EEG Journal Vol. XXVII, No. 3 by Association of Environmental & Engineering Geologists (AEG)

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, 3053 Nationwide Parkway, Brunswick, OH 44212. 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, 3053 Nationwide Parkway, Brunswick, OH 44212. Phone: 844-331-7867. 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 $75 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 $310 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, 3053 Nationwide Parkway, Brunswick, OH 44212. © 2021 by the Association of Environmental and Engineering Geologists

THIS PUBLICATION IS PRINTED ON ACID-FREE PAPER Abdul Shakoor Kent State University Kent, OH 44242 ashakoor@kent.edu

EDITORS

Eric Peterson Department of Geography, Geology, and the Environment Illinois State University Normal, IL 61790 309-438-5669 ewpeter@ilstu.edu

Karen E. Smith, Editorial Assistant, kesmith6@kent.edu

Oommen, Thomas Board Chair, Michigan Technological University Sasowsky, Ira D. University of Akron

ASSOCIATE EDITORS Ackerman, Frances Ramboll Americas Engineering Solutions, Inc. Bastola, Hridaya Lehigh University Beglund, James Montana Bureau of Mines and Geology Bruckno, Brian Virginia Department of Transportation Clague, John Simon Fraser University, Canada Dee, Seth University of Nevada, Reno Fryar, Alan University of Kentucky Gardner, George Massachusetts Department of Environmental Protection

Hauser, Ernest Wright State University Keaton, Jeff AMEC Americas May, David USACE-ERDC-CHL Pope, Isaac Book Review Editor Santi, Paul Colorado School of Mines Schuster, Bob Shlemon, Roy R.J. Shlemon & Associates, Inc. Stock, Greg National Park Service Ulusay, Resat Hacettepe University, Turkey West, Terry Purdue University

Environmental & Engineering Geoscience August 2021 VOLUME XXVII, NUMBER 3

Submitting a Manuscript 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 https://www.editorialmanager.com/EEG/ default.aspx. 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. Manuscripts that do not follow the Style Guide and the Instructions for Authors will be returned. Authors do not pay any charge for color figures that are essential to the manuscript. Manuscripts of fewer than 10 pages may be published as Technical Notes. For further information, you may contact Dr. Abdul Shakoor at the editorial office.

Cover photo View East at Brilliant Cut Rock Slope Failure, Pittsburgh, Allegheny County, Pennsylvania, March 20, 1941. Photo by the late A.C. Ackenheil. See article on page 269.

Volume XXVII, Number 3, August 2021

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.

ADVISORY BOARD Watts, Chester “Skip” F. Radford University Hasan, Syed University of Missouri, Kansas City Nandi, Arpita East Tennessee State University

ENVIRONMENTAL & ENGINEERING GEOSCIENCE

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, 3053 Nationwide Parkway, Brunswick, OH 44212 and additional mailing offices.

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 Volume 27, Number 3, August 2021 Table of Contents 259

Geoprocessing Techniques for the Visualization of Subsurface Geologic Data in Geographic Information Systems Nathan D. Williams

269

Re-evaluation of the 1941 Rock Slide at Brilliant Cut, Pittsburgh, Pennsylvania James V. Hamel

287

An Integrated InSAR-Borehole Inclinometer-Numerical Modeling Approach to the Assessment of a Slow-Moving Landslide Mirko Francioni, Doug Stead, Jayanti Sharma, John J. Clague, Marc-André Brideau

307

Factors Contributing to Landslide Susceptibility of the Kope Formation, Cincinnati, Ohio Michael P. Glassmeyer, Abdul Shakoor

319

Spatiotemporal Evaluation of Flood Potential Indices for Watershed Flood Prediction in the Mississippi River Basin, USA Dorcas Idowu, Wendy Zhou

331

Improved Automated Mapping of Sinkholes Using High-Resolution DEMs Yonathan Admassu, Celestine Woodruff

353

Macrostructural and Microstructural Properties of Residual Soils as Engineered Landfill Liner Materials Lee Li Yong, Vivi Anggraini, Mavinakere Eshwaraiah Raghunandan, Mohd Raihan Taha Technical Note

367

Installation of Utility Trench Wells Using Vacuum Techniques for Urban Groundwater Investigation Nick Schmidt, Martin G. Shepley Book Reviews

371

Timefulness: How Thinking Like a Geologist Can Help Save the World Marcia Bjornerud

373

A Hero on Mount St. Helens: The Life and Legacy of David A. Johnston Melanie Holmes

Geoprocessing Techniques for the Visualization of Subsurface Geologic Data in Geographic Information Systems NATHAN D. WILLIAMS* Geology Section, Civil Design Branch, Engineering and Construction Division, Nashville District, U.S. Army Corps of Engineers, 801 Broadway, Nashville, TN 37203

Key Terms: GIS, site characterization, Python ABSTRACT The ability to visualize subsurface geologic information is critical to sound decision making in many disciplines of geology. While there are numerous commercial off-the-shelf software solutions available to model geologic data in both 2D and 3D, these can be costly and have a steep learning curve. Some of the same functionality of these software packages can be accomplished by workflows that incorporate built-in geoprocessing tools of Geographic Information System (GIS) software. These workflows allow the geologist to plot vertical or inclined borehole data in 2D or 3D, create section views of raster data along section lines, and provide a means to convert contact elevations from existing geologic cross sections into plan-view or 3D space. These workflows have been successfully used to visualize construction data and subsurface geologic information for several embankment dams. Grouting and exploratory borehole data from databases with tens of thousands of records have been transformed into 2D and 3D GIS features. The workflows were instrumental in developing a 3D GIS model of site geology from which a series of geologic cross sections were drawn. These sections were critical in informing risk decisions related to the foundation conditions for a recent risk assessment of an earthen embankment dam.

INTRODUCTION Geographic Information Systems (GIS) are commonly used for plan-view visualization and manipulation of spatial data. The built-in functionality of most GIS software packages is focused largely on working with data in this (plan) dimension. However, many geologic problems involve subsurface data and require modification of existing tools and workflows to accommodate data in profile or section view. The workflows *Corresponding author email: nathan.d.williams@usace.army.mil

described here aim to exploit the built-in functionality of standard GIS software to visualize subsurface geologic data in both 2D and 3D. Some of these processes include the following:

r Converting tabular borehole data to 3D vector features,

r Projecting boreholes to section lines or proﬁle views, r Slicing proﬁle lines along geologic surfaces represented by rasters,

r Projecting section contacts to 3D points for interpolation. These operations rely on a combination of trigonometry and linear referencing. The trigonometry calculations are used to determine the locations of line vertices for boreholes, and the linear referencing operations are necessary for projecting features to and from section views of the data and real-world coordinates. The resulting GIS-based features or models can be integrated with other project information, such as performance monitoring, construction, or laboratory data, within an existing data management system. METHODS The following workﬂows were used to create a custom script toolbox for ArcGIS (Environmental Systems Research Institute [ESRI], 2014) called the Geologic Section Tools (Figure 1). The accompanying tables list the ArcGIS geoprocessing operations used by each tool in the script toolbox and also commands to replicate the functionality of select tools outside of ArcGIS using open-source Python packages. Tabular Borehole Data to 3D Vector Features—Borehole to 3D Digital geologic data are commonly stored database tables or spreadsheet formats. For example, borehole information may be stored in a database where each record represents an interval of the borehole with a

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 259–267

259

Williams

countered in a tunnel or mine roof—have an inclination of 180). 3. Depths to the top and bottom of the interval recorded in the same units as the X, Y, and Z coordinates (typically feet for State Plane).

Figure 1. The Geologic Section Tools script toolbox for ArcGIS.

defined top and bottom and might contain lithology, blow counts, grout volume, rock quality designation, or other types of information associated with the material in the borehole. The input table should be formatted with information about each unique interval in a separate row (Figure 2). The table can include data from a single borehole or multiple boreholes. Most geotechnical database software is likely capable of exporting a table with the required information, though it is possible to compile the necessary information in a spreadsheet program from a field log or historical data. The minimum fields required to locate the borehole in 3D space are the following: 1. X, Y, and Z coordinates for the top of the borehole in a projected coordinate system (e.g., State Plane, UTM). 2. Azimuth and inclination of the borehole in degrees, with azimuth recorded using standard convention relative to true north at 0 and inclination recorded as deviation from vertical downward (vertical downward boreholes have an inclination of 0, and vertical upward boreholes—as might be en-

Some software will require a unique identifier for each interval. A useful identifier might contain the borehole name and top and bottom depths (e.g., BH29_10.5_15.0). In addition, the interval should also contain at least one field that will be symbolized in the GIS—this might be the lithology or formation of the interval, a grout volume, number of blow counts, and so on. Using the coordinates, azimuth, and inclination for the top of the borehole along with the depths to the top and bottom of the interval, the X, Y, and Z coordinates for the vertices of the interval are calculated as follows: XTOP and XBOT = XTOH + sin(A) × (sin(I ) × D),

YTOP and YBOT = YTOH + cos(A) × (sin(I ) × D),

ZTOP and ZBOT = ZTOH − cos (I ) × D, where XTOH, YTOH, and ZTOH are the X, Y, and Z coordinates for the top of the borehole, respectively. The azimuth and inclination are A and I in radians. The depth (D) corresponds to the depth to either the top or the bottom of the interval or distance along the borehole for the case of horizontal borings. After these six ﬁelds are calculated for each record in the table, they can be plotted in the GIS and converted to line features or converted directly to lines using Python (Table 1). The process for displaying point data for the boreholes is identical to the process for displaying the interval data, though the point information requires only a single X, Y, and Z calculation instead of the two calculations required for interval vertices, and points are

Figure 2. Example table format for the Borehole to 3D tool. The tool requires the location and geometry and interval fields, and at least one field should be included that will be symbolized in the section. The supplemental field(s) are optional and can include any additional relevant information.

260

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 259–267

Visualizing Geologic Data in GIS Table 1. Workflow for generating 3D line features from tabular borehole data using ArcGIS and open-source Python packages. Process 1

Add ﬁelds to input borehole data table to contain interval vertex coordinates Calculate vertex coordinates

Create points for interval top and bottom from vertex coordinates

Combine points into single feature class Create lines from points features

Pythona

ArcGIS Tool Add Field (Data Management)

Not applicable

Calculate Field (Data Management) Make XY Event Layer (Data Management)

df[‘XTOP’] = df[‘XTOH’] + np.sin(df[‘Azimuth’]) * (np.sin(df[‘Inclination’]) * df[‘Depth_to_Top’]) df[‘TOP_POINT’] = df.apply(lambda x: shapely.geometry.Point((float(x[‘XTOP’]), float(x[‘YTOP’]), (float(x[‘ZTOP’]))), axis = 1) Not applicable

Merge (Data Management) Points to Line (Data Management) Join Field (Data Management)

Add original ﬁelds to output line features Create Geopandas data frame

Save output line features

Not applicable

df[‘line’] = df.apply(lambda x: shapely.geometry.LineString([x[‘TOP_POINT’], x[‘BOT_POINT’]]), axis = 1) Not applicable df = geopandas.GeoDataFrame(df, geometry = ‘line’) df.to_file(r’output_shapefile_path.shp’, driver = ‘ESRI Shapefile’)

a Input table read into pandas data frame in Python as ‘df’. For Python, Pandas (McKinney et al., 2010) library imported as pd, numpy (Oliphant, 2006) as np. Geopandas (Jordahl, 2014) used to write the data frame to shapeﬁle.

not subsequently converted to line features. Point information in the borehole might include features like geologic contacts, discontinuities, or other information that occurs at a discrete location along the borehole. Project Boreholes to Section Lines or Profile Views—Borehole to Section After 3D boreholes have been generated, they can be projected along one or more section lines. This is accomplished by converting the coordinates of the vertices of each interval of the borehole from projected coordinates (e.g., State Plane or UTM) to coordinates that represent the position (measure) and offset along the section baseline(s). In section view, the X coordinate of the vertex of interest is the measure value along the section line, and the Y coordinate is the same as the point elevation (Z coordinate). ArcGIS (ESRI, 2014) requires that the section baseline be converted to an maware route feature before the point measure and offset can be calculated. In Python, this is accomplished using the shapely (Gillies et al., 2007) “project” method on the baseline object to determine the measure value and the shapely “distance” object to calculate the point offsets (Table 2). Slice Profile Lines along Geologic Surfaces—Surface to Section The Surface to Section tool is used to create profile graphs of raster surfaces along one or more section

lines. The inputs for the tool are one or more raster surfaces and a route features class along which the profile graphs will be created. The output from the tool is a line feature class with coordinates in 2D section-view space referenced to the input section lines. The tool works by densifying the input plan-view section line feature (adding additional vertices), then extracting the underlying raster values to the vertex points (Table 3). Measure values for these points are subsequently referenced to the original section line using the techniques described in step 2 of the Borehole to Section workflow, then converted to lines using the measure value as the X coordinate and the underlying raster elevation as the Y coordinate. Project Section Contacts to 3D Points for Interpolation—Section Contacts to Surface Points The Section Contacts to Surface Points tool in the From Section Toolset is used primarily to facilitate the creation of 3D surfaces from existing geologic sections (Table 4). The inputs for the tool are a line feature class with geologic contacts or surfaces of interest in section view and one or more route feature classes along which the lines will be projected. The output of the tool is a point feature class in plan coordinates that contains an elevation field that can be used to interpolate a surface. This tool functions roughly as an inverse of the Surface to Section tool and permits section-view line geometries to be projected into plan-view points that are subsequently used for raster interpolation.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 259–267

261

Williams Table 2. Workflow for generating section view line features from tabular borehole data using ArcGIS and open-source Python packages. Note that steps 3–8 are identical to the Borehole to 3D workflow except the coordinates for the vertex points (step 3) are the measure value along the baseline and elevation (Z). Process

Pythona

ArcGIS Tool

Create route from section line (baseline) feature as necessary

Create Routes (Linear Referencing)

Calculate measure and oﬀset values for interval vertices relative to the section baseline

Locate Features Along Routes (Linear Referencing)

Create points for interval top and bottom from vertex coordinates using measure (X) and elevation (Y) Combine points into single feature class Create lines from points features

Make XY Event Layer (Data Management)

4 5

6 7 8

Add original ﬁelds to output line features Create Geopandas data frame Save output line features

Merge (Data Management) Points to Line (Data Management) Join Field (Data Management)

Read in section line feature from shapefile or feature class or generate from vertices; convert to shapely LineString object df[‘meas’] = [baseline.project(Point(df[‘Easting’].iloc[i], df[‘Northing’].iloc[i])) for i in range(len(df))] df[‘offset’] = [baseline.distance(Point(df[‘Easting’].iloc[i], df[‘Northing’].iloc[i])) for i in range(len(df))] df[‘TOP_POINT’] = df.apply(lambda x: shapely.geometry.Point((float(x[‘meas_TOP’]), float(x[‘ZTOP’]))), axis = 1) Not applicable df[‘line’] = df.apply(lambda x: shapely.geometry.LineString([x[‘TOP_POINT’], x[‘BOT_POINT’]]), axis = 1) Not applicable df = geopandas.GeoDataFrame(df, geometry = ‘line’) df.to_file(r’output_shapefile_path.shp’, driver = ‘ESRI Shapefile’)

Not applicable Not applicable

a For Python, the section line is represented by a shapely LineString named ‘baseline’. Input borehole information as 3D vertices (points) from step 4 in the 3D borehole process.

APPLICATION EXAMPLE The workﬂows (Figure 3) used in the Geologic Section Tools were successfully used to develop new geologic sections and a geologic map of the foundation excavation for a recent risk assessment of J. Percy Priest Dam.

Table 3. Workflow for generating section view line features from raster surfaces using ArcGIS. Process 1 2 3

262

The data compilation and formatting process involved digitizing borehole data from the geologic exploration program before and during construction in the 1960s. The data were available only in graphical form on scanned as-built drawings and had to be georeferenced and digitized in GIS to extract the relevant information (Figure 4). The resulting tabular data Table 4. Workflow for generating plan-view point features from section-view contact lines using ArcGIS. The resulting point features can be used to interpolate new raster surfaces of the geologic contacts of interest.

ArcGIS Tool

Add additional vertices to plan-view section line Convert vertices to point features Determine the raster value (elevation) underlying each section line vertex Calculate measure and oﬀset values for interval vertices relative to the section baseline Create points from vertex coordinates using measure (X) and raster elevation (Y) Create lines from points features

Process

Densify (Editing) 1 Feature Vertices to Points (Data Management) Extract Multi Values to Points (Spatial Analyst) Locate Features Along Routes (Linear Referencing)

2 3

4 5

Make XY Event Layer (Data Management) 6 Points to Line (Data Management)

Add additional vertices to section-view contact line(s) Convert vertices to point features Calculate the X (measure) and Y (elevation) coordinates for each vertex Convert point features to table Project section-view coordinates to plan-view coordinates using the measure value Create points from vertex coordinates using measure (X) and raster elevation (Y)

ArcGIS Tool Densify (Editing) Feature Vertices to Points (Data Management) Add XY Coordinates (Data Management) Table to Table (Conversion) Make Route Event Layer (Linear Referencing)

Make XY Event Layer (Data Management)

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 259–267

Visualizing Geologic Data in GIS

Figure 3. General process to create 2D and 3D borehole features and geologic sections using the Geologic Section Tools workﬂows.

contained the required location and geometry, interval, and geology (stratigraphy) information for the Borehole to 3D tool. After the borehole data were converted to 3D features using the Borehole to 3D tool (Figure 5), a geologist examined the data extents and regional geology to determine several section lines to be in-

terpreted (Figure 6). The section lines were converted to routes so that M (linear referencing) information was stored with the feature class, which is a requirement of the tools that project features to section view. The 3D boreholes were projected to the section routes using the Borehole to Section tool, and a

Figure 4. Digitized stratigraphic intervals drawn over georeferenced as-built drawings (background image). Colors represent stratigraphic breaks in the Lebanon and Ridley Formations identiﬁed during construction.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 259–267

263

Williams

Figure 5. Result of Borehole to 3D tool shown in 3D space. Colors correspond to stratigraphic breaks identiﬁed in as-built drawings. The gray surface is a hillshade of the dam foundation at the top of the rock excavation. Although the boreholes shown here are vertical, the workﬂow described in the text can also be used to generate inclined borehole features.

geologist interpreted the stratigraphy between the boreholes (Figure 7). After interpretation, the Section Contacts to Surface Points tool was used to generate plan-view point features for each of the subsurface contacts at a specified interval along the section lines (Figure 8). The purpose of the Section Contacts to Surface Points tool is to generate artificial elevation points that allow subsequent interpolation operations to better capture variation in geologic structure from the interpreted sections that may not be adequately represented in the raw borehole data. The natural neighbor interpolation method was used on the elevation field of the points feature class produced by the Section Contacts to Surface Points tool to produce raster surfaces of each geologic con-

tact (Figure 9). These rasters were inputs for new sections generated along the embankment using the Surface to Section tool. The output of this tool consists of line features in section coordinates that were enclosed to form polygons and symbolized by stratigraphy (Figure 10). The locations of piezometers and other instrumentation were projected to the new geologic sections to aid in the interpretation of performance monitoring data using the Borehole to Section tool. Water elevations from piezometers tipped in the embankment and foundation were interpolated to raster surfaces, then added to the geologic sections using the Surface to Section tool. In addition to the new geologic sections, the geologic contact rasters were intersected with a raster

Figure 6. Historical borehole locations and section lines for interpretation in plan view relative to the dam axis baseline.

264

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 259–267

Visualizing Geologic Data in GIS

Figure 7. Stratigraphic interpretation in section view from boring logs projected to a section line. Colors correspond to rock unit boundaries identiﬁed in as-built drawings.

Figure 8. Results of the Section Contacts to Surface Points tool shown in 3D with the points assigned a base height from the contact elevation. Point colors correspond to stratigraphic units. The gray surface is the dam foundation excavation at the top of the rock.

Figure 9. Interpolated rasters and boreholes in 3D from points created from points generated by the Section Contacts to Surface Points tool. Colors correspond to stratigraphic units, and the gray surface is the dam foundation excavation at the top of rock.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 259–267

265

Williams

Figure 10. Cross section through J. Percy Priest Dam generated using the Geologic Section Tools to facilitate dam safety risk decision making. Piezometer geometry and water elevations are shown from upstream to downstream with an interpolated piezometric surface line from readings in March 2019.

of the dam-excavated foundation surface to produce a detailed geologic map of the excavation showing stratigraphic members encountered during construction (Figure 11). The stratigraphic information for this map is exclusively from exploratory borehole information, and no surface mapping information was available in the as-built drawings.

CONCLUSIONS The ability to visualize subsurface information is critical to risk decision making for dam safety projects. The geoprocessing workﬂows described here provide a means of processing and visualizing subsurface geologic information in ArcGIS (ESRI, 2014) using

Figure 11. Plan-view geologic map (top) and proﬁle (bottom) derived from exploratory borehole data. The plan geology is shown with a hillshade of the excavated surface that was generated from topographic contours provided in the construction as-built drawings.

266

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 259–267

Visualizing Geologic Data in GIS

existing geoprocessing tools. Although the tools were developed to facilitate site characterization of dam projects, they are ﬂexible enough to be used on other types of projects. REFERENCES Environmental Systems Research Institute (ESRI), 2014, ArcGIS Desktop, Release 10.3: Environmental Systems Research Institute, Redlands, CA.

Gillies, S.; Bierbaum, A.; Lautaportti, K.; and Tonnhofer, O., 2007, Shapely: Manipulation and Analysis of Geometric Objects: Electronic document, available at https://pypi.org/ project/Shapely. Jordahl, K. 2014, GeoPandas: Python tools for geographic data. URL: https://github. com/geopandas/geopandas. McKinney, W., 2010, Data structures for statistical computing in Python. In Proceedings of the 9th Python in Science Conference, Vol. 445, pp. 51–56. Oliphant, T. E., 2006, A Guide to NumPy, Vol. 1: Trelgol Publishing USA.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 259–267

267

Re-evaluation of the 1941 Rock Slide at Brilliant Cut, Pittsburgh, Pennsylvania JAMES V. HAMEL* Hamel Geotechnical Consultants, 1992 Butler Drive, Monroeville, PA 15146

Key Terms: Brilliant Cut, Pittsburgh, Pennsylvania, Allegheny River, Rock Slide, Progressive Failure, Historic Photographs, Stability Analyses, Residual Strength ABSTRACT On March 20, 1941, more than 110,000 yd3 (84,000 m3 ) of rock slumped from Brilliant Cut in Pittsburgh, Pennsylvania. Failure was triggered by water pressure buildup due to ice blockage of drainage outlets on the slope face. I investigated this slide as part of my Ph.D. research at the University of Pittsburgh in 1968–1969 and have continued to study it. Historical photographs discovered in 1997 provided new insights on the construction and failure of Brilliant Cut and led to this re-evaluation. In this paper, my 1968–1969 work is summarized and then additional geological and historical information is presented along with key observations from the historical photographs. These photographs reveal that slope excavation at Brilliant Cut in 1930–1931 removed lateral support, in turn initiating stress release and progressive failure that loosened or broke bedrock adjacent to the cut. This fractured rock mass remained marginally stable for a decade but then collapsed in March 1941. The 1941 failure was triggered by water held back in rock fractures by a frozen crust over talus and fractured rock on the slope face. A progressive failure mechanism by Brooker and Peck explains the behavior of Brilliant Cut from 1931 to 1941. Sliding Block stability analyses demonstrate the mechanism of progressive failure and suggest that friction angles were reduced gradually to near-residual values along the failure surface, with low water levels in the slope. With drainage blocked in 1941, a water level developed about 30 ft (9 m) above the basal failure surface to initiate the catastrophic failure. This water level is below that previously inferred to have existed at the time of failure. INTRODUCTION On March 20, 1941, a rock mass of more than 110,000 yd3 (84,000 m3 ) slumped from the Brilliant Cut along the Pennsylvania Railroad in Pittsburgh, *Corresponding author email: jvhamel3918@gmail.com

Pennsylvania. The toe of this rock mass displaced three sets of railroad tracks and a train was derailed. It was thought that the failure was triggered by water pressures built up in the slope because natural drainage outlets on the slope face were blocked by ice (Philbrick, 1953; Ackenheil, 1954; and Philbrick, 1960). I analyzed this slide at Brilliant Cut in 1968–1969 as part of my Ph.D. research at the University of Pittsburgh (Hamel, 1969) and presented a paper on it at the 13th U.S. Rock Mechanics Symposium in 1971 (Hamel, 1972). At that time, my focus was on backcalculation of shear strengths from rockslides and assessment of these back-calculated values in terms of peak and residual strength behavior (Skempton, 1964). In 1968–1969, I recognized some shortcomings and uncertainties in my analysis of the slide at Brilliant Cut, but I considered the analysis reasonable based on the general state of knowledge and the information available at that time. During the past 50 years, I have learned a great deal about geology and rockslides in Pittsburgh and the vicinity and found numerous historical photographs relevant to Brilliant Cut that I had not previously known to exist. My re-evaluation of the slide at Brilliant Cut began in 1997 when these photographs were discovered. Because of other priorities, this reevaluation was set aside in 1998 and not resumed until 2019. This paper begins with a summary of my 1968–1969 work. Then I present additional geological and historical information along with key observations from historical photographs. A mechanism of progressive failure (Brooker and Peck, 1993) is suggested to explain the slope behavior at Brilliant Cut from 1931 to 1941. I evaluate water levels in the slope at the time of the 1941 failure from geologic and hydrologic perspectives and then use Sliding Block stability analyses of progressive failure to estimate shear strengths and water levels along the 1941 failure surface. SUMMARY OF 1968–1969 WORK Location and Geology Brilliant Cut is located at the intersection of Washington Boulevard and Allegheny River Boulevard on

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

269

Hamel

along horizontal to sub-horizontal bedding contacts in argillaceous rocks to residual or near residual values (Ferguson and Hamel, 1981; Hamel, 2018). History

Figure 1. A Brilliant Cut location map. Re-drawn from Hamel (1969, 1972).

the east side of Pittsburgh (Figures 1 and 2). The cut is in the nose of a hill at the junction of an abandoned river valley and the present valley of the Allegheny River (Figure 2). Figure 3 is a plan of the 1941 slide. Rocks at Brilliant are ﬂat-lying sedimentary strata of the Pennsylvanian age Conemaugh Group (Johnson, 1929; Wagner et al., 1970, 1975a, b). These strata include sandstones, shales, and claystones with some thin coals and limestones. On older drawings and documents and in Hamel (1969, 1972), the claystones are called “indurated clays.” This latter term was formerly used by the Ohio River Division, U.S. Army Corps of Engineers, for massive, slickensided claystone. Figure 4 is a detailed stratigraphic column and Figure 5 is a generalized cross section through Brilliant Cut before the 1941 slide. This information was obtained by the Pittsburgh District, U.S. Army Corps of Engineers, in 1940–1941 during investigations for a large spillway cut in similar rocks at Youghiogheny Dam (Philbrick, 1960). The failure surface in Figure 5 was determined from borings made by the Pittsburgh District after the 1941 slide. A Pittsburgh District drawing from which Figures 4 and 5 were prepared was provided by Ferguson (1968). The rocks at Brilliant have extensive jointing and fracturing due to stress release accompanying erosion of both the present Allegheny River valley and the abandoned valley (Ferguson, 1967). We can now reasonably infer that movements associated with this valley stress release also reduced shear strengths

270

The original Pennsylvania Railroad tracks at Brilliant were constructed in 1904–1905 with only a slight sidehill cut. In 1930–1931, the railroad was relocated farther into the hillside to permit construction of Allegheny River Boulevard. The completed cut had a height of about 160 ft (48 m), a width of about 330 ft (100 m), and an inclination of 1:1 or 45° (Figures 3 and 5). Sometime during the 1930s, a vertical joint opened to a width of about 1 ft (0.3 m) along the crest of the excavated slope (Figure 3). This joint was filled with concrete, but it had opened at least several inches (mm) more by July 1940. Plate 13 of Ackenheil (1954), from which Figure 3 was developed, indicates this joint “opens again one foot [0.3 m] before slide.” No further information on this joint and its concrete filling is presently available. I observed 1.0 to 1.5 ft (0.3 to 0.5 m) wide remnants of the concrete filling along the slide scarp in 1968. This open joint (attributed to valley stress release and progressive slope movement) formed the tension crack at the rear of the 1941 failure mass (Figure 5). The Slide and Its Cleanup Between 4:00 and 5:30 am on March 20, 1941, the entire nose of the hill broke off and slumped down, displacing three sets of railroad tracks and causing a train derailment. The head of the slide mass dropped about 15 ft (4.5 m). The slide appeared to have occurred rapidly with little movement after the initial slump (Ackenheil, 1954). Recent review of Pittsburgh weather records indicates that approximately 1.60 in. (41 mm) of waterequivalent rain and snow fell from March 1 to March 19. Daytime high temperatures were above freezing and overnight low temperatures were below freezing from March 12 to March 16. Daytime high temperatures were 25°F and 23°F (–4°C and –5°C), with overnight lows of 10°F and 9°F (–12°C and –13°C), respectively, on March 17 and 18. The daytime high was 36°F (2°C), and the overnight low was 18°F (−8°C) on March 19. Normally, surface water infiltrating the upper part of the slope, along with groundwater from behind the slope, would drop down through open stress-release joints and fractures, then drain laterally to the slope face through fractures in the Duquesne coal, Ames

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

Brilliant Cut Slide Re-evaluation

Figure 2. Portion of a City of Pittsburgh 1962 topographic map of Brilliant Cut and vicinity. See Figure 1 for general location. Elevations are in feet; to convert to meters, multiply by 0.305. AR = Allegheny River, ARB = Allegheny River Boulevard, ARC = Abandoned River Channel, BPS = Brilliant Pumping Station, P = Plateau, S = Saddle, SL = Slide, WB = Washington Boulevard.

Limestone, and un-named limestone below the Pittsburgh red beds (Figures 4 and 5). On the day of the slide, ice formations existed on the slope face at the levels of the Duquesne coal and Ames Limestone. It originally appeared that ice plugged essentially all of these outlets and that water pressure in the open joint at the rear of the failure mass (Figure 5) triggered the slide (Philbrick, 1953; Ackenheil, 1954; and Philbrick, 1960). The northerly exposure of the slope face places it in shadow much of the day, contributing to the development and maintenance of ice formations. Excavation of material from the slide toe began on March 21, 1941, and was completed on May 15, 1941, after removal of 109,400 yd3 (83,600 m3 ) of rock debris. During excavation, the failure mass slumped further downward with a slow, continuous rotation and large displacements (several feet or meters), which I believe indicated residual strength behavior. The failure mass was benched at mid-height to decrease its rotational thrust, and the railroad tracks were rebuilt (Ackenheil, 1954).

Stability Analyses The slope cross section in Figure 5 was analyzed using the Morgenstern and Price (1965, 1967) method to back-calculate the eﬀective stress Mohr-Coulomb shear-strength parameters required along the basal clay shale and claystone, through which 72 percent of the failure surface passed. Reasonable shear-strength parameters were assumed for rocks along the remaining 28 percent of the failure surface. For these analyses, I assumed the vertical joint at the rear of the failure mass was open to a depth of 100 ft (30 m) and ﬁlled with water, as inferred by Ackenheil (1954) and Philbrick (1953, 1960). Elsewhere along the failure surface, groundwater was assumed static and everywhere at the level of the ground surface. A total unit weight of 160 lb/ft3 (25.1 kN/m3 ), typical for rocks in the slope, was used in the stability analyses. Other details of these analyses are given in Hamel (1969, 1972). For the above-mentioned conditions, a zerocohesion friction angle of 32° was calculated for the

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

271

Hamel

Figure 3. Plan of the 1941 slide at Brilliant Cut. Elevations are in feet; to convert to meters, multiply by 0.305. Re-drawn from Hamel (1972).

basal clay shale and claystone at limiting equilibrium of the failure mass. This calculated friction angle of 32° corresponds to peak or near-peak friction angles measured in field and laboratory tests on clay shales and claystones similar to those at Brilliant Cut (Hamel, 1969, 1972; Hamel and Adams, 1981). These friction angles are well above typical residual friction angles of 11° to 16° measured in laboratory tests on similar materials (Hamel, 1969, 1972; Hamel and Adams, 1981). ADDITIONAL BACKGROUND INFORMATION Pleistocene Geology and Landforms During Pleistocene time, continental ice sheets extended south to a line approximately 30 miles (50 km) north of Pittsburgh on at least two occasions. Ice and outwash dams, along with torrential flows of meltwater and outwash, profoundly influenced and radically changed drainage systems in the region (Leverett, 1902; Johnson, 1929; Leverett, 1934; Wagner et al., 1970; Jacobson et al., 1988; and Hamel, 1998). Periglacial conditions with heavy precipitation and numerous deep freeze–thaw cycles undoubtedly occurred when ice sheets were not far north of Pitts-

272

burgh. These Pleistocene and periglacial conditions significantly affected landforms in the region, particularly in the vicinity of Brilliant Cut. Figure 2 is a 1962 topographic map of Brilliant Cut and adjacent areas. Washington Boulevard west of Brilliant Cut follows an abandoned channel where water from the Allegheny River flowed south to the Monongahela River (and perhaps water from the Monongahela River flowed north to the Allegheny River) when these rivers were dammed by ice and/or outwash at or downriver from their confluence (Figure 1) at various times during the late Pleistocene (Johnson, 1929; Leverett, 1934; and Wagner et al., 1970). The plateau southeast of Brilliant Cut (Figure 2) was eroded by the Allegheny River in late Pleistocene time. It is likely that the Allegheny River flowed over the saddle at nominal Elevation 1,000 ft (300 m) at the southeast edge of this plateau on various occasions during Pleistocene time. Steep-sided ravines along the east side and southwest edge of the plateau were created by subsequent erosion (probably from heavy periglacial precipitation) (Figure 2). It is apparent that rocks at Brilliant Cut experienced stress release from major valley erosion (Ferguson, 1967; Ferguson and

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

Brilliant Cut Slide Re-evaluation

Hamel, 1981; and Hamel, 2018) on the north and west sides of the cut. Extensive landsliding involving both bedrock and bedrock-derived colluvial soils has occurred along the valley walls of the Allegheny River and its tributaries since Pleistocene time (Ackenheil, 1954; Pomeroy, 1979). Many of these slide masses were partially eroded during late Pleistocene and Holocene time. The Brilliant Pumping Station was constructed on the riverbank bench north of Brilliant Cut (Figure 2). I infer this bench is a partially eroded landslide remnant. Smaller scale topographic maps, e.g., Plate I of Johnson (1929), show it as an anomalous protuberance along the cut-bank side of the Allegheny River.

Brilliant Pumping Station

Figure 4. Generalized stratigraphic column at Brilliant Cut. Redrawn from Hamel (1969, 1972).

The City of Pittsburgh constructed the original water intake and pumping station at Brilliant in 1879 and modified and/or enlarged this facility several times since then. Scharff (1920) stated that a landslide of unspecified size occurred in 1904 during the original construction of Brilliant Cut. He said that this slide, which involved the motion of fill and underlying shale, threatened the Brilliant Pumping Station then under construction. No other information has been found on this slide. It seems likely that this slide involved colluvial soil and underlying rockslide debris. A 1907 photograph (Wikimedia, 2020) shows the Brilliant Pumping Station and the outer edge of the

Figure 5. Cross-section of the 1941 slide at Brilliant Cut. All strata belong to Pennsylvanian age Conemaugh Group. Re-drawn from Hamel and Hamel (1985).

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

273

Hamel

Figure 6. Brilliant pumping station, April 29, 1907. The outer edge of 1904–1905 Brilliant Cut is at the right. From Wikimedia (2020).

original Brilliant Cut (Figure 6). This outer edge of the cut is clearly in colluvium, and there are numerous colluvial bulges and boulders on the slope extending down to the Pumping Station. This colluvium is also visible on 1929 Allegheny County photographs described below. No subsurface information and foundation construction records on the Brilliant Pumping Station are presently available. Allegheny River Boulevard Construction In 1929, Allegheny County arranged with the Pennsylvania Railroad to construct Allegheny River Boulevard (Figure 2). This involved relocating the railroad tracks farther into the hillside at Brilliant. The original (1904–1905) tracks had a sharp 15° curve, which only required a slight sidehill cut. Track relocation flattened the curvature to 11°, moved the railroad right-of-way 70 ft (21 m) into the hillside, and required a rock cut of 310,000 yd3 (237,000 m3 ) (Ackenheil, 1954). This rock excavation, together with other construction along this section of Allegheny River Boulevard, was done in 1930–1931. During excavation at Brilliant Cut, a slide of 12,000 yd3 (9,200 m3 ) occurred on August 12, 1930, and another slide of unspecified size occurred on April 15, 1931 (Ackenheil, 1954). In 1997, I found more than 200 high-quality black and white photographs of the Allegheny River Boulevard construction, all dated and labeled, in files of the Allegheny County Engineering Department in Pittsburgh. These photographs, which included the slope excavation at Brilliant Cut, clarified several issues. I inspected these photographs in 1997–1998 and checked them again in 2019. The Duquesne coal and Ames Limestone (Figures 4 and 5) are key marker beds visible in some of these photographs.

274

Figure 7. View east at 1904–1905 Brilliant Cut, March 19, 1929. Rephotographed from photograph by Allegheny County Department of Public Works.

Photographs taken on March 19, 1929, prior to Allegheny River Boulevard construction, show the original Brilliant Cut with a ridge of colluvium along its riverward side and partially excavated rockslide debris in its lower portion (Figure 7). Other photographs taken from 1929 to 1931 show extensive deposits of partially eroded colluvium and rockslide debris along the valley wall east of Brilliant Cut, as also shown by Pomeroy (1979). Photographs taken from 1930 to 1931 show the excavation of Brilliant Cut with various shallow slides, including the previously mentioned slides of August 1930 (Figure 8) and April 1931 (Figure 9). All of the 1930–1931 photographs show excavation of dry, very broken rock from the slope toe using power shovels with dipper and clamshell buckets. There are no indications that drilling and blasting was used to

Figure 8. View south at the slide during the excavation of Brilliant Cut, September 4, 1930. Re-photographed from photograph by Allegheny County Department of Public Works.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

Brilliant Cut Slide Re-evaluation

Figure 9. View south at the slide during the excavation of Brilliant Cut, April 7, 1931. Re-photographed from photograph by Allegheny County Department of Public Works.

break up any of the rock for excavation. All of these observations indicate that the 1930–1931 excavation, like the 1904–1905 excavation, was done in colluvium and rockslide debris and/or in rock extensively broken by stress release.

Figure 10. Close-up: April 23, 1938, Aerial Photograph 33-59 showing Brilliant Cut. Re-photographed from photograph by Allegheny County Department of Public Works.

April 23, 1938, Aerial Photographs The Allegheny County Department of Public Works obtained aerial photographs of the entire county from 1936 to 1938. Brilliant Cut and nearby areas were photographed on April 23, 1938. I found and inspected stereo-pairs of these photographs at the Allegheny County Engineering Department in 1997 and checked them again in 2019. These are high-quality black and white aerial photographs at a scale of 1:9,600. Scrutiny of pairs of photographs with a stereo-viewer shows many interesting details, but the gross features described next can be seen on the close-up in Figure 10. Figure 10 shows a fresh scar about 250 ft (76 m) wide along the slope just east of Brilliant Cut where a large landslide was excavated during construction of Allegheny River Boulevard (as shown in 1930–1931 construction photographs). This slide scar is visible on the topographic map in Figure 2, and I observed it in the ﬁeld in 1968–1969, 1997, and 2019. This slide scar is clearly visible in some 1941 photographs of the Pittsburgh District, U.S. Army Corps of Engineers, and Allegheny County that are described below. Fieldwork on April 3, 1997, revealed slumped sandstone beds dipping 10°–20° south into the valley wall east of the slide scar at an approximate Elevation 950 ft (290 m). Light-toned graded areas extend south from the slide scar and Brilliant Cut along the west side of the plateau (Figure 10). These areas were apparently

disturbed during the placement of ﬁll or excavation spoil from nearby construction activities, most likely from the Veterans Administration Hospital then being constructed on a hilltop ca. 0.5 mile (0.8 km) eastsoutheast of Brilliant Cut (Figure 2), as shown in other 1938 aerial photographs. The spoil piles are not clear on the topographic map in Figure 2, but I observed them in the ﬁeld in 1997 and 2019. Four circular features on the hilltop southeast of Brilliant Cut (Figure 10) are ring dikes from tanks formerly existing there (Ferguson, 1968). Depressions inside the ring dikes show the locations of the former tanks. These ring dikes are visible on the topographic map in Figure 2, and they (or their tanks) are shown as black dots on the 1903–1904 topographic map (Plate I of Johnson, 1929). A shallow slide scar ca. 150 ft (46 m) wide exists on the newly excavated slope at the west edge of Brilliant Cut (Figure 10). This scar extends from the level of the Duquesne coal down to the base of the Pittsburgh red beds (Figures 4 and 5). The most important features in Figure 10, however, are several open, somewhat linear cracks on the top of the slope at Brilliant Cut. These cracks extend over a distance of approximately 200 ft (60 m), apparently along vertical stress-release joints. The highest crack apparently corresponds to the vertical stress-release joint in the Birmingham shale that opened during the

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

275

Hamel

Figure 11. View southwest at Brilliant Cut, July 1940. Dc = Duquesne coal, AL = Ames Limestone. Re-photographed from photograph by Pittsburgh District, U. S. Army Corps of Engineers.

1930s and formed the scarp of the 1941 slide at Brilliant Cut (Figures 3 and 5). These cracks clearly indicate that the rock nose at Brilliant Cut had loosened by April 1938, but it had not yet collapsed. The rock nose continued creep movement with cracks opening, i.e., progressive failure, from April 1938 to March 1941, when collapse occurred. This collapse was apparently triggered by water pressures in the slope, as indicated by the ice formations and frozen crust in March 1941 photographs described next. The 1938 aerial photographs show the slope surface to be dry, with no visible seepage or ponded water, despite their April 23 date. Records show Pittsburgh had average precipitation in 1937; January 1938 was very dry, but February through April 1938 had typical precipitation. The only indications of wetness seen on the aerial photographs were dark-toned areas of apparent dampness in the bottoms of three ring dike depressions on the hilltop southeast of Brilliant Cut. Photographs of 1941 Slide As noted previously, the Pittsburgh District, U.S. Army Corps of Engineers, studied Brilliant Cut in 1940–1941 during investigations for a large spillway cut in similar rocks. Ferguson (1968) provided highquality black and white photographs of the slide and cleanup: one taken in July 1940 before the slide and 33 taken from March 20 to April 25, 1941. The July 1940 photograph and several 1941 photographs were included in Hamel (1969, 1972). The July 1940 photograph (Figure 11) shows the slope face with small, dark-toned zones of dampness below the Duquesne coal and Ames Limestone. These zones of dampness presumably resulted from the rel-

276

Figure 12. View south at the slide at Brilliant Cut, March 20, 1941. Dc = Duquesne coal, AL = Ames Limestone. Re-photographed from photograph by Pittsburgh District, U. S. Army Corps of Engineers.

atively heavy precipitation of March–June 1940. The Birmingham shale at the top of the slope appears disturbed in July 1940, but the open vertical cracks visible in the 1938 aerial photographs cannot be seen in the July 1940 photograph. This photograph shows no clear indications of progressive failure in the slope face. Pittsburgh District photographs taken on March 20, 1941, show seepage and ice formations at the levels of the Duquesne coal and Ames Limestone and small patches of dampness and snow elsewhere, but overall, the slope and slide debris surfaces appear dry (Figure 12). These March 20, 1941, photographs show a frozen crust of talus and underlying fractured rock about 2 ft (0.6 m) thick on slide debris below the level of the Ames Limestone. This crust was broken along roughly vertical cracks by the slide movement. A thinner frozen crust existed locally below the Duquesne coal (Figure 12). In 1968–1969, I had not recognized this frozen crust nor its signiﬁcance. Pittsburgh District photographs taken on March 26, 1941, during excavation of slide debris show this frozen crust as well as ice formations and minor seepage at Duquesne coal and Ames Limestone levels. Some of these photographs show that the vertical stress-release joint at the slide scarp extended down to Ames Limestone level (Figure 13). Other Pittsburgh District photographs taken from March 27 to April 25, 1941, show the scarp crack extending down to Ames Limestone level as well as minor seepage from the Duquesne coal and Ames Limestone. These photographs show power shovels with dipper and clamshell buckets excavating debris from the slide toe. This debris looks dry and broken from stress-release mechanisms as well as slide movements.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

Brilliant Cut Slide Re-evaluation

Figure 13. View south at the slide at Brilliant Cut, March 26, 1941. Dc = Duquesne coal, AL = Ames Limestone. Re-photographed from photograph by Pittsburgh District, U. S. Army Corps of Engineers.

Rocks exposed in the slide scarp, particularly below Duquesne coal level, appear essentially intact with only minor fracturing. Eight color slides taken in March–April 1941 (Ackenheil, 1998) show features similar to those described earlier. From March 20 to May 19, 1941, the Allegheny County Department of Public Works took approximately 100 black and white photographs of the slide and its cleanup. Some of these photographs show more detailed views of the frozen crust on the lower portion of the slide mass, displaced railroad tracks at the slide toe, patches of snow and ice on the slope, and slide debris on Allegheny River Boulevard (Figure 14).

Figure 14. View southwest at the slide at Brilliant Cut, March 20, 1941. AL = Ames Limestone. Re-photographed from photograph by Allegheny County Department of Public Works.

These photographs also show more detail on the slide debris cleanup. Jackhammers were used to break up some sandstone boulders in the lower part of the slide, but, overall, the slide debris was extremely broken, as in Pittsburgh District photographs. Some Allegheny County photographs show minor seepage from the Duquesne coal and Ames Limestone, but the slide debris appears dry, as in Pittsburgh District photographs. Compared with many other natural and excavated slopes in similar rocks that I have observed over the past 50 years, the exposed rocks shown in the 1929– 1931 and 1941 photographs of Brilliant Cut are extremely dry. The rocks at Brilliant Cut were fractured by valley stress release, then further broken and loosened by progressive failure. The areas of both surface water runoff and groundwater recharge at Brilliant Cut are small (Figure 2). As mentioned earlier, precipitation prior to the 1941 slide was also small. I find it hard to envision that water in the fractured rock of the slope would have risen above the top of the frozen crust at Ames Limestone level (Figures 12–14). The frozen crust itself probably resulted from re-freezing of snow meltwater during the previously mentioned period of temperature fluctuation in mid-March 1941. The icicles at Duquesne coal level probably resulted from localized perched water zones. PROGRESSIVE FAILURE The original railroad slope at Brilliant Cut was excavated in 1904–1905. Then, in 1930–1931, additional railroad slope excavation extended approximately 70 ft (21 m) farther into the rock nose at Brilliant Cut. This excavation removed lateral support, initiated stress release, and led to progressive failure of the slope that continued at least until July 1940. The slope then failed catastrophically on March 20, 1941, when water levels in rock fractures reached critical levels. Brooker and Peck (1993) presented a mechanism of progressive slope failure that is consistent with the behavior of Brilliant Cut ca. 1905–1940. Essential elements of their mechanism are flat-lying sedimentary rocks, including clays or clay shales, with post-failure residual strengths much less than peak strengths; substantial overconsolidation of the clays or clay shales with respect to the general level of uplands; and a valley topography along which lateral pressure in stratified material at the slope face is reduced to zero. These conditions are met for the flat-lying, interbedded strong and weak Pennsylvanian age sedimentary rocks, including claystones and clay shales, along the Allegheny River valley at Brilliant Cut. The soft clay shale and claystone at the base of the slope (Figure 5) are heavily overconsolidated from overburden eroded long ago. My experience elsewhere in the region (e.g.,

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

277

Hamel

Figure 15. Basic features of the stability model. H = slope height, K = coeﬃcient of earth pressure at rest γ = eﬀective unit weight of soil or rock, S = total resisting shear strength along shear surface, L = length of near horizontal shear surface. Re-drawn from Brooker and Peck (1993).

Hamel, 1969, 1972; Hamel and Adams, 1981; and Hamel, 1998) indicates that these argillaceous materials have residual strengths well below peak strengths. Erosion by the Allegheny River during Pleistocene time reduced lateral pressure at the slope face to zero, and the 1904–1905 and 1930–1931 railroad slope excavations removed lateral support farther back into the slope. Basic features of Brooker and Peck’s (1993) stability model are shown in Figure 15. Along the riverbank, conﬁning stresses in the ground have been substantially reduced, whereas at some distance back in the valley wall, high lateral stresses from overconsolidation still exist. Equilibrium along a horizontal or subhorizontal bedding plane is maintained by shearing stresses along the bedding plane. If the shearing resistance along the bedding plane is exceeded, slip occurs and the shear strength is reduced. With large enough movement, the shear strength is reduced to the residual value (Figure 16). Portions of bedding planes along

Figure 16. Shear stress–displacement relationship. Re-drawn from Brooker and Peck (1993).

278

Figure 17. Formation of bedding plane shears. Re-drawn from Brooker and Peck (1993).

which shear strength has been reduced to residual values are termed bedding plane shears (Figure 17). This stability model is consistent with Harry Ferguson’s theory of valley stress release in ﬂat-lying sedimentary rocks (Figure 18) developed in the vicinity of Pittsburgh (Ferguson, 1967; Ferguson and Hamel, 1981; and Hamel, 2018). It is also consistent with the Pleistocene rockslide failure mechanism of Hamel (1998). The potential surface of sliding (d-a-b-c, Figure 17) of a future slide may be caused by further river erosion or construction excavation. Along this surface of sliding, the mobilized shear resistance depends on shear strains or displacements. Near the toe, where bedding plane shears may exist, displacements may have already reduced strengths to residual levels. Farther back in the slope, shear strengths may approach peak levels (Figure 16). At some point a, the potential surface of sliding can depart from the weak beds and extend up and across overlying beds to form a scarp at the ground surface. Brooker and Peck (1993) consider that

Figure 18. Schematic valley wall cross section. Re-drawn from Ferguson and Hamel (1981).

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

Brilliant Cut Slide Re-evaluation

a slide occurs only when an escarpment appears at d in Figure 17. Brooker and Peck (1993) note that an excavation made near the toe in Figure 17 shortens the potential failure surface, increases stresses along this surface, and causes additional shear deformation along this surface. This deformation will increase mobilized shear resistance along portions of the failure surface where the peak strength has not yet been reached. Eventually, the peak strength will be fully mobilized along the failure surface, and with a significant post-peak strength decrease (Figure 16), failure can occur rapidly, with an escarpment formed at the ground surface. Brooker and Peck (1993) term this a “first-time slide” resulting from progressive failure. Brooker and Peck (1993) further note that after a first-time slide has occurred and the strength on the failure surface has been reduced to its residual value, the failure mass rests in a metastable condition. In this case, small increases in driving forces or reductions in resistance can cause additional movement. With the shearing resistance at residual level, these movements occur slowly as force imbalances develop, in contrast with rapid failure that occurs at the time of the initial slip in a first-time slide. Thus, the stability model and progressive failure concepts of Brooker and Peck (1993) explain the behavior of the slope at Brilliant Cut from 1904 to 1941. The original slope excavation in 1904–1905 initiated progressive failure that broke down the rock mass to permit excavation without blasting in 1930–1931. The additional excavation in 1930–1931 initiated a second phase of progressive failure, which further ruptured and loosened rock as observed on the April 1938 aerial photographs and in the field in July 1940. This second phase of progressive failure reduced the strength along the 1941 failure surface to a value below peak resistance. The strength along the argillaceous basal portion of this failure surface was probably reduced to a residual value, whereas the strength along the inclined rear portion of this failure surface (Figure 5) was probably reduced to a post-peak, but not necessarily residual, value. The slope did not collapse fully, however, until March 20, 1941. The 1941 failure was triggered by water pressures that rose to levels not previously experienced by the slope due to frozen drainage outlets on the slope face (Terzaghi, 1950). These higher water pressures overcame peak or post-peak strengths existing on portions of the failure surface where the strengths had not yet been reduced to residual levels. Large movements of this first-time slide mass reduced strength along the entire failure surface to a field residual level (Hamel, 1976; Hamel and Spencer, 1984) as

indicated by slow, continuous, slumping of the failure mass as its toe was excavated. WATER LEVELS IN SLOPE AT FAILURE Water Elevations Inferred in the Past Water in the open vertical stress-release joint at the rear of the failure mass (Figure 5) was thought to have triggered the 1941 slide when drainage outlets on the slope face were frozen. The critical question is: How high was the water level in the joint (and elsewhere in the failure mass) at the time of failure? Philbrick (1953, 1960) and Ackenheil (1954), both of whom inspected the slide, concluded that the open joint at the rear of the failure mass was full or nearly full of water. In my 1968–1969 analyses, I used 100 ft (30 m) of water in the joint. This water extended from assumed joint bottom Elevation 855 ft (261 m) up to ground surface Elevation 955 ft (291 m) at the top of the joint in Figure 5. I now believe that this water level was too high. Geologic Considerations My observations at numerous sites with valley wall stress-release fractures (e.g., Ferguson and Hamel, 1981) and of numerous rock slides related to valley wall stress release (e.g., Hamel, 1998; Hamel et al., 1998) over the past 45 years suggest that vertical or near-vertical stress-release fractures in Birmingham shale at Brilliant Cut would have been too open to hold water to a level much above the Duquesne coal (Figure 5). This inference is supported by the history of the joint at the rear of the failure mass that opened to a width of more than 1 ft (0.3 m) during the 1930s and an additional several inches (millimeters) by 1940 and is supported by photographs, particularly the aerial photographs of April 23, 1938 (Figure 10). This inference is further supported by ﬁeld observations of open joints in Birmingham shale along the Allegheny River valley wall east of Brilliant Cut in March 2019. The frozen crust below the level of the Ames Limestone at nominal Elevation 850 ft (259 m) observed on March 20, 1941, photographs (Figures 12 –14) suggests that water in the slope may have risen as high as this level prior to failure. It seems unlikely that water in the slope rose much higher than the Ames Limestone because no continuous ice formations above this level are seen on the March 20, 1941, photographs. The scattered ice formations at the level of Duquesne coal, nominally at Elevation 880 ft (268 m), probably represent local water discharge perched on coal underclay. Duquesne coal level can be taken as an upper bound on the water level in the slope at the time of failure.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

279

Hamel

Figure 19. Brilliant Cut slide and drainage areas on a 1962 City of Pittsburgh Topographic Map. Elevations are in feet; to convert to meters, multiply by 0.305.

The lower bound on water level at failure is the condition of no water on the failure surface. Without some water, however, to reduce frictional resistance along the failure surface and/or push the failure mass, this mass would simply have remained in place as shown in the April 1938 aerial photograph and the July 1940 ground photograph (Figures 10 and 11, respectively). Hydrologic Considerations The slide area shown on Plate 14 of Ackenheil (1954) and the drainage area above this slide area are plotted on a 1962 City of Pittsburgh Topographic Map in Figure 19. Both of these areas are very small. The slide area is approximately 60,000 ft2 (5,600 2 m ), and the drainage area is approximately 20,000 ft2 (1,900 m2 ). Pre-slide topography from Plate 13 of Ackenheil (1954) was overlaid on post-slide topography from Plate 14 of Ackenheil (1954) to determine the area of pre-slide topography below Elevation 850 ft (259 m) veneered by frozen crust. This area was computed to be approximately 34,000 ft2 (3,200 m2 ). Thus, the portion of the slide area above Elevation 850 ft (259 m) is approximately 26,000 ft2 (2,400 m2 ).

280

Pittsburgh precipitation records indicate long-term average annual precipitation is approximately 37 in. (940 mm). The years 1938 and 1939 were drier than average with total precipitations of 33.22 in. (844 mm) and 31.59 in. (802 mm), respectively. The year 1940, with total precipitation of 41.61 in. (1057 mm), was wetter than average, primarily as a result of heavy rain in April and June. January 1941, with 2.97 in. (75 mm) of precipitation, was about average, whereas February 1941, with 0.37 in. (9 mm) of precipitation, was much drier than average. Precipitation with a water equivalent of 1.60 in. (41 mm or 0.13 ft) fell from March 1 to 19, 1941. Thus, there was relatively little water falling on the slide area immediately prior to the slide of March 20, 1941. Most of this water would have run oﬀ the relative steep slopes (Figure 19). This is particularly true for the portion of the slide area below Elevation 850 ft (259 m) covered by frozen crust. As noted earlier, the drainage area is approximately 20,000 ft2 (1,900 m2 ) and the portion of the slide area above the frozen crust is approximately 26,000 ft2 (2,400 m2 ). The total area of potential inﬁltration is, therefore, approximately 46,000 ft2 (4,300 m2 ).

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

Brilliant Cut Slide Re-evaluation

The 1941 slide had a volume greater than 110,000 yd3 (84,000 m3 ). If the entire 0.13 ft (41 mm) of water equivalent from March 1 to 19, 1941, infiltrated the area of 46,000 ft2 (4,300 m2 ), the volume of water would be 6,000 ft3 (170 m3 ). This is approximately 0.20 percent of the slide volume—not enough to raise the water level in the fractured and loosened rock of the slide mass very much. The volume of long-term, steady-state groundwater flow into the slide mass from the Duquesne coal and Ames Limestone would not have been very great either, based on numerous photographs of the slope taken from 1930 to 1941 and my field observations from 1968–2019. This 6,000 ft3 (170 m3 ) water volume can also be compared to the probable volume of the open joint at the rear of the 1941 failure mass. Plate 13 of Ackenheil (1954) shows pre-slide topography with a 110 ft (33.5 m) long joint at Elevation 950 ft (290 m) at the rear of the failure mass labeled “opens again one foot [0.3 m] before slide.” April 1938 aerial photographs (Figure 10) suggest this joint may have been as long as 200 ft (60 m). Post-slide photographs (Figure 13) show that this joint extended down to the Ames Limestone at nominal Elevation 850 ft (259 m). The volume of an open joint 110 ft (33.5 m) long by 100 ft (30 m) deep by 1 ft (0.3 m) wide is 11,000 ft3 (311 m3 ). The water volume of 6,000 ft3 (170 m3 ) would have filled this joint a little more than half full with none to spare for the remainder of the failure mass. If the joint were 200 ft (60 m) long, as suggested by the aerial photographs, with the previously mentioned depth and width, the water volume of 6,000 ft3 (170 m3 ) would have only filled the joint about a third full with none to spare for the remainder of the failure mass. The actual volume of water available here was much less than 6,000 ft3 (170 m3 ) because much of the water from the March 1–19, 1941, precipitation certainly ran off the slope and some was lost to evaporation. This semi-quantitative hydrologic analysis suggests that the water level in the March 20, 1941, slide was relatively low and certainly nowhere near the top of the open joint at the rear of the failure mass.

STABILITY ANALYSES FOR PROGRESSIVE FAILURE General Here, I use the previously mentioned concepts to provide the basis for additional stability analyses to clarify aspects of progressive failure and ﬁnal catastrophic failure of Brilliant Cut. The analyses were done to assess shear strengths along failure surface segments before and during the 1941 slope failure and

to estimate water levels that existed in the slope at the time of failure. I used the slope cross section in Figure 5 for these two-dimensional limiting equilibrium stability analyses. This same cross section was used in my 1968–1969 analyses (Hamel, 1969). Two rock unit weights were considered. First, analyses were done with a typical intact rock unit weight of 160 lb/ft3 (25.1 kN/m3 ), as used in my 1968–1969 analyses. Then, analyses were done with a unit weight of 145 lb/ft3 (22.8 kN/m3 ) to reflect rock breakage and loosening by stress release and progressive failure. This nominal 9 percent reduction in rock unit weight affected both the driving and resisting forces in the sliding stability analyses so that results were essentially the same for both assumed unit weights. The Sliding Block (Wedge) method of stability analysis (Lambe and Whitman, 1969; Duncan and Wright, 2005) was used. This two-dimensional limiting equilibrium method was used fairly extensively in the pre-computer era for stability analyses of soil and rock slopes. It remains useful for checking computer analyses, analyzing non-circular failure surfaces, and performing specialized analyses such as progressive failure. With the Sliding Block method, the failure surface in a cross section is approximated by two to five straight-line segments with vertical inter-block boundaries. Free body diagrams show forces acting on each block, and force equilibrium of each block is considered. The problem is statically indeterminate, so the inclination of the inter-block forces and the shear strength mobilized on the base of one of the blocks are assumed. The forces acting on each block can then be determined. For a given friction angle (and cohesion intercept, if present), the equilibrium of each block is determined analytically or graphically. Solutions are iterated with different friction angles until all blocks are in equilibrium. This gives a mobilized friction angle. With frictional materials, the factor of safety against sliding is the ratio of the tangent of the available friction angle to the tangent of the mobilized friction angle. With cohesive materials, additional iteration is necessary to obtain the same factor of safety relative to cohesion. Preliminary analyses were done with three blocks corresponding to three linear segments approximating the failure surface in Figure 5. The vertical tension crack at the rear of the failure mass was extended down to Elevation 851 ft (259.4 m) here (Figure 20). I used this three-block system in my 1968–1969 analyses of Brilliant Cut (Hamel, 1969) in order to compare its results with those of the more accurate Morgenstern and Price (1965, 1967) method. My 1968–1969 analyses showed that the Sliding Block method with three

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

281

Hamel

Figure 20. Analysis cross section: three-block system.

blocks and inter-block forces inclined at 45° to the horizontal (approximate inclination of pre-failure slope at Brilliant Cut) gave results within 2 percent of those of the Morgenstern-Price method (Hamel, 1969). This 45° inter-block force inclination was used in all the analyses described next. Subsequent analyses were done with ﬁve blocks (Figure 21), corresponding to ﬁve linear segments

defining the failure surface in Figure 5, to better characterize this failure surface and to investigate progressive failure. Here again, the tension crack at the rear of the failure mass extended down to Elevation 851 ft (259.4 m). Note that Blocks 1 and 2 in the five-block system correspond to Block 1 in the three-block system, Blocks 3 and 4 in the five-block system correspond to Block 2 in the three-block system, and Block

Figure 21. Analysis cross section: ﬁve-block system.

282

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

Brilliant Cut Slide Re-evaluation

5 in the five-block system corresponds to Block 3 in the three-block system. In the present analyses, as in the 1968–1969 analyses, calculations were done by hand. Hand calculations enable better visualization of force interactions and progressive failure development than computer calculations in situations like this. Numerical computations were done with a calculator, and graphical solutions were used for force equilibrium. Preliminary Analyses for the Three-Block System Initial analyses with no water in the slope gave a zero-cohesion friction angle of 15° required along the entire failure surface for limiting equilibrium with both of the previously mentioned rock unit weights. Seventy-five percent of the length of this failure surface along the bases of Blocks 1 and 2 is relatively flat and passes through the basal clay shale and claystone (Figure 20), along which bedding shears with residual strength would likely have developed. This is consistent with the concepts of progressive failure of Brooker and Peck (1993). The 15° friction angle is within the reported range of residual friction angles for similar materials (Hamel, 1969, 1972; Hamel and Adams, 1981). The remaining 25 percent of this failure surface along the base of Block 3 is inclined at 54° to the horizontal and passes through stronger sandstone and shale as well as claystone (Figure 20). The computed 15° frictional angle is too small for the residual strength of these stronger materials sheared across bedding. A more typical residual friction angle for these materials is on the order of 28° to 30° (Deere, 1976; Hamel et al., 1976). Analysis of the three-block system with no water forces, a 15° friction angle along the bases of Blocks 1 and 2, and a 28° friction angle along the base of Block 3 (Figure 20) gave a factor of safety against sliding of 1.20 with both assumed rock unit weights. This implies that the steeply inclined rear portion of the failure surface had some reserve of strength to accommodate water pressures and forces. The three-block system was then analyzed with water forces along the entire failure surface corresponding to static water at Elevations 882 and 851 ft (268.8 and 259.4 m) giving slight heads on the Duquesne coal and Ames Limestone, respectively (Figure 20). With water forces corresponding to Elevation 882 ft (268.8 m), the vertical component of water force on the base of Block 1 exceeds the weight of Block 1 for both rock unit weights. These results imply that Block 1 would lift up and blow out or slide away from the slope toe. With water forces corresponding to Elevation 851 ft (259.4 m) and a rock unit weight of 160 lb/ft3 (25.1 kN/m3 ), a zero-cohesion friction angle of 50° is re-

quired to close the force polygon of Block 1 with no forces from Block 2. With water forces corresponding to Elevation 851 ft (259.4 m) and a rock unit weight of 145 lb/ft3 (22.8 kN/m3 ), a zero-cohesion friction angle of 73° is required to close the force polygon of Block 1 with no forces from Block 2. These friction angles are obviously much larger than those available on the base of Block 1, implying that Block 1 would slide out. Additional calculations showed that, with a rock unit weight 160 lb/ft3 (25.1 kN/m3 ) and water Elevation 861.3 ft (262.5 m), the vertical component of water force on the base of Block 1 equals its weight. This implies that Block 1 will be marginally stable under these conditions. Similarly, with a rock unit weight 145 lb/ft3 (22.8 kN/m3 ) and water Elevation 854.7 ft (260.5 m), Block 1 will be marginally stable. These results strongly suggest that the 1941 failure of Brilliant Cut was triggered by water pressures confined in the slope by the frozen crust face below Ames Limestone level, nominal Elevation 850 ft (259 m). Further analyses were done with the five-block system as described next. Analyses with Five-Block System The five-block system (Figure 21) has 72 percent of its failure surface length fairly flat and passing through the clay shale and claystone on the bases of Blocks 1– 4. The remaining 28 percent of the failure surface is steeply inclined along the base of Block 5. Analyses with the five-block system were done with no water and several assumed static water levels above the basal failure surface (bases of Blocks 1–4) at nominal Elevation 790 ft (240.9 m). Water forces along the entire failure surface were computed from each water level. Most analyses were done only with the rock unit weight of 160 lb/ft3 (25.1 kN/m3 ) because previous analyses had shown a negligible difference in results obtained for sliding with the reduced unit weight of 145 lb/ft3 (22.8 kN/m3 ). Analysis of the baseline case of no water in the slope gave a zero-cohesion friction angle of 14.5° required along the entire failure surface for limiting equilibrium. This result is very similar to the 15° friction angle calculated with the three-block system. Assumed water Elevation 800 ft (243.8 m) gave very small water forces on Blocks 1–4 and no water forces on Block 5. A zero-cohesion friction angle of 15° is required along the entire failure surface for limiting equilibrium. As noted earlier, this is a reasonable residual level friction angle for the clay shale and claystone on the bases of Blocks 1–4. Assumed water Elevation 810 ft (246.9 m) gave somewhat larger water forces on Blocks 1–4 and very small water forces on Block 5. With a zero-cohesion

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

283

Hamel

friction angle of 15° mobilized on the bases of Blocks 1–4, a zero-cohesion friction angle of 27° is required along the base of Block 5 for limiting equilibrium of the failure mass. This latter friction angle is consistent for residual strength of rocks along the base of Block 5 (Deere, 1976; Hamel et al., 1976). With assumed water Elevation 820 ft (249.9 m), Block 1 has very large water forces compared to its weight, based on a rock unit weight of 160 /lb/ft3 (25.1 kN/m3 ). A zero-cohesion friction angle of 31° is required to close the force polygon for Block 1 without any forces acting on it from Block 2. This friction angle exceeds the range of residual friction angles that can reasonably be expected for the clay shale and claystone along the base of Block 1, so Block 1 would likely slide out of the slope. With Block 1 gone and a zerocohesion friction angle of 15° mobilized on the base of Blocks 2–4, a zero-cohesion friction angle of 32.5° is required along the base of Block 5 for limiting equilibrium. This latter friction angle is on the high end of the range of friction angles expected for residual strength of rocks along the base of Block 5 (Deere, 1976; Hamel et al., 1976) and may correspond to post-peak sliding before residual strength is reached. Additional calculations showed that, with a rock unit weight of 160 lb/ft3 (25.1 kN/m3 ), the vertical component of water force on the base of Block 1 will equal the block weight with water Elevation 825.6 ft (251.6 m) and, with a rock unit weight of 145 lb/ft3 (22.8 kN/m3 ), the vertical component of water force on Block 1 will equal the block weight with water Elevation 822.6 ft (250.7 m). These results, along with those above for sliding of the ﬁve-block system, indicate that relatively low water levels in the failure mass were required to initiate the 1941 collapse of Brilliant Cut. CONCLUSIONS Valley stress release, progressive failure, and subsequent sliding broke up rock in the walls of the main Allegheny River valley and the tributary abandoned valley at Brilliant Cut. One of these rock slide masses, with overlying and adjacent colluvial soil material, was partially excavated in 1904–1905 for the original railroad cut in the rock nose at Brilliant. Colluvial soil remnants and a substantial portion of the old rockslide mass were excavated in 1930–1931 as Brilliant Cut was extended approximately 70 ft (21 m) farther into the rock nose. This latter excavation removed lateral support and allowed a portion of the rock mass to de-stress and creep out, progressively opening vertical stress-release joints along the slope crest. By April 23, 1938, the rock nose at Brilliant had loosened but it had not yet collapsed.

284

Calculations presented earlier illustrate the mechanism of progressive failure and final collapse of the slope at Brilliant Cut. Stress release after slope excavation in 1930–1931 initiated progressive failure that continued to 1940–1941. Movements accompanying this progressive failure reduced the strength of clay shale and claystone along the basal failure surface (bases of Blocks 1–4, Figure 21) to a residual value characterized by a friction angle on the order of 15°. This residual strength was sufficient to keep the failure mass marginally stable with low water levels in the slope; higher-strength materials along the steeply inclined rear portion of the failure surface (base of Block 5, Figure 21) provided this marginal stability. Then in March 1941, an icy frozen crust developed on the lower part of the slope face blocking groundwater drainage, and the water level in rock fractures built up to a critical level ca. Elevation 820 ft (250 m). Pressures and forces resulting from this critical, higher water level caused the toe of the failure mass (Block 1) to lift up and/or slide out. Loss of toe support, along with water pressures on the remaining portion of the failure surface, overcame available strength on the steeply inclined rear portion of the failure surface (Block 5). This critical water level, postulated here about 30 ft (9 m) above the basal failure surface, is well below the level previously proposed (Philbrick, 1953, 1960; Ackenheil, 1954; Hamel, 1969,1972) to have existed at time of failure. The lesson here is that if progressive failure significantly reduced strength along the failure surface to a residual or near-residual level, then water did not have to rise very high in rock fractures in the slope to cause the slope collapse on March 20, 1941. This is consistent with the minor amount of precipitation in March 1941 prior to failure, the small areas of surface runoff and surface water infiltration associated with the failure area, and the dry appearance of post-failure slide debris. It has taken me 52 years of intermittent but persistent investigation to develop this most probable scenario for the 1941 rockslide at Brilliant Cut. I have thoroughly enjoyed this challenging and stimulating odyssey. During this time, I continuously revised my hypotheses as more information became available. This is the essence of the scientific method. ACKNOWLEDGMENTS Abdul Shakoor, Co-Editor of Environmental & Engineering Geoscience, and Editorial Assistant Karen Smith provided guidance, assistance, and encouragement in preparation of this paper. John Harper, Pennsylvania Geological Survey (retired); Barry Voight, Emeritus Professor of Geology, Penn State University;

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

Brilliant Cut Slide Re-evaluation

and an anonymous reviewer suggested improvements to the manuscript. Figures 1–5 and 15–21 were redrawn by Jeff Liebdzinski of Rhea Engineers & Consultants, Inc., Valencia, PA. REFERENCES Ackenheil, A. C., 1954, A Soil Mechanics and Engineering Geology Analysis of Landslides in the Area of Pittsburgh, Pennsylvania: Unpublished Ph.D. Dissertation, Department of Geology, University of Pittsburgh, Pittsburgh, PA, 116 p + Appendix. Ackenheil A. C, 1998, personal communication, Senior Engineer, Ackenheil Engineers, Inc., Pittsburgh, PA. Brooker, E. W. and Peck, R. B., 1993, Rational design treatment of slopes in overconsolidated clays and clay shales: Canadian Geotechnical Journal, Vol. 30, No. 3, pp. 526–544. Deere, D. U., 1976, Dams on rock foundations–some design questions. In Rock Engineering for Foundations and Slopes, Vol. II: American Society of Civil Engineers, New York, pp. 55–86. Duncan, J. M. and Wright, S. G., 2005, Soil Strength and Slope Stability: John Wiley & Sons, Inc., Hoboken, NJ. pp. 71–77. Ferguson, H. F., 1967, Valley stress release in the Allegheny Plateau: Bulletin Association Engineering Geologists, Vol. 4, No. 1, pp. 63–71. Ferguson, H. F., 1968, personal communication, District Geologist, Pittsburgh District, U.S. Army Corps of Engineers, Pittsburgh, PA. Ferguson, H. F. and Hamel, J. V., 1981, Valley stress relief in flatlying sedimentary rocks. In Akai, K.; Hayashi, M; and Nishimatsu, Y. (Editors), Weak Rock, Vol. 2: Balkema, Rotterdam, the Netherlands, pp. 1235–1240. Hamel, J. V., 1969, Stability of Slopes in Soft, Altered Rocks: Unpublished Ph.D. Dissertation, Department of Civil Engineering, University of Pittsburgh, Pittsburgh, PA, 305 p. Hamel, J. V., 1972, The Slide at Brilliant Cut. In Cording, E. J. (Editor), Stability of Rock Slopes: American Society of Civil Engineers, New York, pp. 487–510. Hamel, J. V., 1976, Libby Dam left abutment rock wedge stability. In Rock Engineering for Foundations and Slopes, Vol. I: American Society of Civil Engineers, New York, pp. 361–386. Hamel, J. V., 1998, Mechanism of Pleistocene rock slides near Pittsburgh, Pennsylvania: International Journal Rock Mechanics Mining Sciences, Vol. 35, No. 4–5, Paper No. 32. Hamel, J. V., 2018, Harry Ferguson’s theory of valley stress release in flat-lying sedimentary rocks. In Shakoor, A. and Cato, K. (Editors), IAEG/AEG Annual Meeting Proceedings, San Francisco, California, 2018, Vol. 2: Springer Nature, Cham, Switzerland, pp. 121–127. Hamel, J. V. and Adams, W. R., Jr., 1981, Claystone slides, Interstate Route 79, Pittsburgh, Pennsylvania, USA. In Akai, K.; Hayashi, M; and Nishimatsu, Y. (Editors), Weak Rock, Vol. 1: Balkema, Rotterdam, the Netherlands, pp. 549–553. Hamel, J. V. and Hamel, E. A., 1985, Landsliding in Pennsylvania. In Proceedings Fourth International Conference and Field Workshop on Landslides: Japan Landslide Society, Tokyo, pp. 473–480. Hamel, J. V.; Lasko, J. D.; and Ruppen, C. A., 1998, Rock slope evaluation for the Pittsburgh Airport Busway. In Moore, D. P. and Hungr, O. (Editors), Engineering Geology: A Global View

from the Pacific Rim, Vol. 5: Balkema, Rotterdam, the Netherlands, pp. 3121–3128. Hamel, J. V.; Long, S. B.; and Ferguson, H. F., 1976, Mahoning Dam foundation re-evaluation. In Rock Engineering for Foundations and Slopes, Vol. I: American Society of Civil Engineers, New York, pp. 217–244. Hamel, J. V. and Spencer, G. S., 1984, Powerhouse slope behavior, Fort Peck Dam, Montana. In Prakash, S. (Editor), Proceedings, International Conference on Case Histories in Geotechnical Engineering, Vol. 2: University of Missouri - Rolla, Rolla, MO, St. Louis, MO, pp. 541–551. Jacobson, R. B.; Elston, D. P.; and Heaton, J. W., 1988, Stratigraphy and magnetic polarity of the high terrace remnants in the Upper Ohio and Monongahela Rivers in West Virginia, Pennsylvania, and Ohio: Quaternary Research, Vol. 29, pp. 216–232. Johnson, M. E., 1929, Pittsburgh Quadrangle: Pennsylvania Geological Survey Atlas No. 27, 236 p. Lambe, T. W. and Whitman, R. V., 1969, Soil Mechanics: John Wiley & Sons, Inc., New York, pp. 366–369. Leverett, F., 1902, Glacial Formations and Drainage Features of the Erie and Ohio Basins: U.S. Geological Survey Monograph XLI, 781 p. Leverett, F., 1934, Glacial Deposits Outside the Wisconsin Terminal Moraine in Pennsylvania: Pennsylvania Geological Survey Bulletin G-7, 123 p. Morgenstern, N. R. and Price, V. E., 1965, The analysis of the stability of general slip surfaces: Geotechnique, Vol. 15, No. 1, pp. 79–93. Morgenstern, N. R. and Price, V. E., 1967, A numerical method for solving the equations of stability of general slip surfaces: Computer Journal, Vol. 9, No. 4, pp. 388–393. Philbrick, S. S., 1953, Design of deep rock cuts in the Conemaugh Formation. In Proceedings Fourth Symposium on Geology as Applied to Highway Engineering: State Road Commission of West Virginia, Charleston, WV, pp. 79–88. Philbrick, S. S., 1960, Cyclic sediments and engineering geology. In Proceedings Twenty-First International Geological Congress, Pt 20: Det Berlingske Bogtrykkeri, Copenhagen, Denmark, pp. 49–63. Pomeroy, J. S., 1979, Landslide Susceptibility Map of the Pittsburgh East 7.5’ Quadrangle, Allegheny County, Pennsylvania: U.S. Geological Survey Open File Map 79-1314. Scharff, M. R., 1920, Bigelow Boulevard Slide in Pittsburgh: Engineering News-Record, Vol. 85, No. 23, pp. 1076–1080. Skempton, A. W., 1964, Long-term stability of clay slopes:Geotechnique, Vol. 14, No. 2, pp. 77–101. Terzaghi, K., 1950, Mechanism of landslides. In Paige, S. (Editor), Application of Geology to Engineering Practice (Berkey Volume): Geological Society of America, New York, pp. 83–123. Wagner, W. R.; Craft, J. L.; Heyman, L.; and Harper, J. A., 1975a, Greater Pittsburgh Region Geologic Map and Cross Sections: Pennsylvania Geological Survey Map 42. Wagner, W. R.; Heyman, L.; Craft, J. L.; Edmunds, W. E.; and Harper, J. A., 1975b, Greater Pittsburgh Region Structure Contour Map: Pennsylvania Geological Survey Map 43. Wagner, W. R.; Heyman, L.; Gray, R. E.; Belz, D. J.; Lund, R.; Cate, A. S.; and Edgerton, C. D., 1970, Geology of the Pittsburgh Area: Pennsylvania Geological Survey General Geology Report G 59, 145 p. Wikimedia, 2020, https://commons.wikimedia.org/wiki/File: Brilliant_Pumping_Station_(715.071613.CP).jpg

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 269–285

285

An Integrated InSAR-Borehole Inclinometer-Numerical Modeling Approach to the Assessment of a Slow-Moving Landslide MIRKO FRANCIONI* Department of Engineering and Geology, University “G. d’Annunzio” of Chieti-Pescara, Chieti, Italy

DOUG STEAD Department of Earth Sciences, Simon Fraser University, Burnaby, British Columbia, Canada

JAYANTI SHARMA MDA, Richmond, British Columbia, Canada

JOHN J. CLAGUE Department of Earth Sciences, Simon Fraser University, Burnaby, British Columbia, Canada

MARC-ANDRÉ BRIDEAU BGC Engineering, Vancouver, Canada; now at Westrek Geotechnical Services Ltd, Squamish, British Columbia, Canada

Key Terms: Slope Stability, InSAR, Inclinometer, Slope Monitoring, Numerical Modeling, GIS, Engineering Geomorphology ABSTRACT We use results of satellite-based interferometric synthetic aperture radar, Global Positioning System, and borehole inclinometer data to constrain numerical models that improve understanding of slope deformation at the Alexandria landslide, British Columbia, Canada. Surface monitoring data and borehole slope inclinometer measurements provide important insight into the slope failure mechanism. We initially analyzed the data in a geographic information system (GIS) to create thematic maps of the landslide area (hillshade, slope, and aspect), to identify key geological features, and to produce an engineering geomorphology map of the landslide. The monitoring data and the geological/engineering geomorphological models provide important constraints for two-dimensional landslide limit equilibrium and finite difference analyses. The initial limit equilibrium analysis improved understanding of the sliding surfaces. The finite difference models were

*Corresponding author email: mirko.francioni@unich.it

then used to simulate and investigate the potential slope deformation mechanism. The combined slope monitoring/modeling results show that the Alexandria landslide is a slow-moving, multiple-block, retrogressive slope failure. The close agreement between the limit equilibrium and ﬁnite diﬀerence analyses, together with the satellite and ground-based slope monitoring and GIS data, highlight the importance of using a multidisciplinary/integrated approach in landslide studies. INTRODUCTION New slope monitoring technologies and improved numerical analytical techniques over the past two decades have led to a marked increase in the use of remote sensing techniques and numerical simulations in landslide studies (Lorig and Stead, 2018). In contrast, the integrated use of satellite-based interferometric synthetic aperture radar (InSAR), surface Global Positioning System (GPS), and borehole inclinometer monitoring in a geographic information system (GIS) platform to constrain numerical modeling of landslides has received limited attention. In this study, we demonstrate that these techniques, when applied in an integrated manner, can greatly improve understanding of landslide failure mechanisms and slope evolution. Reviews of available state-of-the-art slope analyses and

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

287

Francioni, Stead, Sharma, Clague, and Brideau

their potential application in engineering geology include those of Stead et al. (2006), Lorig and Stead (2018), and Francioni et al. (2018a). Use of numerical modeling in slope analysis requires care to ensure that appropriate analytical methods are selected and input data are representative of ground conditions. It is also important, whenever possible, to constrain the modeling results with surface and borehole deformation data, as such data may provide early warning of incipient catastrophic failure (Scoppettuolo et al., 2020). A comprehensive volumetric slope monitoring system involves deployment of equipment to monitor surface movement (GPS, InSAR, lidar, tiltmeters) and subsurface deformation (inclinometers, extensometers, piezometers, microseismicity systems). Due to the costs involved in deploying such arrays, they are currently used only in large open-pit mines, high-risk dams, and high-risk unstable slopes (e.g., Qi et al., 2004; Tunono et al., 2011; Ramsden et al., 2015; and Lin et al., 2018). For example, the extra cost of such arrays may be warranted in cases where natural gas and oil pipelines cross potentially unstable slopes. Dewar et al. (2016, 2017) provide technical and operational guidelines for monitoring pipelines that cross slow-moving landslides using conventional geotechnical techniques. Guthrie et al. (2018) recently described a toolbox of InSAR techniques for pipeline geohazard investigations. Remote sensing techniques such as airborne lidar and InSAR have been widely used in recent years to overcome the high costs of ground-based monitoring of unstable slopes and, further, to analyze largescale spatial surface deformation (e.g., Sharma et al., 2015, 2016; Mantovani et al., 2016; Bozzano et al., 2017; Bianchini et al., 2018; Francioni et al., 2019; and Raspini et al., 2019). In this study, we acquired and analyzed space-borne InSAR data and reviewed GPS and borehole inclinometer data to document and understand surface and subsurface deformation associated with a deep-seated, slow-moving landslide (Alexandria landslide) crossed by a pipeline in British Columbia. Five inclinometers and approximately 60 GPS stations located along the pipeline corridor were used to deﬁne, respectively, depths of shear surfaces and rates of surface movement of the landslide. High-resolution space-borne InSAR data were acquired using the RADARSAT-2 sensor to understand the areal distribution of slope deformation. We integrated and analyzed all data within a GIS. This tool plays a key role in the study of large landslides, as it enables the creation of thematic maps that can be used to interpret geological and geomorphological features (e.g., Jaboyedoﬀ et al., 2004; Brideau et al., 2011) and to create engineering geomorpholog-

288

ical maps (e.g., Cooke and Doornkamp, 1990; Wolter et al., 2016; and Francioni et al., 2018b). We combined data obtained from the surface and subsurface monitoring systems and GIS analyses to improve understanding of the geometry of the Alexandria landslide and to calibrate and constrain numerical simulations. Two different geomechanical analyses are presented: (1) a limit equilibrium analysis using Slide2 (Rocscience, 2018) to document the geometry of failure surfaces and (2) a 2D continuum numerical simulation using the code FLAC (Itasca, 2018) that allowed us to simulate failure mechanisms and slope deformations constrained by the InSAR, GPS, and borehole inclinometer monitoring systems. The combined use of these two modeling techniques was important because it allowed us to progressively improve knowledge of the behavior of the slope. The limit equilibrium study highlighted the geometry of two sliding surfaces and confirmed the location of backscarps identified both in the field and during preparation of the engineering geomorphology map. This information was included in FLAC numerical model, allowing us to confidently simulate slope behavior. Based on the results of this study, we suggest an iterative and integrated approach in the study of slowmoving landslides. We use a flowchart to suggest a possible integration of monitoring and numerical techniques that has potential for significantly improving the understanding of landslide geohazards. STUDY AREA The Alexandria landslide is located 50 km south of Quesnel, British Columbia, on the east side of the Fraser River (Figure 1). The landslide has a length of about 3.7 km and a width of 2 km, supports a cover of coniferous trees (Figure 2A), and is crossed by a pipeline (Figure 2B). The record of the climate normals of the McLeese weather station (a few kilometers southeast of the Alexandria landslide) indicates that, in the area, the daily temperature during the summer varies from a maximum of 26°C to a minimum of 10°C. During the winter, the temperature drops to a maximum daily temperature of −2°C and a minimum of −10°C. The precipitation range is between about 35 and 50 mm in most months. Only February and March show a lower trend with precipitation up to 15 mm. Five geologic units, ranging in age from Eocene to Pleistocene, are present in the Alexandria area (Figure 3; Rouse and Mathews, 1979; Hora and Hancock, 1994). The oldest unit comprises Eocene basalts (Endako Group) and underlies much of the Fraser River valley within the study area. It is unconformably overlain by lignite, mudstone, diatomite, sandstone,

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

An Integrated InSAR-Borehole Inclinometer-Numerical Modeling Approach to the Assessment

Figure 1. (A) Location of the Alexandria and West Quesnel landslides in British Columbia. (B) Alexandria landslide (delineated by the white line). (C) West Quesnel landslide (delineated by the white line). Images from Google Earth (2020). White arrows indicate the direction of movement of the landslides.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

289

Francioni, Stead, Sharma, Clague, and Brideau

Figure 2. (A) Photograph of the Alexandria landslide area (view to the northeast). Scarps identiﬁed are highlighted; the scarp marking the back of the landslide is delineated by a red line. (B) Pipeline right-of-way on the landslide (view south). The signs mark the location of the buried pipeline.

Figure 3. (A) Geologic map of the study area (after Rouse and Mathews, 1979; Cui et al., 2017). Reference system: NAD 1983, UTM Zone 10N. (B) Geologic column for area shown in (A) (after Hora and Hancock, 1994; AMEC, 2002). (C) Geologic cross section, A–B.

290

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

An Integrated InSAR-Borehole Inclinometer-Numerical Modeling Approach to the Assessment

Figure 4. Locations of (A) borehole inclinometers and (B) Global Positioning System stations on the Alexandria landslide, plotted on a hillshade map (note: boreholes SI0101 and SI0102 are located near each other and cannot be diﬀerentiated at the scale of the image).

and conglomerate, ranging in age from Oligocene (Australian Creek Formation) to Miocene (Crownite and Fraser Bend formations). This clastic sedimentary sequence is overlain by a series of thin, late Miocene basalt flows of the Chilcotin Group (Figure 3B and C). The youngest geologic unit, which caps the plateau surface above the Alexandria landslide, is late Pleistocene in age and comprises till and glaciolacustrine and glaciofluvial sediments deposited during the last period of continental glaciation in British Columbia (the Fraser Glaciation, ∼30,000–11,000 years ago; Clague, 1981). Borehole data show that Alexandria landslide is seated within the Tertiary sedimentary rocks, which reach a depth of up to 160 m below the surface (Figure 3C) and consist of weakly lithified, interbedded conglomerate, sandstone, and mudstone. Three backscarps within the landslide body were identified in the field and verified using aerial photographs and the digital elevation model (Figure 2A). The backscarps are up to 50–60 m high and lie within the displaced Tertiary sedimentary rocks. The trends of the scarps are roughly perpendicular to the direction of movement of the landslide. The Alexandria landslide shares many similarities to the large slow-moving translational West Quesnel landslide in Quesnel, British Columbia (Figure 1A and C), which has damaged a number of houses. This landslide has several slip surfaces that range in depth from about 30 to 70 m and have an average inclination of 4° (AMEC, 2002). GPS annual displacement rates measured over the past 21 years range from 6 mm (in 2010 and 2017) to more than 80 mm (1999 and

2005) (West Quesnel Land Stability Program, 2020). The West Quesnel landslide is seated in the same geologic units as the Alexandria landslide, that is, the Australian Creek, Fraser Bend, and Crownite formations. The slip surfaces are located in weak clay layers within the Tertiary sequence. METHODS We adopted a multidisciplinary approach to improve understanding of the Alexandria landslide. Data from borehole inclinometers were integrated with GIS geomorphic interpretations and InSAR data and used to calibrate 2D numerical simulations. Inclinometers provided information on the depth of the sliding surface and the rate of slope movements, whereas static GPS measurements and InSAR data defined the areal distribution of surface movement. The combined data sets allowed us to establish the slope failure geometry and sliding mechanism, which were then used to constrain numerical simulations of the landslide. Slope Deformation Monitoring GPS and Borehole Inclinometers We analyzed data collected from five slope borehole inclinometers between 2002 and 2008 (Figure 4A). The inclinometers provide horizontal slope displacements over the full depth of the boreholes. The borehole inclinometer data revealed two sliding surfaces. Unfortunately, the five inclinometers are located along the pipeline right-of-way (ROW), which is a straight

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

291

Francioni, Stead, Sharma, Clague, and Brideau

line perpendicular to the direction of movement (Figure 4B); thus, the inclination of the sliding surfaces could not be determined. The 3D form of the sliding surfaces was subsequently estimated through limit equilibrium analyses (see the section “Slope Stability Analyses”). The borehole inclinometer installation was augmented with about 60 GPS stations installed along the pipeline ROW. GPS data used in this research were acquired between 2009 and 2014 (Figure 4B).

Table 1. Summary of RADARSAT-2 InSAR stacks used in this study.

Satellite-Based InSAR

study (SLA15 and SLA10) were acquired in ascending orbital passes and have approximately 2-m ground resolution. Table 1 summarizes the characteristics of the image stacks, and Figure 5 shows the extents of the two radar stacks on an orthoimage of the study area. We used a novel multi-track method (Eppler et al., 2015) to jointly process the InSAR stacks. The method increases the temporal resolution and robustness of the surface displacement estimates compared to estimates obtained by processing each stack individually. In addition to the standard InSAR processing chain (including image co-registration, atmospheric phase compensation, coherence-based ﬁltering, and phase unwrapping; Sharma et al., 2015), we applied spatially adaptive ﬁltering (Parizzi and Brcic, 2011) to suppress phase noise. The output of the InSAR processing chain is a lineof-sight (LOS) time series of displacement for each pixel in the radar data set. To relate the LOS displacements to recorded ground displacements, we assumed that displacements are directly downslope, which is in

We integrated the GPS data set with satellite-based InSAR monitoring data. InSAR is a remote sensing technique that measures the phase diﬀerence (i.e., the interference) between two or more radar data sets. The phase between two radar images acquired with the same viewing geometry at diﬀerent times can be related to the surface displacement between the times of the radar acquisitions. Time series of acquisitions can be combined to better isolate surface displacements from other factors, allowing measurement accuracy of centimeters to millimeters. Rosen et al. (2000) and Pepe and Calò (2017), among others, provide detailed overviews of InSAR fundamentals and InSAR time series methods. Our InSAR data set comprises two image stacks acquired by the Canadian RADARSAT-2 sensor between the fall of 2014 and the fall of 2015 (MDA, 2015). RADARSAT-2 operates at a C-band wavelength (5.6 cm) and has a 24-day orbit repeat cycle. The high-resolution spotlight image stacks used in our

Characteristic

SLA15

SLA10

Pass direction Incidence angle (°) Satellite heading (°) Number of scenes Stack start date Stack end date

Asc 37.8 350.4 16 2014-10-25 2015-10-20

Asc 41.4 349.6 17 2014-10-18 2015-11-06

Figure 5. Footprints of the RADARSAT-2 SLA15 and SLA10 stacks overlain on a Google Earth orthoimage of the area (Digital Globe, July 2010). The location of the Alexandria landslide is indicated by the orange rectangle.

292

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

An Integrated InSAR-Borehole Inclinometer-Numerical Modeling Approach to the Assessment

Figure 6. (A) Section (A–A ) selected for 2D-limit equilibrium analysis. The section intersects borehole inclinometer SI0103. (B) Subsurface displacement data from borehole inclinometer SI0103, showing the two main shear surfaces at 82 and 110 m below the ground surface. (C) Limit equilibrium model.

Table 2. Input properties for Slide2 and FLAC analyses.

Endako Group Australian Creek and Fraser Bend formations

Unit Weight (kN/m3 )

Bulk Modulus (GPa)

Shear Modulus (GPa)

Cohesion (MPa)

Friction angle (°)

Uniaxial Tensile Strength (MPa)

27 20

26.8 0.8

6.99 0.42

27.2 0.02

28 25

1.17 0.01

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

293

Francioni, Stead, Sharma, Clague, and Brideau

Figure 7. Global Positioning System (GPS) displacements recorded for the period 2009–2014. Upper right: Rosette diagram showing azimuths of all GPS slope displacement vectors. The units of the displacement vectors are meters.

agreement with the GPS data. Results must be interpreted with caution in areas where the digital surface model (DSM) might be out of date and in areas that might experience uplift or subsidence that does not conform to the downslope assumption (Sharma et al., 2016). After downslope projection, we post-processed the data to eliminate artifacts due to radar geometry and the DSM. InSAR has limited sensitivity where the radar LOS is orthogonal to the slope, and even small contributions due to noise can be magniﬁed to large displacements on re-projection. To eliminate these areas, we masked out pixels with low sensitivity (within 10° of orthogonal to the slope), which represent about 8% of the total pixels. Additional pixels were masked out due to artifacts in the DSM, including sharp gradients in slope aspect over single pixels and erroneous terrain near the riverbank (roughly 2% of the total pixels). GIS and Engineering Geomorphology We integrated data from borehole inclinometers, GPS instruments, and satellite-based InSAR using

294

ESRI ArcGIS (ESRI, 2018). The GIS was also used to manage the DSM (cell size 20 × 20 m) and generate thematic maps (hillshade, slope, and aspect) of the area. The maps helped us understand the landslide geometry and produce an engineering geomorphological map highlighting linear terrain features, including gullies, backscarps, and back-tilted slopes (i.e., rotated and tilted blocks of failed material with aspects opposite the slope dip direction). The engineering geomorphology interpretation helped us understand GPS and InSAR slope movements and was crucial for subsequent limit equilibrium and numerical modeling of the slope. Slope Stability Analyses We used the information obtained from the GIS analysis to deﬁne the slope proﬁle employed in our numerical slope stability analyses. We initially developed a limit equilibrium model of the landslide using the software Slide2 (Rocscience, 2018). Slide2 allowed us to back-analyze the stability of slip surfaces using limit equilibrium methods. Individual slip surfaces can be analyzed with this software, or search methods can

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

An Integrated InSAR-Borehole Inclinometer-Numerical Modeling Approach to the Assessment

Figure 8. InSAR-measured deformation. Downslope movement is positive, upslope movement is negative, and light-blue areas have near-zero displacements.

be applied to locate the critical slip surface for a given slope. In this research, we investigated both circular and non-circular methods to determine which failure surface is most appropriate. Our simulations showed that the non-circular surface method better ﬁts the geometry of the landslide; thus, we present only the results using this method. The limit equilibrium modeling method used in this study allowed us to properly locate known backscarps, determine the inclination of the sliding surface, and verify the critical stability condition of the slope. The section used for this analysis is orthogonal to the main scarp and passes through borehole inclinometer SI0103 (Figure 6A and C), where shear surfaces occur 82 and 110 m below the ground surface (Figure 6B). Based on the location of inclinometer SI0103 and the depth of the lower sliding surface (110 m below the ground surface), we carried out a sensitivity analysis of the possible inclinations of the sliding surface. Limit equilibrium models using a block search algorithm yielded an inclination of the sliding surface that generated backscarps in agreement with those identiﬁed based on our engineering geomorphological interpretation.

Continuum FLAC models (Itasca, 2018) were then constructed based on the results of the limit equilibrium analysis and constrained by borehole inclinometer, GPS, and InSAR data. Input values used for both limit equilibrium and numerical simulation were taken from a technical report by AMEC (2002) and the Itasca properties database (Itasca, 2018) (Table 2). In contrast to Slide2, FLAC uses an explicit finite difference formulation to model the development of non-linear large slope displacements and strains. The FLAC model results allowed us to interpret and simulate the behavior of the landslide using the shear surfaces delineated with borehole data and the limit equilibrium analysis. RESULTS GPS and InSAR Displacements GPS data for the period 2009–2014 show spatially differing amounts of displacement, suggesting that the Alexandria landslide comprises several independent or semi-independent slide blocks (Figure 7). The highest measured displacements, up to 0.438 m (∼0.07 m/yr),

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

295

Francioni, Stead, Sharma, Clague, and Brideau

Figure 9. (A) Hillshade map. (B) Slope map. (C) Aspect map. The hillshade map highlights linear features, such as gullies, whereas slope and aspect maps accentuate landslide backscarps and steep faces of back-tilted blocks.

are in the central part of the landslide. Displacements are lower nearer the edges of the landslide, and parts of the landslide are stable or moving no faster than <0.005 m/yr, which is the lowest rate that can be measured with the instruments. Slope displacement vectors indicate that the slide blocks are moving mainly toward the southwest (Figure 7). InSAR analysis provided cumulative displacements from October 2014 to October 2015 (Figure 8). The highest displacements (0.05–0.1 m) are in the central part of the landslide and at the toe of the slope. Displacements decrease toward the edges of the landslide, in agreement with the GPS data. Thematic GIS and Engineering Geomorphological Maps Hillshade, slope, and aspect maps (Figure 9) highlight morphological features of the Alexandria landslide. Notably, the hillshade map emphasizes linear features, especially gullies (Figure 9A), and the slope

296

and aspect maps highlight backscarps and back-tilted slopes of large displaced blocks (Figure 9B and C). Backscarps and back-tilted slopes can also be inferred from the slope proﬁle A–A (Figure 10). Using the thematic maps, we produced an engineering geomorphological map of the landslide (Figure 11), showing backscarps, back-tilted block faces, and gullies. The map includes arrows that show local slope directions and their inclinations. Rosette diagrams, which accompany the map, show azimuths of these features. By integrating information on the engineering geomorphological map with the GPS and InSAR monitoring data, we conclude that the following:

r The main landslide body has a length of 3.7 km and a width of 2 km.

r Gullies trend west-southwest, whereas backscarps and back-tilted blocks faces are oriented in a northwest direction (Figure 9).

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

An Integrated InSAR-Borehole Inclinometer-Numerical Modeling Approach to the Assessment

Figure 10. Top: Aspect maps highlighting backscarps and the steep faces of back-tilted blocks. Bottom: Topographic proﬁle showing locations of these features.

r The landslide comprises eight constituent slide blocks (Figure 12A), in agreement with GPS and InSAR data (Figure 12B). r The slide blocks in the central part of the landslide and near the Fraser River (blocks 2 and 3) are moving faster than the other blocks (Figure 12B). Limit Equilibrium Analyses and Numerical Model Simulations The deepest sliding surface, based on measurements from borehole inclinometer SI0103, is 110 m below the ground surface (Figure 6B). We performed a blocksearch sensitivity analysis to determine the inclination of this sliding surface where it intersects the highest backscarp on the slope. We infer the location and dip of the sliding surfaces (4° to the southwest) based on results of the sensitivity analysis, site borehole inclinometer data, and a review of investigation carried out at a similar landslide in West Quesnel (AMEC, 2002), 50 km to the north. Finally, we carried out a sensitivity back analysis to constrain the properties of the sliding surfaces. We

modified the properties of the shear surfaces until a global minimum factor of safety of 1 was obtained using the Morgenstern and Price method (Morgenstern and Price, 1967). Table 3 shows the properties of the sliding surfaces used in the Slide2 and FLAC analyses. At a minimum factor of safety of 1, the limit equilibrium analysis indicates three backscarps along the length of the slope profile (Figure 13), in agreement with the GIS-engineering geomorphological interpretation (Figure 11). Following the limit equilibrium analysis, we incorporated the slope geometry of the same section (A– A ; Figure 6) in the finite difference code FLAC. The backscarps and shear layer identified in the GIS and limit equilibrium analyses were included as discrete structures in the FLAC model (Figure 14A). Displacements obtained using the FLAC model are shown in Figure 14B. The slope deforms as a translational slide, similar to the West Quesnel landslide and in agreement with previous preliminary studies of the Alexandria landslide (AMEC, 2002). Figure 15A provides a comparison of the InSAR monitoring data and the FLAC simulation. There is

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

297

Francioni, Stead, Sharma, Clague, and Brideau

Figure 11. Engineering geomorphological map. Arrows indicate slope dip directions and inclinations. Rosette diagrams show measured azimuths of backscarps, back-tilted block faces, and gullies.

good agreement in the areas with the highest slope displacements, with most of the simulated horizontal displacement occurring at the toe of the slope. We included a second shear surface in the FLAC model to simulate displacements along the other major sliding surface recorded by the borehole inclinometer at 82 m below the ground surface (Figure 6). In this second FLAC model simulation, the InSAR and

modeled displacements again are in good agreement, showing similar areas of movement, with the highest displacements at the slope toe (Figure 15B). We added a vertical proﬁle of equally spaced history points to the FLAC model at the site of borehole slope inclinometer SI0103 to create a virtual inclinometer (Figure 16). We then compared recorded borehole inclinometer data with the virtual borehole inclinometer

Table 3. Assigned properties of the sliding surface layers used in the FLAC analyses.

Shear/sliding surface layers

298

Unit Weight (kN/m3 )

Bulk Modulus (GPa)

Shear Modulus (GPa)

Cohesion (MPa)

Friction Angle (°)

0.04

0.02

0.005

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

An Integrated InSAR-Borehole Inclinometer-Numerical Modeling Approach to the Assessment

Figure 12. (A) Landslide geometry inferred from engineering geomorphological map and Global Positioning System data. (B) InSAR monitoring data.

Figure 13. Result of limit equilibrium analysis, showing three main backscarps, in agreement with the geographic information system engineering geomorphology interpretation (Figure 11).

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

299

Francioni, Stead, Sharma, Clague, and Brideau

Figure 14. (A) FLAC 2D model and (B) simulation results (horizontal displacement in the x-direction).

results from the FLAC model simulation. Figure 16A and B show good agreement between the SI0103 slope inclinometer data and the results of the virtual inclinometer created in the FLAC model for the deeper shear surface (110 m below the ground surface, red dashed line 1). Figure 16C shows the results achieved with a second virtual inclinometer, with the additional sliding surface (line 2) included. We note that borehole inclinometer SI0103 indicates additional minor sliding surfaces that have not been considered in this study due to the dominance of the shear surfaces at 82- and 110-m depth.

DISCUSSION In this study, we have demonstrated that the combined use of a GIS with GPS, space-borne InSAR, and borehole subsurface inclinometer data provides a powerful constraint on the numerical simulation of a landslide. With satellite InSAR data sets, it is possible to monitor surface slope deformation with millimeter-to-centimeter accuracy, depending on error sources and environmental conditions. Although InSAR techniques are now widely used to monitor slowmoving landslides, their integration in a GIS with GPS and borehole inclinometer data to constrain numerical

300

models remains an important opportunity for future landslide research. All geomechanical landslide models involve model and parameter uncertainty. The five borehole inclinometers installed on the Alexandria landslide clearly indicate shear surface depths, and abundant surface displacement data based on GPS monitoring provide precise surface displacement data. However, the instruments were installed along the pipeline ROW, and consequently there is insufficient coverage to resolve potential spatial differences in shear surfaces in three dimensions. We therefore could perform reliable numerical simulations using only 2D methods. Uncertainties that relate to the position of the 3D failure surfaces, groundwater, and geology make it difficult to constrain and calibrate 3D models of the Alexandria landslide; hence, only one example of a FLAC 3D geomechanical model is shown in Figure 17. Notwithstanding the data limitations, the preliminary FLAC 3D model results demonstrate the significant potential of InSAR to constrain 3D models. Future work could include iteration of the 3D failure surfaces to achieve closer agreement with the observed slope damage and displacements, but the multi-block progressive nature of the landslide (both retrogressively and laterally), together with data uncertainty, makes 3D modeling of the landslide extremely challenging.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

An Integrated InSAR-Borehole Inclinometer-Numerical Modeling Approach to the Assessment

Figure 15. (A) InSAR monitoring results. The area selected for slope stability analyses is indicated by the black rectangle. (B) Comparison of InSAR monitoring data and FLAC simulation results (x-displacements) for a single shear layer at 110 m below the ground surface. (C) Comparison of InSAR monitoring data and FLAC simulation results (x-displacements) for two shear layers at 110 and 82 m below the ground surface.

InSAR stacks providing line-of-sight deformation measurements from a single pass direction were used in this study. If both ascending and descending pass directions are available, it is possible to separate eastwest and up-down displacement components, providing further geomechanical model constraints. Remotely sensed monitoring data, however, must be complemented with ﬁeld observations and GIS-based

analyses to determine how geological features control landslide behavior. In the case of the Alexandria landslide, the engineering geomorphological map and monitoring data allowed us to distinguish discrete blocks within the body of the landslide, which would not have been possible with the InSAR data alone. The 2D numerical simulations conﬁrmed that the landslide comprises multiple retrogressive slide blocks and

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

301

Francioni, Stead, Sharma, Clague, and Brideau

Figure 16. Comparison of SI0103 borehole inclinometer data (A) and results for FLAC models (B and C) incorporating a virtual slope inclinometer with, respectively, one shear layer and two shear layers. Depths are meters below the ground surface.

deforms as a multi-block translational slide. This mode of deformation agrees with the failure model previously developed for the both the West Quesnel and the Alexandria landslides (AMEC, 2002). The fastestmoving blocks are located in the central part of the Alexandria landslide and near the Fraser River. The landslide may have initiated with displacements of blocks 2 and 3 (Figure 12), which continue to experience the highest movement rates, and subsequently

involved other blocks retrogressively. River erosion at the toe of the landslide blocks and inﬁltration of water into the slope during rainy weather or snowmelt are probably the main factors contributing to continuing slope deformation. We note that geotechnical laboratory tests and accurate hydrogeological information were not available for this study. However, test data for the similar West Quesnel landslide, hydrogeological data derived from the boreholes, and sensitivity 2D

Figure 17. (A) InSAR downslope displacements overlain on the terrain model. (B and C) FLAC 3D representations of ground displacement after, respectively, 1,509 and 1,559 cycles.

302

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

An Integrated InSAR-Borehole Inclinometer-Numerical Modeling Approach to the Assessment

CONCLUSIONS

Figure 18. Flowchart describing the iterative/integrated InSARgeomechanical modeling approach used in this study.

back analyses allowed us to overcome these limitations and accurately simulate geotechnical and hydrogeological conditions required to induce landslide instability. InSAR-based geomechanical working models should be regularly updated to explain spatial and temporal changes in slope displacement. The kinematics of a slow-moving landslide may continuously change due to displacements and interactions of constituent landslide blocks. InSAR is ideally suited to document these spatial and temporal changes. The geomechanical model can be interrogated and possibly modified to provide an explanation for such changes. Long-term InSAR monitoring would also provide constraints for 3D landslide modeling, decreasing the uncertainties associated with insufficient coverage of borehole slope inclinometers and GPS stations. We suggest that an iterative and integrated InSAR/geotechnical monitoring and geomechanical modeling approach, as outlined in Figure 18 and implemented in this article, has the potential to significantly improve understanding of landslide geohazards.

The Alexandria landslide is a large, slow-moving, translational multi-block, retrogressive slope failure. GPS and InSAR data indicate that displacements differ spatially—parts of the landslide are presently inactive, and displacements within active areas are non-uniform. We performed a GIS analysis that allowed us to identify important geological features and produce an engineering geomorphological map of the landslide. Hillshade, slope, and aspect maps derived from the digital elevation model highlight backscarps and back-tilted blocks that characterize the body of the landslide. Scarps and faces of back-tilted blocks strike northwest, indicating that slope displacements are mainly in a southwest direction and in agreement with surface displacement recorded by longterm GPS measurements and InSAR displacement vectors. Gullies have southwest and west-southwest orientations and appear to mark constituent landslide block boundaries. Erosion associated with these gullies may facilitate lateral release of these blocks and partly explain differences in slope displacements. A 2D limit equilibrium analysis identified three backscarps along a profile across the landslide that coincide spatially with backscarps identified in the field and that are revealed through analysis of the GIS data. The failure geometry inferred through the GIS and limit equilibrium analyses was used in continuum numerical slope simulations. These simulations demonstrate that the Alexandria landslide behaves as a translational slide consisting of multiple sliding blocks with different displacement magnitudes. The area with the highest displacements today is at the toe of the slope, in agreement with InSAR data. This research has demonstrated the importance of using multiple techniques, including surface and subsurface monitoring data and GIS interpretations, when modeling complex slow-moving landslides. InSAR monitoring is a valuable method for constraining numerical simulations, especially if integrated with GPS and subsurface borehole inclinometer data. GIS analyses play a key role in studies of such landslides, as they highlight geological/geomorphic features that are critical in both developing and constraining numerical landslide simulations. ACKNOWLEDGMENTS This work was partially funded by the Canadian Space Agency (CSA) under Earth Observation Application Development Program (EOADP) Contract #9F043-130644/001/MTB. CSA also provided the

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

303

Francioni, Stead, Sharma, Clague, and Brideau

RADARSAT-2 images. The authors thank Westcoast Energy Inc. and Doug Dewar for their contributions to analyzing monitoring data and ﬁeldwork.

REFERENCES AMEC. 2002. West Quesnel Land Stability Study: Quesnel, BC. Technical report submitted to the City of Quesnel, British Columbia, Canada. Bianchini, S.; Raspini, F.; Solari, L.; Del Soldato, M.; Ciampalini, A.; Rosi, A.; and Casagli, N. 2018. From picture to movie: Twenty years of ground deformation recording over Tuscany region (Italy) with satellite InSAR: Frontiers Earth Science, Vol. 6. doi:10.3389/feart.2018.00177. Bozzano, F.; Mazzanti, P.; Perissin, D.; Rocca, A.; De Pari, P.; and Discenza, M. 2017. Basin scale assessment of landslide geomorphological setting by advanced InSAR analysis: Remote Sensing, Vol. 9, No. 3. https://doi.org/10.3390/rs9030267. Brideau, M. A.; Pedrazzini, A.; Stead, D.; Froese, C.; Jaboyedoff, M.; and van Zeyl, D. 2011. Three-dimensional slope stability analysis of South Peak, Crowsnest Pass, Alberta, Canada: Landslides, Vol. 8, pp. 139–158. https://doi.org/10.1007/s10346-010-0242-8. Clague, J. J. 1981. Late Quaternary Geology and Geochronology of British Columbia, Part 2: Summary and Discussion of Radiocarbon-Dated Quaternary History: Geological Survey of Canada Paper 80-35, 41 p. https://doi.org/10.4095/119439. Cooke, R. U. and Doornkamp, J. C. 1990. Geomorphology in Environmental Management: A New Introduction: Clarendon Press, Oxford, U.K. 434 p. https://doi.org/10.1017/S0016756800036967. Cui, Y.; Miller, D.; Schiarizza, P.; and Diakow, L. J. 2017. British Columbia Digital Geology: British Columbia Ministry of Energy, Mines and Petroleum Resources, British Columbia Geological Survey Open File 2017–8, data version 2018-04-05. 9 p. Dewar, D.; McClarty, E.; and Tong, E. 2017. Assessing and monitoring the impacts of very slow moving deep-seated landslides on pipelines. In Proceedings, GeoOttawa 2017, 70 Years of Canadian Geotechnics and Geoscience, Paper 160, 8 p. Dewar, D.; Tong, A.; McClarty, E.; and Van Boven, G. 2016. Technical and operational guidelines when using strain gauges to monitor pipelines in slow moving landslides. In Proceedings, 2016 11th International Pipeline Conference, September 26–30, 2016, Calgary, Alberta, CA, Paper No: IPC2016-64594, V003T04A023; 12 p. https://doi.org/10.1115/IPC2016-64594. Eppler, J.; Kubanski, M.; Sharma, J.; and Busler, J. 2015. High temporal resolution permafrost monitoring using a multiple stack InSAR technique. In Proceedings, International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 36th International Symposium on Remote Sensing of Environment, Berlin, Germany, pp. 1171–1176. doi:10.5194/isprsarchives-XL-7-W3-1171-2015. ESRI. 2018. ArcGIS v. 10. https://www.esri.com/en-us/store/ arcgis-desktop. Francioni, M.; Calamita, F.; Coggan, J.; De Nardis, A.; Eyre, M.; Miccadei, E.; Piacentini, T.; Stead, D.; and Sciarra, N. 2019. A multi-disciplinary approach to the study of large rock avalanches combining remote sensing, GIS and ﬁeld surveys: The case of the Scanno landslide, Italy: Remote Sensing Vol. 11, 1570. https://doi.org/10.3390/rs11131570.

304

Francioni, M.; Salvini, R.; Stead D.; and Coggan, J. J. 2018a. Improvements in the integration of remote sensing and rock slope modelling: Natural Hazards, Vol. 90, pp. 975–1004. https://doi.org/10.1007/s11069-017-3116-8. Francioni, M.; Stead, D.; Clague, J.; and Westin, A. 2018b. Identification and analysis of large paleo-landslides at Mount Burnaby, British Columbia: Environmental Engineering Geoscience, Vol. 24, pp. 221–235. https://doi.org/10.2113/EEG1955. Guthrie, R.; Reid, E.; Richmond, J.; Ghuman, P.; and Cormier, Y. 2018. InSAR and the pipeline geohazards toolbox: Instructions for use as of 2018. In Proceedings of the 2018 12th International Pipeline Conference, September 24–28, 2018, Calgary, Alberta, CA. Volume 3, Paper IPC2018-78571, 8 p. https://doi.org/10.1115/IPC2018-78571. Hora, Z. D. and Hancock, D. D. 1994. Quesnel area—Industrial minerals assessment: British Columbia Geological Survey, Geological Fieldwork 1994, pp. 395–404. Itasca. 2018. FLAC v. 7. https://www.itascacg.com/software/ flac. Jaboyedoff, M.; Baillifard, F.; Couture, R.; Locat, J.; and Locat, P. 2004. Toward preliminary hazard assessment using DEM topographic analysis and simple mechanic modelling. In Lacerda, W. A.; Ehrlich, M.; Fontoura, A. B.; and Sayo, A. (Editors), Proceedings, 9th International Symposium on Landslides: Balkema, Rotterdam, Netherlands, pp. 191– 197. doi:10.1201/b16816-27. Lin, P.; Shi, J.; Zhou, W.; and Wang, R. 2018. 3D geomechanical model tests on asymmetric reinforcement and overall stability relating to the Jinping I super-high arch dam: International Journal of Rock Mechanics and Mining Sciences, Vol. 102, pp. 28–41. doi:10.1016/j.ijrmms.2017.11.017. Lorig, L. and Stead, D. 2018. Numerical analysis. In Wyllie, D. C. (Editor), Rock Slope Engineering—Civil Applications, 5th ed.: CRC Press, Boca Raton, FL, pp. 319–348. https://doi.org/10.4324/9781315154039. Mantovani, M.; Devoto, S.; Piacentini, D.; Prampolini, M.; Soldati, M.; and Pasuto, A. 2016. Advanced SAR interferometric analysis to support geomorphological interpretation of slow-moving coastal landslides (Malta, Mediterranean Sea): Remote Sensing, Vol. 8, No. 6. https://doi.org/10.3390/rs8060443. MDA. 2015. Issue 1/12, RADARSAT-2 product description, RNSP-52-1238, September 28, 2015. Morgenstern, N. R. and Price, V. E. 1967. A numerical method for solving the equations of stability of general slip surfaces: Computer Journal, Vol. 9, pp. 388–393. https://doi.org/10.1093/comjnl/9.4.388. Parizzi, A. and Brcic, R. 2011. Adaptive InSAR stack multilooking exploiting amplitude statistics: A comparison between different techniques and practical results: IEEE Geoscience and Remote Sensing Letters, Vol. 8, pp. 441–445. doi:10.1109/LGRS.2010.2083631. Pepe, A. and Calò, F. 2017. A review of interferometric synthetic aperture RADAR (InSAR) multi-track approaches for the retrieval of Earth’s surface displacements: Applied Sciences, Vol. 7, No. 12. https://doi.org/10.3390/app7121264. Qi, S. W.; Wu, F. Q.; Yan, F. Z.; and Lan, H. X. 2004. Mechanism of deep cracks in the left bank slope of Jinping first stage hydropower station: Engineering Geology, Vol. 73, pp. 129–144. https://doi.org/10.1016/j.enggeo.2003.12.005. Ramsden, R.; Audigé, E.; and Fagan, C. 2015. A comprehensive and integrated slope monitoring approach for enhanced effectiveness: Application at Debswana’s Jwaneng Mine in Botswana. In Slope Stability 2015: South African

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

An Integrated InSAR-Borehole Inclinometer-Numerical Modeling Approach to the Assessment Institute of Mining and Metallurgy, Cape Town, South Africa, pp. 629–640. Raspini, F.; Bianchini, S.; Ciampalini, A.; Del Soldato, M.; Montalti, R.; Solari, L.; Tofani, V.; and Casagli, N. 2019. Persistent scatterer continuous streaming for landslide monitoring and mapping: The case of the Tuscany region (Italy): Landslides, Vol. 16, pp. 2033–2044. https://doi.org/10.1007/s10346-019-01249-w. Rocscience. 2018. Slide2. https://www.rocscience.com/software/ slide2. Rosen, P. A.; Hensley, S.; Joughin, I. R.; Li, F. K.; Madsen, S. N.; Rodriguez, E.; and Goldstein, R. M. 2000. Synthetic aperture radar interferometry: Proceedings IEEE, Vol. 88, pp. 333–382. Rouse, G. E. and Mathews, W. H. 1979. Tertiary geology and palynology of the Quesnel area, British Columbia: Bulletin Canadian Petroleum Geology, Vol. 27, pp. 418–445. Scoppettuolo, M. R.; Cascini, L.; and Babilio, E. 2020. Typical displacement behaviours of slope movements: Landslides, Vol. 17, pp. 1105–1116. https://doi.org/10.1007/s10346-01901327-z. Sharma, J.; Francioni, M.; Busler, J.; Stead, D.; Donati, D.; Onsel, E.; Clague, J. J.; and Brideau, M.-A. 2016. Monitoring landslides in pipeline corridors using a combined satellite based InSAR and geomechanical modelling approach. In Proceedings, GeoVancouver 2016, 69th Canadian Geotechnical Conference, Vancouver, British Columbia, Paper 4200, 10 p.

Sharma, J.; Francioni, M.; Stead, D.; Clague, J.; Brideau M.-A.; Kubanski, M.; Eppler, J.; and Busler, J. 2015. Monitoring geohazards near pipeline corridors with an advanced In-SAR technique and geomechanical modelling. In Proceedings, GEOQuebec 2015, Canadian Geotechnical Society Conference, Quebec, CA, Paper 373, 8 p. Stead, D.; Eberhardt, E.; and Coggan, J. S. 2006. Developments in the characterization of complex rock slope deformation and failure using numerical modelling techniques: Engineering Geology, Vol. 83, pp. 217–235. https://doi.org/10.1016/j.enggeo.2005.06.033. Tunono, A. B.; Ruest, M.; Dimbungu, L.; and Brook, M. 2011. Geotechnical design of the Jwaneng Mine Cut 8. In Proceedings, 2011 International Symposium on Slope Stability in Mining and Civil Engineering (Slope Stability 2011), E. Eberhardt and D. Stead (eds), 18–21 September 2011, Vancouver, Canada, Canadian Rock Mechanics Association, CD-rom only., 12pp. West Quesnel Land Stability Program. 2020. Annual Monitoring Report: Electronic document, available at https://www. quesnel.ca/sites/default/ﬁles/docs/building-development/ monitoringreport2018-2019.pdf. Wolter, A.; Stead, D.; Ward, B. C.; Clague, J. J.; and Ghirotti, M. 2016. Engineering geomorphological characterisation of the Vajont Slide, Italy, and a new interpretation of the chronology and evolution of the landslide: Landslides, Vol. 13, pp. 1067–1081. https://doi.org/10.1007/s10346-015-0668-0.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 287–305

305

Factors Contributing to Landslide Susceptibility of the Kope Formation, Cincinnati, Ohio MICHAEL P. GLASSMEYER Kleinfelder, 180 Sheree Boulevard, Suite 3800, Exton, PA 19341

ABDUL SHAKOOR* Department of Geology, Kent State University, Kent, OH 44242

Key Terms: Kope Formation, Colluvial Soil, Landslide Susceptibility, Shear Strength, Pore Pressure, Slake Durability, Toe Undercutting ABSTRACT The objective of this study was to evaluate the factors that contribute to the high frequency of landslides in the Kope Formation and the overlying colluvial soil present in the Cincinnati area, southwestern Ohio. The Kope Formation consists of approximately 80 percent shale inter-bedded with 20 percent limestone. The colluvium that forms from the weathering of the shale bedrock consists of a low-plasticity clay. Based on field observations, LiDAR data, and information gathered from city and county agencies, we created a landslide inventory map for the Cincinnati area, identifying 842 landslides. From the inventory map, we selected 10 landslides that included seven rotational and three translational slides for detailed investigations. Representative samples were collected from the landslide sites for determining natural water content, Atterberg limits, grain size distribution, shear strength parameters, and slake durability index. For the translational landslides, strength parameters were determined along the contact between the bedrock and the overlying colluvium. The results of the study indicate that multiple factors contribute to landslide susceptibility of the Kope Formation and the overlying colluvium, including low shear strength of the colluvial soil, development of porewater pressure within the slope, human activity such as loading the top or cutting the toe of a slope, low to very low durability of the bedrock that allows rapid disintegration of the bedrock and accumulation of colluvial soil, undercutting of the slope toe by stream water, and steepness of the slopes.

*Corresponding author email: ashakoor@kent.edu

INTRODUCTION Landslide Problem in the Cincinnati Area The Cincinnati area (Hamilton and Clermont counties) comprises the southwestern corner of Ohio and is one of the most landslide-susceptible areas in the United States (Ohio Emergency Management Agency [EMA], 2011). Most of the landslides occur in the Kope Formation and the overlying colluvial soil during late winter and early spring (Fleming, 1975). Landslide damage and mitigation cost the city millions of dollars each year (Rockaway, 2002). According to Schuster (1996), the annual per capita cost for landslide damage in the Cincinnati area was $5.80 in 1981 (equivalent to $17.27 in 2020). This does not include more than $22 million spent in 1981 (equivalent to $65.5 million in 2020) to stabilize a single landslide that occurred on Mount Adams during the construction of Interstate 471. One of the costliest time periods for landslide damage in the Cincinnati area occurred between 1973 and 1978 when, during a 6-year period, an average of $5.1 million in 1981 dollars (equivalent to $15.2 million in 2020) was spent per year to repair landslide damage (Schuster, 1996). Rotational and translational slides are the most frequently occurring slope movements associated with the Kope Formation and the overlying colluvial soil. Rapid earthﬂows, rockfalls, and complex slides (combination of rotational and translational slides), although present, are infrequent. Rotational slides are common where thick colluvium covers the bedrock. They are generally 2–15 m thick, 30–300 m wide (measured perpendicular to the direction of sliding), and 30–150 m long (measured along the direction of sliding). Many rotational slides are associated with springs or marshy areas either beneath or within the slope toes (Fleming and Johnson, 1994). Translational slides are common where thin colluvial soils (2–3 m thick) cover relatively steep slopes (15°–30°). They occur along the colluvium–bedrock contact, are generally 10–150 m wide and 30–130 m long, and vary in shape from

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 307–318

307

Glassmeyer and Shakoor

long and narrow to wide and short (Richards, 1982). Translational slides generally occur during spring because the slide material is almost saturated between the months of January and May (Haneberg, 1991, 1992; Haneberg and Gökce, 1994). The dominant form of deformation in translational slides is longitudinal stretching resulting in a series of scarps. Complex landslides in the Cincinnati area consist of more than one layer of slide material. They are thinner near the slope crest and become thicker near the toe. Rapid earth flows in the Kope Formation (locally known as mudslides) occur on steeper slopes along the Columbia Parkway. They occur during wet periods in areas where the colluvium is <2 m thick and is clayey in nature (Pohana, 1983). Rapid earthflows involve movement of the entire thickness of the colluvium, exposing the bedrock (Richards, 1982; Riestenberg and SovonikDunford, 1983) Geology of the Cincinnati Area The Cincinnati area forms the western flank of the Cincinnati Arch where the bedrock dips gently at less than 1° (Fleming, 1975). The area represents an upland surface, enveloped by the Pre-Illinoian, Illinoian, and Wisconsinan age glacial deposits, that has been dissected by ancient drainage systems as well as the modern Ohio River and its tributaries (Pavey et al., 1992; Potter, 2007). Many of the tributaries have carved broad, terraced valleys with steep slopes. The relief between the Ohio River and the hilltops in the area is approximately 120 m (Baum and Johnson, 1996). Alluvium and glacial outwash cover the valley floors, and colluvium covers most of the hillsides (Baum and Johnson, 1996). The Kope Formation in the Cincinnati area is overlain by the Fairview Formation, both being Upper Ordovician in age. The contact between the two formations is at an elevation between 200 and 215 m (Gibbons, 1973). Figure 1 shows the extent of the Kope Formation in the Cincinnati area, as indicated by the surficial geology map. The formation is >60 m thick and consists of inter-bedded, medium to dark grey shale (80 percent) and coarse-grained fossiliferous limestone (20 percent) (Fleming and Johnson, 1994). It should be noted that what is referred to in earlier studies as “shale” is mudstone/claystone according to the Potter et al. (1980) classification (Sarman, 1991; Dick, 1992; and Hajdarwish, 2006). The shale (mudstone/claystone) consists of illite, chlorite, calcite, and quartz (Sarman, 1991; Dick, 1992; and Hajdarwish, 2006). The limestone layers within the Kope Formation contain three sets of near-vertical joints, occurring at regular spacing. The orientations of the joints, however, vary between different loca-

308

tions (Hofman, 1966; Brett and Algeo, 2001; and Brett et al., 2003). The shale also contains steeply dipping joints (Richards, 1982; Baum, 1983). The colluvium associated with the Kope Formation classifies as Eden silty clay loam according to the Hamilton County Soil Survey Report and as clay of low plasticity according to the Unified Soil Classification System (USCS) (Lerch et al., 1982; Glassmeyer, 2014). The colluvium covers most of the hillsides and generally ranges in thickness from a few centimeters up to 15 m (Fleming and Johnson, 1994), but can be much thicker at some places. Study Objectives Although landslides in the Cincinnati area have been studied extensively, a specific and detailed study regarding the susceptibility of the Kope Formation to landslide occurrence has not been conducted. Thus, the main objective of this study was to investigate the factors that contribute to high landslide susceptibility of the Kope Formation and the colluvium derived from it (Note: in this study, the colluvium is synonymous to the Kope Formation). This objective was accomplished by performing the following tasks: 1. Create a landslide inventory map for the Kope Formation and the associated colluvium. 2. Determine the engineering properties of the Kope Formation and the overlying colluvium. 3. Identify the types of slope movement that affect the Kope Formation. 4. Explain the landslide susceptibility of the Kope Formation and the overlying colluvium in terms of engineering properties, slope characteristics, and hydrologic conditions. RESEARCH METHODS Landslide Inventory We developed a landslide inventory map for the Kope Formation and the overlying colluvial soil using LiDAR data, field observations, and landslidelocations data from city and county governments (Figure 2). A total of 842 landslides were identified in the colluvial soil derived from the Kope Formation. Of these, 542 landslides were identified using the LiDARderived maps and 300 were identified through field observations and data obtained from city and county governments. The LiDAR data, with an accuracy of 0.33 m, were divided into tiles that were 1,524 m by 1,524 m square. Since the LiDAR data are LAS (a blob point file or a collection of binary data stored as a

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 307–318

Landslide Susceptibility of the Kope Formation

Figure 1. Map showing the extent of the Kope Formation (darker brown) in the Cincinnati area. The blue star indicates the location of downtown Cincinnati. The shaded area in the southwest corner of the Ohio map shows Hamilton (left) and Clermont (right) counties.

single entity), the data were converted into usable maps using ArcGIS. The files were first converted from multipoint files to ASCII files. The ASCII files were then converted to raster files. Once the raster files were created, we developed a slope map, a hillshade map, a digital elevation map, and a topography map for the study area. These maps were used to identify landsliderelated features such as scarps and toe bulges. Randomly selected landslides from the inventory map were verified through field observations using the GPS. Before mapping the landslides, three different layers were used to define the area of interest on LiDAR-derived maps: (1) the extent of the Kope Formation in the Cincinnati area as defined by the Ohio Department of Natural Resources (ODNR) bedrock geology map, (2) the extent of the Kope Formation as defined by the ODNR surficial geology map, and (3) the extent of the colluvium as defined by the ODNR soil survey division.

Site Selection, Data Collection, and Sampling for Detailed Investigations From the landslide inventory map, we selected 10 landslide sites for detailed investigations (Figure 3). These included seven rotational landslide sites (Eight Mile Road landslide, Ten Mile Road landslide, Delhi Pike landslide complex, Elstun Road landslide, Nordyke Road landslide, Old US 52 landslide, and Wagner Road landslide) and three translational landslide sites (Nine Mile Road landslide, Berkshire Road landslide, and Columbia Parkway landslide). The selected sites represented a range of landslide sizes and geographic locations. The data collected at each site included slope geometry (slope height, slope angle, and slope length), thickness of the colluvium, type of slope movement, location of the failure plane with respect to slope face, whether the slide occurred in the colluvium or within the bedrock, and landslide dimensions

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 307–318

309

Glassmeyer and Shakoor

Figure 2. Landslide inventory map for the Kope Formation and the overlying colluvial soil within the Cincinnati area.

Figure 3. Locations of the landslide sites selected for detailed study.

310

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 307–318

Landslide Susceptibility of the Kope Formation

(length and width). Where possible, information about the hydrogeologic conditions was obtained. We used the Cruden and Varnes classification system (Cruden and Varnes, 1996) to identify the type of slope movement at each site. For describing landslide features and for measuring landslide dimensions at different sites, we used the standardized terminology recommended by the International Association of Engineering Geology (IAEG) Commission on Landslides (1990). Undisturbed chunk samples of colluvial soil, weighing approximately 5 kg, were collected from each site for laboratory testing. Additionally, bedrock samples were collected from the three translational landslide sites. The samples were immediately sealed in air-tight bags and stored in 19-liter plastic buckets to preserve natural water content of the soil samples and to prevent slaking of the bedrock samples. Laboratory Investigations Laboratory tests were conducted to determine natural water content, grain size distribution, Atterberg limits, shear strength parameters, and slake durability index. All tests were performed according to the American Society for Testing and Materials (ASTM) specifications (ASTM, 2010). Natural water content, an indicator of the void ratio of the soil, was determined as soon as the soil samples were brought to the laboratory. Both sieve analysis and hydrometer analysis were used to determine the grain soil distribution of the colluvial soil samples. The results of grain size distribution analysis helped classify the soil from each site according to the USCS (Casagrande, 1948; Holtz et al., 2011). The Atterberg limits test was performed only on material passing the #200 sieve (0.074 mm) to determine liquid limit, plastic limit, and plasticity index. The test results were used to classify the finegrained fraction of the soil according to the USCS. Two versions of the direct shear test were conducted to determine shear strength parameters. The purpose of the first version was to simulate failure conditions in case of rotational landslides with the failure plane located entirely within the soil, whereas the second version simulated the failure conditions for the translational slides with the failure occurring along the contact between the bedrock and the overlying colluvial soil. The slake durability test was performed on the bedrock samples that were collected from the Nine Mile Road landslide, Berkshire Road landslide, and Columbia Parkway landslide sites where the bedrock is at shallow depths. The purpose of the slake durability test was to evaluate weathering potential of the bedrock. Two cycles of the test were performed on each sample and the second-cycle slake durability index (Id2 ) was calculated. Based on Id2 values, and

Table 1. Natural water content values for the colluvial soil samples from the landslide sites. Sample Location

Natural Water Content (%)

Eight Mile Road landslide Nine Mile Road landslide Ten Mile Road landslide Berkshire Road landslide Columbia Parkway landslide Delhi Pike landslide Elstun Road landslide Nordyke Road landslide Old US 52 landslide Wagner Road landslide Mean Median

13.1 27.1 13.9 23.8 23.0 25.6 18.9 13.6 21.1 23.5 20.4 22.0

using the International Society for Rock Mechanics (ISRM) classification (ISRM, 2007), the durability of the samples was classified as follows: high (Id2 > 95 percent); medium (Id2 = 85 percent to 95 percent); low (Id2 = 60 percent to 85 percent); and very low (Id2 = 0 percent to 60 percent). Stability Analysis The computer program Slide (Rocscience, 2012) was used to perform stability analysis for the 10 sites. The program identified the critical surface of failure and calculated the corresponding factor of safety (FS) for both dry and saturated conditions. We also used Slide to perform sensitivity analysis, i.e., variation of FS with respect to strength parameters and groundwater conditions. RESULTS Laboratory Test Results The natural water content values for the colluvial soils from the 10 landslide sites range from 13.1 percent to 27.1 percent, with a mean value of 20.4 percent (Table 1). The relatively high water content values suggest the presence of a high percentage of fine-grained clayey material in the colluvial soils at the landslide sites. This implies that even a small amount of precipitation can result in buildup of pore pressure and reduction in shear strength, leading to failure. The high water content values also indicate the potential for flow type movement. The results of grain size distribution analysis indicated that, according to USCS, colluvial soils derived from the Kope Formation classify as clayey sand. It should be noted that although the colluvial soil classifies as clayey sand, the sand size particles consist of broken pieces of shale bedrock and fossils, and overall,

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 307–318

311

Glassmeyer and Shakoor Table 2. Atterberg limits of the fine-grained fraction of the colluvial soil from the landslide sites.

Liquid Plastic Plasticity Liquidity Limit Limit Index Index

Sample Location Eight Mile Road landslide Nine Mile Road landslide Ten Mile Road landslide Berkshire Road landslide Columbia Parkway landslide Delhi Pike landslide Elstun Road landslide Nordyke Road landslide Old US 52 landslide Wagner Road landslide Mean Median

23.6 41.9 23.0 40.0 42.6 44.0 37.8 24.4 37.2 34.1 34.9 37.5

10.9 20.1 12.3 23.3 22.5 19.8 18.5 11.6 18.5 18.4 17.6 18.5

12.7 21.8 10.7 16.8 20.1 24.2 19.3 12.8 18.7 15.7 17.3 17.8

0.2 0.3 0.2 0.03 0.02 0.2 0.02 0.2 0.1 0.3 0.2 0.2

the colluvium behaves as a clay of low plasticity during landslide activity. Table 2 presents the Atterberg limits test results for the fine-grained fraction of the colluvium from the 10 sites. A plot of Atterberg limits on Casagrande plasticity chart is shown in Figure 4. The plot shows that the fine-grained fraction of the colluvial soil classifies as clay of low plasticity. Table 2 also lists the liquidity index (LI) values. The LI compares the natural water content with the Atterberg limits and indicates how a soil will behave when sheared. If LI is >1, the soil will behave as a viscous liquid when sheared, if it ranges from zero to one, the soil will behave as a plastic material on shearing, and if it is <0, the soil will behave as a brittle material. The LI values in Table 2 indicate a plastic behavior of colluvial soil during landsliding. The strength parameters of a soil (cohesion and friction angle) are the most important engineering properties of a soil in terms of the slope stability. For the soil alone (rotational slides scenario), the peak cohesion and friction angle range from 24.5 kPa to 47.7 kPa and 22.8° to 39.4°, respectively, and the residual cohesion

Figure 4. Plot of Atterberg limits of the ﬁne-grained fraction of the colluvial soils from the landslide sites on the Casagrande plasticity chart.

and friction angle from 22.2 kPa to 38.9 kPa and 15.6° to 20.8° (Table 3). For soil-bedrock contact (translational slide scenario), the residual cohesion ranges from 6.8 kPa to 13.0 kPa and the residual friction angle from 8.0° to 14.6° (Table 4). We provide only residual strength parameters for soil–bedrock contact because of the slow, continual movement of the thin soil layer over bedrock. These shear strength parameter values are inadequate to maintain stability with respect to both rotational and translational slides. The second cycle slake durability index (Id2 ) ranges from 7.1 percent (very low durability) for the Columbia Parkway landslide to 39.9 percent (low durability) for the 9 Mile Road landslide (Table 5).

Table 3. Shear strength parameters for failure surface through the colluvial soil.

Sample Locations Eight Mile Road landslide Ten Mile Road landslide Delhi Pike landslide Elstun Road landslide Nordyke Road landslide Old US 52 landslide Wagner Road landslide Mean Median

312

Peak Cohesion (Kpa)

Residual Cohesion (Kpa)

Peak Friction Angle (degrees)

Residual Friction Angle (degrees)

24.5 27.5 33.4 26.4 47.7 35.2 27.7 31.8 27.7

23.3 22.5 24.0 24.5 38.9 32.7 22.2 26.9 24.0

31.0 33.8 23.8 50.4 22.8 39.4 27.5 32.7 31.0

20.8 15.6 17.8 19.8 17.8 20.3 18.3 18.6 18.3

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 307–318

Landslide Susceptibility of the Kope Formation Table 4. Shear strength parameters for failure surface along the soilbedrock contact. Residual Residual Friction Cohesion (Kpa) Angle (degrees)

Sample Locations Nine Mile Road landslide Berkshire Road landslide Columbia Parkway landslide Mean Median

11.8 13.0 6.8 10.5 11.8

14.0 8.0 14.6 12.2 14.0

The low to very low durability of the Kope Formation explains the thick accumulation of colluvial soil at many locations. The durability properties of argillaceous rocks are important in slope stability because of the reduction in strength properties as a result of weathering (Dick and Shakoor, 1995). Stability Analysis Results for Selected Slope Failures Rotational landslides constitute the most common type of slope failure in the colluvial soil derived from the Kope Formation. All seven rotational landslides that were studied in detail occurred in colluvial soil. Rotational landslides occur where the colluvial soil is >3 m thick. Translational landslides are the second most common type of failure in the colluvial soil. Translational landslides tend to occur in complexes, aﬀecting widespread areas. The failure plane for a translational slide is located along the contact between the colluvial soil and the underlying bedrock. The sliding mass for the three translational slides studied (Nine Mile Road landslide, Berkshire Road landslide, and Columbia Parkway landslide) consists of colluvial soil. The thickness of colluvial soil at the locations of translational slides was found to be approximately 1.5 m to 3.0 m. Detailed descriptions of both rotational and translational landslides can be found in Glassmeyer (2014). For the sake of brevity, we present stability analysis for one rotational landslide (Ten Mile Road landslide) and one translational landslide (Columbia Parkway landslide). For stability analyses for all 10 landslides, see Glassmeyer (2014). The software program Slide (Rocscience, 2012) was used to perform the stability analysis using residual strength parameters. For the

Figure 5. The Ten Mile Road landslide with well-developed head scarp. Notice the undercutting of the toe by a stream.

Ten Mile Road landslide (Figure 5), the critical surface with the lowest FS is shown in Figure 6, which matches the failure surface location observed in the ﬁeld (Figure 5). The Slide program resulted in a FS of 0.83 for the dry condition and 0.79 for the saturated condition. The stability analysis indicated that for the FS to be >1, the cohesion of the soil should be >61.2 kPa (instead of 22.5 kPa) if the friction angle were to remain constant at 15.6°, or the friction angle needs to be >33.8° if the cohesion remains the same (22.5 kPa) (Table 3). For the Columbia Parkway landslide (Figure 7), the critical surface, as determined by the Slide program, is located along the contact between the colluvial soil and the underlying bedrock (Figure 8). It initiates at the top of the slope and emerges at the top of the retaining wall at the base of the slope (Figure 7). It should be noted that soil–bedrock contact may not be perfectly planar (Figure 8) but we assumed it to be planar for the purpose of stability

Table 5. Slake durability index test results for the bedrock samples from the translational landslide sites. Slake Durability Index (Id2) (%)

Location Berkshire Road landslide Columbia Parkway landslide Nine Mile Road landslide

28.5 7.1 39.9

Durability Rating Very low Very low Low

Figure 6. Critical surface for the minimum factor of safety for dry and saturated conditions for the Ten Mile Road landslide, as determined by the Slide program.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 307–318

313

Glassmeyer and Shakoor

Figure 7. (a) Head scarp of the Columbia Parkway landslide and (b) toe of the Columbia Parkway landslide emerging on the top of the retaining wall.

analysis. Also, we assumed a uniform, average colluvium thickness. Locally, the landslide may change into earthﬂow/mudﬂow. The minimum FS is 1.04 when the colluvium is dry and 0.68 when saturated. Stability analysis results show that the soil–rock friction angle needs to be >18° instead of 14.8° (Table 4) for the FS to be >1, if the cohesion were to remain constant at

Figure 8. Critical Surface for the minimum factor of safety for the Columbia Parkway landslide, as determined by the slide program.

314

Figure 9. (a) Installation of metal mesh and soil nails and (b) section of a new soldier beam retaining wall (photos courtesy of Dr. John Rockaway).

6.8 kPa, or the cohesion should be > 8.9 kPa if the friction angle remains constant (14.8°) (Table 4). These results clearly suggest that strength parameters of the colluvial soil are lower than those required to maintain stability. The Columbia Parkway landslide is currently being stabilized at an estimated cost of $17 million (City of Cincinnati—Transportation & Engineering, 2020). The stabilization project extends from Bains Street (Mt. Adams area) on the west side to beyond Torrence Parkway (East Walnut Hills area) on the east side, a nearly 3.2 km long stretch of the Parkway. Within this stretch, 12 landslide locations have been chosen for stabilization with the stabilization method, involving either metal mesh and soil nails or retaining walls (Figure 9), varying from location to location (City of Cincinnati—Transportation & Engineering, 2020).

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 307–318

Landslide Susceptibility of the Kope Formation

Figure 10. Relationship between strength parameters and factor of safety for the Ten Mile Road Landslide: (a) cohesion vs FS and (b) friction angle vs FS.

Figure 11. Relationship between shear strength parameters and factor of safety for the Columbia Parkway landslide: (a) cohesion vs FS and (b) friction angle vs FS.

The construction started toward the end of 2019 and is expected to be completed by summer 2021 (City of Cincinnati—Transportation & Engineering, 2020).

soil–bedrock contact is an important factor contributing to landslide susceptibility of the Kope Formation.

FACTORS CONTRIBUTING TO LANDSLIDE SUSCEPTIBILITY OF THE KOPE FORMATION Low Shear Strength We believe the residual strength parameters are more important than the peak strength parameters for the long-term stability of slopes composed of colluvial soil derived from the Kope Formation. This is because many of the landslides in the Kope Formation develop progressively over a long period of time. Figures 10 and 11 show the relationships between FS and the residual strength parameters for the Ten Mile Road and Columbia Parkway landslides, respectively. A comparison of these plots with the residual strength parameters (Tables 3 and 4) shows that the residual cohesion and residual friction angle values for both rotational and translational slides are not high enough to support the slopes (i.e., the values in the tables are lower than those required to provide a FS > 1). Therefore, the low shear strength of the colluvial soil and

Porewater Pressure The presence of water within a slope can significantly decrease the stability of a slope. The average amount of precipitation in the Cincinnati area is 107 cm (U.S. Climate Data, 2014). Since the colluvial soil classiﬁes as a clayey sand for all landslides studied, it can be assumed that the material has low permeability and poor drainage characteristics (Holtz et al., 2011). This can lead to buildup of porewater pressure within the slope during prolonged periods of rainfall and snow melt, reducing shear strength and contributing to slope failure. Figure 12 shows the relationship between the location of the water table and the FS for the slopes at the Ten Mile Road and Columbia Parkway landslide sites. In this ﬁgure, 0 (along the vertical axis) represents the water table located at the bedrock level, 1 represents the water table at the ground surface and values in between represent the relative elevations of the water table from the bedrock to the ground surface. The

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 307–318

315

Glassmeyer and Shakoor

top of a slope, the driving forces acting on the slope increase and tend to cause failure. By cutting out the hillsides and the toes of the slopes, the resisting forces decrease. Low to Very Low Durability of the Bedrock The Kope Formation is a clay-bearing rock of low to very low durability against slaking (Id2 = 7.1 percent to 39.9 percent) because of which it easily disintegrates and rapidly erodes. It is the easy disintegration of the Kope Formation that leads to thick accumulation of the colluvial soil on top of bedrock. The nondurable nature of the Kope Formation and the colluvial soil derived from it make these materials susceptible to landsliding. Undercutting of the Slope Toe Many slopes in the Cincinnati area are subject to undercutting of the slope toe by stream erosion (Figure 5). This removes the lateral support, thereby reducing the resisting forces. Undercutting of the slope toe, facilitated by the low durability of the Kope Formation, is a very important factor contributing to high susceptibility of the Kope Formation to landsliding. Figure 12. Relationship between water table location and factor of safety for the (a) Ten Mile Road landslide and (b) Columbia Parkway landslide.

plots in Figure 12 show that as the water table within the slope rises, the FS of the slope gradually decreases. The FS is at its lowest value when the water table is at the ground surface. i.e., the soil is completely saturated. Only partial saturation of the colluvial slopes is required to cause failure, as several other factors also contribute to instability. Many of the slopes in the study area show either continually ﬂowing water or water seeps throughout the year. Thus, development of pore pressure is another important factor that explains the high susceptibility of the Kope Formation to landsliding. Human Activity Human activity is an important factor inﬂuencing the stability of many slopes in the Cincinnati area (Behringer, 1992). Construction activities alter the stability of a slope in two ways: (1) by adding weight to the top of the slope and (2) by removing lateral support at the toe of the slope. Due to the topography of the Cincinnati area, many of the roads are built on tops of hillsides, cut into hillsides, or built in the toe areas by partial removal of the slope toes. By building on

316

Steepness of Slopes The steepness of natural slopes in the Cincinnati area is another contributing factor to landslide susceptibility of the colluvium that is associated with the Kope Formation. The low to very low durability of the bedrock results in rapid down cutting of the valleys, giving rise to steep slopes. Although the bedrock slopes may reach a state of equilibrium at relatively steeper angles, the colluvial soils that cover the bedrock are not strong enough to maintain stability at those angles. Furthermore, many slopes have been over-steepened because of the rapidly eroding streams or human activity. The slope angles in the Cincinnati area range between 20° and 40°, which is generally higher than the residual friction angle values. The results of the stability analysis show that slopes steeper than 15° will not have an adequate factor of safety against failure under wet conditions. This discussion shows that multiple factors, either individually or in combination, contribute to the high susceptibility of the Kope Formation to landsliding. CONCLUSIONS Based on the results of this study, the following conclusions can be drawn:

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 307–318

Landslide Susceptibility of the Kope Formation

1. Rotational and translational landslides are the main types of movement affecting the slopes composed of colluvial soil derived from the Kope Formation. Once a failure has been initiated, both types of movement may transform into earthflows, and occasionally into mudflows, with the addition of water. 2. The factors that contribute to the high susceptibility of the colluvial soil to landslides include low shear strength parameters of the soil or soil/bedrock contact, development of porewater pressure, human activity, low to very low durability of the bedrock, undercutting of the slope toe by stream water, and steepness of the slopes. REFERENCES American Society for Testing and Materials (ASTM), 2010, Annual Book of Standards: Section 4, Construction, 4.08, Soil and Rock (1): ASTM, Conshohocken, PA. Baum R. L., 1983, Engineering Geology and Relative Slope Stability in the Area of the Fay Apartments and in part of Mount Airy Forest, Cincinnati, Ohio: Unpublished M.S. Thesis, University of Cincinnati, Cincinnati, OH, 74 p. Baum R. L. and Johnson, A. M., 1996, Overview of landslide problem, research, and mitigation, Cincinnati, Ohio, Area: U.S. Geological Survey Bulletin 2059-D, 33 p. Behringer, D. W., 1992, A Study of Selected Landslides in the Cincinnati Area in Relation to Human Activity: Unpublished M.S. Thesis, Kent State University, Kent, OH, 164 p. Brett, C. E. and Algeo, T. J., 2001, Stratigraphy of the Upper Ordovician Kope Formation in its type area (northern Kentucky), including a revised nomenclature. In Algeo, T. J. and Brett, C. E. (Editors), Sequence, Cycle and Event Stratigraphy of the Upper Ordovician and Silurian Strata of the Cincinnati Arch Region: Series 12, Guidebook 1, Kentucky Geological Survey, Lexington, KY, pp. 47–64. Brett, C. E.; Algeo, T. J.; and Mclaughlin, P. J., 2003, Use of event beds and sedimentary cycles in high resolution stratigraphic correlation of lithologically repetitive successions: The Upper Ordovician Kope Formation of northern Kentucky and southwestern Ohio. In Harries, P. (Editor), High-Resolution Approaches in Stratigraphic Palaeontology: Plenus Press, Amsterdam, Netherlands, pp. 315–350. Casagrande, A., 1948, Classification and identification of soils: Transactions, American Society Civil Engineers (ASCE), Vol. 113, pp. 901–930. City of Cincinnati—Transportation & Engineering 2020, Columbia Parkway Hillside Stabilization: Electronic document, available at https://www.cincinnati-oh.gov/dote/doteprojects/columbia-parkway-hillside-stabilization/. Cruden D. M. and Varnes D. J., 1996. Landslide types and processes. In Turner A. K. and Schuster R. L. (Editors), Landslides: Investigation and Mitigation: Special Report 247, Transportation Research Board, Washington D.C., pp. 36–75. Dick, J. C., 1992, Relationship Between Durability and Lithologic Characteristics of Mudrocks: Unpublished Ph.D. Dissertation, Kent State University, Kent, OH, 243 p. Dick, J. C. and Shakoor, A., 1995, Characterizing durability of mudrocks for slope stability purposes. In Reviews in Engineer-

ing Geology, Vol. 10: Geological Society of America, Boulder, CO, pp. 121–130. Fleming, R. W., 1975, Geologic Perspectives—The Cincinnati Example: Ohio Valley Soils Seminar Proceedings, Ft. Mitchell, KY, 22 p. Fleming, R. W. and Johnson, A. M., 1994, Landslides in colluvium: U.S. Geological Survey Bulletin 2059-D, 24 p. Gibbons, A. B., 1973, Geologic Map of Parts of the Newport and Withamsville Quadrangles, Campsvill and Kenton Counties, Kentucky: U.S. Geologic Society, Washington D.C. Glassmeyer, M. P., 2014, Geological and Geotechnical Factors Responsible for Landslide Susceptibility of the Kope Formation in Cincinnati, Ohio: Unpublished M.S. Thesis, Kent State University, Kent, OH, 196 p. Hajdarwish, A., 2006, Geologic Controls of Shear Strength Behavior of Mudrocks: Unpublished Ph.D. Dissertation, Kent State University, Kent, OH, 258 p. Haneberg, W. C, 1991, Observation and analysis of pore pressure ﬂuctuations in a thin colluvium landslide complex near Cincinnati, Ohio: Engineering Geology, Vol. 31, pp. 159–184. Haneberg, W. C, 1992, A mass balance model for the hydrologic response of ﬁne-grained hillside soils to rainfall: Geological Society of America Abstracts Programs, Vol. 24, No. 7, p. 203. Haneberg, W. C. and Gökce, A. Ö., 1994, Rapid water-level fluctuations in a thin colluvium landslide west of Cincinnati, Ohio: U.S. Geological Survey Bulletin 2059-C, 16 p. Hofman, H. J., 1966, Deformational structures near Cincinnati, Ohio: Geological Society America Bulletin, Vol. 77, No. 5, pp. 533–548. Holtz, R. D.; Kovacs, W. D.; and Sheahan, T. C., 2011, An Introduction to Geotechnical Engineering, 2nd ed.: Prentice-Hall, Inc., Upper Saddle River, NJ, 853 p. International Association of Engineering Geology (IAEG) Commission on Landslides, 1990, Suggested nomenclature for landslides: Bulletin International Association Engineering Geology, No. 41, p. 13–16. International Society for Rock Mechanics (ISRM), 2007, The complete ISRM suggested methods for rock characterization, testing, and monitoring: 1974–2006. In Ulusay, R. and Hudson, J. A. (Editors), Suggested Methods Prepared by the Commission on Testing Methods: ISRM, Ankara, Turkey, 628 p. Lerch, N. K.; Hale, W. F.; and Lemaster, D. D., 1982, Soil Survey of Hamilton County, Ohio: U.S. Soil Conservation Service, 219 p. Ohio Emergency Management Agency (OHIO EMA), 2011, State of Ohio Emergency Hazard Mitigation Plan: Ohio Department of Public Safety, 345 p. Pavey, R. R.; Goldthwait, R. P.; Brockman, C. S.; Hull, D. N.; and Van Horn, R. G., 1992, The new Quaternary map of Ohio: Geological Society America Abstracts Programs, Vol. 24, No. 7, p. 314. Pohana, R. E., 1983, The Engineering Geologic and Relative Stability Analysis of a Portion of Anderson Township, Cincinnati, Ohio: Unpublished M.S. Thesis, University of Cincinnati, Cincinnati, OH, 132 p. Potter, P. E., 2007, Exploring the Geology of the Cincinnati/Northern Kentucky Region: Kentucky Geological Survey Special Publication 8, Series 12. 128 p. Potter, P. E.; Maynard, J. B.; and Pryor, W. A., 1980, Sedimentology of Shale: Springer-Verlag, New York, 310 p. Richards, K. A., 1982, The Engineering Geology and Relative Stability of Mt. Adams, and Parts of Walnut Hills and Columbia

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 307–318

317

Glassmeyer and Shakoor Parkway, Cincinnati, Ohio: Unpublished M.S. Thesis, University of Cincinnati, Cincinnati, OH, 111 p. Riestenberg, M. M. and Sovonik-Dunford, S., 1983, The role of woody vegetation in stabilizing slopes in the Cincinnati area, Ohio: Geological Society America Bulletin, Vol. 94, pp. 506–518. Rockaway, J., 2002, Southwestern Ohio Landslide Documentation Investigation Report: U.S. Geological Survey, Denver, CO, pp. 2–3. ROCSCIENCE, 2012, Slide: University of Toronto, Toronto, Ontario, Canada

318

Sarman, R., 1991, A Multiple Regression Approach to Predict Swelling in Mudrocks: Unpublished Ph.D. Dissertation, Kent State University, Kent, OH, 365 p. Schuster, R. L., 1996, Socioeconomic signiﬁcance of landslides. In Turner A. K. and Schuster R. L. (Editors), Landslides: Investigation and Mitigation: Special Report 247, Transportation Research Board, Washington D.C., pp. 12–35. U.S. Climate Data, 2014, Electronic document, available at https://www.usclimatedata.com/climate/cincinnati/ohio/ united-states/usoh0188.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 307–318

Spatiotemporal Evaluation of Flood Potential Indices for Watershed Flood Prediction in the Mississippi River Basin, USA DORCAS IDOWU WENDY ZHOU* Department of Geology and Geological Engineering, Colorado School of Mines, 1516 Illinois Street, Golden, CO 80401

Key Terms: GRACE, Flood Potential Index, Hydrological Models, Total Water Storage, Water Budget, Mississippi River Basin, Flood ABSTRACT The Gravity Recovery and Climate Experiment (GRACE) and GRACE Follow-On twin satellites have revealed mass variation over time by measuring Earth’s gravity field anomalies since 2002. These variations can be interpreted as changes in water storage on and beneath the earth’s surface. The water storage anomaly, also known as the Terrestrial Water Storage Anomaly (TWSA), has been utilized successfully to estimate a Flood Potential Index (FPI). The GRACE-based FPI (GRACE-FPI) is a good indicator of when a region is at risk of flooding. However, the GRACE data are limited by low spatial resolution, which has limited applicability to study areas smaller than 200,000 km2 . Since the change in storage derived from the traditional water budget equation has the same physical meaning as the GRACE TWSA, we hypothesize there are similarities in their derived FPIs. In this study, we propose another index, known as the Water Budget–based FPI (WB-FPI), which is derived using higher spatial resolution hydrological datasets and could produce similar but higherresolution results than the GRACE-FPI. When combined with the GRACE-FPI, the index could be valuable in providing flood prediction estimates on a catchment scale. Additionally, a combination of both indices could significantly reveal hydrological details for small basins. Both the GRACE-FPI and WB-FPI were applied to a case study of the flood potential of the U.S. Mississippi River Basin. Our study reveals good correlation between the resampled GRACE-FPI and WB-FPI using the Nash-Sutcliffe efficiency index to compare application efficacy.

*Corresponding author email: wzhou@mines.edu

INTRODUCTION Floods are the most common and widespread of all weather-related natural disasters. Flooding is a natural hazard experienced in all regions of the world that receive a form of precipitation and in every state and territory of the United States. In the United States, fatalities resulting from flooding averaged 95 yearly during the last 10 years, more than any other natural weatherrelated disaster, such as tornadoes, hurricanes, or lightning. The most common cause of flooding is heavy rain and/or snowmelt that accumulates water faster than the ground can absorb or rivers can contain. The Midwestern United States was the wettest (Figure 1) on record between January and May 2019 due to widespread extreme weather and heavy rainfalls. “The Great Flood of 2019,” as named by the New York Times, affected nearly 14 million people in the midwestern and southern states (Almukhtar et al., 2019), and at least three fatalities were reported. Damages were also done to agricultural lands, and at least 4,047 km2 of U.S agricultural land in nine major grainproducing states were flooded. Longer lead-time flood prediction could remarkably reduce flood-related losses but requires accurate information on the hydrologic state of a whole river catchment, that is, its total water storage. Unfortunately, such detailed data are arduous to acquire with existing hydrological networks. However, data from the Gravity Recovery and Climate Experiment (GRACE) twin satellites have exhibited some usefulness for monitoring the total/terrestrial water storage (TWS) and, ultimately, predicting floods (Reager et al., 2014; Molodtsova et al., 2016; and Idowu and Zhou, 2019). It provides the means to detect monthly variations in TWS within large river basins (>200,000 km2 ) based on measurements of the changes in Earth’s gravity field. The TWS indicates the temporal ability of the land to process and account for all water on and beneath the earth’s surface. A satellite and storage-based “flood potential” method using GRACE measurements has also proved invaluable at identifying major flood occurrences

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 319–330

319

Idowu and Zhou

Figure 1. Precipitation averaged across the contiguous United States for the period January to May for all years going back to 1895, with 2019 standing above all prior years to date. Image credit: NOAA (National Oceanic and Atmospheric Administration).

globally (Reager and Famiglietti, 2009). Over the GRACE twin satellite mission length, each region tends to exhibit an effective storage capacity, beyond which marked increases in runoff or evaporation must balance additional precipitation. These saturation periods indicate the possible transition to flooding (Reager et al., 2014). TWS is a major element in the global and continental water cycle (Yin et al., 2019). “GRACE has provided a highly valuable dataset, which allows the study of TWS over larger river basins worldwide” (Yin et al., 2019). However, the GRACE effective spatial resolution of 200,000 km2 makes it only possible to observe monthly variations in TWS at a continental or regional scale. Reager and Famiglietti (2009) introduced a quantitative, effective, terrestrial storage capacity and defined a flood potential index (FPI) to emphasize the information contained within the GRACE data pertinent to regional flooding. Idowu and Zhou (2019), in their study using 1° by 1° grid datasets, showed that some correlation exists between the GRACE-derived TWS and the water storage derived from the hydrological water budget. The effective spatial resolution of GRACE data for resolving the localized hydrologic-related problem has been a debate. Several researchers (Zaitchik

320

et al., 2008; Sun, 2013; Seyoum and Milewski, 2017; and Miro and Famiglietti, 2018) proposed a series of methods to improve its spatial resolution. Because the minimum effective signal-to-noise ratio dictates the effective GRACE spatial resolution, the water budget– based FPI (WB-FPI) could complement the GRACEbased FPI (GRACE-FPI). Combining these two indices could reduce the limitation of the GRACEFPI for studying small areas or local-scale hydrologic events. Therefore, our study evaluated the use of model-based data with finer spatial resolution in deriving a water budget–based TWS and produced a WBFPI that could serve as an auxiliary for the GRACEbased TWS and FPI. The efficacy of the derived water budget–based TWS and WB-FPI was assessed on the continental United States by evaluating how well the predictions compare to streamflow datasets. STUDY AREA The Mississippi River is the second longest river on the North American continent. Its drainage basin is the world’s second largest, draining an approximate area of 4,800,000 km2 that includes tributaries from 32 U.S. states and two Canadian provinces. The Mississippi River watershed encompasses 40 percent of the contiguous United States, and the major

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 319–330

Evaluation of Flood Potential Indices for Watershed Flood Prediction

tributaries include the Missouri, Ohio, Arkansas-RedWhite, and Tennessee Rivers. It consists of five major sub-basins, including the Upper Mississippi, Lower Mississippi, Missouri, Arkansas Red-White, Ohio, and Tennessee River Basins. The Mississippi River and its tributaries have flooded on many occasions. Historically, heavy rainfall and snowmelt events that cause a slow rise in the rivers and extend for days or weeks are seemingly the major culprit responsible for flooding along the river. Floods and flood damages significant enough to merit regional or national attention occurred in the Mississippi River Basin between the 19th and 21st centuries. In 2019, flooding along the Mississippi River was not the result of a single weather event; instead, the Mississippi River flooding of record duration consisted of a series of flood events in tributary basins. Different measures, including structural and non-structural flood control measures (Isaacson et al., 1956; Dehaan et al., 2012; Semmens et al., 2017; Semmens and Zhou, 2020; and Semmens et al., 2021), have been taken to mitigate the devastating effects of floods along the river, yet flooding is persistent. Longer flood warning lead time could give stakeholders enough time to take measures that may reduce the economic impact and damages resulting from floods along the river. Reager et al. (2014) showed that GRACE-based TWS information could increase regional warning lead times to as long as 5 months. However, the GRACE spatial resolution could be a limiting factor for its application to the study of flood potential on a basin-scale or to pinpoint approximate locations within a basin that require emergency flood mitigation measures. Here, we test the accuracy and performance of a 0.125° × 0.125° water budget–based TWS for predicting flooding within the Mississippi River Basin and compare the accuracy to the predictions from the GRACE-based TWS. DATASETS GRACE Terrestrial Water Storage The GRACE satellite and its Follow-On twin measure the changes in Earth’s gravity field on a monthly temporal scale at a spatial resolution of 1° × 1°. These observed changes are related to changes in surface mass resulting from water movement on and/or within the surface of the earth or mass redistribution from earthquakes. The GRACE-based TWS on land represents the combination or sum of all water stored on the land surface. Data acquired by the satellites, which are spherical harmonics, are officially processed by three centers: Jet Propulsion Laboratory (JPL), GeoforschungsZentrum Potsdam, and Center for Space

Research (CSR) at the University of Texas. Although the centers applied different processing methods and techniques on the GRACE spherical harmonics, the resulting products, referred to as solutions, were similar. For our study, we used the 1° × 1° spherical harmonics solution from the CSR at the University of Texas, Austin. The GRACE RL05 TWSA products were downloaded from the GRACE Tellus website (JPL/CIT/NASA, n.d.). The spatial resolution is approximately 111 km × 111 km. We downloaded data for May 2011 for the Mississippi River Basin. We also obtained the GRACE Follow-On (GRACE-FO) level 3 data from January to May 2019 (PODACC, n.d.). North American Land Data Assimilation System Data The North American Land Data Assimilation System (NLDAS) land-surface model datasets, derived from a collaborative project among several groups, are spatially and temporally consistent from the best available observations and are reanalyzed to support modeling activities. This effort is a core project with support from the National Oceanic and Atmospheric Administration (NOAA) climate program office’s Modeling, Analysis, Prediction and Projections (MAPP) program. Monthly 0.125° × 0.125° mosaicked and secondary-forcing NLDAS data for some hydrological variables were retrieved from the NASA Land Data Assimilation System website. The spatial resolution is approximately 11 km × 11 km. Precipitation Data The 1° × 1° Global Precipitation Climatology Centre (GPCC) data provided by the Office of Research and Applications (ORA), the Earth System Research Laboratory (ESRL), Physical Science Division (PSD) at the National Oceanic and Atmospheric Administration (NOAA) were used for deriving the GRACE-FPI. The 0.125° × 0.125° NLDAS secondary-forcing precipitation data downloaded from the NASA NLDAS website were used for deriving the WB-FPI. Dartmouth Flood Observatory Data The Dartmouth Flood Observatory (DFO; Brakenridge, 2018) has an active global archive of large flood events. Information in the archive is derived from news, governmental, instrumental, and remote-sensing sources. We used the zip-compressed GIS shape format, which displays a polygon of the total area affected by flooding for different dates. Polygons of dates corresponding to our study baseline were used for our validation. It is worthy of mention that the DFO source of

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 319–330

321

Idowu and Zhou

Figure 2. Map of the United States showing the Mississippi River and its major drainage basins.

validation is the news media and government sources and may thus be biased toward large flood events. United States Geological Survey Streamflow Data Maps of monthly averaged streamflow from the United States Geological Survey (USGS) water watch data services were also used to validate our analysis. The map displays river discharge in percentile classes on a scale of 100. For instance, “above normal” is streamflow greater than the 75th percentile, whereas “normal” is between the 25th and 75th percentiles and “below normal” is streamflow less than the 25 percentile. These maps were georeferenced in the ArcGIS suite. Sentinel-2 Data The Copernicus Sentinel-2 satellite mission monitors variability in land surface conditions while its swath width (290 km) and revisit time support the monitoring of Earth’s surface changes. The satellite also provides systematic global acquisitions of high spatial and temporal resolution multispectral images or observations for land-cover maps, land-change detection maps, and geophysical variables. The European Space Agency’s Sentinel-2 satellite mission monitors

322

the variability of land surface conditions. The Sentinel2 optical satellite imageries were used for validating our results. The satellite data can be obtained via the Copernicus Open Access Hub (ESA, n.d.). METHODOLOGY In deriving the GRACE-FPIs and WB-FPIs, we applied the methodology proposed by Reager and Famiglietti (2009) and Idowu and Zhou (2019). To compare the index performance, we present two case studies for the Mississippi River Basin (Figure 2), that is, the 2011 and 2019 flood events. Furthermore, for consistency in our analysis, we used the GRACE TWSA baseline because it has a shorter record length (2004–2017) in comparison with the NLDAS datasets (1979–recent). To further test the WB-FPI derived using the 0.125° × 0.125° NLDAS datasets, we studied the recent 2019 flood event (Figure 1) within the basin and validated the predictions using the streamflow datasets and satellite images from the Sentinel-2 satellite. GRACE Flood Potential Index Throughout the GRACE TWSA record length, there are persistent annual maxima that correspond

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 319–330

Evaluation of Flood Potential Indices for Watershed Flood Prediction

Figure 3. Normalized (a) GRACE-FPI and (b) WB-FPI for May 2011 showing areas of high (red) FPI and low (blue) FPI in the Mississippi River Basin.

to hydrological extremes. As a result, we deﬁned our maximum water storage capacity (Smax ) as the historical maximum water storage capacity within the Mississippi drainage basin, our area of interest. The storage deﬁcit (Sdef ), which is the available water on land before attaining the Smax , was derived using Eq. 1: Sdef (t) = Smax − TWSA (t − 1) ,

(1)

where TWSA (t − 1) is the total water storage from the previous month (Idowu and Zhou, 2019), and Sdef implies a shortage or deficit in water storage within a basin or an area on the land. The deficit, Sdef , is expected to increase or decrease during the drier or wetter part of the year. The flood potential, which is the quantity of incoming water above available storage, was calculated using F (t) = Pmon (t) − Sdef (t) ,

(2)

where F(t) is the ﬂood potential and Pmon (t) is the monthly precipitation from the GPCC. The ﬂood potential was further normalized (Reager and Famiglietti, 2009) to derive the GRACE-FPI: FPI (t) =

F (t) . max [F (t)]

(3)

The GRACE-FPI values range from − to 1, with positive values implying the hydrological inputs are above the normal water storage and should be interpreted as a potential ﬂood risk.

Water Budget Flood Potential Index First, we derived the TWSA, which is also ds/dt, by using the traditional water balance equation: ds = P − ET − R − SM, dt

(4)

where ds/dt is TWSA; P is precipitation, which includes rainfall and snow; ET is evapotranspiration; R is the runoff; and SM is soil moisture. Here, we assumed there was no net groundwater flow. Next, we applied the methodology for deriving the GRACE-FPI in calculating the WB-FPI. GRACE-FPI and WB-FPI Comparison To validate the WB-FPI, we tested its prediction skill against the GRACE-FPI for May 2011 (Figure 3) in the Mississippi River Basin and the six U.S. regions (Missouri, Upper Mississippi, Ohio, Lower Mississippi, Arkansas-White-Red, and Tennessee) using the Nash-Sutcliffe efficiency (NSE) index, Pearson’s correlation coefficient (r), and the coefficient of determination (R2 ). The NSE is an efficiency index that is potentially reliable and commonly used for assessing the goodness-of-fit of hydrological models (Nash and Sutcliffe, 1970; McCuen et al., 2006). It is commonly referred to as the efficiency index (Ef ) and is defined as n

(Xi − Yi )2 E f = 1 − n i=1 2 , i=1 Yi − Yavg

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 319–330

(5)

323

Idowu and Zhou Table 1. The contingency table and selected performance metrics. USGS stream ﬂow condition WB-FPI Predicted yes Predicted no

Actual yes

Actual no

True positive (TP) False negatives (FN) True positive rate (TPR) = TP/actual yes

False positives (FP) Trues negatives (TN) False positive rate (FPR) = FP/actual no

where n is the sample size, Xi is the ith observation value, Yi is the ith simulated value, and Yavg is the mean of observed data (Moriasi et al., 2007). Generally, NSE ranges from − to 1. Moriasi et al. (2007) suggested that NSE values generally between 0 and 1 are viewed as acceptable levels of performance, whereas NSE values less or equal to 0 are interpreted as the mean observed value is a better predictor than the simulated value. R2 and r are used to describe the level of collinearity between simulated and measured data. R2 indicates how much variance in the measured data is explained by the model and ranges from 0 to 1, with higher values suggesting less error variance. Commonly, values higher than 0.5 are considered acceptable (Moriasi et al., 2007). Pearson’s correlation coeﬃcient is an index that explains the level of the linear relationship between observed and simulated data (Moriasi et al., 2007). It ranges from −1 to 1. When r is 1 or −1, there exists a perfect positive or negative linear relationship, and when r is 0, no linear relationship exists. For the index comparison, we resampled the GRACE-FPI resolution of 1° × 1° to that of the WBFPI, which is 0.125° × 0.125°. With this, we computed and compared the statistics for both indices.

Georeferencing USGS Streamflow Condition Datasets The USGS streamflow condition maps were obtained in raster format (Graphics Interchange Format) without prior spatial reference information. To view, query, analyze, and accurately compare them to the WB-FPI predictions for our study period, we georeferenced them. Georeferencing is associating a physical map or raster image with spatial locations or map coordinates. To achieve this, we built an attribute table for each of the maps by using their unique values. The values in the USGS maps that were of interest were those having a percentile value of above normal to high streamflow readings. These pixels were isolated by reclassifying the maps and setting irrelevant pixels to NoData, leaving the pixels of interest that were georeferenced.

324

Predictive Performance Test Our classification utilized only two classes. Formally, each instance (e.g., the USGS streamflow condition) is mapped to one element of the set {p (flood), n (no flood)} of positive and negative class (flood or no flood) labels. A classification model (or classifier, which in this case is WB-FPI) is a mapping from instances to predicted classes (Fawcett, 2006). Some models produce discrete class labels, indicating only the predicted class of an instance. To differentiate between the actual class and the predicted class (Table 1), we used the labels {Y, N} for the class predictions produced by our model. Given a classifier (WB-FPI) and an instance (USGS streamflow condition), there are four possible outcomes (Table 1). A positive instance classified as positive is counted as a true positive (TP). If classified as negative, it is counted as false negative (FN). However, if the instance is negative and classified as negative, it is counted as true negative (TN), and if classified as positive, it is counted as false positive (FP). The receiver operative characteristics (ROC) is a technique for visualizing, organizing, and selecting classifiers based on their performance and is generally a useful performance graphing method (Fawcett, 2006). ROC graphs that show the relative tradeoffs between TP and FP are two-dimensional graphs in which the TP rate (TPR) is plotted on the Y-axis, and the FP rate (FPR) is plotted on the X-axis. It is also commonly used as a method for testing the performance of a continuous index (such as the FPI) as a discrete classifier against the USGS streamflow datasets. A discrete classifier is one that outputs only a class label. Each discrete classifier produces a pair of TPR and

Table 2. Statistical Comparison of GRACE-FPI and WB = FPI for May 2011. Statistics of individual layers Minimum Maximum Mean Standard deviation Correlation coeﬃcient

GRACE-FPI

WB-FPI

− 0.2 1.0 0.2 0.2 0.8

− 0.3 1.0 0.2 0.2

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 319–330

Evaluation of Flood Potential Indices for Watershed Flood Prediction

Figure 4. (a) GRACE-FPI at a spatial resolution of 1° × 1° and (b) WB-FPI at a spatial resolution of 0.125° × 0.125° in May 2011 for the Tennessee region within the Mississippi River Basin.

FPR corresponding to a single point in the ROC space. Here, we compared the performance of the WB-FPI against the USGS streamﬂow data from January 2019 to May 2019 within the Mississippi River watershed. Each month represents a point on the ROC space. Several points in the ROC space are important to note, and, informally, one point in the ROC space is better than another (Fawcett, 2006). The 1:1 diagonal line in a ROC plot represents a random guess, which is no predictive skill, whereas areas below the line assume predictive skill worse than a random guess. The closer the points are to the upper left corner of the graph, the better the predictive skill. RESULTS Figure 3 is the visual comparison of both indices at their inherent spatial resolution, which displays a

similar spatial pattern. Qualitatively, both indices predicted flooding in areas where the DFO reported flood for May 2011 in the Mississippi River Basin. The WBFPI, when statistically compared to the GRACE-FPI after resampling to 0.125° × 0.125°, shows some agreement with the index (Table 2). Figure 4 shows the computed GRACE-FPI and WB-FPI at their inherent spatial resolution for the Tennessee region within the Mississippi River Basin. Visually, there exist some similarities in the spatial variation of indices despite having a different spatial resolution. Relatively speaking, the GRACEFPI tends to underestimate the flood potential, especially in the central portion of the Tennessee region. Furthermore, we computed the GRACE-FPI and WB-FPI for each region within the Mississippi River Basin from January to May 2019. These months were

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 319–330

325

Idowu and Zhou Table 3. Statistical comparison of GRACE-FPI and WB-FPI using NSE, r, and R2 methods from January to May 2019. U.S. regions Missouri region Upper Mississippi region Ohio region Lower Mississippi region Arkansas-White-Red region Tennessee region Average

NSE

0.1 0.3 0.9 0.8 0.6 0.9 0.6

0.6 0.9 0.8 0.8 0.8 0.9 0.8

0.4 0.8 0.7 0.7 0.7 0.8 0.7

named the wettest on record. We compared the indices qualitatively and quantitatively (Figure 5). For each month, Table 3 shows that the NSE values for the regions suggest an acceptable performance level of the WB-FPI for ﬂood prediction using the data from January to May 2019. The r and R2 values also indicate some agreement between both indices. The predictive skills of the WB-FPI were tested against the USGS streamﬂow polygons using the con-

Table 4. ROC space graph performance metrics from January to May 2019.

January February March April May Average

Accuracy

TPR

FPR

Precision

0.8 0.7 0.7 0.7 0.9 0.8

0.9 0.9 0.9 0.8 0.9 0.9

0.7 0.7 0.0 0.6 0.8 0.7

0.9 0.7 0.7 0.8 0.9 0.8

tingency table and ROC space graph. The polygons resulted from georeferencing the USGS streamflow maps and converting them from raster to polygon. We then computed a contingency table (confusion matrix) using a discrete classifier with two classes: Yes (flood) or No (no flood). We assigned the class “Yes (flood)” to all areas within the USGS streamflow polygon and assigned the class “No (no flood)” for areas outside the USGS streamflow polygon. Each contingency table produces a TPR, an FPR, accuracy, and precision

Figure 5. Visualization and comparison of the spatial distribution of the (a) GRACE-FPI and (b) WB-FPI for the Mississippi River Basin from January to May 2019.

326

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 319–330

Evaluation of Flood Potential Indices for Watershed Flood Prediction

Figure 6. ROC space graph showing a flood classification of the WB-FPI and the USGS streamflow condition from January to May 2019 and their average point.

for a discrete classifier (Table 4). We plotted the TPR for each month against the FPR on a ROC space graph to test how the WB-FPI predicted floods as compared to the USGS streamflow data from January to May. Figure 6 shows that each point and averaged point plotted on the graph fell on the better half of the graph above the random guess line and suggests that WBFPI is a better classifier (Fawcett, 2006). We qualitatively compared pre-flood and duringflood scenes from Sentinel-2 to areas with positive WB-FPI from January to May 2019. The WB-FPI predicted flooding in areas in the Sentinel-2 scenes. All WB-FPI values compared to the Sentinel-2 scenes have FPI values greater than zero (Figures 7–9), which is indicative of a flood situation. The WB-FPI shows areas of high (red) and low (blue) FPI, and areas of high FPI are shown on the during-flood Sentinel-2 scenes for January and May 2019. The spatial variations in the FPI values seen in Figures 7–9 are a result of the changes in water storage within the basin, which can be associated with heavy rainfall and/or snowmelt. These variations are also reflected in Figure 6.

DISCUSSION AND CONCLUSION Our study suggests a good correlation between the comparison of the resampled GRACE-FPI and WBFPI (Figure 3) for the Mississippi River Basin. The ROC space graph was plotted for WB-FPI to determine how well it predicted flooding at different times when compared to the USGS streamflow data. The points plotting above the diagonal line (random guess line) imply a better classification (Figure 6). The location of the points (upper left above the random guess line) implies the WB-FPI could predict every flood event correctly (high TPR) but also has a tendency to over-predict floods (hence, the high FPR). The GRACE-FPI has a similar downside (Idowu and Zhou, 2019). A machine learning method could be applied to data to reduce the FPR. Furthermore, the differences between the indices possibly resulting in high FPR could be attributed to measurement errors and uncertainties inherent in the data (Güntner, 2008). The statistical similarities between the GRACE-FPI with a spatial resolution of 1° × 1° and WB-FPI having a spatial resolution of 0.125° × 0.125° further suggest

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 319–330

327

Idowu and Zhou

Figure 7. Pre- and during-ﬂood Sentinel-2 scenes with WB-FPI for January and February 2019 for an example section of the Mississippi River.

Figure 8. Pre- and during-ﬂood Sentinel-2 scenes with WB-FPI for March and April 2019 for an example section of the Mississippi River.

328

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 319–330

Evaluation of Flood Potential Indices for Watershed Flood Prediction

Figure 9. Pre- and during-ﬂood Sentinel-2 scenes with WB-FPI for May 2019 for an example section of the Mississippi River.

that the WB-FPI derived using satellite-based highresolution data could predict flooding on a local scale. The ability to effectively predict floods for smaller areas could further improve flood prediction, especially in developing countries where ground flood monitoring systems are sparse or not available. Data continuity presented by the GRACE-FO and by the NLDAS shows the potential for future flood potential or prediction studies using the methodologies presented in this research or a combination of other flood prediction or forecasting models. ACKNOWLEDGMENTS This work was funded by the American Association of University Women (AAUW) scholarship in 2018, Chevron international fellowship in 2019, and the Roy J. Shelmon scholarship of the Geological Survey of America. REFERENCES Almukhtar, S.; Migliozzi, B.; Schwartz, J.; and Williams, J., 2019, The great flood of 2019: A complete picture of a slowmotion disaster: New York Times, available at https://www. nytimes.com/interactive/2019/09/11 / us / midwest-flooding. html. Brakenridge, G. R., 2018, Global active archive of large flood events, Dartmouth Flood Observatory, University of Colorado, available at http://floodobservatory.colorado.edu/Archives/.

DeHaan, H.; Stamper, J.; and Walter, B., 2012, Mississippi River and Tributaries System 2011 Post-Flood Report: Documenting the 2011 Flood, the Corps’ Response, and the Performance of the MR&T System: USACE, Mississippi Valley Division, available at https://hdl.handle.net/11681/19106. ESA (European Space Agency), n.d., Copernicus open access hub, available at https://scihub.copernicus.eu. Fawcett, T., 2006, An introduction to ROC analysis: Pattern Recognition Letters, Vol. 27, No. 8, pp. 861–874. https://doi.org/10.1016/j.patrec.2005.10.010. Güntner, A., 2008, Improvement of global hydrological models using GRACE data: Surveys in Geophysics, Vol. 29, pp. 375– 397. https://doi.org/10.1007/s10712-008-9038-y. Idowu, D. and Zhou, W., 2019, Performance evaluation of a potential component of an early flood warning system— A case study of the 2012 flood, Lower Niger River Basin, Nigeria: Remote Sensing, Vol. 11, No. 17, 1970. https://doi.org/10.3390/rs11171970. Isaacson, E.; Stoker, J.; and Troesch, A., 1956, Numerical solution of flood prediction and river regulation problems: New York University Institute of Mathematical Sciences. Contract DA-33-017.ENG.267 prepared under the sponsorship of U.S. Army Corps of Engineers, Ohio River Division. 132 p. JPL/CIT/NASA (Jet Propulsion Laboratory/California Institute of Technology/National Aeronautics and Space Administration), n.d., Measuring Earth’s surface mass and water changes, available at: https://grace.jpl.nasa.gov/. McCuen, R. H.; Knight, Z.; and Cutter, A. G., 2006, Evaluation of the Nash-Sutcliffe efficiency index: Journal of Hydrologic Engineering, Vol. 11, No. 6, pp. 597–602. doi:10.1061/(ASCE)1084-0699(2006)11:6(597). Miro, E. M. and Famiglietti, S. J., 2018, Downscaling GRACE remote sensing datasets to high-resolution groundwater stor-

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 319–330

329

Idowu and Zhou age change maps of California’s Central Valley: Remote Sensing, Vol. 10, No. 1, p. 143. Molodtsova, T.; Molodtsov, S.; Kirilenko, A.; Zhang, X.; and VanLooy, J., 2016, Evaluating flood potential with GRACE in the United States: Natural Hazards and Earth System Sciences, Vol. 16, pp. 1011–1018. Moriasi, D. N.; Arnold, J. G.; VanLiew, M. W.; Bingner, R. L.; Harmel, R. D.; and Veith, T. L., 2007, Model evaluation guidelines for systematic quantification of accuracy in watershed simulations: Transactions of the ASABE, Vol. 50, No. 3, p. 885e900. Nash, J. E. and Sutcliffe, J. V., 1970, River flow forecasting through conceptual models. 1: Discussion of principles: Journal of Hydrology, Vol. 10, No. 3, pp. 282–290. doi:10.1016/0022-1694(70)90255-6. PODAAC (Physical Oceanography Distributed Active Archive Center/JPL/CIT/NASA), n.d., TELLUS_GRFO_L, available at: http://podaac.jpl.nasa.gov/ dataset/TELLUS_GRFO_L#_JPL_RL06_LND. Reager, J. T. and Famiglietti, J. S., 2009, Global terrestrial water storage capacity and flood potential using GRACE: Geophysical Research Letters, Vol. 36, p. L23402. Reager, J. T.; Thomas, B. F.; and Famiglietti, J. S., 2014, River basin flood potential inferred using GRACE gravity observations at several months lead time: Nature Geoscience, Vol. 7, No. 8, pp. 588–592. Semmens, S. N. and Zhou, W., 2020, Predicting backward erosion piping hazard, Lower Mississippi Valley, USA: Quarterly

330

Journal Engineering Geology Hydrogeology, Vol. 54, No. 1. https://doi.org/10.1144/qjegh2020-035. Semmens, S. N.; Zhou, W.; and Robbins, B., 2021, Empirical assessment of blanket thickness as an indicator of backward erosion piping, Lower Mississippi Valley, USA: Natural Hazards Review, Vol. 22, No. 2, 1–9. https://doi.org/10.1061/(ASCE)NH.1527-6996.0000445. Semmens, S. N.; Zhou, W.; van Wesenbeeck, B. K.; and Santi, P. M., 2017, Application of multiple criteria decision making model for evaluation of levee sustainability: Environmental Engineering Geoscience, Vol. 23, No. 2, pp. 65–78. https://doi.org/10.2113/gseegeosci.23.2.65. Seyoum, W. M. and Milewski, A. M., 2017, Improved methods for estimating local terrestrial water dynamics from grace in the northern high plains: Advances in Water Resources, Vol. 110, pp. 279–290. Sun, A. Y., 2013, Predicting groundwater level changes using GRACE data: Water Resources Research, Vol. 49, pp. 5900– 5912. Yin, W.; Hu, L.; Han, S.-C.; Zhang, M.; and Teng, Y., 2019, Reconstructing Terrestrial Water Storage Variations from 1980 to 2015 in the Beishan Area of China: Geofluids, Vol. 2019, 3874742. https://doi.org/10.1155/2019/3874742. Zaitchik, B. F.; Rodell, M.; and Reichle, R. H., 2008, Assimilation of GRACE terrestrial water storage data into a land surface model: Results for the Mississippi River Basin: Journal of Hydrometeorology, Vol. 9, pp. 535–548.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 319–330

Improved Automated Mapping of Sinkholes Using High-Resolution DEMs YONATHAN ADMASSU* Geology and Environmental Science, James Madison University, 801 Carrier Drive Room, 3232 Harrisonburg, Virginia 22807

CELESTINE WOODRUFF Mathematics and Statistics, James Madison University, 60 Bluestone Drive Room, 305 Harrisonburg, Virginia 22807

Key Terms: LiDAR, DEMs, Sinkholes, Shape Factors, Curvature ABSTRACT Sinkholes are common surface manifestations of the presence of networks of subsurface caverns in areas where the bedrock geology is dominated by soluble rocks such as limestones. Accurate mapping of sinkholes is crucial as they are hazardous to transportation infrastructure and may serve as conduits of contaminants to the groundwater. The use of high-resolution digital elevation models extracted from LiDAR and tools in ArcGIS have made it a simple task to automate the process of identification of closed depressions. However, these automated methods do not differentiate between sinkholes and other man-made depressions. Multivariate statistical methods such as linear discriminant analysis, quadratic discriminant analysis, and logistic regression were used to produce predictive models based on selected shape factor values such as circularity, sphericity, and curvature. Curvature values, especially when combined with circularity, were found to be the most powerful variables in separating closed depressions into sinkholes and other artificial depressions. INTRODUCTION Sinkholes are common ground depressions caused when surface drainage erodes into subsurface networks of openings that are common in areas underlain by limestones, dolomites, and evaporites. In addition to geologic factors, hydrologic and anthropogenic factors such as topography, proximity to geologic structures (faults and fold axes), and hydrologic features play a role in the location and timing of sinkhole development (Doctor et al., 2008; Doctor and Doctor, 2012). In some cases sinkholes can form due to leak-

*Corresponding author email: admassyx@jmu.edu.

ing pipes or groundwater seepage that cause subsurface dissolution and erosion. Sinkholes are hazardous to transportation infrastructure and to the foundations of critical facilities such as hospitals, schools, and power plants (particularly nuclear power plants). They can also create an open passageway for surface contaminants to pollute the groundwater. Mapping sinkholes is therefore important for engineers and geologists during route selection for roads and pipelines. Sinkhole maps are also useful for geologists who study spatial controls of sinkhole development. Sinkholes can be mapped using traditional mapping techniques in the field or remotely using aerial photographs or digital elevation models (DEMs). Highresolution DEMs extracted from airborne LiDAR (light detection and ranging) are capable of scanning the ground surface below vegetation, making them far superior to aerial photography for mapping sinkholes (Figure 1). The basic methods of mapping sinkholes from DEMs are visual identification and automated delineation of closed depressions. Visual identification can be made on Hillshade maps, which are digital surface models created in the ArcGIS software. Other visual approaches include using the Topographic Position Index (TPI). TPI is a moving window method in ArcGIS that calculates the difference between each elevation value and the mean elevation within the window (Jenness et al., 2011). Negative TPI values indicate topographic lows, which may represent closed depressions (Doctor and Young, 2013). Automated mapping tools in ArcGIS have made it a simple process to identify closed depressions (Doctor and Young, 2013; Hofierka et al., 2018; Wu et al., 2016; and Gochenour and Admassu, 2017). One method using ArcGIS is automated identification of closed contours, which are typical indicators of the presence of depressions. Another commonly used method is also accomplished using ArcGIS by filling sinks (sites of flow accumulation) in a DEM and subtracting it from the original DEM so that closed depressions stand out (De Carvalho, 2013; Doctor and Young, 2013). However, the fill sinks-

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

331

Admassu and Woodruﬀ

Figure 1. (a) Hillshade map extracted from a 1-meter, LiDAR-derived DEM showing sinkholes. (b) An aerial photo of the same area, where none of the closed depressions are apparent.

subtract method results in too many non-sinkhole artificial depressions such as roadway/bridge or roadway/culvert intersections, automated processes return too much noise by identifying a large number of closed depressions that are not sinkholes but are artificial depressions at roadway/bridge or roadway/culvert intersections (Doctor and Young, 2013) (Figure 2). A suggested method to reduce the formation of fills at roadway intersections is by lowering the elevation of roads and bridges to disallow filling of artificial depressions that may end up being classified as closed depressions (Maidment, 2002). The objective of this study is to recommend a method to separate these ArcGIS-

generated closed depressions into sinkholes and other artificial depressions (non-sinkholes) based on shape factors (circularity, sphericity, curvature), thereby reducing these false positive depressions. The approach is 1) to produce statistical predictive models using data from a training site and 2) to use the predictive models at a different test site to evaluate the effectiveness of the models. METHOD A study site to be used as a training site was selected in the Shenandoah Valley in the vicinity of

Figure 2. The training site located in the Shenandoah Valley around Harrisonburg, VA. The red areas are example sinkholes whereas the green are non-sinkholes generated through an automated “Fill sinks - subtract” method.

332

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

Improved Automated Mapping of Sinkholes

1) Calculate perimeter, area, and coordinates of the centroid of each polygon in ArcGIS (Figure 3). 2) Convert the boundaries of each closed depression into a series of points using the ”Generate Points” tool in ArcGIS (Figure 3). These boundary points were generated every 5 percent of the length of the perimeter, resulting in 19 points per closed depression.

Figure 3. Flowchart showing the workﬂow for processing a DEM to extract closed depressions and calculate shape factors.

Harrisonburg, VA, where the bedrock geology is underlain by Cambrian-Ordovician age carbonates that are riddled with sinkholes. A 1-meter-resolution LiDAR–derived DEM was used to automatically recognize closed depressions (Figure 2). Using the “Fill” tool in ArcGIS, all local depressions can easily be identified as sites of flow accumulation. The “Fill” tool raises the elevation of sinks where surface flow can accumulate until no flow accumulation is possible, effectively eliminating all local depressions. Subtracting the filled DEM from the original DEM resulted in cells with zero values where there were no sinks (closed depressions) and non-zero values where there were closed depressions. In addition to actual sinkholes, roadway intersections with culverts and bridges were also filled (Figure 2). Reclassification of the differenced DEM was then assigned two unique values: one for cells having non-zero values and another for zero values. The reclassified raster was converted to a polygon shapefile. Because the polygons are converted from raster, their boundaries can be blocky and may need smoothing by applying the PAEK (Polynomial Approximation with Exponential Kernel) algorithm in ArcGIS. This process is summarized in the flowchart in Figure 3. The polygons generated in ArcGIS are processed to:

From the closed depressions generated using the “Fill sinks - subtract” tool, 54 non-sinkhole depressions and 41 sinkhole depressions were selected at the training site (Figure 2). They were identified visually on Hillshade maps and aerial photographs. These selected depressions with their corresponding shape factor attributes were used as training sites to produce statistical predictive models. The models can then be used to predict if a given closed depression, generated using ArcGIS, is classified as a sinkhole or a non-sinkhole. The shape factors selected for the study were circularity, sphericity, and curvature. Other research (Doctor and Young, 2013; Klobal et al., 2015; and Wu et al., 2016) has shown the use of area, perimeter, depth, volume, elongatedness, and compactness threshold values as criteria to separate sinkholes from other man-made depressions. The shape factors selected for this work reflect that sinkholes appear to be more circular and half spherical than non-sinkholes. They also appear to have rounded outlines; hence, curvature can be used to quantify roundedness. Circularity is a measure of the closeness of a twodimensional (2D) shape to a circle and is calculated as follows: Circularity =

perimeter2 . 4π × area

(1)

Sphericity was used to measure how close a threedimensional (3D) shape of a closed depression was to a hemisphere (half sphere). Sphericity is the ratio of the surface area of a perfect hemisphere having the same volume as the closed depression to the actual surface area of the closed depression. The volume of a closed depression is determined by multiplying the depth of each cell from the fill height by the cell area. This volume is then used to determine the radius of a hemisphere having the same volume as the closed depression. Using this radius, the surface area of the volumeequivalent hemisphere is calculated. The actual surface area of the closed depression is the number of cells within the depression multiplied by the cell area. Circularity and sphericity of conjoined sinkholes may be difficult to interpret. To separate conjoined sinkholes, the result of the difference values (Filled DEM – Original DEM) can be reclassified in such a way that the filling depth can be modified, separating conjoined closed

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

333

Admassu and Woodruﬀ

Figure 4. Three consecutive non-collinear points deﬁning a triangle for curvature calculation.

depressions into smaller disconnected depressions. Wu et al. (2016) used ranking of closed contours to separately characterize conjoined and isolated sinkholes. The third shape factor, curvature, is a measure of how segments within the outline of a closed depression are diﬀerent from a straight line (zero curvature). The Menger curvature formula (Léger, 1999) approximates the curvature by considering the triangle deﬁned by the three points (xi−1 , yi−1 ), (xi , yi ), and (xi+1 , yi+1 ) as shown in Figure 4. Menger curvature is calculated as follows: 4 s (s − a) (s − b) (s − c) (2) , k (xi , yi ) = a·b·c where, a, b, and c are the lengths of each of the sides of the triangle and s is half of the perimeter of the triangle. The curvature was then scaled by multiplying it by ri , which is the distance from the point (xi , yi ,) to the centroid of the depression (centroid calculated in ArcGIS). This scaling essentially draws a comparison to a circle, which has a curvature of 1. Curvature values mentioned hereafter are all “Scaled” curvature values. The curvature was calculated in R (See Appendix 2 for script) for every point, using it and its two neighbors, leading to 19 curvature values for every closed depression. From these values, the maximum, minimum, median, and mean curvatures for each closed depression were calculated, as well as the percentage of scaled curvature values less than 0.06. To understand the implications of shape factor values, 14 benchmark shapes were generated (Figure 5) and their circularity and curvature values (maximum, minimum, median, mean, and percentage less than 0.06) were calculated (Table 1). Circularity of the circles (shapes 0, 1, 5) was close to one, whereas elongated shapes (2, 8, and 10) resulted in values greater than two (Figure 5, Table 1). Maximum curvature values for circles (shapes 0, 1, 5) were one, whereas shapes with sharp corners (shapes 2, 6, 7, 8, 9, 10, 13) were all greater than four. Shapes with straight edges had

334

Figure 5. Benchmark shapes used to evaluate shape factor values.

minimum scaled curvature values of zero and higher percentages of scaled curvature values less than 0.06. (Figure 5, Table 1). UNIVARIATE STATISTICS OF INDEPENDENT VARIABLES The differences in the distribution of shape factor values (circularity, sphericity, and curvature) at the training site for the two groups of closed depressions, sinkholes and non-sinkholes, were explored as follows. Circularity The histogram and Q-Q plots show non-normal distributions for both sinkholes and non-sinkholes (Figures 6 and 7). The sinkholes have much lower circularity values than non-sinkholes, as expected, but have a much narrower range (Table 2). Circularity values were log transformed to improve normality (Figures 6 and 7). The box plots for both untransformed and transformed data do not overlap. A t-test of the log (circularity) values showed the two populations to have significantly different means (Table 2).

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

Improved Automated Mapping of Sinkholes Table 1. Shape factor values for benchmark shapes. Curvature values are all scaled values. ID

Circularity

Min. Curv.

Max. Curv.

Median Curv.

Mean Curv.

% < 0.06 Curv.

0 1 2 3 4 5 6 7 8 9 10 11 12 13

1.00 1.00 2.31 2.51 1.15 1.00 1.28 1.27 2.20 1.84 2.15 1.35 1.41 5.17

0.86 0.98 0 0.03 0.17 0.86 0 0 0 0 0 0.09 0.25 0.02

1.21 1.01 6.05 8.77 3.25 1.21 5.03 5 8.73 8.16 5.96 2.05 2.48 4.23

1 1.00 0.01 0.21 0.82 1.00 0 0 0.30 0.41 0 1.46 1.39 1.30

1.00 1.00 1.28 1.29 1.13 1.00 1.04 0.97 1.31 1.14 1.23 1.38 1.40 1.46

0 0 68.42 26.32 0 0 63.16 73.68 31.58 26.32 68.42 0 0 5.26

Curv. = curvature; Min. = minimum; Max. = maximum.

Sphericity The distribution for both groups shows normal distribution based on histograms and Q-Q plots (Figures 6 and 7). The distribution of sphericity values has huge overlaps in the box plots, and the diﬀerence in means for both groups is not signiﬁcant (Table 2). Curvature Minimum, maximum, median, and mean curvature values of each depression were calculated, along

with the percentage of curvature values less than 0.06. The distribution of minimum curvature values for both groups is skewed (Figures 6 and 7). Transformation of minimum curvature values using the square root improved the data to have a more normal distribution (Figures 6 and 7). The t-test deemed the means were signiﬁcantly diﬀerent populations, though the box plots show overlap (Table 2). Median curvature values show normal distribution (Figures 6 and 7). The box plots show that sinkholes have a smaller range of median curvature values than those of non-sinkholes and the populations

Figure 6. Q-Q plots and histograms for circularity, log (circularity), sphericity, maximum curvature, minimum curvature, mean curvature, SQRT (minimum curvature), % < 0.06 curvature, and SQRT (% < 0.06 curvature) for sinkholes. Curvature values are all scaled values. An equal number of data points was used for all variables but some parameters show fewer plots because the values were very close and plot as one point.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

335

Admassu and Woodruﬀ Table 2. Univariate statistics of independent variables. Curvature values are all scaled values.

336

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

Improved Automated Mapping of Sinkholes

Figure 7. Q-Q plots and histograms for circularity, log (circularity), sphericity, minimum curvature, SQRT (minimum curvature), mean curvature, SQRT (minimum curvature), % < 0.06 curvature, median curvature, and maximum curvature for non-sinkholes. Curvature values are all scaled An equal number of data points was used for all variables but some parameters show fewer plots because the values were very close and plot as one point value.

overlap with no clear boundary, though the t-test shows significance. Maximum and mean curvature values show normal distribution (Figures 6 and 7) and their box plots do not overlap, which agrees with the t-test results (Table 2). From the benchmark shapes, straight-edged shapes have very low curvature values, and, therefore, the percentage of curvature values less than 0.06 (a good cutoff value from visual inspection of histograms) was recorded for both sinkholes and non-sinkholes. It is expected that non-sinkholes with long straight edges would have more low curvature values than sinkholes, which are more rounded. The distribution of the percentage of curvature values less than 0.06 was skewed and square root transformation improved the distributions to close to normality, especially for non-sinkholes according to Q-Q plots (Figures 6 and 7). The box plots for these values show no overlap, and the t-test shows significance (Table 2). CLASSIFICATION METHODS Multivariate statistics can be used to classify individuals into groups based on quantitative attributes of independent variables that do not show any multicollinearity or, in other words, inter-correlation among variables (Figure 8). In this case, we have two groups of 95 closed depressions representing 41 sinkholes and 54 non-sinkholes. Using the multivariate statistical methods, predictive models based on the shape factors of

the 95 closed depressions from the training site can be put forward. The resulting predictive models ideally can be applied at other sites to ﬁlter out non-sinkholes from actual sinkholes based on their shape factor values. From the univariate distributions based on histograms, Q-Q plots, and box plots, four shape factor values were selected as classiﬁcation variables. The selected independent variables include log (circularity), maximum curvature, mean curvature, and square root (percentage of curvature values less than 0.06), which show non-overlapping ranges of values for the sinkholes and non-sinkholes (Table 2). Predictive statistical models can be used to determine whether a given depression is a sinkhole or non-sinkhole based on shape factor attributes. Discriminant analysis techniques are types of multivariate statistics to classify individuals into one or more groups on the basis of independent variables. Discriminant analysis and logistic regression methods were used to produce predictive models. Two types of discriminant analysis methods, linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA), were used. LDA calculates discriminant scores based on independent variables for each observation to classify a given closed depression into a group. QDA is a variant of LDA that allows nonlinear grouping of data. Both LDA and QDA assume normal distribution of independent variables. Unlike LDA, QDA does not require equal covariance of variables. Logistic regression (LR) gives the probability that an observation is in one of two categories. On

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

337

Admassu and Woodruﬀ

Figure 8. Scatter plot showing no inter-correlation among the independent variables for (a) sinkholes and (b) non-sinkholes.

its own, LR is not a classifier, but the addition of a cutoff probability enables it to be used for classification. Unlike LDA and QDA, LR does not require normal distribution and equal covariance of the independent variables. The R software was used to create these models (James et al., 2013). A YouTube video demonstrating how LDA, QDA, LR predictive models can be generated using a training dataset and using the models to classify new data is available online (Woodruff, 2020).

RESULTS OF TRAINING SITE CLASSIFICATION Univariate analysis showed that log (circularity), maximum curvature, mean curvature, and the square root of the percentage of curvature values less than 0.06 (SQRT % < 0.06) had mostly non-overlapping box plots for sinkhole and non-sinkhole depressions. Because of this, they were selected as good quantitative attributes to classify between sinkholes and nonsinkholes. Using these variables and combinations of them, 15 different models were tested to determine how successfully they classified the 95 closed depressions into sinkholes and non-sinkholes. More specifically, each of the 15 models were defined by a set of independent variables and their discrimination power was investigated using each of the three multivariate methods.

338

Linear Discriminant Analysis The pre-requisites for LDA are normal distribution and existence of equal covariance. Existence of equal covariance among the independent variables was checked using Box’s M test. All four variables except SQRT % < 0.06 showed normal distribution, though SQRT % < 0.06 seemed to be marginally normal, especially for sinkholes (Figure 6 and 7). All four models with single variables showed greater than 65 percent overall success in discriminating between sinkholes and non-sinkholes. Models 1, 2, and 3 predicted sinkholes at >87 percent, whereas Model 4 predicted non-sinkholes at 81 percent. Eleven combinations of the four variables were also evaluated for discriminant power, all of which showed >82 percent overall success rate (Table 3). Equal covariance was tested for all 15 models using Box’s M test, with only Models 2 (maximum curvature), 4 (SQRT % < 0.06), and 9 (maximum curvature and SQRT % < 0.06) not violating the equal covariance rule (Table 3). Quadratic Discriminant Analysis QDA was attempted on the same 15 models used for LDA. The advantage of QDA is that equal covariance of variables is not required, although normal distribution of variables is still necessary. Among the ﬁrst four models, the results showed an 89 percent success rate for log (circularity), 83 percent for maximum

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

Table 3. Summary of predictive models and results from the training site. Curvature values are all scaled values. Linear Discriminant Analysis

Model

5 6 7

8 9

Log (Circularity) Max. Curvature Mean Curvature Square Root (% Curvatures < 0.06) Log(Circularity) Max. Curvature Log(Circularity) Mean Curvature Log(Circularity) Square Root (% Curvatures < 0.06) Max. Curvature Mean Curvature Max. Curvature Square Root (% Curvatures < 0.06) Mean Curvature Square Root (% Curvatures < 0.06) Log(Circularity) Max. Curvature Mean Curvature Log(Circularity) Max. Curvature Square Root (% Curvatures < 0.06) Log(Circularity) Mean Curvature Square Root (% Curvatures < 0.06) Max. Curvature Mean Curvature Square Root (% Curvatures < 0.06) Log(Circularity) Max. Curvature Mean Curvature Square Root (% Curvatures < 0.06)

% Correctly Predicted as Nonsinkholes

% Correctly Predicted as Sinkholes

Overall % Correctly Predicted

75.9 79.6 68.5 81.5

95.1 87.8 90.2 43.9

84.2 83.2 77.9 65.3

8E-14 0.06 1E-08 0.06

79.6

97.6

87.4

75.9

95.1

81.5

% Correctly Predicted as Sinkholes

Overall Percent Correctly Predicted

% Correctly Predicted as Nonsinkholes

% Correctly Predicted as Sinkholes

Overall % Correctly Predicted

85.2 79.6 70.4 59.3

95.1 87.8 90.2 90.2

89.5 83.2 79.0 72.6

92.6 83.3 83.3 81.5

92.7 80.5 85.4 43.9

92.6 82.1 84.2 65.3

2E-13

87.0

92.7

89.5

92.6

92.7

92.6

84.2

2E-22

81.5

90.2

85.3

92.6

92.7

92.6

97.6

88.4

5E-13

87.0

95.1

90.5

92.6

92.7

92.6

83.3

92.7

87.4

2E-08

83.3

85.4

84.2

90.7

82.9

87.4

81.5

87.8

84.2

0.01

85.2

87.8

86.3

87.0

75.6

82.1

75.9

90.2

82.1

8E-08

79.6

90.2

84.2

83.3

82.9

83.2

79.6

97.6

87.4

4E-21

79.6

90.2

84.2

92.6

92.7

92.6

79.6

95.1

86.3

5E-13

87.0

92.7

89.5

90.7

90.2

90.5

79.6

97.6

87.4

5E-21

83.3

90.2

86.3

92.6

92.7

92.6

83.3

95.1

88.4

5E-08

83.3

85.4

84.2

88.9

82.9

86.3

83.3

100.0

90.5

3E-20

83.3

90.2

86.3

90.7

92.7

91.6

339

Max. = maximum; Sig. = signiﬁcance.

% Correctly Sig. (Box Predicted as Test of Equal NonCovariance) sinkholes

Logistic Regression

Improved Automated Mapping of Sinkholes

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

1 2 3 4

Independent Variables

Quadratic Discriminant Analysis

Admassu and Woodruﬀ Table 4. Result of predictive models on sites from Boyce Quad, VA. Curvature values are all scaled values. Model No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Variables

LDA

QDA

Average Success

Log(Circularity) Max. Curvature Mean Curvature SQRT%<0.06 Curvature Log(Circularity) and Max. Curvature Log(Circularity) and Mean Curvature Log(Circularity) and SQRT%<0.06 Curvature Max. Curvature and Mean Curvature Max. Curvature and SQRT%<0.06 Curvature Mean Curvature and SQRT%<0.06 Curvature Log(Circularity), Max. Curvature, and Mean Curvature Log(Circularity), Max. Curvature, and SQRT%<0.06 Curvature Log(Circularity), Mean Curvature, and SQRT%<0.06 Curvature Max. Curvature, Mean Curvature, and SQRT%<0.06 Curvature Log(Circularity), Max. Curvature, Mean Curvature, and SQRT%<0.06 Curvature

69.4 77.8 58.3 50.0 72.2 63.9 77.8 72.2 77.8 58.3 75.0 72.2 66.7 75.0 72.2

66.7 77.8 55.6 88.9 69.4 58.3 66.7 55.6 77.8 55.6 58.3 66.7 58.3 55.6 58.3

61.1 72.2 44.4 50.0 63.9 61.1 63.9 55.6 75.0 41.7 63.9 66.7 63.9 63.9 66.7

65.7 75.9 52.8 63.0 68.5 61.1 69.5 61.1 76.9 51.9 65.7 68.5 63.0 64.8 65.7

LDA = linear discriminant analysis; QDA = quadratic discriminant analysis; LR = logistic regression; SQRT = square root.

curvature, 79 percent for mean curvature, and 73 percent for SQRT % < 0.06 (Table 3). As stated previously, SQRT % < 0.06 distribution was only near normal, which may cause models containing it to be unstable. Combinations of more than one variable were then tested. All models resulted in >84 percent overall discrimination success, with Model 7 (log (circularity) and SQRT % < 0.06) resulting in overall discrimination success as high as 90 percent (Table 3).

an average overall discrimination power of 76 percent when averaging over the three predictive methods (Table 4). Among the combined variables, Model 9 (maximum curvature and SQRT % < 0.06) resulted in the

Logistic Regression The advantage of logistic regression is that it is unaffected by non-normal distribution and unequal covariance of the independent variables. Almost all of the 15 models showed an overall success rate of >82 percent, with Model 4 (SQRT % < 0.06) having the lowest at 66percent (Table 3). RESULTS OF TEST SITE CLASSIFICATION–BOYCE QUADRANGLE, VA The 15 predictive models were tested on fieldchecked sinkholes in the Boyce quadrangle located in northwestern Virginia. Boyce Quadrangle is a U.S. Geological Survey (USGS) geological mapping area of a 7.5 minute × 7.5 minute quadrangle located in northwestern Virginia (Figure 9). A 1 m × 1 m LiDAR imagery was processed to identify closed depressions, of which 35 were field checked by the USGS. Circularity, sphericity, and curvature values were calculated. The 15 predictive models were applied to test if they would correctly be classified as sinkholes. Among the first four models (Models 1–4) that used individual variables, maximum curvature (Model 2) showed

340

Figure 9. Location map of Boyce Quadrangle.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

Improved Automated Mapping of Sinkholes

highest average overall discrimination power of 77 percent (Table 4). SUMMARY AND DISCUSSION Although it is a relatively simple workflow to identify closed depressions using the “fill sinks and subtract” method, the output combines sinkholes and other artificial depressions, making it difficult to separate them. Doctor and Young (2013) and Maidment (2002) introduced a method by which flow obstructions, such as bridges and roadways, on fill material can be manipulated by lowering their elevation so that artificial closed depressions will not form. Other artificial depressions such as reservoirs, quarries, and ponds may not be filtered out using this method. Previous research (Doctor and Young, 2013; Klobal et al., 2015; and Wu et al., 2016) has shown the use of shape factors (circularity, elongatedness, compactedness, perimeter, etc.) as discrimination tools to differentiate sinkholes from other artificial closed depressions. These studies mainly use shape factors as threshold values to better characterize closed depression sinkholes and to filter out other closed depressions. Gochenour and Admassu (2017) evaluated the use of discriminant analysis with circularity and eccentricity (elongatedness) as discrimination factors to identify sinkholes. This work is an attempt to use statistical methods to evaluate shape factors used in previous research (circularity) and new shape factors (sphericity and curvature) to filter out artificial closed depressions resulting from the “Fill sinks - subtract” method. In this research, circularity, sphericity, and various measures of curvature were evaluated for their potential to discriminate closed depressions. Log (circularity), maximum curvature, mean curvature, and SQRT % < 0.06 were selected as promising discriminating attributes because their box plots for sinkholes and nonsinkholes were distinctly non-overlapping. In general based on the training site data from the Harrisonburg area, all three discriminant methods (LDA, QDA, and LR) showed relatively high success rates in their prediction percentages. In terms of individual variables, log (circularity), maximum curvature, and mean curvature had higher average success rates in predicting sinkholes, whereas SQRT % < 0.06 showed better results for non-sinkholes (Table 3). Applying the predictive models to the sinkholes from the test site, Boyce Quadrangle, maximum curvature showed a high percentage of predicting success (76 percent). It should also be noted that in the case of the conjoined sinkholes in the Boyce Quad, the models that most often accurately categorized them were models containing curvature variables, especially maximum curvature and SQRT % <0.06.

Investigating the physical meaning of the results, circularity is a good descriptor of compactness of a closed depression where most sinkholes have compact shapes. In cases where conjoined sinkholes exist, circularity values may not be effective. Because sinkholes have distinct rounded shapes as compared to nonsinkholes, curvature values describing roundedness of segments were found to be useful discriminating variables. The straight boundaries of non-sinkholes and their sharp corners result from depressions alongside roads and bridges. As shown by the benchmark shapes, rounded corners resulted in lower curvature values than those having sharp corners. Therefore, the maximum curvature values for non-sinkholes with sharper corners will be high. Also, non-sinkholes commonly having straight edges will have very low curvature values; hence, more curvature values less than 0.06 were reflected in SQRT % < 0.06. The choice of variables used in the analyses was based on certain requirements, such as nonoverlapping values between the two populations. Other shape factor values, sphericity, median curvature, and minimum curvature, may prove to be important if one uses another training dataset to produce predictive models. Therefore, it should be noted that the suggested predictive models may not be universally applicable. This is corroborated by the lower success rates of the models for the test site at Boyce Quadrangle. The recommended practice is to select known sinkholes and non-sinkholes at a training site of interest and generate predictive models to be used on other depressions in a geologically similar area. It should be emphasized that the use of predictive models will streamline the result of automated sinkhole mapping. It will reduce the number of artificial sinks that are not sinkholes, but at the same time fail to correctly identify some actual sinkholes. It will not be capable of a 100 percent success rate as there will be shapes that have ambiguous shape factor values. There could be cases where some depressions were originally sinkholes and later modified by human activity such as routing a road across a sinkhole. There will always be the need for field and/or aerial photo verification. Despite some of these shortcomings of mapping sinkholes through automated methods, it is still timesaving to use automated techniques. Therefore, to apply LDA, QDA, and LR to discriminate closed depressions, one has to 1) generate polygons bounding ArcGIS–identified closed depressions, 2) calculate sphericity and circularity, (x, y) points along boundaries using ArcGIS following the methods explained, and 3) Using R, calculate curvature and generate LDA, QDA, and LR models. A short YouTube video (Woodruff, 2020) was prepared to show 1) how to calculate curvature values from

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

341

Admassu and Woodruﬀ

x, y points of boundaries of depressions and 2) how to generate LR, QDA, and LR predictive models. Appendix 2 provides the R script that generates scaled curvature values for region data. Appendix 3 provides the R script that generates LDA, QDA, and LR– based predictive models and allows users to run the predictive models on a given set of data. Both Appendixes 2 and 3 are downloadable from https://drive. google.com/drive/folders/1TR5qVmA29VZSshwOm B4dzigMHGpIlq95?usp=sharing. CONCLUSION From the results of this research, it can be concluded that: 1. Automated mapping of sinkholes from highresolution airborne LiDAR-derived DEMs using ArcGIS is a simple and time efficient technique to identify sinkholes. 2. Multivariate statistics is an excellent tool to filter out closed depressions that are not sinkholes. 3. Discriminant functions and logistic regression methods based on shape factors can yield success rates as high as > 80 percent, which is significant improvement to the ArcGIS-based automated mapping of sinkholes. Therefore, the method reduces a large number of false positives by allowing a practitioner to deal with much fewer depressions. 4. Curvature values are the most powerful shape factor to separate sinkholes from non-sinkholes. ACKNOWLEDGMENT The authors would like to thank Dr. Daniel Doctor of the USGS for providing field checked map of sinkholes from Boyce Quadrangle, Virginia. REFERENCES De Carvalho, O.; Guimarães, R.; Montgomery, D.; Gillespie A.; and Trancoso, R., 2013, Karst depression detection using ASTER, ALOS/PRISM and SRTM-derived digital elevation models in the Bambuí Group, Brazil: Remote Sensing, Vol. 6, pp. 330–351.

342

Doctor, D. and Doctor, K., 2012, Spatial analysis of geologic and hydrologic features relating to sinkhole occurrence in Jefferson County, West Virginia: Carbonates Evaporites, Vol. 27, pp. 143–152. Doctor, D.; Weary, D.; Ordorff, R.; Harlow, G.; Kozar M.; and Nelms, D., 2008, Bedrock structural controls on the occurrence of sinkholes and springs in northern great valley karst, Virginia and West Virginia. In Yuhr L.; Alexander, E.; and Beck, C. (Editors), Sinkholes and the Engineering and Environmental Impacts of Karst: Proceedings of the Eleventh Multidisciplinary Conference: Geotechnical Special Publication No. 183, American Society of Civil Engineers, Tallahassee, Florida, pp. 243–256. Doctor, D. and Young, A., 2013, An evaluation of automated GIS tools for delineating karst sinkholes and closed depressions from 1-meter lidar-derived digital elevation data. In Lewis Land, Daniel Doctor, J. Brad Stephenson (Eds.), Proceedings of the Thirteenth Sinkhole Conference, NCKRI Symposium 2: National Cave and Karst Research Institute, Carlsbad, New Mexico, pp. 449–458. Gochenour, J. and Admassu, Y., 2017, Applying discriminant analysis towards automated sinkhole mapping methods: Geological Society America Abstracts Programs, Vol. 49, No. 6, ISSN 0016-7592, doi: 10.1130/abs/2017AM-304653. Hofierka, T.; Gallay, M.; Bandurab, P.; and Sasak, J., 2018, Identification of karst sinkholes in a forested karst landscape using airborne laser scanning data and water flow analysis: Geomorphology, Vol. 308, pp. 265–277. James, G.; Witten, D.; Hastie, T.; and Tibshirani, R., 2013, An Introduction to Statistical Learning with Applications in R: Springer, New York 434 p. Jenness, J.; Brost, B.; and Beir, P., 2011, Land Facet Corridor Designer: Extension for GIS: Electronic document, available at http://www.jennessent.com Kobal, K.; Bertoncelj, I.; Pirotti, F.; Dakskobler, I.; and Kutnar, L., 2015, Using lidar data to analyse sinkhole characteristics relevant for understory vegetation under forest cover—case study of a high karst area in the Dinaric Mountains: PLoS ONE, Vol. 10, No. 3, DOI: 10.1371/journal. pone.0122070. Léger, J., 1999, Menger curvature and rectifiability: Annals Mathematics, Vol. 149, pp. 831–869. Maidment, D., 2002, Arc Hydro: GIS for Water Resources, Vol. 1: Esri Press, Redlands, CA, 201 p. Woodruff, C., 2020, How to use the sinkhole identification R code, Youtube Video, https://www.youtube.com/watch?v= u757aea_Yc8&feature=youtu.be Wu, Q.; Deng, C.; and Chen, Z., 2016, Automated delineation of karst sinkholes from LiDAR-derived digital elevation models: Geomorphology, Vol. 266, pp. 1–10.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

Improved Automated Mapping of Sinkholes Appendix 1: Shape Factor Values of Sinkholes and Non-Sinkholes.

FID 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Sinkholes (1) Not Sinkholes (0) 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1

Circularity

Log (Circularity)

1.80 1.46 1.39 1.61 1.33 1.46 1.12 2.33 5.35 1.86 2.27 2.43 2.20 1.75 2.05 2.38 1.52 1.56 2.59 2.87 1.79 2.75 3.14 2.41 1.88 5.19 8.04 3.52 2.68 1.91 1.71 2.07 1.48 6.42 1.83 1.64 1.59 1.87 2.05 2.16 1.38 1.70 1.53 2.60 2.92 3.35 3.04 3.14 2.16 2.05 3.06 1.89 2.98 1.44 1.23 1.32 1.31 1.16 1.34 1.16

0.26 0.16 0.14 0.21 0.12 0.16 0.05 0.37 0.73 0.27 0.36 0.39 0.34 0.24 0.31 0.38 0.18 0.19 0.41 0.46 0.25 0.44 0.50 0.38 0.28 0.72 0.91 0.55 0.43 0.28 0.23 0.32 0.17 0.81 0.26 0.21 0.20 0.27 0.31 0.33 0.14 0.23 0.18 0.41 0.47 0.53 0.48 0.50 0.33 0.31 0.49 0.28 0.47 0.16 0.09 0.12 0.12 0.06 0.13 0.07

Sphericity

Min. Curv.

SQRT Min. Curv.

Max. Curv.

Median Curv.

Mean Curv.

% Curv. < 0.06

SQRT % Curv. < 0.06

0.16 0.26 0.29 0.19 0.29 0.18 0.15 0.19 0.22 0.31 0.19 0.17 0.08 0.07 0.24 0.22 0.29 0.29 0.26 0.14 0.11 0.23 0.15 0.17 0.18 0.21 0.16 0.17 0.19 0.15 0.29 0.22 0.40 0.18 0.12 0.20 0.29 0.10 0.09 0.10 0.22 0.09 0.25 0.11 0.08 0.04 0.13 0.13 0.11 0.15 0.24 0.25 0.19 0.32 0.20 0.09 0.13 0.34 0.17 0.17

0.00 0.05 0.02 0.03 0.05 0.10 0.00 0.09 0.02 0.00 0.00 0.05 0.05 0.04 0.00 0.00 0.01 0.00 0.00 0.00 0.16 0.09 0.04 0.00 0.02 0.03 0.02 0.01 0.04 0.06 0.01 0.11 0.00 0.07 0.02 0.00 0.04 0.02 0.07 0.03 0.03 0.08 0.00 0.03 0.03 0.02 0.03 0.01 0.06 0.03 0.00 0.02 0.03 0.01 0.03 0.03 0.02 0.00 0.02 0.05

0.06 0.22 0.13 0.18 0.21 0.32 0.04 0.30 0.12 0.00 0.01 0.22 0.22 0.21 0.00 0.00 0.07 0.00 0.00 0.03 0.40 0.30 0.21 0.03 0.14 0.18 0.13 0.07 0.21 0.24 0.07 0.32 0.02 0.26 0.14 0.00 0.19 0.16 0.26 0.16 0.17 0.29 0.00 0.16 0.16 0.13 0.17 0.08 0.25 0.17 0.00 0.14 0.19 0.10 0.17 0.16 0.15 0.06 0.15 0.21

8.56 4.29 3.39 5.18 6.37 6.70 3.04 8.64 7.17 5.53 8.09 11.99 9.76 5.76 6.96 7.15 6.04 6.41 7.38 8.79 9.00 6.10 7.70 9.77 8.03 8.91 5.54 6.66 5.02 7.70 7.24 7.94 5.42 9.81 5.73 5.18 5.10 10.02 7.14 4.42 6.40 6.97 8.24 8.69 7.58 7.47 8.03 8.10 6.05 10.31 8.43 5.15 8.80 5.61 3.97 2.81 6.07 5.10 5.34 3.55

1.01 0.42 0.80 0.61 0.69 0.39 0.99 0.39 0.27 0.61 0.23 1.07 0.93 0.70 0.88 0.15 0.59 0.52 0.17 0.11 0.63 0.49 0.47 0.16 0.59 0.37 0.56 0.43 0.81 1.08 0.47 0.48 0.65 0.35 1.40 0.21 1.20 0.32 0.54 1.03 0.76 0.47 0.59 0.66 0.61 1.34 1.07 1.18 0.95 0.37 0.15 1.76 0.97 0.50 0.78 0.80 0.64 0.68 0.75 0.79

1.74 1.18 1.25 1.33 1.22 1.21 1.09 1.40 1.22 1.23 1.45 2.11 1.86 1.35 1.36 1.46 1.36 1.36 1.44 1.44 1.34 1.43 1.13 1.32 1.23 1.68 1.37 1.91 1.41 1.72 1.32 1.79 1.31 1.72 1.65 1.26 1.52 2.03 1.74 1.38 1.30 1.29 1.12 2.21 1.55 2.10 1.91 1.68 1.80 1.42 1.40 1.64 1.74 1.26 1.21 1.19 1.20 1.10 1.25 1.11

5.26 5.26 10.53 10.53 10.53 0.00 15.79 0.00 10.53 11.11 26.32 5.26 5.26 15.79 26.32 31.58 10.53 15.79 36.84 42.11 0.00 0.00 10.53 36.84 5.26 21.05 21.05 10.53 5.26 5.26 21.05 0.00 5.26 0.00 10.53 15.79 5.26 5.26 0.00 5.56 10.53 0.00 15.79 5.26 15.79 10.53 15.79 21.05 0.00 15.79 36.84 5.26 5.26 10.53 5.26 5.26 5.26 10.53 5.26 5.26

2.29 2.29 3.24 3.24 3.24 0.00 3.97 0.00 3.24 3.33 5.13 2.29 2.29 3.97 5.13 5.62 3.24 3.97 6.07 6.49 0.00 0.00 3.24 6.07 2.29 4.59 4.59 3.24 2.29 2.29 4.59 0.00 2.29 0.00 3.24 3.97 2.29 2.29 0.00 2.36 3.24 0.00 3.97 2.29 3.97 3.24 3.97 4.59 0.00 3.97 6.07 2.29 2.29 3.24 2.29 2.29 2.29 3.24 2.29 2.29

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

343

Admassu and Woodruﬀ Appendix 1: Continued.

FID 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95

Sinkholes (1) Not Sinkholes (0) 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

Circularity

Log (Circularity)

1.06 1.20 1.18 1.29 1.70 1.16 1.15 1.12 1.13 1.20 1.15 1.12 1.25 1.26 1.13 1.27 1.66 1.22 1.12 1.24 1.24 1.07 1.06 1.49 1.26 1.10 1.13 1.08 1.15 1.07 1.27 1.40 1.37 1.20 1.08

0.02 0.08 0.07 0.11 0.23 0.07 0.06 0.05 0.05 0.08 0.06 0.05 0.10 0.10 0.05 0.10 0.22 0.09 0.05 0.09 0.09 0.03 0.02 0.17 0.10 0.04 0.05 0.03 0.06 0.03 0.10 0.15 0.14 0.08 0.03

Sphericity

Min. Curv.

SQRT Min. Curv.

Max. Curv.

Median Curv.

Mean Curv.

% Curv. < 0.06

SQRT % Curv. < 0.06

0.34 0.31 0.22 0.24 0.19 0.51 0.15 0.26 0.60 0.28 0.31 0.16 0.38 0.12 0.32 0.23 0.25 0.29 0.30 0.20 0.15 0.21 0.20 0.13 0.10 0.22 0.13 0.26 0.23 0.25 0.13 0.25 0.11 0.09 0.28

0.17 0.08 0.05 0.07 0.05 0.00 0.00 0.13 0.11 0.01 0.03 0.27 0.07 0.06 0.01 0.04 0.10 0.01 0.10 0.05 0.20 0.08 0.43 0.03 0.02 0.11 0.00 0.09 0.03 0.08 0.14 0.01 0.04 0.01 0.07

0.42 0.29 0.22 0.27 0.21 0.00 0.00 0.37 0.33 0.08 0.18 0.52 0.27 0.25 0.10 0.21 0.32 0.10 0.32 0.22 0.44 0.28 0.66 0.16 0.16 0.33 0.00 0.30 0.17 0.28 0.38 0.11 0.19 0.12 0.26

1.91 4.84 2.24 4.53 3.92 7.08 3.11 2.27 2.50 2.40 5.94 4.00 2.50 4.91 4.99 4.66 6.64 3.11 3.87 6.69 4.41 2.91 2.20 5.23 5.26 3.78 4.53 3.05 3.71 2.42 3.92 5.48 3.18 2.83 3.23

1.00 1.13 0.98 0.93 1.12 0.91 1.01 0.95 1.06 0.98 0.79 0.77 1.00 0.95 0.90 0.93 0.76 1.14 0.87 0.75 0.75 0.93 1.00 0.75 0.84 0.91 0.91 0.91 0.88 0.79 0.89 0.72 0.64 0.80 0.98

1.07 1.20 1.11 1.27 1.41 1.25 1.10 1.08 1.08 1.14 1.08 1.08 1.06 1.44 1.16 1.18 1.37 1.23 1.20 1.22 1.17 0.99 1.03 1.04 1.35 1.11 1.05 1.09 1.10 0.98 1.21 1.31 1.15 1.12 1.07

0.00 0.00 5.26 0.00 5.26 10.53 5.26 0.00 0.00 5.26 10.53 0.00 0.00 0.00 5.26 5.26 0.00 5.26 0.00 5.26 0.00 0.00 0.00 10.53 5.26 0.00 15.79 0.00 5.26 0.00 0.00 5.26 5.26 5.26 0.00

0.00 0.00 2.29 0.00 2.29 3.24 2.29 0.00 0.00 2.29 3.24 0.00 0.00 0.00 2.29 2.29 0.00 2.29 0.00 2.29 0.00 0.00 0.00 3.24 2.29 0.00 3.97 0.00 2.29 0.00 0.00 2.29 2.29 2.29 0.00

Curvature values are all scaled values. Min. = minimum; Max. = maximum; Curv. = curvature; SQRT = square root.

APPENDIX 2 R Script to Calculate Curvature Usage notes This code assumes that you have two spreadsheets. The ﬁrst has the (x,y) coordinates of the centroids for each of the regions. Ex:

FID

0 1 2

x0 x1 x2

y0 y1 y2

The second has the (x,y) coordinates of each of the boundary points for each of the regions. Ex:

344

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

Improved Automated Mapping of Sinkholes

FID

0 0 .. . 1 1 .. .

x00 x01 .. . x10 x11 .. .

y00 y01 .. . y10 y11 .. .

The regions can be in any order, but the points in each region need to be in order around the boundary, so that they trace out the boundary. In both ﬁles, the code assumes that the column heading for the regions is FID. Most of what is contained in the following code can be left as is. The things that you deﬁnitely want to change are the following: 1. 2. 3. 4. 5.

The name of the working directory The name of the file containing the centroid data The name of the file containing the boundary point data The different curvature statistics that you want to compute The name of the file containing the output

Regarding item 4 above, the code is already pre-loaded with the same curvature statistics that we used in the paper. Be sure to change the variable num_stats if you want more (or fewer) than five different statistics. Code #————————————————————— # Program: scaled_curvature.r # Author: Celes Woodruff # Date: 5/22/20 # Purpose: Calculate the scaled curvatures of the points on the border of many different # regions. Boundary and centroid data for each region is imported. For each region, # the Menger curvature formula is used to approximate the curvature at each point, # which is then scaled by the distance between that point and the centroid. # # NOTE: This code assumes that the points in each region are in order around the # boundary, i.e., if you connected the points in order, you would trace the # boundary. Also, all of the points for a particular region should be together, # but the regions themselves can be in any order. #————————————————————— #—————— # set up workspace #—————— # clear away any variables that are lingering around rm(list = ls()) # set working directory to correct folder setwd(‘∼/Documents’) #————————————– # curvature function to be used later #————————————– #————————————————————– # Function: curve_3pt # Author: Celes Woodruff # Purpose: This function takes in an array of 3 points as # input and estimates the curvature of the curve # that connects them using the Menger curvature

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

345

Admassu and Woodruﬀ

# formula. # Variables: # pts = an array of 3 points, with x-values in the first column # and y-values in the second column # distances = Euclidean distances between each pair of points # a, b, c = distances between pairs of points (separated for readability) # s = semi-perimeter of the triangle connecting the points # A = area of the triangle connecting the points # kappa = approximated curvature curve_3pt <- function(pts){ # calculate the distance between each pair of points distances = dist(pts,method = “euclidean”) a = distances[1] b = distances[2] c = distances[3] # calculate the area of the triangle using Heron’s forumla s = 0.5*(a + b + c) A = sqrt(abs(s*(s-a)*(s-b)*(s-c))) # calculate curvature using Menger’s curvature formula curve_3pt = 4*A/(a*b*c) } #————————————————————– #——————– # BEGIN MAIN PROGRAM #——————– # Variables: # N = total number of regions # centroids = data frame containing x and y values of the centroid in each region (dim: N X 3) # num_ppr = number of points per region # region_pts = data frame containing all of the boundary points in all of the regions (dim: (N*num_ppr) X 3) # curvatures = array holding the curvatures values of every boundary point in each region (dim: N X num_ppr) # region = data frame of boundary points for the current region (dim: num_ppr X 2) # m = array containing current boundary point and its neighbors (dim: 3 X 2) # radius = distance from current point to the centroid # num_stats = number of statistics you want to calculate # curve_stats = array containing the curvature statistics (dim: N X (num_stats + 2)) # curvature_output = data frame made from curve_stats variable #———————————— # import curvature and boundary data #———————————— # import centroid data centroids <- read.csv(“centroids.csv”,header = TRUE) N <- dim(centroids)[1] # number of regions # import boundary point data # NOTE 1: code assumes that the first column is called FID # NOTE 2: the names of the columns containing the x and y values should match the centroid file region_pts <- read.csv(“points.csv”,header = TRUE) #————————– # calculate the curvatures #————————– # determine number of points per region (assumed to be same for all) num_ppr = dim(region_pts)[1]/N # create array to hold curvature values for each boundary point in each region curvatures <- array(0,dim = c(N,num_ppr)) # initialize with zeros 346

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

Improved Automated Mapping of Sinkholes

# go to each region to calculate curvatures for (k in 1:N){ # pick the boundary points for the next region in the centroid list region <- region_pts[region_pts$FID = = centroids$FID[k],2:3] # first point uses last point as “left” point m <- rbind(region[num_ppr,],region[1,],region[2,]) # calculate distance b/w first point and centroid radius <- dist(rbind(region[1,],centroids[k,2:3])) # calculate curvature and scale by radius curvatures[k,1] <- curve_3pt(m)*radius # middle points for (j in 2:(num_ppr-1)){ m <- rbind(region[j-1,],region[j,],region[j+1,]) # calculate distance b/w jth point and centroid radius <- dist(rbind(region[j,],centroids[k,2:3])) # calculate curvature and scale by radius curvatures[k,j] <- curve_3pt(m)*radius } # last point uses first point as “right” point m <- rbind(region[num_ppr-1,],region[num_ppr,],region[1,]) # calculate distance b/w last point and centroid radius <- dist(rbind(region[num_ppr,],centroids[k,2:3])) # calculate curvature and scale by radius curvatures[k,num_ppr] <- curve_3pt(m)*radius } #—————————— # generate curvature statistics #—————————— # generate whichever statistics on the scaled curvatures that you want num_stats <- 5 # number of statistics you want to calculate curve_stats <- array(0,dim = c(N,num_stats+2)) # initialize with zeros curve_stats[,1] <- centroids[,1] # region numbers in first column curve_stats[,2] <- num_ppr # number of points per region in second column # generate statistics on each region and store in array for (k in 1:N){ # minimum scaled curvature curve_stats[k,3] <- min(curvatures[k,]) # maximum scaled curvature curve_stats[k,4] <- max(curvatures[k,]) # median scaled curvature curve_stats[k,5] <- median(curvatures[k,]) # mean scaled curvature curve_stats[k,6] <- mean(curvatures[k,]) # percentage of scaled curvatures less than 0.06 curve_stats[k,7] <- length(which(curvatures[k,] < 0.06))/num_ppr*100 } # for nice looking output, change the array to a data frame and provide column headers curvature_output <- data.frame(curve_stats) # name columns names(curvature_output)[1] <- “FID” names(curvature_output)[2] <- “# of points” names(curvature_output)[3] <- “Min Curv” names(curvature_output)[4] <- “Max Curv” names(curvature_output)[5] <- “Median Curv” names(curvature_output)[6] <- “Mean Curv” Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

347

Admassu and Woodruﬀ

names(curvature_output)[7] <- “% Curv < 0.06” # write output to a file write.csv(curvature_output,“curvature_results.csv”,row.names = FALSE) APPENDIX 3 R Script to Generate LDA-, QDA-, and LR-Based Predictive Models Usage notes This code assumes that you have two spreadsheets. The first spreadsheet contains the variables for the training data, and the second contains the variables for the new data to be classified. It also assumes that the training data has an extra variable called Sink_NotSink that is a 0 if the region is not a sinkhole and 1 if it is. The variable names are the same as the column headings. Example training data spreadsheet columns: FID

Sink_NotSink

log_Circularity

Min.Curv.

...

Example new data spreadsheet columns: FID

log_Circularity

Min.Curv.

...

Most of what is in the code can be left as is. The things that you deﬁnitely want to change are the following: 1. 2. 3. 4. 5. 6.

The name of the working directory The names of the files containing the training data and the new region data The number of models that you want to run The formulas for the different models that you want to run The cutoff value for logistic regression (default is 0.5) The name of the files containing the output data

The code comes pre-loaded with the 15 models that we use in our paper. The formulas contain the names of the variables that you want to include in the model. If you are using more than one variable in a model, separate them with plus signs. The code runs each of the models three times, once for each classification method. The logistic regression version of the model is created using the glm() function, the linear discriminant analysis version is created using the lda() function, and the quadratic discriminant analysis version is created using the qda() function. After the model is created, the predict() function is used to classify the data. If you do not want to use all three classification methods, you can simply comment out undesired versions. Special note: The predict() function for the LDA and QDA methods calculates the discriminant score for both classes, then classifies the region depending on which value is higher, as described in the paper. If the values are sufficiently close, it cannot tell the difference between the scores and randomly picks one of the classes. This could lead to a different classification of a region if run more than once. If you want to be notified of this possibility, you can add a flag into your code when the scores differ by less than a specified tolerance. The code provided here does not include such a flag. Code #————————————————————— # Program: sinkhole_models.r # Author: Celes Woodruff # Date: 5/23/20

348

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

Improved Automated Mapping of Sinkholes

# Purpose: Use variables from different regions to classify each as being a sinkhole or not. # This code creates models based on training data that you provide and uses those # models to classify both the training data and new data. Each model has 3 versions: # one for logistic regression, one for linear discriminant analysis (LDA), and one # for quadratic discriminant analysis (QDA). # # The results are summarized in 3 arrays (one for each version) for the training # data and 3 arrays (one for each version) for the new data. The tables have 1 for a # sinkhole and 0 for a non-sinkhole. # # Note: The predict() function for LDA and QDA calculates discriminant scores for # both classes, then classifies the region depending on which value is higher. # If the values are sufficiently close, it can’t tell the difference between # the values and randomly picks on of the classes. These values are stored in # the posterior variable of the predict() function output. # #————————————————————— #—————— # set up workspace #—————— # clear away any variables that are lingering around rm(list = ls()) # set working directory to correct folder setwd(‘∼/Documents’) # use mass library to get LDA function library(MASS) # Variables: # train_data = data frame containing the variables from the training regions # N1 = number of regions in the training data # new_regions = data frame containing the variables from the regions you want to classify # N2 = number of regions in the new data # M = number of variables that you want to create # vars = vector containing strings defining the models that you want to create # LR_train_probs = array containing the logistic regression probabilities from the models # for the training data # LR_train_class = array containing the logistic regression classifications from the models # for the training data # LDA_train_class = array containing the linear discriminant analysis classifications from the # models for the training data # QDA_train_class = array containing the quadratic discriminant analysis classifications from # the models for the training data # LR_new_probs = array containing the logistic regression probabilities from the models for the # new data # LR_new_class = array containing the logistic regression classifications from the models for # the new data # LDA_new_class = array containing the linear discriminant analysis classifications from the # models for the new data # QDA_new_class = array containing the quadratic discriminant analysis classifications from # the models for the new data # fo1 = model formula for logistic regression models # fo2 = model formula for linear and quadratic discriminant analysis models # cutoff = cutoff probability for classifying logistic regression models #—————————————————— # import training data and new region data to classify #——————————————————

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

349

Admassu and Woodruﬀ

# import training data set train_data <- read.csv(“training_data_set.csv”,header = TRUE) N1 <- dim(train_data)[1] # number of regions # import new region data to classify new_regions <- read.csv(“unclassified_data_set.csv”, header = TRUE) N2 <- dim(new_regions)[1] # number of regions #———————– # create model formulas #———————– M <- 15 # number of models to run # create vector to hold the model formulas as character strings # use plus sign if more than one variable in the model vars <- array(dim = M) vars[1] <- “log_Circularity” vars[2] <- “Max.Curv.” vars[3] <- “Mean.Curv.” vars[4] <- “sqrt_perc_curv_lt06” vars[5] <- “log_Circularity + Max.Curv.” vars[6] <- “log_Circularity + Mean.Curv.” vars[7] <- “log_Circularity + sqrt_perc_curv_lt06” vars[8] <- “Max.Curv. + Mean.Curv.” vars[9] <- “Max.Curv. + sqrt_perc_curv_lt06” vars[10] <- “Mean.Curv. + sqrt_perc_curv_lt06” vars[11] <- “log_Circularity + Max.Curv. + Mean.Curv.” vars[12] <- “log_Circularity + Max.Curv. + sqrt_perc_curv_lt06” vars[13] <- “log_Circularity + Mean.Curv. + sqrt_perc_curv_lt06” vars[14] <- “Max.Curv. + Mean.Curv. + sqrt_perc_curv_lt06” vars[15] <- “log_Circularity + Max.Curv. + Mean.Curv. + sqrt_perc_curv_lt06” #—————————— # create arrays to store output #—————————— # create arrays to store all of the classifications for the training regions for each model LR_train_probs <- array(0,dim = c(N1,M)) # log. reg. probabilities from model predictions for each region LR_train_class <- array(0,dim = c(N1,M)) # log. reg. classifications from model predictions for each region LDA_train_class <- array(0,dim = c(N1,M)) # LDA classifications from model predictions for each region QDA_train_class <- array(0,dim = c(N1,M)) # QDA classifications from model predictions for each region # create arrays to store all of the classifications for the new regions for each model LR_new_probs <- array(0,dim = c(N2,M)) # log. reg. probabilities from model predictions for each region LR_new_class <- array(0,dim = c(N2,M)) # log. reg. classifications from model predictions for each region LDA_new_class <- array(0,dim = c(N2,M)) # LDA classifications from model predictions for each region QDA_new_class <- array(0,dim = c(N2,M)) # QDA classifications from model predictions for each region #———————————————————— # run logistic regression, LDA, and QDA versions of each of the models #———————————————————— for (k in 1:M){ # create formulas using the model variables fo1 <- as.formula(paste(“Sink_NotSink ∼”,vars[k], sep = " ")) # log. reg. formula fo2 <- as.formula(paste(“as.factor(Sink_NotSink) ∼”,vars[k], sep = " ")) # LDA and QDA formulas # create log. reg. model and use it to calculate probababilities for training and new regions lr_model <- glm(fo1, data = train_data, family = “binomial”) # model LR_train_probs[,k] <- predict(lr_model, newdata = train_data, type = “response”) # training data LR_new_probs[,k] <- predict(lr_model, newdata = new_regions, type = “response”) # new data # create LDA model and use it to classify training and new regions lda_model <- lda(fo2, data = train_data, prior = c(0.5,0.5)) # model LDA_train_class[,k] <- predict(lda_model, newdata = train_data)$class # training data 350

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

Improved Automated Mapping of Sinkholes

LDA_new_class[,k] <- predict(lda_model, newdata = new_regions)$class # new data # create QDA model and use it to classify training and new regions qda_model <- qda(fo2, data = train_data, prior = c(0.5,0.5)) # model QDA_train_class[,k] <- predict(qda_model, newdata = train_data)$class # training data QDA_new_class[,k] <- predict(qda_model, newdata = new_regions)$class # new data } #———————————————————– # create arrays of 1’s (sinkholes) and 0’s (not sinkholes) #———————————————————– # convert LDA and QDA class values to 0’s and 1’s # when stored in the array, the class “factors” are changed to 1’s and 2’s LDA_train_class <- LDA_train_class - 1 LDA_new_class <- LDA_new_class - 1 QDA_train_class <- QDA_train_class - 1 QDA_new_class <- QDA_new_class - 1 # classify log. reg. using probability cutoff value cutoff = 0.5 # you can change this if you prefer a different cutoff value # classify training data for(k in 1:N1){ for (j in 1:M){ if (LR_train_probs[k,j] < cutoff){ LR_train_class[k,j] <- 0 }else{ LR_train_class[k,j] <- 1 } } } # classify new data for(k in 1:N2){ for (j in 1:M){ if (LR_new_probs[k,j] < cutoff){ LR_new_class[k,j] <- 0 }else{ LR_new_class[k,j] <- 1 } } } #——————— # write output to files #——————— # training data classifications by model write.csv(LR_train_class,“LR_train_results.csv”,row.names = FALSE) write.csv(LDA_train_class,“LDA_train_results.csv”,row.names = FALSE) write.csv(QDA_train_class,“QDA_train_results.csv”,row.names = FALSE) # new data classifications by model write.csv(LR_new_class,“LR_new_results.csv”,row.names = FALSE) write.csv(LDA_new_class,“LDA_new_results.csv”,row.names = FALSE) write.csv(QDA_new_class,“QDA_new_results.csv”,row.names = FALSE)

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 331–351

351

Macrostructural and Microstructural Properties of Residual Soils as Engineered Landﬁll Liner Materials LEE LI YONG VIVI ANGGRAINI* MAVINAKERE ESHWARAIAH RAGHUNANDAN Civil Engineering Discipline, School of Engineering, Monash University Malaysia, Jalan Lagoon Selatan 47500 Bandar Sunway, Selangor, Malaysia

MOHD RAIHAN TAHA Department of Civil Engineering, National University Malaysia, 43600, Bangi Selangor, Malaysia

Key Terms: Residual Soils, Landfill Liner Materials, Microstructural Properties, Macrostructural Properties, Environmental Geology

sample requires further treatment to enhance its properties in order to comply with the requirements of landﬁll liner materials.

ABSTRACT

INTRODUCTION

This study assessed the performance of residual soils with regard to their macrostructural and microstructural properties and compatibility with leachate in pursuit of exploring alternative cost-effective and efficient landfill liner materials. A series of laboratory investigations was conducted on three residual soil samples by using tap water and leachate as permeation fluid to achieve the objectives of the study. The zeta potential measurements revealed that the presence of multivalent cations in the leachate decreased the diffuse double layer (DDL) thickness around the soil particles. The reduced DDL thickness caused a decrease in Atterberg limits of soilleachate samples and changes in the classification of fine fractions. Additionally, the effects of pore clogging attributed to chemical precipitation and bioclogging were responsible for the reduction in measured hydraulic conductivities of soil-leachate samples. These effects can be clearly observed from the field-emission scanning electron microscopy images of soil-leachate samples with the appearance of less visible voids that led to a more compact and dense structure. The formation of new non-clay minerals and associated changes in the Al and Si ratio as reflected in the x-ray diffraction diffractograms and energy-dispersive x-ray analyses, respectively, were attributed to the effects of chemical precipitation. This study concluded that S1 and S2 residual soil samples are potential landfill liner materials because they possess adequate grading characteristics, adequate unconfined compressive strength, low hydraulic conductivity, and good compatibility with leachate. In contrast, the S3

Exponential population growth contributes to the whopping increase in economic development, urbanization, and industrialization, hence, resulting in a high municipal solid waste (MSW) generation rate over the globe (Khandelwal et al., 2019). This rising global waste production leads to alarming environmental concerns associated with air, soil, and water pollution, highlighting the importance of implementation of appropriate MSW management methods (Brunner, 2013). Studies have shown that landfilling is the prevailing waste management method practiced around the world because of the convenience of its implementation (Townsend et al., 2015; Das et al., 2019). Percolation of water through solid wastes contained in landfill produces leachate, and the production of leachate occurs during the operation period of the landfill and continues for many years even after landfill closure (Pandey and Shukla, 2019). The presence of pollutants such as dissolved organic matter, inorganic macro components, heavy metals, and xenobiotic organic compounds in the leachate can potentially contaminate the soil and groundwater (Christensen et al., 1994, 2001). In fact, municipal solid waste landfills are identified as one of the main sources of groundwater contamination due to the leakage of leachate (Han et al., 2016). Effective landfill systems with utilization of lining systems are recommended to prevent deterioration of environmental issues posed by improper MSW management. The bottom-lining system in an engineered landfill acts as a final defense against the migration of leachate to the surrounding environment. Commonly, natural soils that consist of a significant amount of clay

*Corresponding author email: vivi.anggraini@monash.edu

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 353–366

353

Yong, Anggraini, Raghunandan, and Taha

minerals are compacted to create an effective barrier material in the bottom-lining system (Daniel, 1993). If natural clays or clayey soils are not readily available, it is a common practice to blend onsite soils with clay minerals such as sodium bentonite (USEPA, 1989; Rico et al., 2011; and Tong and Shackelford, 2016). The presence of montmorillonite in bentonite enhances the properties of natural soil in terms of low hydraulic conductivity and high sorption capacity (Shackelford and Sample-Lord, 2014; Ghadr and Assadi-Langroudi, 2018). However, the incompatibility of bentonite with leachate has made this product unfavorable as bentonite is prone to exchange by other divalent and trivalent cations present in leachate that limits the swelling of bentonite, thereby increasing the hydraulic conductivity of the bentonite-soil mixture (Shackelford et al., 2000; Gates et al., 2009; and Bouzza and Gates, 2014). Hence, alternative liner materials with appropriate properties that also show good compatibility with leachate are much sought after and receive growing interest among the researchers. This phenomenon has contributed to various studies assessing the feasibility of (i) natural soils (red mud, marine clay, tropical residual soils, shales, zeolite, sepiolite, and smectite-rich claystones) (Mohamedzein et al., 2005; Chalermyanont et al., 2009; Turan and Ergun, 2009; Musso et al., 2010; Guney et al., 2014; and Rubinos and Spagnoli, 2019); (ii) soil composites (shale-clay mixture, lime-clay mixture, sand-bentoniteglass fiber composite, silica fume-clay composite, activated carbon-clay composite, and lateritic fiber-soil composite) (Kalkan and Akbulut, 2004; Lu et al., 2010; Firoozfar and Khosroshiri, 2016; Li et al., 2017; Ehrlich et al., 2019; and Mukherjee and Kumar Mishra, 2020); and (iii) waste products (spilitic mining wastes, coal gangue, granulated blast furnace slagbentonite-cement mixture) (Maritsa et al., 2016; Wu et al., 2017; and Manikanta and Shankar, 2019) as alternative landfill liner materials over the past few years. Generally, tropical regions experience higher rates of rock weathering as the hot, humid environment expedites the weathering process. Residual soils developed in the tropical climatic regions by severe chemical weathering are referred to as tropical residual soils. Various chemical processes such as hydrolysis, oxidation, reduction, hydration, carbonation, and solution alter the mineral constituents of parent rock and lead to the formation of new compounds (Tan, 2004). The rock weathering process involves leaching of the bases (K, Na, Ca, and Mg) and silica in solution with the insoluble iron and aluminum oxides remaining in place (Lelong et al., 1976; Adeyeri, 2015). As a result, clay minerals (montmorillonite, illite, and kaolinite) form as the silica combines with other weathered products. The presence of clay minerals contributes to low

354

hydraulic conductivity of tropical residual soils, which is one of the pivotal requirements for landfill liner materials. In view of that, tropical residual soils can serve as potential construction materials for a compacted soil liner. Researchers have extensively evaluated the suitability of tropical residual soils as landfill liner materials based on their soil properties, such as hydraulic conductivity (Taha and Kabir, 2005; Osibinu and Charles, 2006; Chalermyanont et al., 2009; Bello, 2013; and Morandini and Leite, 2015) and unconfined compressive strength (Taha and Kabir, 2005; Osibinu and Charles, 2006; and Bello, 2012). The effects of leachate on hydraulic conductivity and the chemical and mineralogical compositions of tropical residual soils have also been evaluated by Frempong and Yanful (2005, 2008). However, the feasibility of using tropical residual soils as landfill liner materials based on their microstructural properties and their associated microstructural changes after interaction with MSW leachate is rarely addressed in the literature. Hence, this study aims at providing a detailed assessment on the macrostructural and microstructural properties of natural tropical residual soils as landfill liner materials. The influence of leachate on the properties of tropical residual soils was also investigated. A series of investigations involving geotechnical and microstructural tests was performed to achieve the stated objectives. MATERIALS Tropical Residual Soils The soil samples used in this study were collected from three construction sites within east peninsular Malaysia (Figure 1a): Sample 1 (S1), National Skills Development Center Serendah; Sample 2 (S2), Bandar Pinggiran Subang; and Sample 3 (S3), Presint 14 Putrajaya (Figure 1b). S1 sample is the weathering product of Silurian to Middle Ordovician Terolak Formation, which comprises schist, phyllite, slate, and limestone (Malaysia Ministry of Natural Resources and Environment, 2010a). S2 and S3 samples are the residual soils weathered from the middle Permian to Carboniferous Kenny Hill Formation that mainly consists of interbedded shale, mudstone, and sandstone (Malaysia Ministry of Natural Resources and Environment, 2010b). The disturbed soil samples were collected using a shovel at a depth of 1 m below the ground surface. The collected soil samples were bagged and transported to the laboratory for testing. MSW Leachate The MSW leachate used in the present study was obtained from Taman Beringin Transfer Station, which

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 353–366

Residual Soils as Engineered Landﬁll Liner Materials

Figure 1. (a) Location of soil sampling (VectorStock, n.d.). (b) Soil samples S1, S2, and S3.

represents the properties of solid wastes in Kuala Lumpur, Malaysia. The average composition of solid wastes generated in the urban areas of Malaysia is comprised of 45 percent food waste, 24 percent plastic, 7 percent paper, 6 percent iron, 3 percent glass and 15 percent others (Ghasimi and Tey, 2010). Taman Beringin transfer station receives and processes municipal solid wastes from Kuala Lumpur by compacting and applying odor and dust control measures prior to transportation to the Bukit Tagar sanitary landfill. The collected leachate was then stored in a refrigerator at 4°C to reduce any chemical reactions prior to testing. The main composition of the leachate is presented in Table 1. The leachate used in this study had a pH of 3.57 and a BOD5 /COD ratio of 0.39. The leachate properties exhibited a young leachate characteristic, which has a pH < 6.5 and a BOD5 /COD ratio > 0.3 (Zainol et al., 2012). Inorganic compounds such as chloride and sulfate and heavy metals were the common pollutants present in collected leachate. The high percentage of biological oxygen demand (23,800 mg/L) can likely be attributed to the presence of a high proportion of active bacteria in the leachate. In view of the high total dissolved solids (9,660 mg/L) present in the leachate, the leachate collected was first filtered through a 75-µm sieve before conducting any experimental test to reduce the pore clogging effects of the soil samples due to the presence of suspended solids. METHODS Soil Sample Preparation The collected soil samples were first oven-dried at 105°C for at least 24 hours. The dried samples were then manually crushed using mortar and pestle

and sieved through a 4.75-mm sieve (for compaction, hydraulic conductivity, and unconfined compressive strength tests), a 425-µm sieve (for Atterberg limits and specific gravity tests), and a 75-µm sieve (for zeta, potential test) prior to testing. Particle Size Distribution, Atterberg Limits, and Specific Gravity The particle size distribution of the natural soil samples was determined by performing wet sieving and hydrometer sedimentation analyses in accordance with BS1377 (BSI, 1990). The Atterberg limits, such as Table 1. Composition of the leachate. Parameter Physical and aggregate properties Total dissolved solids (mg/L) Total hardness as CaCO3 (mg/L) pH Electrical conductivity (mS/cm) Inorganic and nonmetallic properties (mg/L) Chloride Sulfate Metals and major cations (mg/L) Cadmium Chromium Copper Lead Manganese Nickel Zinc Environmental quality (mg/L) Biological oxygen demand Chemical oxygen demand Total suspended solids Ammonia as N

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 353–366

Value 9,660 1,900 3.57 12.61 2,370 497 0.02 0.49 0.39 0.08 5.11 0.36 6.1 23,800 61,200 12,000 462

355

Yong, Anggraini, Raghunandan, and Taha

liquid limit, plastic limit, and plasticity index of the natural and leachate-contaminated soil samples, were determined based on the procedures outlined in BS1377 (BSI, 1990). The liquid limit of the soil samples was obtained by conducting the cone penetration test. The specific gravity of the natural soil samples was determined by adopting the small pycnometer approach as stipulated in British BS1377 (BSI, 1990). Compaction Test The compaction characteristics of natural soil samples (maximum dry density and optimum moisture content) were determined by the light compaction method (Standard Proctor compaction) utilizing a 2.5kg rammer dropped from a height of 300 mm as stipulated in BS1377 (BSI, 1990). The soil samples were first prepared at a different range of water contents and cured for 24 hours in plastic sealed bags. The prepared soil samples were then compacted in three layers with 27 blows applied to each soil layer. The compaction characteristics of the natural soil samples were obtained from the graph dictating the relationship between dry density and moisture content of the soil samples. Hydraulic Conductivity The falling head hydraulic conductivity tests were carried out following the procedures described in ASTM D5856-15 (ASTM, 2007). The soil samples were first prepared by mixing them thoroughly with tap water at their respective optimum moisture content. The prepared samples were then compacted in a rigid-wall compaction-mold permeameter and mounted to the falling head permeameter as shown in Figure 2 with the samples initially permeated with tap water. The permeation process continued until the termination criteria outlined in ASTM D5856-15 (ASTM, 2007) were satisfied before taking the first set of head loss readings across the soil samples. This was to ensure full saturation of the soil samples that gave rise to stable hydraulic conductivity readings over the testing duration. The test was then repeated using MSW leachate as the permeant liquid. The hydraulic conductivity of each soil sample was calculated using Eq. 1 and recorded for 4 weeks consecutively. k=

aL h1 ln At h2

(1)

where k = hydraulic conductivity, cm/s; a = crosssectional area of the standpipe, cm2 ; L = length of soil specimen, cm; A = cross-sectional area of the soil specimen, cm2 ; h1 = head loss across the soil specimen at 356

Figure 2. Experimental conﬁguration for the falling head hydraulic conductivity test.

time t1 , cm; h2 = head loss across the soil specimen at time t2 , cm; t = time difference between h1 and h2 , s. Unconfined Compressive Strength The unconfined compression strength (UCS) test was conducted on natural soil samples compacted at their respective optimum moisture content at a heightto-diameter ratio of 2.0 (height of 100 mm and diameter of 50 mm). The compacted soil samples were wrapped with few layers of plastic wrapping film and kept in a plastic sealed bag for 1-day curing prior to the UCS test. After curing, the compacted soil samples were loaded at an applied strain rate of 1 percent per minute in accordance with ASTM D2166 (ASTM,

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 353–366

Residual Soils as Engineered Landﬁll Liner Materials

2005) method. The applied load and axial deformation were recorded using an automated data acquisition unit. The loading was continued until 15 percent of strain was achieved or when the applied load values decreased with increasing strain. Triplicate samples were tested for each natural soil type to confirm the consistency of the results. Zeta Potential Test Zeta potential indicates the net charge of a clay particle or its aggregate, which is also the primary driving force for clay particle interaction (Zhao et al., 2017). The electrical potential at the intersection between the fixed and mobile parts of the electrical double layer is defined as zeta potential and can be used to characterize the diffusive double layer (Asadi et al., 2011). The soil solution was first prepared by mixing 10 mg of sieved soil in 10 mL of tap water or leachate and then slowly injecting it to the folded capillary cell (DTS1060) using a plastic syringe to avoid the development of air bubbles. The folded capillary cell was then placed in a zeta meter equipped with a microprocessor to perform zeta potential measurements. The zeta meter determined the electrophoretic mobility of the soil particles within the soil solution expressed in terms of zeta potential by applying the Henry equation (Malvern Instruments Ltd, 2004). Field-Emission Scanning Electron Microscopy and Energy Dispersive X-Ray The morphology of the natural and leachatecontaminated soil samples was observed by performing the field-emission scanning electron microscopy (FESEM) test. The air-dried samples for natural and leachate-contaminated soil samples obtained from the hydraulic conductivity test samples were first crushed to relatively small soil chunks before performing the FESEM and energy dispersive x-ray (EDX) tests. The samples were then attached to sample holders using carbon tape and coated with platinum for 40 seconds by using a Quorum/Q150RS Sputter Coater to provide less electrostatically distorted images. The images of the soil samples were captured with a Hitachi SU8010 at magnification of 50k. The elemental compositions of the natural soil and soil-leachate samples were analyzed with the EDX analyzer (Horiba Inca XMax50 EDX). X-Ray Diffraction X-ray diffraction (XRD) tests were performed to determine the mineralogical composition of the natural and leachate-contaminated soil samples. The air-

Figure 3. Particle size distribution curve for S1, S2, and S3 samples.

dried samples for natural and leachate-contaminated soil samples obtained from the hydraulic conductivity test samples were first grinded into powder form. The XRD test was then performed using a Bruker/D8 Discover diffractometer with copper as the x-ray radiation source. The diffracted x-ray from the samples was collected by the x-ray detector at angle of 2θ ranges from 5° to 90° with a step size of 0.03°/min. Characterization of crystalline materials was carried out using DIFFRAC.EVA software. RESULTS AND DISCUSSION Properties of the Natural Soil Samples The physical properties of the natural soil samples are shown in Table 2. Figures 3 and 4 depict the particle size distribution curve and compaction curve for S1, S2, and S3 samples, respectively. The grain size analyses revealed that three of the soil samples contained significant fine fractions. S1 (17.75 percent) and S2 samples (15 percent) had higher proportions of clayey material, whereas the S3 sample contained a higher fraction of silty material (10.03 percent). S1 and S2 samples were classified as clayey sand, and the S3 sample was categorized as silty sand. The grading characteristic of S1 and S2 samples fulfilled the requirements of recommended materials for landfill liner materials as summarized in Table 2. However, the S3 sample failed to meet the requirements as its fine and clay fractions were not within the stated limits. As observed from Table 2, S1 (226.94 kPa) and S2 (398.37 kPa) samples recorded higher UCS than the S3 sample (175.12 kPa). S1 and S2 samples contained higher percentages of fines (S1 sample 31.90 percent, S2 sample 28.90 percent) than sample S3

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 353–366

357

Yong, Anggraini, Raghunandan, and Taha Table 2. Soil properties of three samples. Soil Samples Properties Particle size distribution (%) Gravel Sand Fines fractions (silt + clay) Silt Clay Speciﬁc gravity Maximum dry density (g/cm3 ) Optimum moisture content (%) Unconﬁned compressive strength (kPa)

5.00 63.10 31.90 14.15 17.75 2.75 1.59 24.00 226.94

1.85 69.25 28.90 13.90 15.00 2.70 1.73 18.00 398.37

0 84.10 15.90 10.03 5.87 2.84 1.50 30.50 175.12

(15.90 percent). In fact, the UCS of S1 and S2 samples complied with the minimum UCS requirement of 200 kPa for landfill liner materials as recommended by Daniel and Wu (1993). This suggests that S1 and S2 samples can be utilized as bottom-liner materials in engineered landfills owing to their adequate grading characteristic and strength properties. However, the S3 sample may require further treatment or grinding to improve its grading characteristic and strength to meet the requirements of landfill liner materials.

Requirements ࣘ30% — ࣙ20–30% — ࣙ15% — — — ࣙ200 kPa

Reference Daniel (1993), Benson and Trast (1995), and UKEA (2014)

— — — Daniel and Wu (1993)

Compatibility of the Natural Soil Samples with MSW Leachate Atterberg Limits Table 3 summarizes Atterberg limits for the three soil samples and the classifications of their fine fractions in accordance with BS5930 (BSI, 1999). Samples S1 and S2 were dominated by higher percentages of clay. The fine fractions of the two soils were classified as inorganic clay of intermediate (CI) and

Figure 4. Compaction curve for S1, S2, and S3 samples.

358

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 353–366

Residual Soils as Engineered Landﬁll Liner Materials Table 3. Atterberg limits for soil-water and soil-leachate samples. Soil-Water Samples

Atterberg limit (%) Liquid limit, WL Plastic limit, Wp Plasticity index, IP Classiﬁcation of ﬁne fractions

Soil-Leachate Samples

48.00 23.64 24.36 CI

50.10 21.80 28.30 CH

73.00 36.50 36.50 MH

46.00 23.33 22.87 CI

46.20 19.00 27.20 CI

57.80 25.76 32.04 CH

high plasticity (CH), respectively. The ﬁne fraction of the S3 sample fell into the category of silt with high plasticity (MH). It can be noticed that soilleachate samples showed a lower liquid limit and plasticity index than the soil-water samples. According to the Gouy-Chapman theory, an increase in electrolyte concentration or cation valence in the pore water will decrease the thickness of the diﬀuse double layer (DDL) around the clay particle. The presence in MSW leachate of higher valence cations (e.g., Cadmium Cd2+ , Chromium Cr3+ , Copper Cu2+ , Lead Pb2+ , Nickel Ni2+ , and Zinc Zn2+ , as shown in Table 1) rather than monovalent hydrogen cation in water (H+ ) caused a reduction in the thickness of DDL and surface potential (refer to Figure 5). This phe-

nomenon contributed to reduced liquid and plastic limits in soil-leachate samples. Similar findings were also presented by Frempong and Yanful (2008) and Harun et al. (2013). The changes in Atterberg limits of the soil-leachate samples consequently altered the classification of fine fractions of soil sample S2 (from high plasticity clay to intermediate plasticity clay) and S3 (from high plasticity silt to high plasticity clay). Zeta Potential Figure 6 indicates the zeta potential for soil-water and soil-leachate samples. The three soil samples showed negative zeta potential in the presence of either tap water or MSW leachate. The zeta potential

Figure 5. Eﬀect of MSW leachate on the soil DDL thickness.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 353–366

359

Yong, Anggraini, Raghunandan, and Taha

Figure 6. Zeta potential for soil-water and soil-leachate samples.

of the S3-water sample (–13.5 mV) was less negative than those of the S1-water (–22.2 mV) and S2-water samples (–26.9 mV). This is because the S3 sample consists of a lesser percentage of clay that possesses a negative surface charge. The zeta potential for S1, S2, and S3 samples reduced to –17.3 mV, –17.2 mV, and –11.9 mV, respectively, in the presence of leachate. This reduction in zeta potential can be attributed to the increased electrolyte concentration and cation valence in leachate, reflecting compression of the DDL. Similarly, Chorom and Rengasamy (1995) and Kaya and Yukselen (2005) reported that smaller zeta potential values were observed as a result of increased electrolyte concentration. These observations support the findings of reduced Atterberg limits of soil-leachate samples associated with reduced DDL as explained in the “Atterberg Limits” section. Hydraulic Conductivity Figure 7 presents the variation of hydraulic conductivities for tropical residual soil samples permeated with either tap water or with MSW leachate over 4weeks permeation time. The hydraulic conductivities for S1, S2, and S3 samples exhibited negligible time dependence when permeated with tap water. Conversely, when three of the soil samples were permeated with MSW leachate, significant reductions of hydraulic conductivities with permeation time were observed. All soil samples permeated with leachate showed lower hydraulic conductivity compared to those permeated with tap water after 4 months of permeation time. This observation is in contrast with the results reported by Shackelford et al. (2000) as the author indicated that

360

the reduction in double layer thickness would result in an increase in hydraulic conductivity. This variation is probably due to the nature of the permeation liquid used in the experiment because MSW leachate was used in this study instead of synthetic leachate. Unlike synthetic leachate that consists mainly of inorganic salt solutions, MSW leachates contain a mixture of dissolved organic matter, inorganic macro components, heavy metals, and xenobiotic organic compounds (Christensen et al., 2001; Bradshaw and Benson, 2014). This reduction in measured hydraulic conductivities for soil-leachate samples has been demonstrated by numerous researchers as pore clogging due to (i) the presence of microorganism growth and (ii) chemical precipitation that obstructed the flow of leachate through the compacted soil samples (Frempong and Yanful, 2008; Francisca and Glatstein, 2010; and Aldaeef and Rayhani, 2014). These observations also imply that tropical residual soils still maintain their low hydraulic conductivity properties after being permeated with MSW leachate. It is noted that the measured hydraulic conductivities of S1 and S2 samples permeated with tap water and leachate were within the hydraulic conductivity requirement of liner material, which is less than 1 × 10–7 cm/s as recommended by USEPA (1989), Daniel (1993), and Rowe et al. (2004). In contrast, the hydraulic conductivities of the S3 sample permeated with tap water did not fulfill the requirement of liner materials. Hence, only S1 and S2 samples can be regarded as potential bottom-liner materials in engineered landfill due to their low hydraulic conductivity and good compatibility with the leachate. FESEM FESEM analyses were conducted on soil-water and soil-leachate samples to observe the morphology changes on the soil surface after application of MSW leachate. Figure 8a, c, and e shows the FESEM images for the control sample S1, S2, and S3, respectively. It appears that the control samples had a relatively loose morphology with a considerable quantity of voids in varying sizes. In addition, the kaolinite clay minerals that typically appear in platy forms can be clearly observed in all control samples. This observation is consistent with the higher percentage of kaolinite identified in x-ray diffractograms. At the same magnification scale, the appearance of voids in the soil-leachate samples seems to be less visible, resulting in a more compact and dense structure as observed in Figure 8b, d, and e. The changes in the morphology of soilleachate samples can be attributed to the soil-leachate reactions, which gave rise to chemical precipitation and bioclogging within the soil pores as discussed in the

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 353–366

Residual Soils as Engineered Landﬁll Liner Materials

Figure 7. Variation of hydraulic conductivities of S1, S2, and S3 samples permeated with tap water and MSW leachate over testing time.

“Hydraulic Conductivity” section. Consequently, this phenomenon increased the tortuosity of the leachate ﬂow path and contributed to lower hydraulic conductivities of soil-leachate samples as presented in Figure 7. EDX EDX analyses were conducted to study the changes in elemental compositions of soil samples contaminated with MSW leachate. Table 4 summarizes the elemental compositions of control and soil-leachate samples. It is seen that Al, Fe, and Si were commonly present in all control soil samples, as Al, Fe, and some

of the silica remained insoluble due to the rock weathering processes. Similar findings were reported by Latifi et al. (2017) and Yong et al. (2019), as they observed the presence of Al, Fe, and Si in residual soils due to its lateritic nature. It can be clearly seen that the soilleachate samples contained higher percentages of Al and Si. It is speculated that the presence of cations in the MSW leachate triggers chemical reactions such as cation exchange and physical bonding with the aluminum and silica on the soil surface to form new minerals as found in the diffractograms of soil-leachate samples, resulting in higher percentages of Al and Si in soil-leachate samples. This phenomenon supported the observation of reduced hydraulic conductivities of

Table 4. Elemental composition (%) of soil-water and soil-leachate samples. Element C O Al Si Fe

S1-Water

S1-Leachate

S2-Water

S2-Leachate

S3-Water

S3-Leachate

5.70 49.52 15.59 4.13 25.06

5.71 52.58 17.16 18.06 6.49

4.22 54.28 17.72 4.06 19.72

9.76 43.85 29.91 10.14 6.34

8.58 47.18 11.94 9.79 22.51

7.92 44.52 20.41 18.98 8.17

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 353–366

361

Yong, Anggraini, Raghunandan, and Taha

Figure 8. FESEM images of the (a) S1 control sample, (b) S1-leachate sample, (c) S2 control sample, (d) S2-leachate sample, (e) S3 control sample, and (f) S3-leachate sample.

362

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 353–366

Residual Soils as Engineered Landﬁll Liner Materials

Figure 9. (a) X-ray diﬀractograms for soil-water samples. (b) X-ray diﬀractograms for soil-leachate samples.

soil-leachate samples due to chemical precipitation as elaborated in the “Hydraulic Conductivity” section. Contrarily, a notable decrease in Fe percentage was observed in all soil-leachate samples. These ﬁndings suggest that the acidic leachate dissolved some of the Fe available in the soil samples. On the other hand, the presence of C observed in Table 4 was attributed to the presence of carbon tape that secured the soil samples to the equipment stub. XRD Figure 9a and b illustrates the diﬀractograms for natural and leachate-contaminated soil samples sub-

jected to 1 month of permeation. It can be clearly observed that quartz, kaolinite, and illite are the main minerals present in the natural soil samples. This observation is consistent with the findings reported by Latifi et al. (2015) and Rashid et al. (2017). The predominance of kaolinite in the natural soil samples contributes to (i) the low hydraulic conductivity values observed in the “Hydraulic Conductivity” section, and (ii) the minor swelling or shrinkage upon wetting due to its chemical structure held strongly by hydrogen bonding (Singh and Huat, 2013). This characteristic of kaolinite appears to be beneficial for landfill liner materials, as little swelling or shrinkage implies a lesser tendency for volume change and associated crack

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 353–366

363

Yong, Anggraini, Raghunandan, and Taha

formation. Minor reflections of goethite were also identified in three of the natural soil samples. The presence of goethite explains the appearance of yellowish and blackish brown color of three natural soil samples as observed in Figure 1b. Similarly, diffractograms of leachate-contaminated soil samples showed the presence of quartz, kaolinite, illite, and goethite, suggesting that quartz, kaolinite, illite, and goethite were resistant to leachate exposure. Additionally, the formation of new non-clay minerals with minor reflections, such as hydroxyapatite, pyromorphite, dolomite, and ferrihydrite, were spotted in the soil-leachate samples. This coincides with the observation reported by Frempong and Yanful (2005, 2008) of the similar appearance of non-clay minerals upon leachate exposure on tropical residual soils. The changes in XRD diffractograms of soil-leachate samples indicate that the soil-leachate reactions triggered the precipitation and dissolution of some minerals.

the explanation, as the effects of pore clogging led to less visible voids and a denser soil structure. (iv) Soil-leachate reactions triggered the precipitation and dissolution of some minerals as portrayed in the x-ray diffractograms of soil-leachate samples. The effects of chemical precipitation were also reflected in the elemental compositions of soilleachate samples as higher percentages of Al and Si in soil-leachate samples were recorded. In short, S1 and S2 samples are regarded as potential landfill liner materials due to their adequate properties and a good compatibility with leachate. However, S3 sample may require further treatment to improve its properties prior to its utilization as landfill liner material. It is recommended that further studies can be conducted on the transport parameters (e.g., diffusion coefficient and retardation factor) of leachate pollutants through tropical residual soils, which may also forecast the performance of tropical residual soils as landfill liner materials.

CONCLUSION A series of laboratory investigations was conducted to study the feasibility of tropical residual soils as landfill liner materials in terms of their macrostructural and microstructural properties. Additionally, the influence of MSW leachate on the properties of tropical residual soils was examined to observe the compatibility of tropical residual soils with MSW leachate. The following conclusions can be drawn from the experimental results: (i) S1 and S2 samples fulfilled the requirements of landfill liner materials due to their adequate grading characteristics, unconfined compressive strength, and hydraulic conductivity. However, the S3 sample failed to meet the requirements due to its low percentages of fines and clay fractions, inadequate UCS, and hydraulic conductivity that was not within the stated limits. (ii) The presence of multivalent cations in MSW leachate reduced the DDL thickness around the clay particles. As a result, the Atterberg limits of soil-leachate samples reduced and caused some changes in the classification of fine fractions for soil-leachate samples. The reduction in DDL thickness also decreased the zeta potential of all leachate-contaminated soil samples. (ii) The hydraulic conductivities of tropical residual soil samples reduced after permeated with MSW leachate. This is attributed to chemical precipitation and microorganism growth in the soil pores that obstructed the flow of leachate through soil. FESEM images of soil-leachate samples support

364

ACKNOWLEDGMENTS Financial support for this study is provided by Ministry of Higher Education in Malaysia with grant number FRGS/1/2017/TK01/MUSM/03/01. This support is gratefully acknowledged. CONFLICT OF INTEREST None. REFERENCES Adeyeri, J. B., 2015, Technology and Practice in Geotechnical Engineering, IGI Global, Pennsylvania, United States. 835 p. Aldaeef, A. A. and Rayhani, M. T., 2014, Hydraulic performance of Compacted Clay Liners (CCLs) under combined temperature and leachate exposures: Waste Management, Vol. 34, No. 12, pp. 2548–2560. Asadi, A.; Moayedi, H.; Huat, B. K. K.; Boroujeni, F. Z.; Parsaie, A.; and Sojoudi, S., 2011, Prediction of zeta potential of tropical peat in the presence of diﬀerent cations using artiﬁcial neural networks: International Journal of Electrochemical Science, Vol. 6, pp. 1146–1158. ASTM D2166, 2005, Standard Test Method for Unconfined Compressive Strength of Cohesive Soil: ASTM International, West Conshohocken, PA. ASTM D5856-15, 2007, Standard Test Method for Measurement of Hydraulic Conductivity of Porous Material Using a RigidWall, Compaction-Mold Permeameter: ASTM International, West Conshohocken, PA. Bello, A. A., 2012, Geotechnical evaluation of reddish brown tropical soils: Geotechnical and Geological Engineering, Vol. 30, No. 2, pp. 481–498. Bello, A. A., 2013, Hydraulic conductivity of three compacted reddish brown tropical soils: KSCE Journal of Civil Engineering, Vol. 17, No. 5, pp. 939–948.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 353–366

Residual Soils as Engineered Landfill Liner Materials Benson, C. H. and Trast, J. M., 1995, Hydraulic conductivity of thirteen compacted clays: Clays and Clay Minerals, Vol. 43, No. 6, pp. 669–681. Bouzza, A. and Gates, W. P., 2014, Overview of performance compatibility issues of GCLs with respect to leachates of extreme chemistry: Geosynthetics International, Vol. 21, No. 2, pp. 151–167. Bradshaw, S. L. and Benson, C. H., 2014. Effect of municipal solid waste leachate on hydraulic conductivity and exchange complex of geosynthetic clay liners: Journal of Geotechnical and Geoenvironmental Engineering, Vol. 04013038, pp. 1–17. Brunner, P. H., 2013, Cycles, spirals and linear flows: Waste Management & Research, Vol. 31, pp. 1–2. BS1377, 1990, Methods of Test for Soils for Civil Engineering Purposes: British Standard Institution, London, U.K. BS5930, 1999, Code of Practice for Site Investigations: British Standard Institution, London, U.K. Chalermyanont, T.; Arrykul, S.; and Charoenthaisong, N., 2009, Potential use of lateritic and marine clay soils as landfill liners to retain heavy metals: Waste Management, Vol. 29, pp. 117–127. Chorom, M. and Rengasamy, P., 1995, Dispersion and zeta potential of pure clays as related to net particle charge under varying pH, electrolyte concentration and cation type: European Journal of Soil Science, Vol. 46, pp. 657–665. Christensen, T. H.; Kjeldsen, P.; Albrechtsen, H.; Heron, G.; Nielsen, P. H.; Bjerg, P. L.; and Holm, P. E., 1994, Attenuation of landfill leachate pollutants in aquifers: Critical Reviews in Environmental Science and Technology, Vol. 24, No. 2, pp. 119–202. Christensen, T. H.; Kjeldsen, P.; Bjerg, P. L.; Jensen, D. L.; Christensen, J. B.; Baun, A.; Albrechtsen, H.; and Heron, G., 2001, Biogeochemistry of landfill leachate plumes: Applied Geochemistry, Vol. 16, No. 7–8, pp. 659–718. Daniel, D. E., 1993, Clay liners. In: Daniel, D. E. (Editor), Geotechnical Practice for Waste Disposal: Chapman & Hall, London, U.K., pp. 137–163. Daniel, D. E. and Wu, Y. K., 1993, Compacted clay liners and covers for arid sites: Journal of Geotechnical Engineering, Vol. 119, No. 2, pp. 223–237. Das, S.; Lee, S. H.; Kumar, P.; Kim, K-H.; Lee, S. S.; and Bhattacharya, S. S., 2019, Solid waste management: Scope and the challenge of sustainability: Journal of Cleaner Production, Vol. 228, pp. 658–678. Ehrlich, M.; Almeida, M. S. S.; and Curcio, D., 2019, Hydromechanical behavior of a lateritic fiber-soil composite as a waste containment liner: Geotextiles and Geomembranes, Vol. 47, pp. 42–47. Firoozfar, A. and Khosroshiri, N., 2016, Kerman clay improvement by lime and bentonite to be used as materials of landfill liner: Geotechnical and Geological Engineering, Vol. 35, pp. 559–571. Francisca, F. M. and Glatstein, D. A., 2010, Long term hydraulic conductivity of compacted soils permeated with landfill leachate: Applied Clay Science, Vol. 49, pp. 187–193. Frempong, E. M. and Yanful, E. K., 2005, Geoenvironmental assessment of two tropical clayey soils for use as engineered liner materials. In Rathje, E. M. (Editor), Waste Containment and Remediation (GSP 142): Proceedings of Geo-Frontiers 2005: Geo-Institute of American Society of Civil Engineers, Austin, TX, pp 1–10. Frempong, E. M. and Yanful, E. K., 2008, Interactions between three tropical soils and municipal solid waste landfill leachate: Journal of Geotechnical and Geoenvironmental Engineering, Vol. 134, No. 3, pp. 379–396.

Gates, W. P.; Bouazza, A.; and Churchman, G. J., 2009, Bentonite clay keeps pollutants at bay: Elements, Vol. 5, No. 2, pp. 105–110. Ghadr, S. and Assadi-Langroudi, A., 2018, Structure-based hydro-mechanical properties of sand-bentonite composites: Engineering Geology, Vol. 235, pp. 53–63. Ghasimi, D. S. M. and Tey, B. T., 2010, Municipal solid waste management at Taman Beringin Transfer Station in Malaysia: Waste Management, Vol. 30, pp. 355–359. Guney, Y.; Cetin, B.; Aydilek, A. H.; Tanyu, B. F.; and Koparal, S., 2014, Utilization of sepiolite materials as a bottom liner material in solid waste landfills: Waste Management, Vol. 34, pp. 112–124. Han, Z.; Ma, H.; Shi, G.; He, L.; Wei, L.; and Shi, Q., 2016, A review of groundwater contamination near municipal solid waste landfill sites in China: Science of The Total Environment, Vol. 569–570, pp. 1255–1264. Harun, S. N.; Ali Rahman, Z.; Rahim, S. A.; Lihan, T.; and Idris, W. M. R., 2013, Effects of leachate on geotechnical characteristics of sandy clay soil. In Abdul Murad, A. M. H.; Choong, C. Y.; Ismail, E. S.; Maskat, M. Y.; Md Noorani, M. S.; Ibrahim, N.; Abdul Karim, N. H.; Yahya, R.; Mohd Khalid, R.; Ismail, W. R.; Wee, S. L.; and Ibrahim, Z. (Editors), The 2013 UKM FST Postgraduate Colloquium: Proceedings of AIP Conference: AIP Publishing, Selangor, Malaysia, pp. 530–533. Kalkan, E. and Akbulut, S., 2004, The positive effects of silica fume on the permeability, swelling pressure and compressive strength of natural clay liners: Engineering Geology, Vol. 73, pp. 145–156. Kaya, A. and Yukselen, Y., 2005, Zeta potential of clay minerals and quartz contaminated by heavy metals: Canadian Geotechnical Journal, Vol. 42, pp. 1280–1289. Khandelwal, H.; Dhar, H.; Thalla, A. K.; and Kumar, S., 2019, Application of life cycle assessment in municipal solid waste management: A worldwide critical review: Journal of Cleaner Production, Vol. 209, pp. 630–654. Latifi, N.; Eisazadeh, A.; Marto, A.; and Meehan, C. L., 2017, Tropical residual soil stabilization: A powder form material for increasing soil strength: Construction and Building Materials, Vol. 147, pp. 827–836. Latifi, N.; Marto, A.; and Eisazadeh, A., 2015, Physicochemical behavior of tropical laterite soil stabilized with non-traditional additive: Acta Geotechnica, Vol. 11, No. 2, pp. 433–443. Lelong, F.; Tardy, Y.; Grandin, G.; Trescases, J. J.; and Boulange, B., 1976, Pedogenesis, chemical weathering and processes of formation of some supergene ore deposits. In Wolf, K. H. (Editor), Handbook of Strata-bound and Stratiform Ore Deposits: Elsevier Scientific Publishing Company, Amsterdam, Netherlands, pp. 93–173. Li, L.; Lin, C.; and Zhang, Z., 2017, Utilization of shale-clay mixtures as a landfill liner material to retain heavy metals: Materials & Design, Vol. 114, pp. 73–82. Lu, H.; Luan, M.; and Zhang, J., 2010, Transport of Cr (VI) through clay liners containing activated carbon or acidactivated bentonite: Applied Clay Science, Vol. 50, pp. 99–105. Malaysia Ministry of Natural Resources and Environment, 2010a, Geological Map of Rawang. 1:63360: Malaysia Ministry of Natural Resources and Environment, Kuala Lumpur, Malaysia. Malaysia Ministry of Natural Resources and Environment, 2010b, Kuala Lumpur. 1:63360: Malaysia Ministry of Natural Resources and Environment, Kuala Lumpur, Malaysia. Malvern Instruments Ltd, 2004, Zetasizer Nano Series User Manual: Malvern Instruments Ltd, Worcestershire, U.K.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 353–366

365

Yong, Anggraini, Raghunandan, and Taha Manikanta, D. and Shankar M. U., 2019, Use of ground granulated blast furnace slag blended with bentonite and cement mixtures as a liner in a landfill to retain diesel oil contaminants: Journal of Environmental Chemical Engineering, Vol. 7, pp. 1–13. Maritsa, L.; Tsakiridis, P. E.; Katsiotis, N. S.; Tsiavos, H.; Velissariou, D.; Xenidis, A.; and Beazi-Katsioti, M., 2016, Utilization of spilitic mining wastes in the construction of landfill bottom liners: Journal of Environmental Chemical Engineering, Vol. 4, pp. 1818–1825. Mohamedzein, Y. E. A.; Al-Rawas, A. A.; Al-Aghbari, M. Y.; Qatan, A.; and Al-Rawas, A-H., 2005, Assessment of crushed shales for use as compacted landfill liners: Engineering Geology, Vol. 80, pp. 271–281. Morandini, T. L. C. and Leite, A. D. L., 2015, Characterization and hydraulic conductivity of tropical soils and bentonite mixtures for CCL purposes: Engineering Geology, Vol. 196, pp. 251–267. Mukherjee, K. and Kumar Mishra, A., 2020, Undrained performance of sustainable compacted sand-bentonite–glass fiber composite for landfill application: Journal of Cleaner Production, Vol. 244, pp. 1–14. Musso, T. B.; Roehl, K. E.; Pettinari, G.; and Vallés, J. M., 2010, Assessment of smectite-rich claystones from Northpatagonia for their use as liner materials in landfills: Applied Clay Science, Vol. 48, pp. 438–445. Osibinu, K. J. and Charles, M. O. N., 2006, Design of compacted lateritic soil liners and covers: Journal of Geotechnical and Geoenvironmental Engineering, Vol. 132, No. 2, pp. 203–213. Pandey, L. M. S. and Shukla, S. K., 2019, An insight into waste management in Australia with a focus on landfill technology and liner leak detection: Journal of Cleaner Production, Vol. 225, pp. 1147–1154. Rashid, A. S. A.; Latifi, N.; Meehan, C. L.; and Manahiloh, K. N., 2017, Sustainable improvement of tropical residual soil using an environmentally friendly additive: Geotechnical and Geological Engineering, Vol. 35, pp. 2613–2623. Rico, R. A.; Gilles, B.; and Irini, D., 2011, Hydraulic conductivity determination of compacted sand-bentonite mixture using filter-press: Geotechnical Testing Journal, Vol. 34, No. 1, pp. 9–17. Rowe, R. K.; Quigley, R. M.; and Brachman, R. W. I., 2004, Barrier Systems for Waste Disposal Facilities, 2nd ed.: Spon Press, Florida, U.S., NY, 405 p. Rubinos, D. A. and Spagnoli, G., 2019, Assessment of red mud as sorptive landfill liner for the retention of arsenic (V): Journal of Environmental Management, Vol. 232, pp. 271–285. Shackelford, C. D.; Benson, C. H.; Katsumi, T.; Edil, T. B.; and Lin, L., 2000, Evaluating the hydraulic conductivity of GCLs permeated with non-standard liquids: Geotextiles and Geomembranes, Vol. 18, No. 2, pp. 133–161. Shackelford, C. D. and Sample-Lord, K. M., 2014, Hydraulic conductivity and compatibility of bentonite for hydraulic con-

366

tainment barriers. In Iskander, M.; Garlanger, J. E.; and Hussein, M. H. (Editors), From Soil Behavior Fundamentals to Innovations in Geotechnical Engineering: Honoring Roy E. Olson (GSP 233), Proceedings of Geo-Congress 2014: GeoInstitute of American Society of Civil Engineers, Atlanta, GA, pp. 370–387. Singh, H. and Huat, B. K. K., 2013, Formation and classification of tropical residual soils. In Huat, B. K. K.; Toll, D. G.; and Prasad, A. (Editors), Handbook of Tropical Residual Soil Engineering: CRC Press, Florida, U.S., pp. 23–62. Taha, M. R. and Kabir, M. H., 2005, Tropical residual soil as compacted soil liners: Environmental Geology, Vol. 47, pp. 375–381. Tan, B. K., 2004, Country case study: engineering geology of tropical residual soils in Malaysia. In Huat, B. B. K.; Gue, S. S.; and Faisal, H. A. (Editors) Tropical Residual Soils Engineering: Proceedings of the Symposium on Tropical Residual Soils Engineering (TRSE2004): Taylor & Francis Group, London, UK, Malaysia, pp. 237–244. Tong, S. and Shackelford, C. D., 2016, Standardized hydraulic conductivity testing of compacted sand-bentonite mixtures: Geotechnical Testing Journal, Vol. 39, No. 6, pp. 1015–1029. Townsend, T. G.; Powell, J.; Jain, P.; Xu, Q.; Tolaymat, T.; and Reinhart, D., 2015, Sustainable Practices for Landfill Design and Operation: Springer, New York, 483 p. Turan, N. G. and Ergun, O. N., 2009, Removal of Cu(II) from leachate using natural zeolite as a landfill liner material: Journal of Hazardous Materials, Vol. 167, pp. 696–700. United Kingdom Environment Agency (UKEA), 2014, LFE4Earthworks in Landfill Engineering: UKEA, Bristol, U.K. United States Environmental Protection Agency (USEPA), 1989, Requirements for Hazardous Waste Landfill Design. Construction and Closure: USEPA, Cincinnati, OH. VectorStock, n.d., Map of Malaysia vector image, photograph, available at https://www.vectorstock.com/royaltyfree-vector/map-malaysia-vector-1607161, viewed on 19 November 2019. Wu, H.; Wen, Q.; Hu, L.; Gong, M.; and Tang, Z., 2017, Feasibility study on the application of coal gangue as landfill liner material: Waste Management, Vol. 63, pp. 161–171. Yong, L. L.; Namal Jayasanka Perera, S. V. A. D.; Syamsir, A.; Emmanuel, E.; Paul, S. C.; and Anggraini, V., 2019, Stabilization of a residual soil using calcium and magnesium hydroxide nanoparticles: A quick precipitation method: Applied Sciences, Vol. 9, pp. 1–15. Zainol, N. A.; Aziz, H. A.; and Yusoff, M. S., 2012, Characterization of leachate from Kuala Sepetang and Kulim Landfills: A comparative study: Energy and Environment Research, Vol. 2, No. 2, pp. 45–52. Zhao, Q.; Choo, H.; Bhatt, A.; Burns, S. E.; and Bate, B., 2017, Review of the fundamental geochemical and physical behaviors of organoclays in barrier applications: Applied Clay Science, Vol. 142, pp. 2–20.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 353–366

Technical Note Installation of Utility Trench Wells Using Vacuum Techniques for Urban Groundwater Investigation NICK SCHMIDT Wood Environment and Infrastructure Solutions, 110 James St., Suite 301, St. Catharines, ON L2R 7E8, Canada MARTIN G. SHEPLEY* Martin Shepley Groundwater Inc., 121 Victoria Street, Dundas, ON, L9H 2C1, Canada

Key Terms: Utility Trench Well, Urban Groundwater

INTRODUCTION Sewer networks and their associated utility trenches may strongly influence the effects of urbanization on a groundwater system (Sharp et al., 2003; Garcia-Fresca and Sharp, 2005; and Shepley et al., 2020). Sewer construction methods commonly require placement of coarse granular material in the bedding of the utility trench. Such bedding may have a hydraulic conductivity several orders of magnitude higher than the surrounding geological materials, therefore representing a preferential flow zone (Sharp et al., 2003; GarciaFresca and Sharp, 2005; and Shepley et al., 2020). At the sewer elevation, the width of a utility trench is largely determined by the pipe diameter and the extent of the coarse granular material beneath the sewer does not greatly exceed the physical footprint of the sewer pipes. The placement of adequate monitoring wells in this granular material could be a key element in monitoring the effects of utility trenches on an urban groundwater system, including the assessment of sewer infiltration/exfiltration and the movement of water within the utility trench network (Shepley et al., 2020). Conventional drilling techniques (e.g., Ruda and Farrar, 2006) are generally not suitable because of the high risk of damaging the utilities during the drilling process. It is not well known that vacuum hole excavation, which considerably reduces the risk of utility damage, can be used as a monitoring well installation method. In this technical note, a simple methodology is outlined for installing utility trench wells (UTWs) using readily available industrial vacuum equipment.

*Corresponding author email: martin@msgroundwater.com

UTW INSTALLATION METHODOLOGY The placement of an effective well screen within the coarse granular material of the utility trench bedding requires excavation adjacent to the sewer edge with limited disturbance of the surrounding soils (Figure 1) while minimizing the risk of damage to utilities. The design width of the coarse granular bedding material may vary depending on the sewer diameter and the substrate geotechnical properties (see e.g., ASCE and WEF, 1992); in general, the edge of the bedding material will be within tens of centimeters of the sewer edge. Vacuum techniques are frequently used for daylighting subsurface utilities (e.g., ASTM, 2015) and for excavating holes to install poles and piling. Commercial vacuum daylighting contractors often use either a truck (vac truck) or trailer-mounted industrial vacuum system with a vacuum hose and debris tank for disposal. To improve penetration, a high-pressure airor water-jetting tool is used to erode the soil prior to the removal of the material by vacuum. Figure 2 shows an example of a hydrovac truck excavating a hole prior to well installation. In this example, the operator is using a water-jetting tool that consists of a rigid metal pipe connected via a flexible line to a pressurizing pump (alternatively, a compressor is used for air jetting). The vacuum hose comprises a rigid largediameter steel pipe connected to a flexible hose that connects to the vacuum system and the debris tank. Both the vacuum hose and the rigid jetting tool are extended /lengthened by connecting additional sections of hose/pipe as the hole is deepened. The hole size is controlled by the area eroded by the jetting tool and by the vacuum hose diameter (approximately 0.2 m in Figure 2). The base of a utility trench is commonly within 5 m of the surface; however, depth depends on topography and the required sewer slope. Penetration to 5-m depth is attainable for many soil types. The hydraulic method provides superior erosion power and depth penetration

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 367–369

367

Schmidt and Shepley

tonite seal to the surface for shallow holes (<5 m below ground surface) or a grout for deeper holes. DISCUSSION AND CONCLUDING REMARKS

Figure 1. Generalized cross section of a utility trench.

compared to the air method and works better for very cohesive soils such as compacted tills. Vacuum techniques can excavate holes in excess of 15-m depth given favorable ground conditions. On completion of the hole, a well can be constructed according to the regulations and standards of the jurisdiction in question. Typical monitoring well construction methods may be used, such as installation of a polyvinyl chloride 0.05m (2 in.) internal diameter screen/riser and a sand-ﬁlter pack with a ben-

The vacuum excavation method works best in cohesive soils where the hole diameter can be maintained, thus reducing the amount of well construction supplies (sand/bentonite/grout) needed to backfill and construct the well. In non-cohesive soils, excess material may be removed, resulting in unwanted hole size increases. This is likely to occur when the coarse granular bedding material of the utility trench is encountered. Care is needed to minimize the amount of material removed around the utility to limit settlement, which could damage infrastructure. The debris container size places practical and cost constraints on the number and depth of holes that can be excavated. Container filling may be an issue for non-cohesive soils if the hole diameter cannot be controlled. Using a water-based jetting method creates a slurry that adds to the material requiring disposal; containment and disposal of any in situ contamination will also need consideration. Removal of in situ water from permeable soils, including the permeable bedding material of the utility trench, will also increase the container fill rate. A disadvantage of vacuum hole excavation compared to conventional borehole drilling is the lack of a sampling method to determine soil stratigraphy, particularly at greater depths, where visual observations are not possible. However, although sampling is not practical, audial cues (gravel and coarser materials hitting the steel pipes of the vacuum hose make a clanging sound) can be used to decipher the difference between the coarse granular bedding material and the backfill above, which is usually less coarse grained. Complementary geologic and hydraulic data may be provided from conventionally drilled boreholes/wells in the geological materials adjacent to a utility trench. Shepley et al. (2020) presented results of an investigation that used the present vacuum methodology for UTW construction and showed the importance of using both conventional wells and UTWs for understanding the significance of groundwater interactions between the utility trench system and the surrounding geological materials. ACKNOWLEDGMENTS

Figure 2. Hole excavation using a hydrovac truck as a precursor to well installation.

368

We acknowledge Phil Neville who used this methodology to monitor the migration of contaminants along a sewer utility trench. Two anonymous reviewers are thanked for providing comments to improve this technical note.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 367–369

Installation of Utility Trench Wells

REFERENCES ASTM F3097-15, 2015, Standard Practice for Installation of an Outside Sewer Service Cleanout Through a Minimally Invasive Small Bore Vacuum Excavation: ASTM International, West Conshohocken, PA. American Society of Civil Engineers (ASCE) and Water Environment Federation (WEF), 1992, Design and Construction of Urban Storm Water Systems: ASCE Manuals and Reports of Engineering Practice No. 77, WEF Manual of Practice FD20, 724 p. Garcia-Fresca, B. and Sharp, J. M., 2005. Hydrogeologic considerations of urban development: Urban-induced recharge. In Ehlen, J.; Haneberg, W. C.; and Larson, R. A. (Editors), Humans as Geologic Agents: Reviews in Engineering Geology, Vol. XVI, Geological Society of America, Boulder, CO, pp. 123– 136. doi:10.1130/2005.4016(11)

Ruda, T. and Farrar, J., 2006, Environmental drilling for soil sampling, rock coring, borehole logging, and monitoring well installation. In Nielsen, D. M. (Editor), Practical Handbook of Environmental Site Characterization and Groundwater Monitoring, 2nd ed.: CRC Press, Boca Raton, FL, pp. 297–344. Sharp, J. M.; Krothe, J.; Mather, J. D.; Garcia-Fresca, B.; and Stewart, C. A., 2003, Eﬀects of urbanization on groundwater systems. In Heiken, G.; Fakundiny, R.; and Suter, J. (Editors), Earth Sciences in the City: A Reader: American Geophysical Union, Washington DC, pp. 257–278. Shepley, M. G.; Schmidt, N.; Senior, M. J.; Worthington, S. R. H.; and Scheckenberger, R. B., 2020, Assessing “urban karst” eﬀects from groundwater–storm sewer system interaction in a till aquitard: Groundwater, Vol. 58, No. 2, pp. 269– 277. doi:10.1111/gwat.12908

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 367–369

369

Book Review Timefulness: How Thinking Like a Geologist Can Help Save the World

(Marcia Bjornerud) Review by: Roy Van Arsdale University of Memphis, Department of Earth Sciences, Memphis, TN 38152

Dr. Bjornerud in her Timefulness: How Thinking Like a Geologist Can Help Save the World brings into focus a broad spectrum of Geological Sciences and how its varied parts impact our lives. We are not used to considering ourselves as part of Earth history, but that reality can no longer be ignored. In this beautifully written book, Dr. Bjornerud not only leads us through deep geologic time but also introduces us to the major scientists who led the way. The book starts with a well-founded attack on creationism’s young Earth mentality that impairs people’s time perception. She argues that we cannot appreciate the world’s beauty without knowledge of its past. Are we not all descendants of a long line of people? In similar manner, do not the rocks we walk on contain descendants of a long line of extinct ancestral plants and animals? Lack of geologic knowledge is like visiting modern Rome and being unaware of the history and architecture of the former Roman Empire. To understand the importance of geologic history, Dr. Bjornerud ﬁrst explains critical tools used to explore the past by visiting the inspiring unconformity at Siccar Point in Scotland and the foundations of radiometric dating leading to the determination of our 4.5-billion-year-old planet. The fundamentals of how Earth works is explained in her eloquent discussion of Plate Tectonics and landscape evolution controlled by mountain uplift and countering erosion. Just when a reader may be a bit confused, Dr. Bjornerud slips in a wonderful analogy that clariﬁes the subject. A particularly interesting discussion is the origin and evolution of our atmosphere that is intimately related to major periods of extinction like the end of Permian Period at

250 million years ago that resulted in the demise of 95% of species. Similarly, she discusses the Pleistocene ice ages that enveloped the Earth from approximately 2.6 million years to 11,700 years ago, which is recorded in sea floor cores and ice cores from Greenland and Antarctica. A major lesson to learn from the Pleistocene is the climatic swings. Particularly relevant is the stability of global climates during the Holocene that permitted agriculture and the successful expansion of humanity across the globe. How can we be so cavalier about global climate when we have a very long history of major geologic and climatic variability preserved and clearly observable in the rock and sediment beneath our feet? Although not the first to do so, Dr. Bjornerud concludes her book by expressing the need for a presidential cabinet minister whose job is to focus on the future of humanity and how the United States can attain healthy Earth stewardship. I fear I cannot give justice to this eloquent presentation of how important geologic knowledge is to an intellectually healthy society. I thoroughly enjoyed Timefulness and, more importantly, have been affected by this book. Timefulness: How Thinking Like a Geologist Can Help Save the World has pushed to the front of my mind the importance of looking after the environmental health of my grandchildren and their grandchildren. It is now my desire to find a way to influence people to be environmentally involved and what should be done to assure a healthy planet for our current and future well-being. Marcia Bjornerud, 2018, Timefulness: How Thinking Like a Geologist Can Help Save the World. Princeton University Press, Princeton, New Jersey, 224 p.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, p. 371

371

Book Review A Hero On Mount St. Helens: The Life and Legacy of David A. Johnston

(Melanie Holmes) Review by: Isaac E. Pope Centralia College, 600 Centralia College Boulevard, Centralia, WA 98570

There are few subjects that base their methodology on philosophy as heavily as the sciences, and fewer still that are based in ethics as engineering. Though all fields have had incidents of science gone awry where personal gain was paramount over moral and ethical considerations (such as the Bone Wars), few can tout the bravery recounted in Holmes’ A Hero On Mount St. Helens. Relating the events of the 1980 eruption of Mount St. Helens in southwest Washington, this book uncovers how the actions and ambitions of the young volcanologist David A. Johnston through his ultimate sacrifice helped push geoscience to aid more people. Based on discussions with Johnston’s sister Pat and many other interviews, this book provides an unparalleled examination of Dave Johnston’s career and motivations. Raised in a suburb of Chicago, Johnston was among the smallest and youngest in his age bracket and suffered from the stresses of the Cold War, but his tenaciousness drove him to excel in Boy Scouts and other community ventures. His compassionate and curious nature led Johnston to a desire to explore the sciences and help others, but a shock would come to his altruistic mindset when a tornado killed thirty-seven and injured over five hundred residents of Johnston’s hometown. A budding photographer at age seventeen, Johnston was directly affected as he documented the carnage for his mother’s newspaper. Though this event may have been one of the earliest reasons for his desire to combat natural disasters, Johnston would first major in photojournalism at University of Chicago until his love of geoscience caught up with him in an elective course. After interning with the USGS in 1972 under the guidance of Peter Lipman, Johnston enrolled in a PhD program at University of Washington on the San Juan Mountains volcanic field in southwestern Colorado, though he later switched his thesis topic to the ongoing eruptions of St. Augustine in Alaska. Finally graduating with his PhD in 1978, Johnston’s work on St. Augustine poised him as a burgeoning expert on volatiles in volcanic eruptions, which he continued as a USGS field geologist studying Alaskan volcanoes and as a consultant for geothermal energy on the Azores off the coast of Portugal. After successfully completing

his probationary time at USGS, Johnston was about to leave for his permanent USGS position in Alaska (scheduled for June 1, 1980) when something extraordinary happened. On March 27, 1980, the Seismology Lab at University of Washington detected earthquakes originating two kilometers beneath the northern slopes of Mount St. Helens. A small Quaternary cone along the Cascade Magmatic Arc, Mount St. Helens last erupted in 1857, but the most recent eruption from a Cascade Volcano was Lassen Peak between 1914 and 1917. Besides the volcanism from Lassen Peak at a time when the world focused on World War 1 in Europe, no eruptions had occurred in the continental US where modern volcanologists could observe. USGS scientists were soon on the pulse of the awakening giant, and Johnston was added to the team of researchers based on his experience with volcanic volatiles. Three steam explosions rocked the mountain in April, then silence. Though seismometers continued to sense earthquakes, Mount St. Helens remained surprisingly quiet except for the steady development of a cryptodome on the northern ﬂank of magma was forced into a side vent. Linear fractures soon developed in the developing bulge, which landslide geologist Barry Voight warned may cause a landslide extending to Spirit Lake several kilometers to the north. While USGS scientists debated the nature of a future eruption, Johnston was certain of one thing: an eruption was imminent, and it would be no ordinary eruption. Johnston had studied a report of a lateral eruption in Russia’s Kamchatka Peninsula and, though a lateral eruption had never been closely observed, Johnston attributed curious ash deposits along a ridge twenty km north of Mount St. Helens to a previous lateral blast; it was upon this ridge that Johnston was stationed the morning of May 18. Termed “Coldwater II”, this station was the second observation outpost to be situated for USGS scientists. Over twenty km north of Mount St. Helens, it was deemed by most geologists to be safe for taking measurements of the growing cryptodome and quantify volatiles, but Johnston suspected the site was hazardous. It was to his great relief that his young

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 373–374

373

Holmes and Pope

mentee Harry Glicken, who had been handling the outpost, would have to return to California to meet with his PhD thesis advisor Richard Fischer. Don Swanson was then assigned to the site, but the arrival of a foreign student in the area required Swanson’s absence until 9 am on Sunday, May 18. After helping Glicken pack for his trip to California on the evening of May 17, Johnston found himself alone at Coldwater II, tired but grateful no others were in danger. The next morning at 8:32 am, a 5.2 magnitude earthquake triggered the largest landslide in recorded history as the cryptodome slid into the valley below, releasing the tremendous pressure within the volcano. Johnston heralded the world with the announcement “Vancouver, Vancouver, this is it!” mere seconds before he himself was killed by the lateral blast. Once the ash ﬁnally settled nine hours later, 55 others had been killed and billions of dollars of property had been destroyed by ensuing volcanic mudﬂows.

374

Though the nation was shocked at the sudden calamity, the work of Johnston and others had greatly reduced the loss of life. The tedious hours of research, press conferences, and mapping hazard zones had raised awareness of the danger and prevented people from remaining close to the volcano, and Johnston had himself saved the lives of his fellow scientists through his own sacriﬁce, though he fully understood the danger. Glicken would also follow this example when he was killed by a pyroclastic ﬂow as he helped videographers raise awareness of volcanic hazards. Johnston’s life illustrates how a life devoted to good science and saving lives may impact inestimable people, even at the expense of his own life. As global population grows and geologic disasters impact more communities, how may we likewise make our world a safer, healthier world? Melania Holmes, 2019, A Hero on Mount St. Helens: The Life and Legacy of David A. Johnston. University of Illinois Press, Champaign, IL. 240 pages.

Environmental & Engineering Geoscience, Vol. XXVII, No. 3, August 2021, pp. 373–374

THIS PUBLICATION IS PRINTED ON ACID-FREE PAPER Abdul Shakoor Kent State University Kent, OH 44242 ashakoor@kent.edu

EDITORS

Eric Peterson Department of Geography, Geology, and the Environment Illinois State University Normal, IL 61790 309-438-5669 ewpeter@ilstu.edu

Karen E. Smith, Editorial Assistant, kesmith6@kent.edu

Oommen, Thomas Board Chair, Michigan Technological University Sasowsky, Ira D. University of Akron

Environmental & Engineering Geoscience August 2021 VOLUME XXVII, NUMBER 3

Cover photo View East at Brilliant Cut Rock Slope Failure, Pittsburgh, Allegheny County, Pennsylvania, March 20, 1941. Photo by the late A.C. Ackenheil. See article on page 269.

Volume XXVII, Number 3, August 2021

ADVISORY BOARD Watts, Chester “Skip” F. Radford University Hasan, Syed University of Missouri, Kansas City Nandi, Arpita East Tennessee State University

ENVIRONMENTAL & ENGINEERING GEOSCIENCE