Environmental & Engineering Geoscience AUGUST 2015
VOLUME XXI, NUMBER 3
THE JOINT PUBLICATION OF THE ASSOCIATION OF ENVIRONMENTAL AND ENGINEERING GEOLOGISTS AND THE GEOLOGICAL SOCIETY OF AMERICA SERVING PROFESSIONALS IN ENGINEERING GEOLOGY, ENVIRONMENTAL GEOLOGY, AND HYDROGEOLOGY
Environmental & Engineering Geoscience (ISSN 1078-7275) is published quarterly by the Association of Environmental and Engineering Geologists (AEG) and the Geological Society of America (GSA). Periodicals postage paid at AEG, 1100 Brandywine Blvd, Suite H, Zanesville, OH 43701-7301. Phone: 844-331-7867 and additional mailing offices. EDITORIAL OFFICE: Environmental & Engineering Geoscience journal, Department of Geology, Kent State University, Kent, OH 44242, U.S.A. phone: 330-672-2968, fax: 330-672-7949, ashakoor@kent.edu. CLAIMS: Claims for damaged or not received issues will be honored for 6 months from date of publication. AEG members should contact AEG, 1100 Brandywine Blvd, Suite H, Zanesville, OH 43701-7301. Phone: 844-331-7867. GSA members who are not members of AEG should contact the GSA Member Service center. All claims must be submitted in writing. POSTMASTER: Send address changes to AEG, 1100 Brandywine Blvd, Suite H, Zanesville, OH 43701-7301. Phone: 844331-7867. Include both old and new addresses, with ZIP code. Canada agreement number PM40063731. Return undeliverable Canadian addresses to Station A P.O. Box 54, Windsor, ON N9A 6J5 Email: returnsil@imexpb.com. DISCLAIMER NOTICE: Authors alone are responsible for views expressed in articles. 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 the journal as part of their AEG membership dues. Additional subscriptions may be ordered at $40 per year. GSA members who are not members of AEG may order for $40 per year on their annual GSA dues statement or by contacting GSA. Nonmember subscriptions are $175 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 303-757-2926; contact GSA Subscription Services, c/o American Institute of Physics, 2 Huntington Quadrangle, Suite 1N01, Melville, NY 11747-4502. Single copies are $50.00 each. Requests for single copies should be sent to AEG, 1100 Brandywine Blvd, Suite H, Zanesville, OH 43701-7301. Phone: 844-331-7867. © 2015 by the Association of Environmental and Engineering Geologists All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or by any information storage and retrieval system, without permission in writing from AEG. THIS PUBLICATION IS PRINTED ON ACID-FREE PAPER EDITORS ABDUL SHAKOOR Department of Geology Kent State University Kent, OH 44242 330-672-2968 ashakoor@kent.edu
BRIAN G. KATZ Florida Department of Environmental Protection 2600 Blair Stone Rd. Tallahassee, FL 32399 850-245-8233 eegeditorbkatz@gmail.com
EDITORIAL BOARD ROBERT H. SYDNOR JEROME V. DEGRAFF USDA Forest Service Consulant THOMAS J. BURBEY CHESTER F. WATTS (SKIP) Virginia Polytechnic Institute Radford University SYED E. HASAN University of Missouri, Kansas City ASSOCIATE EDITORS JOHN W. BELL PAUL M. SANTI Nevada Bureau of Mines and Colorado School of Mines Geology ROBERT L. SCHUSTER U.S. Geological Survey RICHARD E. JACKSON (Book Reviews Editor) ROY J. SHLEMON R. J. Shlemon Geofirma Engineering, Ltd. & Associates, Inc. JEFFREY R. KEATON AMEC Americas GREG M. STOCK National Park Service PAUL G. MARINOS National Technical University RESAT ULUSAY Hacettepe University, Turkey of Athens, Greece CHESTER F. “SKIP” WATTS JUNE E. MIRECKI U.S. Army Corps of Radford University Engineers TERRY R. WEST Purdue University PETER PEHME Waterloo Geophysics, Inc NICHOLAS PINTER Southern Illinois University SUBMISSION OF MANUSCRIPTS Environmental & Engineering Geoscience (E&EG), is a quarterly journal devoted to the publication of original papers that are of potential interest to hydrogeologists, environmental and engineering geologists, and geological engineers working in site selection, feasibility studies, investigations, design or construction of civil engineering projects or in waste management, groundwater, and related environmental fields. All papers are peer reviewed. The editors invite contributions concerning all aspects of environmental and engineering geology and related disciplines. Recent abstracts can be viewed under “Archive” at the web site, “http://eeg.geoscienceworld.org”. Articles that report on research, case histories and new methods, and book reviews are welcome. Discussion papers, which are critiques of printed articles and are technical in nature, may be published with replies from the original author(s). Discussion papers and replies should be concise. To submit a manuscript go to http://eeg.allentrack.net. If you have not used the system before, follow the link at the bottom of the page that says New users should register for an account. Choose your own login and password. Further instructions will be available upon logging into the system. Please carefully read the “Instructions for Authors”. Authors do not pay any charge for color figures that are essential to the manuscript. Manuscripts of fewer than 10 pages may be published as Technical Notes. For further information, you may contact Dr. Abdul Shakoor at the editorial office. Cover photo Landslide damaged home in eastern Kentucky. Cut-and-fill slope development, weak rocks, thick clayey soils, and complex groundwater activity were conditions that led to the rainfall triggered landslide. Photo courtesy of Dr. Matthew Crawford (see article on pages 181–195).
Environmental & Engineering Geoscience Volume 21, Number 3, August 2015 Table of Contents 165
Statistical Evaluation of Shoreline Change: A Case Study from Seabrook Island, South Carolina Briget C. Doyle and Madelyne R. Adams
181
Geologic, Geotechnical, and Geophysical Investigation of a Shallow Landslide, Eastern Kentucky Matthew M. Crawford, Junfeng Zhu, and Steven E. Webb
197
Influence of Geologic and Index Properties on Disintegration Behavior of Clay-Bearing Rocks Abdul Shakoor and Tej P. Gautam
211
Seismic Source Characterization for Greater Phoenix Area Earthquake Hazard Simon T. Ghanat, Edward Kavazanjian, Jr., and Ramon Arrowsmith
223
Monitoring Spatial-Temporal Change of Land Desertification in a Fragile Sub-Alpine Rangeland Eco-Environment: A Case Study from China Wei Xian, Zhiying Xiang, Liyang Liu, and Huaiyong Shao
235
Microfabrics-Based Approach to Predict Uniaxial Compressive Strength of Selected Amphibolites Schists Using Fuzzy Inference and Linear Multiple Regression Techniques Esamaldeen Ali, Wu Guang, and Abdelazim Ibrahim
247
Book Reviews Manual of Soil Laboratory Testing By K.H. Head and R.J. Epps Review by: Richard Jackson
Statistical Evaluation of Shoreline Change: A Case Study from Seabrook Island, South Carolina BRIGET C. DOYLE1 Division of Natural Sciences and Engineering, University of South Carolina Upstate, 800 University Way, Spartanburg, SC 29303
MADELYNE R. ADAMS Department of Geology and Environmental Geosciences, College of Charleston, 66 George St., Charleston, SC 29424
Key Terms: Barrier Island, Erosion, Accretion, DSAS, Linear Regression, Geographic Information Systems
ABSTRACT Seabrook Island is a barrier island approximately 28 km south of Charleston, South Carolina (SC). As a residential and resort area, Seabrook Island is important to the economic health of the Charleston, SC, region. Portions of Seabrook Island are impacted by accelerated erosion, which has been combated with engineered stabilization methods, inlet relocation, and beach nourishment. Despite stabilization attempts, localized erosion is continuing, leading to potential economic and ecological losses. We used the Digital Shoreline Analysis System (DSAS) to determine longand short-term erosion rates on the island, focusing on natural and anthropogenic changes to the beach. We assessed beach change by comparing historical aerial photographs and satellite imagery. The use of digital imagery allows for rapid assessment of the shoreline to highlight areas experiencing change, to monitor the impacts of structural controls, and to focus site-specific field investigations. Our results show that since 1939, Seabrook has experienced erosion up to 52.4 m/yr at the north end of the island near Captain Sams Inlet that is likely related to relocations of the inlet, accretion of up to 9.0 m/yr on Seabrook Beach because of nourishment of the beach, and relatively stable conditions near the south end of the island as a result of structural control and nourishments. This method of analysis is useful to support urban planning consistent with preservation of natural resources and thus may be suitable for application and testing elsewhere.
1
Corresponding author email: bdoyle@uscupstate.edu.
INTRODUCTION Many coastal areas with both economic and environmental importance are experiencing significant erosion. It has been estimated that more than 80% of the coastline of the United States is experiencing shoreline retreat or erosion, with the remaining 20% either stable or accreting (Pilkey and Dixon, 1996; Pilkey et al., 1998; Levine and Kaufman, 2008). The offshore, barrier, and sea islands of South Carolina show patterns of active erosion and accretion, with erosion currently dominating in most areas (Pilkey and Dixon, 1996; Pilkey et al., 1998; Harris et al., 2009). The impact of structural and non-structural erosion control methods on coastal erosion and accretion is a matter of much discussion and investigation. Structural control methods, while often effective at protecting property on-shore, may result in erosion and loss of the beach (National Research Council, 1990; Pilkey and Dixon, 1996; Levine et al., 2009). Beach nourishment can stabilize these losses, but frequent renourishment is usually required to maintain the beach (Levine et al., 2009). Because of the importance of coastal tourism on South Carolina barrier islands and beaches, a variety of erosion control methods have been implemented in coastal counties, including in Charleston County. Structural erosion control measures used to protect barrier islands in Charleston County include groins, seawalls, and inlet relocations. In addition, a series of beach nourishments costing over $351 million have been done in Charleston County (Kana, 2012). All of these measures have been undertaken on Seabrook Island, the site of our study. Shoreline change analysis has been a topic of study for decades (Shoshany and Degani, 1992; Thieler and Danforth, 1994a; Fenster and Dolan, 1996; Robertson et al., 2007), but only in recent years has the development of statistical shoreline analysis using digital imagery been used, significantly advancing studies of shoreline change, its causes, and its impacts around the world (BaMasoud and Byrne, 2011;
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
165
Doyle and Adams
Figure 1. Location map for Seabrook Island, South Carolina.
Haunani et al., 2012; Nebel et al., 2012; Virdis et al., 2012). Methods for shoreline change analysis on Seabrook Island have previously focused on repeat beach surveys along 35 profiles (Kana, 1989). These methods are time-intensive and require wide spacing of profiles because of time constraints. Digital methods allow for rapid assessment of the shoreline along closely spaced transects to highlight areas experiencing change, to monitor the impacts of structural controls, and to focus site-specific field investigations. In this paper, we use digital shoreline analysis to analyze rates of erosion and accretion on Seabrook Island, and determine the relative effectiveness and impacts of erosion control methods such as seawalls, inlet relocation, and beach nourishment. STUDY AREA Seabrook Island is a barrier island with an area of 18 km2 and a mean elevation of 3 m, located
166
approximately 28 km south of Charleston, South Carolina (Figure 1). It is a 3-km-long section of a classic drumstick-shaped barrier island complex consisting of Kiawah Island to the northeast and Seabrook Island to the southwest (Harris et al., 2005; Hayes and Michel, 2008). Seabrook Island is bordered by Captain Sams Inlet to the northeast and North Edisto Inlet to the southwest (Figure 1). A large emergent ebb-tidal delta (Deveaux Bank) is located off North Edisto Inlet, and a smaller ebb-tidal delta and ebb-tidal shoals are present off Captain Sams Inlet. Most of Seabrook Island’s sediment originates on Kiawah Island and Folly Island, and is transported to Seabrook Island via the northeast-southwest longshore drift. Sediment transport has been reduced by construction on Kiawah and Folly islands, especially following groin construction on Folly Island in the 1950s and 1960s (Levine et al., 2009). Seabrook Island has a population of approximately 1,300 and consists mainly of residential homes, low
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
Shoreline Change Table 1. History of beach nourishment and sand scraping on Seabrook Island, including location and volume of material. Dates 1983 1990 2001 2002 and 2003 12/2004–2/2005 2005 late 2006
Location1 Seabrook Beach Beach Club Seabrook Beach (just south of Captain Sams Inlet) Beach Club Beach Club Beach Club Beach Club
Type of Stabilization
Volume of Material (m3)
nourishment nourishment beach scrapings (movement of material from below HWL to above HWL) nourishment2 nourishment2 nourishment2 nourishment2
176,000 459,000 unknown volume (‘‘minor’’) 99,400 71,100 71,200 170,500
Source: Kana, 2008. See Figure 1 for locations. 2 Material for nourishment was removed from below the high tide line on Seabrook Beach by beach scraping. 1
density resort accommodations and amenities, and natural areas (both wooded and marshland). Largescale development of the island began in early 1970s and sporadic construction has continued to the present on a much smaller scale. Seabrook Island is of significant economic and environmental importance to the State of South Carolina. Data from 2008 indicates that Charleston County tourism contributed $1.6 billion to South Carolina tourism revenue and supported over 20,000 jobs in that year (SCDPRT, 2011). The bulk of these revenues and jobs are in the downtown and coastal communities, including the residential and resort areas of Seabrook Island. Barrier islands including Seabrook Island protect the mainland from the impacts of hurricanes and other storms. Not only do the island’s marshlands absorb some of the excess water generated in large storms, but the island acts as a ‘‘speed bump’’, attenuating waves and storm surges. The maritime forests on Seabrook Island take the brunt of the energy of damaging storm winds and weaken the storm before it reaches the mainland. Seabrook Island is also important because of the wildlife it supports. Several species that nest and breed on the islands are listed as threatened species by the State of South Carolina and the Federal government, including loggerhead sea turtles, piping plovers, bald eagles, least terns, and Wilsons plovers (USFWS, 2012). Protection of these nesting and breeding grounds is important to the future of the species. Stabilization History In the late 1970s and early 1980s, hard stabilization structures, including temporary sandbag revetments and groins, concrete sheetpile walls, and an extensive 700 m riprap sea wall, were installed at the south end of Seabrook Island along North Edisto Inlet, at a cost to property owners of approximately $3 million
dollars (Kana, 1989; Rinehart and Pompe, 1999; Hayes and Michel, 2008). The hard structures on Seabrook Island have slowed erosion at the south end of Seabrook Island. However, continuing erosion has required repeated beach nourishment to stabilize the beach (Kana, 2008). As public perception has increasingly focused on perceived negative impacts of hard structures on the beach, many states, including South Carolina, have restricted or prohibited construction of new hard structures. The prohibitions have led to an emphasis on soft stabilization to combat coastal erosion and shoreline retreat (National Research Council, 1990; Morton and Miller, 2005; Harris et al., 2009; Levine et al., 2009). Soft stabilization includes beach nourishment, when material is added to the beach, and beach scrapings, when material is removed from the intertidal zone and placed on the upper beach (Carley et al., 2010). Although soft stabilization is effective in the short-term, without changes to the coastal processes that led to the original erosion, these techniques are temporary and need to be repeated. Table 1 summarizes the history of beach nourishments and scrapings on Seabrook Island. In addition to structural control methods and beach nourishment, inlet relocation was attempted on the north end of Seabrook Island to control erosion. Captain Sams Inlet borders Seabrook Island to the north, separating it from Kiawah Island. Captain Sams Inlet was approximately 270 m wide in 2012. In the 1930s Captain Sams Inlet was located close to its present-day position, but scour caused by a major hurricane in the late 1940s cut through the spit on the south end of Kiawah Island, moving the inlet northeast and adding to the shoreline of Seabrook Island (Hayes and Michel, 2008). Since the hurricane, Captain Sams Inlet migrated south again, eroding into the northeast edge of Seabrook Island. In an attempt to mitigate erosion of Seabrook Island, Captain Sams Inlet has been moved to the north twice: in March 1983 and in April 1996. Both relocations involved construction of a new inlet near
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
167
Doyle and Adams
Figure 2. Seabrook Island AoIs for DSAS analysis.
the site of the 1940s breach, and closure of the existing inlet. Since the 1996 relocation, Captain Sams Inlet has migrated south again, eroding into the north end of Seabrook Island. Areas of Interest To focus our study of shoreline change, we divided Seabrook Island into five Areas of Interest (AoIs) (Figure 2). The AoIs are specific geographic regions of the island wherein the erosional history, control methods, and potential negative impacts are consistent across the region. We analyzed shoreline change rates in five AoIs on Seabrook Island described in detail below. Because of the long history of shoreline data (1939–2010) for Seabrook Island, we were able to analyze shoreline data using two methods. First, we analyzed shoreline change over long-term, using all data from 1939–2010. Second, we analyzed shoreline change over short-term intervals to focus
168
on impacts of erosion control on short-term shoreline stability (Kaufman, 2006; Nebel et al., 2012). For example, we analyzed shorelines prior to seawall construction at the Seabrook Beach Club, and then following seawall construction to attempt to identify impacts of the seawall on the beach. Captain Sams Beach and Captain Sams East AoIs Captain Sams Inlet (Figure 2) has experienced the most erosion, as well as both the greatest natural and anthropogenic modifications. It is also the location of habitat for the endangered loggerhead sea turtle, foraging habitat for the threatened piping plover, and the site of frequent dolphin strand feedings (USFWS, 2012). Analysis of the change of the Captain Sams shoreline will provide information to planners, homeowners, the Seabrook Island Property Owners Association, and others concerned with a potential relocation of Captain Sams Inlet.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
Shoreline Change
Figure 3. Maps showing a section of Seabrook Beach on scanned and georeferenced aerial photographs, and the shoreline digitized along the HWL. (A) 1954 aerial photograph and digitized shoreline. (B) 1973 aerial photograph and digitized shoreline. Differences in resolution of the aerial photographs are clearly visible. Resolution was one factor used in determining uncertainties used in DSAS calculations.
The main movement of Captain Sams Inlet, both naturally and following human manipulation, is not inland or seaward, but rather parallel to the shore. Accordingly, to analyze migration direction, we divided the Captain Sams Inlet area into two separate AoIs designated as Captain Sams Beach AoI and Captain Sams East AoI, respectively (Figure 2). The Captain Sams Beach AoI transects are perpendicular to the main NE–SW shoreline trend to calculate island retreat, and the Captain Sams East AoI transects are perpendicular to the mouth of the inlet to determine rates of inlet migration into Seabrook Island. Using these transects, we determined the migration rates and directions of both Captain Sams Beach AoI and Captain Sams East AoI for three periods: 1939–2010, 1949–1979, and 1999–2010. The 1939–2010 period records total movement and rate of change. The 1949–1979 period covers the time from
the inlet’s natural relocation by a hurricane in the late 1940s to just before the first anthropogenic inlet relocation in 1983. No data were available to analyze changes during the period between anthropogenic inlet relocations. The final period, 1999–2010, covers the time from the second inlet relocation in 1996 to the present. The rate-of-change data are for these AoIs and are useful for assessing the possible need for a third inlet relocation. Seabrook Beach AoI The Seabrook Beach AoI is the Atlantic Coast beach of Seabrook Island from Captain Sams Inlet to Renken Point (Figure 2). This approximately 1.5-kmlong section of the island is the longest recreational beach on Seabrook Island. The Seabrook Beach AoI fronts the majority of beach-front homes on the
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
169
Doyle and Adams
Figure 4. Section of Seabrook Beach showing digitized shorelines used in DSAS overlain on the 2010 Digital Globe imagery.
island. The Seabrook Beach AoI is 0.6 km south of the Captain Sams Beach AoI, and is beyond the influence of migration and relocation of Captain Sams Inlet. Like the spit at Captain Sams Inlet, Seabrook Beach is also a nesting habitat for the loggerhead sea turtle and foraging habitat for the piping plover (USFWS, 2012). We analyzed the Seabrook Beach AoI over four periods. The 1939–2010 period allowed assessment of the beach’s long-term rate of change. The other periods allowed us to focus on the impacts of anthropogenic modification of the beach, including nourishments, scraping, and the relocations of Captain Sams Inlet. Specifically, the 1939–1979 period provides information on shoreline characteristics before the first nourishment of the beach; the 1984–2010 period allowed us to analyze change on the beach following the first relocation of Captain Sams Inlet and nourishment of the beach; and the 1999–2010 period included times of beach scraping to remove sand from
170
below the high water line (HWL) on Seabrook Beach for use in nourishment projects father south on the island. Beach Club Beach and Beach Club Seawall AoIs The Beach Club region of beach is 1.1 km long and extends from Renken Point at the south end of Seabrook Beach to the south tip of Seabrook Island at the entrance to North Edisto Inlet (Figure 2). Eleven homes front this section of beach and are located 23–57 m from the 2010 HWL. Other development along this section of beach include: the Seabrook Island Beach Club whose closest point sits 8.5 m from the 2010 HWL; Seabrook Island Road, which is only 25 m from the 2010 HWL at its closest point; and a restaurant and golf club (The Island House) inland of Seabrook Island Road. The Beach Club, Island House, and part of Seabrook Island Road lie behind a rip-rap seawall/revetment constructed in
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
Shoreline Change
Figure 5. Section of Seabrook Beach showing transects used in DSAS calculations overlain on the 2010 Digital Globe imagery. Numbers are transect identification numbers used in DSAS. Figure 5 shows the same location as Figure 4.
1974–1975 in response to beach erosion in 1974 (Hayes and Michel, 2008). We divided the Beach Club region into two AoIs: a northeast portion of the beach (Beach Club Beach AoI) and a southeast portion in front of the seawall (Beach Club Seawall AoI) (Figure 2). The Beach Club Beach AoI was analyzed for three periods: the 1939– 2010 period provides a long-term rate of change for this section of beach; 1939–1984 period predates the first nourishment of the beach in 1990; and the final period, 1999–2010 spans the time from after the first beach nourishment to the present and includes six renourishment projects conducted from 2002–2007 (Table 1). We similarly analyzed beach changes affecting the Beach Club Seawall AoI for the long-term rate-ofchange from 1939–2010; change before the seawall construction as recorded in the 1939–1973 data; and the influence of both the seawall construction and seven beach nourishment projects from the 1999–2010 data.
METHODOLOGY Several methods have traditionally been used to assess shoreline change, including topographic maps, aerial photographs, remote sensing imagery, and ground surveys. Early methods typically relied on measurement with rulers on uncorrected maps and aerial photographs. More recently, Geographic Information Systems (GIS) have been used for shoreline change analysis. One of the more commonly used methods involves calculation of rate-of-change statistics using the GIS-based Digital Shoreline Analysis System (DSAS) of Thieler and Danforth (1994b). DSAS was developed to compute rate-of-change statistics from historic shoreline positions using GIS (Thieler et al., 2009). DSAS is both a user-friendly and powerful tool that has been widely used world-wide (Pires et al., 2009). We completed a statistical analysis of shoreline rate of change on Seabrook Island using DSAS. We
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
171
Doyle and Adams Table 2. Erosion rates for Seabrook Island Areas of Interest. The table shows the calculated LRR and WLR, LSE and WSE, as well as confidence intervals of the rates at 68.3% (LCI683, WCI683). Negative values indicate erosion or shoreline retreat, while positive values indicate accretion. LRR (m/yr)
LCI683
LSE
WLR (m/yr)
WCI683
WSE
Shoreline Change Status
Captain Sams East 1939–2010 1949–1979 1999–2010
28.2 252.4 248.6
8.2 15.4 23.5
567.0 327.0 101.8
29.7 245.9 254.4
5.5 15.7 34.2
184.9 112.4 77.5
very likely eroding very likely eroding very likely eroding
Captain Sams Beach 1939–2010 1949–1979 1999–2010
22.0 212.7 214.7
2.6 5.0 17.9
178.9 106.3 77.8
23.2 211.8 218.8
2.0 4.6 26.1
69.7 33.0 59.2
likely eroding very likely eroding likely eroding
Seabrook Beach 1939–2010 1939–1979 1984–2010 1999–2010
4.5 1.8 9.0 22.7
1.0 1.5 4.8 2.1
69.7 46.3 72.1 9.0
4.8 1.5 8.1 23.6
0.9 1.7 4.1 3.0
29.3 15.3 37.2 6.9
very likely accreting likely accreting very likely accreting very likely eroding
Beach Club Beach 1939–2010 1939–1984 1999–2010
20.3 20.8 1.1
0.5 1.2 1.0
37.7 45.9 4.2
0.2 20.5 1.1
0.3 0.9 1.4
11.3 13.2 3.2
indeterminate likely eroding likely accreting
Beach Club Seawall 1939–2010 1939–1973 1999–2010
0.6 4.0 20.7
0.9 3.8 0.8
63.3 84.9 3.6
0.7 4.4 20.8
0.6 4.1 1.2
18.8 24.7 2.8
likely accreting very likely accreting likely eroding
identified, recorded, and entered the locations of historic shorelines on the island compiled from aerial photography and satellite imagery and dating back to 1939 into a GIS database. Using the DSAS program, we identified rates of shoreline change and located specific areas of vulnerability to erosion over several periods with a focus on determining impacts of anthropogenic changes to the beach. Data Sources The most common data sources for shoreline change analysis are maps, aerial photographs, satellite imagery (Dolan et al., 1991; Thieler and Danforth, 1994a; Virdis et al., 2012). Aerial photographs are the most common source for determining historic shoreline positions (Moore, 2000; Boak and Turner, 2005). Although there are limitations to the use of aerial photography, including distortions and displacements in the imagery (Dolan et al., 1979; Crowell et al., 1991; Shoshany and Degani, 1992; Fletcher et al., 2003), photographs can be orthorectified in a GIS, allowing their use in shoreline analysis (Crowell et al., 1991; Kaufman, 2006; Levine and Kaufman, 2008; Virdis et al., 2012). We obtained shoreline data from aerial photographs, Digital Orthophotographic Quarter Quadrangles (DOQQs), and satellite imagery. The aerial photographs of the coastline were printed large-
172
format mosaiced near-vertical aerial photographs and orthophotographs from 1939, 1949, 1954, 1963, 1973, and 1979, collected by the Natural Resources Conservation Service (NRCS) and provided by the College of Charleston Addlestone Library Special Collections. Digital data included 1984 National High Altitude Photography (NHAP) imagery, 1999 DOQQ from the U.S. Geological Survey (USGS), and 2006 and 2010 Digital Globe imagery from Esri. DSAS requires a specific day, month, and year for each shoreline. When the exact date for imagery was not known, a default date of January 1 was entered for that year. Methods of Analysis We scanned aerial photographs as high resolution TIFF files using a large-format distortion free scanner, imported the files into ArcGIS, and rectified the images to Universal Transverse Mercator (UTM) Zone 17N using a North American Datum (NAD)83 projection to ensure consistent projection in the GIS. Rectification of the photos was completed within the GIS. To provide consistency, 2010 Digital Globe imagery with a resolution of 0.3 m was used as a base for all orthorectification. We incorporated uncertainties in scale and distortion along with the root mean square errors calculated during rectification into the error measurements used
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
Shoreline Change
Figure 6. Map of Captain Sams East AoI showing long-term (1939–2010) WLRs along each transect. Only the 1939 and 2010 shorelines are shown, however all ten mapped shorelines were used to determine WLR. The length of the transect shows the net movement along the shoreline from 1939–2010. The transect color reflects the rate of change along the transect.
by DSAS during analysis. Location uncertainties of aerial photographs range from 3.5–4 m for older photos to 1–2.5 m for more recent photos. Aerial photographic and satellite imagery errors were identified and corrected based on the georectifying and digitizing procedures given in Kaufman (2006), Levine and Kaufman (2008), and Levine et al. (2009). After rectification, the shorelines were digitized using the HWL. The HWL shows the landward extent of the last high tide on the beach (Crowell et al., 1991). The difference in tone between the wet and dry sand is easily recognizable on aerial photographs and satellite imagery (Shoshany and Degani, 1992; Moore, 2000; Leatherman, 2003). As an example, Figure 3 shows the HWL as digitized on 1954 and 1973 aerial images. The digitized shorelines represent the average seasonal shoreline position (Thieler and Danforth, 1994a; Levine and Kaufman, 2008). The
intervals between successive images in our study were closer to decadal than annual. Therefore, lack of information about tidal heights is generally considered to be minor compared to the long-term changes observed in the imagery (Kaufman, 2006; Levine and Kaufman, 2008). Figure 4 is an example of the shorelines as digitized for a section of the Seabrook Beach AoI. Digital Shoreline Analysis System Once all shorelines were digitized, we analyzed the data using DSAS (Thieler, et al., 2009). We first created a baseline offshore of the island and projected a series of equally spaced transect lines from the baseline. Transect lines were spaced every 50 m, with a transect length of 1500 m to ensure that all mapped historic shorelines were crossed. We created a total of
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
173
Doyle and Adams
Figure 7. Map of Captain Sams Beach AoI showing long-term (1939–2010) WLRs along each transect. Only the 1939 and 2010 shorelines are shown, however all ten mapped shorelines were used to determine WLR. Where shorelines intersect transects more than once, only the first shoreline intersection is used. The length of the transect shows the net movement along the shoreline from 1939–2010. The transect color reflects the rate of change along the transect.
150 transects to cover the five AoIs. Figure 5 shows the baseline and numbered transects for the Seabrook Beach AoI (compare with Figure 4). DSAS recorded the locations where the individual transects cross the digitized historic shorelines and provided measurements of erosion and accretion around the islands. We then used DSAS to perform statistical analyses on each AoI on Seabrook Island. DSAS can perform multiple analyses using different segments of shoreline and different sets of shoreline dates. The multiple analyses allowed us to analyze shoreline rates of change between significant natural or anthropogenic events as described above. We determined amounts and rates of shoreline erosion and accretion in the AoIs using three statistical analysis methods in DSAS: End Point Rate (EPR), Linear Regression Rate (LRR), and Weighted
174
Linear Regression Rate (WLR). EPR calculates the rate of change along a particular transect by dividing the distance between the oldest and youngest shoreline by the time between the shorelines; LRR determines rate of change by fitting a least-squares regression line to all shorelines on a transect; and WLR uses the provided uncertainty data to weight the shorelines to determine a best-fit linear regression (Thieler, et al., 2009). Following initial data runs, we determined that the most reasonable and statistically significant results were provided by the LRR and WLR data. We discarded EPR because it only takes into account only the earliest and latest shorelines and neglects intermediate shorelines. Consequently, EPR might miss changes in the sign and magnitude of shoreline movement, and cyclical changes might be neglected (Thieler and Danforth, 1994b; Pires et al.,
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
Shoreline Change
Figure 8. Map of Seabrook Beach AoI showing long-term (1939–2010) WLRs along each transect. Only the 1939 and 2010 shorelines are shown, however all 10 mapped shorelines were used to determine WLR. The length of the transect shows the net movement along the shoreline from 1939–2010. The transect color reflects the rate of change along the transect.
2009). We used LRR because it is commonly used as a robust statistical method and can be readily compared to other studies (Morton and Miller, 2005; Harris et al., 2009). LRR, however, can be biased by outlier effects, especially on shorelines that experience significant change between imagery (Dolan et al., 1991; Thieler et al., 2009). We added WLR because it removes some of the outlying data and gives weight to more reliable data based on the provided uncertainty values (Thieler et al., 2009). RESULTS Our DSAS results show the impact of structural and non-structural erosion controls on Seabrook Island. Accretion occurred where there was beach renourishment but no structural control; shoreline stability was common where both nourishment and
structural controls were emplaced; and erosion took place where there was only minor engineering control and little to no nourishment. We summarize our findings in Table 2. We used the standard error (SE) to forecast relative shoreline stability. A higher SE indicates greater uncertainty in shoreline change estimates. In contrast, low LLR SE (LSE) and WLR SE (WSE) values give higher confidence to estimated shoreline change (Harris et al., 2009). Because shoreline geometry in most of the AoIs showed high variability between individual mapped shorelines, we used LLR and WLR to cross-validate the average rate-of-change of the shoreline, as done by Haunani et al. (2012). If LRR and WLR have the same sign and both are statistically significant, the shoreline is very likely experiencing erosion or accretion (with positive values indicating accretion and negative values indicating erosion). If the rates
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
175
Doyle and Adams
Figure 9. Map of Beach Club Beach AoI showing long-term (1939–2010) WLRs along each transect. Only the 1939 and 2010 shorelines are shown, however all 10 mapped shorelines were used to determine WLR. The length of the transect shows the net movement along the shoreline from 1939–2010. The transect color reflects the rate of change along the transect.
have the same sign but one or both are not statistically significant, the shoreline is likely experiencing erosion or accretion. With this method, we can still use rates that are not significant to indicate the general erosional or accretional trend of the shoreline. We found the highest erosion rates, up to 52.4 m/yr at the north end of the island near Captain Sams Inlet, take place where there is inlet migration and the loss of ebb-tidal delta shoals. The high standard error (Table 2) results from extensive natural and anthropogenic modification of the inlet. Figure 6 shows the individual transects and long-term shoreline change rate (1939–2010) at Captain Sams East AoI. At current rates as high as 254.4 m/yr, inlet migration could impact homes within 8–10 years. Figure 7 shows the individual transects and longterm rates (1939–2010) at Captain Sams Beach AoI. As seen in Table 2, Captain Sams Beach AoI shows
176
much smaller rates of change than Captain Sams East AoI, therefore the main migration of Captain Sams Inlet is parallel to the shore. The short-term erosion rates at both Captain Sams East and Captain Sams Beach AoIs, however, are still higher than short-term rates in the areas to the south. As part of the 1996 inlet relocation, the shoreline of Captain Sams Inlet was moved approximately 15 m seaward (Kana, 2008). We infer that this artificial expansion of the beach is the likely cause for the low WLR rate-ofchange for 1939–2010 (,3.2 m/yr) at Captain Sams Beach; whereas the short-term rates are higher (11.8– 18.8 m/yr). DSAS analysis shows that Captain Sams Beach AoI was very likely eroding between 1949 and 1979, prior to the first anthropogenic inlet relocation in 1983, but following the natural relocation of the inlet in the late 1940s by a hurricane discussed above. This
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
Shoreline Change
Figure 10. Map of Beach Club Seawall AoI showing WLRs along each transect. Only the 1939 and 2010 shorelines are shown, however all 10 mapped shorelines were used to determine WLR. The length of the transect shows the net movement along the shoreline from 1939–2010. The transect color reflects the rate of change along the transect.
result is reasonable, as inlet relocation by the hurricane destroyed the protective ebb-tidal delta, exposing the beach to increased wave action. The most recent rates (1999–2010), although not statistically significant (Table 2), suggest that Captain Sams Beach AoI is likely continuing to erode at rates similar to those before the anthropogenic inlet relocations. The lack of statistical significance in the 1999–2010 period may relate to scrapings done to grade and shape the beach in 2001 (Kana, 2008) (Table 1). Seabrook Beach AoI has been accretional longterm (1939–2010), as well as in the short-term periods prior to nourishment (1939–1979) and following the 1983 Captain Sams Inlet relocation and nourishments (1984–2010) (Table 2). A short-term erosion period (1999–2010) is inferred to be the result of beach scrapings to remove sand from below the HWL in
2001 (Table 1). Figure 8 shows the individual transects and long-term shoreline change rates at Seabrook Beach AoI. The front beach portion of Seabrook Island exhibits a much higher level of statistical significance than the other AoIs over all periods (Table 2). The high significance is because Seabrook Beach AoI is much straighter and exhibited less variability in shoreline location and shape between mapping periods than the other AoIs. The relatively high rate of accretion from 1984 to 2010 is likely related to the relocation of Captain Sams Inlet in 1983. At that time, the ebb-tidal delta offshore of the former inlet site moved onshore and was attached to the island along the upper portion of Seabrook Beach, contributing to accretion of 9.0 m/yr. However, our results indicate that Seabrook Beach AoI, which has historically experienced accretion, was eroding from 1999 to 2010 at rates of 23.6 m/yr. This period
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
177
Doyle and Adams
includes instances of beach scraping to remove sand from below the HWL on Seabrook Beach for use in nourishment projects father south on the island. A total of 412,200 m3 sand was removed between 2002 and 2006 (Table 1; Kana, 2008). Although the scrapings took place below the HWL in an effort to avoid impacts to the beach, the timing indicates that removal of the sand contributed to Seabrook Beach changing from an accretional to an erosional beach. The Beach Club Beach and Beach Club Seawall AoIs, which have a history of engineered erosion control, including construction of a seawall in 1974 and repeated beach nourishment from 2002 to 2007, remained relatively stable with only minor losses. Figures 9 and 10 show the long-term shoreline changes along transects for both Beach Club AoIs. The DSAS data for both AoIs lack statistically significant changes in some periods (Table 2), because of the relative stability of the shoreline in this area. Rates of change calculated in DSAS for the Beach Club AoIs are below the resolution of many of the aerial photographs, and below the uncertainties we used in the DSAS calculations. The standard errors, however, indicate stable shorelines for these AoIs, and the small values of change, although not significant, indicate stability of the shoreline, both at the seawall and at Beach Club Beach AoI, which has no structural control. One set of rates at the Beach Club Seawall AoI is statistically significant. We analyzed the 1939–1973 rate-of-change data, which predate construction of the seawall, and found that beach the beach was very likely accreting at an average of about 4.4 mm/yr. We also found that where the shoreline had been stable in the past, it experienced erosion after seawall construction. The current HWL is on the seawall itself, rather than seaward of the structure. Accordingly, beach width is locally diminished and almost non-existent except during extremely low tides. It appears that following construction of the seawall, erosion of the seawall would have taken place were it not for placement of about 412,200 m3 of sand along the Beach Club Beach and Beach Club Seawall between 2001 and 2007. This result indicates that, although the shoreline is generally stable, the area in front of the seawall continues to lose sediment despite nourishments. DISCUSSION Our use of DSAS to analyze shoreline rates of change is helpful in determining the effectiveness and impacts of the erosion control methods used on Seabrook Island. Our results indicate that Captain Sams Inlet, which has been subject to repeated inlet
178
relocations, but little beach nourishment or hard structural control, continues to experience severe erosion. Current efforts to obtain permits to relocate the inlet a third time are stalled in the court system. A lawsuit has been filed to prevent relocation of Captain Sams Inlet based on aesthetic and environmental grounds (Peterson, 2013). Areas with no hard erosion control structures are either stable (Beach Club Beach AoI) or experience long-term accretion (Seabrook Beach AoI). Our results show that accretion is related to nourishment projects on Seabrook Beach and the Beach Club, and attachment of ebb-tidal shoals onto Seabrook Beach as a result of relocation of Captain Sams Inlet. Recent short-term erosion (1999–2010) at Seabrook Beach is likely related to removal of sand for nourishment projects at the Beach Club. The erosional impact on Seabrook Beach should be considered if future nourishment projects at the Beach Club again use Seabrook Beach as a sand source. The seawall, coupled with beach nourishment, has kept the Beach Club Seawall area relatively stable with periods of minor erosion. Despite public perception that hard structures lead to accelerated erosion, our results indicate that, when coupled with nourishment, the seawall on Seabrook Island has helped to stabilize the shoreline. However, the island will likely experience continued erosion of the shoreline if nourishment is not repeated.
CONCLUSIONS Our case study of shoreline erosion at Seabrook Island is based on analysis of historical aerial photographs and satellite imagery, coupled with statistical techniques. We attempt to quantify, and ultimately forecast, local shoreline erosion and accretion rates using a GIS-based Digital Shoreline Analysis System, a method that is relatively rapid and simple to use. Although some rates lack statistical significance, especially those in areas of frequent modification to the beach, we believe that these analytical methods can be used to better forecast the rate and magnitude of future shoreline change. Understanding the rates of erosion on Seabrook Island, the impacts of erosion control methods, and the time frame over which those processes occur will assist interested parties, including the State of South Carolina, Charleston County, the South Carolina Department of Natural Resources (SCDNR), USGS, the Environmental Protection Agency (EPA), and USFWS to develop future erosion control plans for the island.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
Shoreline Change
ACKNOWLEDGMENTS Research presented in this paper was supported by the University of South Carolina RISE Program and the College of Charleston Office of Undergraduate Research and Creative Activities. We would also like to thank the following for invaluable assistance and information: Addlestone Library Special Collections, University of South Carolina Upstate Division of Natural Sciences and Engineering, College of Charleston Department of Geology and Environmental Geosciences, Mr. Charlie Kaufman, Mr. John Wells of the Seabrook Island Property Owners Association, and the reviewers who commented on this manuscript. REFERENCES BAMASOUD, A. AND BYRNE, M., 2011, Analysis of shoreline changes (1959–2004) in Point Pelee National Park, Canada: Journal Coastal Research, Vol. 27, No. 5, pp. 839–846. BOAK, E. H. AND TURNER, I. L., 2005, Shoreline definition and detection: A review: Journal Coastal Research, Vol. 21, pp. 688–703. CARLEY, J. T.; SHAND, T. D.; COGHLAN, I. R.; BLACKA, M. J.; COX, R. J.; LITTMAN, A.; FITZGIBBON, B.; MCLEAN, G.; AND WATSON, P., 2010, Beach Scraping as a Coastal Management Option: 19th NSW Coastal Conference 2010, Batemans Bay, New South Wales, Australia, 20 p. Electronic document, available at http://www.coastalconference.com/papers.asp CROWELL, M.; LEATHERMAN, S. P.; AND BUCKLEY, M. K., 1991, Historical shoreline change—Error analysis and mapping accuracy: Journal Coastal Research, Vol. 7, pp. 839–852. DOLAN, R.; HAYDEN, B.; REA, C.; AND HEYWOOD, J., 1979, Shoreline erosion rates along the middle Atlantic coast of the United States: Geology, Vol. 7, pp. 602–606. DOLAN, R.; FENSTER, M. S.; AND HOLME, S. J., 1991, Temporal analysis of shoreline recession and accretion: Journal Coastal Research, Vol. 7, No. 3, pp. 723–744. FENSTER, M. AND DOLAN, R., 1996, Assessing the impact of tidal inlets on adjacent barrier island shorelines, Journal Coastal Research, Vol. 12, No. 1, pp. 294–310. FLETCHER, C. H.; ROONEY, J. J. B.; BARBEE, M.; LIM, S. C.; AND RICHMOND, B. M., 2003, Mapping shoreline change using digital orthophotogrammetry on Maui, Hawaii, Journal Coastal Research, Special Issue No. 38, Shoreline Mapping and Change Analysis: Technical Considerations and Management Implications, pp. 106–124. HARRIS, M. S.; GAYES, P. T.; KINDINGER, J. L.; FLOCKS, J. G.; KRANTS, D. E.; AND DONOVAN, P., 2005, Quaternary geomorphology and modern coastal development in response to an inherent geologic framework: An example from Charleston, South Carolina: Journal Coastal Research, Vol. 21, No. 1, pp. 42–43, 49–64. HARRIS, M. S.; WRIGHT, E. E.; FUQUA, L.; AND TINKER, T. P., 2009, Comparison of shoreline erosion rates derived from multiple data types: Data compilation for legislated setback lines in South Carolina (USA). In Pereira da Silva, C. (Editor), Special Issue No. 56. Proceedings 10th International Coastal Symposium (ICS 2009), Vol. II: Coastal Education and Research Foundation and Centro de Estudos de Geografia e Planeamento Regional, Boca Raton, FL, pp. 1224–1228. HAUNANI, H. K.; FLETCHER, C. H.; ROMINE, B. M.; ANDERSON, T. R.; FRAZER, N. L.; AND BARBEE, M. M., 2012, Vulnera-
bility assessment of Hawai’i’s cultural assets attributable to erosion using shoreline trend analysis techniques, Journal Coastal Research, Vol. 28, No. 3, pp. 533–539. HAYES, M. O. AND MICHEL, J., 2008, A Coast for All Seasons: A Naturalist’s Guide to the Coast of South Carolina: Pandion Books, Columbia, SC, 285 p. KANA, T. W., 1989, Erosion and beach restoration at Seabrook Island, South Carolina: Shore Beach, Vol. 57, No. 3, pp. 3–18. KANA, T. W., 2008, Seabrook Island, South Carolina, Captain Sam’s Inlet Relocation Project, Survey Report Number 12: Coastal Science and Engineering, Columbia, SC, 52 p. KANA, T. W., 2012, A brief history of beach nourishment in South Carolina, 2012: Shore Beach, Vol. 80, No. 4, pp. 9–21. KAUFMAN, C. C., 2006, Assessment of Beach Renourishment and Other Shoreline Stabilization Projects in Charleston Co., South Carolina: Unpublished MS Thesis, College of Charleston, Charleston, SC, 90 p. LEATHERMAN, S. P., 2003, Shoreline change mapping and management along the U.S. east coast: Journal Coastal Research, Special Issue No. 38. Shoreline Mapping and Change Analysis: Technical Considerations and Management Implications, pp. 5–13. LEVINE, N. S. AND KAUFMAN, C. C., 2008, Land use, erosion and habitat mapping on an Atlantic barrier island, Sullivan’s Island, South Carolina: Journal Maps, Vol. 2008, pp. 161–174. LEVINE, N.; KAUFMAN, C.; KATUNA, M.; HARRIS, S.; AND COLGAN, M., 2009, Folly Beach, South Carolina: An endangered barrier island. In Kelley, J. T.; Pilkey, O. H.; and Cooper, J. A. G. (Editors), America’s Most Vulnerable Coastal Communities: Geological Society of America Special Paper 460, pp. 91–110. MOORE, L. J., 2000, Shoreline mapping techniques: Journal Coastal Research, Vol. 16, pp. 111–124. MORTON, R. A. AND MILLER, T. L., 2005, National Assessment of Shoreline Change—Part 2, Historical Shoreline Changes and Associated Coastal Land Loss along the U.S. Southeast Atlantic Coast: U.S. Geological Survey Open-file Report 2005-1401, 35 p. NATIONAL RESEARCH COUNCIL, 1990, Managing Coastal Erosion, National Academy Press, Washington D.C., 182 p. NEBEL, S. H.; TREMBANIS, A. C.; AND BARBER, D. C., 2012, Shoreline analysis and barrier island dynamics: Decadal scale patterns from Cedar Island, Virginia: Journal Coastal Research, Vol. 28, No. 2, pp. 332–341. PETERSON, B., 2013, A line in a lot of sand; Seabrook owner disputes re-cutting Capt. Sam’s Inlet, Post & Courier (Charleston, SC): Electronic document, available at http://www. postandcourier.com/article/20130204/PC16/130209729# content PILKEY, O. H. AND DIXON, K. L., 1996, The Corps and the Shore: Island Press, Washington, D.C., 272 p. PILKEY, O. H.; NEAL, W. J.; RIGGS, S. R.; WEBB, C. A.; BUSH, D. M.; PILKEY, D. F.; BULLOCK, J.; AND COWAN, B. A., 1998, The North Carolina Shore and Its Barrier Islands: Restless Ribbons of Sand: Duke University Press, Durham, N.C., 318 p. PIRES, A. C.; GOMES, A.; AND CHAMINE´, H. I., 2009, Dynamics of coastal systems using GIS analysis and geomaterials evaluation for groins: Environmental Engineering Geoscience, Vol. 15, No. 4, pp. 245–260. RINEHART, J. R. AND POMPE, J. J., 1999, Preserving beaches: In South Carolina, doing it privately: PERC Reports, Vol. 17, No. 3, pp. 7–8. ROBERTSON, W. V.; ZHANG, K.; AND WHITMAN, D., 2007, Hurricane-induced beach change derived from airborne laser measurements near Panama City, Florida, Marine Geology, Vol. 237, pp. 191–205.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
179
Doyle and Adams SHOSHANY, M. AND DEGANI, A., 1992, Shoreline detection by digital image processing of aerial photography: Journal Coastal Research, Vol. 8, No. 1, pp. 29–34. SOUTH CAROLINA DEPARTMENT OF PARKS, RECREATION AND TOURISM (SCDPRT), 2011, Tourism Statistics: Electronic document, available at http://www.scprt.com/our-partners/TourismStatistics. aspx THIELER, E. R. AND DANFORTH, W. W., 1994a, Historical shoreline mapping (I): Improving techniques and reducing positioning errors: Journal Coastal Research, Vol. 10, No. 3, pp. 549–563. THIELER, E. R. AND DANFORTH, W. W., 1994b, Historical shoreline mapping (II): Application of the digital shoreline mapping and analysis systems (DSMS/DSAS) to shoreline change mapping in Puerto Rico: Journal Coastal Research, Vol. 10, No. 3, pp. 600–620.
180
THIELER, E. R.; HIMMELSTOSS, E. A.; ZICHICHI, J. L.; AND ERGUL, A., 2009, Digital Shoreline Analysis System (DSAS) version 4.0—An ArcGIS extension for calculating shoreline change: U.S. Geological Survey Open-File Report 2008-1278, 79 p. *current version 4.3 U.S. FISH AND WILDLIFE SERVICE (USFWS), 2012, Species Listed in South Carolina Based on Published Historic Range and Population: Electronic document, available at http://www. fws.gov/ecos/ajax/tess_public/pub/stateListingIndividual.jsp? state5SC&status5listed VIRDIS, S. G. P.; OGGIANO, G.; AND DISPERATI, L., 2012, A geomatics approach to multitemporal shoreline analysis in Western Mediterranean: the case of Platamona-Maritza beach (northwest Sardinia, Italy): Journal Coastal Research, Vol. 28, No. 3, pp. 624–640.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 165–180
Geologic, Geotechnical, and Geophysical Investigation of a Shallow Landslide, Eastern Kentucky MATTHEW M. CRAWFORD1 JUNFENG ZHU STEVEN E. WEBB Kentucky Geological Survey, University of Kentucky, 228 Mining and Mineral Resources Building, Lexington, KY 40506
Key Terms: Landslides, Geophysics, Engineering Geology, Instrumentation, Soil Mechanics
ABSTRACT Shallow colluvial landslides are common in eastern Kentucky, as well as in the east-central Appalachian region. A geological, geotechnical, and geophysical investigation was carried out for a shallow colluvial landslide in Boyd County, KY. The purpose of this project was to assess the geologic conditions, extent, and behavior of a rainfall-triggered landslide in eastern Kentucky and to evaluate the use of electrical resistivity as a tool to characterize a shallow colluvial landslide. This study showed that 1) colluvial landslide movement is correlated to the rainfall and 2) inverted resistivity sections with distinct resistivity contrasts that correlated to landslide stratigraphy, depth of the failure surface, and groundwater regimes. INTRODUCTION Eastern Kentucky is located in the east-central Appalachian Plateau, part of the larger southern Appalachian Basin, and is affected by a wide range of landslide types and magnitudes. Landslides range from small slumps and translational slides along roadways to large earth and debris flows that can be hundreds of meters long. This physiographic region extends from Pennsylvania into parts of Ohio, West Virginia, Kentucky, Virginia, and Tennessee (Gray et al., 1979; Radbruch-Hall et al., 1982; and Outerbridge, 1987a) (Figure 1). The plateau is highly dissected with relief that ranges from approximately 120 to 300 m. Interbedded clastic sedimentary rocks of Paleozoic age dominate the region. Steep slopes have high incidences of landslides, and landslide susceptibility stems from particular bedrock lithologies and colluvial soils (Gray 1
Corresponding author email: mcrawford@uky.edu.
and Gardner, 1977; Outerbridge, 1987b). This region is prone to landslides, particularly during large precipitation events. For example, in 1998 storms produced 165 mm of rain in 72 hours over southeastern Ohio, causing six fatalities and millions of dollars in property and infrastructure damage (Shakoor and Smithmyer, 2005). Landslides damage roadways, infrastructure, and residences, with mitigation costs exceeding $10 million per year in Kentucky, eastern Kentucky in particular (Crawford, 2014; Overfield, 2014). For example, in July of 1939, in Wolfe and Breathitt Counties, KY, 508 mm of rain fell during a thunderstorm over the course of 2 days, causing a reported four debris flows (Wieczorek and Morgan, 2008). Flash flooding in Virgie, KY, in May 1999, caused several damaging debris flows (Harley, 2011). Persistent rainfall totaling 381–457 mm across eastern Kentucky from late April to mid-May 2011 caused more than 60 landslides. A short, intense storm that dropped approximately 90 mm of rain in 3 hours over a very localized area caused a large damaging landslide in Powell County, KY (Crawford, 2012). The majority of landslides induced by heavy rain are shallow, colluvial mass wasting events. This type of landslide is common in Kentucky; however, there are few landslide characterization studies that include a combined geologic, geotechnical, and geophysical analysis. Transportation officials mitigate landslides along roadways, but very few other government agencies analyze landslide hazards, and if they do, their results are not made public or are difficult to access. Private geotechnical engineering firms conduct landslide investigations and provide mitigation services, but the landslide data in their reports are not typically accessible to the public. This study investigated the Meadowview landslide in Boyd County, KY. The landslide occurred in April 2011 and was caused by a combination of natural and anthropogenic factors, and triggered by heavy rain fall. Local geology, steep slope, house foundation
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
181
Crawford, Zhu, and Webb
Figure 1. Locations of the Appalachian Plateau, eastern Kentucky, and project area in Boyd County, KY.
excavation, vegetation removal, and fill placement contributed to the landslide. The purpose of this project was to assess the geologic conditions, geometry, and behavior of a rainfall-triggered landslide and to evaluate the use of electrical resistivity as a tool to characterize a shallow colluvial landslide. A variety of instruments, sensors, and laboratory testing were used to collect information on meteorological and hydrologic conditions and landslide movement. A slope inclinometer and total station monitored landslide movement. Piezometers and a rain gauge collected groundwater and rainfall data, respectively. Laboratory analyses provided material index and strength properties. These included Atterberg limits (ASTM D4318), grain size distribution (ASTM D422), and
182
consolidated undrained triaxial shear tests. The shear test results were not used in slope stability assessment. An eight-channel resistivity meter measured surface and borehole electrical resistivity.
MEADOWVIEW LANDSLIDE The Meadowview landslide is located in Boyd County, eastern Kentucky (Figure 1). The bedrock in the area consists of interbedded shale, underclay, sandstones, and coals. Dobrovolny et al. (1963) stated that plastic and semiplastic shales and underclays are highly impermeable and are the least competent rocks in the area. Most landslides occur along the under-
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
Landslide Investigation, Kentucky
clays, where hillsides are steep. Many small landslides have occurred along these beds in hillside excavations for houses. These rocks develop sandy to clayey colluvial soils on the slopes and residual soils on the ridgetops. The landslide material consists of colluvium with added disturbed material from foundation excavation. Colluvium ranges in thickness from 1 to 3 m. During the excavation of the house foundation, material was pushed down into a naturally concave part of the slope. The concavity was accentuated near the toe by additional excavation for a power line that leads from the base of the slope toward the crown of the slide. The colluvium and excavated material observed at the surface are light brown and clayey to silty, with abundant shale and sandstone fragments. The soft clay soil is mottled gray, and the silty shale fragments are micaceous. During bulldozing, an outcrop of gray, soft clay was exposed near the toe of the slide that correlates to the ‘‘clayey shale’’ described in the boring logs. Large sandstone slabs are also present in the slide material. The Meadowview landslide occurred in late April 2011 as approximately 203 mm of rain fell over the month and triggered the failure (Community Collaborative Rain, Hail, and Snow Network, 2013; Kentucky Mesonet, 2013). The slope containing the slide ranges from approximately 13u near the ridgetop, above the crown, and steepens to 16.7u near the toe of the slide. The landslide occurred in a naturally concave part of the slope that is forested, with the exception of the trees and shrubs that were removed for the house excavation. The landslide is active, containing multiple scarps, seeps, and small localized flows. Rotational movement occurred in the uppermost part of the landslide, and closer to the toe the slide material morphed into a translational flow. The slide measures approximately 44 m long down the axis and 40 m wide near the middle (Figure 2). The main scarp height ranges from a few centimeters at the flanks to approximately 1.5 m near the middle. The volume of displaced material (after the landslide) was calculated as approximately 2,517 m3, assuming a half-ellipsoid shape and using a maximum depth of rupture (approximately 2.7 m) (Working Party on Worldwide Landslide Inventory, 1990; Cruden and Varnes, 1996). A prominent secondary scarp is present approximately 10 m downslope from the head scarp. Small tension cracks occur on the flanks of the upper slide area. High concentrations of water occur at the toe of the landslide. Identifying slope geomorphology is an important part of assessing landslide susceptibility. Natural colluvial soils accumulate in concave parts of slopes and often have high landslide incidences. There is
evidence of pre-existing landslide activity along the ridge, adjacent to the main slide area, including old (historic?) scarps, hummocky topography, and bent tree trunks. GEOTECHNICAL INVESTIGATION Boreholes and Material Properties Six boreholes were drilled into the Meadowview landslide (Figure 2) on March 13 and 14, 2013. The borehole locations were chosen to obtain data near the downslope axis of the landslide and near the main scarp and toe. A 3.25-in. (8.25-cm) hollow-stem auger was used to core all boreholes. Continuous sampling was performed with a Standard Penetration Test split spoon (18 in.; 45.7 cm) to obtain moisture content through two of the boreholes. A summary of the material properties is contained in Table 1. Two boreholes (B1 and B3) were constructed with inclinometer casing, two boreholes (B2 and B4) were converted to open standpipe piezometers, and two boreholes (B5 and B6) were cased with slotted polyvinyl chloride (PVC) and used for borehole electrical-resistivity measurements. B5 and B6 were located such that they lined up with the inclinometer boreholes. Lithologic units in boreholes B1 and B3 were logged, and stratigraphy was interpreted. Borehole B1 was drilled into bedrock to a total depth of 6.5 m and well below the assumed failure surface. The uppermost soil consisted of 2.7 m of disturbed colluvium, and water was encountered at a depth of 1.2 m. The disturbed colluvium was divided into two types: 1.2 m of sandy, lean clay with gravel that overlies 1.5 m of sandy, fat clay. The boundary between the two colluvial types may explain a difference in the disturbed material that came from excavation of the house foundation above the landslide and natural hillslope colluvium. Below the disturbed colluvium are three layers: 0.7 m of stiff to hard, fat clay; 0.7 m of weathered claystone; and 2.4 m of clayey shale. The boring was terminated at 6.5 m in weathered clayey shale. Soil density increased significantly at the contact between the two colluvium types and also between the native fat clay and weathered claystone. The field N-values increase from 4 to 43 and from 18 to 50, respectively. Borehole B3 was drilled to a total depth of 4.7 m. The uppermost soil consisted of 0.6 m of disturbed colluvium, and groundwater was encountered near the surface. Below the fill is 1.2 m of lean clay and 2.7 m of clayey shale. Drilling was terminated when carbonaceous, laminated, weathered shale was encountered at a depth of about 4.7 m. The field
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
183
Crawford, Zhu, and Webb
Figure 2. Aerial image of the Meadowview landslide. The main landslide area is within the dashed outline. Axes show dimensions of the slide. Borehole locations showing instrumentation types are also depicted.
N-values increased at the lean clay–clayey shale contact, indicating an increase in density. Surface and Subsurface Water Observations Elevated groundwater levels cause landslides, and precipitation that elevates these levels to an instability threshold can often be the triggering mechanism. Field reconnaissance at the Meadowview landslide prior to drilling revealed the main landslide area to be very wet, especially near the toe. Several seepage zones existed throughout the landslide. Based on our hydrostratigraphic model for the site, we inferred that shale beds were causing perched water along the slope. Water runs along low-permeability clay shales and seeps out where these beds intersect the surface.
184
Rainfall Rainfall data were collected by a RainWise tippingbucket rain gauge. The rain gauge consists of a standalone collector and recording system. The recorder has the ability to accumulate 1 year of rainfall with 1-minute resolution. The tipping bucket was set with a 0.25 mm/tip threshold. We installed the rain gauge on March 19, 2013. Total rainfall accumulation at the Meadowview landslide from the installation date through May 20, 2014, was 1,227.2 mm (48.3 in.) (Figure 3). Average annual precipitation from 1981 to 2010 in nearby Ashland, KY, was 1,122.6 mm (44.2 in.) (National Climatic Data Center, National Oceanic Atmospheric Administration, [NOAA], 2014). Considering the average
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
Landslide Investigation, Kentucky Table 1. Summary of the material properties from borehole samples and boring logs of the Meadowview landslide. Depth (m) Borehole B1 0–1.2 1.2–1.5 1.5–2.7 2.7–3.4 3.4–4.1 4.1–6.5 Borehole B3 0–0.6 0.6–1.8 1.8–4.6 4.6–4.8
Field Description Sandy lean clay with gravel (CL)—fill Sandy fat clay (CH)—fill Sandy fat clay (CH)—fill Fat clay (CH) Claystone Clayey shale Sandy lean clay with gravel (CL) Lean clay (CL) Clay shale Shale
% Gravel
% Sand
% Silt
% Clay
Plasticity Index
Field N-Value
4.3
45.5
23.8
26.3
16
5
N/A N/A 16
43 5 18 50 N/A
4.2 28.6 23.1 44.1 9.1 41.4 19.4 30.1 Very stiff to hard, residual soil structure Severely weathered, very soft Thinly laminated, weathered, very soft, minor interbedded
sandy shale
Moderately stiff, micaceous, sandstone fragments 6.6 37.6 21.8 Thinly laminated, weathered, very soft Carbonaceous, fissile, weathered, soft
34
8 8 9 N/A
11 24
*N/A 5 not measured/not calculated for this interval.
annual precipitation in the area, the monitoring of the Meadowview landslide occurred during a dry year. Piezometer Data Boreholes B2 and B4 were converted to open standpipe piezometers and were used to measure groundwater levels within the landslide mass (Figure 2). We recorded depth-to-water using a waterlevel meter that consisted of an electronic probe and a cable reel. The cable measured depth from the surface
(at borehole tip) to the water. The initial depth readings in B2 and B4 (both 3 m in total depth) were taken on March 19, 2013. We measured water depth once a week for the first 2 months and then recorded it monthly after that, because water levels did not fluctuate extensively. Beginning on April 12, 2013, we also used a wireless, battery-powered Telog PR-38 Pressure Recorder to measure the groundwater levels in piezometer B2 (below the assumed failure zone). The recorder contains a pressure sensor that is placed
Figure 3. Daily rainfall measured at the Meadowview landslide from March 2013 to May 2014.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
185
Crawford, Zhu, and Webb
Figure 4. Maximum daily groundwater levels measured from the bottom of B2 from the pressure recorder compared with daily rainfall from June 25 through September 23, 2013. Note the slight increase in groundwater level after the rainfall events.
at the bottom of the piezometer, measuring water level above the sensor. The sensor samples the frequency of water levels at user-defined intervals. We correlated groundwater fluctuations (measured in the piezometers) with rainfall. The largest pulses of rainfall caused an increase in groundwater level in the piezometers. A graph from late June to midSeptember 2013 correlates with increases in groundwater level above the bottom of the borehole with rainfall pulses (Figure 4). In B2, groundwater level change above the sensor, after rainfall pulses, varied from 80 mm in the spring of 2013 to 122 mm in the spring of 2014. The time frame for the groundwater increase ranged from 1 to 3 days following a rainfall pulse. The clayey colluvial fill stores a lot of water, which is perched on the low-permeability clay layers, controlling a smaller groundwater-level response to rainfall. Landslide Movement Inclinometer Inclinometer measurements were used to determine the magnitude, rate, direction, and depth of movement at boreholes B1 and B3. We used a Slope Indicator Digitilt Inclinometer System, including a biaxial probe that contains two perpendicular accelerometers, in effect monitoring the displacement normal to the axis of the borehole casing. The baseline inclinometer reading was conducted on March 25, 2013. Readings were taken once a week for the first 2 months and once a month after that. Cumulative horizontal displacement in B1 in the head
186
of the landslide through May 20, 2014, was approximately 2 cm. Cumulative displacement in B3 near the toe of the landslide through May 20, 2014, was approximately 5 cm. The greatest average velocity in B1 (0.05 mm/d) occurred from June 11 to July 2, 2013. This interval corresponded with 78.7 mm of rainfall and had the second highest daily event during monitoring, 36.8 mm on June 26. The two greatest average velocity increases in B3 were 0.16 mm/d from April 19, 2013, to May 8, 2013, and 0.5 mm/d from April 19, 2014, to May 20, 2014. These intervals corresponded with 46.9 and 130.7 mm of rainfall, respectively. Although the inclinometer measured little movement, a correlation was made between landslide movement and rainfall events (Figure 5). Generally, the increase in movement in B3 in the spring of 2013 and 2014 correlated with the obvious pulses of rainfall. The summer months contained pulses of rain that triggered most of the movement in B1. April and May 2014 showed significant increase in movement, backed up by more rainfall in these months (166.5 mm) than in 2013 (92.2 mm). To fully observe seasonal patterns in movement, monitoring should extend beyond the 14 months of data presented here. Total Station Surface displacement at various locations on the landslide was monitored using a Leica TC(R) 403 total station to supplement subsurface displacement information from the inclinometer. Eight survey stakes were leveled and secured with concrete approximately 0.45 m into the ground. The stakes
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
Landslide Investigation, Kentucky
Figure 5. Inclinometer displacement versus time in B1 and B3 plotted with rainfall. June 11 to July 2, 2013, and April 19 to May 20, 2014, contained a high frequency of rainfall events that corresponded with the highest average velocity displacement in B1 and B3.
were distributed along the landslide’s longitudinal axis from near the main scarp down to the landslide toe (Figure 6). The inherent accuracy of total station surveying allows small amounts of movement to be detected even before cracking or tension scarps are apparent (Keaton and DeGraff, 1996). A relative coordinate system was created using the stakes and two known reference base points outside the slide area that were assumed to be stable. Two locations above the headscarp (denoted as pole and garage) were used as the reference points to calculate temporal movements of the eight stakes. Measurements were calculated once a month starting May 1, 2013, and ending November 13, 2013. Displacements were measured using the differences in easting, northing, and height from the initial starting date measurement. This allowed displacement of each stake to be monitored over time; it also allows measurement of the overall average stake displacement over time. The general direction of movement of the eight stakes is to the northeast, which corresponds to the general slope direction and movement of material (Figure 7). Stakes S3, S5, S6, and S8 moved in the expected direction, trending generally northeast. With the exception of S8, these stakes moved horizontally a total of 5.8 cm. S8 had horizontal displacement of approximately 3.74 cm in the northeast direction. S8 is at the toe, where the landslide flows, and more subsurface displacement was measured here. Not all stakes moved in the expected direction, and several had little downslope movement, which was not discernable from the error threshold of the total station (approximately 5 mm). However, several points appeared to move upslope, located on the
slump block, or were located at a hinge and showed no movement. S7, for example, showed backward movement and movement over time that generally trended in the southeast direction. This is reasonable, because S7 lies near the flank of the landslide that faces southeast and may have experienced rotational movement on the steep flank of the landslide. The stakes that moved downslope were all in the lower part of the landslide, below the secondary scarp, where the translational flow is occurring. The relatively small horizontal movement of the stakes agrees with the small subsurface horizontal offset measured by the inclinometer. ELECTRICAL RESISTIVITY The technique of two-dimensional electrical-resistivity tomography (ERT) has been applied successfully for imaging many different types of landslides in order to detect failure surfaces, lithologic interfaces, and moisture regimes (Brooke, 1973; Bogoslovsky and Ogilvy, 1977; McCann and Forster, 1990; Godio and Bottino, 2001; Bichler et al., 2004; Lapenna et al., 2005; Drahor et al., 2006; Sastry et al., 2006; Jongmans and Garambois, 2007; Perrone et al., 2008; Sass et al., 2008; Schrott and Sass, 2008; de Bari et al., 2011; Travelletti et al., 2012; and Van Dam, 2012). We conducted six surface electricalresistivity survey measurements and two borehole resistivity measurements (Figure 8). The borehole and surface measurements were each conducted initially on separate dates on June 14 and July 26, 2013, and both were repeated on November 13, 2013. The surface measurements were set up as two arrays
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
187
Crawford, Zhu, and Webb
Figure 6. Locations of total station stakes and surveying reference points.
perpendicular to the slope direction and one array parallel to the slope direction, down the axis of the landslide. An Advanced Geosciences Supersting eight-channel resistivity meter was used to make the measurements. The surface arrays utilized a dipole-dipole electrode configuration with 1.5-m electrode spacing. Short spacing allows for higher resolution and is optimal for landslides anticipated to be shallow (,10 m). The dipole-dipole array has been proven to be successful for obtaining higherresolution data and for determining shallow interfaces in landslides (Lapenna et al., 2005; Schrott and Sass, 2008). To account for topographic changes, a total station was used to survey points along the arrays. Using those points, a terrain file containing horizontal distance and elevation was created for use in generating the inverted resistivity images. The borehole measurements were made in B5 and B6, the slotted PVC boreholes, and utilized a cross-
188
hole method that measured voltage between electrodes. We used borehole electrodes at 0.5-m intervals. The boreholes were spaced 7.1 m apart and were 5 m deep, resulting in an aspect ratio (depth of hole/ distance between holes) close to 1.5, to maximize resolution (Advanced Geosciences, Inc., 2003). The cables hung in the two open boreholes. The electrodes must be in direct contact with the soil (as with the surface arrays), so the boreholes were filled with water to transmit the current to the soil. The boreholes were aligned with surface array MVS1, which is parallel to the downslope direction of the slide. This allowed comparison with the surface ERT images of MVSI and MVS2, which were arranged perpendicular and parallel to the downslope direction. Resistivity Results Layering and clear resistivity contrasts show that high and low zones were present in the inverted
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
Landslide Investigation, Kentucky
Figure 7. Coordinate system showing surface displacement of all eight stakes in the Meadowview landslide. The general trend of movement is downslope, toward the northeast. Stakes in the area indicated by the dashed circle show approximate area of little discernable movement or movement backward from rotation.
images and reflect the shallow landslide geometry and both rotational and translational styles of movement. Interpreted surfaces coincide with sharp drops in resistivity, indicating high water content (perched water) and/or possibly higher clay content. With saturated soil (disturbed colluvium) encountered at a depth of just 1.2 m, we considered water to be the influential factor in the low resistivity near the surface and were most concerned with identification of the failure zone. These resistivity zones, including the failure surface, correlate with lithologies observed in the boreholes and landslide depth determined from the two inclinometers. The surface and borehole arrays show ranges of electrical-resistivity values that are generally the same with all profiles, and the ranges
do not vary significantly between the two different measurement dates. Very little precipitation had fallen in the 2 days leading up to all the measurements, and little groundwater fluctuation occurred in piezometer B2. Overall precipitation amounts were less in the fall than in the summer, which may account for slight differences in the inverted imagery. Inverted Resistivity Sections MVS1—7/26/2013: Parallel to the Landslide Axis in the Downslope Direction—MVS1 spans 45.7 m and extends downslope from the crown of the slide to the toe (Figure 9). The inverted resistivity section shows that distinct layering and contrasts in resistivity are
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
189
Crawford, Zhu, and Webb
Figure 8. Electrical-resistivity array locations (arrows and yellow circles) in the Meadowview landslide outlined by dashed line.
evident near the headscarp of the slide. A semicontinuous high-resistivity layer (oranges to reds) is present near the surface, ranging between approximately 50 and 600 Ohm-m. An identifiable break in the highresistivity layer occurs at the surface at the headscarp displacement. A thin, lower-resistivity zone (greens) appears below the high-resistivity layer, ranging from 30 to 50 Ohm-m. Perched water on the underlying clay shales creates the lower resistivity (higher conductivity) values. This zone continues downslope, occurring near the surface, where water intersects the surface seeps near the toe of the landslide. A patchy low-resistivity zone (blues) occurs below the highresistivity zone, approximately 2.7 m below the surface in the head of the landslide. This lowresistivity zone ranges from approximately 8 to 19 Ohm-m. Starting at the headscarp, this low-resistivity zone extends downslope for about 22 m and has an undulating, arcuate shape. It becomes shallower
farther downslope and ends abruptly. We interpreted this zone as the failure surface; this was confirmed by inclinometer data that indicated displacement depth at B1 to be about 2.7 m. Below the low-resistivity zone, resistivity increased to a range of approximately 30–50 Ohm-m (greens) down to the bottom of the section. To get a closer look at the resistivity data, we extracted resistivity and depth (x, y, and z) from the raw inverted resistivity data at the location of borehole B1. These data show a resistivity profile through the high- and low-resistivity layers near the headscarp (Figure 10). A sharp peak of a resistivity increase at about 128 Ohm-m correlates to the lithologic change in the disturbed colluvial fill. This material grades from a sandy lean clay into a moderately stiff sandy fat clay. There was also a big jump in density at this interface, as shown by the blow counts in the boring logs. Water was encountered during drilling at this interval, at about 1.2 m. Resistivity then decreased (moisture content increased) to approximately 19 Ohm-m. This interval and the low-resistivity peak correlate with the contact between high-moisture conditions at the colluvial fill and very stiff, fat clay-shale, which is also the inferred failure surface. Below the inferred failure surface, the resistivity increased slightly as the moisture content decreased. Midslope, approximately 17.3 m downslope from the headscarp, resistivity ranged between 14 and 19 Ohm-m in the low-resistivity zone that is the interpreted failure surface. Below the failure surface, resistivity increased toward two distinct high-resistivity zones. One is a continuous arcuate zone that continues downslope; the other deeper zone is lenticular shaped. These may be the deeper, drier (?) clay-shale layers (less conductive). These high-resistivity zones range between approximately 80 and 160 Ohm-m. No borehole was drilled midslope, but the interpreted failure surface (low-resistivity peak) from the resistivity profile from MVS1 correlates with the failure surface determined from the inclinometer data (Figure 11). secondary scarp
0.0
4.6
9.1
13.7
18.3
22.9
27.4
headscarp
32.0
36.6
41.2
Elevation (m)
45.7
Ohm-m
693
12.4
s f a i l u re
-7.9
urfac e
159 36.7
-3.3
B1
-1.2 -5.8
Iteration = 6
B3 RMS = 2.87%
8.4 1.9
L2 = 0.91
Electrode Spacing = 1.52 m
Figure 9. Inverted electrical-resistivity array MVS1. Dashed lines represent multiple failure surfaces. Note locations of boreholes, the headscarp, and secondary scarp.
190
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
Landslide Investigation, Kentucky
Figure 10. Vertical electrical-resistivity profile at borehole B1. Depth starts at the first point, toward the top of the curve, which is at the surface.
Toward the toe (Figure 9), the distinct resistivity zones became more complex. Extracted resistivity and depth data (x, y, and z) from the raw inverted resistivity profiles at the location of borehole B3 showed a high-resistivity peak of 79 Ohm-m just below the surface. At B3, the colluvial fill was only
0.6 m deep, supporting the shallow flow type of slope movement at the toe. The failure surface is difficult to identify in the inverted resistivity section’s correlation to the borehole data. The inclinometer data from borehole B3 indicated that the failure surface was 1.2 to 1.5 m below the surface. The underlying
Figure 11. Vertical resistivity profile taken midslope from section MVS1. The low-resistivity peak correlates with the failure surface depth measured with the inclinometer. Depth starts at the first point, toward the top of the curve, which is at the surface. Vertical axis values on inclinometer reading are depth in feet. Horizontal axis is displacement in inches.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
191
Crawford, Zhu, and Webb
Figure 12. Inverted resistivity profile MVS2 in a transverse direction, below the headscarp of the landslide.
high-resistivity layer (curved yellows and orange layer that start midslope) was approximately 90–130 Ohmm and correlates to the lean clay–clay shale contact, where a stiff, structured lean clay transitions to a very soft, weathered clay shale. A distinct low-resistivity peak of approximately 50 Ohm-m occurred about 4.3 m below the surface, which correlates with the clayey shale–shale contact and a decreasing moisture content, as described in the borehole. A highresolution, lenticular zone was present at the end of the MVS1 array. This zone was approximately 2 m in length and showed significantly higher resistivity values than did the continuous high-resistivity zone that started midslope and curved toward the toe. This feature could be a large sandstone boulder that was dislodged during excavation of the house foundation. Large boulders of that size were identified in the field, at the toe of the slide. MVS2—7/26/2013: Perpendicular to the Downslope Direction, Upper Slope—Electrical-resistivity array MVS2 spanned 36.6 m perpendicular to the downslope direction along the upper part of the slide. This array crosses borehole B1 (Figure 12). There was a clear contrast between a higher-resistivity zone and an underlying low-resistivity zone present. We interpreted this boundary to be the failure surface, which corresponds with the colluvial fill and fat clay bedrock contact, and the landslide depth indicated in the inclinometer data from borehole B1. Two lenticular-shaped, high-resistivity zones (possibly connected) occupied the right side of the inverted section above the failure surface. The right side of the section (toward the end) runs northwest, leading toward the headscarp. A moderately thick sandstone layer crops out behind the headscarp and MVS2 may be intersecting this high-resistivity layer.
Resistivity at this location and along the identified failure surface ranged between approximately 20 and 30 Ohm-m. Similarly to MVS1, a high-resistivity peak from x, y, and z data extracted at the B1 location correlates to the contact among colluvial fill types, sandy lean clay, and sandy fat clay. The highest moisture content was measured at a low-resistivity peak, supporting the location of the failure surface. MVS3—7/26/2013: Perpendicular to the Downslope Direction, Toeslope—Electrical-resistivity array MVS3 spanned 24.4 m in a transverse direction across the toe of the slide. The inverted section shows a complex pattern of resistivity zones (Figure 13). An undulating low-resistivity zone is present near the surface. This zone ranged from approximately 24 to 50 Ohm-m. This low-resistivity zone transitioned to a high-resistivity zone with lenticular regions. The undulating boundary between the low- and highresistivity zones for MVS3 was shallow, about 0.6 m deep, and correlates to the contact between sandy lean clay with gravel fill and stiff, residual, lean clay. The inclinometer measurements from borehole B3 indicate the failure surface is below the colluvial fill– lean clay contact; therefore, the failure zone at the toe may also include the lean clay unit. November Results On November 13, 2013, these same arrays were laid out and the electrical resistivity was measured. In general, the resistivity contrasts, interpreted features, and correlations to stratigraphic boundaries were similar to those measured in July. One change in MVS1 was the presence of a low-resistivity zone (8–26 Ohm-m) that extended down vertically below the inferred failure surface, just in front of the headscarp.
Figure 13. Inverted resistivity profile MVS3 perpendicular to the downslope direction, along the toe of the landslide.
192
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
Landslide Investigation, Kentucky
Figure 14. Borehole resistivity results from June 14, 2013. The middle of the inverted section shows a contrast in resistivity that correlates to the colluvial fill–fat clay stratigraphic boundary.
This zone accentuated the rotational movement in the head. Water may have infiltrated this zone, causing the low resistivity. For MVS3 (November measurement), the measurements from the high-resistivity zones (24–50 Ohm-m) were larger and were spaced differently than the measurements from the July inverted section. Approximately 104 mm less rainfall was measured in the month preceding the November resistivity measurements. This could account for the increased area of higher resistivity in MVS3. Borehole Resistivity We also measured electrical resistivity at the slide using downhole electrode cables in the slotted PVC boreholes. A cross-hole method was used to measure the resistivity. Similar to the surface dipole-dipole array, this method is designed to measure the voltage between all electrodes that hung down in the boreholes. In the center of the inverted section, Figure 14 shows a change in resistivity that correlates with a change in material type in borehole B1 (black dashed line). B1 is between the slotted PVC holes, which are 7.1 m apart. There was no significant difference between the June 14 and November 11 measurements and resulting inverted profiles. Figure 14 also shows the resistivity data at depth taken from the middle of the borehole profile. There is a
slight decrease in resistivity that correlates to the failure surface depth. DISCUSSION For discontinuous, variable bedrock lithologies and heterogeneous soils, drilling boreholes may not provide the data needed to interpret the landslide type and failure surface. Geophysical investigations, specifically electrical-resistivity investigations, can expand landslide hazard research by providing an overall view of the subsurface that can supplement drilling by not only identifying failure planes and moisture regimes but also by relating the electricalresistivity values to mechanical properties. Quality subsurface data, including detailed lithologic logs, an idea of groundwater flow, and the applicable laboratory data, are imperative to using electrical resistivity as a tool for characterizing landslide behavior. The challenge, and possible future work, involves taking a non-unique solution of resistivity measurements in the subsurface and linking it to landslide behavioral properties, such as moisture content, matric suction, clay content, and porosity. Long-term studies on more than one landslide in a region and additional materials testing could improve the use of electrical resistivity as a tool for landslide assessment. Although not addressed in this study, shallow, colluvial landslide
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
193
Crawford, Zhu, and Webb
investigations that aim to correlate electrical resistivity with factors needed to calculate shear strength, ultimately providing a tool for repetitive, effective slope-stability assessments, would be beneficial.
Engineering, for technical advice, and Ed Woolery at the University of Kentucky Department of Earth and Environmental Sciences for review and technical advice.
CONCLUSIONS
REFERENCES
The Meadowview landslide movement corresponded to periods of greatest rainfall. This study showed that increases in groundwater levels corresponded with particular precipitation events. During the study, total displacement observed from the inclinometer in B1 was 2 cm; it measured 5 cm in borehole B2 at the toe. The highest average velocity at borehole B1 occurred between mid-June and early July 2013. During this interval, 78.7 mm of rain fell, and the second greatest daily event during monitoring, 36.8 mm, occurred on June 26. The highest average velocity at borehole B3 occurred from July 2 to July 18, 2013, during which 91.4 mm of rain fell. The rainfall at the site during the year was approximately 127 mm less than the average annual rainfall in the region, which may explain why there was only minor movement over the course of the year. The total station measurements of surface movement supplemented the subsurface inclinometer measurements. An intense or long-duration rainfall has the capability to trigger future movement. This study also showed that the surface and borehole electrical-resistivity measurements across the Meadowview landslide resulted in inverted resistivity sections with distinct resistivity contrasts that correlate to borehole stratigraphy, failure surface depth, and groundwater conditions. Low-resistivity zones were indicators of high moisture content (along with high clay content) and correlated to the failure surface of the landslide. The inverted resistivity profiles confirmed the curviplanar and undulating nature and shallow depth of the failure surface indicated by the inclinometer data.
ADVANCED GEOSCIENCES, INC., 2003, Cross Borehole Electrical Resistivity Tomography (ERT) Measurements: Electronic document, available at https://www.agiusa.com/datasheets. shtml BICHLER, A.; BOBROWSKY, P.; BEST, M.; DOUMA, M.; HUNTER, J.; CALVERT, T.; AND BURNS, R., 2004, Three-dimensional mapping of a landslide using a multi-geophysical approach: The Quesnel Forks landslide: Landslides, Vol. 1, No. 1, pp. 29–40. BOGOSLOVSKY, V. A. AND OGILVY, A. A., 1977, Geophysical methods for the investigation of landslides: Geophysics, Vol. 42, No. 3, pp. 562–571. BROOKE, J. P., 1973, Geophysical investigations of a landslide near San Jose, California: Geoexploration, Vol. 11, No. 1, pp. 61–73. COMMUNITY COLLABORATIVE RAIN, HAIL, AND SNOW NETWORK, 2013. Electronic document, available at http://www.cocorahs. org/State.aspx?state5KY CRAWFORD, M. M., 2012, Understanding landslides in Kentucky: Tools and methods to further landslide hazard research. In Eberhardt, E.; Froese, C.; Turner, K. A.; and Leroueil, S. (Editors), Landslides and Engineered Slopes: Proceedings of the 11th International and 2nd North American Symposium on Landslides, Banff, Alberta, Canada, Vol. 1, pp. 467–472. CRAWFORD, M. M., 2014, Kentucky Geological Survey Landslide Inventory: From Design to Application: Kentucky Geological Survey Information Circular IC_31_12, 18 p. 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: Transportation Research Board, National Research Council, Special Report 247, Washington, DC, pp. 36–75. DE BARI, C.; LAPENNA, V.; PERRONE, A.; PUGLISI, C.; AND SDAO, F., 2011, Digital photogrammetric analysis and electrical resistivity tomography for investigating the Picerno landslide (Basilicata region, southern Italy): Geomorphology, Vol. 133, No. 1, pp. 34–46. DOBROVOLNY, E.; SHARPS, J. A.; AND FERM, J. C., 1963, Geology of the Ashland Quadrangle, Kentucky-Ohio, and Catlettsburg Quadrangle in Kentucky: U.S. Geological Survey Geologic Quadrangle Map GQ-196, scale 1:24,000. DRAHOR, M. G.; GOKTURKLER, G.; BERGE, M. A.; AND KURTULMUS, T. O., 2006, Application of electrical resistivity tomography technique for investigation of landslides: A case from Turkey: Environmental Geology, Vol. 50, No. 2, pp. 147– 155. GODIO, A. AND BOTTINO, G., 2001, Electrical and electromagnetic investigation for landslide characterization: Physics Chemistry Earth, Part C: Solar Terrestrial Planetary Science, Vol. 26, No. 9, pp. 705–710. GRAY, R. E.; FERGUSON, H. F.; AND HAMEL, J. V., 1979, Slope stability in the Appalachian Plateau, Pennsylvania and West Virginia, USA. In Voight, B. (Editor), Rockslides and Avalanches, Elsevier North Holland Inc., New York, N.Y. Vol. 2, pp. 447–471. GRAY, R. E. AND GARDNER, G. D., 1977, Processes of colluvial slope development at MC Mechen, West Virginia: Bulletin
ACKNOWLEDGMENTS The Kentucky Geological Survey provided financial support. We would like to thank Terracon Consultants, Inc., for drilling and laboratory services, and especially Benjamin Taylor and Christopher Yohe for cooperation and geotechnical advice throughout the project. Francis Ashland of the U.S. Geological Survey Landslide Hazards Program assisted with site selection, technical advice, and editorial review. We would like to thank Zhenming Wang, Mike Lynch, Max Hammond, and Meg Smath at the Kentucky Geological Survey, Sebastian Bryson at the University of Kentucky, Department of Civil
194
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
Landslide Investigation, Kentucky International Association Engineering Geology, Vol. 16, No. 1, pp. 29–32. HARLEY, S. B., July 13, 2011, personal communication, NOAA, National Weather Service, Jackson, KY. JONGMANS, D. AND GARAMBOIS, S., 2007, Geophysical investigation of landslides: A review. Bulletin Socie´te´ Ge´ologique France, Vol. 178, pp. 101–112. KEATON, J. R. AND DEGRAFF, J. V., 1996, Surface observation and geologic mapping. In Turner, A. K. and Schuster, R. L. (Editors), Landslides: Investigation and Mitigation: Transportation Research Board, National Research Council, Special Report 247, Washington, DC, pp. 178–230. KENTUCKY MESONET, 2013, Monthly Climatological Summary: Electronic document, available at http://www.kymesonet.org/ historical_data.php LAPENNA, V.; LORENZO, P.; PERRONE, A.; PISCITELLI, S.; RIZZO, E.; AND SDAO, F., 2005, 2D electrical resistivity imaging of some complex landslides in the Lucanian Apennine chain, southern Italy: Geophysics, Vol. 70, No. 3, pp. B11–B18. MCCANN, D. M. AND FORSTER, A., 1990, Reconnaissance geophysical methods in landslide investigations: Engineering Geology, Vol. 29, No. 1, pp. 59–78. NATIONAL CLIMATIC DATA CENTER, NOAA, 2014, 1981–2010 Normals: Electronic document, available at http://www. ncdc.noaa.gov/cdo-web/datasets OUTERBRIDGE, W. F., 1987a, The Logan Plateau, a Young Physiographic Region in West Virginia, Kentucky, Virginia, and Tennessee: U.S. Geological Survey Bulletin 1620, 19 p. OUTERBRIDGE, W. F., 1987b, Relation between landslides and bedrock in the Central Appalachian Plateaus. In Schultz, A. P. and Southworth, C. S. (Editors), Landslides of Eastern North America: U.S. Geological Survey Circular 1008, 43 p. OVERFIELD, B. L., 2014, The Geologic Context of Landslide and Rockfall Maintenance Costs in Kentucky—2002 to 2009: Kentucky Geological Survey Information Circular [in press]. PERRONE, A.; VASSALLO, R.; LAPENNA, V.; AND CATERINA, D., 2008, Pore water pressures and slope stability: A joint geophysical
and geotechnical analysis: Journal Geophysics Engineering, Vol. 5, No. 3, pp. 323–337. RADBRUCH-HALL, D. H.; COLTON, R. B.; DAVIES, W. E.; LUCCHITTA, I.; SKIPP, B. A.; AND VARNES, D. J., 1982, Landslide Overview Map of the Conterminous United States: U.S. Geological Survey Professional Paper 1183, 25 p. SASS, O.; BELL, R.; AND GLADE, T., 2008, Comparison of GPR, 2D-resistivity and traditional techniques for the subsurface ¨ schingen landslide, Swabian Alb exploration of the O (Germany): Geomorphology, Vol. 93, No. 1, pp. 89–103. SASTRY, R. G.; MONDAL, S. K.; AND PACHAURI, A. K., 2006, 2D electrical resistivity tomography of a landslide in Garhwal Himalaya. In Proceedings of 6th International Conference and Exposition on Petroleum Geophysics (Kolkata): Society of Petroleum Geophysicists, Uttarakhand, India, pp. 997–1001. SCHROTT, L. AND SASS, O., 2008, Application of field geophysics in geomorphology: Advances and limitations exemplified by case studies: Geomorphology, Vol. 93, No. 1, pp. 55–73. SHAKOOR, A. AND SMITHMYER, A. J., 2005, An analysis of storminduced landslides in colluvial soils overlying mudrock sequences, southeastern Ohio, USA: Engineering Geology, Vol. 78, No. 3, pp. 257–274. TRAVELLETTI, J.; SAILHAC, P.; MALET, J. P.; GRANDJEAN, G.; AND PONTON, J., 2012, Hydrological response of weathered clayshale slopes: Water infiltration monitoring with time-lapse electrical resistivity tomography: Hydrological Processes, Vol. 26, No. 14, pp. 2106–2119. VAN DAM, R. L., 2012, Landform characterization using geophysics—Recent advances, applications, and emerging tools: Geomorphology, Vol. 137, No. 1, pp. 57–73. WIECZOREK, G. F. AND MORGAN, B. A., 2008, Debris-Flow Hazards within the Appalachian Mountains of the Eastern United States: U.S. Geological Survey Fact Sheet, 20083070, 4 p. WORKING PARTY ON WORLDWIDE LANDSLIDE INVENTORY, 1990, A suggested method for reporting a landslide: Bulletin International Association Engineering Geology, No. 41, pp. 5–12.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 181–195
195
Influence of Geologic and Index Properties on Disintegration Behavior of Clay-Bearing Rocks ABDUL SHAKOOR1 Department of Geology, Kent State University, 221 McGilvrey Hall, Kent, OH 44242
TEJ P. GAUTAM Department of Petroleum Engineering and Geology, Marietta College, 215 Fifth St., Marietta, OH 45750
Key Terms: Disintegration Behavior, Disintegration Ratio, Natural Climatic Conditions, Clay Mineralogy, Index Properties, Regression Analysis
ABSTRACT Clay-bearing rocks exhibit a broad range of disintegration behaviors, with second-cycle slake durability index values from 0 percent for some claystones to nearly 100 percent for some siltstones. This variability comes from their geologic makeup, especially clay content, clay mineral composition, and related index properties. To investigate the influence of geologic and index properties on disintegration behavior under natural climatic conditions, 12 replicate samples of each of 20 clay-bearing rocks (five claystones, five mudstones, five siltstones, and five shales) were exposed to natural conditions for 12 months. After each month, one replicate sample of each rock was tested for grain size distribution. Disintegration ratio (DR), the ratio of the area under the grain size distribution curve of slaked material for a given rock to the total area encompassing all grain size distribution curves of the tested samples, was used to quantify the amount of disintegration. Additionally, clay mineralogy, natural water content, dry density, void ratio, absorption, adsorption, Atterberg limits, and slake durability index were determined. Clay mineralogy and index properties were correlated with DR values after 1, 6, and 12 months of exposure. Results show that clay content, percentage of expandable clay minerals, plasticity index, adsorption, and slake durability index are better indicators of short-term disintegration behavior, whereas expandable clay mineral content and absorption influence long-term be havior. Regression models suggest that short-term disintegration behavior can be predicted more accurately 1
Corresponding author email: ashakoor@kent.edu.
than long-term behavior, as indicated by decreasing R2 values for samples exposed for 1, 6, and 12 months, respectively. INTRODUCTION The most commonly encountered clay-bearing rocks include claystones, mudstones, siltstones, and shales. Upon interaction with water, these rocks disintegrate to varying degrees through the processes of slaking and swelling (Dick and Shakoor, 1992; Sarman et al., 1994; Czerewko and Cripps, 2001; Pejon and Zuquette, 2002; Sadisun et al., 2005; Fahimifar and Soroush, 2007; Doostmohammadi et al., 2008; Erguler and Shakoor, 2009a; Summa et al., 2010; and Hajdarwish et al., 2013). The second-cycle slake durability index (Id2), as determined by the slake durability test (American Society for Testing and Materials [ASTM], 1996; International Society for Rock Mechanics [ISRM)] 2007), can range from 0 percent for some claystones to nearly 100 percent for some siltstones (Dick and Shakoor, 1992; Erguler and Shakoor, 2009a; and Gautam and Shakoor, 2013). This extreme variation in the disintegration behavior of clay-bearing rocks is influenced by the diversity of their geologic characteristics (clay content, amount of expandable clay minerals, micro-fabric) and index properties (natural water content, dry density, specific gravity, void ratio, absorption, adsorption, Atterberg limits) (Olivier, 1979; Russell, 1981; Shakoor and Sarman, 1992; Dick et al., 1994; Santi and Higgins, 1998; Gemici, 2001; Moradian et al., 2010; Wuddivira et al., 2010; Youn and Tonon, 2010; Broto´ns et al., 2013; and Nahazanan et al., 2013). A significant amount of research has been conducted to investigate the relationships between the slake durability index of clay-bearing rocks and their geologic and index properties (Russell, 1981; Dick and Shakoor, 1992; Dick et al., 1994; Bell et al., 1997; Santi and Doyle, 1997; Santi, 1998, 2006; Santi and Higgins, 1998;
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 197–209
197
Shakoor and Gautam
Koncagu¨l and Santi, 1999; Go¨kc¸eog˘lu et al., 2000; Singh et al., 2005; and Moradian et al., 2010). According to Russell (1981), clay content has the greatest influence on disintegration behavior. In addition to clay content, the type of clay minerals influences the amount of disintegration, being higher for rocks containing smectite than for other clay minerals (Bell et al., 1997). Additionally, microfabric, micro-fractures, and the nature of pores influence the nature and amount of disintegration (Russell, 1981; West and Shakoor, 1984; Dick et al., 1994; Molina et al., 2011; Bryson et al., 2012; and Weng and Li, 2012). Although extensive research has been done on investigating the relationships between laboratory values of the slake durability index and relevant geologic and index properties of clay-bearing rocks, a detailed study has not been conducted to investigate the influences of geologic and index properties on long-term disintegration behavior under natural climatic conditions. The disintegration behavior under laboratory conditions can be quite different than the disintegration behavior under field conditions (Santi, 2006; Gautam and Shakoor, 2013). Moreover, there is a need to develop a better index than the slake durability index for quantifying the amount of disintegration, especially for disintegration occurring under field conditions. The objective of this study was to investigate the influence of geologic and index properties on the disintegration behavior of clay-bearing rocks under natural climatic conditions. This study uses disintegration ratio (DR), instead of Id2, to quantify the amount of disintegration under natural climatic conditions. DR is defined as the ratio of the area under the grain size distribution curve of slaked material for a given rock sample to the total area encompassing all grain size distribution curves of the tested samples (Erguler and Shakoor, 2009b). We used DR because it provides a better quantification of the amount of disintegration by taking into account the grain size distribution of all slaked material from a sample than does Id2, which is based on material retained on 2-mm mesh sieve, regardless of the sizes of particles retained (Erguler and Shakoor, 2009b; Gautam and Shakoor, 2013). RESEARCH METHODS Sample Selection Twenty clay-bearing rocks—five claystones (CST), five mudstones (MST), five siltstones (SLT), and five shales (SHL)—were selected from 11 different states in the United States. Sample selection was based on information available from literature review to allow
198
for maximum diversity in lithologic characteristics and geologic ages. With the exception of one sample that was obtained from a dam site excavation, all samples were obtained from cut slopes along roadways. Maximum care was taken to assure that the samples collected were as fresh as possible by removing the weathered material to a depth of 0.3 to 1.3 m (1 to 4 ft). Ten to 20 individual chunks of rock, totaling approximately 23–27 kg (50–60 lb), were collected from each site. The sample pieces were individually wrapped, sealed in ‘‘Zip-lock’’ storage bags, and stored in 5-gallon snap-lid plastic buckets to preserve natural water content and to prevent breakage. Table 1 lists the sample number, geologic formation, geologic age, and geographic location for each sample. The classification of Potter et al. (1980), modified by Dick and Shakoor (1992), was used to classify each rock sample into one of the four groups. Laboratory Investigations Laboratory investigations relating to geologic properties included the following: determination of clay content by hydrometer analysis, determination of clay mineralogy by x-ray diffraction analysis, and examination of micro-fabric by scanning electron microscope (SEM) analysis. Index properties tested in the laboratory included natural water content, void ratio, dry density, absorption, adsorption, Atterberg limits (liquid limit, plastic limit, plasticity index), and slake durability index. All tests were conducted according to ASTM specifications (ASTM, 1996), where applicable. Exposing Clay-Bearing Rocks to Natural Climatic Conditions Disintegration behavior of clay-bearing rocks under natural climatic conditions was investigated by placing 12 replicate samples of each of the 20 rocks on the roof of McGilvrey Hall, a five-story building at Kent State University. Each sample consisted of 10– 12 pieces, each piece weighing 40–60 g, with the total sample weighing 450–550 g. Aluminum pans containing the samples had screen-covered holes for water to drain. The samples were exposed to natural climatic conditions for 1 year, from September 2009 to September 2010. During the period of exposure, the temperature ranged from 24.3uC (24.3uF) in January 2010 to 23.0uC (73.4uF) in July 2010, and the average annual precipitation was 101.5 cm (40 in.). The 5-year (2008–2012) monthly average temperature range was from 24.1uC (24.6uF) to 22.4uC (72.3uF), and the average precipitation was 118.2 cm (46.5 in.). January and February are the coldest months of the year in
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 197–209
Disintegration of Clay-Bearing Rocks Table 1. Formation name, age, and location for clay-bearing rocks studied. Total clay content and amounts of expandable versus nonexpandable clay minerals are also shown.
No. Sample
Clay Mineral in Bulk (%)
,0.002 mm ,0.004 mm
Exp. Non-Exp.
Formation
Age Penn
I-77 S, Ripley/Fairplan exit, West Virginia
41
53
2.6
38.4
Cret Cret Penn Cret
SR-159 S, 700 m S of I-70, Kansas I-90, 1 mile N of Oacoma, South Dakota Dam site, Point Marion, Pennsylvania SR-139 N, 27 miles N of Loma Carfield, Colorado SR-50 S, mile marker 23, Ohio I-77 N, 1 mile to Dexter exit, Ohio SR-191 S, 1 mile E of Flaming Gorge Dam, Utah SR-7 S, 1 mile off Loganport exit, Ohio SR-260, 3 miles from SR-7 S, Ohio I-77 N, 2 miles N of Dexter City exit, Ohio SR-22 E, 13 miles W of SR-7, Ohio I-77 S, Ramp of Eden Fork exit, West Virginia I-70 E, mile marker 156, Colorado I-80 W, mile marker 216, Wyoming SR-11 N, 2 miles from Radford, Virginia SR-75, 15 miles S of Loganport, Ohio SR William H. Natcher Parkway, mile marker 52, Kentucky SR-10, 8 miles S of Price Town, Utah SR-381 N, Bristol Motor Speedway, Tennessee
77 38 38 45
82 52 51 57
20.4 5.1 0 22.7
56.5 32.9 38 22.3
22 20 29 27 28 17 21 17 12 18 19 16 28
34 47 33 35 36 22 27 22 22 22 24 19 36
2.7 0.6 3.9 1 12.7 19.8 2.2 0 3.3 4.5 0.7 7.3 6.7
19.5 19.7 25.1 16 15.3 4.2 18.8 17 8.7 13.4 18.3 19.7 21.3
25 18
33 24
16.6 1.7
8.4 16.3
1
CST-1
2 3 4 5
CST-2 CST-3 CST-4 CST-5
6 7 8 9 10 11 12 13 14 15 16 17 18
MST-1 MST-2 MST-3 MST-4 MST-5 SLT-1 SLT-2 SLT-3 SLT-4 SLT-5 SHL-1 SHL-2 SHL-3
Conemaugh Formation Dakota Group Mowery Shale Conemaugh Fruitland Formation Conemaugh Monogahela Red Pine Shale Dunkard Dunkard Conemaugh Monongahela Conemaugh Wasatch Green River Milboro Shale Conemaugh Tradewater
19 20
SHL-4 SHL-5
Straight Cliffs Rome
Penn Penn Prot Perm Perm Penn Penn Penn Eoce Eoce Devo Penn Penn Cret Camb
Location (in United States)
Clay Content in Bulk (%)
CST 5 claystone; MST 5 mudstone; SLT 5 siltstone; SHL 5 shale; Exp. 5 expandable; Non-Exp. 5 non-expandable.
northeast Ohio, and May to August are the warmest months. Most of the snowfall in the region occurs during the months of December, January, and February. After each month of exposure to natural climatic conditions, one replicate sample of each of the four types of clay-bearing rocks was taken to the laboratory, oven-dried at 50uC, and its grain size distribution determined. Figure 1 is an example of the disintegration behavior of a claystone sample after varying periods of exposure to natural climatic conditions. Grain size distributions were used to quantify the amount of disintegration of each sample,
in terms of DR, after varying periods of exposure to natural conditions. Figure 2 describes the procedure for determining DR from grain size distribution. Observations of disintegration behavior were made on a biweekly basis, and a photographic record was maintained. Details of disintegration behavior of the four groups of clay-bearing rocks studied can be found in Gautam and Shakoor (2013). Data Analysis Bivariate, principal-component, and multivariate regression analyses were used to analyze the labora-
Figure 1. Disintegration of a claystone sample (CST-1) after exposure to natural climatic conditions.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 197–209
199
Shakoor and Gautam
Figure 2. Procedure for determining DR using grain size distribution curves of slaked material (Erguler and Shakoor, 2009b).
tory test and DR data. First, the data were tested for normality. For non-normal distributions, logarithmic and other transformations were applied (Davis, 2002). If a transformation was found to result in normal distribution, the transformed data were selected for the multivariate analysis. The data that did not show normal distribution, even after transformation, were used without transformation, and caution was exercised in interpreting the results. Bivariate analysis was used to determine the relationship between a pair of variables, whereas principal component analysis (PCA) was used to explain the variability within a data set. The Statistical Package for the Social Sciences (SPSS, 2011) software was used to perform the principal component analysis. PCA extracts the components that explain the greatest variability so that the influencing properties can be described. Multivariate regression uses a combination of independent variables to predict the dependent variable. In this study, DR was used as the dependent variable, whereas geologic and index properties were used as the independent variables. Different regression models were developed to describe the disintegration behavior in terms of DR values after 1, 6, and 12 months of exposure to natural climatic conditions. GEOLOGIC, INDEX PROPERTY, AND DR DATA Geologic Properties—Clay Mineralogy Table 1 presents the total amount of clay as well as percentages of expandable versus non-expandable clay minerals in the rocks studied. Claystones have the
200
highest mean amounts of 0.002-mm and 0.004-mm size clay particles—50 percent (range 38–77 percent) and 62 percent (range 51–82 percent), respectively—whereas siltstones have the least amount of 0.002-mm and 0.004-mm size clay particles—17 percent (range 12–21 percent) and 23 percent (range 22–27 percent), respectively. Mudstones and shales have intermediate amounts of clay size particles, with mudstones having slightly higher percentages (0.002 mm: 25 percent; 0.004 mm: 37 percent) than shales (0.002 mm: 21 percent; 0.004 mm: 27 percent). Clay minerals present in the rock samples were grouped into two categories: expandable and nonexpandable (Table 1). Expandable clay minerals in the study samples consist of mixed-layer clay minerals such as illite-smectite and montmorillonite, whereas non-expandable clay minerals include illite, kaolinite, and chlorite, with kaolinite being the most common. Index Properties Index property data, including natural water content, dry density, void ratio, absorption, adsorption, Atterberg limits, and Id2, are presented in Table 2. Claystones have the lowest mean value of dry density and the highest mean value of void ratio, 2.71 Mg/m3 (145.5 lb/ft3) and 0.112, respectively. The mean values of dry density for mudstones, siltstones, and shales are 2.54 Mg/m3 (158.3 lb/ft3), 2.49 Mg/m3 (155.5 lb/ft3), and 2.49 Mg/m3 (155.2 lb/ft3), respectively, whereas mean values of void ratio for these three rock types are 0.071, 0.084, and 0.099, respectively. The mean absorption values are 46.5 percent for claystones, 18.3 percent for mudstones, 5.0 percent for siltstones, and 10.4 percent for shales.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 197–209
Disintegration of Clay-Bearing Rocks Table 2. Natural water content, dry density, void ratio, absorption, adsorption, Atterberg limits, and second-cycle slake durability index data for clay-bearing rocks. Atterberg Limits Sample No.
Water Content (%)
Dry Density (pcf)
CST-1 CST-2 CST-3 CST-4 CST-5 MST-1 MST-2 MST-3 MST-4 MST-5 SLT-1 SLT-2 SLT-3 SLT-4 SLT-5 SHL-1 SHL-2 SHL-3 SHL-4 SHL-5
4 14.5 19.5 1.3 10.9 5.3 9.2 2.3 2.3 4.2 1.5 1.2 1.2 1 4.1 2.6 2.5 7 2.8 1.3
157 139.2 139.8 160 131.3 — — 164.3 153.4 157.2 159.3 160.3 160.1 148.4 149.5 144.5 169.6 149.8 153.3 158.7
Void Ratio
Absorption (%)
Adsorption (%)
Liquid Limit
Plastic Limit
Plasticity Index
Slake Durability Index (Id2)
0.04 0.21 0.11 0.06 0.31 — — 0.02 0.13 0.06 0.08 0.06 0.07 0.10 0.11 0.17 0.02 0.10 0.10 0.11
44.0 62.2 45.8 14.1 66.3 5.1 31.6 10.8 7.5 36.7 7.7 3.8 3.1 1.9 8.8 4.2 5.3 24.5 14.6 3.2
3.9 10 5.2 2.5 5.6 — — 1.7 3 3.5 1.9 1.4 1.7 1.2 2.6 1.7 2.1 3.5 3.3 1.3
28.1 55.1 48.4 32.7 75.0 29.7 26.5 21.5 26.7 28.2 22.3 21.9 21.8 26.1 20.8 24.3 26.6 34.0 30.8 27.8
20.4 21.3 28.9 23.4 30.0 21.3 20.7 14.8 19.8 19.8 17.2 14.6 15.4 20.1 15.7 20.6 24.8 27.6 18.4 23.3
7.7 33.8 19.5 9.3 45.0 8.4 5.8 6.7 6.9 8.4 5.1 7.3 6.4 6.0 5.1 3.7 1.8 6.4 12.4 4.5
80.1 43.3 76.3 64.0 1.5 93.3 3.4 73.2 95.2 75.7 96.7 99.2 95.0 99.4 86.9 99.0 95.8 28.0 95.8 97.7
CST 5 claystone; MST 5 mudstone; SLT 5 siltstone; SHL 5 shale; pcf 5 1b/ft3; 62.4 pcf 5 1 Mg/m3.
The mean adsorption value is the highest for claystones (5.4 percent) and the lowest for siltstones (1.8 percent). Claystones exhibit the highest mean value of liquid limit (47.9) and siltstones the lowest value (22.6). Mudstones and shales have intermediate values. The mean values of plasticity index for claystones, mudstones, siltstones, and shales are 23.1, 7.3, 6.0, and 5.8, respectively. Claystones have the lowest Id2 values, with a mean of 53.0 percent, whereas siltstones exhibit the highest values, with a mean Id2 of 95.4 percent. Disintegration Ratio Data Table 3 presents the monthly DR data for all samples exposed to natural climatic conditions. The mean values of DR after 1 month of exposure are 0.333, 0.499, 0.900, and 0.664 for claystones, mudstones, siltstones, and shales, respectively. Generally, all four types of clay-bearing rocks exhibit a wide variation in their monthly DR values. This variation in DR values can be attributed to the inherent variations in the geologic makeup of the rocks tested, especially the presence or absence of micro-fractures. Clay-bearing rocks vary within the same group (i.e., claystones, mudstones, siltstones, shales) and within the same sample.
STATISTICAL ANALYSIS OF PROPERTIES INFLUENCING DISINTEGRATION BEHAVIOR Bivariate regression analysis, PCA, and multivariate regression analysis were performed to identify the index properties that have the greatest influence on the amount of disintegration of clay-bearing rocks under natural climatic conditions. Bivariate Regression Analysis Table 4 shows the bivariate correlation matrix between DR and geologic and index properties, including clay content (0.002-mm and 0.004-mm size material), amounts of expandable and non-expandable clay minerals, natural water content, dry density, void ratio, absorption, adsorption, liquid limit, plastic limit, plasticity index, and Id2. Table 5 shows the correlation between selected geologic and index properties of the rocks studied and their DR values after 1, 6, and 12 months of exposure. As can be seen from the table, absorption exhibits the highest correlation coefficient (R2) values for the three periods of exposure: 0.66, 0.64, and 0.62, respectively, indicating that absorption is a better indicator of the variability in DR than are other properties (adsorption, total clay content, expandable clay content, and plasticity index). This may be due to the fact that absorption reflects both clay content and texture
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 197–209
201
Shakoor and Gautam Table 3. DR values for clay-bearing rocks exposed to natural climatic conditions for 1 to 12 months. Disintegration Ratio for Months (M) 1–12
Sample No.
M1
M2
M3
M4
M5
M6
M7
M8
M9
M10
M11
M12
CST-1 CST-2 CST-3 CST-4 CST-5 MST-1 MST-2 MST-3 MST-4 MST-5 SLT-1 SLT-2 SLT-3 SLT-4 SLT-5 SHL-1 SHL-2 SHL-3 SHL-4 SHL-5
0.312 0.229 0.850 0.197 0.079 0.715 0.095 0.597 0.941 0.147 0.751 0.993 0.994 0.991 0.772 0.988 0.883 0.281 0.198 0.969
0.172 0.029 0.851 0.079 0.064 0.719 0.018 0.528 0.721 0.104 0.329 0.992 0.996 0.997 0.574 0.985 0.666 0.160 0.081 0.911
0.116 0.036 0.889 0.053 0.019 0.438 0.004 0.269 0.759 0.037 0.626 0.769 0.991 0.984 0.121 0.978 0.511 0.107 0.042 0.831
0.085 0.027 0.600 0.596 0.005 0.417 0.003 0.193 0.648 0.040 0.796 0.859 0.994 0.978 0.060 0.972 0.466 0.070 0.041 0.700
0.074 0.017 0.323 0.050 0.007 0.376 0.004 0.136 0.468 0.023 0.105 0.792 0.980 0.993 0.039 0.976 0.437 0.062 0.104 0.892
0.060 0.014 0.155 0.069 0.020 0.475 0.001 0.109 0.381 0.031 0.145 0.486 0.990 0.997 0.037 0.966 0.391 0.048 0.034 0.604
0.061 0.012 0.154 0.029 0.010 0.350 0.003 0.095 0.694 0.149 0.895 0.610 0.996 0.998 0.050 0.973 0.397 0.041 0.019 0.498
0.050 — 0.095 0.016 0.004 0.190 0.001 0.086 0.363 0.018 0.974 0.771 0.984 0.998 0.028 0.934 0.344 0.039 0.008 0.605
0.045 — 0.049 0.011 0.003 0.416 — 0.078 0.518 0.114 0.866 0.963 0.989 0.974 0.028 0.971 0.329 0.016 0.006 0.454
0.038 — 0.072 0.003 0.000 0.337 — 0.086 0.381 0.011 0.436 0.964 0.980 0.989 0.015 0.961 0.377 0.010 0.011 0.526
0.042 — 0.072 0.001 — 0.584 — 0.076 0.380 0.154 0.364 0.922 0.968 0.989 0.009 0.951 0.306 0.014 0.004 0.324
0.042 — 0.022 0.376 — 0.208 — 0.079 0.416 0.110 0.914 0.828 0.938 0.990 0.006 0.950 0.328 0.011 0.005 0.831
CST 5 claystone; MST 5 mudstone; SLT 5 siltstone; SHL 5 shale.
(pore volume) (Dick and Shakoor, 1992). Figure 3 shows the correlation between DR, determined after 6 months of exposure, and the index properties of clay content, absorption, and adsorption. The R2 values for the three correlations are 0.40, 0.66, and 0.63, respectively, again indicating that absorption has the greatest influence on the variability of disintegration behavior. A higher value of absorption indicates that more of the void volume in the rock is filled with water. The water in the voids compresses air and interacts with the clay minerals, causing slaking and progressively breaking the rock (Olivier, 1979; Dick and Shakoor, 1992; and Czerewko and Cripps, 2001). Highly indurated rocks, such as siltstones, have lower values of absorption and higher values for density, specific gravity, and durability (DR). Correlation of DR with clay content (0.002 mm or 0.004 mm) shows that DR decreases with increasing clay content (Table 5 and Figure 3a). Clay-bearing rocks with greater clay content have not only higher absorption values but also higher adsorption values. Principal Component Analysis PCA extracts eigenvectors, called principal components, at successive levels from a variance-covariance matrix or a correlation matrix. It identifies the individual variables having common relationships and evaluates their contributions to the overall variance. Three different PCA tests were performed
202
on the relationships that geologic and index property data have with DR after 1, 6, and 12 months of exposure. For DR, after 1 month of exposure, 12 variables were used for PCA. These include clay content (,0.002 mm and ,0.004 mm), amount of nonexpandable clay minerals, amount of expandable clay minerals, natural water content, density, void ratio, absorption, adsorption, plasticity index, Id2, and DR. The analysis extracted three components, designated as components 1, 2, and 3 (Table 6). Component 1 alone accounts for 77.4 percent of the total variance in the data set, and components 1 and 2 together explain 94.1 percent of the total variance. Loadings for component 1 suggest that the most important variables contributing to the total variance in a data sets are clay-related properties such as clay content (loading for ,0.004 mm clay is 0.89 and for ,0.002 mm clay is 0.87); log of percent absorption (0.84); and log of plasticity index (0.82) (Table 6). Rescaled loadings are the correlation coefficients and are shown in parentheses for different variables. Variables with negative loadings suggest inverse relationships. For example, density, Id2, and log of DR after 1 month show a negative relationship with component 1. Further, an increase in the clay content, plasticity index, or absorption results in a decrease in values for variables such as Id2, density, and log of DR, their loadings being 20.93, 20.67, and 20.73, respectively.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 197–209
0.81 0.79 0.71 20.49 20.64 20.18 20.57 0.58 20.62 20.73 20.64 20.52 1 20.54 20.42 20.3 0.78 0.8 0.64 0.67 20.74 0.67 0.67 0.78 1 20.52 20.56 20.7 20.69 0.93 0.9 0.69 0.81 20.64 0.58 0.81 1 0.78 20.64 Log (DR of Mx) 5 log of disintegration ratio of a rock sample after 1, 6, 12 months of exposure, Log (PI) 5 log of plasticity index.
20.78 20.8 20.7 0.75 0.83 0.47 0.73 20.5 0.38 1 0.81 0.67 20.73 20.41 20.32 0 0.48 0.49 0.39 0.52 20.89 1 0.38 0.58 0.67 20.62 0.36 0.37 0.32 20.52 20.56 20.41 20.72 1 20.89 20.5 20.64 20.74 0.58 Log (DR of M1) Log (DR of M6) Log (DR of M12) % ,0.002 mm clay % ,0.004 mm clay Log (Exp. clay content) % Water content Density (pcf) Void ratio Log (% Absorp.) % Adsorp. Log(PI) Id2
1 0.86 0.66 20.49 20.63 20.34 20.36 0.36 20.41 20.78 20.56 20.54 0.81
0.86 1 0.87 20.46 20.61 20.17 20.48 0.37 20.32 20.8 20.7 20.42 0.79
0.66 0.87 1 20.28 20.48 20.12 20.55 0.32 0 20.7 20.69 20.3 0.71
20.49 20.46 20.28 1 0.94 0.58 0.64 20.52 0.48 0.75 0.93 0.78 20.49
20.63 20.61 20.48 0.94 1 0.44 0.72 20.56 0.49 0.83 0.9 0.8 20.64
20.34 20.17 20.12 0.58 0.44 1 0.36 20.41 0.39 0.47 0.69 0.64 20.18
20.36 20.48 20.55 0.64 0.72 0.36 1 20.72 0.52 0.73 0.81 0.67 20.57
Void Ratio Log Log Log % ,0.002 mm % ,0.004 mm Log (Exp. % Water Density (DR of M1) (DR of M6) (DR of M12) Clay Clay Clay Content) Content (pcf)
Table 4. Correlation matrix for the geologic and index properties tested.
Log (% Absorp.)
% Adsorp. Log(PI)
Id2
Disintegration of Clay-Bearing Rocks
Component 2 accounts for 16.8 percent of the total variance. The important variables exhibiting a positive relationship with component 2 are Id2 (0.36), clay content of ,0.004-mm size (0.44), and non-expandable clay content (0.61). Variables indicating a negative relationship with this component are ,0.002-mm clay content (20.93), log of expandable clay content (20.04), and void ratio (20.15). Component 3 accounts for 4.12 percent of the total variance. Influence for this component is provided by void ratio, water content, and log of plasticity index, with component loadings of 0.51, 0.40, and 0.36, respectively. In the next step, DR after 1 month of exposure was replaced by DR values after 6 months and 12 months of exposure and PCA tests were performed. Important variables contributing to the variance explained by extracted components were found to be absorption, clay content, and log of plasticity index, with only a minor influence by void ratio and water content. Multivariate Regression Analysis Stepwise procedure, involving the forward selection and the backward elimination methods, was used to conduct multivariate regression analysis. DR, determined after 1, 6, and 12 months of exposure, was used as the dependent variable to evaluate the short-term (1-month) and long-term (6–12-month) disintegration behavior. For DR of samples exposed for 1 month, the regression analysis used log of DR as the dependent variable. The independent variables used were percent clay (,0.002 mm and ,0.004 mm), percent non-expandable clay, log of percent expandable clay, log of absorption, adsorption, density, void ratio, water content, log of plasticity index, and Id2. Sixteen different models were generated using the stepwise procedure. Table 7 provides the correlation coefficient (R) value, the R2 value, the variables involved, the adjusted R2 value, and the standard error of the estimate for each model. Based on the R2 and adjusted R2 values in Table 7, Model 12 was selected as the best model to predict the dependent variable (Log of DR after 1 month of exposure, as shown in Equation 1). As the number of variables included increases, the R2 value increases. However, the adjusted R2 is increased only if the new variable improves the regression model. Table 8 presents the results of regression, including the selected model summary, analysis of variance (ANOVA) table, and coefficients. The equation to predict DR for all clay-bearing rocks after 1 month of exposure is as follows:
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 197–209
203
Shakoor and Gautam Table 5. R2 values for correlation between DR and selected geologic and index properties. Months of Exposure (No.)
Absorption
Adsorption
Clay Content (%, ,0.004 mm)
Clay Content (%, ,0.002 mm)
Expandable Clay Content
Plasticity Index
1 6 12
0.66 0.64 0.62
0.46 0.46 0.65
0.51 0.37 0.36
0.39 0.36 0.45
0.31 0.19 0.41
0.27 0.16 0.17
Log of DR after 1 month of exposure~
Log of DR after 6 months of exposure~
3:201{0:638 Log (%Absorption) z0:078 (%Adsorption)
0:264{0:871 Log (%Absorption) ð1Þ
{0:324 Log (%Expandable Clay Content) {0:016 (Density){3:19 (Void Ratio) Eq. 1 reveals that variation of DR after 1 month of exposure is explained by five independent variables: expandable clay content, absorption, adsorption, density, and void ratio. The measured and predicted values for log of DR are plotted in Figure 4a. A plot of the corresponding standardized residuals is presented in Figure 4b to show that there is no clear trend exhibited by the residuals. Although claybearing rocks are highly variable, ranging from claystones of high clay content to siltstones or shales of low to very low clay content, the model produces a significant R2 value of 0.81. This suggests that the DR is an appropriate indicator of disintegration behavior and that its variability can be explained by the five variables listed above. The multivariate regression discussed above was also used to investigate the geologic and index properties that had the greatest influence on the amount of disintegration after 6 and 12 months of exposure. The best models, with the highest R2 values of 0.70 and 0.52, respectively, resulted in the following regression equations:
z0:018 (%Adsorption)
ð2Þ
{0:413 Log (%Expandable Clay Minerals) Log of DR after 12 months of exposure~ 0:541{0:921 Log(%Absorption)
ð3Þ
{0:922 Log (%Expandable Clay Minerals) In order to verify the validity of selected models, the predicted values of log of DR after 6 and 12 months of exposure were plotted against the measured values, as shown in Figures 5 and 6, respectively. The figures also include plots of residuals. The data in Figures 5 and 6 show a relatively strong relationship between predicted and measured values in the case of 6 months of exposure (R2 5 0.76), but the relationship weakens for 12 months of exposure (R2 5 0.27). Eqs. 2 and 3 suggest that the 6month disintegration behavior of clay-bearing rocks under natural climatic conditions is influenced mostly by absorption, adsorption, and expandable clay content, but only absorption and expandable clay content influence the amount of disintegration after 12 months. Influence of Micro-Fabric on Disintegration Behavior SEM was used to study micro-fabric and incipient fractures at varying magnifications. Claystones, most
Figure 3. Relationship between DR, after 6 months exposure of samples to natural climatic conditions, and (a) absorption, (b) adsorption, and (c) clay content.
204
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 197–209
Disintegration of Clay-Bearing Rocks Table 6. Principal component matrix for DR, after 1 month of exposure to natural climatic conditions and other geologic and index properties. Raw Principal Components and their Loadings Variables S.N.
Variance Explained (%) Log(DR_M1) ,0.002 mm clay (%) ,0.004 mm clay (%) Log (Expandable clay content) Non-Expandable (%) Wn natural water content (%) Density (pcf) Void ratio Log (absorption) Adsorption % Log(PI) Id2
1 2 3 4 5 6 7 8 9 10 11 12
Rescaled Principal Components and their Loadings
1
2
3
1
2
3
77.36
16.76
4.16
77.36
16.76
4.16
20.26 13.9 15.11 0.29 8.4 3.94 26.75 0.05 0.42 1.92 0.29 227.08
0.07 7.65 7.55 20.02 7.23 1.02 0.48 20.01 0.06 0.9 0.05 10.47
0.02 20.33 1.24 0.05 23.18 2.16 27.08 0.04 0 0.31 0.13 1.64
20.73 0.87 0.89 0.65 0.7 0.74 20.67 0.66 0.84 0.86 0.82 20.93
0.19 0.48 0.44 20.04 0.61 0.19 0.05 20.15 0.12 0.4 0.15 0.36
0.07 20.02 0.07 0.11 20.27 0.4 20.71 0.51 0 0.14 0.36 0.06
Log (DR_M1) 5 log of disintegration ratio after 1 month of exposure; ,0.002 mm 5 % of clay content ,0.002 mm; ,0.004 mm 5 % of clay content ,0.004 mm; Log (Exp. clay content) 5 log of % of expandable clay content; Non-exp 5 % of non-expandable clay content; Log(PI) 5 log of plasticity index.
susceptible to rapid disintegration, show highly flocculated clay particles, with large voids between them and an abundance of smectite (Figure 7a). The mineral smectite is highly expandable upon exposure to moisture, contributing to the rapid disintegration. The claystone sample CST-5 in Figure 7a has a low DR of 0.019 after 6 months of exposure (Table 3).
For mudstones, SEM analysis revealed varying degrees of interlocking or flocculation of mineral grains and cementation. This corresponds to the variable nature of disintegration behavior exhibited by mudstones. Mudstone sample MST-3 has quartz overgrowth, with matrix support providing weaker interlocking of the grains and showing micro-fissures
Table 7. Multivariate regression models relating geologic and index properties to disintegration ratio after 1 month of sample exposure to natural climatic conditions. Model No.
R
R2
1 2 3 4 5
0.781 0.785 0.846 0.850 0.855
0.610 0.616 0.716 0.722 0.731
6
0.875
0.766
7
0.884
0.782
8
0.905
0.818
9
0.904
0.817
10
0.904
0.817
11
0.902
0.814
12
0.901
0.811
13 14 15 16
0.878 0.846 0.828 0.781
0.771 0.716 0.685 0.610
Variables Log (% Absorp.) Log (% Absorp.), % Adsorp. Log (% Absorp.), % Adsorp., Log (Exp. Clay Content) Log (% Absorp.), % Adsorp., Log (Exp. Clay Content), % ,0.002 mm Clay Log (% Absorp.), % Adsorp., Log (Exp. Clay Content), % ,0.002 mm Clay, Log(PI) Log (% Absorp.), % Adsorp., Log (Exp. Clay Content), % ,0.002 mm Clay, Log(PI), Id2 Log (% Absorp.), % Adsorp., Log (Exp. Clay Content), % ,0.002 mm Clay, Log(PI), Id2, Density (pcf), % ,0.004 mm Clay Log (% Absorp.), % Adsorp., Log (Exp. Clay Content), % ,0.002 mm Clay, Log(PI), Id2, Density (pcf), % ,0.004 mm Clay, Void Ratio Log (% Absorp.), % Adsorp., Log (Exp. Clay Content), Log(PI), Id2, Density (pcf), % ,0.004 mm Clay, Void Ratio Log (% Absorp.), % Adsorp., Log (Exp. Clay Content), Log(PI), Id2, Density (pcf), Void Ratio Log (% Absorp.), % Adsorp., Log (Exp. Clay Content), Id2, Density (pcf), Void Ratio Log (% Absorp.), % Adsorp., Log (Exp. Clay Content), Density (pcf), Void Ratio Log (% Absorp.), % Adsorp., Log (Exp. Clay Content), Void Ratio Log (% Absorp.), % Adsorp., Log (Exp. Clay Content) Log (% Absorp.), Log (Exp. Clay Content) Log (% Absorp.)
Adjusted R2
Standard Error of the Estimate
0.582 0.582 0.645 0.621 0.596
0.231 0.238 0.213 0.220 0.227
0.610
0.223
0.533
0.244
0.546
0.241
0.607
0.224
0.656
0.210
0.690
0.199
0.717
0.190
0.688 0.645 0.637 0.582
0.200 0.213 0.215 0.231
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 197–209
205
Shakoor and Gautam Table 8. Results of multivariate regression analysis relating geologic and index properties to disintegration ratio after 1 month of sample exposure to natural climatic conditions. Summary of Selected Model No. 12
R
R2
Adjusted R2
Standard Error of the Estimate
0.901
0.811
0.717
0.190
Sum of Squares
df
Mean Square
F
Significance
1.553 0.362 1.915
5 10 15
0.311 0.036 —
8.584 — —
0.002 — —
ANOVA Regression Residual Total
Unstandardized Coefficients Coefficients (Constant) Log(% Absorp.) % Adsorp. Log (Exp. Clay Content) Density (pcf) Void Ratio
Standardized Coefficients
B
Standard Error
Beta
t
Significance
3.201 20.638 0.078 20.324 20.016 23.190
1.898 0.182 0.045 0.159 0.011 1.476
— 20.888 0.491 20.405 20.461 20.654
1.686 23.501 1.753 22.043 21.458 22.161
0.123 0.006 0.110 0.068 0.175 0.056
along the grain contacts (Figure 7b). The sample has a DR value of 0.109 after 6 months of exposure. Siltstones are characterized by a high degree of interlocking and well-cemented quartz grains. However, fragmentation of some siltstone samples can generally be attributed to the presence of incipient microfractures. Shales exhibit well-defined mineral alignment (fissility), parallel to the direction of bedding (Figure 7c). Fissility contributes to cleaving, resulting in thin, sheet-like fragments upon disintegration. Siltyshales were found to be highly indurated as compared to
clayey shales. SEM analysis corroborates the disintegration behavior observed under natural climatic conditions, indicating that both texture and mineral composition are important factors influencing their disintegration behavior. DISCUSSION This study investigated the influence of geologic and index properties on the disintegration behavior of clay-bearing rocks under natural climatic conditions.
Figure 4. (a) Measured vs. predicted values of log of DR (Model 12) for all rocks after 1 month of exposure to climatic conditions; (b) plot of residuals.
206
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 197–209
Disintegration of Clay-Bearing Rocks
Figure 5. (a) Measured vs. predicted values of log of DR for all rocks after 6 months of exposure to climatic conditions; (b) plot of residuals.
The purpose was to determine the extent to which the amount of disintegration, quantitatively expressed as the DR, can be explained by geologic and index properties. Results show that rocks with high clay content, such as claystones, are more susceptible to disintegration upon interaction with water (average DR after 1 month of exposure 5 0.333; Table 3) because clay content influences the amount of water a rock absorbs or adsorbs. These findings are supported by previous research by Grainger (1984), Dick and
Shakoor (1992), Sarman et al. (1994) and Czerewko and Cripps (2001). SEM analysis revealed that rocks with large pore volume, flocculated texture, and micro-fractures, such as claystones, disintegrate more readily upon exposure to water than do rocks, such as siltstone, with a higher degree of interlocking and with the presence of well-cemented quartz. Molina et al. (2011) also found that large pore volume and flocculated texture facilitate disintegration upon exposure to water.
Figure 6. (a) Measured vs. predicted values of log of DR for all rocks after 12 months of exposure to climatic conditions; (b) plot of residuals.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 197–209
207
Shakoor and Gautam
Figure 7. SEM photomicrographs of (a) a claystone sample (CST-5) showing highly flocculated clay minerals and large voids between them (magnification 3700), (b) a mudstone sample (MST-3) showing large quartz grains and their overgrowth (magnification 3190), and (c) a silty-shale sample (SHL-1) showing a high degree of mineral alignment (fissility) and the presence of fractures (magnification 3900).
Mudstones show variable disintegration behaviors, with DR values ranging from 0.095 to 0.941 after 1 month of exposure (Table 3). The variable behavior of mudstones is most likely caused by their varying clay content, varying degree of compactness, and the presence or absence of micro-fractures. Siltstones are the most durable. They have mean DR values of 0.900 and 0.735 after 1 month and 12 months of exposure, respectively (Table 3). The main properties contributing to the high durability of siltstones are their highly interlocked texture, large amount of quartz, high degree of compactness, and small amount of clay. Shales show a wide range of disintegration behavior, with 1-month DR values ranging from 0.198 to 0.988 (Table 3). The disintegration of shales is influenced mainly by their degree of fissility, compactness, and clay content (i.e., whether the shale is silty or clayey in nature). The disintegration behavior of clay-bearing rocks depends on local climatic conditions and can vary significantly from one climatic region to another. Therefore, the results of this study will be applicable to climatic regions similar to Ohio (warm and humid summers, cold winters with frequent below-freezing temperature cycles).
clay content are better indicators of long-term (6– 12-month exposure) disintegration behavior. 2. DR is a better indicator of short-term (1 to 3– month) disintegration behavior than of long-term (6–12-month) behavior. 3. Micro-fabric, including micro-fractures, plays an important role in influencing the amount and rate of disintegration of clay-bearing rocks. Upon exposure to water, rocks such as claystones, with large pore volume, flocculated texture, and microfractures, disintegrate to a greater extent and more rapidly than do rocks such as siltstones, with a higher degree of interlocking and the presence of well-cemented quartz. 4. Mudstones and shales exhibit highly variable disintegration behavior, intermediate between that of claystones and siltstones. ACKNOWLEDGMENTS The authors would like to thank Dr. Ira Sasowsky, Co-Editor; Dr. Peter Hudec; Dr. Jeffrey Dick; and an anonymous reviewer for their valuable comments that helped improve the quality of this manuscript. The authors also express their appreciation to Karen Smith for her help in formatting the manuscript and with the submission process.
CONCLUSIONS Based on the results of this study, the following conclusions can be derived: 1. Clay content, percentage of expandable clay minerals, absorption, adsorption, density, void ratio, and second-cycle slake durability index are the best indicators of short-term (1 month of exposure) disintegration behavior under natural climatic conditions. Absorption and expandable
208
REFERENCES AMERICAN SOCIETY FOR TESTING and MATERIALS (ASTM), 1996, Annual Book of ASTM Standards 04.08: American Society for Testing Materials, West Conshohocken, PA, 950 p. BELL, F. G.; ENTWISLE, D. C.; AND CULSHAW, M. G., 1997, A geotechnical survey of some British coal measures mudstones, with particular emphasis on durability: Engineering Geology, Vol. 46, pp. 115–129. BROTO´NS, V.; TOMA´S, R.; IVORRA, S.; AND ALARCO´N, J. C., 2013, Temperature influence on the physical and mechanical
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 197–209
Disintegration of Clay-Bearing Rocks properties of a porous rock: San Julian’s Calcarenite: Engineering Geology, Vol. 167, pp. 117–127. BRYSON, S. L.; GOMEZ-GUTIERREZ, I. C.; AND HOPKINS, T. C., 2012, Development of a new durability index for compacted shale: Engineering Geology, Vol. 139–140, pp. 66–75. CZEREWKO, M. A. AND CRIPPS, J. C., 2001, Assessing the durability of mudrocks using the Modified Jar Slake Index Test: Quarterly Journal Engineering Geology Hydrogeology, Vol. 34, pp. 153–163. DAVIS, J. C., 2002, Statistics and Data Analysis in Geology, 3rd ed: John Wiley & Sons, New York, 638 p. DICK, J. C. AND SHAKOOR, A., 1992, Lithologic controls of mudrock durability: Quarterly Journal Engineering Geology, Vol. 25, pp. 31–46. DICK, J. C.; SHAKOOR, A.; AND WELLS, N., 1994, A geological approach toward developing a mudrock-durability classification system: Canadian Geotechnical Journal, Vol. 31, No. 1, pp. 17–27. DOOSTMOHAMMADI, R.; MOOSAVI, M.; AND ARAABI, B. N., 2008, A model for determining the cyclic swell-shrink behavior of argillaceous rock: Applied Clay Science, Vol. 42, pp. 81–89. ERGULER, Z. A. AND SHAKOOR, A., 2009a, Relative contribution of climatic processes in disintegration of clay-bearing rocks: Engineering Geology, Vol. 108, No. 1–2, pp. 36–42. ERGULER, Z. A. AND SHAKOOR, A., 2009b, Quantification of fragment size distribution of clay-bearing rocks after slake durability testing: Environmental Engineering Geoscience, Vol. 15, No. 2, pp. 81–89. FAHIMIFAR, A. AND SOROUSH, H., 2007, A moisture index classification system for rocks: Rock Mechanics Rock Engineering, Vol. 40, No. 1, pp. 63–79. GAUTAM, T. P. AND SHAKOOR, A., 2013, Slaking behavior of claybearing rocks during a one-year exposure to natural climatic conditions: Engineering Geology, Vol. 166, pp. 17–25. GEMICI, U., 2001, Durability of shales in Narlidere Izmir, Turkey, with an emphasis on the impact of water on slaking behavior: Environmental Geology, Vol. 41, pp. 430–439. GO¨KC¸EOG˘LU, C.; ULUSAY, R.; AND SO¨NMEZ, H., 2000, Factors affecting durability of the weak and clay-bearing rocks selected from Turkey with particular emphasis on the influence of the number of drying and wetting cycles: Engineering Geology, Vol. 57, pp. 215–237. GRAINGER, P., 1984, The classification of mudrocks for engineering purposes: Quarterly Journal Engineering Geology, Vol. 17, pp. 381–387. HAJDARWISH, A.; SHAKOOR, A.; AND WELLS, N. A., 2013, Investigating statistical relationships among clay mineralogy, index engineering properties, and shear strength parameters of mudrocks: Engineering Geology, Vol. 159, pp. 45–58. 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: International Society for Rock Mechanics, Ankara, Turkey, 628 p. KONCAGU¨L, E. C. AND SANTI, P. M., 1999, Predicting the unconfined compressive strength of the Breathitt Shale using slake durability, shore hardness and rock structural properties: International Journal Rock Mechanics Mining Sciences, Vol. 36, No. 2, pp. 139–153. MOLINA, E.; CULTRONE, G.; SEBASTIA´N, E.; ALONSO, F. J.; CARRIZO, L.; GISBERT, J.; AND BUJ, O., 2011, The pore system of sedimentary rocks as a key factor in the durability of buitrun "delete_author_querieslding materials: Engineering Geology, Vol. 118, pp. 110–121.
MORADIAN, Z. A.; GHAZVINIAN, A. H.; AHMADI, A.; AND BEHNIA, M., 2010, Predicting slake durability index of soft sandstone using indirect tests: International Journal Rock Mechanics Mining Sciences, Vol. 47, pp. 666–671. NAHAZANAN, H. B.; CLARKEA, S.; ASADIB, A.; YUSOFFB, Z. M.; AND HUATB, B. K., 2013, Effect of inundation on shear strength characteristics of mudstone backfill: Engineering Geology, Vol. 158, pp. 48–56. OLIVIER, H. J., 1979, A new engineering-geological rock durability classification: Engineering Geology, Vol. 14, pp. 255–279. PEJON, O. J. AND ZUQUETTE, L. V., 2002, Analysis of cyclic swelling of mudrocks: Engineering Geology, Vol. 67, pp. 97–108. POTTER, P. E.; MAYNARD, J. B.; AND PRYOR, W. A., 1980, Sedimentology of Shale: Springer-Verlag, New York, 306 p. RUSSELL, D. J., 1981, Controls on shale durability: The response of two Ordovician shales in the slake durability test: Canadian Geotechnical Journal, Vol. 9, pp. 1–13. SADISUN, I. A.; SHIMADA, H.; ICHINOSE, M.; AND MATSUI, K., 2005, Study on the physical disintegration characteristics of Subang claystone subjected to a modified slaking index test: Geotechnical Geological Engineering, Vol. 23, pp. 199–218. SANTI, P. M., 1998, Improving the jar slake, slake index, and slake durability test for shales: Environmental Engineering Geoscience, Vol. 4, No. 3, pp. 385–396. SANTI, P. M., 2006, Field methods for characterizing weak rock for engineering: Environmental Engineering Geoscience, Vol. 12, No. 1, pp. 1–11. SANTI, P. M. AND DOYLE, B. C., 1997, The locations and engineering characteristics of weak rock in the U.S. In Santi, P. M. and Shakoor, A. (Editors), Characterization of Weak and Weathered Rock Masses: Association of Engineering Geologists, Special Publication, No. 9, pp. 1–22. SANTI, P. M. AND HIGGINS, J. D., 1998, Methods for predicting shale durability in the field: Geotechnical Testing Journal, Vol. 21, No. 3, pp. 195–202. SARMAN, R.; SHAKOOR, A.; AND PALMER, D. F., 1994, A multiple regression approach to predict swelling in mudrocks: Bulletin Association Engineering Geologists, Vol. 31, No. 1, pp. 107–121. SHAKOOR, A. AND SARMAN, R., 1992, Swelling potential classification of mudrocks based on geotechnical properties. In Hudson, J. A. (Editor), Rock Characterization: ISRM Symposium Eurock’92, England, U.K., pp. 75–80. SINGH, T. N.; VERMA, A. K.; SINGH, V.; AND SAHU, A., 2005, Slake durability study of shaly rock and its predictions: Environmental Geology, Vol. 47, pp. 246–253. STATISTICAL PACKAGE FOR THE SOCIAL SCIENCES (SPSS), 2011, IBM SPSS Statistics For Windows, Version 20.0: SPSS, Armonk, NY. SUMMA, V.; TATEO, F.; GIANNOSSI, M. L.; AND BONELLI, C. G., 2010, Influence of clay mineralogy on the stability of landslide in Plio-Pleistocene clay sediments near Grassano (Southern Italy): Catena, Vol. 80, pp. 75–85. WENG, M. C. AND LI, H. H., 2012, Relationship between the deformation characteristics and microscopic properties of sandstone explored by the bonded-particle model: International Journal Rock Mechanics Mining Sciences, Vol. 56, pp. 34–43. WEST, T. R. AND SHAKOOR, A., 1984, Influence of petrography of Argillaceous carbonates on their frost resistance in concrete: Cement, Concrete, Aggregates, American Society for Testing and Materials, Vol. 6, pp. 84–89. WUDDIVIRA, M. N.; EKWUE, E. I.; AND STONE, R. J., 2010, Modeling slaking sensitivity to assess the degradation potential of humid tropic soils under intense rainfall: Land Degradation Development, Vol. 21, pp. 61–73. YOUN, H. AND TONON, F., 2010, Effect tf air-drying duration on the engineering properties of four clay-bearing rocks in Texas: Engineering Geology, Vol. 115, pp. 58–67.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 197–209
209
Seismic Source Characterization for Greater Phoenix Area Earthquake Hazard SIMON T. GHANAT1 Department of Civil and Environmental Engineering, The Citadel, 171 Moultrie Street, Charleston, SC 29409, 843-953-7692, sghanat@citadel.edu
EDWARD KAVAZANJIAN, JR. Arizona State University, School of Sustainable Engineering and Built Environment, Mail Code 5306, Tempe, AZ 85251, edward.kavazanjian@asu.edu
RAMON ARROWSMITH Arizona State University, School of Earth and Space Exploration, Office 773, ISTB4 – Bldg. 75, Tempe, AZ 85251, ramon.arrowsmith@asu.edu
Key Terms: Quaternary Faults, Seismic Source Characterization, Probabilistic Seismic Hazard Analysis
ABSTRACT We have developed a model for active seismogenic faults for the greater Phoenix area. The Quaternary geologic record was used to delineate the local seismic sources, assess potential earthquake size, and estimate earthquake recurrence to support an integrated regional probabilistic evaluation of the seismic hazard. The probabilistic seismic hazard analysis is summarized as uniform hazard spectra and contour maps of peak ground acceleration and spectral acceleration at one second for weak rock site conditions at two percent probability of exceedance in 50 years. As building codes typically allow some level of reduction in design spectral accelerations from values obtained from the National Seismic Hazard Maps, we compared our results to values obtained from the National Seismic Hazard Maps. No reduction in building code seismic design requirements is justified in the areas addressed herein on the basis of our site specific analysis. INTRODUCTION Assessment of the seismic hazard in areas of relatively low seismicity and modest geologic evidence for recent deformation, such as the greater Phoenix area (Figure 1), is often given little attention. However, when factors such as high population density, a large number of critical facilities, and building code requirements are considered, it becomes apparent that accurate seismic 1
Corresponding author email: sghanat@citadel.edu
assessments are desirable for large urban regions in such seismic regimes. Furthermore, increases in the return period for the design ground motions specified by the International Building Code (IBC, 2012) from 500 years to approximately 1,000 years (i.e., to two-thirds of the ground motions for a 2,500 year return period) have resulted in more stringent seismic design requirements in many areas, including greater Phoenix. However, the IBC allows reductions of up to 20 percent in the design ground motion values derived from the seismic hazard maps produced by the United States Geological Survey (USGS) under the National Earthquake Hazard Reduction Program (NEHRP) and that are employed by the IBC, if justified by site-specific analysis (IBC, 2012). In some cases, even small changes in ground motion values can result in a change in Seismic Design Category, which in turn, changes the seismic design requirements specified by the building code (IBC, 2012). The recent increases in the return period for seismic ground motions in the IBC have, in fact, increased seismic design requirements for many structures in the greater Phoenix area based upon the NEHRP maps. In particular, the increase in design ground motions have moved critical facilities (Occupancy Class III, e.g. hospital and hazardous material storage facilities) in the greater Phoenix area from Seismic Design Category (SDC) A, requiring no seismic design to SDC-C, requiring lateral force analyses (SDC-B does not apply to Occupancy Class III). Even a 10 percent decrease in design ground motions would drop design requirements for critical facilities back to SDC-A (IBC, 2012). The NEHRP probabilistic ground motion maps (Petersen et al., 2014) for a two percent probability of exceedance in 50 years, corresponding to ground motions with a return period of approximately 2500 years, the basis for current IBC seismic design
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 211–222
211
Ghanat, Kavazanjian, and Arrowsmith
Figure 1. (A) Active faults and seismicity of Arizona overlain on shaded relief. White rectangle indicates the location of B. (B) Active faults, seismicity, cities, highways, and railroads for north-central Arizona. U.S. Geological Survey and Arizona Geological Survey, 2010.
requirements (IBC, 2012), are not based on a detailed description of the local faults in the greater Phoenix area. (As noted previously, IBC employs two-thirds of the ground motion with a two percent probability of exceedance in 50 years as the design basis ground motion). Therefore, we conducted an assessment of the seismic potential for the faults capable of causing strong ground motions in the Phoenix area. This assessment was employed in a state-of-the-practice probabilistic seismic hazard analysis (PSHA) for the Phoenix area to determine if, as recommended by some local consultants, reductions in spectral accelerations from the NEHRP-mapped levels were justified. We conducted a geologic and seismologic evaluation of existing literature and data to characterize the seismic sources capable of generating strong ground motions
212
in the greater Phoenix area. Results of the PSHA described herein do not justify a reduction in Seismic Design Category. This paper describes the seismic source characterization as well as the associated PSHA, conducted to arrive at this conclusion. Seismic Setting The greater Phoenix area is located between the more tectonically active regions of the North American-Pacific plate boundary in California to the west, the Northern Arizona Basin and Range province to the north, and New Mexico’s Rio Grande Rift to the east. It is on the edge of the Arizona Transition Zone which is a northwest trending belt of similar geology and physiography separating the southern Basin and
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 211–222
Phoenix Seismic Source Characterization
Range province to the southwest from the Colorado Plateau to the northeast. The continental lithosphere in the Basin and Range province has been stretched and thinned, resulting in a distinctive physiography of narrow mostly north-south trending mountain ranges bounded by active or inactive normal faults separated by broad, sediment-filled basins (Menges and Pearthree, 1989). This extension was in two phases (ENEWSW mostly along low angle normal faults [, 30– 17 Ma] followed by slip along steep N–S normal faults [13–8 Ma principally]; (Eberly and Stanley, 1978; Reynolds, 1985; Menges and Pearthree, 1989). Widespread and rapid extension in the area of greater Phoenix ended about 8 Ma. Some faults in the central transition zone and adjacent Basin and Range have either been reactivated during the Quaternary or continued to be active at lower rates (Menges and Pearthree, 1989). Seismic Sources The known active seismic sources that may affect greater Phoenix are the Carefree, Horseshoe, Sugarloaf, Verde, Sand Tank, and Cottonwood Basin Fault (Figure 2). Based on the U.S. Geological Survey and Arizona Geological Survey (2010) database, the closest active seismic source to the greater Phoenix area is the Carefree Fault Zone. The Carefree Fault has a total length of 11 km; where exposed it shows middle and late Quaternary activity. The length of the Horseshoe Fault Zone is approximately 20 km and it shows Holocene activity. The length of the Sugarloaf Fault is only about 9 km and it reveals late Quaternary activity. The length of the Verde Fault Zone is 10 km and this fault indicates late Quaternary activity. The Cottonwood Fault Zone displays middle and late Quaternary activity and has a length of about 5 km. The Sand Tank Fault indicates late Quaternary activity and is approximately 5 km long. In addition to these known fault sources, a random area source (e.g. floating) was used in the PSHA to account for earthquake sources that had not been identified. Historical seismicity and regional deformation not assigned to active faults were assigned to the random area source. A description of the characteristics of each of these fault zones follows. Carefree Fault Zone The Carefree Fault Zone is located in the northeastern part of the Phoenix Basin, a complex physiographic basin near the Sonoran Desert subprovince of the Basin and Range province. The Carefree Fault Zone consists of a series of northwest trending normal faults within and along the western
margin of an extensive bedrock pediment formed on Precambrian granite (Pearthree and Scarborough, 1984; Skotnicki et al., 1997). The Carefree Fault Zone has a slip rate of 0.01 mm/yr based on the topographic relief of less than 3 m across the fault and the faulting of inferred middle Pleistocene deposits (Pearthree, 1998; Pearthree and Scarborough, 1984; Skotnicki et al., 1997). The slip rate of 0.01 mm/yr as stated in the literature was assigned to the Carefree Fault Zone in this study. Horseshoe Fault Zone The Horseshoe Fault Zone is located in the upland portion of the Basin and Range province in central Arizona (Figure 2). The Horseshoe Fault Zone defines the southern and western margins of the small, dissected Horseshoe basin between the Mazatzal and Humboldt Mountains (Hells Canyon section). The basin is probably an asymmetric graben, with the Horseshoe Fault being the master fault (U.S. Geological Survey and Arizona Geological Survey, 2010). The interior of the basin is divided into several sections. The mountain front associated with the Hells Canyon section is linear with less than 5-m-high scarps. The basin has been deeply dissected in response to down-cutting of the Verde River, which flows through it (Pearthree and Scarborough, 1984). Middle Pleistocene Verde River alluvium is faulted down-to-the-northeast by 1.5–2 m. Two or three faulting events have occurred in the past 300 ka; the youngest faulting event is about 10 to 20 ka (Piety and Anderson, 1991). According to Pearthree (1998), the slip rate of the Hells Canyon section is less than 0.03 mm/yr and the slip rate of the Hells Canyon section is 0.01 mm/yr. The slip rates stated in the Pearthree (1998) were assigned to the Horseshoe Fault Zone in this study. Sugarloaf Fault Zone The Sugarloaf Fault Zone is located in the eastern portion of Phoenix Basin (Figure 2). This normal fault defines the western margin of a small basin that represents a shallow, asymmetric graben, just west of the Mazatzal Mountains (Pearthree and Scarborough, 1984). The Sugarloaf Fault forms low but fairly sharply-defined, east-facing scarps as much as five meters high at the boundary between weathered Precambrian granite and Tertiary basin-fill sediment (Pearthree et al., 1983). Pearthree (1998) stated that slip rate is 0.01–0.02 mm/yr based on less than 1 m of vertical displacement in the past 50–100 ka. The slip rate of 0.01–0.02 mm/yr as stated in the Pearthree
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 211–222
213
Ghanat, Kavazanjian, and Arrowsmith
Figure 2. Historic (brown dots) and instrumental (purple dots) seismicity and active faults are shown within 100 km of one of the cities in greater Phoenix area (Scottsdale; blue polygon). Blue curves show 50 and 100 km buffer around city of Scottsdale. Most historic seismic events have occurred to the north of Phoenix in the Prescott and southern Colorado Plateau areas. Active faults (orange, yellow, white traces) are short (mostly less than 10 km) and to the northeast of the metropolitan area (except Sand Tank Fault). Seismicity is from Arizona Earthquake Information Center, 2011, and the active faults are from USGS and Arizona Geologic Survey, 2010. See Figure 1 for an explanation of other symbols.
(1998) was assigned to the Sugarloaf Fault Zone in this study. Verde Fault Zone The Verde Fault, a northwest-striking Basin and Range normal fault, is located in the Basin and Range province near the margin of Colorado Plateaus (Figure 2). The Verde Fault forms the boundary between the Black Hills block and Verde Valley. The Verde Valley is a large, asymmetric, southwest-tilted graben. Subsidence of the Verde Basin and displacement along the Verde Fault occurred during the late Miocene and Pliocene periods (Bressler and Butler, 1978). The only clear evidence of Quaternary fault
214
motion exists along the southern part of the fault zone, where fault scarps as much as about 7 m high are formed on high, dissected, probable lower to middle Pleistocene alluvial fans (House, 1994). A low slip rate is inferred based on about five meters of vertical displacement of deposits estimated to be about 300 to 500 ka (Pearthree et al., 1983). The most recent deformation is considered late Quaternary and is based on morphologic scarp analysis and rough estimates for the age of faulted deposits. The morphologic analyses of scarp data suggest an early to middle Holocene time for the youngest movement (Pearthree et al., 1983). According to Pearthree (1998), the slip rate is 0.01–0.02 mm/yr based on 5 m of vertical displacement in the past 300–500 ka. The
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 211–222
Phoenix Seismic Source Characterization
slip rate of 0.01–0.02 as stated in the Pearthree (1998) was assigned to the Verde Fault Zone in this study. Cottonwood Basin Fault Zone The Cottonwood Basin Fault Zone, a normal fault in Verde Valley, is located in the Highlands portion of the Basin and Range province (Figure 2). The Cottonwood Basin Fault is a large, asymmetric, southwest-tilted graben. Most of the displacement along the fault occurred during the late Miocene and Pliocene (Bressler and Butler, 1978) as recorded by displacement of Tertiary volcanic rocks and accumulation of the Verde Formation sediments in Verde Valley. This fault zone shows middle and late Quaternary activity with slip rate of less than 0.01 mm/yr (Pearthree, 1998). Pearthree (1998) also states that that slip rate is less than 0.01 mm/yr based on 2–3 m vertical displacement in 1 Ma. The slip rate of 0.002–0.003 mm/yr from Pearthree (1998) was assigned to the Cottonwood Basin Fault Zone in this study. Sand Tank Fault Zone The Sand Tank Fault is a down-to-west, normal fault on the east side of the broad Gila Valley in south-central Arizona (Figure 2). It is located in the heart of the Sonoran Desert sub-province of the Basin and Range province, which is characterized by low, pediment, deeply embayed mountain fronts indicative of long-term tectonic stability (Schell and Wilson, 1982). The fault vertically displaces middle to upper Pleistocene alluvium by less than two meters, and latest Pleistocene and Holocene alluvium is not faulted (Demsey and Pearthree, 1990). The total length of the fault is about 5 km with a dip of 45–90 NW (U.S. Geological Survey and Arizona Geological Survey, 2010). Pearthree (1998) states that that slip rate is 0.01–0.03 mm/yr based on 1.5–2 m of vertical displacement in the past 70–200 ka. The slip rate of 0.01–0.03 as stated in the Pearthree (1998) was assigned to the Sand Tank Fault Zone in this study. FAULT CHARACTERIZATION Seismic source characterization requires evaluation of three fundamental characteristics of regional seismicity: (1) the identification, location, and geometry of sources of potentially damaging earthquakes (seismic source identification); (2) the maximum size of the earthquake associated with these sources (maximum, or characteristic magnitude); and (3) the rate at which earthquakes occur along these sources (recurrence) (Reiter, 1990; Kramer, 1996).
Each fault zone in this study was defined by its surface trace coordinates (latitudes and longitudes), mode of faulting, orientation (depth and dip angles), minimum and maximum moment magnitudes (all magnitudes reported in this paper are moment magnitudes) for use in seismic hazard analysis, and two recurrence models (data came from U.S. Geological Survey and Arizona Geological Survey, 2010). The fault depths had to be assumed, due to a lack of significant instrumental seismicity. A focal depth of 10 km was assumed for all the fault sources based upon an average of all instrumentally recorded earthquake depths located within 50 km of the local fault sources. A sensitivity analysis on the focal depth value was performed in the PSHA. It was found that changing the focal depth of faults from 10 km to 14 km caused a negligible variation in the spectral acceleration values. The sense of movement of each fault source and area source was taken as normal, consistent with the style of faulting in the region. Doser and Smith (1989) determined that the dip of normal faults in the western United States ranges from 40 to 90 degrees, with an average dip value of 60 degrees. Therefore, a dip value of 60 degrees was assumed for each seismic source and the dip directions were taken from the geological literature (Pearthree and Scarborough, 1984). The widths of the faults were determined from the focal depth and the dip of the fault, so that width equaled 10 km divided by the sine of the dip. It was assumed that the fault ruptured all the way to the ground surface from 10 km depth. Minimum and maximum moment magnitudes were assigned to each source. Two different magnitude recurrence relationships were analyzed for each source: a truncated Gutenberg–Richter (G–R) exponential model (Gutenberg and Richter, 1954), and a characteristic earthquake model (Schwartz and Coppersmith, 1984). The seismic hazard analysis employed both models, with a weight of 50 percent applied to each model. A minimum moment magnitude of 4.6, considered to be the smallest magnitude of engineering interest, was assigned to each fault. The Pearthree (1998) slip rate estimates, which are based on field evidence, were assigned to the seismic sources in the PSHA. The uncertainty in slip rates for all fault sources, except Carefree Fault Zone, was addressed by using three slip rate values: an upper bound value, an average value, and a lower bound value, all equally weighted in the PSHA. Estimation of Maximum Magnitude from Geologic Data The maximum earthquake magnitude is an important parameter for both characteristic and truncated
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 211–222
215
Ghanat, Kavazanjian, and Arrowsmith Table 1. Summary of fault characteristics for the greater Phoenix area. All faults have been assigned a normal sense of slip, 60u dip, 10 km maximum rupture depth, and 4.6 minimum magnitude. Length and age of latest movement compiled from U.S. Geological Survey and Arizona Geological Survey 2010. Slip rates compiled from Pearthree, 1998; Maximum magnitudes from Wells and Coppersmith, 1994. Fault Sources
Length (km)
Age of Latest Movement
Slip Rate (mm/yr)
Mmax
Horseshoe Carefree Verde Sugarloaf Cottonwood Sand Tank
20 11 10 9 5 5
Late Quaternary (,130 ka) Middle & Late Quaternary Late Quaternary Late Quaternary Middle & Late Quaternary Late Quaternary
0.01–0.03 0.01 0.01–0.02 0.01–0.02 0.002–0.003 0.01–0.03
6.6 6.3 6.2 6.1 5.9 5.9
Gutenberg–Richter recurrence models. Preferably, it would be estimated on the basis of either an extensive historical catalog of earthquakes or comprehensive paleoseismic data. However, due to the shortage of historical seismicity, the maximum magnitude cannot be determined reliably by either of these methods in the Phoenix region. The surface rupture length of a fault can been used to estimate the moment magnitude of an earthquake. A number of studies (e.g., Tocher, 1958; Bonilla et al., 1984; and Wells and Coppersmith, 1994) have illustrated the general nature of the relationship between fault rupture length and earthquake magnitude. The Wells and Coppersmith, 1994 empirical relations are based on more comprehensive data set than previous studies and are globally relevant as opposed to more plate boundary-oriented datasets. The Wells and Coppersmith (1994) for all fault types was selected over the Wells and Coppersmith (1994) for the normal faults only for the following reasons: (1) the regression for all fault types is based upon a larger sample size and (2) it also has a smaller standard error than the empirical relationship for the normal faults only. The empirical relationship of Wells and Coppersmith (1994) relating surface rupture length (SRL) to the earthquake magnitude (M) for all fault types, presented in Equation 1 was used to calculate the moment magnitude (M) for each of the known faults.
the end is mapped, but propagation of faults and ruptures at very low slip rates might not be preserved in the landscape.
M~5:08z1:16 logðSRLÞ
where N(M) is the number of earthquakes with moment magnitude greater than M occurring in a given time. The b value describes how the number of earthquakes varies with magnitude and the a value represents the baseline activity rate (Kramer, 1996). The b value of the G–R relationship derived from simple statistical counts of historical events was considered inadequate due to insufficient historical seismicity data in Central Arizona. Values of the b parameter developed from worldwide averages of earthquake data generally fall within the range of 0.9 to 1.0 (Reiter, 1990). Lockridge et al., 2012 analyzed a large dataset of events recorded by the Earth Scope Transportable Array (TA) and determined a b value
ð1Þ
Table 1 presents the maximum magnitude derived using Equation 1 for each fault employed in this study along with maximum fault length (assuming all fault segments rupture together), age of latest movement, and fault slip rates. As shown in Table 1, Equation 1 resulted in maximum moment magnitudes of 6.6 for the Horseshoe Fault, 6.3 for the Carefree Fault, 6.2 for the Verde Fault, 6.2 for the Sugarloaf Fault, 5.9 for the Sand Tank Fault, and 5.9 for the Cottonwood Fault. We recognize that these mapped lengths of fault could link up as they rupture, and the ends of some sections may not terminate exactly where
216
Magnitude-Recurrence Relations As noted previously, the annual number of earthquakes of various magnitudes that were assigned to each fault was based on the slip rate information and was defined using a combination of two different distributions: (1) a truncated exponential model that implies that earthquakes on a given fault follow the logarithmic earthquake frequency-magnitude of the G–R relationship up to a maximum magnitude (the truncation point) (Kramer, 1996), and (2) the characteristic earthquake model that implies that a typical maximum size of earthquake occurs repeatedly along a particular segment of the fault (Schwartz and Coppersmith, 1984). Derivation of Activity Rates from Geologic Data In the PSHA, the magnitude recurrence model is defined by minimum and maximum moment magnitudes, G–R b value, and the activity rate. The G–R relationship can be explained mathematically as: log N ðM Þ~a{b M
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 211–222
ð2Þ
Phoenix Seismic Source Characterization
of 0.906 6 0.017 from a catalog of 995 earthquakes (minimum Ml was 1.2) in Arizona during the TA deployment from April 2006 to March 2009. A b value of 1.0 was adopted by this study for the greater Phoenix region based on observations of global seismicity in areas of low to moderate seismicity. However, Ghanat (2008) investigated the sensitivity of the results to this value and showed that the results of the seismic hazard analysis were relatively insensitive to its value over the range of 0.9 to 1.0. Therefore, b value of 1.0 was used in this study. The activity rate can be determined from the magnitude range, fault slip rate, fault parameters, b value, and seismic moment (Molnar, 1979). The procedure outlined by Shedlock et al. (1980) was employed to determine the activity rate for each seismic source. It was assumed that the G–R distribution applies from the minimum moment magnitude up to the maximum moment magnitude assigned to the fault. Using the minimum and maximum moment magnitude values, the probability density function of G–R can be expressed as Equation 3: :
(ln 10:b)e{ln 10 bðM{Mmin Þ fM (M)~ : 1{e{ln 10 bðMmax {Mmin Þ
ð3Þ
The cumulative distribution function on M can be expressed as Equation 4: :
1{e{ln 10 bðm{Mmin Þ FM (m)~ : 1{e{ln 10 bðMmax {Mmin Þ
ð4Þ
The moment magnitude is related to seismic moment Mo through Equation 5 (Hanks and Kanamori 1979). ð5Þ
M~0:67 log10 Mo {10:73
The cumulative distribution function can be written as a function of Mo. : :
1{eð10:73b ln 10zln 10 b Mmin Þ Mo FM0 ðMo Þ~ : 1{e{ln 10 bðMmax {Mmin Þ
b {1:5
E ½Mo ~
Mo,max ð
Mo fMo (Mo )dMO
ð8Þ
Mo,min
Mo,max and Mo,min can be determined by substituting Mmax and Mmin into the Equation (9): Mo ~10(1:5 Mz16:1)
ð9Þ
Expected seismic moment can be expressed as Equation 10: 2 3 b ð10:73 ln 10:bzln 10:b:Mmin Þ e 6 1:5 7 E ðMo Þ~ 4 5 :b {ln 10 1{e ðMmax {Mmin Þ 1{ b 1:5 ð10Þ h b | 10ð1:5Mmax z16:1Þð1{1:5Þ i b {10ð1:5Mmin z16:1Þð1{1:5Þ The total rate of seismic moment release is the product of the rate of occurrence of earthquakes and expected seismic moment. Brune (1968) expressed the rate of seismic moment build up as a function of slip rate of fault as shown in Equation 11. _ o ~m:L:W :u_ M
ð11Þ
Where m is shear modulus, L is length of the fault, W is width of the fault, and u˙ is the fault slip rate. The shear modulus can be assumed to be 3.0 3 1011 dynes/cm2. Equating the rate of seismic moment release with the rate of seismic moment build up, and solving for the activity rate, gives Equation 12: N ðMmin Þ~
m:L:W :u_ E(Mo )
ð12Þ
The activity rate, N (Mmin), for each seismic source in the greater Phoenix area was determined using the method described and is presented in Table 2.
ð6Þ
The cumulative density function, Equation 6, can be differentiated, resulting in probability density function (Equation 7). b : : { b {1 eð10:73b ln 10zln 10 b Mmin Þ 1:5 Mo 1:5 ð7Þ fMo ðMo Þ~ : 1{e{ln 10 b ðMmax {Mmin Þ The expected seismic moment is E (Mo) and is presented as Equation 8:
Background Seismicity A random area source was also defined to model seismicity not associated with the six known local fault sources. Random earthquakes were assigned an equal likelihood of occurrence at any point within the random area source. The boundaries for this area source were latitude 32.5 to 35 degrees north and longitude 111 to 112.5 degrees west (Figure 2). A minimum magnitude of 4.6 and a maximum magnitude of 6.1 (average maximum magnitude of all seismic sources) were assigned to the random area
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 211–222
217
Ghanat, Kavazanjian, and Arrowsmith ˙ o, b value, and activity rate for the greater Phoenix area seismic sources. Table 2. Summary of slip rate, M
Fault Sources
Slip Rates (mm/yr)
˙ o 3 1021 M (dyne-cm/yr)
b Value
Activity Rate (event/yr)
Horseshoe Carefree Verde Sugarloaf Cottonwood Sand Tank
0.01–0.03 0.01 0.01–0.02 0.01–0.02 0.002–0.003 0.01–0.03
0.69–2.07 0.38 0.35–0.69 0.31–0.62 0.035–0.052 0.017–0.52
1.0 1.0 1.0 1.0 1.0 1.0
0.00034–0.0012 0.00032 0.00033–0.00066 0.00034–0.00069 0.000049–0.000073 0.00024–0.00073
source. A depth of seismicity of 10 km, a normal mechanism of faulting, a b value of 1.0 and an a value of 0.23 were assigned to the random area source. The a value was adopted from the recurrence model developed for this seismic zone in a statewide hazard analysis by the Arizona Department of Transportation in 1992 (Euge et al., 1992). Characteristic Earthquake Model The characteristic earthquake hypothesis states that seismic moment release along a fault segment is dominated by a characteristic earthquake rupturing the entire length of the segment and that energy release from smaller events is negligible (Schwartz and Coppersmith, 1984). From a seismic hazard assessment perspective, this offers a great simplification because only the characteristic earthquake is considered for each fault segment. The characteristic recurrence model allows more large magnitude earthquakes than the exponential recurrence distribution and generally assigns a small range of magnitudes (Figure 3) to the characteristic earthquake on a fault (Schwartz and Coppersmith, 1984).
Ground Motion Estimation One of the basic inputs to seismic hazard computations is the ground motion prediction equation (equations). The ground motion prediction equations are used to calculate the range of the expected ground motion at a given site based upon the distance from an earthquake of a given magnitude with magnitude usually expressed as moment magnitude (Kramer, 1996). The earthquake ground motion prediction equations used in this study are those developed for the shallow crustal earthquakes in Western North America for weak rock site conditions (average shear wave velocity in the upper 30 m equal to 760 m/s, the B-C boundary in NEHRP site classification). This study employed both the Next Generation Attenuation (NGA) and traditional ground motion prediction equations. The NGA project is five sets of groundmotion models developed for shallow crustal earthquakes in the western United States and similar active tectonic regions. The NGA models were developed for wider ranges of magnitudes, distances, and response spectral periods than had been used in traditional ground-motion relations (Power et al., 2008). The NGA ground motion prediction equations employed in this study are those developed by Campbell and Bozorgnia (2008) and Abrahamson and Silva (2008). The traditional ground motion prediction equations used in this study are those developed by Boore et al. (1997) and Abrahamson and Silva (1997). These four ground motion prediction equations were given equal weight in the PSHA. RESULTS OF PSHA
Figure 3. To capture both small magnitude events that conform to the G–R relationship and large characteristic events, the Characteristic and the Gutenberg Richter models may be combined, as shown.
218
The parameters for the active seismic sources presented in Tables 1 and 2 were employed by this study to conduct a comprehensive PSHA for the greater Phoenix area on grid points at spacing of about one kilometer, using the computer program EZ-FRISK (McGuire, 1976). The PSHA methods used here are fairly standard; however, analyses with the level of detail employed herein are rarely conducted for an area of low seismicity.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 211–222
Phoenix Seismic Source Characterization
Figure 4. Contour map of PGA and five percent damping for two percent probability of exceedance in 50 years (values in percent g).
Results of the probabilistic analysis are presented herein in the form of contour maps of the peak ground acceleration (PGA) and spectral acceleration for five percent damping at a spectral acceleration of one second for weak rock site conditions, with a two percent probability of being exceeded in 50 years (a return period of approximately 2,500 years) and uniform hazard spectra for several representative locations within the greater Phoenix area. Results are presented in Figures 4 and 5. The PGA and the spectral acceleration at one second in the greater Phoenix area range from about 0.07 g to 0.08 g and 0.02 g to 0.03 g, respectively. Figure 6A and B compare the uniform hazard spectra from this study to that of the USGS National
Seismic Hazard Maps (Petersen et al., 2014) for cities of Goodyear and Scottsdale for a two percent probability of exceedance in 50 years for weak rock site conditions. Note that these two cities (Goodyear and Scottsdale) are in the opposite quadrant of the greater Phoenix area. It is important to note that the USGS results are not used as a baseline, but rather as a check on the reasonableness and validity of the results. For the PGA, the ground motions are within approximately 10 percent of the USGS values. The difference between the results of this study and the USGS map values (Petersen et al., 2014) is more substantial at one second spectral period. For the spectral acceleration at one second, the ground motions in some parts of the greater Phoenix area
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 211–222
219
Ghanat, Kavazanjian, and Arrowsmith
Figure 5. Contour map of one-second spectral acceleration and five percent damping for two percent probability of exceedance in 50 years (values in percent g).
are approximately 30 percent lower than the USGS map values (Petersen et al., 2014). Despite these differences, no reduction in building code seismic design requirements can be justified based upon the results of our study. We attribute the differences between the results of this study and the USGS National Hazard maps (Petersen et al., 2014) primarily to differences in the seismic sources and the rate of occurrence of earthquakes. The seismic sources used in this study are not included in the USGS National Hazard maps (Petersen et al., 2014) due to insufficient documentation. To quantify the seismic hazard in the greater Phoenix area, USGS uses a smoothed historical
220
seismicity model and a large background source zone model. Both USGS models are characterized by using historical seismicity occurrence and magnitude data (Petersen et al,, 2008; Frankel et al., 1996). The smoothed historical seismicity model uses a 0.1u source grid to count number of events and the a value for each grid is calculated from the maximum likelihood method (Weichert, 1980), based on events with magnitudes of 4.0 and larger (Frankel et al., 1996). To address the possibility of having large earthquakes, USGS uses a large background zone based on broad geologic criteria. The earthquake rate for the background zone is determined by counting earthquakes with a magnitude greater than
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 211–222
Phoenix Seismic Source Characterization
design requirements for the greater Phoenix area can be justified based upon our study. The PSHA methods used here are fairly standard. However, analyses with the level of source characterization detail employed herein are rarely conducted for an area of low seismicity. The major contribution of this study is the use of the best available information on active faults to explicitly identify and delineate earthquake sources, assess earthquake size, estimate earthquake recurrence, and evaluate spatial distribution of the ground shaking hazard in the greater Phoenix area. Studies with this level of detail are valuable in areas like Phoenix which has a high population density and a large number of critical facilities and where slight reduction in design ground motions can significantly reduce building code seismic design requirement. ACKNOWLEDGMENT We acknowledge an anonymous reviewer for his valuable input to this manuscript. REFERENCES
Figure 6. (A and B) Comparison of the Uniform Hazard Spectra results from this study and those from USGS at cities of Goodyear and Scottsdale, respectively.
4, computing an annualized rate, and prorating this rate uniformly across the entire zone (Frankel et al., 1996, 2002). USGS uses a maximum moment magnitude of 7.0 and a b value of 0.8 in their background seismicity model versus a maximum moment magnitude of 6.1 and a b value of 1.0 used in this study. However, this study used the best available geologic data to characterize the local fault seismic sources. CONCLUSIONS Results of the site specific seismic hazard analysis for the greater Phoenix area presented herein are generally consistent with the USGS National Seismic Hazard Maps (Petersen et al., 2014). However, the spectral acceleration at one second period in some parts of the greater Phoenix area are approximately 30 percent lower than the USGS map values. Despite these differences, no reduction in IBC 2012 seismic
ABRAHAMSON, N. A. AND SILVA, W. J., 1997, Empirical response spectral attenuation relations for shallow crustal earthquakes: Seismological Research Letters, Vol. 68, pp. 94–127. ABRAHAMSON, N. A. AND SILVA, W. J., 2008, Summary of the Abrahamson & Silva NGA ground motion relations: Earthquake Spectra, Vol. 24, No. 1, pp. 67–97. ARIZONA EARTHQUAKE INFORMATION CENTER, 2011, Seismicity data: Electronic document, available at http://www.cefns. nau.edu/Orgs/aeic/eq_history.html BONILLA, M. G.; MARK, R. K.; AND LIENKAEMPER, J. J., 1984, Statistical relations among earthquake magnitude, surface rupture length, and surface rupture displacement: Bulletin Seismological Society America, Vol. 74, pp. 2379–2411. BOORE, D. M.; JOYNER, W. B.; AND FUMAL, T. E., 1997, Equations for estimating horizontal response spectra from western North American earthquakes: Seismological Research Letters, Vol. 68, pp. 128–153. BRESSLER, S. L. AND BUTLER, R. B., 1978, Magnetostratigraphy of the late tertiary Verde formation: Earth and Planetary Science Letters, Vol. 38, No. 2, pp. 319–330. BRUNE, J. N., 1968, Seismic moment, stress, and source dimensions for earthquakes in the California-Nevada region: Journal Geophysics, Res., Vol. 73, pp. 4681–4694. CAMPBELL, K. W. AND BOZORGNIA, Y., 2008, NGA ground motion model for the geometric mean horizontal component of PGA, PGV, PGD, and 5% damped linear elastic response spectra for periods ranging from 0.01 to 10 s: Earthquake Spectra, Vol. 24, No. 1, pp. 139–171. DEMSEY, K. A. AND PEARTHREE, P. A., 1990, Late Quaternary Surface Rupture of the Sand Tank Fault and Associated Seismic Hazard for the Proposed Super-conduction Super Collider Site Maricopa County, Arizona: Arizona Geological Survey Open-File Report 90-1, 43 p. DOSER, D. I. AND SMITH, R. B., 1989, An assessment of the source parameters of earthquakes in the Cordillera of western
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 211–222
221
Ghanat, Kavazanjian, and Arrowsmith United States: Bulletin Seismological Society America, Vol. 79, pp. 1383–1409. EBERLY, L. D. AND STANLEY, T. B., JR., 1978, Cenozoic stratigraphy and geologic history of southwestern Arizona: Geological Society America Bulletin, Vol. 89, No. 6, pp. 921–940. EUGE, K. M.; SCHELL, B. A.; AND LAM, I. P., 1992, Development of Seismic Acceleration Maps for Arizona: Arizona Department of Transportation Report AZ92-344, 327 p. FRANKEL, A.; MUELLER, C.; BARNHARD, T.; PERKINS, D.; LEYENDECKER, E.; DICKMAN, N.; HANSON, S.; AND HOPPER, M., 1996, National Seismic Hazard Maps: U.S. Geological Survey Open-File Report 96–532, 110 p. FRANKEL, A. D.; PETERSEN, M. D.; MUELLER, C. S.; HALLER, K. M.; WHEELER, R. L.; LEYENDECKER, E. V.; WESSON, R. L.; HARMSEN, S. C.; CRAMER, C. H.; PERKINS, D. M.; AND RUKSTALES, K. S., 2002, Documentation for the 2002 update of the National Seismic Hazard Maps: U.S. Geological Survey Open-File Report 2002–420, 33 p. GHANAT, S. T., 2008, Probabilistic Seismic Hazard Assessment of Greater Phoenix Area: Unpublished M.S. Thesis, Arizona State University, Tempe, AZ, 283 p. GUTENBERG, B. AND RICHTER, C. F., 1954, Seismicity of Earth and Associated Phenomena: Princeton University Press, Princeton, NJ, 310 p. HANKS, T. C. AND KANAMORI, H., 1979, A moment magnitude scale: Journal Geophysics, Vol. 84, B5, pp. 2348–2350. HOUSE, P. K., 1994, Surficial Geology of Southern Verde Valley Yavapai County, Arizona: U.S. Geological Survey Open-File Report 94-23, 20 p. INTERNATIONAL CODE COUNCIL (ICC), 2012, International Building Code: 689 p. KRAMER, S. L., 1996, Geotechnical Earthquake Engineering: Prentice-Hall, Inc., Upper Saddle River, NJ, 653 p. LOCKRIDGE, J.; FOUCH, M.; AND ARROWSMITH, J. R., 2012, Seismicity within Arizona during the deployment of EarthScope U.S. Array Transportable Array: Bulletin Seismological Society America, Vol. 102, No. 4, pp. 1850–1863. MCGUIRE, R. K., 1976, Evaluation of Sites for Earthquake Risk: U.S. Geological Survey Open-File Report 76-67, 68 p. MENGES, C. M. AND PEARTHREE, P. A., 1989, Late Cenozoic tectonism in Arizona and its impact on regional landscape evolution, In Jenney, J. P. and Reynolds, S. J. (Editors), Geologic Evolution of Arizona: Arizona Geological Society Digest 17, pp. 649–680. MOLNAR, P., 1979, Earthquake recurrence intervals and plate tectonics: Bulletin Seismological Society America, Vol. 69, pp. 115–133. PEARTHREE, P. A., 1998, Quaternary Fault Data and Maps for Arizona: Arizona Geological Survey Open-File Report 98-24, 122 p. PEARTHREE, P. A.; MENGES, C. M.; AND MAYER, L., 1983, Distribution, Recurrence, and Possible Tectonic Implications of Late Quaternary Faulting in Arizona: Arizona Bureau of Geologic Mineral Technical Open-File Report 83-20, 51 p. PEARTHREE, P. A. AND SCARBOROUGH, R. B., 1984, Reconnaissance Analysis of Possible Quaternary Faulting in Central Arizona:
222
Arizona Bureau of Geology and Mineral Technology OpenFile 85-4, 75 p. PETERSEN, M. D.; FRANKEL, A. D.; HARMSEN, S. C.; MUELLER, C. S.; HALLER, K. M.; WHEELER, R. L.; WESSON, R. L.; ZENG, Y.; BOYD, O. S.; PERKINS, D. M.; LUCO, N.; FIELD, E. H.; WILLS, C. J.; AND RUKSTALES, K. S., 2008, Documentation for the 2008 Update of the United States National Seismic Hazard Maps: U.S. Geological Survey Open-File Report 2008-1128, 61 p. PETERSEN, M. D.; MOSCHETTI, M. P.; POWERS, P. M.; MUELLER, C. S.; HALLER, K. M.; FRANKEL, A. D.; ZENG, Y.; REZAEIAN, S.; HARMSEN, S. C.; BOYD, O. S.; FIELD, N.; CHEN, R.; RUKSTALES, K. S.; LUCO, N.; WHEELER, R. L.; WILLIAM, R. A.; AND OLSEN, A. H., 2014, Documentation for the 2014 Update of the United States National Seismic Hazard Maps: U.S. Geological Survey Open-File Report 2014-1091, 243 p. PIETY, L. A. AND ANDERSON, L. W., 1991, The Horseshoe Fault, evidence for prehistoric surface rupture earthquakes in central Arizona: Arizona Geology, Vol. 21, No. 3, pp. 1–8. POWER, M.; CHIOU, B.; ABRAHAMSON, N.; BOZORGNIA, Y.; SHANTZ, T.; AND ROBLEE, C., 2008, An overview of the NGA project: Earthquake Spectra, Vol. 24, No. 1, pp. 3–21. REITER, L., 1990, Earthquake Hazard Analysis Issues and Insights: Columbia University Press, New York, NY, 254 p. REYNOLDS, S. J., 1985, Geology of the South Mountains, Central Arizona: Arizona Bureau of Geology and Mineral Technology Bulletin 195, 61 p. SCHELL, B. A. AND WILSON, K. L., 1982, Regional Neotectonic Analysis of Sonoran Desert of Arizona and California: U. S. Geological Survey Open-File Report 82-57, 61 p. SCHWARTZ, D. P. AND COPPERSMITH, K. J., 1984, Fault behavior and characteristic earthquakes examples from Wasatch and San Andrea Fault: Journal Geophysics, Vol. 89, No. B7, pp. 5681–5698. SHEDLOCK, K. M.; MCGUIRE, R. K.; AND HERD, D. G., 1980, Earthquake Recurrence in the San Francisco Bay Region, California, From Fault Slip and Seismic Moment: U. S. Geological Survey Open-File Report 80-999, 21 p. SKOTNICKI, S. J.; LEIGHTY, R. S.; AND PEARTHREE, P. A., 1997, Geologic Map of the Wildcat Hill Quadrangle, Maricopa County, AZ: Arizona Geological Survey Open-File Report 97-2, 17 p. TOCHER, D., 1958, Earthquake energy and ground breakage: Bulletin Seismological Society America, Vol. 48, pp. 147–153. UNITED STATES GEOLOGICAL SURVEY AND ARIZONA GEOLOGICAL SURVEY, 2010, Quaternary Fault and Fold Database for the United States: Electronic document, available at http:// earthquake.usgs.gov/regional/qfaults/ WEICHERT, D. H., 1980, Estimation of earthquake recurrence parameters for unequal observation periods for different magnitudes: Bulletin Seismological Society America, Vol. 70, pp. 1337–1356. WELLS, D. L. AND COPPERSMITH, K. J., 1994, New Empirical relationships among magnitude, rupture length, rupture width, rupture area, and surface displacement: Bulletin Seismological Society America, Vol. 84, No. 4, pp. 974–1002.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 211–222
Monitoring Spatial-Temporal Change of Land Desertification in a Fragile Sub-Alpine Rangeland Eco-Environment: A Case Study from China WEI XIAN1 College of Resources and Environment, Chengdu University of Information Technology, No. 24, Section 1, Xue Fu Road, the Southeast Airport Economic Development Zone, Chengdu, Sichuan Province 610225, China
ZHIYING XIANG LIYANG LIU HUAIYONG SHAO Key Laboratory of Geoscience Spatial Information Technology, Ministry of Land and Resources of the P. R. China, Chengdu 610059, China
Key Terms: Land Desertification, Zoige County, Spectral Mixture Analysis, Change Vector Analysis, Ecological Suggestions
ABSTRACT Zoige County, China, represents a fragile sub-alpine rangeland eco-environment with a severe land desertification problem. This paper aims at detecting land desertification change in Zoige County over 15 years with quantitative remote-sensing techniques using multi-spectral imagery. Landsat images acquired in 1994 and 2009 were analyzed using the following methodology: (1) image pre-processing; (2) spectral mixture analysis (SMA) to obtain precise sub-pixel classification results of land cover; and (3) change vector analysis (CVA) to conduct a multi-temporal comparison process. Change detection results depict the land desertification conditions and vegetation re-growth conditions. In this way, we characterized the spatialtemporal change pattern of land desertification in Zoige County between 1994 and 2009. After categorizing ecological regions based on change detection results, we analyzed the driving factors of both land desertification conditions and vegetation re-growth conditions, finding out that grasslands under intense grazing pressure tend to suffer severe desertification, while topographic relief has an obvious influence on vegetation re-growth. Specific suggestions for each ecological region are proposed, which can assist the development of environmental restoration measures and 1
Corresponding author email: xianwei@cuit.edu.cn.
environmental protection measures in Zoige County in an effective way. Furthermore, this methodology for monitoring land desertification could be carried out across neighboring counties or in other regions with similar sub-alpine rangeland and land desertification problems. INTRODUCTION China is one of the many countries around the globe facing a serious problem of land desertification (Heshmati and Squires, 2013). Land desertification areas in China cover 2.632 million km2, accounting for 27.33 percent of its total national territory (SFA, 2011). Due to its wide distribution, land desertification threatens the living conditions of nearly 200 million people in China (DPSSTS, 2002). Land desertification affects a wide range of benefits provided by the environment to humans: products such as food and water, natural processes such as climate regulation, and also non-material services such as recreation, and supporting services such as soil conservation (Millennium Environment Assessment, 2005). Moreover, land desertification has environmental impacts that go beyond the areas directly affected. For instance, wind erosion of the desertified areas can increase the formation of large dust clouds that can cause health problems in more densely populated areas thousands of kilometers away (Wang, 2004). To combat desertification, large-scale and longterm monitoring is needed to understand desertification processes and determine the extent of land desertification. Remote-sensing techniques stand out as a time- and cost-efficient method for monitoring
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 223–234
223
Xian, Xiang, Liu, and Shao
early 1990s, threatening its fragile sub-alpine rangeland eco-environment and its water conservation functions (Dong et al., 2010). In the remote-sensing images of Zoige County where grassland, wetland, and sandy soil are distributed closely, one pixel usually contains mixed spectral information due to the high variability in the distribution of land-cover components. According to a previous case study in Zoige (Qiu et al., 2009), supervised classification, which has been commonly used for monitoring land desertification, is sensitive to pixel-level changes of land-cover components. Therefore, this paper applies a sub-pixel classification technique, spectral mixture analysis (SMA), to improve accuracy of land desertification monitoring in Zoige. SMA is designed to derive the proportions of vegetation, sandy soil, and water that compose a mixed pixel to monitor land desertification in a precise way. SMA has been proven effective in a variety of quantitative applications with multi-spectral imagery (Peddle et al., 1999; Small, 2001; Okin et al., 2004; Powell et al., 2007). Therefore, applying SMA to monitor land desertification in a fragile sub-alpine rangeland eco-environment with humid plateau climate has much potential. MATERIALS AND METHODS Study Area
Figure 1. Location of Zoige County and field photos (A, B, C, D indicate the locations of field photos).
land desertification (Hellde´n, 1984; Tripathy et al., 1996; and Collado et al., 2002). Using remote-sensing techniques, previous research has mostly focused on desertification in arid and semi-arid areas. Commonly applied approaches to detecting land desertification are vegetation indices such as the NDVI (Normalized Difference Vegetation Index) (Piao et al., 2005; Huang and Siegert, 2006) which is calculated from visible and near-infrared bands and image classification (Wu and Ci, 2002; Qi et al., 2012). However, land desertification of subhumid and humid areas in China is still lacking research (Wang, 2004). Zoige County, through which the upper Yellow River flows, is located on the northeastern part of Qinghai-Tibet Plateau, with a typical humid plateau climate in a frigid temperate zone. Under the influence of climate change and stock over-grazing, land desertification has been a severe environmental problem in Zoige County since the
224
Zoige County (seen in Figure 1), which is located on the northeastern part of the Qinghai-Tibet Plateau, covers 10,436.58 km2 (latitude 32u569– 34u199N, longitude 102u089E–103u399E). Plateau hills and alpine valleys form its landscape. Containing the headwaters of the Yellow River, Zoige County has a crucial water conservation function for southwest China. It enjoys a humid plateau climate in a frigid temperate zone with annual rainfall ranging from 600 to 750 mm. Average annual temperature is around 1uC, with the lowest 210.3uC in January and the highest 10.9uC in July. The major vegetation types in Zoige County are sub-alpine meadow and marshmeadow, dominated by Festuca nivina, Kobresia setchuanensis, Elymus nutans, Carex muliensis, and Kobresia tibetica; soil types in Zoige County include peat moor soil, alpine meadow soil, sub-alpine meadow soil, swamp soil, and swampy meadow soil (Yong et al., 2003). Grassland accounts for more than 60 percent of the total land cover area in Zoige County; therefore, pastoral farming has been a major part of local husbandry for centuries. Wetland is distributed widely in Zoige County: 178 plant and 218 animal species have been identified in the Zoige National Wetland Nature Reserve. Among them,
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 223–234
Monitoring Land Desertification in Fragile Sub-Alpine Rangeland
about one third of birds (46 species) and 40 percent of mammals (10 species) belong to international or national protected birds and animals (McNamee, 2003). With the impacts of global climate change and stock over-grazing, land desertification has been a severe environmental problem in this area for decades. Data Acquisition and Pre-Processing Landsat-5 Thematic Mapper (TM) images of the studied area on the dates of August 4, 1994, and July 28, 2009, were acquired. For each date, acquired images were on the tracks of path130/row37, path131/row36, and path131/row37, respectively. Multi-temporal images were analyzed to monitor spatial-temporal change of land desertification in Zoige County in the 15 year period. Landsat-5 TM has a spatial resolution of 30 m with six visible/near infrared bands and one thermal band. Bands 1 (0.45–0.52 mm), 2 (0.52–0.60 mm), 3 (0.63–0.69 mm), 4 (0.76–0.90 mm), 5 (1.55–1.75 mm), and 7 (2.08–2.35 mm) from the Landsat-5 TM sensor for both dates were used for the analysis. Band 6 (10.4– 12.5 mm) was not included in the analysis because the thermal infrared wavelengths are not required for performing atmospheric correction. For the application of SMA, conversion from digital values to reflectance was carried out by atmospheric correction for all TM images. The atmospheric correction is based on a revision of the dark-object method, which estimates atmospheric transmissivity as a function of the cosine of the zenith angle (Chavez, 1996). Reflectance for the six nonthermal channels is then computed as follows: rk ~
KpðLsen,k {La,k Þ E0,k ðcos hi Þ2
ð1Þ
where rk is the reflectance for band k, K is a factor that takes into account the variation of the Sun-Earth distance; Lsen,k is the radiance detected by the sensor (computed from the digital values using the calibration coefficients included in the image); La,k is the atmospheric radiance, computed from the minimum (dark-object) value of that band; and E0,k is the solar irradiance at the top of the atmosphere and the solar zenith angle. K is computed as a function of the Julian day (D): K~1z0:0167ðsenð2pðD{93:5Þ=365ÞÞ
ð2Þ
It can be seen from the formula above that the darkobject method does not consider the situations of atmospheric multiple scattering or multiple scattering
between objects. In addition, the dark-object method does not consider topographic influences. Therefore, the results of dark-object atmospheric correction can be affected by these facts. Geometric precision correction was applied to all TM images. Each TM image was geo-registered to an existing geo-referenced image using nearest-neighbor re-sampling with 20 control points, and an average root mean square (RMS) error of 0.5 was calculated, which was appropriate for multi-temporal comparisons. In addition, the Shuttle Radar Topography Mission (SRTM) 90 m digital elevation model (DEM) on the track of column 56/row 6 was also acquired for the purpose of topographic analysis in the study area. Spectral Mixture Analysis SMA is a technique to derive sub-pixel cover fractions of surface materials using high-spectralresolution reflectance measurements collected from airborne or space-borne spectrometers (Asner et al., 2003). This method is ideal for use in a sub-alpine, sub-humid rangeland eco-environments where subpixel cover variation is high. The goal of SMA is to identify primary spectral contributions within each pixel (Adams et al., 1993). It provides a means to determine the relative abundance of land-cover materials present in any pixel based on the spectral characteristics of the materials. SMA transforms radiation or reflectance data into fractions of a few dominant end members, which are fundamental physical components of the scene and not themselves a mixture of other components (Elmore et al., 2000). Each end-member component contributes to the pixel-level spectral reflectance, expressed as: Ri ~
n X
Fj :REij zei and
j~1
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u B uX ðei Þ2 RMS~t B i~1
n X
Fj ~1
ð3Þ
j~1
ð4Þ
where Ri is the reflectance of the mixed spectrum in band i, Fj is the fractional abundance of end-member j, REij is the reflectance of the end-member spectrum j in band i, n is the number of spectral end members, and ei is the error of the fit for band i. Thus, for this analysis with TM data, there will be six equations, one for each spectral band (B 5 6). Equation 4 is the total root-mean square error (RMSE), where B is the total number of spectral bands.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 223–234
225
Xian, Xiang, Liu, and Shao
End-Member Selection The crucial step to a successful SMA is the selection of appropriate end members. End-member selection techniques, which directly impact modeling performance, vary depending on the tradeoffs among classification accuracy, library size, and computation time (Roth et al., 2012). End members must define a coherent set of spectra that are representative of physical components on the surface, but they must also model the spectral variability inherent to the scene (Elmore et al., 2000). End members can be identified using (1) libraries of known spectra collected with a spectrometer in the field or in a laboratory, (2) libraries of known spectra from previous SMA studies, or (3) spectrally pure or ‘‘extreme’’ pixels identified within the images being analyzed (Schweik and Green, 1999). Although image end members cannot be entirely pure, their degree of pureness is more accurate because they represent the dimensionality of the corresponding data set. Thus, they are more suitable for multi-temporal change detection. Aiming at monitoring land desertification, this paper applies image endmembers that were derived with three steps: (1) spectral reduction by the minimum noise fraction (MNF) transform, (2) spatial reduction with the pixel purity index (PPI) method, and (3) manual identification of the end members using the N-dimensional visualizer. The MNF transform, which consists of two consecutive data reduction operations, aims to ensure valid dimensions of imagery data by separating noise from it, thereby reducing the calculation amount of later procedures (Green et al., 1988). The PPI, which has been widely used in multi-spectral and hyper-spectral images analysis for end-member extraction, aims to search for a set of vertices of a convex geometry in a given data set that are supposed to represent pure signatures present in the data (Chaudhry et al., 2006). The N-dimensional visualizer is an interactive tool; by adding in PPI results (relatively pure pixels), it can interactively assist researchers to select image end members in N-dimensional space. In this case study, five image end members were manually selected: bright vegetation (BV), bright soil (BS), dark vegetation (DV), dark soil (DS), and water. The fractions of soil and vegetation can facilitate the analysis of land desertification and vegetation re-growth in the studied area. Change Vector Analysis CVA is a radiometric technique that examines the corresponding pixels of two satellite images by comparing two bands of each image to produce
226
images of change magnitude and change direction (Kuzera et al., 2005). In this study, bright vegetation (BV) and bright soil (BS) fraction images were used to monitor the land desertification and vegetation regrowth between 1994 and 2009. The change magnitude of the vector is calculated from the Euclidean distance. The results show the difference between the pixel values of the fraction images for bright vegetation (BV) and bright soil (BS) cover, respectively, between 1994 and 2009. It is shown as follows: qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi R~ (BS1 {BS2 )2 z(BV1 {BV2 )2
ð5Þ
where R is the magnitude of vector change, and subscripts 1 and 2 indicate the fraction covers in 1994 and 2009. Change direction is measured as the angle (a) of the change vector from a pixel measurement in 1994 to the corresponding pixel in 2009 according to: tan a~
(BS1 {BS2 ) (BV1 {BV2 )
ð6Þ
Angles measured between 90 and 180 degrees indicated an increase in sandy soil and decrease in vegetation cover and therefore represent land desertification conditions. Meanwhile, angles measured between 270 and 360 degrees indicate a decrease in sandy soil and an increase in vegetation cover and therefore represent vegetation re-growth conditions (Lorena et al., 2002). Angles measured between 0 and 90 degrees and between 180 and 270 degrees indicate either an increase or decrease in both sandy soil and vegetation cover, and consequently persistent conditions. Field Survey The field survey was conducted in August 2011 in order to evaluate the accuracy of SMA using ground vegetation data as a reference. In total, 30 sampling sites (size 60 3 60 m for each site, corresponding to four pixels of the Landsat image) were located and established in the study area. At each site, trees and bushes were geo-referenced with a global positioning system (GPS), and then the percentage of ground vegetation cover including sub-alpine meadow, marshmeadow, and shrubs was estimated using the line-point intercept sampling method. Measurements were taken along 30 60-m-long transects oriented in a N-S direction. Pin flags were lowered at 60 cm intervals along the entire length of the transect. At each point, the types of cover were recorded, and the percentage of vegetation cover was calculated. The accuracy of SMA
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 223–234
Monitoring Land Desertification in Fragile Sub-Alpine Rangeland
Figure 2. End-member spectra, where BV represents bright vegetation component, DV represents dark vegetation component, BS represents bright soil component, and DS represents dark soil component.
was estimated by scatter plotting correlations of the total percentage of vegetation cover in each plot and the vegetation fraction image. RESULTS AND DISCUSSIONS End-Member Spectra The MNF transform was applied on each TM image to complete spectra reduction. It is found that the first four bands that are generated after transform contain 92 percent of the variance. In the meanwhile, spatial reduction for the MNF transform result was performed with the PPI method. Finally, using the Ndimensional visualizer, five end members were manually selected: bright vegetation (BV), dark vegetation (DV), bright soil (BS), dark soil (DS), and water. BV consists of vegetation with high water content, such as meadow, marsh-meadow, and shrubs. DV consists of vegetation with low water content, such as senescing meadow and senescing shrubs. BS consists of sandy soil with high reflectance and low water content. DS consists of bare soils with low reflectance and high water content. The water end member consists of rivers, lakes, and water areas in wetlands. A set of end-member spectra extracted from one of the August 4, 1994, images is shown in Figure 2. SMA Process To achieve the best quality of fraction images, three sets of end members were tested in the SMA process for each TM image. The sets were: (1) all five end members; (2) BV, BS, DS, and water; and (3) BV, BS, and water. Fraction images derived from the different sets of end members were evaluated using visual interpretation, and error extent and distribution in
the error fraction image. The set of four end members (BV, BS, DV, and DS) was chosen, since it provided the best distinction of land-cover types and relatively low errors. Proportions of BV, BS, DS, and water for each TM image, presented as fraction images, were determined after the SMA process. The vegetation cover information carried by BV fraction images and the sandy soil cover information carried by BS fraction images separately indicate land desertification (as shown in Figure 3) and vegetation re-growth (as shown in Figure 4). Also, in combination, they contribute to the analysis of wetland degradation in wetland areas (as shown in Figure 5), because detection of an increase of both vegetation cover and soil cover near water implies the degradation of wetland. The scatter plot correlation between the percentage of vegetation derived from SMA on the TM image (July 2009) and field data (August 2011) is shown in Figure 6. As shown, the R2 of 0.8942 represents an appropriate correlation between them. In addition, there are possible sources of error that may have affected the correlation result. First of all, the imprecise registration of multi-date images is potentially the largest source of error (Elmore et al., 2000), especially in our case, as the geometric rectification was done with 20 ground control points for each image. In addition, the application of the line-point intercept sampling method in the field survey contains error. Additionally, although the average cloud coverage percentage of images is quite low, cloud shadow in images can possibly affect SMA results lightly. In spite of the existing slight error, the correlation between SMA data and field data in our case shows an acceptable capacity to conduct the multi-date images comparison. Change Detection CVA applies BV and BS fraction images to monitor land desertification and vegetation re-growth between 1994 and 2009. The change directions that were derived from CVA indicate both land desertification and vegetation re-growth. The magnitude ranges from low level to high level for both land desertification condition and vegetation re-growth condition (shown in Figure 7). Overall, land desertification prevailed over vegetation re-growth. The land desertification area covers 2,609.66 km2 in total in Zoige County, including 466.57 km2 high-level area, 949.36 km2 medium-level area, and 1,193.73 km2 low-level area. The high-level areas of land desertification are concentrated in the northwest part of Zoige County. This is caused by long-term integrated driving factors such as climate change and stock over-grazing (Zhang et al., 2007; Shi and Tu, 2009),
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 223–234
227
Xian, Xiang, Liu, and Shao
Figure 3. BV and BS fraction images and differences between 15 years: (a) BV in 1994, (b) BV in 2009, (c) difference in BV, (d) BS in 1994, (e) BS in 2009, (f) difference in BS.
threatening the fragile sub-alpine environment in Zoige County. The medium-level areas of land desertification are distributed near high-level areas as transition zones to low-level areas of land de-
sertification. Meanwhile, the vegetation re-growth areas cover 2,383.90 km2 in total, which consists of 299.33 km2 high-level areas, 917.09 km2 medium-level areas, and 1,167.48 km2 low-level areas. Medium-level
Figure 4. BV fraction images and differences between 15 years: (a) BV in 1994, (b) BV in 2009, (c) difference in BV.
228
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 223–234
Monitoring Land Desertification in Fragile Sub-Alpine Rangeland
Figure 5. BV and BS fraction images and differences between 15 years: (a) BS in 1994, (b) BS in 2009, (c) difference in BS, (d) BV in 1994, (e) BV in 2009, (f) difference in BV.
and low-level vegetation re-growth conditions were detected mainly in the southeast and northeast of study area where alpine valleys are distributed. A high level of vegetation re-growth conditions was detected in national ranches located in the western part of
Zoige County, where land desertification combating measures have been carried out for years. ECOLOGICAL SUGGESTIONS Categorizing Ecological Regions Based on CVA Result
Figure 6. Scatter plot correlation between measured and SMA estimated vegetation fraction in 2009.
For a further understanding of the spatial-temporal change of land desertification in Zoige County between 1994 and 2009, a more specific and visualized classification for land desertification conditions and vegetation re-growth conditions is needed. Considering the concentrated distribution of similar levels for both land desertification conditions and vegetation re-growth conditions, we categorized Zoige County into three regions (shown in the left part of Figure 8). Respectively, region I represents concentrated areas of land desertification conditions, region II represents concentrated areas of vegetation re-growth conditions, and region III represents concentrated areas of persistent conditions.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 223–234
229
Xian, Xiang, Liu, and Shao
Figure 7. Distribution map and statistical table of land desertification and vegetation re-growth areas by applying change vector analysis.
230
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 223–234
Monitoring Land Desertification in Fragile Sub-Alpine Rangeland
Figure 8. Concentration areas of land desertification, vegetation re-growth, and persistence, categorized according to the change results between 1994 and 2009 in Zoige County (on the left). Slope distribution map of Zoige County (on the right).
Analysis of Driving Factors Driving Factors of Land Desertification Climate Change— Under the influence of global climate change, the average annual temperature in Zoige County has been rising significantly for decades at an average rate of 0.28uC/10 yr (Zhang et al., 2007), which is faster than the average rate of global temperature rising 0.03–0.06uC/10 yr (Houghton et al., 2001). Meanwhile, the average annual precipitation in Zoige County has been declining with an average rate of 211.559 mm/10 yr; the average annual evaporation in Zoige County has been increasing with an average rate of 7.621 mm/10 yr (Guo and Li, 2007). As a result of the long-term effects of temperature rises, precipitation declines, and evaporation increases, Zoige County has been suffering from strong winds and droughts in winter and spring (Shi and Tu, 2009). Therefore, there has been a decrease of both the ground vegetation cover (Wang and Zhao, 2005) and the organic matter content of surface soil (Zhao and He, 2000). Stock Over-Grazing— Pastoral farming is the pillar industry of Zoige County, contributing more than 90 percent of local husbandry income (Jiang and Li, 2012). The ideal maximum of stock capacity in Zoige County is 1.865 3 106 sheep (Shen and Wang,
2003); however, according to the provincial survey of rangeland resources, the actual stock capacity in Zoige County had already reached up to 3.412 3 106 sheep by the end of 2006, overloading the capacity by about 80.1 percent. The overloaded stock capacity in Zoige County leads to stock overgrazing. Long-term stock over-grazing in Zoige County badly affects the growth of vegetation, causing a decrease of grassland productivity and ecological resilience; over-trampling upon grassland by large amounts of stock causes soil hardening and vegetation decrease in Zoige County (Wang and Bao, 2002). Driving Factors of Vegetation Re-Growth Protection Measures of National Wetland Nature Reserve— Approved by the State Council, Zoige National Wetland Nature Reserve was established in 1998 (Wang, 2012). Since then, protection measures towards land desertification, such as vegetation restoration and fencing protection, have been effectively carried out within and beyond the boundaries of National Wetland Nature Reserve in Zoige County. In this paper, CVA results shows a fine vegetation re-growth condition around the National Wetland Nature Reserve, which confirms that the protective measures towards land desertification have improved the ground vegetation cover percentage a lot.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 223–234
231
Xian, Xiang, Liu, and Shao
Topographic Influences— After analyzing SRTM 90 m DEM data, it is clear that grassland and wetland dominate the land-cover types in Zoige County, while southeastern and northeastern Zoige County contains large areas of alpine valley with an altitude ranging from 2,400 m to 4,200 m. In addition, a slope distribution map of Zoige County (shown in the right part of Figure 8) was generated from DEM data, classified into four types: flat area (0 to 7 degrees), gentle slope (7 to 15 degrees), medium slope (15 to 25 degrees), and steep slope (.25 degrees). Due to the precipitous topography, the alpine valley areas in Zoige County, which are dominated by medium and steep slopes, are spared from heavy human use. Therefore, high forest cover remains there, and fine water conservation function is preserved.
Specific Suggestions for Ecological Regions Suggestions for Region I (Area of Concentrated Land Desertification Conditions) Region I, where grassland is widely distributed, has experienced long-term stock over-grazing, which has caused the serious problem of land desertification. Stock capacity control measures such as rotating grazing could alleviate rangeland pressure from stock overloading. At the same time, sparing zones where serious land desertification is caused by livestock grazing and also planting wind-resistant and droughttolerant species of grasses or shrubs could improve the vegetation cover percentage and rangeland ecoenvironmental restoration for land desertification areas. Suggestions for Region II (Area of Concentrated Vegetation Re-Growth Conditions) Region II, including alpine valley areas and wellrestored grassland or wetland areas, presented fine vegetation re-growth conditions. Natural resources protection is necessary for areas with high vegetation cover, especially for alpine forests. Vegetation restoration measures that have proven effective in these areas, such as establishing barriers to protect vegetation during growth period and planting wind-resistant and drought-tolerant vegetation, should be consistently enforced. In this way, it is viable to accomplish the sustainable development of natural resources in region II. Suggestions for Region III (Area of Concentrated Persistent Conditions) Region III covers areas where slight land desertification conditions and slight vegetation re-growth
232
conditions have simultaneously been detected, resulting in a relatively persistent condition. For the slight land desertification conditions that are detected in region II, timely and efficient desertificationcombating measures should be carried out. For instance, it is recommended to spare zones where land desertification have shown signs because of livestock grazing and to plant appropriate species of grasses until vegetation restoration is completed. CONCLUSIONS This paper points out the spatial-temporal change pattern of land desertification in Zoige County by applying SMA and CVA. Between 1994 and 2009, land desertification conditions in Zoige County were concentrated in grassland areas where pastoral farming dominates. Vegetation re-growth conditions were detected in protected areas (National Wetland Reserve and National Ranches) and alpine valley areas, which enjoy fine forest cover. In this study, SMA provided us with precise land-cover information in Zoige County in both 1994 and 2009, which ensures the accuracy of the multi-temporal comparison performance along with the quality of our monitoring results. After analyzing the driving forces of land desertification in Zoige County, specific suggestions for each ecological region are proposed to help the development of local environment management and efficient measures such as stock capacity control and vegetation restoration to combat land desertification. Combined with long-term remote-sensing monitoring and specific desertification-combating measures, we believe it is possible to restore vegetation cover in land desertification areas in a fragile sub-alpine ecoenvironment, which will benefit regional eco-environment stability as well as protect land for future generations. ACKNOWLEDGMENTS This study is supported and funded by the Key Laboratory of Geoscience Spatial Information Technology, Ministry of Land and Resources of China (Grant No. KLGSIT2013-07), the Research Start-up Funds for Brain Gain of Chengdu University of Information Technology (Grant No. KYTZ201304), the Science & Technology Department of Sichuan Province (Grant No. 2014ZR0145), the National Natural Sciences Foundation of China (Grant No. 41302282), the Research Fund for the Doctoral Program of Higher Education of China (Grant No. 20115122120007), and the National Undergraduate
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 223–234
Monitoring Land Desertification in Fragile Sub-Alpine Rangeland
Training Programs for Innovation (Grant No. 201310616010). REFERENCES ADAMS, J. B.; SMITH, M. O.; AND GILLESPIE, A. R., 1993, Imaging spectroscopy: Interpretation based on spectral mixture analysis: Remote Geochemical Analysis: Elemental and Mineralogical Composition, Vol. 7, pp. 145–166. ASNER, G. P.; BORGHI, C. E.; AND OJEDA, R. A., 2003, Desertification in central Argentina: Changes in ecosystem carbon and nitrogen from imaging spectroscopy: Ecological Applications, Vol. 13, No. 2, pp. 629–648. CHAUDHRY, F.; WU, C. C.; AND LIU, W., 2006, Pixel purity index– based algorithms for endmember extraction from hyperspectral imagery: Recent Advances in Hyperspectral Signal and Image Processing, Vol. 37, No. 2, pp. 29–62. CHAVEZ, P. S., 1996, Image-based atmospheric corrections— Revisited and improved: Photogrammetric Engineering and Remote Sensing, Vol. 62, No. 9, pp. 1025–1035. COLLADO, A. D.; CHUVIECO, E.; AND CAMARASA, A., 2002, Satellite remote sensing analysis to monitor desertification processes in the crop-rangeland boundary of Argentina: Journal of Arid Environments, Vol. 52, No. 1, pp. 121–133. DEPARTMENT OF POPULATION, SOCIAL, SCIENCE AND TECHNOLOGY STATISTICS, P.R. CHINA (DPSSTS), 2002, China Population Statistics Yearbook: China Statistics Press, Beijing (in Chinese). DONG, Z.; HU, G.; AND YAN, C., 2010, Aeolian desertification and its causes in the Zoige Plateau of China’s Qinghai–Tibetan Plateau: Environmental Earth Sciences, Vol. 59, No. 8, pp. 1731–1740. ELMORE, A. J.; MUSTARD, J. F.; AND MANNING, S. J., 2000, Quantifying vegetation change in semiarid environments: Precision and accuracy of spectral mixture analysis and the normalized difference vegetation index: Remote Sensing of Environment, Vol. 73, No. 1, pp. 87–102. GREEN, A. A.; BERMAN, M.; AND SWITZER, P., 1988, A transformation for ordering multispectral data in terms of image quality with implications for noise removal: Geoscience and Remote Sensing, Vol. 26, No. 1, pp. 65–74. GUO, J. AND LI, G., 2007, Climate change in Zoige plateau marsh wetland and its impact on wetland degradation: Plateau Meteorology, Vol. 26, No. 2, pp. 422–428 (in Chinese). HELLDE´N, U., 1984, Drought impact monitoring: A remote sensing study of desertification in Kordofan, Sudan: Rapporter och Notiser-Lunds Universitets Naturgeografiska Institution, Vol. 61, pp. 61. HESHMATI, G. A. AND SQUIRES, V., 2013, Combating Desertification in Asia, Africa and the Middle East: Springer Press, Netherlands, Berlin, Germany. HOUGHTON, J. T.; DING, Y.; GRIGGS, D. J.; AND NOGUER, M., 2001, Climate Change 2001: The Scientific Basis: Cambridge University Press, Cambridge, U.K. HUANG, S. AND SIEGERT, F., 2006, Land cover classification optimized to detect areas at risk of desertification in North China based on SPOT VEGETATION imagery: Journal of Arid Environments, Vol. 67, No. 2, pp. 308–327. JIANG, J. AND LI, A., 2012, Change of wetlands in Zoige Country from 1974 to 2007: Wetland Science, Vol. 10, No. 3, pp. 318–326 (in Chinese). KUZERA, K.; ROGAN, J.; AND EASTMAN, J. R., 2005, Monitoring vegetation re-growing and deforestation using change vector analysis: Mt. St. Helens study area. Proceedings, The ASPRS Annual Conference: Baltimore, MD, pp. 8.
LORENA, R. B.; SANTOS, J. R.; AND SHIMABUKURO, Y. E., 2002, A change vector analysis technique to monitor land use/land cover in SW Brazilian amazon: Acre state. Proceedings, PECORA 15-Integrating Remote Sensing at the Global, Regional and Local Scale, Denver, CO, USA, pp. 8–15. MCNAMEE, P., 2003, Management plan for Ruoergai National Nature Reserve, report to GEF. In UNDP-PR/98G32 Wetland Biodiversity Conservation and Sustainable Use in China. MILLENNIUM ENVIRONMENT ASSESSMENT, 2005, Environments and Human Well-Being: Island Press, Washington, DC. OKIN, G. S. AND ROBERTS, D. A., 2004, Remote sensing in arid regions: Challenges and opportunities: Manual of Remote Sensing, Vol. 4, pp. 111–146. PEDDLE, D. R.; HALL, F. G.; AND LEDREW, E. F., 1999, Spectral mixture analysis and geometric-optical reflectance modeling of boreal forest biophysical structure: Remote Sensing of Environment, Vol. 67, No. 3, pp. 288–297. PIAO, S.; FANG, J.; AND LIU, H., 2005, NDVI-indicated decline in desertification in China in the past two decades: Geophysical Research Letters, Vol. 32, No. 6, pp. L06402. POWELL, R. L.; ROBERTS, D. A.; AND DENNISON, P. E., 2007, Subpixel mapping of urban land cover using multiple endmember spectral mixture analysis: Remote Sensing of Environment, Vol. 106, No. 2, pp. 253–267. QI, Y.; CHANG, Q.; AND JIA, K., 2012, Temporal-spatial variability of desertification in an agro-pastoral transitional zone of northern Shanxi Province, China: Catena, Vol. 88, No. 1, pp. 37–45. QIU, P.; WU, N.; AND LUO, P., 2009, Analysis of dynamics and driving factors of wetland landscape in Zoige, eastern Qinghai-Tibetan Plateau: Journal of Mountain Science, Vol. 6, No. 1, pp. 42–55 (in Chinese). ROTH, K. L.; DENNISON, P. E.; AND ROBERTS, D. A., 2012, Comparing endmember selection techniques for accurate mapping of plant species and land cover using imaging spectrometer data: Remote Sensing of Environment, Vol. 127, pp. 139–152. SCHWEIK, C. M. AND GREEN, G. M., 1999, The use of spectral mixture analysis to study human incentives, actions, and environmental outcomes: Social Science Computer Review, Vol. 17, No. 1, pp. 40–63. SHEN, S. AND WANG, J., 2003, Principal factors in retrogression of the Zoige plateau marsh wet land: Acta Geologica Sichuan, Vol. 23, No. 2, pp. 123–125 (in Chinese). SHI, C. AND TU, J., 2009, Remote sensing monitory study on land desertification in Ruoergai Plateau of Sichuan Province during 40 years: Southwest China Journal of Agriculture Sciences, Vol. 22, No. 6, pp. 1662–1664 (in Chinese). SMALL, C., 2001, Estimation of urban vegetation abundance by spectral mixture analysis: International Journal of Remote Sensing, Vol. 22, No. 7, pp. 1305–1334. STATE FORESTRY ADMINISTRATION, P.R. CHINA (SFA), 2011, A Bulletin of Status of Desertification and Sandification in China: Electronic document, available at http://www.forestry. gov.cn// TRIPATHY, G. K.; GHOSH, T. K.; AND SHAH, S. D., 1996, Monitoring of desertification process in Karnataka state of India using multi-temporal remote sensing and ancillary information using GIS: International Journal of Remote Sensing, Vol. 17, No. 12, pp. 2243–2257. WANG, F., 2012, Study of Evaluation in Wetland Protection and Restoration Project: Unpublished Ph.D. Thesis, Chinese Academy of Forestry, Beijing. WANG, Q. AND BAO, W., 2002, Basic types and characters of the western Zoige meadows and their changes in recent decades: Chinese Journal of Applied & Environmental Biology, Vol. 8, No. 2, pp. 133–141 (in Chinese).
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 223–234
233
Xian, Xiang, Liu, and Shao WANG, T., 2004, Study on sandy desertification in China: Key regions for studying and combating sandy desertification: Journal of Desert Research, Vol. 24, pp. 1–9 (in Chinese). WANG, Y. AND ZHAO, Z., 2005, Characteristics of the climatic variation in Zoige in the past 45 years and its effects on the eco-environment in the area: Journal of Geo-mechanics, Vol. 11, No. 4, pp. 328–332 (in Chinese). WU, B. AND CI, L., 2002, Landscape change and desertification development in the Mu Us Sandland, Northern China: Journal of Arid Environments, Vol. 50, No. 3, pp. 429– 444.
234
YONG, G.; SHI, C.; AND QIU, P., 2003, Remote sensing monitoring of sand expansion at fragile grassland ecosystem in Ruoergai Plateau, Sichuan, China: Journal of Mountain Science, Vol. 21, No. 6, pp. 758–762 (in Chinese). ZHANG, S.; GUO, H.; AND LUO, Y., 2007, Assessment on driving force of climate change & livestock grazing capacity to grassland sanding in Ruoergai: Chinese Journal of Grassland, Vol. 29, No. 5, pp. 64–71 (in Chinese). ZHAO, K. AND HE, C., 2000, Influence of human activities on the Mire in Zoige Plateau and countermeasure: Scientia Geographica Sinica, Vol. 20, No. 5, pp. 444–449 (in Chinese).
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 223–234
Microfabrics-Based Approach to Predict Uniaxial Compressive Strength of Selected Amphibolites Schists Using Fuzzy Inference and Linear Multiple Regression Techniques ESAMALDEEN ALI1 Department of Geology, Faculty of Petroleum and Minerals, Al Neelain University, Khartoum, Sudan
WU GUANG Department of Engineering Geology, Faculty of Geosciences and Environmental Engineering, Southwest Jiaotong University, Chengdu 610031, China
ABDELAZIM IBRAHIM Department of Geology, Faculty of Petroleum and Minerals, Al Neelain University, Khartoum, Sudan
Key Terms: Fuzzy Inference System, Microfabrics, Multi-variate Regression, Uniaxial Compressive Strength, Amphibolite Schists
ABSTRACT In this paper, we established the prediction capability for uniaxial compressive strength (UCS) from microfabric characterization of banded amphibolite schists using fuzzy inference system (FIS) and multiple linear regression (MLR) techniques. In this study, the method of semi-automatic petrographic image analysis (PIA) was adopted to calculate and measure the microfabric parameters. Based on statistical analysis, more influential microfabrics parameters that affect the UCS more than the others have been selected to predict UCS, which include grain size, shape factor, and quartz content. Multi-variate regression relations were established using the same input variables as the FIS model. To assess the performance of both models, some performance indices such as correlation coefficient (R), variance accounted for (VAF), and root mean square error (RMSE) were calculated and compared for the two models. The results show that both models reliably predict the UCS, with the multiple regression model being better based on the performance indices criteria. One of the most significant findings to emerge from this study is that the microfabrics-based PIA approach can be easily extended to the modeling of 1 Corresponding author email: esameldeen77@yahoo.com; phone: +249912992185.
strength and deformation behavior of rocks in the absence of adequate geological information or abundant data. INTRODUCTION Uniaxial compressive strength (UCS) is one of the most important intact rock parameters, and it is commonly used for a variety of engineering applications, such as rock mass classification and rock failure criteria, etc. (Jumikis, 1992). However, such tests require high-quality core samples, which cannot always be obtained, particularly from weak, stratified, highly fractured, and weathered rocks. Thus, many researchers use conventional statistical methods to estimate UCS from simple index parameters such as Schmidt hammer, point load, block punch, and petrographic properties (Ulusay et al., 1994; Raisa¨nen, 2004; Yagiz, 2008; Basu and Kamran, 2010; and Mishra and Basu, 2012). Although often neglected, microfabric characteristics of rocks are of great significance for the failure of rocks, especially if intense deformation is present (Seo et al., 2002). In fact, one obvious reason for this is the complexity of rock materials and the variation in quantitative relationships between different rock types. However, the geometrical parameters of rock-forming minerals (microfabrics) known to affect the mechanical properties of rocks include grain size, grain shape, grain packing, porosity, matrix type, and spatial arrangement of minerals visible in microscale (Jones, 1987). In the last few years, quantitative microfabric properties of rocks have been used as an acceptable
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 235–245
235
Ali, Guang, and Ibrahim
approach only for preliminarily estimation of rock strength (Singh et al., 2001; Gokceoglu, 2002; Gokceoglu and Zorlu, 2004; Jeng et al., 2004; Zorlu et al., 2008; Gokceoglu et al., 2009; Cevik et al., 2011; and Yesiloglu-Gultekin et al., 2013). For metamorphic rocks, Singh et al. (2001) constructed a neural network model to estimate UCS from mineral composition and texture properties for some schistose rocks similar to the rock units used for this study. However, one of the main disadvantages of their model is that they used less effective input parameters such as rock type and area weighting, as well as the secondary mineral contents (e.g., chlorite and mica). Although many attempts have been made in the past to predict UCS from mineral composition and the microfabric properties of rocks, still special attention should be given to understand the microfabric behavior of anisotropic rocks (e.g., metamorphic rocks) under stress. For safe and reliable design of engineering projects, anisotropic rocks represent a most crucial factor. For this purpose, banded amphibolite schists were selected in the current study due to their high rock-fabric complexity and their significant impact on engineering design. However, far too little attention has been paid to this rock unit. Thus, contrary to Singh et al.’s (2001) UCS model, which used less effective input parameters for predicting rock strength of different schistose rocks, this study aims at selecting relatively smaller numbers of rock microfabrics of amphibolite schists but more influential ones, which adequately represent the UCS of a given rock type. Therefore, in this study, the microfabric properties of selected rocks were obtained using petrographic image analysis (PIA). Data obtained from PIA and UCS were subjected to a series of comprehensive statistical analyses using SPSS V.19.0 (SPSS Statistics, 2010). Results of statistical analyses were employed to develop the predictive models using a combination of linear multivariate regression (MR) and fuzzy inference system (FIS) techniques. This effort is a complementary contribution to the study presented herein. The amphibolite schist samples were collected from varied rock slopes along the Chengdu-Wenchuan highway in Sichuan Province, China. Macroscopically, the rocks exhibit a distinct planar syn-metamorphic fabric that is characterized by near-perfect tectonic metamorphic layering and penetrative stretching lineation (Figure 1). However, the foliation bands vary in thickness from 1 to 3 mm. Recently, the study of Ali et al. (2013) has shown that the dynamic mechanical properties of these rocks have different values concerning the banding plane. They reported that banded amphibolite rocks have a U-shaped anisotropy with maximum strength at b 5 90 degrees,
236
Figure 1. Outcrop and hand specimen views of tectonic metamorphic layering and banding structure of the studied amphibolite schist rocks.
and minimum strength is obtained when b 5 30 degrees. Thus, neglecting microfabrics of such rocks in the engineering analysis may lead to unreliable results. METHODOLOGY AND DATABASE CONSTRUCTION Uniaxial Compressive Strength (UCS) Thirty representative blocks of banded amphibolite schists were collected for compressive strength tests. Each block sample was inspected for macroscopic defects such as fractures, partings, or alteration zones that would result in unreliable UCS values. Core specimens with length to diameter ratios of 2 were carried out perpendicular to the foliation plane using a conventional laboratory drilling machine. The core samples were prepared and tested following the ASTM (1995, 2001) suggested methods. The test procedure was repeated at least three times for each subtype sample, and the average value of UCS was recorded in the database. In this study, the UCS is considered as the dependent variable. Microscopic Description of Investigated Rocks In order to evaluate the microfabric properties of the investigated rocks, thin sections across the schistosity plane were prepared and observed under a high-power polarized microscope. Slight compositional heterogeneity and textural characteristics of the typical rocks were studied down to the centimeter scale. Mineral assemblages of subtype samples were determined using the X-ray diffraction method (XRD). The predominant mineralogical composition is quartz, hornblende, actinolite, orthoclase, and plagioclase, with rare pyroxene, and minor amounts of calcite, epidote, and chlorite. Iron oxides occur as accessory minerals. Microscopically, thin sections show that the rocks contain varying proportions of xenoblastic hornblende set in a fine- to mediumgrained quartz-plagioclase matrix. In some thin
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 235–245
Prediction of Uniaxial Compressive Strength
automated PIA was employed using a standard polarizing microscope and specialized software named TIGER 3000P (2012). The whole process of semi-automatic PIA is described in detail by Prˇikryl (2006), which consists of the following stages: Image Acquisition Image acquisition refers to the selection of a measured area (observation window) of the thin section that is photographed using a digital camera attached to the optical microscope. The dimension of each image is 1.15 mm (1,024 pixels) in height and 1.4 mm (768 pixels) in length with special resolution. In total, 25–36 images are captured from each thin section and then mosaicked to represent the thin section. The average grain size of the rock represents the main factor for the selection of the analyzed area, since at least 300–500 grains should be measured. Figure 2. Petrographic panorama view of thin section of amphibolite schists: (A) aligned hornblende crystals with the foliation as the symmetry plane, (B) preferred orientation of hornblende and deformed quartz and feldspar grains in quartz-rich zone; (C) quartz and plagioclase porphyroblasts in highly deformed schistose amphibolite.
sections, aligned hornblende crystals with an average length of 0.1 mm occurred, although large crystals reached 0.25 mm and usually contained numerous small inclusions of quartz (Figure 2A). The main microstructure of these rocks is a well-developed layered lattice structure that is defined by strong preferred orientation of hornblende and elongated quartz grains in a quartz-rich zone (Figure 2B and C). This well-defined layering in microscopic scale indicates that these rocks are characterized by a higher degree of anisotropy. In some thin sections, numerous porphyroblasts of quartz and plagioclase considerably deflect the schistosity plane (Figure 2C), which may cause some variation in strength measurements. Some parts of fine-grained rocks were composed almost entirely of small granular epidotes, actinolites, and chlorites, which are formed after hornblende. The presence of slight amounts of highpressure/low-temperature mineral assemblages (e.g., chlorite, actinolite and epidote), along with the aforementioned minerals, indicates that these rocks are low-grade amphibolites of igneous origin.
Image Pre-Processing Image pre-processing includes the step in which the images are graphically modified to increase the contrast between the measured objects and the background by image analysis. This requires identifying hidden grain boundaries showing similar optical features on a micrograph. However, the accuracy of data analysis increases by increasing the identification of precise grain boundaries. Image Digitizing Image digitizing focuses on drawing outlines of grain boundaries and preparation of a map of mineral grains from the thin sections. Due to the high degree of metamorphism and schistose texture of amphibolites, the study concentrated on more resistant minerals, which include quartz, feldspars, and plagioclase grains (Figure 3). Edges that did not form closed boundaries were neglected. Grain boundaries were manually drawn from the photomicrographs using Corel Draw Software. In this stage, it is necessary to adjust the map resolution to get real values of grain parameters compared with the original thin section. However, an accurate map of the grains forms the most crucial factor for obtaining perfect analysis of selected geometrical parameters.
Quantitative Petrographic Image Analysis In recent years, the use of quantitative PIA as an efficient technique for the numerical description of rock microfabrics has become more common (Ersoy and Waller, 1995; Prˇikryl, 2006). In this study, semi-
Image Measurement and Data Analysis For measurement purposes, a reference area was selected based on the rock grain sizes that contain a number of grains. Grains touching the observation
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 235–245
237
Ali, Guang, and Ibrahim
Figure 3. Grain boundaries and preparation of a map of mineral grains from the thin sections.
window were not measured. The polarized software TIGER 3000P was employed, which offers featurespecific measurement during which each grain is measured independently. The grain parameters were measured directly, including grain area (GA), grain size (GS), perimeter, and maximum and minimum axis length for each grain (Figure 4). The two secondary geometrical parameters, aspect ratio (AR) and shape factor (SF), were calculated indirectly, which describe grain shape parameters. The definitions and calculation of these parameters are illustrated in Table 1. The average value for each parameter was calculated and recorded in the database. STATISTICAL ANALYSES OF THE EXPERIMENTAL DATA The statistical data consist of eight microfabric properties set as explanatory variables and UCS as the dependent variable. The microfabric parameters used in this study include grain size, grain area, shape factor, aspect ratio, K-feldspar (%), plagioclase (%), quartz (%), and hornblende (%). However, the statistical
Figure 4. Geometric parameters of grains measured by petrographic image analysis.
238
analysis consists of determining any meaningful correlations among these variables. The variables with the most significant influence compared to the others were considered useful indicators for predicting rock strength. The descriptive statistics shown in Table 2 provide the minimum, maximum, median, means, and standard deviations for each independent and dependent variable used in this study. To visualize the relationship between USC and each measured microfabric property, the original data set was subjected to bivariate correlation and curve fitting analyses (Figure 5). Pearson’s correlation coefficient (r) was investigated to see whether a trend was sufficiently correlated (Table 3). Results from this linear correlation revealed five of the eight predictive variables exhibited statistically significant correlations with UCS at the 95 percent confidence level, which include grain size, grain area, shape factor, quartz content, and hornblende content (Table 3). Other parameters of microfabrics that showed poor correlation with UCS include aspect ratio (r 5 0.13), as well as contents of feldspar and plagioclase (r 5 0.27 and 0.3, respectively) at the 95 percent confidence level. The results indicate that grain size is the most significant microfabric property affecting UCS, with a negative r-value of 0.85 for UCS, which is in line with the previous studies by Prˇikryl (2006) and Raisa¨nen (2004). In this study, there was moderate correlation between quartz content of amphibolite schists and its strength (r 5 0.52). Fahy and Guccione (1979); Shakoor and Bonelli (1991) reported that the quartz content of sandstones shows a weak positive correlation with the rock strength. To establish a proper simple regression model for UCS using each independent variable, different empirical equations, such as logarithmic, power, and
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 235–245
Prediction of Uniaxial Compressive Strength Table 1. Basic microstructural parameters measured by the petrographic image analysis of thin sections. Parameter
Unit
Computation 2
Grain area (A) (Grain size) Equivalent diameter
pixels, mm pixels, mm
Major axis length (Dmax)
pixels, mm
Minor axis length (Dmin)
pixels, mm
Perimeter (LP) Aspect ratio (AR)
pixels, mm –
Shape factor (SF) (Howarth and Rowlands, 1986)
Area reports the area for the selected object. Equivalent diameter is calculated as a diameter of circle, which has the same area (A) as measured object 2 Dequiv 5 (4 3 A/p)1/2 Major axis of the object defined by the two most distant points on the object (reports the length of the axis). Minor axis of the object defined by the two most distant points on the object that creates a line perpendicular to the major axis. Length of all edge pixels outlining the objects (length of grain boundary) Dmax The ratio between the minor and major axis lengths (A~ ) Dmin Shape factor is a measure of the grain’s deviation from circularity. Ideal circle shows a shape parameter 1; objects with elongated or irregular shape show a shape factor close to 0. SF 5 4 3 p 3 A/LP2.
exponential, along with the linear model, were tried to fit the data. Models fitting the data best for the predictions of UCS for five independent variables are given in Table 4. The results revealed that linear relations were fitted with higher correlations coefficients than the other models. Thus, in this study, a linear regression model was preferred for predicting UCS and further will be adopted in the multiple regression analysis. MODEL DEVELOPMENT STUDIES Linear Multi-Variate Regression (LMR) Most problems in geology involve complex and interacting factors, which are impossible to isolate and study individually (Davis, 1973). Therefore, the LMR technique is used to construct such complex prediction models. In addition, multiple regression analysis may also be used to make a choice among the existing independent variables in order to obtain the most appropriate model equation that explains the dependent variable according to the aim of usage (Harrell, 2001). Based on bivariate correlation, microfabric parameters that significantly correlated with UCS at a 95 percent confidence were selected to build the model. Therefore, the multiple regression
model was calculated using five independent variables, which include grain size, grain area, shape factor, quartz content, and hornblende content. From the results of the simple regression analysis, the multiple regression function was assumed to be linear. Thus, the equation representing this model can be written in the following form: Y ~b0 zb1 X1i zb2 X2i z . . . zbn Xki
ð1Þ
where Y is the predicted value corresponding to the dependent variable, b0 is the intercept (constant), X1, X2, and Xn are the predictors, and b1, b2, and bn are the regression coefficients of X1, X2, and Xn. An important step in a multiple regression is to ensure that the assumption of no multi-collinearity has been met (Draper and Smith, 1981). However, the effect of any given explanatory variable depends on which other variables have been included in the regression model; this situation indicates the multicollinearity phenomenon (Harrell, 2001). Thus, the statistical criteria, F-test and t-test, tell us, respectively, whether there is a linear relationship between the response and all explanatory variables taken together, and whether any given explanatory variable has an influence on the dependent variable over that of the other explanatory variables at 5 percent or 10 percent
Table 2. Basic descriptive statistics for the original data set. Variables
Min
Max
Mean
Median
Std. Deviation
Variance
Quartz content (%) Plagioclase content (%) K-feldspar content (%) Hornblende content (%) Grain size (mm) Grain area (mm2) Aspect ratio Shape factor UCS (MPa)
26.6 4.23 8.70 3.89 0.017 226.9 0.612 0.499 78.0
53.0 13.00 17.01 30.96 0.056 2462 0.685 0.698 123.0
39.66 8.06 12.69 18.00 0.033 962.5 0.657 0.618 105.56
39.35 7.46 12.00 17.93 0.032 1.221 0.670 0.620 106.75
5.34 2.58 2.45 7.12 0.011 619.3 0.024 0.043 10.70
28.54 6.66 6.00 50.71 0.000 383558 0.001 0.002 114.49
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 235–245
239
Ali, Guang, and Ibrahim Table 3. Results of correlation coefficient (r) for simple regression analyses for UCS with the different independent variables. Petrographic Characteristics
Correlation Coefficient (r)
Quartz (%) Hornblende (%) Plagioclase (%) K-feldspar (%) Grain size (mm) Grain area (mm2) Aspect ratio Shape factor
0.52a 0.69a 0.30 0.27 20.85a 20.84a 0.13 0.75a
a
Variables have significant relation at 95 percent confidence level.
then grain area (GA), and so on. By using this process, the four prediction models that emerged from multiple regression analysis are shown below: UCS~126:405{416:308|GS {0:007|GA R2 ~0:733
ðmodel 1Þ
UCS~76:75{616:803|GS z79:643|SF R2 ~0:783
ðmodel 2Þ
UCS~52:214{527:77|GSz80:86|SF ðmodel 3Þ z0:526|Qtz R2 ~0:844
Figure 5. Bivariate linear regression and cure fitting of UCS against the microfabric characteristics of rocks.
significance levels using the null hypothesis (Draper and Smith, 1981). Stepwise Regression Method The main purpose of the stepwise regression method is to select a subset of variables entirely by statistical criteria (e.g., t-test). The method combines forward selection and backward elimination. In this study, forward selection is used. The selection starts with the independent variable that is the best predictor of UCS and checks that the coefficient is significant at the 5 percent confidence level based on a t-test. The model adds another variable that improves the prediction level and that passes the significance criteria at the 5 percent limit. At each step, all variables currently in the model are checked to see if any can be removed using t-test. The process continues until no more variables can be added or removed. As mentioned above, grain size (GS) was the best predictor of UCS and was entered first, and
240
UCS~43:1{399:08|GSz78:84|SF ðmodel 4Þ z0:552|Qtzz0:28|Hb R2 ~0:86 The statistical summary of these four models is given in Table 5. One can see that R2 value always increases with the additional predictor variables. Thus, the R2 statistic is not useful as an indicator for comparison between different sets of data. Therefore, the significance of R2 values can be determined by the t-test, assuming that both variables are normally distributed. Based on the F-test, the four models can be used to predict UCS, as they are statistically significant at the 5 percent confidence limit. By considering the t-test, the grain area (GA) and/or hornblende (Hb) variables, when included in the model, are insignificant at the 95 percent confidence level (models 1 and 4). Nevertheless, shape factor (SF) is good predictor of UCS at the 95 percent confidence level; it turns out that, in the presence of grain size (GS) (model 2), it made an insignificant contribution to predict UCS at the 5 percent confidence level (Table 5). From this result, one can conclude that model 3 is good among these five models to predict the UCS of amphibolite schists. However, the model revealed that approximately 84.4
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 235–245
Prediction of Uniaxial Compressive Strength Table 4. Empirical equations (simple regression models) for prediction of UCS from micropetrographic properties of studied rocks. Function Type Linear
Logarithmic
Power
Exponential
Independent Variable (x)
Simple Regression Models
Correlation Coefficient (r2)
Significance Level
Qtz Hb GS GA SF Qtz Hb GS GA SF Qtz Hb GS GA SF Qtz Hb GS GA SF
UCS 5 64.332 + 1.04x UCS 5 86.776 + 1.043x UCS 5 132.654 2 820.197x UCS 5 116.3 2 8.379x UCS 5 187.319x 2 10.122 UCS 5 42.3lnx 2 49.74 UCS 5 66.9 + 13.8lnx UCS 5 17.8 2 25.3lnx UCS 5 105.2 2 9.76lnx UCS 5 159.76 + 111.9lnx UCS 5 21.024x0.438 UCS 5 72.49x0.133 UCS 5 44.59x20.25 UCS 5 104.6x20.09 y 5 178.78x1.1 UCS 5 68.59e0.011x UCS 5 87.66e0.01x UCS 5 137.06e28.06x UCS 5 116.83e20.08x UCS 5 33.79e1.84x
0.27 0.48 0.73 0.70 0.56 0.30 0.39 0.67 0.65 0.55 0.33 0.37 0.66 0.64 0.54 0.30 0.46 0.72 0.70 0.55
0.003 0.000 0.000 0.000 0.000 0.002 0.000 0.000 0.000 0.000 0.001 0.000 0.000 0.000 0.000 0.002 0.000 0.000 0.000 0.000
Qtz 5 quartz, Plg 5 plagioclase, Kfds 5 K-feldspar, Hb 5 hornblende, GS 5 grain size, GA 5 grain area, AR 5 aspect ratio, SF 5 shape factor.
percent of the variance of the UCS could be estimated by the linear combination of grain size, shape factor, and quartz content. However, a higher value of grain size predicts a lower value of UCS, while a higher value of the other two variables predicts a higher value of UCS. A scatter plot of the fitted UCS values against those observed for the multiple linear regression model is given in Figure 6, with correlation coefficient r 5 0.919, which is statistically significant at the 95 percent level. Furthermore, these three selected micro-petrographic properties will be employed for predicting UCS using the fuzzy inference system. Fuzzy Inference System (FIS) The fuzzy inference system (FIS) emerged as a tool to deal with uncertain, imprecise, or qualitative decision-making problems that can be easily expressed through a set of linguistic fuzzy rules (IFTHEN rules) (Zadeh, 1994). However, one of the biggest advantages of fuzzy logic is its capability to describe complex problems in a transparent way (Setnes et al., 1998). Recently, FIS progressed as a powerful technique in the areas of rock mechanics and engineering geology, where several problembased models have been carried out (e.g., Finol et al., 2001; Gokceoglu and Zorlu, 2004; Tutmez and Hatipoglu, 2007; and Gokceoglu et al., 2009). There are four typical components required to design a fuzzy system (Zadeh, 1994):
1. Fuzzification: The fuzzification module converts the crisp numerical values into the degrees of membership (i.e., fuzzy values) related to the corresponding fuzzy sets. A fuzzy set is represented by a linguistic term such as ‘‘small,’’ ‘‘medium,’’ or ‘‘high’’ (Zadeh, 1973). Fuzzy sets allow an object to be a partial member of a set (membership function), which is the curve that defines how each point in the universe of discourse is mapped to a membership value in the range of [0…1]. 2. Rule base: It is a process to map the input space to the output space. Rules here are the bridge between the input and the output space. The rule base is simply a series of IF-THEN rules that relate the input fuzzy variables and the output variable using linguistic variables, each of which is described by a fuzzy set and fuzzy implication operator AND. 3. Inference engine: It attempts to simulate human decision making based on fuzzy concepts. It involves matching fuzzy facts against the antecedents-part in the rules. 4. Defuzzification: Defuzzification is such inverse transformation that maps the output from the fuzzy domain back into the crisp domain. There are several defuzzification methods available, such as center of gravity, center of area, middle of maximum, etc. In the last two decades, several rule-based fuzzy modeling methods have been proposed to solve
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 235–245
241
0.554 0.554
.514 .554 .884
.331 .553 .876 .476
.000
.000
.000
3.02 1.809 1.142 2.101
1.945 1.805 1.131
1.804 1.804
0.031 0.031 .000
32.56 32.56
VIF Tolerance Sig.
38.48
46.76
48.644
Sig.
.000 .452 .459 .001 .000 .014 .015 .000 .005 .004 .042 .004 .005 .002 .097
t
14.109 20.762 20.751 3.573 25.322 2.628 2.597 25.073 3.086 3.181 2.145 23.194 3.118 3.45 1.726
F
37.017
F-test
4
3
2
GS 5 grain size; SF 5 shape factor; GA 5 grain area; Qtz 5 quartz; Hb 5 hornblende.
20.415 0.314 0.276 0.187
20.549 0.322 0.262
20.641 20.317
20.433 20.427
Beta
8.959 546.017 0.010 21.480 115.893 30.301 20.109 104.036 26.2 .165 20.091 124.968 25.284 0.16 0.163
Std. Error B
126.405 2416.308 20.007 76.75 2616.803 79.643 52.214 2527.769 80.86 0.526 43.103 2399.084 78.84 0.552 0.281 Constant GS GA Constant GS SF Constant GS SF Qtz Constant GS SF Qtz Hb
Model
1
242
Figure 6. Scatter plots of the values predicted by the linear multiple regression model and those observed for UCS.
T-test Standardized Coefficients Unstandardized Coefficients
Table 5. Statistical summary and hypothesis tests of the multiple regression models.
Collinearity Statistics
Ali, Guang, and Ibrahim
engineering problems (Mamdani and Assilan, 1975; Pedrycz, 1984; and Sugeno and Kang, 1988). In this study, the Mamdani fuzzy algorithm was preferred, because this method is perhaps the most appealing fuzzy method to employ in engineering geological problems (Alvarez Grima, 2000). The rules of Mamdani fuzzy are of the general form: IF antecedent(s) THEN consequent(s), where antecedents and consequents are propositions containing linguistic variables. In this study, the FIS model includes three inputs (grain size, shape factor, and quartz content) and one output (UCS) (Figure 7). The graphical illustrations of the membership functions are given in Figure 8. The total number of data-driven linguistic rules of the fuzzy inference system is 125 rules. The total number of rules is defined by the following equation (Zadeh, 1973): RI ~X A
ð2Þ
where RI is total number of rules used in the model, X is the number of membership functions for each variable, and A is number of input variables. An example of the IF-THEN rules is as follows: IF GS is high and SF is very low, and Qtz is very low, THEN UCS is very low. In this study, MATLAB Version R2010a (MATLAB, 2010) was utilized during the construction process of the fuzzy inference model. The centroid
Figure 7. The general architecture of the fuzzy inference system constructed in the study.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 235–245
Prediction of Uniaxial Compressive Strength
Figure 8. Membership graphs for the fuzzy inference system used in the study.
method is preferred for calculating the output from the fuzzy domain back into the crisp domain (defuzzification) (Hellendoorn and Thomas, 1993). The influence of input parameters on the output in the FIS model can be illustrated through graphical representation for visual perception. Consequently, the variation of UCS is plotted against each of three independent variables (Figure 9). Finally, the crosscorrelation relationship between predicted and measured UCS values was carried out, and a strong correlation coefficient was obtained (r 5 0.91) (Figure 10). Generally, during experimental studies, measurement of high values may result in some data scatter. However, this fair agreement is not sourced from the modeling, but the main source of this observation is the nature of the measurements.
Figure 9. Graphical representation showing the variation of UCS against the different independent variables.
accounted for (VAF) (Eq. 3) and root mean square error (RMSE) (Eq. 4) were calculated and compared for each model, and the results are provided in Table 6. ! 0 var y{y VAF ~ 1{ |100 ð3Þ varð yÞ
Performance Assessment of the Two Models Similar input parameters as those used in LMR were used to construct the two models. To assess the models performances, some performance indices (Alvarez Grima and Babuska 1999; Finol et al., 2001) such as correlation coefficient (r), variance
Figure 10. Scatter plots of the values predicted by fuzzy models and those observed for UCS.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 235–245
243
Ali, Guang, and Ibrahim Table 6. The statistical performance indices for both prediction models in this study. Performance Indices Model Multi-variate regression model Fuzzy inference system
R
RMSE
VAF
0.92 0.91
4.16 6.1
97.0 95.2
*VAF 5 Variance Account For.
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u N uI X RMSE~t ðy{y0 Þ2 N i~1
ð4Þ
where y and y9 are the measured and the predicted values, respectively, and N is the number of data. The above performance indices are interpreted as follows: the higher the VAF, the better the model performs. The lower the RMSE, the better the model performs. When the two models are compared, the results produced by the proposed multiple regression model are observed to be better in predicting UCS (Table 6). Although, good cross-correlations between predicted and measured UCS for both multiple linear regression and FIS prediction models exist, the Rvalue for the regression model (R 5 0.92) is slightly greater than that for the FIS model (R 5 0.91). As can be seen in Table 6, the RMSE statistic for the multiple linear regression model is lower than that for FIS. Moreover, the VAF in the multiple regression is higher than that of the FIS model. Therefore, this study indicates that the UCS model that was developed using the linear regression model to predict the actual UCS values is better than those developed using the fuzzy inference system. CONCLUSION This study was undertaken to formulate the UCS of banded amphibolite schists from their microfabric attributes using linear multivariate regression and fuzzy inference system (FIS) techniques. The following results and conclusions can be drawn from the present study: 1. The study revealed that the technology of PIA is an acceptable approach in the preliminarily estimation of rock strength. The biggest advantage of this technique is its significant time saving and cost-effectiveness, while an experimental work is needed to accommodate such practice. However, this approach is easy to use in the absence of adequate geological information or abundant data.
244
2. Contrary to previous UCS models from textural characteristics, which use less effective input parameters (e.g., area weighting and secondary minerals such as chlorite and mica), this study employed a relatively smaller number of variables but more influential ones, which adequately predict the UCS of a given rock type. However, grain size, shape factor, and quartz content were defined as dominant microfabric parameters, which are tailored to the specific estimation of UCS in this study. This can increase popularity of this technique for evaluating mechanical properties of rocks from textural properties. 3. Although both models (LMR, FIS) are reliable to estimate the UCS, the study revealed that the LMR model is better than the FIS model, as indicated by the consistency of the performance indices for the two models. 4. Results of this study are valid for the tested rock unit; however, further study would be helpful to improve the reliability of concluded results on the prediction of UCS from microfabric properties for different rock types.
ACKNOWLEDGMENTS The authors greatly acknowledge Southwest Jiaotong University, China, which financially supported and provided logistics for this work. Special thanks are due to Professor Abdul Shakoor, co-editor of Journal of Environmental and Engineering Geoscience, for his excellent comments and follow up on this article. REFERENCES ALI, E.; WU, G.; ZHAO, Z. M.; AND JIANG W. X., 2013, Assessments of strength anisotropy and deformation behavior of banded amphibolite rocks: Journal of Geotechnical and Geological Engineering, Vol. 32, pp. 429–438, DOI:10.1007/ s10706-013-9724-5. ALVAREZ GRIMA, M., 2000, Neuro-fuzzy Modeling in Engineering Geology: A.A. Balkema, Rotterdam, 244 p. ALVAREZ GRIMA, M. AND BABUSKA, R., 1999, Fuzzy model for the prediction of unconfined compressive strength of rock samples: International Journal of Rock Mechanics and Mining Science, Vol. 36, pp. 339–349. ASTM, 1995, Standard Test Method for Unconfined Compressive Strength of Intact Rock Core Specimens: ASTM D2938-95 (Reapproved 2002), United States. ASTM, 2001, Standard Practice for Preparing Rock Core Specimens and Determining Dimensional and Shape Tolerances: ASTM D4543, United States. BASU, A. AND KAMRAN, M., 2010, Point load test on schistose rocks and its applicability in predicting uniaxial compressive strength: International Journal of Rock Mechanics & Mining Sciences, Vol. 47, pp. 823–828.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 235–245
Prediction of Uniaxial Compressive Strength CEVIK, A.; SEZER, E. A.; CABALAR, A. F.; AND GOKCEOGLU, C., 2011, Modelling of the uniaxial compressive strength of some clay-bearing rocks using neural network: Applied Soft Computing, Vol. 11, No. 2, pp. 2587–2594. DAVIS, J. C., 1973, Statistics and Data Analysis in Geology: Wiley, New York, 550 p. DRAPER, N. AND SMITH, H., 1981, Applied Regression Analysis, 2nd ed.: John Wiley & Sons, Inc., New York. ERSOY, A. AND WALLER, M. D., 1995, Textural characterization of rocks: Engineering Geology, Vol. 39, No. 3–4, pp. 123–136. FAHY, M. P. AND GUCCIONE, M. J., 1979, Estimating the strength of sandstone using petrographic thin section data: Bulletin International Association of Engineering Geology, Vol. 16, No. 4, pp. 467–485. FINOL, J.; GUO, Y. K.; AND JING, X. D., 2001, A rule based fuzzy model for the prediction of petrophysical rock parameters: Journal of Petroleum Science and Engineering, Vol. 29, pp. 97–113. GOKCEOGLU, C., 2002, A fuzzy triangular chart to predict the uniaxial compressive strength of the Ankara agglomerates from their petrographic composition: Engineering Geology, Vol. 66, No. 1–2, pp. 39–51. GOKCEOGLU, C.; SONMEZ, H.; AND ZORLU, K., 2009, Estimating the uniaxial compressive strength of some clay bearing rocks selected from Turkey by nonlinear multivariable regression and rule-based fuzzy models: Expert Systems, Vol. 26, No. 2, pp. 176–190. GOKCEOGLU, C. AND ZORLU, K., 2004, A fuzzy model to predict the uniaxial compressive strength and the modulus of elasticity of a problematic rock: Engineering Applications of Artificial Intelligence, Vol. 17, pp. 61–72. HARRELL, F. E., 2001, Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression and Survival Analysis: Springer-Verlag, New York. HELLENDOORN, H. AND THOMAS, C., 1993, Defuzzification of fuzzy controllers: Journal of Intelligent Fuzzy Systems, Vol. 1, pp. 109–123. HOWARTH, D. F. AND ROWLANDS, J. C., 1986, Development of an index to quantify rock texture for qualitative assessment of intact rock properties: Geotech Testing, Vol. 9, pp. 169–179. JENG, F. S.; WENG, M. C.; LIN, M. L.; AND HUANG, T. H., 2004, Influence of petrographic parameters on geotechnical properties of Tertiary sandstones from Taiwan: Engineering Geology, Vol. 73, pp. 71–91. JONES, M. P., 1987, Applied Mineralogy: A Quantitative Approach: Graham and Trotman Limited, London, 259 p. JUMIKIS, A. R., 1992, Rock Mechanics: Trans Tech Publications, Clausthal, Germany. MAMDANI, E. H. AND ASSILAN, S., 1975, An experiment in linguistic synthesis with a fuzzy controller: International Journal of Man-Machine Studies, Vol. 7, No. 1, pp. 1–13. MATLAB, 2010, Statistics Toolbox for Use with MATLAB, User’s Guide Version: The MathWorks, Inc, United States. MISHRA, D. A. AND BASU, A., 2012, Use of the block punch test to predict the compressive and tensile strengths of rocks: International Journal of Rock Mechanics & Mining Sciences, Vol. 51, pp. 119–127, electronic document available at http:// dx.doi.org/10.1016/j.ijrmms.01.016. PEDRYCZ, W., 1984, An identification algorithm in fuzzy relational systems: Fuzzy Sets and Systems, Vol. 13, pp. 153–167.
PRˇIKRYL, R., 2006, Assessment of rock geomechanical quality by quantitative rock fabric coefficients: Limitations and possible source of misinterpretations. Engineering Geology, Vol. 87, No. 3–4, pp. 149–162. RAISA¨NEN, M., 2004, Relationships between texture and mechanical properties of hybrid rocks from the Jaal–Iitti complex, southeastern Finland: Engineering Geology, Vol. 74, No. 3–4, pp. 197–211. SEO, Y. S.; JEONG, G. C.; KIM, J. S.; AND ICHIKAWA, Y., 2002, Microscopic observation and contact stress analysis of granite under compression: Engineering Geology, Vol. 63, pp. 259–275. SETNES, M.; BABUSKA, R.; AND VERBRUGGEN, H. B., 1998, Rulebased modeling: Precision and transparency: IEEE Transactions on Systems Man and Cybernetics, Vol. 28, No. 1, pp. 165– 169. SHAKOOR, A. AND BONELLI, R. E., 1991, Relationship between petrographic characteristics, engineering properties, and mechanical properties of selected sandstones: Bulletin International Association of Engineering Geology, Vol. 26, No. 1, pp. 55–71. SINGH, V. K.; SINGH, D.; AND SINGH, T. N., 2001, Prediction of strength properties of some schistose rocks from petrographic properties using artificial neural networks: International Journal of Rock Mechanics & Mining Sciences, Vol. 38, pp. 269– 284. SPSS STATISTICS 19.0, 2010, Core System User’s Guide: SPSS Inc, United States. SUGENO, M. AND KANG, G. T., 1988, Structure identification of fuzzy models: Fuzzy Sets and Systems, Vol. 28, pp. 15–33. TIGER 3000P, 2012, TIGER 3000P Polarized software TC-M013: Chongqing Science and Technology Company (In Chinese), Zhongqing, China. TUTMEZ, B. AND HATIPOGLU, Z., 2007, Spatial estimation model of porosity: Computers and Geosciences, Vol. 33, pp. 465–474. ULUSAY, R.; TURELI, K.; AND IDER, M. H., 1994, Prediction of engineering properties of a selected litharenite sandstone from its petrographic characteristics using correlation and multivariate statistical techniques: Engineering Geology, Vol. 37, pp. 135–157. YAGIZ, S., 2008, Predicting uniaxial compressive strength, modulus of elasticity and index properties of rocks using the Schmidt hammer: Bulletin of Engineering Geology and Environment, Vol. 68, pp. 55–63, DOI:10.1007/s10064-0080172-z. YESILOGLU-GULTEKIN, N.; SEZER, E. A.; GOKCEOGLU, C.; AND BAYHAN, H., 2013, An application of adaptive neuro fuzzy inference system for estimating the uniaxial compressive strength of certain granitic rocks from their mineral contents: Expert Systems with Applications, Vol. 40, No. 3, pp. 921– 928. ZADEH, L. A., 1973, Outline of a new approach to the analysis of complex systems and decision processes: IEEE Transactions on Systems Man and Cybernetics, Vol. SMC-3, pp. 28–44. ZADEH, L. A., 1994, Soft computing and fuzzy logic: IEEE Software, Vol. 11, No. 6, pp. 48–56. ZORLU, K.; GOKCEOGLU, C.; OCAKOGLU, F.; NEFESLIOGLU, H. A.; AND ACIKALIN, S., 2008, Prediction of uniaxial compressive strength of sandstones using petrography based models: Engineering Geology, Vol. 96, No. 3–4, pp. 141–158.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 235–245
245
Book Review Manual of Soil Laboratory Testing (K.H. Head and R.J. Epps) Review by: Richard Jackson 11 Venus Crescent, Geofirma Engineering Ltd., Heidelberg, Ontario N0B 2M1, Canada
Over the past 25 years, K. H. Head and Roger Epps have prepared three editions of their classic geotechnical laboratory text in three volumes. For the environmental and engineering geoscientist, the manual is most helpful in explaining the principles of testing soil materials and how the numerical values are generated. Thus, expertise in data interpretation is enhanced by inspection and mastery of the testing methods. The third edition of Manual of Soil Laboratory Testing. Volume III: Effective Stress Tests (2014), was recently published by the Scottish publishing house Whittles, thus completing the updating of the test methods. Therefore, this is a suitable time to review this multi-volume textbook. Some may recognize the engineering geologist Roger Epps as the lead author of the chapter on ground investigation and testing in the Geological Society of London’s 2012 monograph on Hot Deserts, reviewed in this journal by Robert Watters (2013) of the University of Nevada. Volume I of the series was originally published in 1980 to describe the U.K. standards, but it has been revised over the years along with the other volumes to account for changes in the British standards and the adoption of tests in 2010 developed by the European Union, i.e., Eurocode 7, Geotechnical Design, which are replacing the British Standards Institute tests. Most importantly for North American readers, the manual identifies those tests that are similar to or the same as the standardized tests of the American Society of Testing Materials (ASTM). Thus, in Volume I, Chapter 4, on particle size, Procedure 4.8.3 Hydrometer Analysis identifies both the BS 1377 standard and the ASTM D 422 standard as references. The text clarifies that two types of hydrometer are specified by ASTM. Similarly, consolidation testing (BS 1377) is described in Volume II in detail, and then a section is devoted to describing the specific differences in the ASTM consolidation
test (D 2435), e.g., specimen size, load pressures, and equipment. The third edition of Volume I (soil classification and compaction tests; Head, 2006) includes seven chapters: (1) scope and general requirements; (2) moisture content and index tests; (3) density and particle density; (4) particle size; (5) chemical tests; (6) compaction tests; and (7) description of soils. The third edition of Volume II (permeability, shear strength, and compressibility tests; Head and Epps, 2011) has a further seven chapters: (8) scope, equipment, and laboratory practice; (9) preparation of test specimens; (10) permeability and erodibility tests; (11) California bearing ratio; (12) direct shear tests; (13) undrained compression tests; and (14) oedometer consolidation tests. The third edition of Volume III (effective stress tests; Head and Epps, 2014) consists of a final eight chapters: (15) effective stress testing principles: theory and practice; (16) stress paths in triaxial testing; (17) test equipment; (18) calibrations, corrections, and general practice; (19) routine effective stress triaxial tests; (20) further triaxial shear strength tests; (21) triaxial consolidation and permeability tests; and (22) hydraulic consolidation and permeability tests. The three-volume set is not a cookbook for laboratory technicians, although that is certainly one of the uses considered by the authors. Each volume contains the theoretical basis of measurement that is most helpful to geoscientists who may never have taken a course in geotechnical engineering or soil mechanics. Thus, Volume II, Chapter 8 describes the use, calibration, and care of geotechnical laboratory instruments from pressure transducers to load rings. Later in this volume, Chapter 10 presents a useful introduction to the laboratory measurement of hydraulic conductivity, therein referred to by the old and incorrect terminology ‘‘permeability’’ or ‘‘coefficient of permeability,’’
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 247–248
247
Book Review
which faded from use in North America in the 1970s as we realized the terminological error and the ambiguity when considering multiphase flow. The introductory chapter in Volume III, ‘‘Effective Stress Testing Principles: Theory and Applications,’’ provides a very useful introduction for geoscientists to the concept of effective stress and the theory of triaxial testing. The descriptions are clear, the chapters are well organized, and the illustrations are superb. This splendid manual belongs in every university library patronized by geoscientists and engineers. Through the kindness of Whittles Publishing, the review set now is resident here in the library of the University of Waterloo.
248
REFERENCES HEAD, K. H., 2006, Manual of Soil Laboratory Testing. Volume I: Soil Classification and Compaction Tests, Vol. 3rd ed.: Whittles Publishing, distributed in North America by CRC Press LLC, Boca Raton, FL. HEAD, K. H. AND EPPS, R. J., 2011, Manual of Soil Laboratory Testing. Volume II: Permeability, Shear Strength and Compressibility Tests, Vol. 3rd ed, Whittles Publishing, distributed in North America by CRC Press LLC, Boca Raton, FL. HEAD, K. H. AND EPPS, R. J., 2014, Manual of Soil Laboratory Testing. Volume III: Effective Stress Tests, Vol. 3rd ed.: Whittles Publishing, distributed in North America by CRC Press LLC, Boca Raton, FL. WATTERS, R. J., 2013, Book Review of Hot Deserts: Engineering, Geology and Geomorphology, (edited by M. J. Walker. (Editor), Environmental & Engineering Geoscience, Vol. 19, No. 4, pp. 394–395.
Environmental & Engineering Geoscience, Vol. XXI, No. 3, August 2015, pp. 247–248