from two different North American iron mines, regressions using sample temperature and ambient humidity were able to predict surface moisture content (R2 > 0.9).

1 Corresponding

author email: bzwissler@barr.com.

INTRODUCTION Massive volumes of waste materials are produced each year by mining operations. According to the U.S. Environmental Protection Agency (USEPA), between one and two billion metric tons of mine waste are produced annually in the United States, excluding coal (USEPA, 2014(b)). Mine tailings result from ore beneficiation, in which crushing and grinding operations yield finely crushed rock particles (average particle size is 20 ␮m, with a significant portion between 1 and 10 ␮m) to separate the mineral from the waste rock. Once produced, the tailings are often deposited in slurry form into permanent tailings impoundments (Vick, 1983).

Over 3,500 active tailings impoundments exist worldwide (Davies et al., 2000). The impoundments constructed to contain tailings are among some of the largest earthen structures in the world, often with tailings stored in impoundments that are hundreds of meters thick and tens of square kilometers in surface area.

These impoundments are subject to intense regulatory and public attention because of the land areas they disturb and the hazards that can be associated with the mining waste they contain (Vick, 1983).

One of the many important environmental hazards associated with tailings impoundments is air pollution from blowing dust (Figure 1a; Buck and Gerard, 2001; Stovern et al., 2014). Blowing dust due to wind erosion of mine tailings can negatively impact human activity by creating poor visibility for drivers and respiratory health problems (Bang et al., 2009). Blowing dust is regulated by the USEPA as airborne particulate matter based on particle size and particle concentration. Particles of concern include coarse particles

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 299–312

299

Zwissler, Oommen, Vitton, and Seagren

Figure 1. Hazards associated with surface strength at mine tailings impoundments: (a) fugitive dust emissions, and (b) trafficability issues during application of dust control measures related to bearing capacity loss due to high moisture contents.

(diameters 2.5–10 ␮m) and fine particles (diameters less than 2.5 ␮m). The latter are of particular concern because they are small enough to reach the lower portions of the human respiratory tract and can cause respiratory health problems, including damage to lung tissue, cancer, and premature death (USEPA, 2012).

Under the USEPA Clean Air Act, the current national air quality standard limits particles smaller than 2.5 ␮m to 12 ␮g/m3 for an annual mean, or 35 ␮g/m3 for a 24 hour concentration, and it limits particles 2.5–10 ␮m to 150 ␮g/m3 for a 24 hour concentration (USEPA 2014(a)). Dusting events at tailings impoundments can lead to violations of these regulations.

Studying the concentration of particulate matter released from mine tailings impoundments is often desirable, especially if site conditions are favorable for dust generation, and if the tailings impoundment is in close proximity to residential areas, as is the case for the Iron King Mine tailings impoundment, located in Dewey-Humboldt, AZ (Stovern et al., 2014).

To control fugitive dust emissions, it is crucial to understand how the tailings are affected by wind erosion and how to detect when they may be susceptible to dusting. Typically, the lower the moisture content of the tailings, the higher is the dust susceptibility (Greeley and Iversen, 1987; Nickling and Neuman, 2009).

However, even for well-studied tailings impoundments, it is difficult to make static and general characterizations about the tailings properties without considering spatial and temporal variations in the dynamic impoundment. Understanding how the moisture content changes spatially and temporally is complex, so monitoring the tailings to determine if and when certain regions of the impoundment may be susceptible to dusting is important. The traditional approach for monitoring dust emissions from tailings is reactive in nature rather than proactive, with a focus on collecting

300

dust samples from gauges. This approach provides limited spatial coverage of the tailings, and it is also costly, labor intensive, and affected by trafficability. Because of these limitations, the problem may not be detected before the dusting event occurs.

In addition to the hazard of blowing dust (Figure 1a), trafficability can also be a concern for tailings impoundment managers, as shown in Figure 1b, which shows that high surface moisture can result in mine equipment losing bearing capacity during application of dust control measures. Trafficability is also related to the surface moisture content of the tailings, and it can be defined as the ability of the surface tailings to sup¨ port the traffic of heavy equipment (Muller et al., 2011).

Trafficability concerns include the ability of tailings to provide adequate traction for vehicles, and the ability to resist excessive compaction and structural damage (Paul and de Vries, 1979; Earl, 1997). While the hazard of fugitive dust emissions is caused by areas of tailings with low moisture content, accessing these regions for mitigation activities can be limited by regions of high moisture and low trafficability. Being able to identify regions of low trafficability is critical during the planning and execution of dust mitigation operations.

The ever-changing surface conditions of any active tailings impoundment make it difficult to characterize the surface conditions of the tailings. For dust susceptibility and trafficability monitoring, remote sensing may be an effective method for monitoring the surface of tailings impoundments, because it provides a synoptic view of surficial processes with high spatial and temporal resolutions, making it useful for monitoring and detecting change. To utilize the high spatial and temporal resolution of remote sensing for monitoring the impoundment surface moisture content and strength, it is critical to indirectly measure the moisture content in the top layer of the tailings. In

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 299–312

TIR Remote Sensing of Tailings—Lab Study

previous studies, thermal remote sensing has been used to directly detect soil moisture (Liu and Zhao, 2006; Minacapilli et al., 2012), but to the best of the authors’ knowledge, no study has looked at its applicability for mine tailings. It is hypothesized that a relationship between the surface strength and moisture content of the tailings can be developed, and that thermal remote sensing data can then be indirectly related to strength.

If proven true, this would be a key development for dust susceptibility monitoring as well as trafficability characterization in tailings impoundments.

The objective of this study was to perform a preliminary laboratory-scale analysis to verify whether a relationship exists between moisture content and strength for the surface of mine tailings, as well as to verify whether thermal remote sensing can be used to derive the spatial variation in moisture content in surface tailings impoundments. Development of the laboratory-scale relationship between moisture content and strength required identification of one or more critical climatic variable(s) and evaluation of their influence on the remotely measured moisture content.

Future efforts will scale the study to utilize high-spatialresolution thermal data collected with satellites or unmanned aerial vehicles (UAVs) in the field to detect changes in moisture content on the surface of tailings impoundments.

EXISTING TECHNIQUES UTILIZING REMOTE SENSING TO DETECT MOISTURE CONTENT Thermal remote sensing relies on passively collecting thermal infrared electromagnetic energy that is emitted from objects in the 3–14 ␮m portion of the electromagnetic radiation (EMR) spectrum. When kinetic energy exits an object, it is converted to thermal infrared radiant energy, which can be measured using remote sensing (Price, 1980; Salisbury and D’Aria, 1992; Becker and Li, 1995; Prata et al., 1995; Di Girolamo et al., 1998). Being able to remotely detect an object’s true kinetic temperature is a powerful tool: It enables the ability to know an object’s surface temperature without having to measure it in situ, and it also provides knowledge about the object’s thermal properties. One such thermal property of interest is thermal inertia, which is a measure of the thermal response of a material or object to temperature change. Thermal inertia varies with soil type and soil moisture content, and it is commonly measured quantitatively for soils. Thermal remote sensing (particularly thermal inertia) has been shown to have a linear relationship with moisture content (Liu and Zhao, 2006), and it is a common method used to indirectly determine relative soil moisture (Minacapilli et al., 2012).

Numerous studies have used remote sensing to determine soil (apparent) thermal inertia (Xue and Cracknell, 1995; Wang et al., 2010; Murray and Verhoef, 2007a, 2007b; Putzig and Mellon, 2007; Ramakrishnan et al., 2013; Soliman et al., 2013). Cracknell and Xue (1996) provided a thorough review of the early work performed on satellite-based mapping of thermal inertia, but airborne (Maltese et al., 2013) and groundbased remote sensing systems (Liu and Zhao, 2006) have also been used. Many studies have also used remotely sensed (apparent) thermal inertia to indirectly determine soil moisture (Price, 1980, 1985; Kahle et al., 1984; Zhang et al., 2002; Cai et al., 2005, 2007; Liu and Zhao, 2006; Scheidt et al., 2010; Minacapilli et al., 2012; Soliman et al., 2013).

Although there is significant evidence in the literature that thermal remote sensing can be related to soil moisture, there is no existing evidence in the literature that this holds true for mine tailings. If anything, remote sensing of tailings impoundments should be easier to implement than remote sensing of other soil applications, because the lack of vegetation on the impoundment makes it possible to sense soil surface conditions, which is often a limitation to using remote sensing for soil. In addition, the tailings contained in a given impoundment have exceptional uniformity when compared to most other heterogeneous soils and are mineralogically similar, so the primary variables within an impoundment are the tailings grain size distribution and moisture content. This study aims to fill that gap and to detect the moisture content of mine tailings using thermal remote sensing.

Before performing this study, we conducted preliminary laboratory tests with mine tailings that showed remote sensing for (apparent) thermal inertia was not an appropriate method for capturing variations in moisture content of the tailings, as supported by Price (1985). Rather than trying to modify the methods used by other studies (as described above) that were often developed for use with a specific satellite or sensor, this research tried to identify the critical remote sensing and atmospheric variables that allow for variations in moisture content of surface tailings to be detected; in this way, these variables could be obtained and used for remote sensing data collected from any platform (e.g., laboratory, UAV, or satellite) and at any scale (e.g., laboratory or field scale).

LABORATORY METHODS The laboratory testing was conducted using tailings samples collected from two North American iron mines: magnetite tailings from an impoundment at a mine in Michigan (these tailings will be referred to as MI-magnetite tailings in this paper, or MI-mag for

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 299–312

301

Zwissler, Oommen, Vitton, and Seagren

short), and magnetite tailings from an impoundment at a mine in Minnesota (these tailings will be referred to as MN-magnetite tailings, or MN-mag for short, in this paper). These samples of magnetite tailings had an average specific gravity of 3.09 (MI-mag) and 2.91 (MN-mag), respectively, as determined using a Micromeritics AccuPyc 1330 helium pycnometer. The MI-mag and MN-mag tailings were determined to be 95 percent and 82 percent silt-sized particles and 5 percent and 6 percent clay-sized particles, respectively, based on hydrometer analysis following ASTM D42263 (ASTM, 2007). These values are similar to those found during laboratory characterization of other mine tailings samples (Qiu and Sego, 2001). All laboratory samples were prepared as much as possible to mimic the actual conditions in the tailings impoundment.

Tailings were deposited into the impoundments in a 55 percent solid/45 percent liquid slurry (Muszynski, 2000; Price, 1998; Vasher, 1999) and allowed to gradually dry from fully saturated conditions. For the MI-mag tailings, surface moisture contents were measured in situ in the top 2 cm of tailings from 5 percent to 30 percent by mass, following ASTM D2216-10 (ASTM, 2010), and surface void ratios were measured in situ in the top 15 cm of tailings to range between 0.7 and 1.2, with a mean of 0.9 (Price, 1998). Therefore, all laboratory samples were prepared at void ratios between 0.8 and 1.0 in 76 × 76 × 76 mm polycarbonate soil boxes and fully saturated using a Mariotte tube constant head device. Even though deposition of field tailings occurs via wet pluviation and leads to particle-size sorting across the impoundment, we chose to constitute samples using dry pluviation followed by saturation for sample repeatability, and we do not believe that significant variation from field conditions exist in these laboratory specimens due to their small size.

Once the MI-mag and MN-mag tailings samples were prepared to fully saturated conditions, they were exposed to an artificial diurnal heating cycle daily, and they were monitored and tested until the samples were dry. This approach provided a range of moisture/ strength conditions per sample over the duration of testing. Diurnal heating cycles were simulated in the laboratory by heating samples for 6 hours daily under two 500 W halogen lamps and two full-spectrum lights mounted at a 45 degree angle to the samples at a height of 0.7 m above the samples. Only the surface of the samples was exposed to the lamp. A schematic of the test setup is shown in Figure 2. Measurements were taken twice daily for samples: once before heating, called “pre-heating,” and once after heating, called “post-heating.” These measurements are outlined in Table 1 and are described in more detail in the following text and in Zwissler (2016).

302

Figure 2. Schematic of laboratory thermal remote sensing testing setup, featuring two halogen lamps and two full-spectrum lamps mounted at 45 degree angles from iron samples.

Gravimetric moisture content was determined by monitoring sample mass loss using a balance. The sample mass measured at a given sampling event was subtracted from the initial sample mass recorded during sample preparation to determine the mass of water lost, which was used to back-calculate moisture content.

The reported moisture content is the average moisture content for the entire 76 × 76 × 76 mm sample, and it is worth noting that the surface moisture content conditions were likely lower than the bulk moisture content. However, because the samples started at the same conditions and were exposed to the same heating conditions, it is expected that all samples dried in the same way, and therefore the bulk moisture content should have a direct relationship with surface moisture content. For ease of data collection, bulk moisture content was measured rather than surface moisture content.

Relative soil strength was quantified using the penetration depth (cm) from a Humboldt H-1240 universal needle penetrometer with Humboldt H-1280 40–45 mm standard hardened stainless-steel needle, 150 g drop weight, and drop duration of 0.5 seconds.

Five measurements were taken at each sampling event,

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 299–312

TIR Remote Sensing of Tailings—Lab Study Table 1. Details of thermal remote sensing laboratory testing conducted per sampling event.

Measurement Collected

Instrumentation

Mass (g) Penetration depth (cm)

AND EK Balance Humboldt H-1200 needle penetrometer FieldSpec HandHeld Pro Spectroradiometer with plant probe General Tools PTH8708 Thermohygrometer FLIR SC640 thermal camera

Spectral reflectance (%)

Atmospheric temp. (K) & humidity (%) Thermal image (K)

one in each quadrant of the soil box and one in the center, with no less than 3 cm between penetration positions. The penetration depth measurements were then averaged to yield an average penetration depth, which was used to represent the relative surface strength of each sample.

Surface albedo was measured using an ASD R HandHeld Pro (325–1,075 nm) with plant FieldSpec probe; the average of 10 spectral reflectance curves per sample was used to calculate albedo as the mean reflectance from 400 to 1,075 nm. Albedo was obtained to quantify the color change associated with change in moisture of the tailings.

Atmospheric temperature (K) and humidity (percent) were measured using a thermohygrometer, both pre-heating and post-heating.

Change in surface temperature (K) was measured R ThermaCAM SC640 thermal camera using a FLIR (where FLIR indicates forward-looking infrared) that measures temperature in the 8–14 ␮m region of the EMR spectrum. Studies have shown that the 8–14 ␮m region of the EMR spectrum is not much affected by atmospheric interaction (Lillesand et al., 2014). The thermal remote sensing utilizes the principle that objects above absolute zero (0 K = −273◦ C) emit radiation in the thermal infrared region of the EMR. The emitted thermal radiation is a function of the emissivity, geometry, and the temperature of the object. Because the emissivity and the geometry of the object remain constant between the time points of the thermal remote sensing, a difference of the thermal images would indicate a change in temperature.

Purpose of Measurement

Tests per Sample

Moisture content calculation Represent relative surface strength Albedo calculation

1 5 10

Quantify atmospheric variation

1

Quantify surface temp & T due to diurnal heating

1

The raw data collected from the FLIR thermal camera was emitted electromagnetic radiation, as reflected radiation was found to be negligible for the samples and test setup used in the 8–14 ␮m region of the EMR spectrum (reflected radiation was found to be less than 3 percent using a diffuse infrared reflector as described in ASTM E1862-97: Standard Test Methods for Measuring and Compensating for Reflected Temperature Using Infrared Imaging Radiometers), which is supported by the work of Allison et al. (2016), Blacket (2014), Prakash (2000), and Settle (1981). In order to use the thermal camera to determine surface temperature, FLIR object parameters were measured as described in Table 2. The surface temperature of each sample, which was used to calculate the change in surface temperature between the diurnal heating cycles, was determined using the thermal imagery, the R ThermaFLIR object parameters, and the FLIR CAM Researcher software. The magnitude of soil temperature change, with respect to atmospheric temperature change, is a function of soil moisture. This is why this study measured change in temperature due to diurnal heating cycles, rather than just an average daily temperature or an instantaneous temperature. Figure 3 contains thermal images collected for an MN-mag sample before and after heating, which were used to calculate change in sample temperature (for these images, change in sample temperature was 11.7 K).

Testing of the MI-mag tailings occurred during the summer of 2014, while testing of the MN-mag tailings occurred in the summer of 2015. Between two and four

Table 2. Description of each FLIR object parameter and how the parameter was quantified.

Object Parameter

How to Quantify Each Parameter

Emissivity Reflected apparent temperature (K) Object distance (m) Atmospheric temperature (K) Relative humidity (%) External optics temperature (K) and transmission

Calculated as complement of albedo from spectroradiometer Same value as atmospheric temperature 0.7 m (distance between sample surface and camera lens) Measured with thermohygrometer between sample and lens Measured with thermohygrometer between sample and lens 1.0 (value to be used when no external optics for camera)

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 299–312

303

Zwissler, Oommen, Vitton, and Seagren

Figure 3. Thermal images collected for an MN-mag sample on June 8, 2015: (a) pre-heating, with a measured sample moisture content of 8.6 percent, a measured sample penetration depth of 0.40 cm, and an average sample temperature of 295.1 K; and (b) post-heating, with a measured sample moisture content of 7.3 percent, a measured sample penetration depth of 0.35 cm, and an average sample temperature of 306.8 K.

samples were tested at a time, with one data point collected per day per sample, yielding 45 total data points for the MI-mag tailings and 48 total data points for the MN-mag tailings. After all data were collected, the relationships between the measured variables for each set of tailings were explored using regression analysis. The validity of the regressions was assessed using tools such as root mean squared error (RMSE) and coefficient of determination (R2 ).

etration depth for MI-mag tailings using the following relationship: Penetration Depth (cm) = 0.1996e0.1087w ,

(1)

where R2 is 0.76, and the RMSE is 0.06 cm. Similarly, moisture content can be used to predict penetration depth for the MN-mag tailings using the following relationship: Penetration Depth (cm) = 0.2072e0.0718w ,

(2)

LABORATORY RESULTS As stated previously, it was hypothesized that if a relationship between the surface strength and moisture content of the iron tailings exists, then it can be used to indirectly relate thermal remote sensing data to the strength/dusting susceptibility of the tailings. To validate this hypothesis, the first relationship explored was that between moisture content and strength for the iron tailings.

The relationship between daily average moisture content and penetration depth, which is used to represent surface strength, is shown in Figure 4 for the MI-mag tailings and Figure 5 for the MN-mag tailings. Penetration depth has an inverse relationship with surface strength, because a stronger surface means lower penetration depth. The relationship between moisture content and penetration depth is not the same for the MImag and MN-mag tailings, demonstrating the need to consider site-specific differences between tailings samples. Nevertheless, the overall shape of the curves is similar, which indicates that it is reasonable to relate moisture content and strength for mine tailings. Gravimetric moisture content (w) can be used to predict pen-

304

Figure 4. Relationship between moisture content and surface strength (penetration depth) for MI-mag tailings (R2 = 0.76). Each data point represents the daily average measurement for a sample.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 299â&#x20AC;&#x201C;312

TIR Remote Sensing of Tailingsâ&#x20AC;&#x201D;Lab Study

Figure 5. Relationship between moisture content and surface strength (penetration depth) for MN-mag tailings (R2 = 0.84). Each data point represents the daily average measurement for a sample.

where R2 is 0.84, and RMSE is 0.04 cm. These relationships suggest that relationships using thermal remote sensing to predict moisture content can be indirectly applied to determine the relative surface strength of mine tailings, as long as the relationship between moisture content and strength is investigated for each new tailings sample being considered. For instance, the moisture content range for these two samples differs (approximately 0 to 20 percent for MI-mag, and 0 to 30 percent for MN-mag) for the same strength range (penetration depths of approximately 0 to 2 cm), which is assumed to capture strength variations that would be needed to detect trafficability and dust susceptibility issues.

Once it was validated that a relationship exists between moisture content and strength for mine tailings,

the relationship between the thermal remote sensing variables and moisture content, the dependent variable, was explored. Table 3 contains statistical information for each variable considered based on the laboratory data collected for the MI-mag tailings, and Table 4 contains the same information for the MN-mag tailings. All observed values in the laboratory data are representative of values observed in the field for soil (Scheidt et al., 2010; Soliman et al., 2013; Zwissler, 2016). Albedo was considered as a way to quantify the color change associated with change in moisture of the tailings. Sample temperature was directly determined with thermal remote sensing. While atmospheric temperature and atmospheric humidity were inputs for the thermal camera used to detect sample temperature, the trends in atmospheric conditions were also considered. The atmospheric temperature and humidity during laboratory testing were monitored, and the trend is shown in Figure 6 for the MI-mag testing and Figure 7 for the MN-mag testing. These data reveal that the atmospheric conditions of temperature and relative humidity varied widely throughout the testing periods. Furthermore, there was no direct relationship between the variation in temperature/humidity in the pre- and post-heating measurements. These variations reflected in the data can be explained by the fact that testing was performed during the summer in a laboratory that was not air conditioned and did not have any means for temperature/humidity control. These laboratory data were used to develop a multivariate linear regression to relate thermal remote sensing variables and atmospheric variables to the moisture content of the iron tailings, which is discussed in the following section.

MULTIVARIATE REGRESSION MODEL RESULTS AND DISCUSSION For each set of tailings, a multivariate regression was developed using sample temperature, sample albedo,

Table 3. Statistical information for each variable considered in relationship between moisture content and thermal remote sensing for MI-mag tailings, n = 45.

Variable (Units)

Symbol

Mean

Median

Min.

Max.

Standard Deviation

Albedo Pre-heating atmospheric temp. (K) Post-heating atmospheric temp. (K) Change in atmospheric temp. (K) Pre-heating atmospheric humidity (%) Post-heating atmospheric humidity (%) Change in atmospheric humidity (%) Pre-heating sample temperature (K) Post-heating sample temperature (K) Change in sample temperature (K) Gravimetric moisture content (%)

â?Ł Tatm,pre Tatm,post Tatm Hpre Hpost H Tsamp,pre Tsamp,post Tsamp w

0.0642 297.6 320.7 23.13 49.58 20.33 29.24 295.7 307.2 11.48 11.87

0.0626 297.6 320.2 23.10 47.00 21.00 29.00 295.6 307.0 11.00 12.64

0.0535 294.6 314.0 18.30 31.00 13.00 16.00 293.2 302.4 7.10 1.89

0.0812 300.0 325.4 26.20 64.00 28.00 42.00 298.2 316.2 18.80 20.45

0.0069 1.6381 3.0731 1.9976 9.9190 4.3432 6.8694 1.6385 2.5696 2.2554 5.3353

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 299â&#x20AC;&#x201C;312

305

Zwissler, Oommen, Vitton, and Seagren Table 4. Statistical information for each variable considered in relationship between moisture content and thermal remote sensing for MN-mag tailings, n = 48.

Variable (Units)

Symbol

Mean

Median

Min.

Max.

Standard Deviation

Albedo Pre-heating atmospheric temp. (K) Post-heating atmospheric temp. (K) Change in atmospheric temp. (K) Pre-heating atmospheric humidity (%) Post-heating atmospheric humidity (%) Change in atmospheric humidity (%) Pre-heating sample temperature (K) Post-heating sample temperature (K) Change in sample temperature (K) Gravimetric moisture content (%)

␣ Tatm,pre Tatm,post Tatm Hpre Hpost H Tsamp,pre Tsamp,post Tsamp w

— 295.9 311.7 15.86 40.23 22.42 17.81 294.0 306.2 12.24 16.04

— 295.9 313.0 17.25 40.80 22.25 17.00 294.4 305.6 12.00 15.28

— 294.4 299.4 4.80 30.80 11.80 9.80 290.4 302.0 9.50 2.67

— 297.5 327.2 30.26 53.50 35.80 27.80 297.2 312.6 17.30 32.20

— 0.9544 7.4722 7.3901 6.0573 6.0324 4.4687 1.6079 2.6766 1.7649 8.3110

and atmospheric conditions to predict the moisture content of the MI-mag and MN-mag tailings using thermal remote sensing variables. The program RStudio (R Core Team, 2014; R Studio Team, 2015) was

used to explore the relationships between the variables, to determine a multivariate linear regression for moisture content, and to perform statistical analysis of the regression model. The multivariate linear regressions

Figure 6. Atmospheric conditions during laboratory testing of MI-mag tailings (results shown in Figure 4). Each data point represents a single measurement.

306

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 299–312

TIR Remote Sensing of Tailings—Lab Study

Figure 7. Atmospheric conditions during laboratory testing of MN-mag tailings (results shown in Figure 5). Each data point represents a single measurement.

were developed using the data sets of 45 data points (Table 3) for the MI-mag tailings and 48 data points (Table 4) for the MN-mag tailings.

The development of a multivariate linear regression to predict the moisture content of the tailings was carried out in the same way for both sets of tailings. Rather than using advanced machine-learning techniques, a manual, iterative approach was used to assess different combinations of variables so that the significance of the variables selected, individually and in combination, could be assessed. Statistical tools were used to assess the validity of the regressions and the significance of the variables used. The validity of the regressions was assessed using tools such as RMSE and R2 . The significance of the contribution of the variables was assessed using tools such as adjusted R2 and regression parameter hypothesis test (t-test). Statistical tools, like Q-Q plots and histograms, were used to identify variables that could benefit from transformation. The original variables, as well as multiple transformations

(e.g., log and square root transformations), were tested using the same process to assess which transformation, if any, added the most value to the regression. A more complete description of the regression development technique can be found in Zwissler et al. (2014).

For brevity, only the final regression equation that provided the best statistical performance for data set is presented in this paper.

The equation for the multivariate linear regression developed to predict the moisture content of MI-mag tailings is: w = −8.8754 log( Tsamp ) + 23.4879 (Tsamp,post )1/3 + 0.0290(Hpre )2 − 51.6874 log(Hpost ) − 0.0474( H)2 .

(3)

The predicted versus observed plot for this regression is shown in Figure 8, and the variable assessment statistics for the regression are displayed in Table 5. Based on Eq. 3, the moisture content of the MI-mag tailings

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 299–312

307

Zwissler, Oommen, Vitton, and Seagren Table 5. Variable assessment statistics for moisture content multivariate linear regression developed for MI-mag tailings.

Variable

Estimate

log( Tsamp ) (Tsamp,post )1/3 (Hpre )2 log(Hpost ) ( H)2 *

−8.8754 23.4879 0.0290 −51.6874 −0.0474

Std. Error

t Value

Pr(>|t|)

Significance*

3.5601 4.8918 0.0071 13.6049 0.0122

−2.493 4.801 4.102 −3.799 −3.884

0.01691 2.22e-05 0.00020 0.00048 0.00038

+ +++ +++ +++ +++

Significance level codes: 0–0.1 percent: +++ ; 0.1–1 percent: ++ ; 1–5 percent: + ; 5–10 percent: “.”; 10–100 percent: blank.

can be predicted by using pre- and post-heating sample temperature and pre- and post-heating atmospheric humidity. In this regression, post-heating sample temperature and atmospheric humidity are the most significant variables, while change in sample temperature is less significant but still adds value to the regression (Table 5). The predicted versus observed plot (Figure 8) shows a good relationship between modeled and measured moisture content for the MI-mag tailings, with a high R2 (0.91), adjusted R2 (0.90), and RMSE (3.89 percent).

Similarly, the equation for the multivariate linear regression developed to predict the moisture content of MN-mag tailings is: w = 314.4710 1/( Tsamp ) + 0.2091 (Tsamp,post ) − 2.7136(Hpre )2 + 0.0403(Hpost )2 + 0.0396( H)2 .

(4)

Figure 8. Multivariate linear regression validation to predict moisture content for MI-mag tailings with thermal remote sensing and atmospheric variables (R2 = 0.91). Points indicate daily average values for each sample, and the line shown is the 1:1 line for the data.

308

The predicted versus observed plot for this regression is shown in Figure 9, and the variable assessment statistics for the regression are displayed in Table 6. Based on Eq. 4, the moisture content of the MN-mag tailings can also be predicted by using pre- and post-heating sample temperature and pre- and post-heating atmospheric humidity. In this regression, change in sample temperature is the most significant variable, while postheating sample temperature and pre-heating atmospheric humidity are less significant, and post-heating atmospheric humidity and change in atmospheric humidity have the least significance but still add value to the regression (Table 6). The predicted versus observed plot (Figure 9) shows a good relationship between modeled and measured moisture content for the MN-mag tailings, with a high R2 (0.92), adjusted R2 (0.91), and RMSE (5.07 percent).

Figure 9. Multivariate linear regression validation to predict moisture content for MN-mag tailings with thermal remote sensing and atmospheric variables (R2 = 0.92). Points indicate daily average values for each sample, and the line shown is the 1:1 line for the data.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 299–312

TIR Remote Sensing of Tailings—Lab Study Table 6. Variable assessment statistics for moisture content multivariate linear regression developed for MN-mag tailings.

Variable

Estimate

Std. Error

1/( Tsamp ) Tsamp,post Hpre (Hpost )2 ( H)2

314.4710 0.2091 −2.7136 0.0403 0.0396

67.2272 0.0595 0.7846 0.0162 0.0187

*

t Value 4.678 3.512 −3.459 2.490 2.117

Pr(>|t|)

Significance*

2.89e-05 0.00106 0.00124 0.01673 0.04013

+++ ++ ++ + +

Significance level codes: 0–0.1 percent: +++ ; 0.1–1 percent: ++ ; 1–5 percent: + ; 5–10 percent: “.”; 10–100 percent: blank.

There are notable differences between the regressions developed for MI-mag and MN-mag tailings. Each set of tailings requires its own equation to predict moisture content from thermal remote sensing variables.

Both the magnitude of the coefficients and the variable transformations used to most effectively predict moisture content vary between the MI-mag tailings (Eq. 3) and the MN-mag tailings (Eq. 4). This further demonstrates the need to consider site-specific conditions, suggesting that all relationships need to be validated, calibrated, or re-developed for each type of tailings to be studied.

However, there are also a number of striking similarities between these regressions. While the coefficients and the variable transformations used to predict moisture content vary between the MI-mag (Eq. 3) and MN-mag (Eq. 4) tailings, the actual variables used are consistent. The variables that were considered in regression development, but ultimately not utilized in the final regressions, are the same for both tailings samples; neither albedo nor atmospheric temperature added significant value to the regressions predicting moisture content. Albedo was considered in an attempt to quantify the color change associated with a change in moisture of the tailings, but the actual color change (and therefore, variation in albedo) observed during testing was less significant than anticipated, so it is not surprising that albedo was not helpful in predicting moisture content.

After observing the variation in atmospheric temperature during testing (Figures 6 and 7), it was expected that the variation might have an effect on the diurnal heating of the samples. Ambient temperature, however, was an input parameter when sample temperature was being measured using the FLIR thermal camera (Table 2). It is possible that the internal correction for ambient temperature made by the thermal camera was enough to account for these variations.

Another similarity between the two regressions is that sample temperature and atmospheric humidity were the most effective variables to predict moisture content for both tailings impoundment. Sample temperature was expected to contribute heavily to the prediction of moisture content, because sample tempera-

ture is the variable being remotely sensed and the variable that has been shown to have a relationship with moisture content (Liu and Zhao, 2006; Minacapilli et al., 2012). It is also reasonable that the atmospheric humidity would contribute to the prediction of sample moisture content. This indicates that, for future studies, when predicting moisture content from thermal remote sensing, both sample temperature and ambient humidity should be expected to contribute significantly to the sample-specific regressions being developed for tailings samples or even other soil samples.

PRACTICAL APPLICATIONS For both the MI-mag (Eq. 3) and MN-mag (Eq. 4) tailings, this paper presents a way to predict surface gravimetric moisture content using thermal remote sensing and atmospheric variables. Using Eq. 1 (MImag) and Eq. 2 (MN-mag), surface strength can then be estimated for gravimetric moisture contents of 0 to 20 percent (MI-mag) and 0 to 30 percent (MN-mag).

Similar relationships could be studied for other soil types, using the same methods presented in this paper, which take advantage of site-specific and atmospheric conditions.

While these relationships were developed in a laboratory setting, they were developed in a way that they can be directly applied to field scale. The most practical method for these relationships to be applied to tailings impoundment monitoring is through the use of UAV-based platforms. UAV technology is rapidly developing, with UAVs becoming more affordable and sensors (including thermal sensors) becoming more lightweight and UAV-friendly. The UAV industry is advancing rapidly, and the application of this research demonstrates one of many ways that UAVs can be used in the future, often with fewer limitations than seen with the use of traditional satellite remote sensing data (Colomina and Molina, 2014; Liu et al., 2014). For the application of this specific research, a UAV that can carry the payload of a thermal sensor is all that is required to collect the thermal remote sensing data needed to determine surface temperature for use in Eq. 3 and Eq. 4.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 299–312

309

Zwissler, Oommen, Vitton, and Seagren

To assess the accuracy of the application to the field before widespread application, UAV data collection must be paired with a ground-truthing campaign that measures surface moisture content and strength conditions.

In addition to the thermal data that are required to apply this research, atmospheric data are also needed.

Mining operations monitor atmospheric conditions such as temperature and humidity at or near their tailings impoundments, so those data are readily available to use in Eq. 3 and Eq. 4. For field application, the pre-heating temperature and humidity would be collected in the early morning and could be identified as the daily low temperature, and the post-heating temperature and humidity would be collected in the afternoon and could be identified as the daily high temperature.

SUMMARY AND CONCLUSIONS This study utilized laboratory testing to verify that: (1) a relationship exists between moisture content and strength for the surface of mine tailings, and (2) thermal remote sensing can be used to derive spatial variations in moisture content for the surface of two types of iron tailings. Multivariate regressions were developed to identify the critical remote sensing and climatic variables and evaluate their influence in the measurement of remotely measured moisture content.

Based on this work, the following conclusions can be drawn: 1. A relationship does exist between moisture content and surface strength for both MI-mag and MN-mag tailings. The relationship for each tailings sample is unique, but it follows similar trends, indicating that a similar relationship could be developed for other tailings impoundments being studied.

2. Thermal remote sensing data, specifically (pre- and post-heating) sample temperature and ambient humidity, can be used to predict moisture content for both MI-mag and MN-mag tailings. Again, the relationship for each tailings impoundment is unique, but a similar relationship could be developed for other tailings samples being studied.

3. Site-specific variation between tailings must be considered when studying moisture content and strength. All relationships need to be validated, calibrated, or re-developed for each type of tailings to be studied.

Future efforts will expand this study to the field scale by utilizing high-spatial-resolution thermal data collected with UAVs to detect changes in moisture content for the surface of tailings impoundments.

310

ACKNOWLEDGMENTS This material is based upon work supported by the National Science Foundation under Grant No. 1234126, with additional support from the Michigan Space Grant Consortium. The authors would like to acknowledge the Cliffs Natural Resources, Inc., of Cleveland, OH, for allowing access to their mine tailings impoundments in both northern Michigan and Minnesota.

REFERENCES ALLISON, R. S.; JOHNSTON, J. M.; CRAIG, G.; AND JENNINGS, S., 2016, Airborne optical and thermal remote sensing for wildfire detection and monitoring: Sensors, Vol. 16, No. 8, pp. 1310– 1339.

ASTM D422-63, 2007, Standard Test Method for Particle-Size Analysis of Soils: ASTM International, West Conshohocken, PA. ASTM D2216-10, 2010, Standard Test Methods for Laboratory Determination of Water (Moisture) Content of Soil and Rock by Mass: ASTM International, West Conshohocken, PA. ASTM E1862-97, 2010, Standard Test Methods for Measuring and Compensating for Reflected Temperature Using Infrared Imaging Radiometers: ASTM International, West Conshohocken, PA. BANG, S. S.; BANG, S.; FRUTIGER, S.; NEHL, L. M.; AND COMES, B. L., 2009, Application of novel biological technique in dust suppression: In Proceedings of the Transportation Research Board 88th Annual Meeting: Transportation Research Board, Washington, D.C. BECKER, F. AND LI, Z. L., 1995, Surface temperature and emissivity at various scales: Definition, measurement and related problems: Remote Sensing of Environment, Vol. 12, pp. 225–253.

BLACKET, M., 2014, Early ANALYSIS of Landsat-8 thermal infrared sensor imagery of volcanic activity: Remote Sensing, Vol. 6, pp. 2282–2295.

BUCK, S. AND GERARD, D., 2001, Cleaning Up Mining Waste: Political Economy Research Center Research Study, Bozeman, MT. CAI, G.; WU, J.; XUE, Y.; HU, Y.; GUO, J.; AND TANG, J, 2005, Soil moisture retrieval from MODIS data in Northern China Plain using thermal inertia model (SoA-TI). In Proceedings of the Geoscience and Remote Sensing Symposium, 2005 (IGARSS’05): IEEE, pp. 4501–4504.

CAI, G.; XUE, Y.; HU, Y.; WANG, Y.; GUO, J.; LUO, Y.; WU, C.; ZHONG, S.; AND QI, S., 2007, Soil moisture retrieval from MODIS data in Northern China Plain using thermal inertia model: International Journal Remote Sensing, Vol. 28, No. 16, pp. 3567–3581.

COLOMINA, I. AND MOLINA, P., 2014, Unmanned aerial systems for photogrammetry and remote sensing: A review: ISPRS Journal Photogrammetry Remote Sensing, Vol. 92, pp. 79–97.

CRACKNELL, A. AND XUE, Y., 1996, Thermal inertia determination from space—A tutorial review: International Journal Remote Sensing, Vol. 17, No. 3, pp. 431–461.

DAVIES, M.; MARTIN, T.; AND LIGHTHALL, P., 2000, Tailings dam stability: Essential ingredients for success. In Hustrulid, W. A; McCarter, M. K.; and Van Zyl, D. J. A. (Editors), Slope Stability in Surface Mining: Society for Mining, Metallurgy, and Exploration, Denver, CO, pp. 365–377.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 299–312

TIR Remote Sensing of Tailings—Lab Study DI GIROLAMO, L.; VARNAI, T.; AND DAVIES, R., 1998, Apparent breakdown of recriprocity in reflected solar radiances: Journal of Geophysical Resaerch, Vol. 103, pp. 8795–8803.

EARL, R., 1997, Prediction of trafficability and workability from soil moisture deficit: Soil and Tillage Research, Vol. 40, No. 3, pp.

155–168.

GREELEY, R. AND IVERSEN, J. D., 1987, Wind as a Geological Process: On Earth, Mars, Venus and Titan: Cambridge University Press Archive, Cambridge, U.K. KAHLE, A. B.; SCHIELDGE, J. P.; AND ALLEY, R. E., 1984, Sensitivity of thermal inertia calculations to variations in environmental factors: Remote Sensing of Environment, Vol. 16, No. 3, pp.

211–232.

LILLESAND, T.; KIEFER, R. W.; AND CHIPMAN, J., 2014, Remote Sensing and Image Interpretation: John Wiley & Sons, New York.

LIU, P.; CHEN, A. Y.; HUANG, Y.-N.; HAN, J.-Y.; LAI, J.-S.; KANG, S.-C.; WU, T.-H.; WEN, M.-C.; AND TSAI, M.-H., 2014, A review of rotorcraft unmanned aerial vehicle (UAV) developments and applications in civil engineering: Smart Structures and Systems, Vol. 13, No. 6, pp. 1065–1094.

LIU, Z. AND ZHAO, Y., 2006, Research on the method for retrieving soil moisture using thermal inertia model: Science in China Series D, Vol. 49, No. 5, pp. 539–545.

MALTESE, A.; CAPODICI, F.; CIRAOLO, G.; AND LA LOGGIA, G., 2013, Mapping soil water content under sparse vegetation and changeable sky conditions: Comparison of two thermal inertia approaches: Journal Applied Remote Sensing, Vol. 7, No. 1, pp.

073548.

MINACAPILLI, M.; CAMMALLERI, C.; CIRAOLO, G.; D’ASARO, F.; IOVINO, M.; AND MALTESE, A., 2012, Thermal inertia modeling for soil surface water content estimation: A laboratory experiment: Soil Science Society America Journal, Vol. 76, No. 1, pp. 92–100. ¨ , L.; LIPIEC, J.; KORNECKI, T. S.; AND GEBHARDT, S., 2011, MULLER Trafficability and workability of soils. In Glinksi, J.; Horabik, J.; and Lipiec, J. (Editors), Encyclopedia of Agrophysics: Springer, The Netherlands, pp. 912–924.

MURRAY, T. AND VERHOEF, A., 2007a, Moving towards a more mechanistic approach in the determination of soil heat flux from remote measurements: I. A universal approach to calculate thermal inertia: Agricultural and Forest Meteorology, Vol. 147, No. 1, pp. 80–87.

MURRAY, T. AND VERHOEF, A., 2007b, Moving towards a more mechanistic approach in the determination of soil heat flux from remote measurements: II. Diurnal shape of soil heat flux: Agricultural and Forest Meteorology, Vol. 147, No. 1, pp. 88–97.

MUSZYNSKI, M. R., 2000, Void Ratio Distribution of Normally Consolidated Coarse-Grained Magnetite Tailings as a Funcion of Aging Time: M.S. thesis, Michigan Technological University, Houghton, MI. NICKLING, W. G. AND NEUMAN, C. M., 2009, Aeolian sediment transport. In Parsons, A. J. and Abrahams, A. D. (Editors), Geomorphology of Desert Environments: Springer, The Netherlands, pp. 517–555.

PAUL, C. AND DE VRIES, J., 1979, Effect of soil water status and strength on trafficability: Canadian Journal Soil Science, Vol. 59, No. 3, pp. 313–324.

PRAKASH, A., 2000, Thermal remote sensing: Concepts, issues and applications: International Archives Photogrametry and Remote Sensing, Vol. 33, pp. 239–243.

PRATA, A. J.; CASELLES, V.; COLL, C.; SOBRINO, J.; AND OTTLE, C., 1995, Thermal remote sensing of land surface temperature from satellites: Current status and future prospects: Remote Sensing Reviews, Vol. 12, pp. 175–224.

PRICE, J. C., 1980, The potential of remotely sensed thermal infrared data to infer surface soil moisture and evaporation: Water Resources Research, Vol. 16, No. 4, pp. 787–795.

PRICE, J. C., 1985, On the analysis of thermal infrared imagery: The limited utility of apparent thermal inertia: Remote Sensing of Environment, Vol. 18, No. 1, pp. 59–73.

PRICE, J. C., 1998, Evaluation and Characterization of the Effectiveness of Dust Suppressants on Iron Ore Mine Tailings: Unpublished M.S. Thesis, Michigan Technological University, Houghton, MI. PUTZIG, N. E. AND MELLON, M. T., 2007, Apparent thermal inertia and the surface heterogeneity of Mars: Icarus, Vol. 191, No. 1, pp. 68–94.

QIU, Y. AND SEGO, D., 2001, Laboratory properties of mine tailings: Canadian Geotechnical Journal, Vol. 38, No. 1, pp. 183–190.

R CORE TEAM, 2014, R: A Language and Environment for Statistical Computing: R Foundation for Statistical Computing, Vienna, Austria.

R STUDIO TEAM, 2015, RStudio: Integrated Development for R: RStudio, Inc., Boston, MA. RAMAKRISHNAN, D.; BHARTI, R.; SINGH, K.; AND NITHYA, M., 2013, Thermal inertia mapping and its application in mineral exploration: Results from Mamandur polymetal prospect, India: Geophysical Journal International, Vol. 195, No. 1, pp. 357–368.

SALISBURY, J. W. AND D’ARIA, D. M., 1992, Emissivity of terrestrial materials in the 8–14 ␮m atmospheric window: Remote Sensing of Environment, Vol. 42, pp. 83–106.

SCHEIDT, S.; RAMSEY, M.; AND LANCASTER, N., 2010, Determining soil moisture and sediment availability at White Sands Dune Field, New Mexico, from apparent thermal inertia data: Journal of Geophysical Research: Earth Surface (2003–2012), Vol. 115, No. F2, pp. 1–23.

SETTLE, M., 1981, Geological Applications of Thermal Infrared Remote Sensing Techniques: Lunar and Planetary Institute (LPI) Technical Report 81-06, Houston, TX, pp. 1–138.

SOLIMAN, A.; HECK, R. J.; BRENNING, A.; BROWN, R.; AND MILLER, S., 2013, Remote sensing of soil moisture in vineyards using airborne and ground-based thermal inertia Data: Remote Sensing, Vol. 5, No. 8, pp. 3729–3748. ´ , A. E.; VILLAR, O. I. F.; STOVERN, M.; BETTERTON, E. A.; SAEZ RINE, K. P.; RUSSELL, M. R.; AND KING, M., 2014, Modeling the emission, transport and deposition of contaminated dust from a mine tailing site: Reviews on Environmental Health, Vol. 29, No. 1–2, pp. 91–94.

U.S. ENVIRONMENTAL PROTECTION AGENCY (USEPA), 2012, Particulate Matter (PM-10): Electronic document, available at http://www.epa.gov/airtrends/aqtrnd95/pm10.html U.S. ENVIRONMENTAL PROTECTION AGENCY (USEPA), 2014a, Fine Particule Matter (PM2.5) Designations: Electronic document, available at http://www.epa.gov/pmdesignations/ U.S. ENVIRONMENTAL PROTECTION AGENCY (USEPA), 2014b, Mine Waste Technology: Electronic document, available at http://www.epa.gov/nrmrl/std/mwt/ VASHER, D. R., 1999, A Technique for Assessing Fugitive Particulate Emission Potential from an Active Mine Tailings Basin: Unpublished M.S. Thesis, Michigan Technological University, Houghton, MI. VICK, S. G., 1983, Planning, Design, and Analysis of Tailings Dams: John Wiley & Sons, Inc., New York.

WANG, J.; BRAS, R.; SIVANDRAN, G.; AND KNOX, R., 2010, A simple method for the estimation of thermal inertia: Geophysical Research Letters, Vol. 37, No. 5, pp. L05404.

XUE, Y. AND CRACKNELL, A., 1995, Advanced thermal inertia modelling: Remote Sensing, Vol. 16, No. 3, pp. 431–446.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 299–312

311

Zwissler, Oommen, Vitton, and Seagren ZHANG, R.; SUN, X.; AND ZHU, Z., 2002, Remote sensing information model in surface evaporation from differential thermal inertia and its validation in Gansu Province: Science in China Series D, Vol. 32, No. 1, pp. 1041–1050.

ZWISSLER, B., 2016, Dust Susceptibility at Mine Tailings Impoundments: Thermal Remote Sensing for Dust Susceptiblity Characterization and Biological Soil Crusts for Dust Susceptibility

312

Reduction: Ph.D. Dissertation, Michigan Technological University, Houghton, MI. ZWISSLER, B.; OOMMEN, T.; AND VITTON, S., 2014, A study of the impacts of freeze–thaw on cliff recession at the Calvert Cliffs in Calvert County, Maryland: Geotechnical and Geological Engineering, Vol. 32, No. 4, pp. 1133–1148.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 299–312

Deterministic Three-Dimensional Rock Mass Fracture Modeling from Geo-Radar Survey: A Case Study in a Sandstone Quarry in Italy MOHAMED ELKARMOTY1 CAMILLA COLLA ELENA GABRIELLI ` STEFANO BONDUA ROBERTO BRUNO Department of Civil, Chemical, Environmental and Materials Engineering, University of Bologna, Via Terracini 28, 40131 Bologna, Italy; email: camilla.colla@unibo.it; elena.gabrielli4@unibo.it; stefano.bondua@unibo.it; roberto.bruno@unibo.it

Key Terms: Ground Penetrating Radar, Engineering Geology, Fracture Modeling, Visualization, Quarrying, Ornamental Stones ABSTRACT Rock mass fractures adversely affect the production of ornamental stone quarries. Fractures cause natural rock blocks, which threaten extraction of the required commercial block size of ornamental stones. Accurate subsurface detection and modeling of fractures are required for pre-exploitation evaluation and planning. This paper introduces a new three-dimensional deterministic fracture modeling approach using ground penetrating radar (GPR) as a data acquisition tool. A case study was performed in a fractured bench of a sandstone quarry in Firenzuola, Italy, using a 400 MHz GPR antenna. To accurately detect fractures at true depth, an in situ calibration based on previous knowledge of the depth of a subsurface reference reflector allowed us to estimate a bulk dielectric constant of the rock mass during the time of data acquisition. A data interpretation tracing technique was developed to model fractures as 3-D surfaces in two forms, either irregular or planes. The modeled fractures were visualized through a multi-platform visualization software package (ParaView). A comparison between the orientations of the fractures measured by the traditional manual method and the orientations of the modeled fractures is presented as a possible geologic validation for the detection and interpretation of fractures. For the objective of pre-exploitation evaluation, a distribution analysis provided an evaluation-based fracture index for the bench in the case study.

1 Corresponding

314

author email: mohamed.elkarmoty2@unibo.it

INTRODUCTION Geo-engineering applications require an accurate characterization of rock mass discontinuities (fractures, joints, cracks, foliations, etc.) in order to carry out safe and feasible surface or underground geoengineering projects. The classical characterization of rock mass discontinuities is based on the detection of outcropping fracture traces in rock faces. The detection of discontinuities in spatial intensity (amount and spacing) and the modeling of geometric features (extension, shape, and orientation) of the entire rock mass remain challenging research topics.

The methods for detection of discontinuities in different geo-engineering applications have been the topic of a number of publications. They can be mainly classified in destructive and non-destructive methods. Noteworthy destructive methods include those based on boring the rock mass, such as drill-core analysis, borehole tele-viewer, and well logs (Lau et al., 1987; Dezayes et al., 2000; Annavarapu et al., 2012; Zazoun, 2013 ; Gao et al., 2016). These methods are time consuming and expensive. However, they are used as data sources for the stochastic modeling of fractures and for huge projects (such as petroleum reservoir characterization), but not when boring the rock mass may be considered a partial deposit destruction (for example, in applications such as ornamental stone quarrying production). Non-destructive methods can be divided into contact and non-contact methods. Non-contact nondestructive surveying methods (laser scanning, photogrammetric, and optical methods; Assali et al., 2014; Deliormanli et al., 2014; Fisher et al., 2014; Lai et al., 2014; Vasuki et al., 2014) currently have widespread applications, such as landslides and slope stability analysis, particularly when the survey area is not accessible and the surveyed rock face is not flat. These methods can survey wide surfaces of rock faces, and often

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314â&#x20AC;&#x201C;331

3-D Fracture Modeling from GPR Survey

stochastic modeling of fractures is applied to the measured data. Contact geophysical non-destructive methods (ground penetrating radar [GPR], seismic tomography, and resistivity profiling methods; D´erobert and Abraham, 2000; Seol et al., 2001; Zou and Wu, 2001; Theune et al., 2006; Zajc et al., 2014; Mineo et al., 2015; Walton et al., 2015; Zimmer and Sitar, 2015) have also been applied in various applications to detect fractures.

A traditional method of contact non-destructive methods is the manual survey by compass (International Society For Rock Mechanics [ISRM], 1978; Priest and Hudson, 1981; Stavropoulou, 2014). It is used to characterize outcropping fractures and is trusted for results comparison and validation of other methods (Park and West, 2002; Assali et al., 2014; Fisher et al., 2014).

Fractures in rock masses decrease the economic value of quarries and cause a huge amount of waste during the exploitation of ornamental stone deposits.

Accurate three-dimensional (3-D) detection and modeling of rock mass fractures are required (i) to obtain a preliminary deposit evaluation before quarrying, and (ii) to optimize the cutting orientation of stones, which minimizes the waste. We define a deterministic model for fractures as a model that represents the geometric set and behavior of fractures as close as possible to reality without stochastic processes. However, the modeling of rock mass fractures, in a full 3-D deterministic way, is quite difficult because of: (i) data acquisition resolution, (ii) interpolations, and (iii) assumptions or manual interpretations of fractures data that may exist in a model. This paper presents the application of the GPR method to fracture detection and modeling in a bench of a sandstone quarry. We introduce improvements to the deterministic modeling of fractures and a new analytical methodology to obtain an evaluationbased fracture index for the quarry bench from GPR measurements.

PRINCIPLES OF GPR GPR was selected as the data acquisition tool in this research because it is a non-destructive fast method and because of its proven ability to detect subsurface discontinuities. GPR is based on the transmission of electromagnetic wave pulses (frequency ranging from 10 MHz to 2.6 GHz) in the material being surveyed.

Simultaneously, the receiver gains the reflected energy from discontinuities where there is a change in the dielectric properties (Annan, 2003). When radar waves hit a boundary between two materials with different dielectric properties, the reflected waves from this boundary are received by the antenna. The magnitude of the amplitude from a reflector such as a discontinuity depends on the reflection coefficient, which is controlled by the discontinuity aperture size

and the dielectric constant of the filling material. The relation between the dielectric constant of a medium (ε r ) and the signal propagation velocity in the medium (v) is defined by Eq. 1 (Reynolds, 2011), √ εr = c/v, (1) where c is the electromagnetic velocity in free space (299,792,458 m/s). The time for the signal to reach the target and return back after reflection is known as double reflective time of the signal (t), which is later converted to a vertical depth (d) in a radargram by Eq. 2, v = 2d/t.

(2)

The maximum penetration depth mainly depends on the frequency of the antenna. There are other factors that control the maximum penetration depth (signal attenuation), such as the water content and the medium homogeneity. Low-frequency antennas allow greater depth ranges to be obtained, but with a low resolution and vice versa. Further theoretical basis for GPR and signal processing can be found in Daniels (2004), Reynolds (2011), and Yelfm (2007).

PREVIOUS STUDIES Several applications for detecting and characterizing fractures through GPR in quarries of different deposits have been presented in the literature (Botelho and Mufti, 1998; Porsani et al., 2006; Kadioglu, 2008; Luodes, 2008; Mysaiah et al., 2011; Zajc et al., 2013; Elkarmoty et al., 2016a, 2016b). With respect to 3-D deterministic modeling of fractures from GPR survey in quarries, the most notable contributions are the works by Grandjean and Gourry (1996), Grasmueck (1996), and Grasmueck et al. (2013). The fracture modeling approach proposed in this paper aims at improving the level of deterministic accuracy and decreasing the gap among these contributions in the application to highly fractured rock masses. We define the deterministic accuracy level as the level of modeling that most closely approximates the geometric features of the fracture surfaces.

The application of Grandjean and Gourry (1996) was carried out in a marble quarry. It interpolated the picked fractures from parallel two-dimensional (2-D) radargrams in a gridded domain in which a 3-D fracture surface was modeled. The conditioned grid domain in this type of approach may affect the deterministic accuracy level of the modeled fracture surface.

The approach proposed in our study models a fracture surface in a domain in which just the reflections belonging to the fracture surface are detected. The application of Grasmueck (1996) was performed in a gneiss quarry. The migrated GPR cuboid was loaded into an interactive 3-D processing and interpretation software

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314–331

315

Elkarmoty, Colla, Gabrielli, Bondu`a, and Bruno

Figure 1. The location of the quarry in a map of Italy (Bargossi et al., 2008) and a panoramic top view of the quarry; the blue star indicates the bench under study (Google Maps GPS Coordinates Website, 2016).

package. Then, an automatic tracking program tracked the continuity of a fracture surface (horizon) from manual picks of the uppermost continuous troughs of the reflection bands in the GPR cuboid (Grasmueck, 1996). In this approach, a modeled fracture surface is the result of two successive phases of interpolations: (i) building the 3-D GPR cuboid by the processing software and (ii) using the tracking algorithm. The approach proposed in our study takes into consideration the modeling of small surface area fractures that do not appear as uppermost continuous troughs in the reflection bands in the GPR cuboid. The application of Grasmueck et al. (2013) performed a 3-D GPR survey that produced high-resolution imaging of sub-vertical fractures in a limestone quarry. By inserting the migrated GPR cuboid into a visualization and interpretation software package (GeoprobeTM ), sub-vertical fractures were identified. The modeling approach proposed in our study is feasible when no interpretation software package is available.

SITE CHARACTERISTICS This study was carried out on a sandstone quarry bench in Firenzuola, located in the province of Florence, Italy (Figure 1). It is an active quarrying city, especially for compacted grayish sandstone. There are various uses for this kind of stone: It is considered ideal for renovation work, prestigious interior and exterior residential work, commercial buildings with a minimalist style, or large modern architectural works (La Borghigiana Srl website, 2016).

316

The sandstones in Firenzuola belong to the Marnoso-Arenacea Formation, which is one of the numerous geologic formations that crop out in the Northern Apennine region. The sandstones extracted from Firenzuola are also known as “Pietra Serena di Ferenzuola;” this name is well-known in the historical heritage of Italy. The sediments of this formation were deposited in the Langhian geologic age, about 17 million years ago in a deep marine basin, extending in a NW-SE direction for a length of more than 300 km and for a width of more than 40 km (Bargossi et al., 2008).

The geo-coordinates of the bench center under study correspond to a latitude of 44◦ 8 8.5452 N, a longitude of 11◦ 25 8.3172 E, and a height of 857 m. The rock mass of the bench is highly fractured. The vertical fractures have aperture sizes larger than the horizontal fractures. In general, the aperture size of the fractures ranges from about 0.4 cm to 3.0 cm. The owner company of this quarry considered this bench as an un-mineable area. The bench surface is quite flat and suitable for the GPR survey activities. The pictures in Figure 2 show the bench in the case study (Figure 2A), some of the irregular outcropping fracture surfaces in the bench surface traced by yellow dashed lines and indicated by black arrows (Figure 2B), and some of the outcropping fractures in the bench face (Figure 2C).

There are three different colors of sandstone strata that crop out in the face of the bench (Figure 3). The stratification from up to down is blackish gray sandstone, gray sandstone, and black sandstone. The measured vertical outcropping thicknesses of these strata are about 0.3 m, 1.25 m, and 0.3 m, respectively. The

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314–331

3-D Fracture Modeling from GPR Survey

Figure 2. The bench in the case study with some outcropping fractures in the bench surface and face marked.

visual identification of the sandstone colors was carefully recognized due to the slight difference between the grayish and blackish grayish strata.

ture model can be visualized by coding the obtained geometric features in a visualization software package.

METHODOLOGY

Field Survey

Overview

The GPR instrument used in this study (Figure 4) consisted of a subsurface interface radar (SIR) System3000, equipped with a 400 MHz shielded antenna adapted to a cart (all produced by Geophysical Survey Systems GSSITM ). The 400 MHz antenna was selected considering the required penetration depth, the resolution range, and previous published works accomplished with the objective to detect fractures in sandstone rock mass (Maerz and Kim, 2000; Aqeel et al., 2013). It was shown that only the 400 MHz antenna, among 400, 500, and 900 MHz antennas, penetrated

The key focus of this research was to use GPR as a data source for the 3-D modeling of fractures in a deterministic approach. The modeling of fractures from GPR data can be carried out using an interpretation tracing technique that follows a fracture surface based on continuity indicators in parallel radargrams. Geometric features (dip, dip direction, vertices coordinates) of the modeled fractures can then be obtained through geometric formulations. Consequently, the 3-D frac-

Figure 3. Recognition of different sandstone beds by visual observation of the bench face. On the right-hand side of this picture, an irregular sub-vertical fracture surface crops out.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314â&#x20AC;&#x201C;331

317

Elkarmoty, Colla, Gabrielli, Bondu`a, and Bruno

parameters were estimated. A time range of 100 ns, 512 samples/scan, a format of 16 bits/sample, a rate of 120 scans/s, high band-pass filter of 100 MHz, low band-pass filter of 800 MHz, and a linear range gain of two points (−20 dB, 61 dB) were set up in the GPR system. Four lines (F01, F02, F03, and F04) parallel to the bench face, with an interval of 2.0 m and a length of 25.0 m each, were surveyed by the GPR (Figure 5).

The bearing of the parallel survey lines was measured

by Brunton compass with respect to north and was about 130◦ . The first survey line (F01) was 2.0 m away from the vertical bench face to move the GPR cart safely. Radargram F01 had a surveying delay of 1.25 m because of a non-smooth and non-safe surface area that was ignored during the surveying. The survey lines were oriented to be parallel to the bench face in order to be perpendicular to the outcropping sub-vertical fractures in the bench surface (following Tsoflias et al., 2004; Kadioglu, 2008; Seren and Acikgoz, 2012).

Figure 4. The GPR unit.

deeply enough to clearly identify the fractures up to 3.0 m depth in a sandstone rock mass (Maerz and Kim, 2000). Similarly, an application to detect fractures in a sandstone rock cut was carried out using a 400 MHz antenna reaching a penetration depth of 4.0 m (Aqeel et al., 2013).

The surveying was carried out on a cold winter day at a temperature of 5.6◦ C and a relative humidity of 61 percent. The rock mass was relatively wet because of rains in the previous days, leading to higher-amplitude reflections from fracture apertures filled with water (Toshioka et al., 1995) and limiting the penetration depth, since water attenuates the signal.

After testing the response of the signal in the rock mass of the bench, the preliminary data acquisition

Estimation of a Representative Dielectric Constant for the Rock Mass Estimation of the signal propagation velocity of the medium is required to estimate ε r from Eq. 1. The signal processing software package (RADANTM ) used in this study allows the use of a single value of ε r , for a radargram, to convert the double travel time to depth.

An inaccurate estimation of the parameter ε r strongly affects the true location of reflectors.

Rock mass formations are a mix of materials with different dielectric properties; for example, fractured rock mass may contain different strata and different filling materials in fracture apertures such as water, air, gas, oil, or sediments. That is why, finding the ε r of an intact rock in the laboratory will not accurately

Figure 5. A plan view of the survey lines over the bench surface. Points A and B are reference points used in section “Evaluation-based fracture index.”

318

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314–331

3-D Fracture Modeling from GPR Survey

Figure 6. Determining the double reflective time of the boundary between the black and gray sandstone strata from radargram F01. The blue line is the interpretation of this boundary. The gray sandstone has the weakest amplitude reflections (mostly the white zone), and the blackish gray sandstone has an intermediate phase of water content. On the right-hand side of this figure, there is the interpretation of the stratification boundaries.

represent the rock mass. Even if many different rock samples were collected from the site with preserved physical conditions, they would not accurately represent the heterogeneous and discontinuous character of the rock mass. It is recommended, when possible, to obtain a representative ε r of the rock mass during the time of data acquisition for applications that require high accuracy in detecting subsurface anomalies at accurate depth, rather than using published values or values obtained in a laboratory from intact rock samples. The following paragraphs describe how a representative ε r was obtained for this case study through a technique named in the literature as “known depth to reflector” (American Society for Testing and Materials [ASTM]: D6432-11, 2011). This technique was used by Maerz et al. (2015) to find the dielectric constant for a sandstone block in the laboratory. We applied this technique in situ to obtain a more representative result for the rock mass body.

In a correlation study between sandstone colors and water adsorption performed by Mubiayi (2013), it was found that blackish sandstone had the maximum water absorption capability among six sandstone samples with different colors. In contrast, the grayish sandstone has a minimum water absorption capability. From this point and according to the wet rock mass condition, it was expected for the blackish sandstone stratum to have the highest water content.

A preliminary analysis of radargrams showed that the boundaries of the strata were almost perpendicular to the bench surface. The detection of the strata boundaries in the radargrams depends on the amount of amplitude reflections. The reflections in the blackish sandstone stratum were the strongest, as expected, because of the high water content in this stratum (see F01 as an example in Figure 6). The lower boundary of the grayish sandstone was selected as the reference horizontal reflector because it is the deepest reflector that crops out in the face. Therefore, a more represen-

tative value of ε r could be estimated, since the signal passes through the major part of the rock mass under study. The lower boundary of the grayish sandstone was detected in the radargrams almost at double reflective times of 27.4, 28.1, 27.7, and 27.9 ns in F01, F02, F03, and F04, respectively.

Knowing the depth of the lower boundary of the grayish sandstone strata that is visible from the bench surface (1.55 m) and the average double reflective time of this boundary (27.77 ns), the average signal propagation velocity in the rock mass could be calculated from Eq. 2. Thus, a representative ε r for the rock mass of 7.2 was obtained from Eq. 1. This estimated value is acceptable, since as given by Daniels (2004), the dielectric constant of sandstone in wet conditions is in the range of 5 to 10.

The method used for estimating the bulk dielectric constant for this rock mass body may lead to some uncertainty regarding the subsurface location of anomalies in the radargrams. Assuming that the depth of the lower boundary of the grayish sandstone strata (1.55 m) is constant through the whole rock body, absolute errors in the subsurface depth of anomalies were estimated as shown in Table 1.

GPR Data Interpretations After applying standard signal processing functions, noise was removed. A deconvolution processing function was also applied, because it is important for Table 1. Absolute error estimation for the proposed method of estimating the bulk dielectric constant.

Radargram F01 F02 F03 F04

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314–331

Absolute Error (cm) 2.15 1.76 0.47 0.64

319

Elkarmoty, Colla, Gabrielli, Bondu`a, and Bruno

Figure 7. The processed radargrams with interpretation of the stratification boundaries of the sandstone strata. The yellow lines refer to the boundary between the blackish gray sandstone strata and the gray sandstone, while the red lines refer to the boundary between the gray sandstone strata and the black sandstone.

optimally enhancing the range resolution below a quarter of the wavelength (Widess, 1973) by maximizing the bandwidth and minimizing the GPR pulse dispersion. The reconstruction of the true shape of the detected fractures was carried out by applying migration (Botelho and Mufti, 1998).

The boundaries of the three sandstone strata are interpreted in Figure 7. The boundaries of the strata appear in the radargrams as slightly irregular surfaces with a dominant orientation parallel to the bench surface. Therefore, the boundaries were interpreted as lines in the 2-D radargrams.

A fracture tracing interpretation technique is proposed to model fracture surfaces. A fracture trace is the line (fracture reflection) resulting from the intersection between a fracture surface and a cross-sectional radargram. The proposed interpretation technique traces a fracture surface based on the continuity indicators of a fracture trace in successive radargrams. The continuity indicators are determined by the amount of reflection

320

amplitude, inclination, location, and length of fracture trace reflections. The vertical and the horizontal fractures were interpreted separately because of the highly fractured status of the rock mass.

The tracing technique was applied to interpret vertical fracture surfaces in such a way that each single detected fracture trace belonging to a fracture surface was highlighted by a particular color in each radargram (Figure 8). Some of the interpreted vertical fractures (black color) could not be traced in more than one radargram. These fractures are without a continuity indicator, while others, such as the yellow and cyan fracture surfaces, extend through the entire surveyed area.

Some vertical fractures, such as the cream-colored one (Figure 8), were detected only in a particular surveyed volume (F01 and F02) without more continuity indicators. The interpretations showed that not all the outcropping vertical fractures in the bench extend through the entire penetration depth (see as examples the green and pink fractures on the left-hand side of Figure 8).

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314â&#x20AC;&#x201C;331

3-D Fracture Modeling from GPR Survey

Figure 8. Interpretation of vertical or sub-vertical fractures by using the fracture tracing interpretation technique.

The horizontal and sub-horizontal fracture reflections in this rock mass showed a random fracture pattern with higher intensity and variety in characteristics.

For this reason, the proposed tracing technique is not applicable to these fractures. The horizontal fracture reflections were interpreted as single-appearing reflections (Figure 9) because of difficulties in finding continuity indicators. It is worth mentioning that the reflection spots from the vertical fractures were avoided in the interpretation of horizontal fracture reflections.

Modeling and Visualization Fracture Modeling Formulations Fractures propagate in the rock mass as irregular surfaces; however, they are usually modeled as planes for simplicity. The proposed deterministic model represents the fractures not only as planes, but also as 3-D surfaces. The geometric features of fractures (dip, dip direction, length, vertices coordinates) were obtained by developing Excel spreadsheet formulations

for two schemes. The first scheme was for fractures interpreted by the tracing technique (FITT), mostly vertical fractures in this case study (Figure 8). The second scheme was for fractures interpreted as single appearance (FISA) in a radargram, mostly horizontal fractures in this case study, or few vertical fracture traces without continuity indicators (Figure 9). The input and output parameters of the FITT and FISA formulations are listed in Table 2.

The fracture modeling concept used in this paper is close to the one published in Grandjean and Gourry (1996), in particular, for the modeling of vertical fracture surfaces. The fracture modeling approach in Grandjean and Gourry (1996) is based on correlating the detected fracture reflections (spots) in parallel radargrams and then interpolating a 3-D fracture surface in a regular-spaced grid (Figure 10A).

Accordingly, some spots-free areas in a modeled fracture surface are included as a part of a fracture surface, neglecting that there may be multiple separate surfaces of fractures. The modeling approach presented in this paper models a fracture surface without a dimensioned

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314â&#x20AC;&#x201C;331

321

Elkarmoty, Colla, Gabrielli, Bondu`a, and Bruno

Figure 9. Interpretation of horizontal or sub-horizontal fractures.

grid and just depends on the interpolation of the detected traced extensions (Figure 10B). The reason for this was to decrease the amount of assumptions, particularly for regions without data.

For the FITT formulation, interpreted fracture traces of a fracture surface often have different plunges

(␦) and coordinates (x, z) in the successive radargrams (Figure 11). The FITT formulation calculates the dip direction (ф) and strike (␣) of the regression plane of a fracture surface. The bearing of the survey lines is necessary to calculate the dip direction of a fracture plane. The dip angle (␪) of a regression plane was

Table 2. Inputs and outputs of fractures interpreted by the tracing technique (FITT) and fractures interpreted as single appearance (FISA) formulations.

Formulation

Inputs

Outputs

FITT

• Extremity coordinates of fracture traces (x, z).

• Bearing of the survey lines.

• Interval between the survey lines.

• Trend and plunge of fracture traces.

• Length of all fracture traces.

• Dip, dip direction, and strike of modeled surfaces.

• Generating and arranging of vertices coordinates of fracture surfaces.

FISA

• Extremity coordinates of fracture traces (x, z).

• Bearing of the survey lines.

• Interval between the survey lines.

• Number of surveying lines.

• Offset from the bench face.

• Trend and plunge of fracture traces.

• Length of all fracture traces.

• Dip, dip direction, and strike of modeled planes.

• Generating and arranging of vertices coordinates of fracture planes.

322

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314–331

3-D Fracture Modeling from GPR Survey

Figure 10. On the left, modeling of a fracture surface in a conditioned regular-spaced grid (Grandjean and Gourry, 1996). On the right, reconstruction of the modeled fracture surface on the left by the proposed method (3-D red surfaces of fractures), ignoring the reflection-free areas.

calculated by taking the average of ␦. The trend (␥ ) of a fracture trace was assumed to be either in the bearing of the survey lines (130◦ ) or in the opposite direction (310◦ ).

On the other hand, for FISA, the formulation is different because it models each single fracture trace as a plane extending to half the interval distance before and after the radargram in which the fracture trace was detected. Each fracture plane takes dip angle and dip direction equivalent to ␦ and ␥ for the fracture trace being detected. The single fracture traces detected in the first and last radargram were modeled as planes extending just inside the limit of the surveyed area of 6.0 m × 25.0 m.

For a 3-D visualization of the modeled fractures, the vertices coordinates of each fracture surface or plane were generated and arranged by the two formulations, in the particular order necessary for a visualization programming code. In this case study, the FITT formulation was applied to 66 fracture traces that led us to define the geometric features of 22 vertical fracture surfaces, while the FISA formulation was applied to 133 single fracture traces producing 133 modeled fracture planes. Some numerical data from the FITT and

Figure 11. A sketch illustrating how the FITT formulation obtains the dip and dip direction of a fracture surface.

FISA formulation spreadsheets are shown in Table 3 and Table 4, respectively.

Coding and 3-D Visualization The visualization of fractures plays a key role in the evaluation of the fracture status inside a rock mass.

Among the various 3-D visualization software packages, ParaView (ParaView website, 2016) was selected because it is an open-source and multi-platform application for visualizing and analyzing scientific data sets. Among the various file formats accepted by ParaView, the polygon file format (.PLY; Bourke, 2011) was chosen. This format is used by researchers to describe objects in 3-D models in a code that defines points, lines, polygons, etc. In this case study, the objects being visualized were the planes and surfaces of fractures and the volumetric boundary of the bench.

The vertices computed by the FITT and FISA formulations were coded to describe the fractures’ planes and surfaces. In this case study, the coding of the modeled fracture planes obtained from FISA was included in one file (.PLY), while the modeled fracture surfaces obtained from FITT were included in a separate file. In addition, the geometry of the surveyed bench (25.0 m × 6.0 m × 2.5 m) was also coded to visualize the bench boundaries jointly with the fractures.

The modeled fracture surfaces were coded keeping the same colors used in the interpretation (Figure 8).

The gray color was given to all the modeled fracture planes (FISA). ParaView enables the user to freely turn the whole body of objects in different directions (Figure 12), and it also enables the user to temporarily remove objects so as to clearly study any other object; for instance, removing the modeled horizontal fracture planes allows for a better visualization of the geometric features of the modeled vertical fracture surfaces (Figure 13). The framework of this research methodology is summarized in a process chart in Figure 14.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314–331

323

324

.. ..

..

22

1 2 3 4 5 6 133

Frac.

Trace no.

..

3

..

11

2.8 5.0 1.3 1.6 1.7 3.7 —

x2 (m)

1 2 3 4 ..

2 3 4 .. ..

Radargram No. 1.4 0.8 0.5 0.1 ..

10.1 10.2 10.2 .. ..

x2 (m) 0.2 0.0 0.0 0.0 ..

0.0 0.0 0.0 .. ..

z2 (m) 1.5 1.1 1.0 0.4 ..

10.3 10.4 10.5 .. ..

x1 (m) 2.0 2.2 2.5 2.3 ..

2.1 1.8 1.9 .. ..

z1 (m)

(x1 , z1 ) Lower Point

86.7 82.2 78.7 81.3 ..

84.8 84.0 82.5 .. ..

␦ (0) 130.0 130.0 130.0 130.0 ..

130.0 130.0 130.0 .. ..

␥ (0)

.. ..

..

83.8

82.3

␪ (0)

.. ..

..

38.6

28.0

␣ (0)

.. ..

..

128.6

118.0

ф (0) 1.8 2.2 2.5 2.3 ..

2.1 1.8 1.9 .. ..

Length (m) 1.4 0.8 0.5 0.1 ..

10.1 10.2 10.2 .. ..

x3 (m) 0.2 0.0 0.0 0.0 ..

0.0 0.0 0.0 .. ..

z3 (m) 0.0 2.0 4.0 6.0 ..

4.0 6.0 8.0 .. ..

y3 (m)

0.4 0.5 0.9 1.8 2.0 1.6 —

z2 (m) 4.8 1.3 3.1 3.1 3.1 6.5 —

x1 (m) 0.4 0.6 0.9 1.8 2.0 1.8 —

z1 (m)

(x1 , z1 ) Lower Point

0.0 0.8 0.0 1.1 2.1 4.2 —

␦ (0) 130.0 310.0 130.0 130.0 130.0 130.0 —

␥ (0) 2.0 3.8 1.8 1.5 1.4 2.8 —

Length (m) 2.8 5.0 1.3 1.6 1.7 3.7 —

x3 (m)

0.4 0.5 0.9 1.8 2.0 1.6 —

z3 (m)

0.0 0.0 0.0 0.0 0.0 0.0 —

y3 (m)

2.8 5.0 1.3 1.6 1.7 3.7 —

x4 (m)

0.4 0.5 0.9 1.8 2.0 1.6 —

z4 (m)

1.0 1.0 1.0 1.0 1.0 1.0 —

y4 (m)

4.8 1.3 3.1 3.1 3.1 6.5 —

x5 (m)

0.4 0.6 0.9 1.8 2.0 1.8 —

z5 (m)

Planes Vertices Coordinates

0.0 0.0 0.0 0.0 0.0 0.0 —

y5 (m)

4.8 1.3 3.1 3.1 3.1 6.5 —

x6 (m)

1.5 1.1 1.0 0.4 ..

10.3 10.4 10.4 .. ..

x4 (m)

0.4 0.6 0.9 1.8 2.0 1.8 —

z6 (m)

2.0 2.2 2.5 2.3 ..

2.1 1.8 1.9 .. ..

z4 (m)

Surfaces Vertices Coordinates

Table 4. Numerical data from FISA formulation for modeled horizontal fracture planes from detected fracture traces in radargram F01.

(x2 , z2 ) Higher Point

4

No. of Frac. Traces

1

Fracture Surface No.

(x2 , z2 ) Higher Point

1.0 1.0 1.0 1.0 1.0 1.0 —

y6 (m)

0.0 2.0 4.0 6.0 ..

4.0 6.0 8.0 .. ..

y4 (m)

Table 3. Numerical data from FITT formulation for modeled fracture surfaces. The fracture surfaces 1 and 11 are data from fracture surfaces interpreted in red and blue colors in Figure 7, respectively.

Elkarmoty, Colla, Gabrielli, Bondu`a, and Bruno

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314–331

3-D Fracture Modeling from GPR Survey

Figure 12. Screenshots, from ParaView, showing the visualized 3-D fracture model (FISA and FITT) in different orientations.

ANALYSIS OF RESULTS Fracture Classifications The classical method for presenting and classifying rock mass discontinuities is through a stereographic projection. Such a graphical method allows us to represent fractures and classify the fracture systems in the rock mass into sets according to their geo-spatial orientations (dip and dip direction). All the detected fracture traces here are presented in a lower-hemisphere stereographic projection, using the Schmidt method of equal area (Figure 15A). The geo-spatial data used in this

representation are (␥ , ␦) of the fracture traces. A discrimination analysis, based on the variation in plunge allowed us to classify the sets into three groups. The first group varies in plunge from 0◦ to 16◦ , the second group varies in plunge from 32◦ to 48◦ , and the third one varies in plunge from 60◦ to 90◦ (Figure 15B). The horizontal and sub-horizontal fracture traces (set 1) are predominant, representing 62 percent of fractures, while the vertical and sub-vertical fracture traces (set 3) represent 33 percent. A small percentage of 5 percent (set 2) has an intermediate inclination.

The mean trace length was considered as a character of fracture sets (Sonntag et al., 2012; Zeeb et al., 2013).

Figure 13. A screenshot, from ParaView, showing the visualized vertical fracture surfaces (FITT), filtering the fractures from the FISA formulation.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314–331

325

Elkarmoty, Colla, Gabrielli, Bondu`a, and Bruno

Figure 14. Research methodology framework.

The three sets of fracture traces are characterized by the following mean trace lengths: set 1 = 2.76 m, set 2 = 0.86 m, and set 3 = 1.82 m. Thus, the analysis of fracture trace lengths may be considered as a kind of geologic validation of the GPR interpretations of fracture reflections.

Spacing between discontinuities is a characteristic of each fracture set (Priest and Hudson, 1976). Since the proposed fracture modeling formulations identify the fractures in 3-D space, spacing can be estimated with additional geometric formulations. However, spacing was not estimated in the presented case study. The high intensity of horizontal fractures is the main reason for this bench being completely un-mineable, since it inhibits extraction of commercial-size blocks.

Comparison with the Traditional Manual Method The manual survey method is the traditional method used to characterize fracture orientations from rock faces (International Society for Rock Mechanics [ISRM], 1978; Priest and Hudson, 1981). The manual method was applied to the outcropping vertical fracture traces in the bench face. It led to the measurement of, through a digital inclinometer, the plunges of 25 vertical and sub-vertical fracture traces, since the bench face is vertical and flat. This manual survey was car-

ried out to compare its results with the orientations of the modeled fracture surfaces (FITT). This comparison can be used as a possible validation for the proposed approach (following Park and West, 2002; Assali et al., 2014; Fisher et al., 2014). The comparison between the modeled horizontal fracture planes (FISA) and the traditional manual method was not considered, since few horizontal fracture traces outcrop in the bench face.

The comparison was carried out through a graphical representation of fracture orientations in a lower-hemisphere stereographic projection, using the Schmidt method of equal area. The orientations of the 22 modeled vertical fracture surfaces are presented in Figure 16A. The stereograph shows that there are two sets of vertical fractures, with set 1 being the predominant one. The mean direction of these data is 124◦ , with a maximum frequency of 59.09 percent; the direction of set 1 varies from 112◦ to 135◦ , while the direction of set 2 varies from 292◦ to 307◦ (Figure 16B).

The stereographic representation of the manual method data (Figure 17A) demonstrates that their behavior is similar to the modeled fracture surfaces; however, the difference between the fracture sets is not as clear as in Figure 16A. The mean direction is 130◦ , with a maximum frequency of 60 percent (Figure 17B), which are close to the mean direction and the maximum frequency of the modeled vertical fractures. It is worth

Figure 15. Graphical representation of the whole set of detected fracture traces through a stereographic projection (A) and through a discrimination analysis shown in a pie chart (B).

326

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314–331

3-D Fracture Modeling from GPR Survey

Figure 16. Graphical representation of the orientations of the modeled vertical fracture surfaces in a stereographic projection (A) and in a rose diagram (B).

mentioning that the GPR method, followed by modeling, is able to characterize the orientations of hidden or outcropping fracture surfaces, while the traditional method is limited to characterizing the orientations of outcropping fracture traces in a rock surface.

EVALUATION-BASED FRACTURE INDEX For the purpose of pre-exploitation evaluation, it is important to classify the rock mass into volumetric zones, based on the amount of fractures. An evaluation-

based fracture index for the bench in the case study was obtained through a distribution statistical analysis of the detected fractures inside the bench. The bench was divided, parallel to the bench surface, into five regions with equal intervals of 0.5 m (Table 5), since the maximum penetration depth was 2.5 m. For each region, the number of the detected vertical and horizontal fracture traces was determined from the radargrams. When a fracture trace existed in a region, this region was given a load of “1,” representing a fracture trace. In the case where a fracture trace extended through several regions, each region was given a load of “1.” Thus, each

Figure 17. Graphical representation of the data obtained by the manual method in a stereographic projection (A) and in a rose diagram (B).

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314–331

327

Elkarmoty, Colla, Gabrielli, Bondu`a, and Bruno Table 5. Identification of the depth regions. Each region is identified by a subsurface depth, where 0.0 m refers to the surface of the bench.

Depth Region Region 1 Region 2 Region 3 Region 4 Region 5

Subsurface Depth (m) 0.0–0.5 0.5–1.0 1.0–1.5 1.5–2.0 2.0–2.5

region had a certain integer load of fractures for each radargram.

The histograms of the load of vertical and horizontal fracture traces in each depth region enabled us to study the fracture distribution behavior within the depth regions (Figure 18). With regard to the vertical fractures (Figure 18A), region 1 is the most fractured one in all the radargrams, and the frequency decreases with increasing depth. Radargram F03 shows the lowest frequency of vertical fractures over the majority of depth regions, while radargram F02 shows the opposite behavior. Regarding the general trend, the fracture frequency decreases gradually with increasing depth in all the radargrams.

On the other hand, the histogram of the horizontal fracture frequency (Figure 18B) shows that the fracture frequency, through all the radargrams, mostly increases gradually with increasing depth up to a transition point in the third region, and then it decreases dramatically with increasing depth. Radargram F03 has the highest frequency in the first three depth regions, and interestingly, the horizontal fractures in radargram F03 disappear in region 4. The normal distributions of fracture traces are clear in radargrams F02 and F04.

In order to study the fracture distribution behavior in the bulk volume of the bench, the total load of horizontal, vertical, and all fractures, for each depth region, through the four radargrams is graphically pre-

Figure 19. Graphical representation of the total load of all fractures within the depth regions.

sented by the histogram in Figure 19. This shows that the bulk volume of regions 4 and 5 (named bulk volume 1) has a lower fracture load than the bulk volume of the remaining regions (named bulk volume 2).

For a more detailed classification of volumetric zones, we assumed that the load of fractures in a radargram is the same in a volumetric zone clipped by two imaginary planes parallel to the vertical cross section of the radargrams. The first imaginary plane is spaced half the interval between radargrams (1.0 m) in the + y direction, while the second one is in the −y direction.

Accordingly, this led to division of the bench vertically into four main volumetric zones, and eight volumetric zones were obtained in total. Each volumetric zone in bulk volume 1 has dimensions of 25.0 m × 2.0 m × 1.0 m, while the volumetric zones in bulk volume 2 have dimensions of 25.0 m × 2.0 m × 1.5 m. Through these dimensions, each volumetric zone

Figure 18. Graphical representation of the fracture load of vertical fractures (A) and horizontal fractures (B) inside the depth regions through the radargrams.

328

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314–331

3-D Fracture Modeling from GPR Survey

Figure 20. An illustrative sketch showing the evaluation-based fracture index (in red-colored text) for each volumetric zone. The locations of points A and B, in the top view of the bench surface, were graphically introduced in Figure 5. The numbers in brackets refer to the total fracture load in each volumetric zone. The green hatching lines highlight bulk volume 1, while the turquoise hatching lines highlight bulk volume 2.

can be cut into ornamental stone blocks of commercial size, since the standard dimensions of a commercial block size range between 2.0 m × 1.0 m × 0.5 m and 3.0 m × 2.0 m × 1.0 m (United Nations, 1976). We defined the evaluation-based fracture index (Figure 20) as an index of integer numbers starting from 1 and ranking the fracture load in each volumetric zone. The fracture load in a volumetric zone is the total fracture load of the depth regions inside it. An index of 1 is given to the volumetric zone with the lowest fracture load. The index number increases when the fracture load inside a volumetric zone increases. In the case where several volumetric zones have the same load of fractures, each of them has the same index.

However, the bench in this case study is un-mineable because of fractures, and the proposed evaluationbased fracture index can be used in the preliminary pre-exploitation planning for mineable benches. Particularly, by adding further technical parameters to the index, in a large survey area, the index can be used for production scheduling.

CONCLUSIONS The limitations of the proposed method are mostly the common limitations of GPR; however, the effect of these limitations can be minimized. The limitations of the proposed method can be listed as follows: (i) the surveying area should be suitable for a GPR survey: accessible, flat, with a reasonable angle of inclination of the surveyed surface; (ii) the size of the detected fractures is controlled by the antenna frequency, since with low-frequency antennas, small or closed fractures do not show up well in radargrams; (iii) antenna frequency also controls the penetration depth, where closed or small fractures can be detected with the use of highfrequency antennas but with very limited penetration depth; (iv) the operator’s experience and skill control the quality of the GPR signal processing and data interpretation; (v) GPR survey is fast, but other elements in

the methodology consume time, such as fracture modeling formulations; and (vi) the modeling assumptions used in the FISA scheme may cause some uncertainties, but results can be improved when all the detected fractures can be modeled following the FITT scheme.

Using a wide range of GPR antenna frequencies can provide a compromise solution to the problem of penetration depth and resolution. The time-consuming issue can be tackled by surveying large areas, since it is expected to consume almost the same amount of time. To increase the deterministic accuracy level, minimizing the spacing between parallel survey lines is recommended.

GPR is a successful tool for detecting fractures, even in wet and fractured rock mass conditions. In this study, the processed radargrams led to detection of fractures in a sandstone rock mass up to a depth of 2.5 m using a 400 MHz GPR antenna. Estimation of a representative dielectric constant for a rock mass in situ is a promising way to optimize the accuracy for detecting fractures at their real subsurface locations. The use of parallel GPR survey lines that cross the main sub-vertical outcropping fractures is necessary for detecting hidden surfaces of fractures.

The proposed tracing interpretation technique allowed us to successfully model fractures as 3-D surfaces. A combination of FITT and FISA formulations allowed us to build a full 3-D model for the existing fracture sets with different orientations. For a better perception of the modeled fractures in the rock mass, the geometric features of the modeled fractures could be coded in polygon file format (.PLY), which is readable for visualization software packages (e.g., ParaView).

As a validation of the GPR measurements and interpretations, a comparison of the fracture orientation with the manual method shows that the GPR measurements results are close to the manual method results. However, the comparison is limited to the dipping angles since measuring the dip direction of

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314–331

329

Elkarmoty, Colla, Gabrielli, Bondu`a, and Bruno

fractures, by the manual method, is not possible when rock faces are flat.

The proposed method of 3-D fractures modeling and visualization can be used in different rock engineering applications, when suitable GPR surveying conditions exist, not only for the purpose of evaluating and planning quarries, but also, for example, to collect structural data to be used in rock slope stability analyses and the design of underground structures. However, for some applications, due to the limitations of GPR, it may be difficult to apply the method, for example, in open pits, particularly in deep mines.

As a contribution for pre-quarrying evaluation, a frequency analysis of the detected fractures in the radargrams provided a 3-D evaluation-based fracture index for the bench. This index is a representation of the quantitative mapping of fractures; other geologic and economic factors will be considered in future research to improve this index.

Future research based on fracture modeling by the approach proposed here can target production optimization of ornamental stone quarries through developing a 3-D quarrying optimization model based on the proposed fracture modeling approach. This quarrying model can minimize waste production by searching for the best cutting orientation that maximizes the number of extracted non-fractured commercial-size blocks.

ACKNOWLEDGMENTS The authors are grateful to “La Borghigiana” quarrying company for allowing u this research to be carried out and providing the authors with all in situ needs. Special thanks go to Prof. Giuseppe Raspa (Department of Chemical, Materials and Environmental Engineering, Sapienza University of Rome) for the technical suggestions and advice given to Mohamed Elkarmoty. The authors also thank the EU-METALIC II–ERASMUS MUNDUS program, which funded the Ph.D. scholarship for the lead author, Mohamed Elkarmoty.

REFERENCES ANNAN, A. P., 2003, Ground Penetrating Radar Principles, Procedures, and Applications: Sensors & Software Inc., Mississauga, Canada.

ANNAVARAPU, S.; KEMENY, J.; AND DESSUREAULT, S., 2012, Joint spacing distributions from oriented core data: International Journal Rock Mechanics and Mining Sciences, Vol. 52, pp. 40– 45.

AQEEL, A.; ANDERSON, N.; AND MAERZ, N., 2013, Mapping subvertical discontinuities in rock cuts using a 400-MHz ground penetrating radar antenna: Arabian Journal Geosciences, Vol. 7, pp. 2093–2105.

ASSALI, P.; GRUSSENMEYER, P.; VILLEMIN, T.; POLLET, N.; AND VIGUIER, F., 2014, Surveying and modeling of rock discontinuities by terrestrial laser scanning and photogrammetry: Semi-

330

automatic approaches for linear outcrop inspection: Journal Structural Geology, Vol. 66, pp. 102–114.

AMERICAN SOCIETY FOR TESTING AND MATERIALS (ASTM): D643211, 2011, D6432-11: Standard guide for using the surface ground penetrating radar method for subsurface investigation.

In Book of ASTM Standards 04.09: American Society for Testing Materials, West Conshohocken, PA. BARGOSSI, G. M.; GAMBERINI, F.; FABBRI, S.; PEDDIS, F.; RAVAGLIA, B.; VESCOGNI, C.; MARINI, P.; AND BELLOPEDE R., 2008, Mineralogical-petrographic and physical-mechanical characterization. In: Bartolomei, A.; Montanuri, F. (Eds.), Pietra Serena, Materia Della Citta: Edizioni Aida, FIRENZE, Italy, pp. 57–84.

BOTELHO, M. A. B. AND MUFTI, I. R., 1998, Exploitation of limestone quarries in Brazil with depth migrated groundpenetrating radar data: Society of Exploration Geophysicists, Expanded Abstracts, Vol. 2, pp. 898–903.

BOURKE, P., 2011, PLY—Polygon File Format: Electronic document, available at http://paulbourke.net/dataformats/ply/ DANIELS, D. J., 2004, Ground Penetrating Radar, 2nd ed.: The Institute of Electrical Engineers, London, U.K. DELIORMANLI, A. H.; MAERZ, N. H.; AND OTOO, J., 2014, Using terrestrial 3D laser scanning and optical methods to determine orientations of discontinuities at a granite quarry: International Journal Rock Mechanics and Mining Sciences, Vol. 66, pp. 41– 48.

D´EROBERT, X. AND ABRAHAM, O., 2000, GPR and seismic imaging in a gypsum quarry: Journal Applied Geophyscis, Vol. 45, pp.

157–169.

DEZAYES, C.; VILLEMIN, T.; AND PˆECHER, A., 2000, Microfracture pattern compared to core-scale fractures in the borehole of Soultz-sous-Forˆets granite, Rhine graben, France: Journal Structural Geology, Vol. 22, pp. 723–733.

ELKARMOTY, M.; COLLA, C.; GABRIELLI, E.; BONDUA` , S.; AND BRUNO R., 2016a, Application of low frequency GPR antenna to fractures detection and 3D visualization in a new quarry bench (E-poster). In Proceedings of the International Conference on Geosciences and Geophysics, Oct. 06-07, 2016, Orlando, USA: Journal Geology and Geophysics, Vol. 5, No. 5 (Suppl), pp. 47.

ELKARMOTY, M.; BONDUA` , S.; BRUNO, R.; AND CANGIOLI S., 2016b, Three dimensional fractures detection by geo-radar for sustainable production of ornamental stones. In Proceedings of the 1st International Sustainable Stone Conference: Carrara, Italy, May 20, 2016, pp. 18.

FISHER, J. E.; SHAKOOR, A.; AND WATTS, C. F., 2014, Comparing discontinuity orientation data collected by terrestrial LiDAR and transit compass methods: Engineering Geology, Vol. 181, pp. 78–92.

GAO, M.; JIN, W.; ZHANG, R.; XIE, J.; YU, B.; AND DUAN, H., 2016, Fracture size estimation using data from multiple boreholes: International Journal Rock Mechanics and Mining Sciences, Vol. 86, pp. 29–41.

Googlemaps GPS Coordinates website, 2016, Visualization of an Address, Online, from Given Latitude and Longitude: Electronic document, available at http://www.gps-coordinates.net/ GRANDJEAN, G. AND GOURRY, J. C., 1996, GPR data processing for 3D fracture mapping in a marble quarry (Thassos, Greece): Journal Applied Geophysics, Vol. 36, pp. 19–30.

GRASMUECK, M., 1996, 3-D ground-penetrating radar applied to fracture imaging in gneiss: Geophysics, Vol. 61, pp. 1050– 1064 GRASMUECK, M.; QUINTA` , M. C.; POMAR, K.; AND EBERLI, G. P., 2013, Diffraction imaging of sub-vertical fractures and karst with full-resolution 3D ground-penetrating radar: Geophysical Prospecting, Vol. 61, pp. 907–918.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314–331

3-D Fracture Modeling from GPR Survey INTERNATIONAL SOCIETY FOR ROCK MECHANICS (ISRM), 1978, International Society for Rock Mechanics Commission on Standardization of Laboratory and Field Tests: International Journal Rock Mechanics and Mining Sciences & Geomechanics Abstracts, Vol. 15, pp. 319–368.

KADIOGLU, S., 2008, Photographing layer thicknesses and discontinuities in a marble quarry with 3D GPR visualisation: Journal Applied Geophyscis, Vol. 64, pp. 109–114.

La Borghigiana Srl website, 2016, Pietra Serena: Electronic document, available at www.laborghigiana.com/gamma/ LAI, P.; SAMSON, C.; AND BOSE, P., 2014, Visual enhancement of 3D images of rock faces for fracture mapping: International Journal Rock Mechanics and Mining Sciences, Vol. 72, pp. 325–335.

LAU, J. S. O.; AUGER, L. F.; AND BISSON, J. G., 1987, Subsurface fracture surveys using a borehole television camera and acoustic televiewer: Canadian Geotechnical Journal, Vol. 24, pp. 499– 508.

LUODES, H., 2008, Natural stone assessment with ground penetrating radar: Estonian Journal Earth Sciences, Vol. 57, pp. 149– 155.

MAERZ, N. H.; AQEEL, A. M.; AND ANDERSON, N., 2015, Measuring orientations of individual concealed sub-vertical discontinuities in sandstone rock cuts integrating ground penetrating radar and terrestrial LIDAR: Environmental and Engineering Geosciences,Vol. XXI, No. 4, pp. 293–309.

MAERZ, N. H. AND KIM, W., 2000, Potential use of ground penetrating radar in highway rock cut stability. In Proceedings of Geophysics 2000: St. Louis, MO, USA, Dec. 11–15, 2000, pp. 9.

MINEO, S.; PAPPALARDO, G.; RAPISARDA, F.; CUBITO, A.; AND DI MARIA, G., 2015, Integrated geostructural, seismic and infrared thermography surveys for the study of an unstable rock slope in the Peloritani Chain (NE Sicily): Engineering Geology, Vol. 195, pp. 225–235.

MUBIAYI, M. P., 2013, Characterisation of sandstones: Mineralogy and physical properties. In Proceedings of the World Congress on Engineering,Vol. III: London, U.K., July 3–5, 2013, pp.

2171–2176.

MYSAIAH, D.; MAHESWARI, K.; SRIHARI RAO, M.; SENTHIL KUMAR, P.; AND SESHUNARAYANA, T., 2011, Ground-penetrating radar applied to imaging sheet joints in granite bedrock: Current Science.Vol. 100, No. 4, pp. 473–475.

PARAVIEW WEBSITE, 2016, ParaView Software Program: Electronic document, available at http://www.paraview.org/.

PARK, H. J. AND WEST, T. R., 2002, Sampling bias of discontinuity orientation caused by linear sampling technique: Engineering Geology, Vol. 66, pp. 99–110. ´ , A. O. S., 2006, GPR PORSANI, J. L.; SAUCK, W. A.; AND JUNIOR for mapping fractures and as a guide for the extraction of ornamental granite from a quarry: A case study from southern Brazil: Journal Applied Geophyscis, Vol. 58, pp. 177–187.

PRIEST, S. D. AND HUDSON J. A., 1976, Discontinuity spacings in rock: International Journal Rock Mechanics and Mining Sciences & Geomechanics Abstracts, Vol. 13, pp. 135–148.

PRIEST, S. D. AND HUDSON, J. A., 1981, Estimation of discontinuity spacing and trace length using scanline surveys: International Journal Rock Mechanics and Mining Sciences & Geomechanics Abstracts, Vol. 18, No. 5, pp. 82.

REYNOLDS, J. M., 2011, An Introduction to Applied and Environmental Geophysics, 2nd ed.: John Wiley & Sons, Ltd., New York.

SEOL, S. J.; KIM, J.; SONG, Y.; AND CHUNG, S., 2001, Finding the strike direction of fractures using GPR: Geophysical Prospecting, Vol. 49, No. 3, pp. 300–308.

SEREN, A. AND ACIKGOZ, A. D., 2012, Imaging fractures in a massive limestone with ground penetrating radar, Hay-

mana, Turkey: Scientific Research and Essays, Vol. 7, No. 40, pp. 3368–3381.

SONNTAG, R.; EVANS, J. P.; LA POINTE, A.; DERAPS, M.; SISLEY, H.; AND RICHEY, D., 2012, Sedimentological controls on the fracture distribution and network development in Mesaverde Group sandstone lithofacies, Uinta Basin, UT, USA: Special Publication 374, Geological Society, London, U.K., pp. 23–50.

STAVROPOULOU, M., 2014, Discontinuity frequency and block volume distribution in rock masses: International Journal Rock Mechanics and Mining Sciences, Vol. 65, pp. 62–74.

THEUNE, U.; ROKOSH, D.; SACCHI, M. D.; AND SCHMITT, D. R., 2006, Mapping fractures with GPR: A case study from Turtle Mountain: Geophysics, Vol. 71, No.5, pp. 139–150.

TOSHIOKA, T.; TSUCHIDA, T.; AND SASAHARA, K., 1995, Application of GPR to detecting and mapping cracks in rock slopes: Journal Applied Geophyscis, Vol. 33, pp. 119–124.

TSOFLIAS, G. P.; VAN GESTEL, J.; STOFFA, P. L.; BLANKENSHIP, D. D.; AND SEN, M., 2004, Vertical fracture detection by exploiting the polarization properties of ground penetrating radar signals: Geophysics, Vol. 69, pp. 803–810.

UNITED NATIONS, 1976, The Development Potential of Dimension Stone: Department of Economic and Social Affairs, United Nations, New York, 95 p.

VASUKI, Y.; HOLDEN, E. J.; KOVESI, P.; AND MICKLETHWAITE, S., 2014, Semi-automatic mapping of geological structures using UAV-based photogrammetric data: An image analysis approach. Computers and Geosciences, Vol. 69, pp. 22–32. ¨ , H.; PERRAS, M. A.; AND WALTON, G.; LATO, M.; ANSCHUTZ DIEDERICHS, M. S., 2015, Non-invasive detection of fractures, fracture zones, and rock damage in a hard rock excavation— ¨ o¨ Hard Rock Laboratory in Sweden: Experience from the Asp Engineering Geology, Vol. 196, pp. 210–221.

WIDESS, M. B., 1973, How thin is a thin bed?: Geophysics, Vol. 38, pp. 1176–1180.

YELFM, R. J., 2007, Application of ground penetrating radar to civil and geotechnical engineering: Electromagntic Phenomena, Vol. 7, pp. 103–117.

ZAJC, M.; GOSAR, A.; AND POGACNIK, Z., 2013, Analysis of tectonic and karst formations as geological hazard for exploitation of flyschoid rocks by ground penetrating radar, the case of Anhovo-Rodez quarry (W Slovenia). In Spence, G. H.; Redfern, J.; Aguitero, R.; Bevan, T. G.; Cosgrove, J. W.; Couples, G. D.; and Daniel, J.-M. Proceedings of the 7th International Workshop on Advanced Ground Penetrating Radar, Institute of Electrical and Electronic Engineering (IEEE), Nantes, France, pp. 1–6. ˇ AND GOSAR, A., 2014, Ground penetrating ˇ , Z.; ZAJC, M.; POGACNIK radar and structural geological mapping investigation of karst and tectonic features in flyschoid rocks as geological hazard for exploitation: International Journal Rock Mechanics and Mining Sciences, Vol. 67, pp. 78–87.

ZAZOUN, R. S., 2013, Fracture density estimation from core and conventional well logs data using artificial neural networks: The Cambro-Ordovician reservoir of Mesdar oil field, Algeria: Journal African Earth Sciences, Vol. 83, pp. 55–73.

ZEEB, C.; GOMEZ-RIVAS, E.; AND BONS, P. D., 2013, Evaluation of sampling methods for fracture network characterization using outcrops: Bulletin American Association of Petroleum Geologists, Vol. 97, No. 9, pp. 1545–1566.

ZIMMER, V. L. AND SITAR, N., 2015, Detection and location of rock falls using seismic and infrasound sensors: Engineering Geology, Vol. 193, pp. 49–60.

ZOU, D. H. AND WU, Y. K., 2001, Investigation of blast-induced fracture in rock mass using reversed vertical seismic profiling: Journal of Applied Geophysics, Vol. 48, pp. 153–162.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 314–331

331

Analysis and Prediction of Gas Recovery from Abandoned Underground Coal Mines in China WEI LI Key Laboratory of Coal Methane and Fire Control, Ministry of Education, China University of Mining and Technology, Xuzhou 221116, China; Key Laboratory of Mining Engineering in Regular Institutions of Higher Education of Heilongjiang Province, Heilongjiang University of Science & Technology, Harbin 150027, China; and National Engineering Research Center for Coal & Gas Control, China University of Mining and Technology, Xuzhou 221116, China

ER-LEI SU YUANPING CHENG1 RONG ZHANG ZHENGDONG LIU Key Laboratory of Coal Methane and Fire Control, Ministry of Education, China University of Mining and Technology, Xuzhou 221116, China; and National Engineering Research Center for Coal & Gas Control, China University of Mining and Technology, Xuzhou 221116, China

PAUL L. YOUNGER School of Engineering, James Watt Building (South), University of Glasgow, Glasgow G12 8QQ, United Kingdom

DONGMING PAN School of Resource and Earth Science, China University of Mining and Technology, Xuzhou 221116, China

Key Terms: Abandoned Coal Mine, Gas Recovery, Mining, Isotope, Emission

ABSTRACT Mine closures are likely to become especially widespread in eastern China. However, because of relatively high residual coal-bed methane content, the abandoned coal mine methane (ACMM) reserves of China are huge, and from a greenhouse gas–control perspective it is preferable that they be developed and utilized rather than allowed to vent to the atmosphere. The exploitation and development of ACMM in China is still in its infancy, with theory and practice undergoing rapid development. Four factors are particularly influential in the design of ACMM recovery strategies. The first factor is what may be termed the “enrichment space,” which reflects the final state of the strata after completion of longwall extraction and subsequent strata settlement and is here defined as the region between the outer limit of stress relief and the limit of the extracted panels. Quan-

1 Corresponding

author email: ypcheng@cumt.edu.cn

titative analysis of the components of gas mixtures recovered from the enrichment space can be tracked using stable carbon isotope techniques. The second factor is the permeability field surrounding the abandoned mine voids. The thick mudstones that commonly overlie the coal seams serve to confine the water and gas within the enrichment space and old mine voids. The geometry of these confining layers can be confirmed by seismic reflection or other geophysical methods, which can reveal the extent of the zone affected by fracture development. On this basis, models of methane movement in abandoned mines can be constrained, allowing valuable predictions of availability of ACMM resources under different mining and post-closure drainage conditions.

INTRODUCTION Coal occupies the first place in China’s primary energy consumption, and more than 95 percent of the coal is won by underground mining. As of 2015 there were more than 11,000 coal mines in China, and the total annual coal production is about 3.8 billion tons (Wang et al., 2013). However, as depletion of coal resources and/or localized problems with mining

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 333–344

333

Li, Su, Cheng, Younger, Pan, Zhang, and Liu

technology and geological conditions arise there will inevitably be more and more closures of underground coal mines. By 2020, the number of closed state-owned coal mines in China is expected to be nearly 1,000; most of these mines will have high gas content. Because the coal mines in China mainly adopt underground mining, the coal recovery ratio is low (the average ratio of actual recovery is about 30 percent), so the residual coal deposits in abandoned mines are huge in number. Consequently, methane resources within mining-influenced terrain are already estimated at as much as one trillion cubic meters (Yongjun, 2005).

Currently there are abandoned coal mines all around the world: the United States has 7,582 abandoned mines, some of which host gas-extraction operations (Karacan, 2015); Britain, Germany, and Ukraine have started gas extraction and utilization in abandoned mines (Shi et al., 2006; Palchik, 2014; and Younger, 2014). To date, gas extraction in China is restricted to a few individual abandoned mining areas; as mine closures expand, there will be a need for far more extensive operations to control abandoned coal mine methane (ACMM).

After mines have been abandoned, gas will gradually accumulate in the void space of the abandoned mines.

The abandoned mines may be partially or completely flooded if they are connected to aquifers, and the degree of flooding will change over time until equilibrium water levels are attained. Only unflooded portions of the mines will emit significant quantities of methane (Younger, 2014). Uncontrolled migration of ACMM is problematic. If it migrates into operating mines, it may easily lead to exceedance of the methane limits permitted in the mine atmosphere, or even to gas explosions (Creedy, 1991). Where ACMM migrates upward via subsidence fractures and faults, it may enter the built environment at surface, where it can accumulate in confined spaces and pose an explosion hazard (Robinson, 2000; Thielemann et al., 2001; Didier, 2009; and Rudolph et al., 2010). Coal-bed methane is not only a clean energy but also a greenhouse gas, being 27 times more potent than CO2 . Therefore, through the extraction and utilization of coal-bed methane in abandoned mines, the reduction of greenhouse gas emissions and the utilization of energy resources can be simultaneously achieved. Methane hazards are also of potential relevance to other uses of abandoned mines that have been increasingly considered in recent years, such as CO2 sequestration and mine water geothermal energy recovery in the abandoned mines (Piessens and Dusar, 2006; Renz et al., 2009; Preene and Younger, 2014; and Li et al., 2015).

Although many factors affect coal seam occurrence and conditions in Chinese abandoned coal mines (such as coal seam stratigraphy, depth of burial, gas content,

334

and the geomechanical impacts of multi-seam mining techniques), there appears to be significant potential for recovery and use of ACMM in China. The earliest coal mines in China lie mainly in the east, where large-scale industrial mining has taken place for more than 100 years in some places. The number of abandoned mines is now gradually increasing in the eastern region. As shown in Figure 1, the coal-bed methane content in the eastern region of China is generally greater than 10 m3 /ton in an area with coal mining depths greater than 700 m and significant risks of coal and gas outbursts (He et al., 2005). (In contrast, in the western region of China, the depth of coal mining is relatively shallow at 300–500 m, and ACMM is correspondingly less there.)

At present, the development of coal-bed methane in abandoned mines in China is still in its infancy, and further developments in theory and practice still require further research. This article explores some of the main factors that need to be investigated to facilitate design of operations for the extraction and utilization of ACMM under Chinese conditions. This article will not consider the case of abandoned mines that are already flooded, as these will not yield significant methane without a restoration of mining space area (Younger, 2014).

METHANE OCCURRENCE IN AND AROUND ABANDONED COAL MINES In most underground coal mines in China, longwall extraction techniques are used. During longwall coal mining, the balance of stress between the overlying strata and the floor strata is disturbed, resulting in the movement and deformation of the strata, largely due to the weight of the roof rock. When the internal tensile stress of the immediate roof strata exceeds the rock strength, the strata breaks, is crushed, and falls, forming “goaf” (the American term for goaf is “gob”) in the mine void. The overlying strata tends to sag along a line normal to the bedding according to conventional bending beam or cantilever beam deflection, resulting in fractures and bed separation cavities. In this way, as shown in Figure 2, the goaf (caving) (I), fracture (II), and bending (III) zones of the coal seam roof are formed. The vertical extent of each of these zones is correlated not only to the thickness of the worked coal seam but also to the strength of the roof strata and the mining method. As the working face advances, the newly disturbed overlying strata begins to move. When the goaf reaches a certain size, rock strata disturbances can expand to the surface, resulting in shallow subsidence. As the mining working face continues to advance, the ground surface can eventually deform and form a large, closed basin (Figure 3).

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 333–344

Gas Recovery from Abandoned Underground Coal Mines

Figure 1. The distribution of coal resource and coal seam methane contents in China (Zhang and Wu, 2013). The different colors indicate different gas contents. The eastern region of China has a longer coal mining history (in Shandong, Jiangsu, and Anhui Provinces) and higher seam gas contents.

As shown in Figure 3, three zones of deformation are identified:

• In zone I, the roof of the coal seam breaks first and caves into the void resulting from coal extraction to form the goaf. The downward warping of the overlying strata then leads to compaction of the goaf. The

permeability of coal and rock in this zone is relatively poor because the mining-induced fractures inside the compression zone are closed again.

• In zone II, the fallen and broken rocks near the edge of the extracted panel are not compacted; thus, the permeability of this area remains relatively high.

In the plan view, zone II forms an approximate

Figure 2. In the early stages of longwall coal mining, following the removal of the coal, the rock mass overlying the seam is deformed and forms a caving zone (I), fracture zone (II), and bending zone (III).

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 333–344

335

Li, Su, Cheng, Younger, Pan, Zhang, and Liu

Figure 3. Characteristics of the gas enrichment space developed in and around an abandoned longwall coal panel. 1 , 2 is the mining pressure relief angle, and ␤0 ,␥ 0 is the rock strata movement angle. After mining the coal seam, areas of decreasing and concentrating stress are generated on both sides of the coal seam. Thus, swelling deformation and compressive deformation of the corresponding overlying rock strata occur, and the rock strata in the bending zone will undergo continuous bending deformation until a subsidence basin is formed at the ground surface. The shaded area in Figure 3 represents the gas enrichment space in the abandoned mine system.

“O” shape or halo around the extracted panel zone; hence, zone II is termed the “O-ring fractured zone” (Qin et al., 2015). When methane drainage boreholes target the O-ring, the best gas extraction rates can be obtained.

• In zone III, after mining the coal seam, the outermost zone of strata disturbance is a certain distance outside the mining boundary. The coal and rock mass outside the fracture line are damaged as a result of the decreasing horizontal stress resulting from the concentrated stress, and several secondary fractures are produced in the coal and rock mass that substantially increase the permeability coefficient of the coal and rock mass. Consequently, some side directions have better developed fractures within a certain range outside of the mining boundary. In the corresponding regions of the unloading boundary and within the rock strata moving boundary lines, the strata are in a complex mechanical state because they suffer from horizontal tension and vertical compression. As a result of the influences of factors such as the mechanical properties of different coal and rock seams and cover depth, the development of fractures in this zone is associated with the mechanical histories of the coal and rock seams. These factors also influence permeability, as summarized in Figure 4 below.

The relaxation of roof and floor strata allows gas to flow from all gas sources to the goaf areas and the

336

underground workings. The intensity of the gas flow depends on the type and strength of the rocks and the degree of strata relaxation. Coal extraction and subsequent caving of the roof strata create a low-stress, fractured zone with significantly increased permeability in the overlying strata and, to a lesser extent, in the underlying strata. Consequently, the permeability distribution in the affected zones around the mine workings is very heterogeneous.

Using Forster’s model (Forster and Enever, 1992) of strata deformation in the overburden areas together with descriptions of the stress and permeability distriet al., 2014), the butions (Black and Aziz, 2009; Szlazak

high-probability zone of gas migration in the fractured and constrained zones can be determined, as shown in Figure 4. In longwall mines, the zone of disturbance can be large, depending on several factors, including the dimensions of the longwall panels, depth of mining, and thickness of the coal extracted. It has been estimated that the three zones of disturbance may extend up to 150–170 m into the roof rock and 40–70 m below the seam being worked (Tauziede et al., 2002; Han et al., 2004; and Adhikary and Guo, 2015).

Vertical fractures are highly developed within the caving zone, which means that the methane in the coal seam experiences full pressure–relief, which stimulates vigorous desorption. Because it has a low density, the released methane migrates upward in the fractured zone and accumulates beneath the less permeable strata

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 333–344

Gas Recovery from Abandoned Underground Coal Mines

Figure 4. Strata deformation above a longwall panel with a degasification zone (Forster and Enever, 1992; Black and Aziz, 2009; and Szlazak

et al., 2014).

above the top section of the zone. Gas flows freely in large cracks and by seepage in tiny fractures.

Pressure relief and desorption also occur in the zone of strata bending and bed separation, but the horizontal fractures that are dominant in this zone lead to gas flow in the direction of the bedding in the form of seepage along the tensile fractures between the layers.

Consequently, the gas rarely enters the goaf, and large amounts of highly concentrated methane can accumulate in the bed separation voids.

equals zero. At this time, the methane will “float” in a large area, forming a wide zone of gas accumulation.

A large number of production practices and observations show that a wide range of gas accumulation often occurs in the upper part of the mining fractured zone.

However, given the heterogeneity of porosity in the goaf and vertically fractured zones, localized pockets of trapped methane may also occur.

The physical state of the ACMM essentially resembles the states of the gases in unmined coal seams (Figure 5) and can be divided into the adsorbed state, the

THE SOURCES AND OCCURRENCE OF GAS IN ABANDONED MINES The Provenance and States of Methane in Abandoned Mines The gas in abandoned mines is derived from multiple sources, including the residual coal inside the goaf, the adjoining coal pillars, and the adjacent panel emissions.

More recently, the possibility of microbial methanogenesis as an ongoing source of methane in abandoned ¨ coal mines has been suggested (Kruger et al., 2008).

The concentration of methane inside the goaf is the highest during its early stage of formation. However, the methane concentration is affected by diffusive motion (in response to the concentration gradient) and the flow induced by density differences with time; thus, the methane concentration in the aging goaf gradually stabilizes at a lower level. Vertically, the methane rises continuously through the pores of the goaf medium (or cracks) until it is physically blocked by low permeability or until the density difference with the ambient gas

Figure 5. Methane adsorption isotherms of Chinese coal with different ranks within 5 MPa at 30◦ C. RO is the vitrinite reflectance (percent); VL is the Langmuir volume (m3 /t); and PL is the Langmuir pressure (MPa).

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 333–344

337

Li, Su, Cheng, Younger, Pan, Zhang, and Liu

Figure 6. Gas emission curves (Noack, 1998) (a) and residual gas contents of the coal seams (b) at different distances from the working seams and with different initial gas contents typical of Chinese conditions (cf. Figure 1).

free state, and the water-soluble state. However, because of the stress relief caused by mining, the adsorbed and water-soluble fractions tend to have much lower ACMM concentrations than do the virgin coal seams, and the proportion of gas in the free state is much greater.

Thus, the residual mine voids form an impressive underground reservoir for free gas, which is equivalent to 25â&#x20AC;&#x201C;35 percent of the total volume of coal extracted, as shown by various surveys and as demonstrated during mine flooding (Tauziede et al., 2002). The backfilled and goaf areas of the formerly mined panels contain effective porosities of 3 to 8 percent (Van Tongeren and Dreesen, 2004). These investigations also show that the backfilled panels have a larger porosity than do the surrounding unmined rocks, and they also have porous pathways.

Gas Emissions from Abandoned Mines The stress relief induced by mining leads to coal expansion and extension of the cleat opening. In response to the negative atmospheric pressure caused by ventilation in the workings, coal seam gas flows into the working faces. Noack (1998) has provided empirical evidence for roof and floor gas emissions (Figure 6).

As shown in Figure 1, the residual gas contents of the coal seams were estimated at different distances from the working seams and were 20 m3 /t, 15 m3 /t, and 10 m3 /t.

Mines with relatively low active methane emission rates are likely to have less gas pressure buildup in the mine after closure and, consequently, negligible

338

emissions after closure. Conversely, mines with high emission rates during operation are likely to have high emissions after abandonment, even though the emissions will decrease with time since abandonment. These emissions can vary daily with changes in atmospheric pressure. Gas emissions coming from old mines have a positive effect just after flooding. The gas emissions of abandoned mines in the United States were previously measured, revealing a linear relationship between the gas emissions of an abandoned mine and the gas emissions of the same mine when it was operating.

When emissions occur, they are generally between 2 and 30 percent of the active mine emissions, with an average rate of 17 percent (Kirchgessner et al., 2000). The gas emissions in the Huainan and Huaibei mining areas in China were estimated based on this finding, as shown in Figure 7. These areas were chosen because the mines in these two coalfields are prone to gas outbursts, which reflect the high original gas contents of the coal seam (Wang and Cheng, 2012). According to the average value of 17 percent, the anticipated ACMM emissions in the Huainan mining area were estimated to be 20â&#x20AC;&#x201C;40 m3 /min.

Fingerprinting the Methane Emissions of Specific Seams Using Carbon Isotopes In multi-seam systems and after underground coal mine abandonment, large volumes of gas are released from coal seams and gas-bearing strata above and below the mined seam within the emission zone affected by mining. Gas is emitted from both the mined seam and the influenced zones in the roof and floor of the

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 333â&#x20AC;&#x201C;344

Gas Recovery from Abandoned Underground Coal Mines

Figure 7. Predicted methane emissions for abandoned coal mines in the Huainan and Huaibei mining areas in China based on the emission rates of working mines and estimated emissions based on the observations of Kirchgessner et al. (2000).

mined seam. Hence, the volume of emitted gas is generally much larger than the volume of gas contained in the mined seam. The amounts of emissions from un-

bon isotopic composition of coalbed CH4 varies widely and is related to the process of formation and the thermal maturation level (Clayton, 1998). Rice (1993) reported ␦13 C values for coalbed CH4 that ranged from approximately −80 to −17 . The ␦13 C values reported for CO2 in coal beds ranges from approximately −27 to + 19 (Rice et al., 1993). The differential carbon isotope ratios of CO2 and CH4 gases desorbed from coal are widely used for determining the origins of coal seam gas (Faber and Stahl, 1984). The different ␦13 C values reflect differences in sedimentary sources and diagenetic processes and changes in the coalification rank. The method of using the ␦13 C of coalbed gaseous hydrocarbon to calculate the ratios of different gas sources was developed based on this knowledge.

The carbon isotope values of the source coal and coal seam gas in a certain area are measured in advance without considering the fractionation of the gas carbon isotopes during the migration process. Therefore, according to the different characteristic values of the carbon isotopes of different gas sources, the carbon isotope values of the mixed gas can be used in the following formula to calculate the component mixture:

⎧ V1 ␦13 C1CH +V2 ␦13 C2CH +V3 ␦13 C3CH +···+Vn ␦13 CnCH 4 4 4 4 ⎪ ␦13 CCH4 mix = ⎪ ⎪ V1 +V2 +V3 +···+Vn ⎪ ⎪ ⎪ ⎪ ⎪ ␹ 1 V1 ␦13 C1C H +␹ 2 V2 ␦13 C2C H +␹ 3 V3 ␦13 C3C H +···+␹ n Vn ␦13 CnC H ⎪ 13 2 6 2 6 2 6 2 6 ⎪ ␦ C = ⎪ C H mix 2 6 ␹ 1 V1 +␹ 2 V2 +␹ 3 V3 +···+␹ n Vn ⎪ ⎨ ␥1 V1 ␦13 C1

+␥2 V2 ␦13 C2

+␥3 V3 ␦13 C3

+···+␥n Vn ␦13 Cn

C3 H8 C3 H8 C3 H8 C3 H8 ⎪ ␦13 CC3 H8 mix = ⎪ ␥1 V1 +␥2 V2 +␥3 V3 +···+␥n Vn ⎪ ⎪ ⎪ ⎪ ⎪ ␩1 V1 ␦13 C1CO +␩2 V2 ␦13 C2CO +␩3 V3 ␦13 C3CO +···+␩n Vn ␦13 CnCO ⎪ 2 2 2 2 ⎪ ␦13 CCO2 mix = ⎪ ⎪ ␩1 V1 +␩2 V2 +␩3 V3 +···+␩n Vn ⎪ ⎩ ······

mined coal seams within zones influenced by mining depend on the distance of the seams from the mined seam and the variations of the empirical model for delineating the zone that is influenced by mining. In this study, we used the geochemical properties (molecular and carbon isotopic compositions) of the main components of methane emissions from an abandoned mine to identify the gas emission rate adjacent to the mined seam.

The gaseous hydrocarbons emitted from different seams are distinct, which permits them to be used to identify the coal seam from which the components of a methane mixture originate. Coalbed gas consists of various proportions of hydrocarbons (C1 to C4 ) (C2 + from 0 to approximately 70 percent), with CO2 contents ranging from 0 to more than 99 percent. The car-

where ␦13 CCn Hm mix (and ␦13 CCO2 mix ) is the mixed gas ␦13 C value for the different gaseous hydrocarbons ( ); ␦13 C N Cn Hm (and ␦13 CnCO ) is the original 2 coalbed gas ␦13 C value for the different gaseous hydrocarbons ( ); VN is the methane volume fraction of the original coalbed (percent); and ␹ N , ␥ N , and ␩N are the different gaseous hydrocarbon volume fractions compared with the methane contents in the different original coal seam gas mixtures (percent).

Thus, when characterizing the mixed gas samples collected from the gas enrichment space of the abandoned mines, the components of the mixed gas and carbon isotope values of a variety of hydrocarbon gases (and CO2 ) were obtained. This ␦13 C method could also be used to determine the depth of the mining stress

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 333–344

339

Li, Su, Cheng, Younger, Pan, Zhang, and Liu

relief gas-emitting zone and the source of goaf gas emissions.

to spontaneous combustion of the underground coal (Song and Kuenzer, 2014).

Detecting Mine Voids That Host Gas Accumulations

GAS ENTRAPMENT IN THE ABANDONED MINES The Sealing Effect of the Overlying Mudstones Gas that accumulates in underground mine voids tends to migrate to the surface in significant quantities when the pressure within the reservoir is greater than the atmospheric pressure prevailing at the ground surface. Above abandoned longwall workings, the fracture system that develops (Figures 2 through 4) can provide a migration path to the surface, particularly where the workings are shallow and/or not confined beneath a low-permeability rock layer. In many coalfields, thick mudstones within the Coal Measures serve as a cap to the ACMM reservoir. In several coalfields in East China, such as the coalfields in Anhui, Shandong, and northern Jiangsu, a thick Quaternary clay deposit (dozens to hundreds of meters thick) overlies the coal and performs the same role.

Formed in the sedimentary system, coal-bearing series depend on the specific conditions of the cap rock. In China, coal-bearing strata are usually deposited in intercratonic basins, intracontinental depression basins, and rift-valley basins. The types of deposition include the six following systems: barrier island–lagoons and tidal flats, coastal plains, deltas, rivers, lakes, and alluvial fans. The most common cap rocks of coal are mudstone and silty mudstone, and overlying carbonate or sandstone strata occur in some areas. Mudstone and sandstone have low and high permeability, respectively. As a result of the mining depth of coal seams in the eastern region of China (up to 1,300 m), this Quaternary clay layer often lies in or above the bending zone of the mining deformation (Figure 3). Hence, the integrity and stratified structure of the clay in this region remain the same. In addition, given the great thickness and low permeability of the clay, it provides an effective cap that prevents gas from leaking from the abandoned mine. Subsidence basin lakes are often formed at the surface, which indicate that the clay layer can effectively isolate water and gas. Indeed, in waste management, a layer of only 1 m of such clay (which can have a permeability of up to k = 10−9 m/s after compaction) is sufficient for containing gas in a landfill (Robinson, 2000), as shown in Figure 8.

In West China, the mining depth of the coal seams is shallow. Thus, after mining the coal seam, the overlying coal and rock mass are within the fracture zone, and direct communication can occur between the underground goaf and the ground surface, which can lead

340

The distinct physical properties of goaf (e.g., high porosity, fluid content, compressibility) mean that it is quite different from the intact surrounding rock, which is a pre-condition for various geophysical exploration methods (Johnson et al., 2002; Gochioco et al., 2008).

The seismic exploration method is one exploration method based on the homogeneous medium geological model, which requires that each layer of the medium should be uniform, two adjacent layers of the medium should be continuous, and different media should only be considered as significantly different when a physical difference exists. In this case, the characteristics of the seismic reflection wave field received by the ground would obviously change. Through research, and based on the contrast of the reflection wave field characteristics, the occurrence, location, and form of the underground geological hazard can be deduced. Normally, the reflection coefficient of the reflected waves from coal seams is between 0.3 and 0.5, which is a significant wave impedance interface. A set of continuous strong reflection waves and regular events appears on the seismic time section. When the coal seam is excavated, the medium, form, and depth of the reflecting layer change result in the abnormal frequency, amplitude, and time of the reflection wave. The response characteristics of seismic waves in goaf are interrupted and the wave can disappear, and the energy reflectivity of the layer shows a decreasing phenomenon. Furthermore, analysis of the layer properties reveals the weakened energy reflectivity of the layer.

According to the basic principle of seismic exploration, when a seismic wave encounters the wave impedance interface, a reflection wave is generated, which provides information about the subsurface structure. Therefore, the co-phase axis of the reflected wave can be formed in the seismic signal in the peripheral portion of the goaf. When the coal seam has been removed or the seam roof has been damaged, the seismic reflection wave group will appear interrupted or will disappear in the corresponding seismic time section. In contrast, the rock strata overlying the goaf has much stronger energy absorption, dispersion, and attenuation properties; thus, the frequency of the reflected wave will decrease, and the reflected wave will disappear or its energy will become weak. The reaction to the attenuation of the seismic wave of the broken surrounding rock and fractures also shows that the waveform of the reflected wave becomes irregular, disordered, or even distorted. The location of the interface with the goaf occurs where the reflected waves in the overlying

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 333–344

Gas Recovery from Abandoned Underground Coal Mines

Figure 8. The effects of isolating water and gas of the Quaternary clay layer for abandoned mines in the eastern region of China that form ground subsidence basin lakes and provide effective capping of ACMM in old workings at depth.

rock strata appear scrambled, have low energy, and are discontinuous. These features mainly result from the development of diffraction and scattering phenomena induced by caving, collapse, and fractures developed in the overlying rock.

In addition, corresponding to the mining influence zone, the energies of the reflected waves in the rock strata in the boundary angle range of the updip and downdip directions are conspicuously reduced, appearing locally as a blank reflection band. According to these characteristics, the extent of the goaf, the development height of the mining-induced fracture zones, and the areas affected by mining can be identified.

Thus, seismic reflection techniques provide a method for detecting features that control ACMM recovery (Figure 9).

GAS RESOURCE AND EMISSION PREDICTION IN ABANDONED MINES Existing resource prediction methods include the adjacent-layer residual gas pressure method, the mass

balance method, and the empirical estimation method (Cote, 2000; Han et al., 2004). However, these methods are all based on directly related parameters such as the residual coal volume, gas content, spatial extent of the abandoned mines, etc. These factors are subject to uncertainty and/or are difficult and expensive to measure. Hence, the resulting prediction of actual resources is not accurate. Numerical reservoir simulation techniques can also be used to model abandoned mines and to predict their emissions and methane production at the regional scale. Therefore, before establishing a theoretical model to predict the amount of resources in the abandoned mines, the following parameters related to the abandoned mines should be obtained:

1) Time since abandonment; 2) The shape, size, and porosity of the methane enrichment space in and around the abandoned mines; 3) The thickness and spacing of each coal seam; and 4) The gas adsorption characteristics, residual coal seam gas pressure, and coal seam permeability.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 333â&#x20AC;&#x201C;344

341

Li, Su, Cheng, Younger, Pan, Zhang, and Liu

Figure 9. The characteristics of seismic waves in coal seams and goaf obtained using the seismic reflection method. (a) Dip direction; (b) Parallel direction.

Utilizing these data, modeling must consider multiple factors, including the gas-diffusion and seepage characteristics between the coal matrix and the cracks in the matrix (cleat and induced fractures), the dynamic impacts of gas adsorption and desorption on coal seam permeability, the influence of stress changes on coal seam permeability within the range of movement, and the deformation of rock strata. This issue is simplified in Figure 10, which only considers the case in which Darcyâ&#x20AC;&#x2122;s flow is applicable in the coal seam.

In the scope of the gas storage space and under the effect of a gas pressure gradient, there are three coalbed methane sources from three coal seams that flow into the coalbed methane enrichment space in the abandoned mines. These sources are present because the permeability of the coal mine methane enrichment spaces in abandoned mines is greater than the pre-mining permeability of the three coal seams.

By solving the model, the pressure of the coalbed methane enrichment space in the abandoned mines at different times and the distribution characteristic of

342

pressure after the flow of the coalbed methane are obtained (Figure 10). Based on the results of this model, the gas resources within the scope of the coal mine methane enrichment space in the abandoned mines can be obtained at the different times and for different drainage conditions.

When only considering Darcyâ&#x20AC;&#x2122;s flow in the coal seam and by applying the model results, the amounts of methane in the abandoned mines can be predicted. To more accurately predict the quantity of methane based on different conditions, the computational model should be improved.

One precondition of the computational model is that the abandoned mine is not flooded. However, many coal mines are filled with water soon after they are abandoned. For a coal mine in which the roof is the main source of water for filling the mine, the aquifer leakage in the affecting range of the transmission fissure zone during mining should be investigated first, and grouting and plugging should be conducted where water inflows from the roof. Considering China as an

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 333â&#x20AC;&#x201C;344

Gas Recovery from Abandoned Underground Coal Mines

Figure 10. The gas pressure distributions of three coalbeds and an abandoned coal mine 1,000 days after abandonment. a, b, and c represent the different coalbeds, and d represents the gas-enriched area affected by mining in the lowermost seam.

example, for a coal mine in which floor water fills the mine, emphasis should be placed on the thin and thick limestone on the floor of the mine, and the distance between the aquifer and mining space should be analyzed. In addition, the current water inflow sites should be sealed, and grouting reinforcement should be conducted in the areas with high inflow risk, such as faults and collapsed columns. For surface lakes, rivers, reservoirs, seawater, etc., emphasis should be placed on evaluating the long-term stability of the protective seam to prevent water from inflowing into the coal mine because of the failure of the protective seam.

CONCLUSION AND SUMMARY Over time, after longwall mechanized mining the stress balance between the overlying strata and floor strata is destroyed, resulting in strata deformation. This gives rise to a “gas enrichment space” in and around abandoned mines, delimited by the region between the stress-decreasing boundary line under the influence of mining (i.e., the fractured area of O-ring) and the fracture line.

The gas found in this gas enrichment space occurs mainly in the free state, and the amount of gas emission from each coal seam can be quantified using carbon isotope techniques.

Once abandoned mines have been closed it can be determined whether the ACMM can be exploited. In the eastern region of China, the extra-thick clay layer overlying the coal seam roof that lies in the bending zone of strata movement and deformation serves to isolate water and gas. Based on the current situation of China’s abandoned coal mines and ACMM

resources, this article has presented a preliminary analysis of coal mine methane enrichment space, gas occurrence, and gas state characteristics that should be useful in determining where the collection of ACMM gas is practicable.

ACKNOWLEDGMENTS This research was financially supported by the Fundamental Research Funds for the Central Universities (grant 2015XKMS006), the National Science Foundation of China (grant 51574229), and a project funded by the priority academic development program of Jiangsu Higher Education Institutions.

REFERENCES ADHIKARY, D. AND GUO, H., 2015, Modelling of longwall mininginduced strata permeability change: Rock Mechanics Rock Engineering, Vol. 48, pp. 345–359.

BLACK, D. J. AND AZIZ, N., 2009, Developments in coal mine methane drainage and utilisation in Australia. In Proceedings of the Ninth International Mine Ventilation Congress, pp. 1–12.

CLAYTON, J., 1998, Geochemistry of coalbed gas—A review: International Journal Coal Geology, Vol. 35, pp. 159–173.

COTE, M. M., 2000, Abandoned Coal Mine Emissions Estimation Methodology: U.S. EPA Report.

CREEDY, D., 1991, An introduction to geological aspects of methane occurrence and control in British deep coal mines: Quarterly Journal Engineering Geology Hydrogeology, Vol. 24, pp. 209– 220.

DIDIER, C., 2009, Postmining management in France: Situation and perspectives: Risk Analysis, Vol. 29, pp. 1347–1354.

FABER, E. AND STAHL, W., 1984, Geochemical surface exploration for hydrocarbons in North Sea: AAPG Bulletin, Vol. 68, pp.

363–386.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 333–344

343

Li, Su, Cheng, Younger, Pan, Zhang, and Liu FORSTER, I. AND ENEVER, J., 1992, Hydrogeological response of overburden strata to underground mining: Office Energy Report, Vol. 1, p. 104.

GOCHIOCO, L. M.; MILLER, T.; AND RUEV, F., JR., 2008, Highresolution 2D surface seismic reflection survey to detect abandoned old coal mine works to improve mine safety: Leading Edge, Vol. 27, pp. 80–86.

HAN, B.; ZHANG, X.; AND ZHANG, Q., 2004, Theoretical study on calculation limits of CBM resource of abandoned coal mine: Coal Geology Exploration, Vol. 1, p. 9.

HE, M. C.; XIE, H.; PENG, S.; AND JIANG, Y., 2005, Study on rock mechanics in deep mining engineering: Chinese Journal Rock Mechanics Engineering, Vol. 24, pp. 2803–2813.

JOHNSON, W. J.; SNOW, R.; AND CLARK, J. C., 2002, Surface geophysical methods for deduction of underground mine workings. In Symposium on Geotechnical Methods for Mine Mapping Verifications, Charleston, WV. ¨ 2015, Modeling and analysis of gas capture from KARACAN, C. O., sealed sections of abandoned coal mines: International Journal Coal Geology, Vol. 138, pp. 30–41.

KIRCHGESSNER, D. A.; PICCOT, S. D.; AND MASEMORE, S. S., 2000, An improved inventory of methane emissions from coal mining in the United States: Journal Air Waste Management Association, Vol. 50, pp. 1904–1919. ¨ , M.; BECKMANN, S.; ENGELEN, B.; THIELEMANN, T.; KRUGER CRAMER, B.; SCHIPPERS, A.; AND CYPIONKA, H., 2008, Microbial methane formation from hard coal and timber in an abandoned coal mine: Geomicrobiology Journal, Vol. 25, pp.

315–321.

LI, W.; YOUNGER, P. L.; CHENG, Y.; ZHANG, B.; ZHOU, H.; LIU, Q.; DAI, T.; KONG, S.; JIN, K.; AND YANG, Q., 2015, Addressing the CO2 emissions of the world’s largest coal producer and consumer: Lessons from the Haishiwan Coalfield, China: Energy.

Vol. 80, pp. 400–413.

NOACK, K., 1998, Control of gas emissions in underground coal mines: International Journal Coal Geology, Vol. 35, pp. 57–82.

PALCHIK, V., 2014, Time-dependent methane emission from vertical prospecting boreholes drilled to abandoned mine workings at a shallow depth: International Journal Rock Mechanics Mining Sciences, Vol. 72, pp. 1–7.

PIESSENS, K. AND DUSAR, M., 2006, Feasibility of CO2 sequestration in abandoned coal mines in Belgium: Geologica Belgica, Vol. 7, pp. 3–4.

PREENE, M. AND YOUNGER, P., 2014, Can you take the heat?— Geothermal energy in mining: Mining Technology, Vol. 123, pp. 107–118.

QIN, Z.; YUAN, L.; GUO, H.; AND QU, Q., 2015, Investigation of longwall goaf gas flows and borehole drainage performance by CFD simulation: International Journal Coal Geology, Vol. 150, pp. 51–63. ¨ ¨ , W.; SCHATZL , P.; AND DIERSCH, H.-J., 2009, RENZ, A.; RUHAAK Numerical modeling of geothermal use of mine water: Chal-

344

lenges and examples: Mine Water Environment, Vol. 28, pp.

2–14.

RICE, D.; LAW, B.; AND CLAYTON, J., 1993, Coalbed Gas—An Undeveloped Resource: United States Geological Survey, Professional Paper 1570.

RICE, D. D., 1993, Composition and origins of coalbed gas. Hydrocarbons from coal: AAPG Studies Geology, Vol. 38, pp.

159–184.

ROBINSON, R., 2000, Mine gas hazards in the surface environment: Mining Technology, Vol. 109, pp. 228–236.

RUDOLPH, T.; MELCHERS, C.; MINKE, A.; AND COLDEWEY, W. G., 2010, Gas seepages in Germany: Revisited subsurface permeabilities in the German mining district: AAPG Bulletin, Vol. 94, pp. 847–867.

SHI, J. Q.; DURUCAN, S.; AND KORRE, A., 2006, Numerical modelling and prediction of abandoned mine methane recovery: Field application at the Saar coalfield, Germany: Geologica Belgica, Vol. 7, pp. 207–213.

SONG, Z. AND KUENZER, C., 2014, Coal fires in China over the last decade: A comprehensive review: International Journal Coal Geology, Vol. 133, pp. 72–99. , N.; OBRACAJ, D.; AND SWOLKIEN´ , J., 2014, Methane SZLAZAK

drainage from roof strata using an overlying drainage gallery: International Journal Coal Geology, Vol. 136, pp. 99–115.

TAUZIEDE, C.; POKRYSZKA, Z.; AND BARRIE` RE, J.-P., 2002, Risk assessment of surface emission of gas from abandoned coal mines in France and techniques of prevention: Mining Technology, Vol. 111, pp. 192–196.

THIELEMANN, T.; KROOSS, B. M.; LITTKE, R.; AND WELTE, D. H., 2001, Does coal mining induce methane emissions through the lithosphere/atmosphere boundary in the Ruhr Basin, Germany?: Journal Geochemical Exploration, Vol. 74, pp. 219–231.

VAN TONGEREN, P. AND DREESEN, R., 2004, Residual space volumes in abandoned coal mines of the Belgian Campine basin and possibilities for use: Geologica Belgica, Vol. 7, pp. 157–164.

WANG, J.; FENG, L.; AND TVERBERG, G. E., 2013, An analysis of China’s coal supply and its impact on China’s future economic growth: Energy Policy, Vol. 57, pp. 542–551.

WANG, L. AND CHENG, Y.-P., 2012, Drainage and utilization of Chinese coal mine methane with a coal–methane co-exploitation model: Analysis and projections: Resources Policy, Vol. 37, pp.

315–321.

YONGJUN, C., 2005, Abandoned coal mine methane—A noteworthy CBM resource: China Coalbed Methane, Vol. 3, p. 6.

YOUNGER, P. L., 2014, Hydrogeological challenges in a low-carbon economy: Quarterly Journal Engineering Geology Hydrogeology, Vol. 47, pp. 7–27.

ZHANG, Z. AND WU, Y., 2013, Tectonic-level-control rule and areadividing of coalmine gas occurrence in China: Earth Science Frontiers, Vol. 2, pp. 237–245.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 333–344

Optimizing Digital Elevation Model Resolution Inputs and Number of Stream Gauges in Geographic Information System Predictions of Flood Inundation: A Case Study along the Illinois River, USA ANAS RABIE1,2 ERIC PETERSON3 JOHN KOSTELNICK1 REX ROWLEY1 Illinois State University, Department of Geography-Geology, Campus Box 4400, Normal, IL 61790-4400

Key Terms: Hydrology, Geographic Information System (GIS), Spatial Analysis, Flood Hazards, River Management ABSTRACT Spatial analysis using Geographic Information Systems (GISs) is evaluated for its ability to predict the potential hazard of a flood event in the Illinois River region in the state of Illinois. The data employed in the analysis are available to the public from trusted organizations such as the Illinois State Geological Survey and the U.S. Geological Survey. Since available GIS data may be limited for flood risk modeling in some parts of the world, the purposes of this study are to examine the use of spatial analysis in a GIS to determine flood inundation risk and to produce an accurate flood inundation vulnerability map employing the least amount of data. This study concentrates on areas that have stream gauge data with definable flood stage(s) and utilizes the inverse distance weighted interpolation method on different digital elevation models (DEMs) with different spatial resolutions (1 m, 10 m, and 30 m) to determine the extent of flooding over the study area. Resulting maps created for the Illinois River region yielded about 80 percent agreement with the effects of an actual flood event on the Illinois River near Peoria, IL, on April 23, 2013. A four-gauge distribution scenario using a 10-m DEM produced the most accurate results, but all scenarios generated reasonable flood simulation. Thus, we speculate that it is possible to create a flood prediction map with a reasonable amount of accuracy using only two initial input data layers: stream gauges and a DEM. 1 Emails: arabie@kau.edu.sa; jckoste@ilstu.edu; rjrowley@ilstu.edu 2 Present address: Indiana University, Department of Earth and

Atmospheric Sciences, 1001 East 10th Street, Bloomington, IN 47405-1405.

3 Corresponding author email: ewpeter@ilstu.edu.

INTRODUCTION Rainfall and runoff gauges are not readily available for every river system, which affects the availability and credibility of hydrological data. Vigorous urbanization of areas coupled with temporal and spatial variation in hydrological characteristics makes the quantitative assessment of runoff characteristics in most areas unattainable (El-Hames and Richards, 1998). Disasters due to natural hazards are subject to many types of uncertainty, which complicates how these disasters are predicted and represented on maps and geovisualizations (Kostelnick et al., 2013). For example, natural variability of streamflow and uncertainty of even “best available” elevation data create ambiguity in defining the floodplain boundary for flood hazard maps.

The term “flood risk” indicates the perceived or actual exposure to loss from a river flooding event during a natural disaster. The level of risk depends on the natural disaster’s overall impact on human lives and/or the economy (Safaripour et al., 2012). In order to identify that risk, however, accurate maps showing potential inundation (hazard) are required. Simple maps depicting floodwater distribution allowing real-time and rapid simulations, which can be considered “an effective real-time flood modeling and prediction system” (Al-Sabhan et al., 2003), could give decision makers an understanding of the threatened areas.

For thorough flood modeling to be successful, many models require detailed information, including discharge, precipitation, ambient soil water content, land use, evaporation intensity, watershed infiltration, and the geology and geomorphology of the area. Each of the factors affects the others significantly, and their complex relationship affects the stream runoff. To create an accurate hydrological model, a good grasp of the interaction between such factors is manda-

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 345–357

345

Rabie, Peterson, Kostelnick, and Rowley

tory (Kia et al., 2012). However, the limitations of available and reliable data constrain flood modeling in many locations and physical environments around the world. This article postulates that a limited, but critical, set of data may be used in a Geographic Information System (GIS) to generate flood hazard maps with a reasonable level of accuracy. Additionally, we evaluate the use of such data in creating potential inundation scenarios at various scales, or spatial resolutions.

Flood modeling with GIS first requires the selection of suitable input data, including a Digital Elevation Model (DEM) to represent surface topography.

DEMs are subject to several potential sources of error and uncertainty, including methods for how elevation values are generated and interpolation methods used to derive a DEM from these elevation values (Fisher and Tate, 2006). Many DEM sources exist today for flood modeling, including those created through Light Detection and Ranging (LiDAR) (typically 1- to 3-m resolution) and those created through other methods (typically 10â&#x20AC;&#x201C;30-m resolution or coarser). Resolution will have an effect on point-specific and topographic attributes (Deng et al., 2008). A comparison among the different resolutions of DEMs is necessary to enrich the understanding of any value added given the high expense in acquiring and processing LiDAR data sets (Galzki et al., 2008), especially in parts of the world where LiDAR data are not readily available for flood modeling.

Studies in various environmental applications have found mixed results regarding comparisons of GIS analyses with DEMs at varying spatial resolutions and derived through different elevation sources. For example, Jacoby et al. (2013) concluded that 10-m DEMs were noticeably beneficial for delineating geomorphic features such as cave levels when compared to 30-m DEMs. However, in the context of coastal inundation predictions for sea level rise, Kostelnick et al. (2013) found comparable results in predicted inundation for analyses in coastal Maine for both 10-m and 30-m DEMs from the National Elevation Dataset, which they attributed to similar source elevation data sources for both DEMs (see Gesch [2007]). In another sea level rise inundation study, Gesch (2009) compared four DEMs of varying accuracy and resolution and found significant improvement of predicted inundations derived from DEMs generated from LiDAR compared to those that were not. In contrast, Schroeder et al.

(2015) reported no differences in stream profiles generated from 1-m and 3-m DEMs created from the same LiDAR data. An added, and sometimes overlooked, advantage of LiDAR DEMs in the context of hydrologic modeling is their ability to account for features such as levees and drainage canals that may im-

346

pact predicted inundation extents (Poulter and Halpin, 2008).

The present study provides a demonstration of the viability of creating a flood hazard map using stream gauge spacing and DEMs as the two significant components in determining vulnerability and inundation using a GIS platform. Our primary purpose is to evaluate the minimum amount of data required to produce an accurate GIS model by comparing the accuracy of flood-model results generated from different combinations of stream gauges and DEM resolutions. In other words, we generate several predictive flood models with different DEM and stream gauge input data sources and then systematically reduce and modify input data until the minimum data required to generate an acceptable accuracy level is reached. Validity of the individual models is tested by comparing the predicted flooded areas to actual flooded areas through an accuracy assessment. In short, the approach taken in this study illustrates the development of a methodology for a rapid, easy-to-use, and cost-effective means for implementing flood hazard models. The developed models are practical and can be applied to a wide variety of scenarios for which flood hazards data may be limited.

MATERIALS AND METHODS Study Area The study incorporated a portion of the Illinois River, a major tributary of the Mississippi River (Lian et al., 2010), in the State of Illinois (Figure 1). The area was chosen because of the availability of all the needed data, specifically DEMs at different spatial resolutions, a network of stream gauges with predefined flood levels (Table 1), and satellite data for a flood event that inundated the area in spring 2013. The examined segment of the Illinois River has a length approaching 220 km and drainage of roughly 36,350 km2 . Geographically, the study area along the Illinois River extends from Cass and Schuyler Counties in the south to La Salle County in the north. At the southern end of the region, floodplain deposits (alluvium) dominate the bottomlands, with a width ranging from about 5 to 6 km.

This belt covers the southern banks of the Illinois River (Worthen, 1868). The southernmost parts are prairies that have thin wood belts skirting the channel. To the north, broken, hilly bluffs run parallel to the river. Humans have heavily modified the Illinois River watershed to support agriculture and urban growth (Akanbi and Singh, 1997). The investigated segment resides in the lower Illinois River, where the floodplain is used for agriculture. To preserve a suitable water depth for ship navigation, seven locks and dams were built on

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 345â&#x20AC;&#x201C;357

GIS Predictions of Flood Inundation

Figure 1. Map of the Illinois River study area showing the stream gauges used in the different distribution scenarios. Map by Anas Rabie.

the Illinois River. Levees and drainage constructions are present that influence the degree of flooding (Lian et al., 2010), which may also have an impact on floodplain modeling.

GIS Data Use GIS data were collected from different online and no-cost data sources that provided hydrographical,

Table 1. U.S. Army Corps of Engineers (USACE) and National Weather Service (NWS) stream gauges positioned along the Illinois River and incorporated in this study.

Station Name

Code

River Mile*

Latitude (◦ N)

Longitude (◦ W)

Flood Stage (m.a.s.l.)

Illinois River—LaSalle Illinois River—Henry Illinois River—Peoria Illinois River—Peoria Lock and Dam Illinois River—Havana Illinois River—Beardstown

lsli2 hnyi2 piai2 prai2 havi2 beai2

224.7 196 164.6 157.9 119.6 88.6

41.323611 41.107222 40.702222 40.633333 40.292778 40.020278

89.110833 89.356111 89.564444 89.625000 90.068611 90.436667

137.2 136.8 136.1 136.2 133.6 132.5

m.a.s.l. = meters above sea level. * Miles above the mouth of the Illinois River.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 345–357

347

Rabie, Peterson, Kostelnick, and Rowley

Figure 2. The three DEMs at different spatial resolutions (1 m, 10 m, and 30 m) used in the analysis. The square area highlights the location of the inset DEMs and is presented in Figure 8.

topographical, and related data for the study area.

DEMs for the study area were acquired from the U.S. Geological Survey (USGS) National Map Viewer with different resolutions—30 m, 10 m, and 1 m—or with the native resolution of 1 arc second, 1/3 arc second, and 1/9 arc second, respectively. Hydrography data, including polygon water bodies and flow paths as lines, were obtained from the National Hydrography

348

Dataset. Prior to spatially analyzing the data layers, the coordinate systems of the different layers were converted to Universal Transverse Mercator to minimize areal distortion from the map projection for all area calculations. Furthermore, a seamless DEM was created for each of the three different resolutions (Figure 2).

Stream gauges maintained by the U.S. Army Corps of Engineers (USACE) and monitored by the National

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 345–357

GIS Predictions of Flood Inundation

Weather Service (NWS) (http://www.weather.gov) provide continuous stage data, along with metadata including longitude and latitude coordinates and flood stage levels for the Illinois River system (Table 1 and Figure 1). Six stream gauges along the Illinois River were incorporated into the study: at La Salle in La Salle County, at Henry in Marshall County, two stream gauges east and southeast of Peoria in Peoria County, at Havana in Mason County, and at Beardstown in Cass County (Table 1).

For the accuracy assessment, Landsat 8 Operational Land Imager (OLI) data were acquired from the USGS Global Visualization Viewer (GLOVIS). Between April 18 and May 16, 2013, the Illinois River experienced a significant flood event (Figure 3). Peak flooding occurred on April 23, but as a result of the 16-day satellite revisit periods, available Landsat imagery tiles were from April 29 for the central and northeastern portions of the study area and from April 20 for the southwestern portion of the study area (Figure 3). Landsat 8 was selected because it provided the only no-cost optical imagery available for the time period of the flood.

Neither of the Landsat 8 image dates is optimal: The April 29 images for the central and northeastern portions of the study area occurred 6 days following the peak event, whereas the April 20 images were occurred 3 days before the flood peak for the southwestern portion (Figure 4). Furthermore, the April 20 images had over 33 percent cloud coverage, which increased error in the image classification. The accuracy assess-

Figure 3. Landsat 8 OLI images as they appear on GLOVIS (February 2014). The study area is shaded in blue.

ment model simulated conditions for April 29 using the stage levels provided by NWS for each stream gauge to coincide with the Landsat images of April 29, 2013.

This provided uniformity but, of course, restricted our

Figure 4. Hydrograph of gauge Peoria (PIAI2). Highlighted dates correspond to the available Landsat imagery. The peak stage drops from 139.14 meters above sea level (m.a.s.l.) on April 23 to 138.99 m.a.s.l. on April 29.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 345â&#x20AC;&#x201C;357

349

Rabie, Peterson, Kostelnick, and Rowley

accuracy assessment to only that portion of the study area covered by the April 29 images (see Figure 3). The Landsat 8 OLI imagery, with a spatial resolution of 30 by 30 m (USGS, 2014), was merged into a single image to generate a map of the “actual” flood extent for ground-truthing purposes. The data were classified using a supervised classification approach (where the user creates “training sites” to specify the desired land cover classes) to delineate the locations that were inundated from those that were not inundated. Then, the water associated with the Illinois River, which in this case is defined by the actual water pixels on the day of interest, was isolated from the Landsat reclassified imagery.

The study uses five different stream gauge distributions (Table 2). For each of the five gauge distributions, flood prediction interpolations were generated for the three different DEM resolutions (30 m, 10 m, and 1 m), resulting in a total of 15 different scenarios. The distribution of the stream gauges in each scenario was selected for optimal geographic spacing. For instance, the two–stream gauge distribution used gauges “hnyi2” and “havi2.” This decision was made to avoid using the two stream gauges on the edges or the two in the middle that minimize additional errors that may have resulted because of the interpolation process.

Stream-stage data, site name, longitude, and latitude for each gauge station were organized into a table, and five different data sets were created to accommodate the different stream gauge distribution scenarios (Table 2).

The data sets were imported into ESRI’s ArcGIS and plotted as point locations. The working environment in the GIS model was set so that the results would have the same extent of the Illinois River portion used in this study and to have the same cell size of the desired DEM resolution (30 × 30 m; 10 × 10 m; 1 × 1 m).

Inundation simulations were conducted for the entire study area; however, the accuracy assessment was conducted only on the portion covered by the April 29, 2013, Landsat images.

Simulation Procedure Figure 5 presents a flow chart of procedures used in this study. A series of raster-based GIS analyses were

Figure 5. Flow chart of the methodology used to develop the flood hazard model.

used to predict areas affected by a flood informed by stream gauge data. Potential inundation surfaces were interpolated for each DEM resolution from the stream stage data using an inverse distance weighted (IDW) technique. The IDW interpolation determines raster cell values using a linearly weighted combination of a set of sample points. IDW interpolation is based upon an assumption that the modeled variable loses influence with distance from data locations (Watson and Philip, 1985). IDW was used to predict water levels at a particular flood stage between gauge locations along the river.

Table 2. Breakdown of stream gauges used for the different distribution scenarios.

Scenario 6-Gauge 5-Gauge 4-Gauge 3-Gauge 2-Gauge

350

La Salle lsli2

Henry hnyi2

Peoria piai2

Peoria Lock and Dam prai2

Havana havi2

Beardstown beai2

X X X X

X X X

X

X X X X

X X

X X X X

X

X

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 345–357

GIS Predictions of Flood Inundation Table 3. Flood hazard simulation results.

Landsat Imagery

Simulation Prediction

Condition

Score

Condition

Score

Calculated Score

Result

Not flooded Not flooded Flooded Flooded

0 0 10 10

Not flooded Flooded Not flooded Flooded

0 1 0 1

0 1 10 11

Agreement Overestimation Underestimation Agreement

The Map Algebra Raster Calculator tool in ArcGIS was used to predict inundation at the different DEM resolutions with the different stream gauge distributions. This was done using an expression that subtracts the IDW results from the desired DEM, providing a flood prediction based on the stream gauge information where any positive value is “Flooded” and any negative value is “Not Flooded.” To make the raster calculator results easier to compare visually, the results were reclassified using the Spatial Analyst Reclassify tool.

Generating Comparable Results An accuracy assessment of the projected inundation from the DEM was performed by comparing results to the actual flood extents, as extracted from Landsat images using ERDAS IMAGINE Supervised Image Classification, based on over 40 training sites on each image. Areas defined as water were designated the numerical score of “10” and those with no water were assigned the score of “0” (Table 3). The results of the simulation prediction were also reclassified into two groups. “Not Flooded” areas were assigned the score of “0” and “Flooded” areas a score of “1.”

The two layers were evaluated to determine the agreement in areas designated “Flooded” and “Not Flooded” from the different DEM resolutions by using an expression that simply added the pixel scores of the two reclassifications. The new calculated values indicate either agreement between the simulation and the Landsat image classification (a pixel value of either 11 or 0 reflecting a match between predicted and actual flooding or not, respectively) or a disagreement (values of either 1 or 10 reflecting a mismatch between predicted and actual flooding or not, respectively) (Table 3).

RESULTS Fifteen flood scenarios were generated, allowing for the comparison of actual flooded areas to those predicted to be inundated on April 29, 2013, for three DEM resolutions and five stream gauge spacings. For

each scenario, we computed areas and percentages associated with agreement (flooded and not flooded, in both actual and predicted) and disagreements (overestimation and underestimation) (Table 4). Each of the 15 scenarios exhibited flood inundation agreement over 70 percent of the area simulated.

While the two–stream gauge scenarios produce the lowest overestimation error, at nearly 9 percent for all DEM resolution, the scenarios consistently produced the highest total error (between 29.5 and 29.7 percent, depending on resolution) because of much larger underestimation errors compared to those associated with other stream gauge scenarios (all DEM resolutions around 21 percent) (Figure 6). Scenarios with three, four, five, and six gauges all reported largely similar total error percentages between 22 and 23 percent, depending on DEM resolution, each with relatively consistent over- and underestimation errors (Table 4 and Figure 7). Locations farther from the employed gauges had higher error than did areas near the used gauges, a function of the imperfections in the spatial interpolation process. For every scenario, the area of disagreement for underestimation of potential flood was greater than the area for overestimation (Table 4 and Figure 7b and c).

DISCUSSION Underestimation or Overestimation In planning flood mitigation and resiliency measures, it is, of course, crucial to have accurate models available.

When such accuracy may not be possible as a result of lack of data or data precision, then an overestimation of hazard may be preferable to an underestimate.

It is true that overestimations may prompt planners to believe that some mitigation measures will be costeffective when they may be unnecessary, but in general the costs of under-preparedness are much higher than being over-prepared, especially when dealing with the consequences of avoidable events (Rosner et al., 2014).

The stream gauge scenarios with more than two gauges generated the highest overestimation and the lowest underestimation consistently among the different

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 345–357

351

Rabie, Peterson, Kostelnick, and Rowley Table 4. Flood hazard simulation disagreements (in km2 ) and error percentages, with best models highlighted in bold. Total area of the study location is 610.62 km2 .

Overestimate Number of Gauges* DEM resolution, 30 m 2 3 4 5 6 DEM resolution, 10 m 2 3 4 5 6 DEM resolution, 1 m 2 3 4 5 6

Underestimate

km2

%

km2

%

Total Disagreement (km2 )

Total Error (%)

53.42 58.83 58.84 58.28 58.24

8.75 9.63 9.64 9.54 9.53

127.92 78.97 76.75 80.80 82.33

20.95 12.93 12.57 13.23 13.48

181.36 137.80 135.60 139.08 140.57

29.70 22.57 22.21 22.78 23.02

53.21 58.61 58.64 58.07 58.06

8.72 9.59 9.60 9.51 9.50

126.63 76.19 74.08 78.02 79.44

20.74 12.47 12.13 12.77 13.01

179.84 134.81 132.71 136.09 137.51

29.45 22.08 21.73 22.29 22.52

53.32 58.73 58.75 58.20 58.19

8.73 9.61 9.62 9.53 9.53

126.65 76.24 74.13 78.04 79.46

20.74 12.49 12.14 12.78 13.01

179.98 134.99 132.89 136.24 137.66

29.47 22.11 21.76 22.31 22.54

DEM = Digital Elevation Model. * See Table 2 for distribution of gauges.

DEM comparisons, with the four-gauge scenario yielding slightly lower error (Figure 7b). We hypothesize that such a result may be connected to the locations and/or the distances between those stream gauges in the combination of stream gauges used in that scenario.

One possibility for the four–stream gauge density scenario yielding lower error is that the IDW interpolation method used in the study area’s settings has a threshold that was reached with four gauges, and the results do not necessarily improve with more gauges.

Visual comparison of the extent of the predicted floods (Figure 6) and the error data (Table 4) suggests that no apparent difference exists between the different stream gauge distributions or with use of different DEM resolutions. Underestimation is seen across the edges of the river, whereas overestimation is clustered in the southwest end as well as the northeast end of the Illinois River, which may represent edge effect error.

Although error estimations across the DEM resolutions were relatively consistent, a closer examination of the role of DEM resolution reveals a larger difference between the 30-m DEM and 10-m DEM resolutions than between the 10-m and 1-m DEM resolutions (Figures 7b and 7c). The results are similar to those reported by Jacoby et al. (2013) in that the 10m DEM generated lower error and was more accurate than were 30-m DEMs for modeling cave entrances.

While the use of the 10-m DEM consistently generated

352

models with marginally lower error than those derived from the 1-m DEM, the individual differences among the various gauge distribution scenarios were too small to have significance in terms of the results. A simple quantitative comparison between the results of 10-m DEM models and 1-m DEM models shows that the average difference in this point-to-point comparison is an area of 0.16 km2 , or only 0.03 percent. Generally, higher resolution data provide better sampling and improve the elevation delineations (Hammer et al., 1995; Zhang et al., 1999). However, the differences between 1-m DEMs and more coarse DEMs has not always led to improved accuracy in the models here. For example, Schroeder et al. (2015) found that streambed elevations generated from 1-m and 3-m DEMs were alike.

For this work, the difference between 10-m and 1-m DEM–generated models is found in a release note provided by the USGS regarding the National Elevation Dataset (NED). The NED metadata states that the study area extent is within the Missouri and Mississippi River Basin flood project for the USACE for the Upper Midwest and Plains States, which lasted from 1997 to 2001, using 1/9 arc second (3-m) NED. Later on, the 3-m LiDAR DEMs were used to create the 10-m and 1-m DEM resolutions (Gesch et al., 2002).

The re-sampling explains why the results between the two DEM resolutions were comparable to each other: they are derived from the same elevation source. Hence,

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 345–357

GIS Predictions of Flood Inundation

Figure 6. Flood hazard simulation map for the 10-m DEM and selected stream gauge scenarios, where (a) shows the results at La Salle, (b) shows the results at Henry, and (c) shows the results at Peoria Lock and Dam. Refer to Figure 1 for stream gauge locations. Maps by Anas Rabie.

these results suggest that the source elevation data may be more important than merely the spatial resolution of the DEM. Alternatively, two different versions of standard 30-m DEM modules are available in the United States: “Level 1” and “Level 2.” The 30-m DEM used in this study is a “Level 1” 30-m DEM, which was derived from 7.5 U.S. Topo maps, also created by the USGS. “Level 2” 30-m DEMs are derived from 1/3 arc second DEMs, which are usually 10-m DEMs. This different elevation source explains, at least in part, why the 30-m DEM used in this study yielded results dif-

ferent from the other DEM resolutions. However, the difference between the 10-m DEM and the 1-m DEM simulations is not significant (Table 3). Our results do not prioritize the choice of DEM resolution to employ when generating inundation maps in areas with similar characteristics to the study locale. Despite the fact that the 30-m DEM data are not as accurate as the LiDARderived 10-m and 1-m DEM data, the differences in predicted flood areas among the resolutions is small.

Considering the cost of acquiring LiDAR data and the longer processing times required for the larger file sizes

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 345–357

353

Rabie, Peterson, Kostelnick, and Rowley

Figure 7. (a) Total disagreement, in percent, for the various stream gauge scenarios for all 15 gauge scenarios. (b) Overestimation and (c) underestimation in the simulated flood hazard errors for the 30-m DEM (black circle), 10-m DEM (dark gray square), and 1-m DEM (light gray triangle).

(5 minutes for 30-m, 12.5 minutes for 10-m, and 38.5 minutes for 1-m DEMs in this study), the results from this study suggest that the 10-m DEM resolution is the preferred resolution for the generation of the inundation maps. The 10-m DEM models produced less error than did the 30-m DEM in one-third the processing time of the 1-m DEM. Given that the model output and error analyses show the 1-m and the 10-m results are very similar, the spatial resolution is not a significant control if the source elevation data (LiDAR) are the same for both DEMs, as is the case in the study in hand, or for an area with similar characteristics to those of the study area. Thus, this work supports the conclusion by Galzki et al. (2008) that lower resolution DEMs are preferred over the 1-m DEM as a result of the computational requirements and lack of availability of data in some areas.

354

Focus on Disagreement It is important to realize that the model in this study simulated the flood based solely on elevation and does not examine the role of human influences on water flow dynamics. Human modification of the land surface resulted in major areas of disagreement between observed conditions and modeled conditions in this study (Figure 8). To better show such differences, two elevation profiles were created. Profile A-A (Figure 9) is for the agricultural fields within the flood plain, whereas Profile B-B (Figure 10) is for a reservoir. Along Profile A-A , the green areas that represent overestimations are actually agricultural fields protected by levees. This area would be overestimated even at base flow because the stage of the river would be higher than the land elevation. Similarly, along Profile B-B , the elevation

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 345â&#x20AC;&#x201C;357

GIS Predictions of Flood Inundation

Figure 10. Line B-B elevation profile with the extent of river and the reservoir.

the natural conditions. Detailed modeling and vetting can overcome these errors but is beyond the scope of this work. It is, however, important to recognize that hazard mitigation studies on floodplains must take into account such human modification of the landscape that may not be discernable in topography-based simulation of inundation.

Figure 8. Map showing an example of the total disagreement associated with human modification of nature. Lines A-A and B-B show the location of the elevation profiles used for the agricultural fields and the reservoir, respectively. Map by Anas Rabie.

of the reservoir perimeter is designed to contain water at an elevation higher than the stage of the river. The models simulated that flood water will not inundate the reservoir, but the reservoir area always appears to be flooded. For both of these locations, the errors are related to human influences, which modify, mask, or alter

Figure 9. Line A-A elevation profile with the extent of river and agricultural fields.

Figure 11. Underestimation map for the fourâ&#x20AC;&#x201C;stream gauge scenario, 10-m DEM resolution. To provide spatial detail of the results, the fourth gauge (bea12) is not shown (off map). Refer to Figure 1 for stream gauge locations. Map by Anas Rabie.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 345â&#x20AC;&#x201C;357

355

Rabie, Peterson, Kostelnick, and Rowley

To validate the methods used in this study, Landsat imagery was acquired, but could not be used to represent the actual water distribution on the day of the flood peak in the City of Peoria (April 23, 2013) since that day coincided with heavy rain and severe weather conditions, with clouds covering the extent of the imagery, making it unusable for ground condition assessment.

Thus, a later date, April 29, was chosen to represent the peak inundation even though the stage was about 0.8 m below the peak flood (Figure 4). This could explain some of the underestimation in the simulation. In other words, areas predicted as “not flooded” that were actually flooded may be a result of residual flooding on April 29 associated with the peak conditions on April 23. Another possible explanation for the high underestimation is because the land still appears wet following the peak conditions, thus being classified as water in the binary classification of the imagery. An overview of the underestimated areas for the four-gauge scenario developed with the 10-m DEM is shown in Figure 11.

Furthermore, the mismatch in resolution between the Landsat imagery (at 30 m) and the 10-m and 1-m DEMs, given the unfortunate absence of other feasible satellite data sets, could have also influenced the error estimates in the accuracy assessment.

CONCLUSIONS The study found that it is possible to create flood prediction maps with a reasonable level of accuracy using few data inputs. The largest total error percentage is less than 30 percent, and if the two–stream gauge case scenario was excluded, the largest total error percentage was only 23 percent. It is noteworthy that even though the errors may seem large, the study error percentages represent the least amount of data possible to create a flood hazard analysis map, based on DEMs and stream gauge water levels. Using this simple approach with projected peak stages will allow for accurate prediction of areas to be flooded, which can be considered a good first solution with which to start planning flood emergency management situations. The simplicity of this model makes it a useful asset in urban planning and future flood predictions, addressing the concerns presented by Al-Sabhan et al. (2003). For instance, if an area is expecting a flood of a certain intensity, all emergency planners need to do is to use a DEM and available data from stream gauges to simulate flood conditions.

Given access to minimal data, DEMs, and stream gauge data, the methods we present allow for quick, simple, and accurate vulnerability predictions. Furthermore, this simplified model is easier to implement by a wide range of staff and personnel, especially those who are not flood engineers. While the scenarios us-

356

ing the 30-m DEM resolutions or two-gauge scenarios were not optimal, they still produced results similar to the LiDAR-derived 10-m and 1-m resolution DEMs. Using highly detailed data does not necessarily lead to better simulations or always produce better results; this was clearly shown by the accuracy assessment. Furthermore, the scenarios using the 1-m DEM also had a slightly larger total disagreement than did the 10-m DEM, which was unexpected. The computational difference between 30-m DEM use and 10-m DEM use is consistent with other hydrologic work (Jacoby et al., 2013). Schroeder et al. (2015) drew similar conclusions when developing stream profiles using 1-m and 10-m DEMs. This study showed that higher stream gauge density (using more stream gauges) does not necessarily produce better results, as this statement was largely unsupported by our analysis. In fact, the four–stream gauge scenario has the highest overestimate and the lowest underestimate across all three DEM resolutions, as well as the least total error percentage. Regardless of the DEM resolution used, the four–stream gauge distribution has the best results based on total error (underestimation or overestimation). One or a combination of factors may have led to these results. The four-gauge scenario may have been an optimal distance between and/or spatial distribution of the gauges. Another possibility is that in the process of going from five- or six-gauge scenarios, a gauge with higher error may have been removed from the simulations. A future study may try different distributions and combinations than those used in the current study. This would further determine if accuracy is correlated to the location and distance between individual stream gauges.

ACKNOWLEDGMENTS The authors acknowledge and thank two reviewers for their suggestions, which have improved the manuscript. The authors wish to thank King Abdulaziz University for the financial assistance afforded Anas Rabie while pursuing his M.S. degree.

REFERENCES AKANBI, A. A. AND SINGH, K. P., 1997, Managed Flood Storage Option for Selected Levees along the Lower Illinois River for Enhancing Flood Protection, Agriculture, Wetlands, and Recreation: Second Report, Validation of the UNET Model for the Lower Illinois River: Illinois State Water Survey, Contract Report 608, 110 p.

AL-SABHAN, W.; MULLIGAN, M.; AND BLACKBURN, G. A., 2003, A real-time hydrological model for flood prediction using GIS and the WWW: Computers, Environment Urban Systems, Vol. 27, No. 1, pp. 9–32, doi:10.1016/S0198-9715(01)00010-2.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 345–357

GIS Predictions of Flood Inundation DENG, Y.; WILSON, J. P.; AND GALLANT, J. C., 2008, Terrain analysis. In Wilson, J. P. and Fotheringham, A. S. (Editors), The Handbook of Geographic Information Science: Blackwell, Oxford, U.K., pp. 417–435.

EL-HAMES, A. S. AND RICHARDS, K. S., 1998, An integrated, physically based model for arid region flash flood prediction capable of simulating dynamic transmission loss: Hydrological Processes, Vol. 12, No. 8, pp. 1219–1232, doi:10.1002/(SICI)10991085(19980630)12:8<1219::AID-HYP613>3.0.CO;2-Q. FISHER, P. F. AND TATE, N. J., 2006, Causes and consequences of error in digital elevation models: Progress Physical Geography, Vol. 30, No. 4, pp. 467–489, doi:10.1191/0309133306pp492ra.

GALZKI, J.; MULLA, D.; JOEL, N.; AND WING, S., 2008, Targeting Best Management Practices (BMPs) to Critical Portions of the Landscape: Using Selected Terrain Analysis Attributes to Identify High-Contributing Areas Relative to Nonpoint Source Pollution: Minnesota Department of Agriculture.

GESCH, D. B., 2007, The national elevation dataset. In Maune, D. (Editor), Digital Elevation Model Technologies and Applications: The DEM Users Manual, 2nd ed.: American Society for Photogrammetry and Remote Sensing, Bethesda, MD, pp. 99– 118.

GESCH, D. B., 2009, Analysis of Lidar elevation data for improved identification and delineation of lands vulnerable to sea-level rise: Journal Coastal Research, Special Issue 53. pp. 49–58, doi:10.2112/si53-006.1.

GESCH, D. B.; OIMOEN, M.; GREENLEE, S.; NELSON, C.; STEUCK, M.; AND TYLER, D., 2002, The national elevation dataset: Photogrammetric Engineering Remote Sensing, Vol. 68, No. 1, pp.

5–32.

HAMMER, R. D.; YOUNG, F. J.; WOLLENHAUPT, N. C.; BARNEY, T. L.; AND HAITHCOATE, W., 1995, Slope class maps from soil survey and digital elevation models: Soil Science Society America Journal, Vol. 59, No. 2, pp. 509–519.

JACOBY, B.; PETERSON, E. W.; KOSTELNICK, J. C.; AND DOGWILER, T., 2013, Approaching cave level identification with GIS: A case study of Carter Caves: ISRN Geology, Vol. 2013, No. 160397, p. 7, doi:10.1155/2013/160397.

KIA, M.; PIRASTEH, S.; PRADHAN, B.; MAHMUD, A.; SULAIMAN, W.; AND MORADI, A., 2012, An artificial neural network model for flood simulation using GIS: Johor River Basin, Malaysia: Environmental Earth Sciences, Vol. 67, No. 1, pp. 251–264, doi:10.1007/s12665-011-1504-z.

KOSTELNICK, J. C.; MCDERMOTT, D.; ROWLEY, R. J.; AND BUNNYFIELD, N., 2013, A cartographic framework for visualizing risk: Cartographica, Vol. 48, No. 3, pp. 200–224, doi:10.3138/carto.48.3.1531.

LIAN, Y; CHAN, I.; XIE, H.; AND DEMISSIE, M., 2010, Improving HSPF modeling accuracy from FTABLES: Case study for the Illinois River Basin: Journal Hydrologic Engineering, Vol. 15, No. 8, pp. 642–650, doi:10.1061/(ASCE)HE.19435584.0000222.

POULTER, B. AND HALPIN, P. N., 2008, Raster modelling of coastal flooding from sea-level rise: International Journal Geographical Information Science, Vol. 22, No. 2, pp. 167–182, doi:10.1080/13658810701371858.

ROSNER, A.; VOGEL, R. M.; AND KIRSHEN, P. H., 2014, A risk-based approach to flood management decisions in a nonstationary world: Water Resources Research, Vol. 50, No. 3, pp. 1928– 1942, doi:10.1002/2013WR014561.

SAFARIPOUR, M.; MONAVARI, M.; ZARE, M.; ABEDI, Z.; AND GHARAGOZLOU, A., 2012, Flood risk assessment using GIS (case study: Golestan province, Iran): Polish Journal Environmental Studies, Vol. 21, No. 6, pp. 1817–1824.

SCHROEDER, K.; PETERSON, E. W.; AND DOGWILER, T., 2015, Field validation of DEM and GIS derived longitudinal stream profiles: Journal Earth Science Research, Vol. 3, No. 3, pp. 43–54, doi:10.18005/JESR0303002.

WATSON, D. F. AND PHILIP, G., 1985, A refinement of inverse distance weighted interpolation: Geo-processing, Vol. 2, No. 4, pp.

315–327.

WORTHEN, A. H., 1868, Geology of Illinois: Legislature of Illinois.

ZHANG, X.; DRAKE, N. A.; WAINWRIGHT, J.; AND MULLIGAN, M., 1999, Comparison of slope estimates from low resolution DEMS: Scaling issues and a fractal method for their solution: Earth Surface Processes Landforms, Vol. 24, pp. 763–779.

Environmental & Engineering Geoscience, Vol. XXIII, No. 4, November 2017, pp. 345–357

357