EEG Journal Vol. XXVI, No. 4 by Association of Environmental & Engineering Geologists (AEG)

EDITORIAL OFFICE: Environmental & Engineering Geoscience journal, Department of Geology, Kent State University, Kent, OH 44242, U.S.A. phone: 330-672-2968, fax: 330-672-7949, ashakoor@kent.edu. CLAIMS: Claims for damaged or not received issues will be honored for 6 months from date of publication. AEG members should contact AEG, 3053 Nationwide Parkway, Brunswick, OH 44212. Phone: 844-331-7867. GSA members who are not members of AEG should contact the GSA Member Service center. All claims must be submitted in writing. POSTMASTER: Send address changes to AEG, 3053 Nationwide Parkway, Brunswick, OH 44212. Phone: 844-331-7867. Include both old and new addresses, with ZIP code. Canada agreement number PM40063731. Return undeliverable Canadian addresses to Station A P.O. Box 54, Windsor, ON N9A 6J5 Email: returnsil@imexpb.com. DISCLAIMER NOTICE: Authors alone are responsible for views expressed inarticles. Advertisers and their agencies are solely responsible for the content of all advertisements printed and also assume responsibility for any claims arising therefrom against the publisher. AEG and Environmental & Engineering Geoscience reserve the right to reject any advertising copy.

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

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

ASSOCIATE EDITORS Brankman, Charles, Consultant Boston MA Bruckno, Brian Virginia Department of Transportation Clague, John J. Simon Fraser University, Canada De Graff, Jerome V. California State University, Fresno Fryar, Alan University of Kentucky Hauser, Ernest Wright State University Hutchinson, Jean Queens University, Canada Keaton, Jeff AMEC Americas Marinos, Vassillis Aristotle University of Thessaloniki, Greece

SUBSCRIPTIONS:

McBride, John Brigham Young University Mwakanyamale, Kisa Illinois State Geological Survey Santi, Paul Colorado School of Mines Dee, Seth University of Nevada, Reno Shlemon, Roy R.J. Shlemon & Associates, Inc. Stephenson, William U.S. Geological Survey Stock, Greg National Park Service Sukop, Michael Florida International University Ulusay, Resat Hacettepe University, Turkey Watts, Chester F. “Skip,” Radford University West, Terry Purdue University

Environmental & Engineering Geoscience November 2020 VOLUME XXVI, NUMBER 4

SUBMISSION OF MANUSCRIPTS

Member subscriptions: AEG members automatically receive digital access to the journal as part of their AEG membership dues. Members may order print subscriptions for $75 per year. GSA members who are not members of AEG may order for $60 per year on their annual GSA dues statement or by contacting GSA.

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.

Nonmember subscriptions are $310 and may be ordered from the subscription department of either organization. A postage differential of $10 may apply to nonmember subscribers outside the United States, Canada, and Pan America. Contact AEG at 844-331-7867; contact GSA Subscription Services, Geological Society of America, P.O. Box 9140, Boulder, CO 80301.

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.

Single copies are $75.00 each. Requests for single copies should be sent to AEG, 3053 Nationwide Parkway, Brunswick, OH 44212.

To submit a manuscript go to https://www.editorialmanager.com/EEG/ default.aspx. If you have not used the system before, follow the link at the bottom of the page that says New users should register for an account. Choose your own login and password. Further instructions will be available upon logging into the system. Please carefully read the “Instructions for Authors”.

THIS PUBLICATION IS PRINTED ON ACID-FREE PAPER EDITORS

Brian G. Katz Environmental Consultant Tallahassee, FL 32309 eegeditorbkatz@gmail.com

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 Base flow in a perennial reach of the San Pedro River, Cascabel, Arizona, bordered by a riparian forest of willow, cottonwood and invasive tamarisk. Photo courtesy of Christopher Eastoe. See article on page XXX.

Volume XXVI, Number 4, November 2020

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.

Abdul Shakoor Department of Geology Kent State University Kent, OH 44242 330-672-2968 ashakoor@kent.edu

ENVIRONMENTAL & ENGINEERING GEOSCIENCE

Environmental & Engineering Geoscience (ISSN 1078-7275) is published quarterly by the Association of Environmental & Engineering Geologists (AEG) and the Geological Society of America (GSA). Periodicals postage paid at AEG, 3053 Nationwide Parkway, Brunswick, OH 44212 and additional mailing offices.

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

Environmental & Engineering Geoscience Volume 26, Number 4, November 2020 Table of Contents 383

2020 Student Professional Paper - Graduate Level: Integrating Design Parameters for Reseeding and Mulching after Wildfire: An Example from the 416 Fire, Colorado Andrew Graber

393

Case Study: Reconstructing the 2015 Dulcepamba River Flood Disaster Jeanette Newmiller, Wesley Walker, William E. Fleenor, Nicholas Pinter

405

Landslide Mapping Using Multiscale LiDAR Digital Elevation Models Javed Miandad, Margaret M. Darrow, Michael D. Hendricks, Ronald P. Daanen

427

Evaluating the Use of Unmanned Aerial Systems (UAS) for Collecting Discontinuity Orientation Data for Rock Slope Stability Analysis Rachael K. Delaney, Abdul Shakoor, Chester F. Watts

449

UAS-Derived Surficial Deformation around the Epicenter of the 2016 Mw 5.8 Pawnee, Oklahoma, USA, Earthquake Olufeyisayo Ilesanmi, Xue Liang, Francisca E. Oboh-Ikuenobe, J. David Rogers, Mohamed Abdelsalam, Jordan Feight, Emitt C.Witt III

463

Sources of Perennial Water Supporting Critical Ecosystems, San Pedro Valley, Arizona Christopher J. Eastoe

481

A Robust and Efficient Method of Designing Piles for Landslide Stabilization Yang Yu, Xingmin Li, Xiaohua Pan, Qing LuĚ&#x2C6;

493

The Effectiveness of Reactive Materials for Contaminant Removal in the Process of Coal Conversion Jacek Grabowski, Aleksandra Tokarz

507

Effects of Laumontite Hydration/Dehydration on Swelling Deformation and Slake Durability of Altered Granodiorite Junsong Yan, Junhui Shen, Kaizhen Zhang, Jianjun Xu, Weifeng Duan, Richang Yang

517

Notice of Retraction

2020 Student Professional Paper - Graduate Level Integrating Design Parameters for Reseeding and Mulching after Wildﬁre: An Example from the 416 Fire, Colorado ANDREW GRABER* Department of Geology and Geological Engineering, Colorado School of Mines, 1500 Illinois Street, Golden, CO 80401

Key Terms: Erosion Mitigation Design, Reseeding, Mulching, Wildfire, Colorado, 416 Fire, GIS ABSTRACT Wildfires burn vegetation over large areas of land, removing ground cover and frequently increasing potential for erosion-related hazards, such as debris flows, soil loss, and increased sedimentation downstream. Reseeding and mulching are two techniques used to prevent erosion and foster re-establishment of native plant species. However, design guidelines and specifications for reseeding and mulching programs are scattered in the literature, impeding efforts to follow best practices when preparing mitigation plans. This article summarizes guidelines and specifications for both reseeding and mulching and applies them in a sample GIS-based reseeding and mulching design for a basin burned in 2018 by the 416 Fire, Colorado. In addition to relying only on remote data, the method presented here aids operators and management personnel in making quick assessments of mitigation needs and areas suitable for mitigation, allowing for prompt responses to time-sensitive erosion hazards. INTRODUCTION Reseeding and mulching are techniques that are commonly used to prevent erosion, aid in reestablishing ground cover and native plant communities, and prevent invasions of non-native species after a wildfire has burned an area. However, guidelines, specifications, and methodologies for designing mitigation plans for these techniques are scattered or absent in the literature. In terms of hazards, reseeding and mulching are used to prevent soil erosion and debris flows as well as to decrease risk of downstream flooding by decreasing runoff (MacDonald, 1989). These methods are often applied in conjunction with other *Corresponding author email: andrewgraber66@gmail.com

erosion mitigation methods, such as log barriers (i.e., contour felling), debris racks, debris flow check dams, and silt fences (deWolfe et al., 2008; Moench and Fusaro, 2012). Debris flows and flooding can have severe impacts on downstream communities, including injuries, fatalities, and damage to buildings and infrastructure. Soil erosion can impede re-establishment of native plant species and remove topsoil. As the frequency and severity of wildfires, in the western United States and abroad, increase due to climate change, engineering geologists and geological engineers are more and more likely to be called on to design erosion mitigation plans for burned areas because more and more areas will likely require mitigation. Widespread wildfire damage and the resulting time-sensitive hazards require mitigation design methods that can provide design plans quickly to assist management officials in promptly responding to erosion-related hazards in burned areas. This article collects design guidelines and specifications from a variety of sources and presents a method for combining these guidelines and considerations to prepare a GIS-based reseeding and mulching design using only remotely available data. Included in this article are relevant background on both reseeding and mulching, design principles and considerations for applying each technique to erosion control, and a sample GIS-based design method applied to a basin that was burned by the 416 Fire in La Plata County, Colorado, in 2018.

BACKGROUND Existing Design Guidance Based on a review of available literature, the problem of how to design a reseeding and mulching program for a burned area has been partly solved, but uniﬁed design guidelines with speciﬁcations do not appear to be readily available. Current guidance is available in the form of general design guidelines and fact sheets

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 383–392

383

Graber

for post-wildfire erosion mitigation as well as technical publications on conservation planting. While reseeding and mulching are frequently applied methods for wildfire rehabilitation, design of reseeding and mulching plans is not well codified. The U.S. Department of the Interior (USDI) Bureau of Land Management (BLM) gives brief general principles for both reseeding and mulching but provides no design specifications for reseeding and mulching plans (USDI BLM, 2007). Design procedures and rehabilitation goals are laid out in general terms with some specifications by the U.S. Department of Agriculture (USDA) Natural Resources Conservation Service (NRCS) for reseeding (USDA NRCS, 2014) and for mulching (USDA NRCS, 2018). General principles of wildfire rehabilitation planning are also presented, without design specifications, by the USDI (2006a, 2006b). Many of the available resources on reseeding and mulching are fact sheets and short publications intended for use by rural landowners in managing fire recovery on their own properties (e.g., USDA NRCS, 2005; Moench and Fusaro, 2012; Shive and Kocher, 2017; California Native Plant Society [CNPS], 2019; and USDA NRCS, n.d.). Many of these documents also give basic design guidelines; however, these guidelines are presented in fairly general terms. A variety of scientific publications have also addressed the subject of reseeding and/or mulching (e.g., deWolfe et al., 2008; Peterson et al., 2009; Peppin et al., 2010; and Morgan et al., 2014), frequently focusing on evaluating the effectiveness of these erosion mitigation methods. Guidance on “conservation planting” is also available from the USDA NRCS in a series of technical notes published in conjunction with the State of Colorado (USDA NRCS, 2011, 2012a, 2015a, 2015b) These technical notes address conservation planting, which refers to “establishing and maintaining permanent vegetative cover” (USDA NRCS, 2014) on a variety of land types. As a result, these guidelines can apply to planting for erosion control, soil stabilization, and plant regeneration in many situations, not just after wildfire. These technical notes focus on conservation planting in Colorado, though similar documents may also be available for other states. Reseeding Reseeding after wildfire is performed with the goal of reducing erosion, re-establishing native plant communities, reducing weed growth that can occur when soil is disturbed, rebuilding wildlife habitats, and reestablishing forage plants for grazing livestock (Hudson, 2016). Reseeding can be accomplished by hand, using broadcast seeders or seed drills, using aerial

384

spreading, or spraying seed mixed with hydromulch (USDI BLM, 2007). Generally, reseeding on wildlands is done using only native species to avoid introducing non-native or invasive species that might disrupt local ecosystems (USDI BLM, 2007; USDA NRCS, 2014; and CNPS, 2019). Mulching Mulching refers to the application of mulch (hay, straw, wood chips, or synthetic fibers) to a soil surface for the purposes of reducing erosion, fostering plant growth, retaining moisture, slowing runoff, or increasing infiltration time (CNPS, 2019). Mulch can be spread by hand, using mechanical spreaders, as hydromulch (a sprayed mix of mulch and a tackifying agent, sometimes also including seed), or aerially. Mulch can be transported off-site or piled into drifts by the wind. Therefore, it is often secured to the ground using pins, nets, or a tackifying agent or by crimping it into the soil with a shovel. Mulch can be applied with seed to help increase the effectiveness of a reseeding program by protecting the seeds from erosion. DESIGN SCOPE To design a reseeding and mulching plan for a burned area, several questions should be considered during the planning process: 1. Is reseeding and mulching required for this area? 2. What portions of the burned area need to be seeded? Are they accessible? 3. Reseeding a. What mix of plant species should be used? b. What density of seed should be spread? c. What application method should be used? d. How will the timing of the fire affect opportunities for seed germination? 4. Mulching a. Is mulch necessary to stabilize soil or protect seed? b. What kind of mulch should be used? c. What thickness of mulch? d. How much area should be mulched? e. What application method should be used? f. Is an additional stabilization technique required for mulching? 5. What equipment is available for reseeding and mulching? 6. What will the cost of reseeding and mulching be? Answering these questions will help evaluate the feasibility of remediation, estimate required amounts of seed and mulch, choose application methods, and evaluate costs.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 383–392

Integrating Design Parameters for Reseeding and Mulching after Wildﬁre

DESIGN PROCESS Design Process Overview Based on a review of various guideline documents, a general outline of the design process for a reseeding and mulching program was prepared. This outline is not directly taken from any one of the resources cited in this article but rather combines elements that are common themes in the literature reviewed for this report (USDA NRCS, 2005; USDI BLM, 2007; Moench and Fusaro, 2012; USDA NRCS, 2012a, 2014; Shive and Kocher, 2017; USDA NRCS 2018; CNPS, 2019; and USDA NRCS, n.d.). The design process outline is as follows: 1. Evaluation of burned area and identifying areas that are suitable for reseeding/mulching 2. Consideration of erosion control methods 3. Reseeding design 4. Mulching design 5. Implementation, operation, and maintenance 6. Cost estimation Details of Design Process Evaluation of Burned Area and Identifying Areas That Are Suitable for Reseeding/Mulching Several studies agree that natural plant recovery should be prioritized over artificial recovery whenever possible because not all burned areas require a reseeding/mulching program (Hudson, 2016; CNPS, 2019; and USDA NRCS, n.d.). Many fires do not destroy all vegetation or seeds in a given area. And even in case of wildfires that do cause total or near total vegetation mortality, severely burned areas are typically patchy, and other burned areas are often less severely affected. The entire burned area should be evaluated for reseeding and mulching based on the severity of the burn, the accessibility for remediation, and the suitability of the burned slopes for reseeding and mulching. Different resources report conflicting guidelines for maximum suitable slope angles for reseeding and mulching. The USDA NRCS (n.d.) suggests that reseeding is often not effective on slopes steeper than 60º. The CNPS (2019) advises seeding only on slopes less than 35º. Some portions of burned areas can be particularly vulnerable to erosion, such as drainages, areas exposed to wind, and fire lines (breaks in vegetation cut to contain a wildfire), and these areas should be given special consideration for erosion control (Miles et al., 1989). In addition, areas that had poor perennial grass coverage prior to the wildfire can be particularly vulnerable to soil loss and should be prioritized for erosion control (Hudson, 2016).

Consideration of Erosion Control Methods Since this article focuses on reseeding and mulching, other possible erosion control methods are not discussed. However, other soil loss mitigation methods, such as straw wattles, check dams, and contour log felling, are typically available options for erosion control eﬀorts in burned areas and should be considered. In the event that other erosion mitigation methods are chosen, this GIS-based design method could be adapted for use with those other methods.

Reseeding Design Seed mixes for erosion control should be suitable for the local ecosystem that existed before the wildfire (or that may partially survive the fire). Species chosen for seeding should be those native to the area and should be suitable for local soil, elevation, and climatic conditions. In some cases, non-persistent or sterile species (such as cereal grains) can be used to quickly stabilize soil where emergency stabilization is required (Hudson, 2016). For erosion control purposes, a seed mix should consist of at least 50 percent sod-forming plants (USDA NRCS, 2012a). The seeding rate used for a given reseeding project is affected by the size and germination rate of the seeds of individual species and the percent of each species desired in the seed mix. Another important consideration for reseeding designs is the timing of seeding. In many parts of the western United States, wildfires usually occur in summer or fall. The timing of the fire and when the fire is completely extinguished will affect which plants are chosen for reseeding (USDA NRCS, 2012a). Warmseason plants germinate in the early summer. So, a fire that burns from mid- to late summer will often preclude the seeding of warm-season plants unless the seeds can be protected from erosion and animals until the next summer. Cool-season plants are more versatile in germination timing and may be a more suitable choice for reseeding after fires that occur later in the year (Hudson, 2016). The timing of the wildfire may also affect whether additional erosion mitigation (either mulch or other methods) will be necessary to stabilize the soil until seeds can germinate. Reseeding designs should also consider whether weed control or other site preparation, such as removing dead material or breaking up hydrophobic soil layers, will be necessary prior to seeding in order to improve the effectiveness of reseeding a burned area (Hudson, 2016; CNPS, 2019). See USDA NRCS (2012a) for additional project-specific seeding considerations, such as soil texture, CaCO3 tolerance, and rate of spread.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 383–392

385

Graber

Mulch Design Because mulch is intended to help prevent erosion and retain seeds and moisture, mulch coverage is key to the effectiveness of mulching in a burned area (USDA NRCS, 2018). Mulch should be applied to achieve at least 70 percent ground coverage to aid plant establishment and up to 90 percent coverage or more for reducing evaporation and improving water retention in the soil (USDA NRCS, 2018). Miles et al. (1989) suggest applying mulch in the areas that are most vulnerable to erosion, such as slopes adjacent to perennial streams and fire lines. Some consideration should also be given to whether mulch is likely to wash into drainages and create debris in streams. Mulch should be applied evenly to a thickness of about 1–2 in. (2.5–5 cm) or thicker in areas where heavy rains or high winds are likely (USDA NRCS, 2018). Mulch materials include wood chips, grass hay, straw, and synthetic fibers (USDA NRCS, 2018). The choice of mulch material will be affected by what is available at the project site. If chipping equipment is available, downed trees can be chipped to provide mulch for erosion mitigation. Hay and straw are widely available and relatively easy to transport, but only certified weed-free hay and straw should be used as mulch (USDA NRCS, 2005; CNPS, 2019). Synthetic mulch materials may be useful for some applications but also require retrieval following re-establishment of native plants. Anchoring of mulch can be beneficial for avoiding loss of mulch by wind or water transport (USDA NRCS, 2018). Anchoring methods include use of tackifiers, netting, crimping into the soil with a shovel blade, or pinning. Use of these methods in a mulching program will likely depend on budget availability for personnel hours and whether wind and water transport are likely to affect the amount of mulch retained on the soil surface. See USDA NRCS (2018) for additional project-specific mulching considerations, including the effects of the carbon-to-nitrogen ratio of mulch to local ecosystems and guidelines for gaps in mulch around plant stems. Implementation, Operation, and Maintenance Application methods for seeding and mulching are similar and include spreading by hand, broadcast or mechanical spreading, hydromulching (can also include seed), and aerial spreading. In many cases, the same application method is used to apply both seed and mulch to a given burned area, even if the two components are applied in separate steps. Onground methods, such as application of seed and

386

mulch by hand, using small broadcast spreaders, or using all-terrain vehicle–mounted spreaders, are typically cheaper but require more personnel hours and road access to the burned area. These methods have the additional advantage of giving an opportunity for tilling/raking of seeds and anchoring mulch. Hydromulching is more expensive and typically limited only to areas with good road access and where erosion control is particularly critical to protect structures or other assets. Aerial spreading is also a relatively expensive method and does not allow for securing seeds and mulch but does allow for large areas to be seeded and mulched quickly. A reseeding and mulching design should consider which application method is most suitable for the situation at hand and most cost effective. Typically, seeding will occur first, followed by mulching, unless the seed and mulch are applied in one step. For reseeding and mulching, operation and maintenance are relatively small components of the design. Reseeding and mulching plans are typically designed for life spans on the order of 1–3 years and typically do not require any more operation and maintenance than inspections and spot reapplications of seed or mulch. A basic plan that includes several inspections of the remediated area to evaluate plant re-establishment, weed invasion, and soil loss will likely be sufficient for the majority of cases. An additional budget may be required to retrieve mulching materials if synthetic mulches are used for the project. Cost Estimation Estimating the cost of this kind of design is relatively simple because the implementation is relatively simple. Per area costs should be estimated for seed and mulch as well as the total cost of each based on the expected coverage and density of application. Costs of application, such as personnel hours, equipment hours, fuel, and equipment maintenance, should also be considered. Maintenance costs for reseeding and mulching designs are also expected to be low relative to the initial implementation costs because these projects have short implementation and operation phases; that is, once native plants have been re-established (generally within 1–3 years), erosion hazards have likely returned to background levels, and the reseeding and mulching project is typically considered complete. Additional Caveats and Considerations for Reseeding and Mulching Careful consideration should be given to whether a slope requires erosion mitigation at all. In some cases, such as for lightly to moderately burned slopes,

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 383–392

Integrating Design Parameters for Reseeding and Mulching after Wildﬁre

reseeding may be unnecessary because unburned roots and seeds below the surface can help quickly reestablish native vegetation (CNPS, 2019). Soil thickness should be considered, as areas of exposed bedrock will likely not require reseeding, and soil shallower than 2–4 in. (10–15 cm) may not provide a suitable seedbed. An additional caveat for any post-wildfire erosion mitigation effort is that the mitigation plan should be designed so as to minimize the disturbance to the burned soil in order to avoid accelerating erosion by breaking it up or by making it less favorable for vegetation re-establishment. As a result, foot and vehicle traffic in burned areas should be limited to reduce compaction of the soil; leaves, dead trees, and ash should be left to provide a natural erosion barrier; and burned material should be left in place to allow existing seeds an opportunity to germinate (CNPS, 2019). Finally, it is important to bear in mind that reseeding and mulching almost always include a non-zero risk of introducing non-native plants or weeds to a burned area. Even certified weed-free seed mixes or mulches have sometimes been found to contain seeds and stems of non-native species (Shive and Kocher, 2017). In some cases, non-native species that are similar to native species can cross-pollinate and modify the local gene pool, potentially causing significant effects on local ecosystems (CNPS, 2019). SAMPLE DESIGN FOR A BASIN BURNED BY THE 416 FIRE, COLORADO Overview This section presents a sample reseeding and mulching design that was prepared, using the considerations discussed in the previous section, for a basin that was partly burned by the 416 Fire, which occurred in La Plata County near Hermosa, Colorado, beginning on June 1, 2018. The fire burned a total of about 54,000 acres (∼218 km2 ) in the San Juan National Forest, west of Colorado State Highway 550 (Figure 1). Several basins along the Hermosa River were burned, and burn severity ranged from low to severe.

Figure 1. Outline of the area burned by the 416 Fire overlaid on 2016 satellite imagery (imagery credit: ESRI). Burned area outline from USDI USFS (2018). Locator map hillshade from USDI USGS (2005). 1 mi = 1.609 km.

area (measured at breast height). As a result, burned areas that experience severe basal area loss (ࣙ75 percent) most likely also experience total or near total destruction of ground cover. Gray lines in the RAVG map indicate areas where RAVG values were not mapped in the USDI USFS (2018) analysis due to insuﬃcient data. For this example design, percent basal area loss is used to select the most severely burned areas for mitigation because the areas that lost most or all vegetative ground cover are the most likely to need soil stabilization. Out of the several basins burned by the 416 Fire, one basin was selected for this sample mitigation plan. The target basin (Figure 2) was selected because it contained the highest density of basal area loss of any of the basins burned by the 416 Fire. As a result, this basin is expected to be the most likely candidate for

Evaluation of Burned Area and Selection of Reseeding/Mulching Areas Rapid Assessment of Vegetation Condition after Wildﬁre (RAVG) analysis was performed by the U.S. Forest Service (USDI USFS, 2018) using Landsat imagery from before and after the wildﬁre. RAVG analysis produces several indicators of burn severity, including a map of percent basal area loss (Figure 2). Basal area loss indicates loss of cross-sectional tree

Figure 2. Map of percent basal area loss for the 416 Fire, La Plata County, Colorado, overlaid on 2016 satellite imagery (imagery credit: ESRI). The highlighted basin was chosen for the example reseeding/mulching design because it has the highest density of basal area loss ࣙ75 percent. Twelve-digit basin outlines from USDA NRCS (2013). Burn area outline and RAVG data from USDI USFS (2018). 1 mi = 1.609 km.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 383–392

387

Graber

Figure 3. Map of surface slope (º) for the target basin overlaid on hillshade basemap. Slope and hillshade layers calculated from USDI USGS (2017, 2018). 1 mi = 1.609 km.

erosion mitigation. To estimate the total area of the basin that was suitable for mitigation, two criteria were used: severe burn damage (basal area loss ࣙ75 percent) and a slope less than 35º. The CNPS (2019) slope recommendation of <35º was chosen over the cutoff of <60º suggested by USDA NRCS (n.d.) because is often difficult to access and properly maintain seeded and mulched areas when the slope is steeper than 35º. A map of surface slope (Figure 3) was prepared using a 1/3 arc-second (approximately 33 ft [10 m]) resolution DEM of the area. Figure 4A shows all the pixels from the RAVG raster that represent a basal area loss of 75 percent or more. Figure 4B shows all the pixels from the slope raster that represent slopes less than 35º. Figure 4C represents the combination of these two rasters, showing all pixels where both the basal area loss ࣙ75 percent and slope <35º conditions are met. These pixels represent the area suitable for mitigation based on the chosen criteria. It should be noted that there are some gaps in the map of suitable mitigation area (Figure 4C) because of gaps in the basal area loss data. These gaps are caused by regions in which the Landsat data coverage is insufficient to calculate percent basal area loss (shown in gray in Figure 2). A count of the cells suitable for mitigation in Figure 4C yields a total of 13,166 cells. Multiplying this number of cells by the cell size (98 × 98 ft [30 × 30 m]) gives a total remediation area of 2,930 acres (∼11.8 km2 ). Because this is a relatively large area and the areas suitable for remediation are fairly concentrated, aerial seed/mulch application is expected to be the best method for this remediation plan. Consideration of Erosion Control Methods Since this example is focused on reseeding and mulching design, no other erosion control methods are

388

Figure 4. (A) Map of all pixels in the target basin that have a slope <35º. (B) map of all pixels in the target basin that have a basal area loss ࣙ75 percent. (C) Overlay of Figure 4A and B showing all pixels that meet both criteria. These pixels (Figure 4C) mark the areas to be reseeded and mulched. Basin outline from USDA NRCS (2013). Burn area outline from USDI USFS (2018). Basemap satellite imagery credit: ESRI. 1 mi = 1.609 km.

considered here. However, as previously discussed in this article, other erosion methods can be combined with reseeding and mulching if desired. Seed Mix Design To select the seed mix for remediation in this basin, several pieces of information were considered along with the seeding speciﬁcations given in Technical Note 59 (USDA NRCS, 2012a). Species for mitigation in this area should meet the following criteria: 1. They should be suitable for planting in the local Colorado Major Land Resource Area (MLRA). In this case, the MLRA is 48A, Southern Rocky Mountains. 2. They should tolerate precipitation within the expected annual average for this site. In this case, mean annual precipitation for the target basin ranges from 25 to 30 in/yr (63.5–76.2 cm/yr) (Figure 5). 3. They should be suitable for the elevation of the burned basin, ∼7,300–9,700 ft (∼2,230–2,960 m).

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 383–392

Integrating Design Parameters for Reseeding and Mulching after Wildﬁre

The other species were assigned smaller proportions in the mix to reﬂect the fact that forbs and bunch grasses are typically more numerous in forest environments than shrubs. Non-irrigated, broadcast seeding rates from USDA NRCS (2012a) were used to calculate the pounds live seed (PLS) per acre of each species, the total weight of seed required, and the cost per acre of seed based on the expected relative cost of each type of seed. Actual costs per pound of each kind of seed were not available, so relative costs were assigned based on the expected diﬀerences in cost among grasses, forbs, and shrubs. Table 1 summarizes the seed weight calculations. The weight of each species in the seed mix (column 4) was computed by multiplying the percent in the mix (column 2) by the recommended weight (PLS) per acre (column 3). The total amount of seed required per species (column 6) was computed by multiplying the weight in the seed mix (column 4) by the number of acres suitable for mitigation (column 5). The cost per acre for each species (column 8) was computed by multiplying the assumed cost per pound (column 7) by the weight of that species in the mix (column 4). Total weight and cost of seed was computed by summing the individual weights and costs from each species. As a result, the total cost of seed (not including application) is $35.50 per acre ($8,875/km2 ). The USDA Soil Survey was referenced to check for areas of very thin soil where seeding may not be effective for re-establishing vegetation (USDA NRCS, 2019). Based on the soil survey data, the target basin does not include extensive areas of thin soil shallower than 9.8 in. (25 cm).

Figure 5. Mean annual precipitation based on the PRISM (Parameter-Elevation Regressions on Independent Slopes Model) model. PRISM data from USDA NRCS (2012b). The annual precipitation in the target basin varies from 25 to 30 in/yr (63.5– 76.2 cm/yr). Basin outline from USDA NRCS (2013). Fire burn boundary from USDI USFS (2018). Hillshade layer calculated from USDI USGS (2017, 2018). 1 mi = 1.609 km; 1 in. = 2.54 cm.

4. They should be suitable for range conservation planting. 5. Post-ﬁre erosion control seed mixes should include at least 50 percent sod-forming species (CNPS, 2019). 6. They should be cool-season plants. The 416 Fire burned uncontained from June 1 to July 31, 2018, and portions of the area continued to burn until the ﬁre was called out on November 29, 2018. Coolseason plants can be seeded in late summer or as dormant during the winter months (USDA NRCS, 2012a). Based on these criteria, six species were chosen for the seed mix (column 1 in Table 1) from the table in USDA NRCS (2012a). The two sod-forming species chosen were assigned proportions of 25 percent each to meet the guidelines from CNPS (2019).

Mulch Design Because aerial seeding was chosen as the application method and because cool-season plants may not germinate until spring, mulching is likely to be beneﬁcial

Table 1. Seed mix data and calculations. Costs/Pound (column 7) are assumed based on the expected relative cost diﬀerences between grasses, forbs, and shrubs. 1 acre = 0.004 km2 ; 1 lb = 0.454 kg.

Common Species Name (Type) Alpine Bluegrass (sf) Mountain Brome (bg) Russet Buﬀaloberry (s) Silvery Lupine (f) Saskatoon Serviceberry (s) American Vetch (sf) Total weight of seed Total cost/acre

Percent in Mix

Recommended Weight/Acre

Weight of Species in Mix (PLS/Acre)

Total Acres

Total Pounds of Seed Required

Cost/Pound ($)

25 15 10 10 15 25

2 25 2 0.2 0.4 6

0.5 3.75 0.2 0.02 0.06 1.5

2,930 2,930 2,930 2,930 2,930 2,930

1,460 10,970 585 58.5 175 4,390 ∼17,600

5 6 12 12 8 5

Cost/Acre ($) 2.5 22.5 2.4 0.24 0.48 7.5 ∼35.50

PLS = pounds live seed; sf = sod former; bg = bunchgrass; s = shrub; f = forb.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 383–392

389

Graber Table 2. Cost estimate table (application and maintenance per acre costs are estimated). Total cost was calculated as cost/acre × 2,930 acres suitable for mitigation.

Item Seed mix Seed application (includes fuel, helicopter/pilot hours) Mulch Mulch application (includes fuel, helicopter/pilot hours) Maintenance (helicopter or drone surveys), spot repair of seed or mulch by hand Total

Cost per Acre ($)

Total Cost ($)

35.50 20

104,000 58,600

86 20

252,000 58,600

29,300

502,500

in the entire seeded area to protect the seeds from being washed or blown away and to increase soil contact. As a result, aerial mulching using a straw mulch was chosen because straw can be easily spread from the air. A typical mulching thickness of 1–2 in. was chosen, requiring about 43 bales of straw per acre (USDA NRCS, 2005). Assuming that straw costs $2 per bale, the cost of mulch is $86/acre. Implementation, Operation, and Maintenance Seed and mulch should be applied to the area highlighted in Figure 4C using aerial application by helicopter. Because the mitigation area includes the majority of the burned portion of the basin, application of seed and mulch can be done systematically across the basin by flying back and forth, dropping and refilling seed and mulch as needed. Seed and mulch should not be applied directly in or adjacent to Hermosa Creek, which flows through the burned area. It is also expected that some of the areas that were unmappable in the basal area loss assessment (Figure 2) may require seed and mulch as well. Cost Estimate Using the calculated per acre costs for seed and mulch, the total cost of this remediation plan was estimated (Table 2). Application costs were estimated based on the expected cost of helicopter use, pilot pay, and fuel. Like many engineering designs, this remediation plan includes a maintenance component. However, because seeding and mulching are short-term design projects, the maintenance component of this design is relatively small and includes money for aerial surveys of the seeded and mulched area and for additional seed and mulch that could be spread by hand to fill gaps in the reseeded/mulched area.

390

DISCUSSION This sample design has demonstrated two advantages of using a GIS-based design for planning reseeding and mulching. Because it is GIS based, designs can be completed using only remotely sensed data if no onground data are available. The widespread availability of elevation rasters, soil survey data, and precipitation data means that most areas in the United States will have some kind of data available. Second, this design method can be completed within a matter of hours for a given basin once the relevant data are collected. This allows management authorities to quickly review designs and make mitigation decisions and helps speed up responses to time-sensitive post-wildfire hazards. As in many applications of earth science, data collected in the field can greatly improve understanding of an area of interest. Because only remote data were used for this sample design, some information is unknown, such as the extent of the area to avoid mulching and reseeding along Hermosa Creek and whether there are areas of very shallow soil or exposed bedrock at a resolution too fine for the USDA Soil Survey to represent. However, even a first-pass assessment of mitigation needs and the volumes of materials required to meet them can be invaluable for mitigation workers, who are then better equipped to make on-ground adjustments to designs. For the purposes of this sample design, only one seed mix was prepared for the entire suitable area. Full-scale reseeding designs may benefit from the inclusion of different seed mixes for different portions of the seeding area since different soil types may be suitable for different species or mixes of species. This could potentially also reduce costs since some soils may require fewer or cheaper seeds to re-establish perennial vegetation. Some uncertainty arises in the calculated area for reseeding and mulching because of the gaps in the basal area loss data due to unmappable areas in the RAVG analysis (Figure 2). As a result, this sample design may underestimate the amount of area that is suitable for mitigation. To account for the gaps in RAVG evaluations, workers using RAVG data based on Landsat imagery may choose to increase suitable area estimates by a fixed percentage—such as [(area of gaps in RAVG map within area of interest [AOI])/(area of AOI)] × 100 percent—to ensure that sufficient materials are included in the budget. The suggested ratio uses the conservative assumption that all of the land in the gaps in the RAVG data is severely burned. In many cases, this value will likely be around 5 to 15 percent. It should be noted that the evidence that reseeding helps prevent erosion is somewhat ambiguous (Peppin et al., 2010; Morgan et al. 2014; and Shive and

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 383–392

Integrating Design Parameters for Reseeding and Mulching after Wildﬁre

Kocher, 2017). For example, Peppin et al. (2010) surveyed 94 studies on the effect of reseeding on soil erosion and found that the majority of the highest-quality studies (the quality levels of these studies were determined by Peppin et al.) showed no decrease in erosion rates after reseeding, while lower-quality studies had a greater tendency to show a decrease in erosion rates. These results do not prove that reseeding is ineffective for erosion control but do indicate ambiguity in the evidence. Other sources conclude that the best way for wildlands to recover is for native plants to be re-established naturally and that reseeding mitigation either for erosion control or for re-establishing native species should be implemented only under extreme circumstances (USDI, 2006b; USDI BLM, 2007; and CNPS, 2019). Settling the ambiguity as to whether reseeding and mulching are effective is outside the scope of this article. However, the mere fact that reseeding and mulching are commonly applied to erosion control is reason enough for land managers and decision makers to be equipped with as much guidance as possible. This article seeks to assist by collecting and summarizing useful references in one place as well as by providing a design framework for asking and answering the questions that should be addressed in reseeding/mulching design. Many of the questions and principles presented here can be easily adapted to other post-wildfire erosion control methods. CONCLUSION This example demonstrates the usefulness of GISbased design and provides a framework for future designs of reseeding and mulching for erosion mitigation. In cases where more detailed or complicated erosion mitigation strategies are required, GIS-based design can provide a quickly developed and useful initial design for mitigating erosion-related hazards, allowing workers to make necessary adjustments in the field. As wildfires become more and more common, the need for mitigation of post-wildfire hazards is likely to increase, calling on geological engineers and engineering geologists to prepare mitigation designs. GIS provides a useful tool for quickly responding to this need. ACKNOWLEDGMENTS Special thanks go to Paul Santi, Claire Graber, and several anonymous reviewers at AEG for their reviews of the manuscript and helpful comments. REFERENCES California Native Plant Society (CNPS), 2019, Fire Recovery Guide: Electronic document, available from: https://www. cnps.org/give/priority-initiatives/fire-recovery

deWolfe, V. G.; Santi, P. M.; Ey, J.; and Gartner, J. E., 2008, Effective mitigation of debris flows at Lemon Dam, La Plata County, Colorado: Geomorphology, Vol. 96, pp. 366–377. Hudson, T., 2016, Seeding after Fire: Bulletin FS 206E, Washington State University Extension, Pullman, WA, 8 p. MacDonald, L. H., 1989, Rehabilitation and recovery following wildfires: A synthesis. In Berg, N. H. (Editor), Proceedings of the Symposium on Fire and Watershed Management: General Technical Report PSW-109, US Department of Agriculture, Forest Service, Pacific Southwest Forest and Range Experiment Station, Berkeley, CA, pp. 141–144. Miles, S. R.; Haskins, D. M.; and Ranken, D. W., 1989, Emergency burn rehabilitation: Cost, risk, effectiveness. In Berg, N. H. (Editor), Proceedings of the Symposium on Fire and Watershed Management: General Technical Report PSW-109, US Department of Agriculture, Forest Service, Pacific Southwest Forest and Range Experiment Station, Berkeley, CA, pp. 97–102. Moench, R. and Fusaro, J., 2012, Soil Erosion Control after Wildfire: Factsheet No. 6.308, Colorado State University Extension, Fort Collins, CO, Electronic document, available from: https://extension.colostate.edu/docs/pubs/natres/06308.pdf Morgan, P.; Moy, M.; Droske, C. A.; Lentile, L. B.; Lewis, S. A.; Robichaud, P. R.; and Hudak, A. T., 2014, Vegetation response after post-fire mulching and native grass seeding: Fire Ecology, Vol. 10, pp. 49–62. Peppin, D.; Fulé, P. Z.; Sieg, C. H.; Beyers, J. L.; and Hunter, M. E., 2010, Post-wildfire seeding in forests of the western United States: An evidence-based review: Forest Ecology Management, Vol. 260, pp. 573–586. Peterson, D. W.; Dodson, E. K.; and Harrod, R. J., 2009, Fertilization and seeding effects on vegetative cover after wildfire in north-central Washington State: Forest Science, Vol. 55, No. 6, pp. 494–502. Shive, K. and Kocher, S., 2017, Recovering from Wildfire: A Guide for California’s Forest Landowners: ANR Publication 8386, University of California, Department of Agriculture and Natural Resources, Berkeley, CA, Electronic document, available from: https://anrcatalog.ucanr.edu/pdf/8386.pdf U.S. Department of Agriculture, Natural Resources Conservation Service (USDA NRCS), 2005, Wildfire Rehabilitation Assistance: Electronic document, available from: https:// www.nrcs.usda.gov/Internet/FSE_DOCUMENTS/nrcs144p2_ 050242.pdf U.S. Department of Agriculture, Natural Resources Conservation Service (USDA NRCS), 2011, Seedbed Preparation for Conservation Plantings: Plant Materials Technical Note No. 74, 4 p. U.S. Department of Agriculture, Natural Resources Conservation Service (USDA NRCS), 2012a, Plant Suitability and Seeding Rates for Conservation Plantings in Colorado: Plant Materials Technical Note No. 59 (Revised), 5 p. U.S. Department of Agriculture, Natural Resources Conservation Service (USDA NRCS), 2012b, Precipitation Rasters for Each Month plus yearly for 1981–2010: Raster Spatial Dataset: Electronic document, available from: https://datagateway.nrcs.usda.gov/GDGHome.aspx U.S. Department of Agriculture, Natural Resources Conservation Service (USDA NRCS), 2013, 12 Digit Watershed Boundary Dataset in HUC8: Vector Spatial Dataset: Electronic document, available from: https://datagateway. nrcs.usda.gov/GDGHome.aspx

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 383–392

391

Graber U.S. Department of Agriculture, Natural Resources Conservation Service (USDA NRCS), 2014, Conservation Practice Standard: Conservation Cover: Code 327, 3 p. U.S. Department of Agriculture, Natural Resources Conservation Service (USDA NRCS), 2015a, Conservation Cover Assessment: Plant Materials Technical Note No. 78, 3 p. U.S. Department of Agriculture, Natural Resources Conservation Service (USDA NRCS), 2015b, Conservation Cover Establishment: Plant Materials Technical Note No. 77, 4 p. U.S. Department of Agriculture, Natural Resources Conservation Service (USDA NRCS), 2018, Conservation Practice Standard: Mulching: Code 484, 4 p. U.S. Department of Agriculture, Natural Resources Conservation Service (USDA NRCS), 2019, Soil Survey Geographic (SSURGO) database for Animas-Dolores Area, Colorado, Parts of Archuleta, Dolores, Hinsdale, La Plata, Montezuma, San Juan, and San Miguel Counties: Vector Spatial Dataset: Electronic document, available from: https://websoilsurvey. sc.egov.usda.gov U.S. Department of Agriculture, Natural Resources Conservation Service (USDA NRCS), n.d., After the Fire: Seeding: Electronic document, available from: https://www.nrcs.usda. gov/wps/portal/nrcs/detail/wa/home/?cid = STELPRDB 1259629 U.S. Department of the Interior, 2006a, Interagency Burned Area Emergency Response Guidebook, Version 4.0: Electronic document, available from: https://www.nps.gov/ archeology/npsGuide/ﬁre/docs/18%20Interagency%20BAER %20Handbook.pdf

392

U.S. Department of the Interior, 2006b, Interagency Burned Area Rehabilitation Guidebook, Version 1.3: Electronic document, available from: https://www.fws.gov/ﬁre/ postwildﬁre/Files/Interagency%20BAR%20Guidebook.pdf U.S. Department of the Interior, Bureau of Land Management (USDI BLM), 2007, Burned Area Emergency Stabilization and Rehabilitation Handbook (Public): BLM Handbook H-1742-1, 87 p. U.S. Department of the Interior, U.S. Forest Service (USDI USFS), 2018, RAVG Data Bundle for the 416 Fire Occurring on the San Juan National Forest—2018 (CO3746110780820180601): Raster Spatial Dataset: Electronic document, available from: https://fsapps.nwcg.gov/ ravg/data-access U.S. Department of the Interior, U.S. Geological Survey (USDI USGS), 2005, USGS Small-Scale Dataset—Color Conterminous United States Shaded Relief—200-Meter Resolution 200512 GeoTIFF: Raster Spatial Dataset: Electronic document, available from: https://www.sciencebase.gov/ catalog/item/4f4e47a3e4b07f02db496601 U.S. Department of the Interior, U.S. Geological Survey (USDI USGS), 2017, USGS NED 1/3 Arc-Second n38w109 1 × 1 Degree ArcGrid 2017: Raster Spatial Dataset: Electronic document, available from: https://www.sciencebase.gov/ catalog/item/58b7bf9ee4b01ccd5500b51e U.S. Department of the Interior, U.S. Geological Survey (USDI USGS), 2018, USGS NED 1/3 Arc-Second n38w108 1 × 1 Degree ArcGrid 2018: Raster Spatial Dataset: Electronic document, available from: https://www.sciencebase.gov/ catalog/item/58b7bf9de4b01ccd5500b51a

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 383–392

Case Study: Reconstructing the 2015 Dulcepamba River Flood Disaster JEANETTE NEWMILLER WESLEY WALKER WILLIAM E. FLEENOR* Department of Civil and Environmental Engineering, One Shields Avenue, University of California, Davis, Davis, CA 95616

NICHOLAS PINTER Department of Earth and Planetary Sciences, One Shields Avenue, University of California, Davis, Davis, CA 95616

Key Terms: Model, Hydraulic, Hydrologic, Flood, Hydropower ABSTRACT In March 2015, the village of San Pablo de Amalí on the Dulcepamba River in Ecuador was hit by a flood that killed three residents, destroyed five homes, and eroded several hectares of farmland. Residents asserted that the recent construction of a run-of-the-river hydroelectric facility built in the river channel directed flood flows toward the village, causing the associated damage and fatalities. We conducted a forensic hydrologic and hydraulic analysis of the catchment to assess potential causal mechanisms affecting flooding, including the construction of the hydroelectric facility. Hydrologic analysis demonstrated that the river flows produced by the March 2015 storm were equivalent to a 6-year return interval event, with a discharge of 58.6 cms, not the much more extreme 33-year return interval, 400-cms event that had been suggested in a report produced by the hydroelectric company. Hydraulic modeling determined an ∼2-m elevation surcharge of water attributable to the hydroelectric facility, suggesting that damage to the village would not have occurred without the obstruction created by debris blockage of the hydroelectric plant intake. Hydrologic modeling also quantified monthly totals of water availability in the Dulcepamba watershed, including average dry-season flow volumes. When compared to flow volumes allocated to the hydroelectric operator, the modeling indicated that the seasonal water availability in the Dulcepamba watershed is not sufficient to collectively meet the minimum in-stream environmental flow requirements, the agriculture demands from local subsistence irrigators, and the flow volumes allocated to the hydroelectric operator.

*Corresponding author email: weﬂeenor@ucdavis.edu

INTRODUCTION The Dulcepamba watershed is located in central Ecuador in Bolívar Province. The catchment covers nearly 500 km2 and extends from the ∼3,200-melevation highlands of the Andean plateau down to the coastal foothills at ∼100 m above sea level. The northern and eastern areas of the watershed are particularly rugged and largely forested or covered with shrub brush. Land uses in the gentler sloping downstream portions of the watershed are dominated by agriculture, including a variety of crops. The agrarian town of San Pablo de Amalí is located along the left bank of the Dulcepamba River at ∼420 m elevation. The location is shown in Figure 1. In 2012, Hidrotambo, S.A., began construction of a water intake facility for a low-head, run-of-the-river hydroelectric plant without pondage in the channel of the Dulcepamba River. The intake of the hydroelectric structure was located in the original riverbed adjacent to San Pablo de Amalí, and flow not passing through the facility was directed in a channel toward the town. In addition, as part of the operating license for the hydroelectric facility, Hidrotambo had obtained a right to seasonal diversions from the Dulcepamba River in the following volumes: up to 6.50 cubic meters per second (cms) from December 15 to June 14 and up to 1.96 cms from June 15 to December 14. A storm and resulting flood on March 19–20, 2015, sent water into the town of San Pablo de Amalí, causing severe bank erosion and undermining of the town, which caused several deaths, loss of property, and damage to the only road access to the town. Eyewitness accounts and photographic evidence indicate that the flood mobilized large volumes of coarse sediment and other debris along the Dulcepamba channel and floodplain. Witness reports and photographs (Figure 2) taken after the flood suggested complete or near-complete blockage of the hydroelectric intake and

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 393–404

393

Newmiller, Walker, Fleenor, and Pinter

Figure 1. Study location map, San Pablo de Amalí identiﬁed by the yellow star. Map inset shows the boundaries of the hydraulic model.

394

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 393–404

2015 Dulcepamba River Flood Disaster

three purposes: (1) to determine the recurrence interval of the 2015 ﬂood and other high-water events; (2) as input to the hydraulic model of the Dulcepamba River to create a more detailed analysis of the hydraulic and geomorphic processes that occurred prior to, during, and after the March 2015 event; and (3) to assess water availability for Dulcepamba basin water rights. Basin Data and Model Parameters

Figure 2. Debris blocking the intake structure in the original river channel.

adjacent channel by coarse sediment, woody material, and other debris. A report produced by the power plant owner, Hidrotambo S.A. (Soria, 2015), suggested that the March 2015 flood had a peak flow of 400 cms and a return interval of 33 years. According to Soria (2015), flood flows through San Pablo de Amalí began on March 20 and lasted 4 days. In contrast, eyewitness accounts and the available precipitation and discharge measurements indicate that the flood wave began to pass through San Pablo de Amalí on March 19, with peak flows lasting only 1 to 2 days. Soria (2015) ascribed the 2015 flood damages in San Pablo de Amalí to extreme precipitation in the watershed. METHODS Hydrologic and hydraulic models were developed to simulate conditions in the Dulcepamba watershed and in the Dulcepamba River through San Pablo de Amalí. Precipitation and stream flow data were obtained from local sources and from the Instituto Nacional de Meteorología e Hidrología (INAMHI) in Ecuador. Input data, simulation durations, and sources for both the hydrologic and the hydraulic models are described in the sections below. Available data were used to simulate historical hydrologic conditions in the basin as well as hydrologic and hydraulic conditions during select storm events, including the March 2015 flood. Hydrologic Model A hydrologic model was developed for the historical record in the Dulcepamba River basin. The Hydrologic Modeling System software, HEC-HMS (Hydrologic Engineering Center [HEC], 2016d), was used to analyze the period of record (through continuous simulation), several historic flood events, and the March 2015 flood. Model results were generated for

Shapefiles of the watershed boundaries, precipitation isohyets, stream paths, land use, soil type, and vegetative cover were used to delineate sub-basins within the watershed and to estimate initial hydrologic infiltration, storage, and loss parameters (Gobierno Autonomo Decentralizado Municipal Chillanes [GADMC], 2012a, 2012b). Considering differences in land use, vegetation, and slope, the Dulcepamba model was split into three sub-basins: Upper, Middle, and Lower, shown in Figure 3. Cross sections were obtained for the following reaches of the river channel and floodplain, including San Pablo de Amalí, Amalí-Salunguiri, Chima Guapo, Chima Pesqueria, Congon Tendal, Limon, and Sicoto. The cross-section data were used in the hydrologic model as input geometry for hydraulic routing. Within the model, the Simple Canopy, Simple Surface, Deficit and Constant Loss, Clark Unit Hydrograph, Recession Baseflow, and Monthly Evapotranspiration methods were used to represent hydrologic processes in the basin. Annual precipitation in the study area averages around 1,450 mm per year but can range up to 2,000 mm per year in the higher elevations of the watershed (GADMC, 2012a). Most of the annual precipitation occurs from December to April, as shown in the average monthly rainfall histogram in Figure 4 (monthly data from the Chillanes precipitation gage). Precipitation in the basin ultimately accumulates in the Dulcepamba River, which flows southwesterly toward the coastal plains. Rainfall data were available for several locations within and near the watershed. Precipitation gages used in the study are listed in Table 1 and shown in Figure 3. Data for Chillanes (Code# M0130), San Pablo de Atenas (Code# M0131), and Table 1. Precipitation gages used in the Dulcepamba HMS model. Gage Name

Latitude

Longitude

Frequency

Chillanes San Pablo de Atenas Sanabanan San Pablo de Amalí San Vicente San Jose Del Tambo

−1.981 −1.822 −1.974 −1.951 −1.919 −1.943

−79.068 −79.069 −79.102 −79.167 −79.138 −79.236

Daily Daily Daily 5 min Daily Daily

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 393–404

395

Newmiller, Walker, Fleenor, and Pinter

Figure 3. Locations of precipitation and discharge gages in the Dulcepamba watershed; precipitation data.

396

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 393â&#x20AC;&#x201C;404

2015 Dulcepamba River Flood Disaster

period of study, these were intermittent, instantaneous measurements, with significant and variable time periods between measurements. These data provided helpful snapshots, but they were not used for calibration. Two exceptions were measurements from the Congon Tendal tributary and for San Pablo de Amalí. Data of representative dry-season flows and storm events were available for Congon Tendal. More consistent flow discharge measurements were available for San Pablo de Amalí from March 2014 to February 2017. Hydraulic Model Figure 4. Monthly average precipitation histogram for the Chillanes precipitation gage; discharge data.

San Jose Del Tambo (Code# M0384) were available from INAMHI. Data from Chillanes and San Pablo de Atenas were particularly useful for the historical study, as records go back to January 1, 1963, and August 1, 1968, respectively. Data for the remaining sites were recorded and provided by a local nongovernmental organization (Instituto de Estudios Ecologistas del Tercer Mundo [IEETM]). The sites maintained by IEETM also recorded evapotranspiration data, which were useful for verification of evapotranspiration in the model and as input for the more recent historical simulation events. Given the available gages and data, the inverse distance method was most appropriate. Homogeneous and continuous historical river discharge data were available for only one gage site in the basin at Sicoto. The gage site (Code# H0334) and associated data are maintained by INAMHI. Daily average flow rates (cms) were available beginning January 1, 1968. Although several days are missing from the record, sufficient data existed for calibration of the hydrologic model and frequency analysis. The Sicoto gage site is located nearly 12 km upstream of and about 1,500 m higher in elevation than San Pablo de Amalí. Daily flow data were also available for a gage site downstream of San Pablo de Amalí at San Jose del Tambo, also maintained by INAMHI (Code# H0395). Although measurements are available only from 1968 to 1973, these data were still useful for calibrating the historical period of record as well as providing an expectation of typical flow magnitudes downstream of San Pablo de Amalí. Both INAMHI gage locations are shown in Figure 3. Instantaneous discharges were measured at several additional sites within the watershed, located primarily at Chima Guapo, Chima Pesqueria, Chima Villa Mora, Congon Tendal, Amalí-Salnuguiri, and San Pablo de Amalí (locations shown in Figure 3; data measured and provided by IEETM). For most of the

A two-dimensional (2D) hydraulic model was developed using the River Analysis System software developed by the U.S. Army Corps of Engineers, HEC-RAS 5.0.3, to inform hydraulic conditions and geomorphic processes near San Pablo de Amalí during the March 2015 flood (HEC, 2016a, 2016b, 2016c). The model used the output from the hydrologic model and hydrologic and topographic measurements from the study reach. The hydraulic model simulations were then used to determine flow depths, velocities, and shear stresses for the March 2015 flood. To assess the impact of the hydroelectric facility, the hydraulic simulation of the 2015 event was run as if the full flow volume were passed through the reach in the absence of blockage, and these water surface profiles were compared with measured elevations during the 2015 flood. The terrain model was developed from topographic survey data of the study reach channel and floodplain (measured by Facultad de Ingeniería, Pontificia Universidad Católica del Ecuador, and provided by IEETM). A comparison of the river location for pre- and post-construction as well as post-flood conditions is presented in Figure 5. The model computational grid spacing was typically 5 m and reduced to as little as 1 m in areas with the steepest slopes. Hydrologic data for the upstream boundary of the hydraulic model were entered for model calibration, flood simulation, and flood capacity testing. The computational time step was 0.5 seconds, and roughness (Manning’s n) was initially set to a uniform value of 0.1 based on recommended values for modeling flows on a steep slope (Jarrett, 1984). For model calibration, four measured flow rates were used. Manning’s n calibration focused on average velocity and depth at channel cross sections. Establishing matching cross-section profiles between the measured data and modeled results was not possible due the several discrepancies in spatial data, including that (1) the topographic survey did not include bathymetric measurements for the riverbed, (2) the crosssection measurements provided by the Dulcepamba Project Team indicated that the profile changes over

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 393–404

397

Newmiller, Walker, Fleenor, and Pinter

Figure 5. San Pablo de Amalí on the Dulcepamba River showing riverbed, hydroelectric power plant, and areas of housing damage (a) before construction, (b) after construction, and (c) after the ﬂood event of March 2015.

time, and (3) the location of the velocity and depth measurements was acquired by IEETM using a GPS unit of unknown precision. For this reason, calibration was deemed sufficient for the purpose of this study when all four calibration flows produced depth and velocity results in the range of observed measurements at the approximate location they were taken. Additionally, since the reach is relatively small, it was assumed that secondary inflow and outflow would be negligible to the magnitude of the flood event, and therefore flow rates at both boundaries should be consistent with each other. The mean velocity and depths in the reach for each of the calibration flows are presented in Table 2. As a final verification of the capacity of the natural system, simulations were run using synthetic flows,

incrementing inﬂow rates from 20 to 500 cms over a 7-hour time period.

RESULTS Hydrologic Model The HEC-HMS model was calibrated to historical rainfall-runoﬀ events and to the period-of-record data. Model parameters were adjusted (within known acceptable limits) until the model best reproduced the observed discharges at the available gage locations. Two groups of calibration studies were completed: a period-of-record study and individual events.

Table 2. Measured and modeled flows, depths, and velocities used for pre-flood condition calibration. Measured Date January 7, 2014 January 31, 2014 March 25, 2014 October 16, 2014

398

Calculated Mean

Flow Rate (cms)

Depth (m)

Velocity (m/s)

Depth (m)

Velocity (m/s)

8.39 38.89 12.38 2.77

0.45 1.18 0.66 0.53

0.67 1.15 0.79 0.32

0.66 1.55 0.82 0.34

0.93 1.75 1.11 0.55

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 393–404

2015 Dulcepamba River Flood Disaster Table 3. Final HMS model parameters. Sub-Basin Parameter

Upper

Middle

Lower

Area (km2 ) Maximum canopy storage (mm) Maximum surface storage (mm) Maximum deficit (mm) Constant loss rate (mm/hr) % impervious Time of concentration (hr) Storage coefficient (hr) Baseflow recession constant Baseflow ratio

221.5 3 3 18 3.3 4 5 22 0.95 0.5

204.1 3 4 20 2.6 6 7 25 0.95 0.45

48.2 — — 18 2.7 6 6 15 — —

A continuous period-of-record simulation was completed with two objectives. The first was to determine baseline model values that most accurately reproduced observed flow data at the Sicoto and San Jose del Tambo flow measurement sites. A summary of the finalized model parameter values is given in Table 3. The second objective was to perform a flood-frequency analysis at San Pablo de Amalí through development of a historical hydrograph. Accordingly, a simulation was completed for the period of record from January 1, 1969, to December 31, 2016. Results were produced at a daily time step to compare with the available historical data. Several historical events were modeled to further refine the calibration parameters and assess model accuracy. Simulations were completed for runoff events in April 1970, March 1989, January 1993, February 2008, and February 2017. These events were chosen to demonstrate model response to a range of hydrologic and meteorological conditions. Model parameters related to base-flow levels and soil moisture conditions were slightly adjusted to account for antecedent basin conditions for each simulation. All simulations were run at a daily time step to provide an appropriate comparison to the available observed flow data. Error in peak flows and flow volumes at Sicoto and San Jose del

Tambo between modeled and observed flow values for each calibration event is given in Table 4. The model simulated the peak flows with high accuracy. In some cases, the timing of peak flow did not match; these mismatches are attributable largely to errors associated with the daily resolution of the data. The modeled flow volume for each event was also in acceptable agreement with observed values, except for the January 1993 event (a drought-year event). Although the 1993 percent error was high relative to the other simulations, the overall volume of error was still low. Results of the calibration events demonstrate that the model accurately simulated both high- and low-flow events. Subsequent to calibration, the model was run for the period March 1–31, 2015, which includes the March 2015 flood event. Model results were again produced at a daily time step to provide appropriate comparison to the daily gage data at Sicoto. The full month was modeled to assess the following: antecedent conditions in the basin, response to pre-storm conditions, response to the peak event, and flow recession after the storm. The model produced highly accurate results of the event for the Sicoto gage location, as shown in Figure 6. Peak flow error was only 1.56%, total flow volume error was 0.86%, and the model accurately simulated the timing of the peak flow and flow recession. A hydrograph of the March 2015 event was also produced for the river at San Pablo de Amalí; however, gage data were not available during this event. Modeled peak flow during the March 2015 event was 58.6 cms (shown in Figure 7). After completing calibration based on the historical record, the maximum daily flow for each year of the analysis (1969–2016) was calculated. A flood frequency analysis was then completed on the peak flows using a log-Pearson type III distribution shown in Table 5. For the March 2015 event, the frequency analysis indicates a return interval of approximately 6 years for the flow at San Pablo de Amalí, confirming eyewitness and other local reports that the March 2015 event was not an extraordinary storm.

Table 4. Calibration event error analysis of hydrologic model. Discharge Location and Event Sicoto April 1970 March 1989 January 1993 February 2008 March 2015 San Jose del Tambo April 1970

Computed Peak (cms)

Observed Peak (cms)

Percent Error

Computed Volume (1,000 m3 )

Observed Volume (1,000 m3 )

Percent Error

7.8 13.3 3.2 27.8 19.5

7.4 12.5 3.0 25.8 19.2

5.41 6.40 6.67 7.75 1.56

13,195 23,532 1,870 23,445 14,217

13,501 23,827 1,559 25,807 14,340

2.27 1.24 19.95 9.15 0.86

32.7

31.7

3.15

61,933

63,420

2.34

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 393–404

399

Newmiller, Walker, Fleenor, and Pinter

Figure 6. Modeled and observed ﬂows at Sicoto for the March 2015 ﬂood event.

The Dulcepamba hydrologic model was also used to estimate monthly flow volumes and daily flows available for water diversion. Cyclical analysis was completed for the modeled discharge at San Pablo de Amalí to determine the average, 90%, and 10% exceedance levels for daily and monthly flows over the period 1969–2016 (shown in Table 6). Similar analysis could be extended to other locations in the Dulcepamba watershed. Hydraulic Model Hydraulic model simulations were run using flows for the March 2015 flood from the hydrologic analyses in this study. As-built plans were not available

for the Hidrotambo intake facility. A conservative approach for the 2015 hydraulic simulation was run as if the full flow volume was passed through the reach without blockage. The result of this modeling was a water surface elevation approximately 2 m lower than the water surface elevation measured during the 2015 flood. The 2-m difference demonstrates the approximate hydraulic impact of blockage in and near the Hidrotambo intake facility. Modeling also provided a 2D distribution of velocities that were used to calculate the size of boulder where motion would occur (Julien 2002). Flow velocities and depths demonstrate that the peak flow of the March 2015 event could mobilize submerged boulders up to 1 m in diameter.

Table 5. Sicoto and San Pablo de Amalí flow return intervals. Percent Chance Exceedance Event (ﬂow cms) Location Sicotoa Amalíb a b

50% (2-yr)

20% (5-yr)

10% (10-yr)

4% (25-yr)

2% (50-yr)

1% (100-yr)

0.5% (200-yr)

9.63 26.19

14.93 48.91

18.06 71.16

21.56 110.50

23.84 150.10

25.87 200.62

27.69 264.48

Observed data. Modeled data.

400

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 393–404

2015 Dulcepamba River Flood Disaster

Figure 7. Modeled ﬂow at San Pablo de Amalí for the March 2015 ﬂood event.

DISCUSSION AND CONCLUSIONS Hydrologic modeling of the Dulcepamba watershed accurately simulated observed hydrologic conditions and flood events. Results of the period-of-record simulation and simulations of specific flood events demonstrate that the March 2015 flood event was not an extreme flood, nor was it even a particularly rare event, confirming eyewitness accounts (R. Conrad, pers. comm., 2015, 2017). The model results and statistical flood analysis contradict the values suggested by the Hidrotambo report (Soria, 2015). The modeled peak flow in March 2015 was 58.6 cms (compared to the Hidrotambo report of 400 cms), and the peak flow lasted only 1 day (compared to the 4 days suggested by

the Hidrotambo report). Furthermore, Soria (2015) indicated a return interval of 33 years for the 2015 flood event, while the hydrologic model demonstrates a return interval of just 6 years. The hydrologic model ultimately demonstrated that flooding and damage in San Pablo de Amalí cannot be ascribed simply to rare and extreme meteorological or hydrologic conditions in the watershed. Additionally, the modeled flows at San Pablo de Amalí over the period of January 2010–February 2017 were compared to the hydroelectric facility seasonal water allocations (up to 6.50 cms for December 15– June 14 and up to 1.96 cms for June 15–December 14) and the minimum in-stream environmental flow requirement, 1.059 cms, which is set in statute by the

Table 6. Average, 10%, and 90% exceedance flows at San Pablo de Amalí. Monthly Flow Averages at Amalí (cms)

Monthly average 90% monthly Average exceedance ﬂow 10% monthly Average exceedance ﬂow

Jan.

Feb.

Mar.

Apr.

May

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

7.51 4.39 10.14

10.6 5.12 20.2

12.13 5.11 19.95

12.85 5.85 21.33

8.69 4.6 12.81

4.91 2.99 8.09

3.21 2.43 4.45

2.71 2.31 3.21

2.82 2.27 3.57

3.08 2.42 3.87

3.42 2.43 4.24

4.42 2.58 5.48

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 393–404

401

Newmiller, Walker, Fleenor, and Pinter

Figure 8. Comparison of available and allocated ﬂow rates.

Ecuadorian government. The 8-year period was chosen to demonstrate water availability given current land use and hydrologic conditions in the context of the environmental flow requirement and the hydroelectric facility water rights. Model results indicate that the daily average flow at San Pablo de Amalí is below the sum of the environmental flow requirement and the Hidrotambo water allocation in 69% of days during the 2010–2017 period. These results suggest that the hydroelectric water rights would allow the facility to claim all of the flow in the river during part of the year, leaving little or no flow available for downstream

402

irrigators, particularly during the dry summer season, when irrigation demands are highest. A comparison of available and allocated flow rates is shown in Figure 8. Despite recognition of this shortfall, Hidrotambo reportedly received an updated Dulcepamba water allocation in September of 2017 of 6.50 cms year-round (R. Conrad, pers. comm., 2019). Hydraulic modeling of the Dulcepamba River near San Pablo de Amalí was successfully calibrated to available measurements. Because spatial and flow data were not available for the intake facility, the modeling here assumed that the full flow volume was passed

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 393–404

2015 Dulcepamba River Flood Disaster

Figure 9. Flow capacity test with pre-ﬂood geometry and depth for ﬂow rates up to 500 cms. Model domain shown in Figure 1.

through the reach in the absence of the Hidrotambo hydroelectric construction. There was initial concern due to the 1:8 slope of the modeled reach, which is greater than the maximum slope of 1:10 recommended by the HEC-RAS. However, we determined through testing that (1) using relatively small cells in the steepest areas and (2) utilizing relatively high values of

Manning’s n, as suggested by Jarret (1984), the model would provide good approximations. For verification of these two assumptions, an additional study was performed in which a flow 3D model was produced for comparison. The comparison demonstrated that the HEC-RAS 2D model performed robustly when compared to the flow 3D model (Newmiller, 2017).

Figure 10. Flow capacity test with pre-ﬂood geometry and velocity distributions for ﬂow rates up to 500 cms. Model domain shown in Figure 1.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 393–404

403

Newmiller, Walker, Fleenor, and Pinter

In the report released by Hidrotambo (Soria, 2015), the March 2015 flood peak was estimated at 400 cms. This value is significantly higher than the 58.6 cms calculated by the hydrologic modeling described above. In order to assess the 400-cms flow suggested in the above report, a synthetic hydrograph with flows increasing to 500 cms was simulated with the hydraulic model. The depth of flow results of this modeling are shown in Figure 9, and the flow velocity is shown in Figure 10. The 2015 flood, as calculated by the hydrologic model and as reported by Soria, was modeled through the modified channel without the added capacity of the intake facility. Even for the extreme flow rate reported by Soria (2015), an unobstructed flow would not have inundated the village of San Pablo de Amalí without additional obstruction of the channel at the hydroelectric intake. During the March 2015 flood, no significant obstruction was observed at any other location along the Dulcepamba River as occurred at the power plant intake. The obstruction directed the bulk of the flow toward San Pablo de Amalí and caused the 2-m increase in water elevations. ACKNOWLEDGMENTS Funding for the project was provided by the Center for Watershed Sciences at the University of California, Davis, as part of its international outreach efforts.

404

REFERENCES Gobierno Autonomo Decentralizado Municipal Chillanes, 2012a, Plan de Desarrollo y Ordenamiento Territorial, Tipos de Suelo (Taxonomia): Gobierno Autonomo Decentralizado Municipal Chillanes, Chillanes, Ecuador, November. Gobierno Autonomo Decentralizado Municipal Chillanes, 2012b, Plan de Desarrollo y Ordenamiento Territorial, Aptitud del Suelo: Gobierno Autonomo Decentralizado Municipal Chillanes, Chillanes, Ecuador, November. Hydrologic Engineering Center, 2016a, CPD-68A, Hydraulic Reference Manual, Version 5.0: U.S. Army Corps of Engineers, Davis, CA, February. Hydrologic Engineering Center, 2016b, CPD-69, HEC-RAS River Analysis System User’s Manual, Version 5.0: U.S. Army Corps of Engineers, Davis, CA, February. Hydrologic Engineering Center, 2016c, CPD-70, River Analysis System Application Guide, Version 5.0: U.S. Army Corps of Engineers, Davis, CA, February. Hydrologic Engineering Center, 2016d, CPD-74A, Hydrologic Modeling System HEC-HMS User’s Manual, Version 4.2: U.S. Army Corps of Engineers, Davis, CA, August. Jarrett, R. D., 1984, Hydraulics of high-gradient streams: Journal of Hydraulic Engineering, Vol. 110, No. 11, pp. 1519–1539. Julien, P. Y., 2002, River Mechanics. Cambridge University Press, Cambridge, U.K. Newmiller, J., 2017, River Hydraulics on a Steep Slope: Can a 2D Model Push the Limits of the Hydrostatic Assumption?:Master’s Thesis, University of California, Davis. https:// pqdtopen.proquest.com/doc/2026279256.html?FMT = ABS Soria, D. R., 2015, Engineer: HTSA 0078-2015. Hidrotambo S.A., Bolivar, Ecuador. September 30.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 393–404

Landslide Mapping Using Multiscale LiDAR Digital Elevation Models JAVED MIANDAD* NewFields Mining Design and Technical Services, 9400 Station Street, Suite 300, Lone Tree, CO 80124

MARGARET M. DARROW University of Alaska Fairbanks, P.O. Box 755800, Fairbanks, AK 99775-5800

MICHAEL D. HENDRICKS RONALD P. DAANEN Alaska Department of Natural Resources, Division of Geological and Geophysical Surveys, 3354 College Road, Fairbanks, AK 99709

Key Terms: Landslide, Object-Oriented Image Analysis, LiDAR, Alaska ABSTRACT This study presents a new methodology to identify landslide and landslide-susceptible locations in Interior Alaska using only geomorphic properties from light detection and ranging (LiDAR) derivatives (i.e., slope, profile curvature, and roughness) and the normalized difference vegetation index (NDVI), focusing on the effect of different resolutions of LiDAR images. We developed a semi-automated object-oriented image classification approach in ArcGIS 10.5 and prepared a landslide inventory from visual observation of hillshade images. The multistage work flow included combining derivatives from 1-, 2.5-, and 5-m-resolution LiDAR, image segmentation, image classification using a support vector machine classifier, and image generalization to clean false positives. We assessed classification accuracy by generating confusion matrix tables. Analysis of the results indicated that LiDAR image scale played an important role in the classification, and the use of NDVI generated better results. Overall, the LiDAR 5-m-resolution image with NDVI generated the best results with a kappa value of 0.55 and an overall accuracy of 83 percent. The LiDAR 1-m-resolution image with NDVI generated the highest producer accuracy of 73 percent in identifying landslide locations. We produced a combined overlay map by summing the individual classified maps that was able to delineate landslide objects better than the individual maps. The combined classified map from 1-, 2.5-, and 5-m-resolution LiDAR with NDVI generated producer accuracies of 60, 80, *Corresponding author email: rakibjaved@gmail.com

and 86 percent and user accuracies of 39, 51, and 98 percent for landslide, landslide-susceptible, and stable locations, respectively, with an overall accuracy of 84 percent and a kappa value of 0.58. This semi-automated object-oriented image classification approach demonstrated potential as a viable tool with further refinement and/or in combination with additional data sources. INTRODUCTION A landslide is defined as “the movement of a mass of rock, earth, or debris down a slope” (Cruden, 1991). Landslides can initiate within bedrock or within the soil mass that covers bedrock. They drastically change the morphology of a landscape (Pradhan and Lee, 2009; Lee et al., 2012; and Kavzoglu et al., 2014), changing the slope, aspect, and curvature of the ground surface. These changes provide identifying characteristics to geomorphologists and, together with meteorological factors, help to identify potential landslide zones (Guzzetti et al., 1999; Kavzoglu et al., 2014). A substantial number of studies have used various techniques to identify landslides and to develop landslide susceptibility maps. Researchers around the world have used Geographic Information System (GIS) tools to identify and map landslides; Guzzetti et al. (1999) provided a review of these GIS techniques. The one common philosophy in developing susceptibility maps is that future landslides will occur under similar conditions as past and present landslides. Hence, a landslide inventory is an important first step in landslide susceptibility mapping (Guzzetti et al., 1999). Elevation data derived from light detection and ranging (LiDAR), interferometric synthetic aperture radar, and satellite imagery (e.g., WorldView-2, Spot5) have made it possible to identify landslides from

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

405

Miandad, Darrow, Hendricks, and Daanen

before-and-after images of an area; however, the scale or resolution of these data plays an important factor in identifying landslides, as small-scale landslides are often not visible in coarser-resolution images. The models used by different researchers to identify landslides can be divided into two broad categories: qualitative and quantitative. The qualitative approach relies on visual identification of landslides from satellite and/or aerial imagery, field surveying, and interpretation of historical photographs. Quantitative analysis can be divided into three categories: (1) probabilistic models (e.g., Lee and Pradhan, 2007; Yilmaz, 2009; Choi et al., 2012; Lee et al., 2012; Nourani et al., 2014; Kumar and Anbalagan, 2015; Ramesh and Anbazhagan, 2015; Youssef et al., 2015; Chen et al., 2016; Son et al., 2016; and Zhang et al., 2016b), (2) statistical models (e.g., Van Westen et al., 2003; Lee and Sambath, 2006; Lee and Pradhan, 2007; Neuhäuser and Terhorst, 2007; Falaschi et al., 2009; Yilmaz, 2009; Regmi et al., 2010, 2014; Dieu Tien et al., 2011; Choi et al., 2012; Lee et al., 2012; Mohammady et al., 2012; Devkota et al., 2013; Felicisimo et al., 2013; Ozdemir and Altural, 2013; Xu et al., 2013; Kavzoglu et al., 2014; Nourani et al., 2014; Pourghasemi et al., 2014; Budimir et al., 2015; Chen et al., 2016; Zhang et al., 2016a; and Ahmed and Dewan, 2017),), and (3) machine learning algorithms (e.g., Lee et al., 2004; Yao et al., 2008; Falaschi et al., 2009; Pradhan and Lee, 2009; Yilmaz, 2009; Marjanović et al., 2011; Ballabio and Sterlacchini, 2012; Choi et al., 2012; Lee et al., 2012; Li et al., 2012; Xu et al., 2012, 2016; Pradhan, 2013; Bi et al., 2014; Micheletti et al., 2014; Moosavi et al., 2014; Nourani et al., 2014; Tsangaratos and Benardos, 2014; Wu et al., 2014; Dou et al., 2015; Kavzoglu et al., 2014; Gelisli et al., 2015; Su et al., 2015; Hong et al., 2016; and Samodra et al., 2017). With advancements in processor computing power, machine learning techniques are becoming more and more feasible. Some researchers have used very different approaches in landslide identification. For example, Leshchinsky et al. (2015) developed the contour connection method. This method is a new algorithm to identify landslides using a Python script in GIS. A LiDAR-derived digital elevation model (DEM) is used to make contours, and each contour has nodal connections with other contours based on a set of values. Leshchinsky et al. (2015) demonstrated that the density of nodal connections is representative of different parts of a landslide (e.g., head scarp, body, and fan deposit). As the landslide disrupts the topography, it generates features with distinct contour shapes, such as concave contours in the head scarp and convex contours in the deposit area. The contour connection method generates connecting vectors between the con-

406

tours, and each landslide feature generates a different density of connecting vectors that are used as signatures to identify the features. Image classification techniques, also called pixelbased approaches, have been implemented in landslide identification. Supervised image classification methods can be applied using tools in GIS (Kasai et al., 2009). Guzzetti et al. (2012) point out that one problem with pixel-based approaches is that they do not consider the local geomorphological context (e.g., size, shape, and position) of the extracted feature. The problems of pixel-based approaches can be overcome using the more advanced technique of object-oriented image analysis (OOIA), which has been used successfully by Aksoy and Ercanoglu (2012), Li et al. (2015), and Martha et al. (2010), to identify landslides. This approach is a semi-automatic method of identifying landforms and other objects by extracting features using spectral, spatial, and morphometric attributes of segmented images. Specialized software is available that can handle OOIA well. Recently, the ArcGIS software was equipped with OOIA, and its capability in landslide identification is worth investigating. The goal of this research was to develop a methodology that can be used in Interior Alaska to identify landslides and potential landslide zones with limited available data. The road system in Alaska covers significant distances, often without redundancy. For example, should a landslide block one highway, vehicles may have to detour several hundred kilometers to reach their intended destination. These significant potential detours also would have an impact on the state’s economy. Our study area consists of 39.76 km2 of rugged terrain without much road access; hence, it is difficult to map landslides using traditional field mapping techniques. Additionally, no detailed landslide inventory exists for Interior Alaska at the current time. The inventory prepared for this study provides only locations of landslides and landslide-susceptible areas; the information necessary to use some of abovereferenced techniques (e.g., distance to faults and lineaments, rainfall density, etc.) is not always available in Alaska due to its size and complexity (Alaska Geospatial Council, 2011). As an example, only 20 percent of Alaska’s geology has been mapped at the inch-to-mile scale or larger (Alaska Division of Geological and Geophysical Surveys, 2019). Hence, traditional techniques are not suitable for use in this study. Given this data limitation, we developed a modified image classification approach to identify landslide locations from available data, which were LiDAR derivatives (e.g., slope, curvature, and roughness) and the normalized difference vegetation index (NDVI) obtained from color-infrared imagery. Van Den Eeckhaut et al. (2012) used LiDAR derivatives to identify

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

Landslide Mapping Using LiDAR

Figure 1. Location of study sites: Slate Creek, Copper River, Richardson, and Yukon. Highways are indicated by red lines, and in the inset, areas with LiDAR coverage are indicated by a green hatched pattern. Nearby communities are indicated in italicized text.

landslides in Belgium using eCognition software. In contrast, we used LiDAR derivatives in ArcGIS 10.5 and a supervised image classification approach with segmented images prepared from OOIA classification techniques. The samples were trained using a support vector machine (SVM) classifier, which follows a machine learning algorithm. The developed methodology was implemented at different resolutions (i.e., 1, 2.5, and 5 m) of LiDAR data. Finally, we used accuracy assessment techniques to quantify the results. Here we describe this new method developed using limited available data and discuss its advantages and limitations. STUDY AREA The study area broadly consists of road corridors within Interior Alaska (Figure 1). We prepared a landslide inventory for four study sites (i.e., Slate Creek, Copper River, Richardson, and Yukon) (Figure 2). We chose multiple training sites in different geological settings to improve landslide identification within the long transportation corridors of Interior Alaska (Figure 3). Although the surficial and bedrock geology vary considerably along the length of the Alaskan road system, the geology of each chosen site is representative of its respective study area as indicated in Figure 3.

Slate Creek The Slate Creek landslide (Figures 2a and 3a) is located on an east-facing river terrace near Mile Post (MP) 258 along the Parks Highway and adjacent to the Nenana River, approximately 9 mi north of Healy, Alaska. It is located in the northern foothills of the Alaska Range, in which permafrost is present (Wahrhaftig, 1965). The landslide occurred in the Tertiary Nenana Gravel Unit, which is a poorly consolidated ﬂuvial deposit consisting mostly of pebble to boulder conglomerate and coarse-grained sandstone, with interbedded mudﬂow deposits, thin claystone layers, and local thin lignite beds (Wahrhaftig, 1970; Wilson et al., 2015). As four distinct glacial advances have been recognized in the Nenana River area, the landslide area also contains outwash gravel that overlays the Nenana Gravel on the top of the river terrace (Wahrhaftig and Black, 1958). The Slate Creek landslide area is representative of an approximately 3,700-km2 region underlain by the Tertiary Nenana Gravel deposit, mostly located within the Healy, Mount McKinley, and Fairbanks quadrangles (Wilson et al., 2015). Major faults near the study site are the Denali fault, the Healy Creek fault, and the Northern Foothills thrust fault (Koehler, 2013); however, we did not investigate if seismic events along any of these faults were responsible for initiating the Slate

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

407

Miandad, Darrow, Hendricks, and Daanen

Figure 2. Landslide inventory prepared using visual inspection of LiDAR data: (a) Slate Creek, (b) Copper River, (c) Richardson, and (d) Yukon. The base layer is a hillshade image derived from 1-m-resolution LiDAR (Hubbard et al., 2011).

Creek landslide, as that was beyond the scope of this study. Copper River The Copper River training site (Figures 2b and 3b) is located to the west of Richardson Highway near MP 112 along the Tazlina River, which is a tributary of the Copper River. This training site is in the Copper River Lowland physiographic province (Wahrhaftig, 1965) and consists of a relatively smooth plain that is entrenched by the Copper River and its tributaries. The mountains of the Alaska, Talkeetna, Chugach, and Wrangell ranges surround the Copper River Basin. During the Pleistocene glaciations,

408

glaciers advanced from these mountains and dammed the basinâ&#x20AC;&#x2122;s drainage, forming a proglacial lake known as Lake Atna (Ferrians et al., 1983). Thus, the soils of the Copper River Basin consist mainly of glaciolacustrine sediments that cover approximately 13,000km2 area (Ferrians, 1989). Permafrost began to form in these sediments following the retreat of glaciers and drainage of the lake (Wahrhaftig, 1965; Ferrians et al., 1983; and Wiedmer et al., 2010). Richardson Highway MP 296 This site is located along the Richardson Highway, starting at MP 296 and extending to the north to MP 299 (Figures 2c and 3c). It is in the Yukon-Tanana

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405â&#x20AC;&#x201C;425

Landslide Mapping Using LiDAR

Figure 3. Geology of the study sites: (a) Slate Creek, (b) Copper River, (c) Richardson, and (d) Yukon (Wilson et al., 2015).

Upland physiographic province, which is characterized by rounded even-topped ridges with gentle side slopes (Wahrhaftig, 1965). The landslides on the north side of the highway are in Devonian and older pelitic schist and quartzite (Figure 3c). Similar bedrock also is present through much of Interior Alaska, with approximately 13,350 km2 identiﬁed in the Fairbanks, Livengood, Circle, Big Delta, Eagle, and Charley River quadrangles (Wilson et al., 2015). To the south of the highway, the slopes are covered with unconsolidated surﬁcial deposits of Quaternary

age (Figure 3c; Wilson et al., 2015). The study site runs parallel to the Tanana River, and south-facing slopes in the low terraces of the Tanana River may contain widespread, shallow, and locally ice-rich permafrost (Reger and Solie, 2008; Reger et al., 2008). Yukon River The training site along the Yukon River is located west of the E. L. Patton Bridge (Figures 2d and 3d). This area is on the boundary between

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

409

Miandad, Darrow, Hendricks, and Daanen Table 1. Summary of data used in this study. Data Type

Sensor

Spatial Reference

Scale

Data Derivatives

LiDAR DEM

Airborne

NAD 1983 UTM

Color-infrared image

Spot-5

NAD 1983 UTM

1m×1m 2.5 m × 2.5 m 5m×5m 2.5 m × 2.5 m

Slope Proﬁle curvature Roughness NDVI

DEM = digital elevation mode; NDVI = normalized diﬀerence vegetation index.

the Kokrine-Hodzana Highlands, the Rampart Trough, and the Yukon-Tanana Upland physiographic provinces (Wahrhaftig, 1965; Koehler et al., 2013). Bedrock outcrops exposed along the river consist mainly of Mississippian-Triassic intrusive and extrusive mafic igneous rocks with some sedimentary rocks, such as argillite, chert, greywacke, shale, and limestone (Figure 3d; Wilson et al., 2015). Along the road corridor, similar rocks outcrop along the Dalton Highway in the northwestern part of the Livengood quadrangle, representing approximately 18,700 km2 of Interior Alaska. Frozen surficial deposits consist mainly of loess, which is 1.5 to 15 m thick on ridge crests and ice-rich in stream valleys (Weber et al., 1992; Koehler, 2011; Koehler et al., 2013; and Wilson et al., 2015). DATA SETS A publicly available LiDAR data set was used in this research. The Alaska Division of Geological and Geophysical Surveys hosts 1-m-resolution LiDAR data for an area of 7,770 km2 (3,000 mi2 ) with 1.6-km (1-mi) width coverage along major infrastructure corridors (Figure 1; Hubbard et al., 2011). LiDAR DEMs were resampled to 2.5- and 5-m scales using the bilinear interpolation method, which calculates the value of each pixel by averaging the values of the four surrounding pixels. To calculate the NDVI, we used Spot 5 colorinfrared imagery. This imagery is available for the entire state from the Alaska Statewide Digital Mapping Initiative. Table 1 summarizes the details of the data used in this study. METHODOLOGY Slope, profile curvature, and roughness (Figure 4) were derived from the LiDAR DEMs at different scales (e.g., 1, 2.5, and 5 m). Surface roughness describes the variability of a topographic surface at a certain scale (Grohmann et al., 2011). Some common surface roughness measures are standard deviation of residual topography, standard deviation of profile curvature, and standard deviation of slope. Some methods

410

can show local relief, while other methods are good for delineating regional relief. Since a landslide body has a higher roughness index than its surroundings, we used the topographic roughness index (Figure 4c; Jenness, 2004). Equation 1 was used to calculate roughness using a focal statistics tool with a 5 × 5-cell moving window from a LiDAR DEM: mean − minimum roughness = , (1) maximum − minimum where mean, minimum, and maximum are the statistics derived from the DEM. The focal statistics tool computes a statistic value in each cell of the DEM by performing a neighborhood operation. The 5 × 5 moving window considers five cells surrounding the input cell; computes statistics such as the mean, minimum, or maximum of all values encountered in that neighborhood; and assigns those statistics to the particular input cell. The three thematic rasters (i.e., slope, profile curvature, and roughness) derived from LiDAR data were combined to create a multispectral composite band image where bands 1, 2, and 3 were slope, profile curvature, and roughness rasters, respectively. The composite band image was used to generate a segmented image. We also used NDVI in the subsequent training phase (see Figure 4d using the Slate Creek study site as an example). NDVI is an index parameter that indicates the health of different plants. Very low and negative NDVI values indicate areas without vegetation, such as bodies of water, snow, rock, or bare soil. Since our observed landslide head scarps have little to no vegetation, they have low NDVI values (e.g., the Slate Creek landslide head scarp in Figure 4d). We hypothesized that the NDVI may work as an identifying parameter. Figure 5 illustrates the work flow for the OOIA. Image Segmentation A segmented image is generated using the “Segment Mean Shift” tool in ArcGIS 10.5. Accuracy of the classification depends on the segmentation process. In this process, an image is segmented on the basis of its spectral, spatial, and morphometric properties. Prior to segmentation, we determined values for

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

Landslide Mapping Using LiDAR

Figure 4. Examples of derived data for the Slate Creek study site: (a) slope, (b) proﬁle curvature, and (c) roughness derived from the 1m-resolution LiDAR digital elevation model; (d) normalized diﬀerence vegetation index derived from the 2.5-m-resolution color-infrared imagery.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

411

Miandad, Darrow, Hendricks, and Daanen

the “minimum segment size.” ArcGIS merges segments smaller than this size with their best-matched neighbors. This “minimum segment size” depends on the scale of the image. Higher segment sizes result in a smoother image. Since the smallest landslide polygon in our inventory contained approximately 300 pixels, we chose this as the minimum segment size. Next, we generated segmented images for 1-, 2.5-, and 5-mresolution LiDAR data (see Figure 6 for examples of the 1-m segmented images). Training Samples

Figure 5. Work ﬂow of object-oriented image analysis and classiﬁcation in ArcGIS 10.5.

spectral detail, spatial detail, and minimum segment size. Appropriate values for these parameters are often found by trial and error (Drăguţ and Blaschke, 2006; Aksoy and Ercanoglu, 2012); the range available in ArcGIS for spectral and spatial values is between 1 and 20. Although we did not perform a sensitivity analysis, we did test multiple combinations of these parameters to determine the best result. For example, a low spectral value results in a smoother image and will cause the landslide bodies and stable locations to have the same spectral signature. Since we wanted to distinguish the landslide bodies from other areas, we used a high spectral value of 18. A high spatial value is used where features of interest are small and clustered together. The landslide bodies had variable pixel characteristics, and we wanted to cluster them together to form an object; however, water bodies, roads, and other stable locations had less variable pixels. We used a value of 10 to get an optimum number of features with similar characteristics. The ﬁnal input in the “Segment Mean Shift” function is

412

After the segmented images were prepared, training samples were extracted into three categories: landslides, landslide susceptible, and stable. In this study, the phrase “landslide susceptible” was used to define places where future landslides can occur, and the word “landslide” was used to define locations of recent and/or historic landslides as determined by visual examination of the 1-m LiDAR DEM. All other places were considered stable along with water bodies and roads. This training was perhaps the most important step since the accuracy of the classified image depended on how carefully the classifier was trained. It also depended on the judgment of the person selecting the samples from the segmented images, making this process semi-automatic. We chose training samples as area-based objects represented by polygons instead of individual pixels to follow the OOIA classification technique. In the landslide inventory, 21 polygons were identified as landslides and 75 as landslide-susceptible locations. All other areas were designated as stable (Figure 2). From this inventory, we chose 15 landslide polygons, 57 landslide-susceptible polygons, and 52 stable polygons for training. The training sites were chosen randomly, and that made this process semiautomatic. In one instance, we had to sub-sample from a mapped landslide location because that landslide body had a similar spectral signature to the surrounding stable features (Figure 7a). We had less confidence in identifying landslide locations in the inventory since field verification was not possible for all locations. This resulted in a smaller number of both landslide polygons and training polygons. Many paleo-landslides could be present in the 75 polygons that were mapped as landslide susceptible. SVM Classifier ArcGIS 10.5 offers a range of classifiers for image classification purposes. The SVM classifier is a wellknown tool that uses machine learning algorithms. It

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

Landslide Mapping Using LiDAR

Figure 6. Segmented image generated from 1-m-resolution LiDAR. Study sites are (a) Slate Creek, (b) Copper River, (c) Richardson, and (d) Yukon.

requires a large number of training samples, but the samples do not need to be normally distributed. We used the SVM classifier to train the signature files, which then were used to classify the entire study area. The classify raster function was used to generate classified images using these signature files. We implemented a generalization technique on the classified images to expand eight neighboring cells of the stable locations. We used the expand tool to clean up false positives from stable locations in the classified raster. With the expand tool, a cell can expand into its neighboring cells. In our case, false positives were the landslide/landslide-susceptible pixels in the stable zones. The number of cells needed to expand depended on the concentration of false positives in an area. Ex-

panding by eight cells removed most of the false positives and improved the overall accuracy. Accuracy Assessment As with any image classification technique, accuracy assessment is an important measure to quantify classification accuracy. At present, no standard evaluation method for assessing classified images from image segmentation exists (Drăguţ et al., 2011; Van Den Eeckhaut et al., 2012). We followed the same procedure used in traditional image classification techniques to quantify accuracy. In this method, a matrix is generated consisting of classified and referenced points as a tally of rows and columns, respectively. The major

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

413

Miandad, Darrow, Hendricks, and Daanen

Figure 7. Training sample locations in the study area. Study sites are (a) Slate Creek, (b) Copper River, (c) Richardson, and (d) Yukon.

diagonal of the confusion matrix indicates the points that are correctly classified, and the overall accuracy of the classification is obtained by dividing the sum of the major diagonal values by the total number of samples. The other two ways to determine the accuracy of individual categories are producer’s accuracy and user’s accuracy. Producer’s accuracy is obtained by dividing the total number of correctly classified data for a category by the total number of ground-truthed points in that category. It is called producer’s accuracy because the producer of an image or map is concerned about how well a certain category on the ground can be classified (Congalton, 1991). User’s accuracy is obtained by dividing the total number of correctly classified data for a category by the total number of classified data sampled in that category. User’s accuracy represents how well a map represents the original ground (Story and Congalton, 1986; Congalton, 1991). Another measure of accuracy is kappa analysis (Congalton et al., 1983), which is often used in remotely sensed data. The kappa

414

analysis generates a statistic called KHAT, which measures the agreement between referenced and classified data. KHAT values greater than 80 percent, between 40 and 80 percent, and less than 40 percent indicate strong, moderate, or poor agreement, respectively, between classified and referenced data (Environmental Systems Research Institute, 2017). In this study, we assessed the accuracy of 3,000 randomly generated points using the stratified random function. This function generates points that are proportional to the relative area of each class (e.g., landslide and stable). Since stable locations had the highest proportion, these areas contained the highest number of random points. RESULTS AND DISCUSSION Figures 8 and 9 are examples of the classified images from the 1- and 5-m segmented images, respectively, that incorporated the NDVI, and Figures 10

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

Landslide Mapping Using LiDAR

Figure 8. Classiﬁed image from 1-m-resolution LiDAR with normalized diﬀerence vegetation index. Study sites are (a) Slate Creek, (b) Copper River, (c) Richardson, and (d) Yukon.

and 11 are the classified images from the 1- and 5m segmented images without the NDVI, respectively. From visual inspection, the finer-resolution 1-m classified image with NDVI (Figure 8) delineated landslide polygons better than the coarser-resolution 2.5and 5-m classified images; however, it also produced more false positives, leading to low user accuracy. At the other extreme, the 5-m classified image without NDVI did not identify any landslide polygons, with all landslide locations identified as landslide susceptible (Figure 11). This may be attributed to the oversampling of training samples as landslide-susceptible pixels. Overall, the addition of NDVI produced better results, and the 5-m-resolution LiDAR with NDVI generated fewer false positives than did the 1- and 2.5-m

scales. Note that we used the same training samples for all scales. Tables 2 and 3 summarize the accuracy assessment results for the 1-m classified images with and without NDVI, respectively. Similar tables were generated for 2.5- and 5-m-resolution classified images. Table 4 summarizes overall accuracy and kappa statistics for the 1-, 2.5-, and 5-m-resolution LiDAR data. The overall accuracy of 1-, 2.5-, and 5-m-resolution classified images varies between 80 and 83 percent, with the highest accuracy obtained from the 5-m-resolution image with NDVI and the lowest accuracy from the 1-m-resolution LiDAR without NDVI (Table 4). The high overall accuracy at each scale may be attributed to the fact that the majority of the accuracy assessment

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

415

Miandad, Darrow, Hendricks, and Daanen

Figure 9. Classiﬁed image from 5-m-resolution LiDAR with normalized diﬀerence vegetation index. Study sites are (a) Slate Creek, (b) Copper River, (c) Richardson, and (d) Yukon.

Table 2. Confusion matrix table for 1-m-resolution LiDAR with normalized diﬀerence vegetation index. Referenced Data Classiﬁed Data

Stable

Landslide Susceptible

Landslide

Total

User Accuracy

Stable Landslide susceptible Landslide Total Producer accuracy Kappa

2,129 114 149 2,392 0.89

111 204 180 495 0.41

16 15 83 114 0.73

2,256 333 412 3,001

0.94 0.61 0.20

416

Kappa

0.81 0.48

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

Landslide Mapping Using LiDAR

Figure 10. Classiﬁed image from 1-m-resolution LiDAR without normalized diﬀerence vegetation index. Study sites are (a) Slate Creek, (b) Copper River, (c) Richardson, and (d) Yukon.

Table 3. Confusion matrix table for 1-m-resolution LiDAR without normalized diﬀerence vegetation index. Referenced Data Classiﬁed Data

Stable

Landslide Susceptible

Landslide

Total

User Accuracy

Stable Landslide susceptible Landslide Total Producer accuracy Kappa

2,094 229 54 2,377 0.88

121 281 123 525 0.54

16 45 37 98 0.38

2,231 555 214 3,000

0.94 0.51 0.17

Kappa

0.80

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

0.48

417

Miandad, Darrow, Hendricks, and Daanen

Figure 11. Classiﬁed image from 5-m-resolution LiDAR without normalized diﬀerence vegetation index. Study sites are (a) Slate Creek, (b) Copper River, (c) Richardson, and (d) Yukon.

points were generated in stable locations that were correctly classified. The overall accuracy was not a good indicator of how landslide and landslide-susceptible locations were correctly classified. The highest kappa value, 55 percent, was obtained from the 5-m-resolution LiDAR with NDVI, and the lowest kappa value of 43 percent was obtained from the 5-m-resolution LiDAR without NDVI. The 1- and 2.5-m-resolution classified images generated kappa values of 48 and 49 percent, respectively, for both with and without NDVI (Table 4). Producer and user accuracies provided a good indication of landslide and landslide-susceptible locations

418

in this accuracy assessment. The additional NDVI input generated better results in every instance of landslide identiﬁcation. The 1-m-scale LiDAR with NDVI generated a producer accuracy of 73 percent for landslide locations, whereas the same scale LiDAR without the addition of NDVI generated a producer accuracy of only 38 percent. Similarly, the 2.5-m LiDAR with NDVI generated a user accuracy of 32 percent and without NDVI generated a user accuracy of only 22 percent in identifying landslide locations (Table 5). Overall, the addition of NDVI yielded better results in user accuracy. The proposed method generally performed better in identifying landslide-susceptible

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

Landslide Mapping Using LiDAR Table 4. Summary of overall accuracy and kappa statistics for multi-resolution LiDAR data. NDVI Scale LiDAR, 1 m LiDAR, 2.5 m LiDAR, 5 m

Without NDVI

Overall Accuracy

Kappa

Overall Accuracy

Kappa

0.81 0.81 0.83

0.48 0.49 0.55

0.80 0.81 0.82

0.48 0.49 0.43

NDVI = normalized diﬀerence vegetation index.

locations (resulting in higher producer and user accuracy) than for identifying landslide locations as discussed above, and the highest producer and user accuracies were obtained for stable locations (Table 5). For example, a user accuracy of 95 percent was achieved in identifying stable locations in both the 5-m-resolution LiDAR with NDVI and the 2.5-m-resolution LiDAR without NDVI. In general, the addition of NDVI with segmented images from slope, profile curvature, and roughness yielded better results in identifying landslide locations. We combined the 1-, 2.5-, and 5-m-resolution classified images with NDVI and without NDVI using the raster calculator to generate a combined classification map. For each individual classified image, we assigned pixels in landslide, landslide-susceptible, and stable locations values of 2, 1, and 0, respectively. Through raster addition, this resulted in a combined map with six class values. Next, we reclassified the combined map into a new map using only the three original classes (see Figure 12 for an example with NDVI). We conducted an accuracy assessment on the combined classified image to generate a confusion matrix. The combined classified map from 1-, 2.5-, and 5m LiDAR with NDVI generated a kappa value of

58 percent and user accuracy of 39 percent for landslide locations (Table 6). This user accuracy is higher than for any of the individual classified maps. From visual inspection, the entire extent of landslide masses is identified in the combined map (Figure 12), whereas in the individual maps, often only the head scarp, toe, or a smaller portion of the landslide was identified (Figures 8 and 9). The combined classified map from 1-, 2.5-, and 5-m LiDAR without NDVI generated a kappa value of 53 percent and user accuracy of 32 percent for landslide locations (Table 7); however, this map produced a large number of false-positive landslide locations, especially for the Richardson and Yukon sites. It is possible that some of the falsepositive landslide locations are paleo-landslides. One limitation of this study is that the landslide inventory was not field checked with the exception of the Slate Creek study site and some of the landslides at the Richardson site. The accuracy assessment method used for the classified images is usually done for land cover classification. No standard classification procedure has yet been developed for object-oriented image classification. Since training samples were taken as area-based objects represented by polygons in this study, it may have been

Table 5. Summary of producer accuracy and user accuracy for multi-resolution LiDAR data. NDVI Location and Scale Landslide LiDAR, 1 m LiDAR, 2.5 m LiDAR, 5 m Landslide susceptible LiDAR, 1 m LiDAR, 2.5 m LiDAR, 5 m Stable LiDAR, 1 m LiDAR, 2.5 m LiDAR, 5 m

Without NDVI

Producer Accuracy

User Accuracy

Producer Accuracy

User Accuracy

0.73 0.50 0.61

0.20 0.32 0.29

0.38 0.23 0.00

0.17 0.22 0.00

0.41 0.67 0.65

0.61 0.48 0.58

0.54 0.69 0.55

0.51 0.47 0.50

0.89 0.85 0.88

0.94 0.94 0.95

0.88 0.86 0.92

0.94 0.95 0.89

NDVI = normalized diﬀerence vegetation index.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

419

Miandad, Darrow, Hendricks, and Daanen

Figure 12. Combined classiﬁed image from 1-, 2.5-, and 5-m-resolution LiDAR data with normalized diﬀerence vegetation index. Study sites are (a) Slate Creek, (b) Copper River, (c) Richardson, and (d) Yukon.

Table 6. Confusion matrix table for combined classified image from 1-, 2.5-, and 5-m-resolution LiDAR with normalized difference vegetation index. Referenced Data Classified Data

Stable

Landslide Susceptible

Landslide

Total

User Accuracy

Stable Landslide susceptible Landslide Total Producer accuracy Kappa

2,102 309 37 2,448 0.86

31 357 60 448 0.80

8 34 62 104 0.60

2,141 700 159 3,000

0.98 0.51 0.39

420

Kappa

0.84 0.58

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

Landslide Mapping Using LiDAR Table 7. Confusion matrix table for combined classified image from 1-, 2.5-, and 5-m-resolution LiDAR without normalized difference vegetation index. Referenced Data Classified Data

Stable

Landslide Susceptible

Landslide

Total

User Accuracy

Stable Landslide susceptible Landslide Total Producer accuracy Kappa

2,107 270 31 2,408 0.88

99 330 62 491 0.67

17 40 44 101 0.44

2,223 640 137 3,000

0.95 0.52 0.32

more appropriate to generate a confusion matrix using ground-truthed objects (e.g., landslide, landslide susceptible, and stable). While both producer and user accuracy quantify the accuracy of individual categories, producer accuracy perhaps is more useful from a hazard management perspective. This is because producer accuracy tells us how many ground-truthed points are correctly classified as landslides. Additionally, the question of “what is good enough” for accuracy may arise. The answer depends on the specific need. Kappa values usually provide a measure of the overall performance and strength of agreement of the classification. Landis and Koch (1977) suggested the following ranges for kappa values: 0 to 0.2 is poor, 0.2 to 0.4 is fair, 0.4 to 0.6 is moderate, 0.6 to 0.8 is substantial, and 0.8 to 1 is nearly perfect. Our results demonstrated kappa values between 0.43 and 0.58, which were moderate. We achieved an overall accuracy of more than 80 percent at all scales, which can be considered good. Producer accuracy for landslide locations was between 50 and 73 percent, which is moderate to good; however, user accuracy for landslide locations was between 20 and 30 percent, which can be improved by incorporating more training samples. The low number of landslide polygons in the inventory affected the accuracy of the classification, as it left only six polygons for verification (we decided to use more samples as training polygons in case a landslide inventory would be available in the future to verify the classification). This resulted in lower user accuracy than producer accuracy at every scale for landslide locations (Table 5). This is because producer accuracy depends on the number of ground-truthed polygons, and in this case, we had more ground-truthed or landslide training polygons compared to verification polygons. Hence, user accuracy is less representative of the final accuracy of this model. The roughness values depend on the method used for calculation. The method we used for roughness demonstrated extreme local relief even in waves generated on the Yukon River surface. Other methods may produce different results. The spatial resolution and

Kappa

0.83 0.53

quality of the DEM and color-infrared imagery may affect the results of this study. Scale plays an important role in identifying landslides. From the combined classified map, it is evident that different parts of the same landslide body may be not be identifiable at different scales. This is the reason that the combined classified map gives a fuller picture since it incorporates results from all three scales. The combined classified approach can be taken to generate landslide hazard zonation maps. The image segmentation approach used in this study will need further development. Landslide identification using this technique depends on the minimum size of the segments. If a landslide body is smaller than the minimum segment generated from the segment mean shift function, it is possible that it would not be identified in subsequent phases. The spectral fingerprints of landslides and landslide-susceptible locations are similar; however, the shape of the training objects helped to separate these different areas. It is important to select training samples that resemble all possible shapes in the study area. For our work, each training polygon included multiple segments. The low user accuracy in identifying landslide objects can be attributed to the fact that both landslide and landslide-susceptible polygons consisted of similar segments. Further study is necessary to generate a segmented image that will have more distinct segments for the landslide and landslidesusceptible locations. CONCLUSIONS Landslide identification is important, as it helps to manage this type of geohazard in a planned way. The transportation corridors in Interior Alaska have significant importance to the state’s economy and transportation sector. Landslides located within these corridors may represent serious hazards to the roadways. Field identification of these landslides is often not feasible due to budget and time constraints. Hence, it is important to develop an expedient and sufficiently accurate methodology that can be implemented us-

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

421

Miandad, Darrow, Hendricks, and Daanen

ing the limited data available. For this research, we developed a new method for landslide mapping using an object-oriented image classification approach in ArcGIS 10.5. We chose this approach since it may be able to identify landslides rapidly using only LiDAR and color-infrared imagery. Analysis of the results indicated that the combination of different derivatives from a LiDAR digital elevation model and the NDVI from color-infrared imagery can be used to identify landslide and landslidesusceptible locations. The results also demonstrated that the scale of the LiDAR data plays an important role. The finer-resolution image generated more false positives. The overall accuracy of LiDAR 1-, 2.5-, and 5-m classified images with NDVI were 81, 81, and 83 percent, respectively. The 2.5-m-resolution LiDAR data with NDVI performed well in identifying landslide locations compared to the other scales with a user accuracy of 32 percent, a producer accuracy of 50 percent, and an overall accuracy of 81 percent. In general, the 5-m LiDAR with NDVI performed well in identifying landslide, landslide-susceptible, and stable locations and generated the highest kappa value of 55 percent. The results indicate that image segmentation plays an important role in object-oriented image classification. Different scales of data sets identified different parts of landslides, and the combined map generated the best result in identifying each class by combining all the pixels from individual maps. The combined map with NDVI attained user accuracies of 39, 51, and 98 percent and producer accuracies of 60, 80, and 86 percent for landslide, landslide-susceptible, and stable locations, respectively. The overall accuracy of the combined map was 84 percent. The method produced in this research is fast and can be implemented easily over a large study area. It requires only LiDAR-derived products and NDVI from color-infrared imagery. Although not as accurate, the method can be implemented even without the NDVI. It is similar to the conventional image classification approach since it is supervised; however, it eases the rigorous procedure of visually identifying individual pixels and assigning them to different classes by developing training files using large objects containing many pixels. This method can be improved by fine-tuning the segmented image generation, and, of course, the results should be verified through field investigations. Because we used mainly LiDAR data, this approach relies heavily on the geomorphic expression of landslides. As a result, the age of the landslide becomes important, as the method may not easily identify paleo-landslides that have subtle surface expression. A future step to address this issue is to conduct radiometric dating of the identified landslides within the

422

transportation corridors. Another limitation with the LiDAR data is their narrow width along the corridors; this limited the size and orientation of landslides that we were able to identify. This method used the 2011 LiDAR data set. Should another LiDAR data set be collected along these transportation corridors, the two digital elevation models could be differenced to determine movement that occurred between data set acquisitions. This change detection would be one desktop approach to validate our methodology. The purpose of this study was to develop a methodology that could be used to identify landslides with only LiDAR-derived parameters and NDVI since spatial data in Alaska are typically sparse. Thus, the methodology presented here is inherently limited. To develop a robust landslide inventory and/or landslide susceptibility map, we recommend including other data sets (e.g., hydrology, soil, geology, and material properties) to incorporate other environmental and geologic parameters that are important to landslide occurrence. Additionally, the use of historic optical imagery will help to determine the timing of landslide events. Furthermore, in Interior Alaska, the presence of permafrost is an important parameter. Incorporating permafrost distribution and slope aspect considerations would result in a more comprehensive methodology, especially when coupled with historic optical imagery and changes in temperature distribution and amount of precipitation. Despite these limitations, this semi-automated object-oriented image classification approach demonstrated potential as a viable tool with further refinement and/or in combination with additional data sources.

REFERENCES Ahmed, B. and Dewan, A., 2017, Application of bivariate and multivariate statistical techniques in landslide susceptibility modeling in Chittagong city corporation, Bangladesh: Remote Sensing, Vol. 9, No. 4, p. 304. Aksoy, B. and Ercanoglu, M., 2012, Landslide identification and classification by object-based image analysis and fuzzy logic: An example from the Azdavay region (Kastamonu, Turkey): Computers and Geosciences, Vol. 38, No. 1, pp. 87–98. Alaska Division of Geological and Geophysical Surveys, 2019, Geologic Mapping: state Coverage: Electronic document, available at http://dggs.alaska.gov/maps-data/statecoverage.html (accessed December 4, 2019). Alaska Geospatial Council, 2011, Alaska Geospatial Strategic Plan: Electronic document, available at http://dnr.alaska. gov/agc/documents/AKGeospatialStrategicPlanFINAL.pdf (accessed December 4, 2019). Ballabio, C. and Sterlacchini, S., 2012, Support vector machines for landslide susceptibility mapping: The Staffora River Basin case study, Italy: Mathematical Geosciences, Vol. 44, No. 1, pp. 47–70.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

Landslide Mapping Using LiDAR Bi, R.; Schleier, M.; Rohn, J.; Ehret, D.; and Xiang, W., 2014, Landslide susceptibility analysis based on ArcGIS and artificial neural network for a large catchment in Three Gorges region, China: Environmental Earth Sciences, Vol. 72, No. 6, pp. 1925–1938. Budimir, M. E. A.; Atkinson, P. M.; and Lewis, H. G., 2015, A systematic review of landslide probability mapping using logistic regression: Landslides, Vol. 12, No. 3, pp. 419–436. Chen, W.; Ding, X.; Zhao, R.; and Shi, S., 2016, Application of frequency ratio and weights of evidence models in landslide susceptibility mapping for the Shangzhou District of Shangluo City, China: Environmental Earth Sciences, Vol. 75, No. 1, Article 64. Choi, J.; Oh, H. J.; Lee, H. J.; Lee, C.; and Lee, S., 2012, Combining landslide susceptibility maps obtained from frequency ratio, logistic regression, and artificial neural network models using ASTER images and GIS: Engineering Geology, Vol. 124, pp. 12–23. Congalton, R. G., 1991, A review of assessing the accuracy of classifications of remotely sensed data: Remote Sensing Environment, Vol. 37, No. 1, pp. 35–46. Congalton, R. G.; Oderwald, R. G.; and Mead, R. A., 1983, Assessing Landsat classification accuracy using discrete multivariate analysis statistical techniques: Photogrammetric Engineering and Remote Sensing, Vol. 49, No. 12, pp. 1671–1678. Cruden, D. M., 1991, A simple definition of a landslide: Bulletin International Association Engineering Geology, Vol. 43, pp. 27– 29. Devkota, K. C.; Regmi, A. D.; Pourghasemi, H. R.; Yoshida, K.; Pradhan, B.; Ryu, I. C.; Dhital, M. R.; and Althuwaynee, O. F., 2013, Landslide susceptibility mapping using certainty factor, index of entropy and logistic regression models in GIS and their comparison at Mugling-Narayanghat road section in Nepal Himalaya: Natural Hazards, Vol. 65, No. 1, pp. 135–165. Dieu Tien, B.; Lofman, O.; Revhaug, I.; and Dick, O., 2011, Landslide susceptibility analysis in the Hoa Binh Province of Vietnam using statistical index and logistic regression: Natural Hazards, Vol. 59, No. 3, pp. 1413–1444. Dou, J.; Yamagishi, H.; Pourghasemi, H. R.; Yunus, A. P.; Song, X.; Xu, Y.; and Zhu, Z., 2015, An integrated artificial neural network model for the landslide susceptibility assessment of Osado Island, Japan: Natural Hazards, Vol. 78, No. 3, pp. 1749–1776. Drăguţ, L. and Blaschke, T., 2006, Automated classification of landform elements using object-based image analysis: Geomorphology, Vol. 81, No. 3, pp. 330–344. Drăguţ, L.; Eisank, C.; and Strasser, T., 2011, Local variance for multi-scale analysis in geomorphometry: Geomorphology, Vol. 130, No. 3–4, pp. 162–172. Environmental Systems Research Institute, 2017, ArcGIS Desktop Release 10.5: Environmental Systems Research Institute, Redlands, CA. Falaschi, F.; Giacomelli, F.; Federici, P. R.; Puccinelli, A.; D’Amato Avanzi, G.; Pochini, A.; and Ribolini, A., 2009, Logistic regression versus artificial neural networks: Landslide susceptibility evaluation in a sample area of the Serchio River valley, Italy: Natural Hazards, Vol. 50, No. 3, pp. 551–569. Felicisimo, A. M.; Cuartero, A.; Remondo, J.; and Quiros, E., 2013, Mapping landslide susceptibility with logistic regression, multiple adaptive regression splines, classification and regression trees, and maximum entropy methods: A comparative study: Landslides, Vol. 10, No. 2, pp. 175–189. Ferrians, O. J., Jr., 1989. Glacial Lake Atna, Copper River Basin, Alaska. Late Cenozoic History of the Interior Basins of

Alaska and the Yukon; U.S. Geological Survey Circular 1026, US Government Printing Office, Washington, pp. 85–88. Ferrians, O. J., Jr.; Nichols, D. R.; and Williams, J. R., 1983, Copper River Basin. In Péwé, T. L. and Reger, R. D. (Editors), Guidebook to Permafrost and Quaternary Geology along the Richardson and Glenn Highways, between Fairbanks and Anchorage, Alaska: Guidebook 1, Alaska Division of Geological and Geophysical Surveys, Fairbanks, AK. pp. 39–44. Gelisli, K.; Kaya, T.; and Babacan, A. E., 2015, Assessing the factor of safety using an artificial neural network: Case studies on landslides in Giresun, Turkey: Environmental Earth Sciences, Vol. 73, No. 12, pp. 8639-8646. Grohmann, C. H.; Smith, M. J.; and Riccomini, C., 2011, Multiscale analysis of topographic surface roughness in the Midland Valley, Scotland: IEEE Transactions Geoscience Remote Sensing, Vol. 49, No. 4, pp. 1200–1213. Guzzetti, F.; Carrara, A.; Cardinali, M.; and Reichenbach, P., 1999, Landslide hazard evaluation: A review of current techniques and their application in a multi-scale study, Central Italy: Geomorphology, Vol. 31, No. 1–4, pp. 181–216. Guzzetti, F.; Mondini, A. C.; Cardinali, M.; Fiorucci, F.; Santangelo, M.; and Chang, K. T., 2012, Landslide inventory maps: New tools for an old problem. Earth-Science Reviews, Vol. 112, No. 1–2, pp. 42–66. Hong, H.; Pradhan, B.; Jebur, M. N.; Bui, D. T.; Xu, C.; and Akgun, A., 2016, Spatial prediction of landslide hazard at the Luxi area (China) using support vector machines: Environmental Earth Sciences, Vol. 75, No. 1, pp. 1–14. Hubbard, T. D.; Koehler, R. D.; and Combellick, R. A., 2011, High-Resolution Lidar Data for Alaska Infrastructure Corridors: Raw Data File 2011-3, Alaska Division of Geological and Geophysical Surveys, Fairbanks, AK, 291 p., available at http://doi.org/10.14509/22722. Jenness, J. S., 2004, Calculating landscape surface area from digital elevation models: Wildlife Society Bulletin, Vol. 32, No. 3, pp. 829–839. Kasai, M.; Ikeda, M.; Asahina, T.; and Fujisawa, K., 2009, LiDAR-derived DEM evaluation of deep-seated landslides in a steep and rocky region of Japan: Geomorphology, Vol. 113, No. 1, pp. 57–69. Kavzoglu, T.; Sahin, E. K.; and Colkesen, I., 2014, Landslide susceptibility mapping using GIS-based multi-criteria decision analysis, support vector machines, and logistic regression: Landslides, Vol. 11, No. 3, pp. 425–439. Koehler, R. D., 2011, Application of lidar to mapping geologic hazards along gas pipelines in Alaska (poster): Geological Society of America Abstracts with Programs, Vol. 43, No. 4, p. 83, Paper No. 30-2, Alaska Division of Geological and Geophysical Surveys, Fairbanks, AK, available at http://doi.org/10.14509/22623. Koehler, R. D., 2013, Quaternary Faults and Folds (QFF): Digital Data Series 3, Alaska Division of Geological and Geophysical Surveys, Fairbanks, AK, available at http://doi.org/ 10.14509/qff. http://doi.org/10.14509/24956. Koehler, R. D.; Reger, R. D.; Sicard, K. R.; and Spangler, E. R., 2013, Yukon River Bridge Landslide: Preliminary Geologic and Geotechnical Evaluation: Preliminary Interpretive Report 2013-6, Alaska Division of Geological and Geophysical Surveys, Fairbanks, AK, 69 p., available at http://doi.org/10.14509/25642. Kumar, R. and Anbalagan, R., 2015, Landslide susceptibility zonation in part of Tehri reservoir region using frequency ratio, fuzzy logic and GIS: Journal Earth System Science, Vol. 124, No. 2, pp. 431–448.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

423

Miandad, Darrow, Hendricks, and Daanen Landis, J. R. and Koch, GG., 1977. The measurement of observer agreement for categorical data: International Biometric Society: Biometrics, Vol. 33, No. 1, pp. 159–174. Lee, M. J.; Choi, J. W.; Oh, H. J.; Won, J. S.; Park, I.; and Lee, S., 2012, Ensemble-based landslide susceptibility maps in Jinbu area, Korea: Environmental Earth Sciences, Vol. 67, No. 1, pp. 23–37. Lee, S. and Pradhan, B., 2007, Landslide hazard mapping at Selangor, Malaysia using frequency ratio and logistic regression models: Landslides, Vol. 4, No. 1, pp. 33–41. Lee, S.; Ryu, J. H.; Won, J. S.; and Park, H. J., 2004, Determination and application of the weights for landslide susceptibility mapping using an artificial neural network: Engineering Geology, Vol. 71, No. 3–4, pp. 289–302. Lee, S. and Sambath, T., 2006, Landslide susceptibility mapping in the Damrei Romel area, Cambodia using frequency ratio and logistic regression models: Environmental Geology, Vol. 50, No. 6, pp. 847–855. Leshchinsky, B. A.; Olsen, M. J.; and Tanyu, B. F., 2015, Contour connection method for automated identification and classification of landslide deposits: Computers Geosciences, Vol. 74, pp. 27–38. Li, X.; Cheng, X.; Chen, W.; Chen, G.; and Liu, S., 2015, Identification of forested landslides using lidar data, object-based image analysis, and machine learning algorithms: Remote Sensing, Vol. 7, No. 8, pp. 9705–9726. Li, Y.; Chen, G.; Tang, C.; Zhou, G.; and Zheng, L., 2012, Rainfall and earthquake-induced landslide susceptibility assessment using GIS and artificial neural network: Natural Hazards and Earth System Sciences, Vol. 12, No. 8, pp. 2719–2729. Marjanović, M.; Kovačević, M.; Bajat, B.; and Voženílek, V., 2011, Landslide susceptibility assessment using SVM machine learning algorithm: Engineering Geology, Vol. 123, No. 3, pp. 225–234. Martha, T. R.; Kerle, N.; Jetten, V.; van Westen, C. J.; and Kumar, K. V., 2010, Characterizing spectral, spatial and morphometric properties of landslides for semi-automatic detection using object-oriented methods: Geomorphology, Vol. 116, No. 1, pp. 24–36. Micheletti, N.; Foresti, L.; Robert, S.; Leuenberger, M.; Pedrazzini, A.; Jaboyedoff, M.; and Kanevski, M., 2014, Machine learning feature selection methods for landslide susceptibility mapping: Mathematical Geosciences, Vol. 46, No. 1, pp. 33–57. Mohammady, M.; Pourghasemi, H. R.; and Pradhan, B., 2012, Landslide susceptibility mapping at Golestan Province, Iran: q comparison between frequency ratio, Dempster–Shafer, and weights-of-evidence models: Journal Asian Earth Sciences, Vol. 61, pp. 221–236. Moosavi, V.; Talebi, A.; and Shirmohammadi, B., 2014, Producing a landslide inventory map using pixel-based and objectoriented approaches optimized by Taguchi method: Geomorphology, Vol. 204, pp. 646–656. Neuhäuser, B. and Terhorst, B., 2007, Landslide susceptibility assessment using “weights-of-evidence” applied to a study area at the Jurassic escarpment (SW-Germany): Geomorphology, Vol. 86, No. 1, pp. 12–24. Nourani, V.; Pradhan, B.; Ghaffari, H.; and Sharifi, S. S., 2014, Landslide susceptibility mapping at Zonouz Plain, Iran using genetic programming and comparison with frequency ratio, logistic regression, and artificial neural network models: Natural Hazards, Vol. 71, No. 1, pp. 523–547. Ozdemir, A. and Altural, T., 2013, A comparative study of frequency ratio, weights of evidence and logistic regression

424

methods for landslide susceptibility mapping: Sultan Mountains, SW Turkey: Journal Asian Earth Sciences, Vol. 64, pp. 180–197. Pourghasemi, H. R.; Moradi, H. R.; Fatemi Aghda, S. M.; Sezer, E. A.; Goli Jirandeh, A.; and Pradhan, B., 2014, Assessment of fractal dimension and geometrical characteristics of the landslides identified in north of Tehran, Iran: Environmental Earth Sciences, Vol. 71, No. 8, pp. 3617–3626. Pradhan, B., 2013, A comparative study on the predictive ability of the decision tree, support vector machine and neuro-fuzzy models in landslide susceptibility mapping using GIS: Computers and Geosciences, Vol. 51, pp. 350–365. Pradhan, B. and Lee, S., 2009, Regional landslide susceptibility analysis using back-propagation neural network model at Cameron Highland, Malaysia: Landslides, Vol. 7, No. 1, pp. 13–30. Ramesh, V. and Anbazhagan, S., 2015, Landslide susceptibility mapping along Kolli Hills Ghat road section (India) using frequency ratio, relative effect and fuzzy logic models: Environmental Earth Sciences, Vol. 73, No. 12, pp. 8009–8021. Reger, R. D. and Solie, D. N., 2008, Reconnaissance Interpretive Map of Permafrost, Alaska Highway Corridor, Delta Junction to Dot Lake, Alaska: Preliminary Interpretive Report 2008-3C, Alaska Division of Geological and Geophysical Surveys, Fairbanks, AK, 10 p. Reger, R. D.; Stevens, D. S. P.; and Solie, D. N., 2008, SurficialGeologic Map, Delta Junction to Dot Lake, Alaska Highway Corridor: Preliminary Interpretive Report 2008-3A, Alaska Division of Geological and Geophysical Surveys, Fairbanks, AK, 48 p., available at http://doi.org/10.14509/17961. Regmi, N. R.; Giardino, J. R.; McDonald, E. V.; and Vitek, J. D., 2014, A comparison of logistic regression-based models of susceptibility to landslides in western Colorado, USA: Landslides, Vol. 11, No. 2, pp. 247–262. Regmi, N. R.; Giardino, J. R.; and Vitek, J. D., 2010, Modeling susceptibility to landslides using the weight of evidence approach: Western Colorado, USA: Geomorphology, Vol. 115, No. 1, pp. 172–187. Samodra, G.; Chen, G.; Sartohadi, J.; and Kasama, K., 2017, Comparing data-driven landslide susceptibility models based on participatory landslide inventory mapping in Purwosari area, Yogyakarta, Java: Environmental Earth Sciences, Vol. 76, No. 4, Article 184. Son, J.; Suh, J.; and Park, H. D., 2016, GIS-based landslide susceptibility assessment in Seoul, South Korea, applying the radius of influence to frequency ratio analysis: Environmental Earth Sciences, Vol. 75, No. 4, Article 310. Story, M. and Congalton, R. G., 1986, Accuracy assessment: A user’s perspective: Photogrammetric Engineering and Remote Sensing, Vol. 52, No. 3, pp. 397–399. Su, C.; Wang, L.; Wang, X.; Huang, Z.; and Zhang, X., 2015, Mapping of rainfall-induced landslide susceptibility in Wencheng, China, using support vector machine: Natural Hazards, Vol. 76, No. 3, pp. 1759–1779. Tsangaratos, P. and Benardos, A., 2014, Estimating landslide susceptibility through an artificial neural network classifier: Natural Hazards, Vol. 74, No. 3, pp. 1489–1516. Van Den Eeckhaut, M.; Kerle, N.; Poesen, J.; and Hervás, J., 2012, Object-oriented identification of forested landslides with derivatives of single pulse LiDAR data: Geomorphology, Vol. 173, pp. 30–42. Van Westen, C. J.; Rengers, N.; and Soeters, R., 2003, Use of geomorphological information in indirect landslide susceptibility assessment: Natural Hazards, Vol. 30, No. 3, pp. 399– 419.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

Landslide Mapping Using LiDAR Wahrhaftig, C., 1965. Physiographic Divisions of Alaska: Geological Survey Professional Paper 482, 52 p. Wahrhaftig, C., 1970, Geologic Map of the Healy D-5 Quadrangle, Alaska: U.S. Geological Survey Geologic Quadrangle Map 804. Wahrhaftig, C. and Black, R. F., 1958, Quaternary Geology of the Nenana River Valley and Adjacent Parts of the Alaska Range: U.S. Geological Survey Professional Paper 293, 78 p. Weber, F. R.; Wheeler, K. L.; Rinehart, C. D.; Chapman, R. M.; and Blodgett, R. B., 1992, Geologic Map of the Livengood Quadrangle, Alaska: U.S. Geological Survey Open-File Report 92-562, 20 p. Wiedmer, M.; Montgomery, D. R.; Gillespie, A. R.; and Greenberg, H., 2010, Late Quaternary megaﬂoods from Glacial Lake Atna, Southcentral Alaska, U.S.A.: Quaternary Research, Vol. 73, No. 3, pp. 413–424. Wilson, F. H.; Hults, C. P.; Mull, C. G.; and Karl, S. M., 2015, Geologic Map of Alaska: U.S. Geological Survey Scientiﬁc Investigations Map SIM 3340. Wu, X.; Ren, F.; and Niu, R., 2014, Landslide susceptibility assessment using object mapping units, decision tree, and support vector machine models in the Three Gorges of China: Environmental Earth Sciences, Vol. 71, No. 11, pp. 4725–4738. Xu, C.; Dai, F.; Xu, X.; and Lee, Y. H., 2012. GIS-based support vector machine modeling of earthquake-triggered landslide susceptibility in the Jianjiang River watershed, China: Geomorphology, Vol. 145, pp. 70–80. Xu, C.; Shen, L.; and Wang, G., 2016, Soft computing in assessment of earthquake-triggered landslide susceptibility: Environmental Earth Sciences, Vol. 75, No. 9, Article 767.

Xu, C.; Xu, X.; Dai, F.; Wu, Z.; He, H.; Shi, F.; Wu, X.; and Xu, S., 2013, Application of an incomplete landslide inventory, logistic regression model and its validation for landslide susceptibility mapping related to the May 12, 2008 Wenchuan earthquake of China: Natural Hazards, Vol. 68, No. 2, pp. 883– 900. Yao, X.; Tham, L. G.; and Dai, F. C. 2008, Landslide susceptibility mapping based on support vector machine: A case study on natural slopes of Hong Kong, China: Geomorphology, Vol. 101, No. 4, pp. 572–582. Yilmaz, I., 2009, Landslide susceptibility mapping using frequency ratio, logistic regression, artiﬁcial neural networks and their comparison: a case study from Kat landslides (Tokat— Turkey): Computers and Geosciences, Vol. 35, No. 6, pp. 1125– 1138. Youssef, A. M.; Pradhan, B.; Pourghasemi, H. R.; and Abdullahi, S., 2015, Landslide susceptibility assessment at Wadi Jawrah Basin, Jizan region, Saudi Arabia using two bivariate models in GIS: Geosciences Journal, Vol. 19, No. 3, pp. 449– 469. Zhang, M.; Cao, X.; Peng, L.; and Niu, R., 2016a, Landslide susceptibility mapping based on global and local logistic regression models in Three Gorges Reservoir area, China: Environmental Earth Sciences, Vol. 75, No. 11, Article 958. Zhang, Z.; Yang, F.; Chen, H.; Wu, Y.; Li, T.; Li, W.; Wang, Q.; and Liu, P., 2016b, GIS-based landslide susceptibility analysis using frequency ratio and evidential belief function models: Environmental Earth Sciences, Vol. 75, No. 11, Article 948.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 405–425

425

Evaluating the Use of Unmanned Aerial Systems (UAS) for Collecting Discontinuity Orientation Data for Rock Slope Stability Analysis RACHAEL K. DELANEY Resource Development Consultants Ltd., Unit 2, 182 Main Road, Tawa, Wellington, 5249, New Zealand

ABDUL SHAKOOR* Department of Geology, Kent State University, Kent, OH 44240

CHESTER F. WATTS Geohazards & Unmanned Systems Research Center, Radford University, Radford, VA 24142

Key Terms: Unmanned Aerial System (UAS), Unmanned Aerial Vehicle (UAV), Transit Compass, Discontinuity Orientation, Stereonet Analysis, Principal Discontinuity Sets, Kinematic Analysis, Ground Control Points

sion lines at Site 1, and biases in human versus artiﬁcial interpretations. Compensating for some identiﬁed factors is now simpler by adding surveyed ground-control points to the model. Representative compass data should always be acquired for quality assurance.

ABSTRACT INTRODUCTION This study compared the reliability of discontinuity orientation data collected in 2016 using unmanned aerial systems (UAS) with traditional transit compass data as a control, to evaluate UAS-generated results for rock slope stability analysis. The 2016 UAS operations were primitive by 2020 standards. Lessons learned are reviewed and related to UAS procedures common today. Two sites in Virginia were selected: a cut slope along State Route 629 (Site 1) and an abandoned shale quarry below Cove Mountain (Site 2). For logistical reasons, a different UAS was used at each of the two sites. Overlapping images from UAS flights were used to create point clouds from which discontinuity orientation data were extracted. Discontinuity data from both UAS and transit compass methods were imported into Dips 7.0, RocPlane 3.0, and Swedge 6.0 software for stereonet, statistical, and kinematic analyses. Statistical evaluation of the data sets suggested better overall reliability of UAS results for Site 2 than for Site 1. Compared to transit compass data, Site 1 UAS results are reliable for plane failure analysis only, whereas Site 2 UAS results are reliable for analyzing all types of failure. Possible reasons for this include different drone systems used at the two sites with different navigation and camera characteristics, interference from high-voltage electrical transmis*Corresponding author email: ashakoor@kent.edu

Unmanned aerial system (UAS) remote sensing is an emerging technology for the collection and characterization of geologic structure data in hardto-access sites. In this study, photogrammetry alone was employed as the UAS remote-sensing method. Digital point clouds of rock masses were generated from dozens of overlapping images and treated as virtual rock slopes from which structural data were extracted. Drone-based light detection and ranging (LiDAR) is another option, although far more expensive, for generating similar point cloud models. Each has its own advantages and disadvantages. UAS refers to not only the unmanned aerial vehicle (UAV, or drone), but more completely to the entire ﬂight system. The system includes sensors like cameras, software applications for controlling or assisting ﬂight, the remote controller, and, especially, the human operator. This is all in addition to the UAV itself. Photogrammetry works on the principle that locations can be determined in three-dimensional (3D) space by making measurements of features on overlapping photographs (McCarthy, 2014). Current software, such as Pix4Dmapper Pro (Pix4D, 2018), can automatically process overlapping photographs to rapidly extract the 3D coordinates of millions of surface points to create a point cloud model (Bemis et al., 2014). In comparison to traditional photogrammetry,

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427–447

427

Delaney, Shakoor, and Watts

which requires previous knowledge of camera positions, structure-from-motion (SfM) is a photogrammetric technique that automatically solves for original 3D camera positioning (Vasuki et al., 2014). Once the point cloud model created from UAS data is available and colorized, if desired, using reference field-site images, models can be used to identify discontinuities present in rock faces of interest, either manually or through programmed algorithms (Vasuki et al., 2014). Orientations of identified discontinuities can then be used to evaluate potential modes of failure for the slope. The traditional method used to accurately measure discontinuity orientations is the transit compass. Therefore, orientation data obtained by UAS were compared here to transit compass data as the standard, in order to evaluate the reliability of UAS methods for acquiring structural data. Two commonly used procedures for collecting orientation data by transit compass are line mapping and window mapping (Wyllie and Mah, 2004). Line mapping involves placing a 100 ft (30 m) measuring tape along the slope face and recording data (orientation, spacing, continuity, surface irregularities, infilling material, groundwater conditions) for every discontinuity that intersects the tape. Window mapping involves delineating “windows” along the slope face, with typical dimensions for windows being 5 ft (1.5 m) × 25 ft (8 m) and with an inter-window distance of 10 ft (3 m) to 25 ft (8 m). All discontinuities that fall within a window are measured. A major limitation of the transit compass method is that it can only be used on accessible portions of the slope. The UAS method of discontinuity data collection has the benefit of easily accessing portions of a slope that may be difficult or impossible to access physically. Given that UAS photogrammetric surveys can cover the entire rock slope, they provide more comprehensive data than field scan lines close to the slope toe. This advantage is particularly important in the case of irregularly jointed rock masses, where discontinuity orientations measured along a scan line at the toe do not necessarily reflect the discontinuity orientations higher up the slope. However, remotesensing techniques do not work well in rock masses that are, in general, characterized by a low geological strength index (GSI) and/or that lack rock slope surface relief (Sturzenegger and Stead, 2009). Clearly, UAS methods work best for rock slopes in competent rock where stability is controlled by discontinuities from geologic structure. A general flight plan for a UAS photogrammetry survey requires images that overlap in terms of slope coverage and angular change between overlapping images (Bemis et al., 2014).

428

STUDY OBJECTIVES The main objectives of this study were to: 1. Compare discontinuity orientation data obtained by the UAS photogrammetry method to data obtained by traditional transit compass methods, using graphical and statistical stereonet analyses, for two sites in Virginia; and 2. Perform kinematic analyses, using data gathered by the two methods, to determine the potential modes of failure at both sites and determine the factor of safety (FS) values for potential failures. METHODS Site Selection Two sites in Virginia, designated as Site 1 and Site 2, were selected for the study (Figure 1). Site 1 is a cut slope along State Route 629, near Deerﬁeld in Augusta County. This site experienced a rockslide in 2009, involving 10,000 cubic yards (7,645 cubic meters) of rock material and resulting in road closure for 3 months (Niemann, 2013). Site 2 is an abandoned shale pit at the base of Cove Mountain in Wythe County. The criteria used for selecting the two study sites were: (a) sites were accessible for collecting discontinuity data by transit compass, (b) sites exhibited well-developed discontinuities, and (c) permission was granted for accessing the sites. Site Geology The cut slope at Site 1 consists of dark-gray shale belonging to the Brallier Formation of Devonian age, which is a series of interbedded shale, siltstone, and minor micaceous sandstone (Rader and Wilkes, 2001). These strata are folded around the axes of an anticlinesyncline pair (Niemann, 2013). The most obvious discontinuities present at the site are bedding planes dipping roughly 25° to 35° towards the southeast and near-vertical orthogonal joints dipping to the northeast and northwest (Figure 2a). The beds range in thickness from 2 in. (5 cm) to 6 in. (15 cm) and are heavily jointed. Discontinuities are generally tight and smooth-surfaced, although a few have an aperture >0.4 in. (1cm). Rocks of the Devonian Millboro Shale outcrop at Site 2 (Webb, 1965). The black shale beds at the site are intensely laminated, with individual beds up to 8 in. (20 cm) thick and lamination within the beds about 0.4 in. to 1.2 in. (1 mm to 3 cm) thick. Bedding is well developed, dipping at 30° to 45° to the southeast (Figure 2b). The orthogonal joints, although not as well developed as at Site 1, dip steeply

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427–447

UAS for Collecting Rock Slope Discontinuity Data

Figure 1. General locations of the two study sites in Virginia, with inset maps showing the corresponding topography (composite from d-maps.com, 2018).

to the northwest and southwest. Generally, the discontinuities are tight and smooth-surfaced. Field Investigations For logistical reasons, different UAV models were used to obtain imagery at each of the two sites. The use of different UAVs allowed for a comparison of the capabilities, advantages, and disadvantages of each system. Specifics are described below. At Site 1, a DJI “Phantom 3 Pro” with the stock camera was used. This UAV model is lightweight, includes a 3-axis gimbal for camera stabilization, and has a maximum flight time of approximately 23 minutes per battery pack (DJI Official, 2017). Due to Federal Aviation Administration (FAA) and Virginia Department of Transportation (VDOT) regulations in effect

at the time, a third-party licensed pilot was required to fly the UAV during 15 minute temporary road closures at this site. A spotter was used to maintain proper distance from nearby power lines and guy wires. Data were collected mostly for the bottom half of the slope face (because of its accessibility for comparison data collected by transit compass), taking special care to scan the cut slope obliquely at varying angles. At Site 2, the DJI “Phantom 3” was no longer available, and so a 3D Robotics “Solo” was used. This UAV had a GoPro Hero 4 camera with 3-axis gimbal and a maximum flight time of approximately 16 minutes per battery pack (Anderson, 2015). Because of the remote nature of Site 2, co-author Chester Watts was able to fly the drone under FAA guidelines for educational use. The focus of data collection was again oblique imagery of the bottom portion of the slope that

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427–447

429

Delaney, Shakoor, and Watts

Figure 2. (a) Close-up view of the right portion of the slope at Site 1 focusing on an individual bed with red marking the bedding and blue and green marking the orthogonal joints. (b) Close-up view of the slope face at Site 2 with red stripes representing the bedding and blue and green representing the orthogonal joints.

430

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427â&#x20AC;&#x201C;447

UAS for Collecting Rock Slope Discontinuity Data

contained repeating orthogonal joints and bedding planes that could be measured easily with a transit compass. A transit compass was used to collect discontinuity orientation data at both sites to provide a set of control data, for comparison with the UAS photogrammetry results, and to evaluate UAS accuracy and reliability. Typically, either the detailed line survey mapping method (Piteau and Martin, 1977) or the window mapping method is used to collect discontinuity data with a transit compass. For this study, however, it was decided that a more thorough mapping method was required in order to fully characterize the orientations of bedding and joints present at both sites. In order to distinguish between joint sets, discontinuities were visually separated by dip direction and dip angle. Discontinuity faces of similar orientation were assigned a unique number. Specifically, all bedding was designated as “1,” the first joint set was designated as “2,” the next joint set was designated as “3,” and so on. Chalk was used to mark the bedding and joint sets accordingly (Figure 2). A transit compass was used to measure each marked discontinuity. Measurements were categorized by bearing or dip direction, dip angle, discontinuity type number, and general slope face region (right side, right middle, left middle, and left side). Reference sketches of slope face regions were drawn at the time of data collection. This method of bookkeeping allowed for data to be easily distinguished later during processing and analysis. In addition to transit compass readings, the iPhone app GeoID (Sang Ho Lee, 2014) was used to collect discontinuity data. The accuracy of the GeoID app was tested frequently by comparing several measurements to measurements made by the transit compass. Using these methods, 395 discontinuities were measured at Site 1, and 232 discontinuities were measured at Site 2. The difference in sample size is due to the smaller size of Site 2. At both field sites, parameters such as rock type(s) and discontinuity characteristics relevant to slope stability (surface roughness, aperture, nature of infilling material, groundwater conditions) were recorded. Slope orientations were measured at several locations to obtain average values. Three representative rock block samples were taken at each site for determining dry density in the laboratory. Data Analysis UAS images were first processed using Pix4Dmapper software (Pix4D, 2018). Image analysis software seeks key points in each image that might also be found in other images. When key points are matched to multiple images, they become tie points and are assigned x-y-z coordinates, forming a sparsely popu-

lated point cloud. Interpolation between tie points allows a dense point cloud to be generated. The process resulted in a 3D model of points in space, where each point had color attributes and was tied to camera position (McCarthy, 2014). The 3D models generated from the UAS images were rotated to their correct compass orientations, only if necessary, using easily identifiable features in the UAS images that were also visible in online georeferenced databases from other dates. One example is the concrete block at the Deerfield site that anchors the wire stays for the power transmission line tower. The point clouds for the two sites were imported into Split-FX software (Split Engineering, 2014) to generate a triangulated irregular network (TIN) (or mesh) of connected triangles. Neighboring triangles were compared and combined to form patches if they were of similar orientation. A patch is a polygon representing the plane of the surface on which it was drawn. Split-FX software automatically determines the dip direction, dip, and surface data of patches within point clouds (Fisher et al., 2014). Lato et al. (2009) recommended setting the number of data points per triangle as 10 to 12, and the angle difference as 3° to 5°, when forming patches within the triangle mesh. However, automated patch generation resulted in seemingly erroneous discontinuity surfaces in this study. Hence, the option to manually outline discontinuity patches within the point clouds with a mouse was used. Those patches more accurately matched field observations and were then used to extract discontinuity orientation data. Patch size was not deemed relevant for comparison of data collection methods. Discontinuity orientation data from both the UAS and transit compass methods were imported into the Dips 7.0 program (ROCSCIENCE, 2018a) to generate contoured pole plots. Dips 7.0 automatically assigned contour intervals for each data set. The counting circle size was set to 1.00°. Principal discontinuity sets (PDS), indicated by closely spaced contours around pole concentrations, were identified using the cluster analysis tool. This tool is based on the fuzzy K-means algorithm, which “basically seeks for regions of high density in data” (ROCSCIENCE, 2018f). The centroid of each cluster is the vector most representative of the cluster, i.e., the geometric mean of all vectors belonging to that cluster (Hammah and Curran, 1998). ROCSCIENCE (2018a) reports that, “a useful rule of thumb is that any cluster with a maximum concentration of greater than 6% is very significant. Four to six percent represents a marginally significant cluster. Less than 4% should be regarded with suspicion unless the overall quantity of data is very high (several hundreds of poles).” Following this rule of thumb, the PDS were determined by manually selecting the centers of contoured pole clusters within a maximum radius. In

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427–447

431

Delaney, Shakoor, and Watts

this study, the radius ranged from 6° to 25°, as necessary, in order to capture only data with a maximum concentration of 4 percent or greater. Statistical analysis included 95 percent variability cone plots, 95 percent confidence cone plots, and the Fisher’s dispersion coefficient (Fisher’s K) values for each of the two data sets. Dips software generated variability cones and confidence cones based on the PDS of each data set. The variability value reflects the natural variability of the data, assuming the mean is correct. For example, a 24 percent variability indicates that any pole, selected from the full population of data, has a 24 percent probability of falling within the variability cone (ROCSCIENCE, 2018e). A higher variability value indicates a higher degree of scatter for that set. The confidence limit is an angular value that reflects confidence in the location of the true PDS mean within a specified degree of certainty (95 percent in this study) (ROCSCIENCE, 2018e). The equations for the confidence and variability limits are as follows: Variability limit angle: cos (α) = 1.0 + ln (1 − P) / (K ) .

Conﬁdence limit angle: cos (α) = 1.0 + ln (1 − P) / (RK ) .

(1)

RESULTS Statistical Analysis of Discontinuity Orientation Data

(2)

For variability, P is the probability that a vector selected at random makes an angle theta with the calculated mean. For confidence, P is the probability that the calculated mean is within theta of the true population mean. P ranges from 0 to 1 (i.e., 0 percent to 100 percent). K (Fisher’s constant) = (N − 1)/(N − R), where N is the total length of pole vectors in a set, and R is the length of the resultant vector (upon vector addition of all poles in the set) (ROCSCIENCE, 2018d). The K value describes the “tightness” of a cluster of orientation data points. A larger K value indicates a tighter cluster of points, whereas a smaller K value indicates a greater level of data dispersion away from the mean orientation of the cluster (Fisher, 1953). Kinematic Analysis and FS Calculations Using the PDS and Dips software, kinematic analysis was performed to determine modes of rock slope failure for the two sites. The purpose was to check the extent to which any differences in discontinuity orientation influenced the results of kinematic analysis. The Dips software uses poles from the PDS for analysis of plane failure and flexural toppling failure, but it uses the intersections of the PDS great circles for analysis of wedge failure and a combination of PDS poles and great circle intersections for analysis of direct/oblique

432

toppling failure. Other input parameters for kinematic analysis included the average slope orientation and the basic friction angle. For the rock types exposed at the study sites and considering that most joints are smooth or slickensided, a basic friction angle of 27° was assumed for simple kinematic analysis (Wyllie and Mah, 2004; West and Shakoor, 2018). Computer programs RocPlane (ROCSCIENCE, 2018b) and Swedge (ROCSCIENCE, 2018c) were used to compute FS values against plane and wedge failures, respectively, for both Sites 1 and 2. Input parameters for FS calculations included the mean orientation values for the relevant principal discontinuities and their intersections, the assumed basic friction angle, the slope heights, slope angles, and the rock dry density values measured on core samples obtained from rock blocks collected in the ﬁeld. FS calculations for direct/oblique and ﬂexural toppling were not carried out due to lack of required input parameters such as the base inclinations of the rock blocks and the widths of potential toppling blocks.

Principal Discontinuity Sets In order to avoid under-representation of some discontinuities, identical repetitive bedding measurements, typical of dominant discontinuities, were deleted for both sites. Consequently, the statistical analysis involved 243 UAS data points for Site 1, 211 UAS data points for Site 2, 362 transit compass data points for Site 1, and 193 transit compass data points for Site 2. The transit compass data served as control data for evaluating the reliability of the UAS data collection method. The contoured stereonet plots of UAS data indicate three PDS for Site 1 and four PDS for Site 2 (Figure 3). For transit compass data, the stereonet plots indicate four PDS for Site 1 and three PDS for Site 2 (Figure 4). Table 1 summarizes the mean dip and dip direction values for the PDS for both sites. It should be noted that although joints were numbered numerically for ﬁeld measurements, they are labelled for analysis purposes as A, B, and C on the stereoplots and in the tables. Variability Cone Plots Table 2 provides a summary of statistical analysis results (variability limits, conﬁdence limits, Fisher’s K values) of the PDS for the two sites. Figure 5 shows the PDS variability cone plots for the UAS data from Sites 1 and Site 2, whereas Figure 6 shows the same plots

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427–447

UAS for Collecting Rock Slope Discontinuity Data

Figure 3. Pole plots and principal discontinuity sets (PDS) indicated by UAS data for (a) Site 1 and (b) Site 2. Warmer colors denote higherdensity concentrations, with a maximum density of 22 percent for Site 1 and 12 percent for Site 2. The plots indicate three PDS for Site 1 and four PDS for Site 2 (bounded by dark red) along with their corresponding great circles.

for the transit compass data. In these plots, a higher degree of scatter in a given PDS results in a larger area of the variability cone. For Site 2 UAS data, joint sets A, B, and C, with overlapping cones, show a higher de-

gree of scatter (Figure 5b). The variability limits range from 5.5° to 21.8° (Table 2). The variability limits indicate how high the degree of variability is for each principal discontinuity. For example, if a discontinu-

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427–447

433

Delaney, Shakoor, and Watts

Figure 4. Pole plots and principal discontinuity sets (PDS) indicated by transit compass data for (a) Site 1 and (b) Site 2. Warmer colors denote higher-density concentrations, with a maximum density of 10 percent for Site 1 and 16 percent for Site 2. The plots indicate four PDS for Site 1 and three PDS for Site 2 (bounded by dark red) along with their corresponding great circles.

434

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427â&#x20AC;&#x201C;447

UAS for Collecting Rock Slope Discontinuity Data

Figure 5. Variability cone plots for principal discontinuities derived from UAS data for (a) Site 1 and (b) Site 2. Dark blue indicates variability cones, and dark red indicates principal discontinuities. Notice the overlap between variability cones for joint sets A, B, and C for Site 2.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427â&#x20AC;&#x201C;447

435

Delaney, Shakoor, and Watts

Figure 6. Variability cone plots for principal discontinuities derived from transit compass data for (a) Site 1 and (b) Site 2. Dark blue indicates variability cones, and dark red indicates principal discontinuities.

436

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427â&#x20AC;&#x201C;447

UAS for Collecting Rock Slope Discontinuity Data Table 1. Mean dip and dip direction values of the principal discontinuities for both sites. Site 1

Data Collection Method

Principal Discontinuity

Dip (degrees)

Dip Direction (degrees)

UAS photogrammetry

Bedding Joint set A Joint set B Bedding Joint set A Joint set B Joint set C Bedding Joint set A Joint set B Joint set C Bedding Joint set A Joint set B

28.7 83.3 88.5 26.3 88.9 88.0 81.8 35.9 49.8 84.5 72.2 37.1 55.3 78.8

119.0 29.6 328.5 128.0 37.0 335.2 269.4 164.5 283.6 246.6 269.6 163.9 299.7 247.6

Transit compass

UAS photogrammetry

Transit compass

ity has a 95 percent variability limit of 10°, orientation values of data points in that discontinuity can differ from the mean discontinuity orientation by up to 10°. Confidence Cone Plots Confidence cones are conical projections of the confidence limit values centered about the mean orientation values (ROCSCIENCE, 2018f). The confidence limits indicate where the mean orientation of the center of a discontinuity lies with 95 percent certainty. For example, a confidence limit of 5° indicates that the mean orientation of a discontinuity center is within 5°. A lower confidence limit indicates a higher degree of precision. The 95 percent confidence limits range from 0.8° to 3.8° for both types of data for the two sites (Table 2). The confidence cone plots were overlaid to create summary plots for Sites 1 and 2 (Figure 7).

These overlays were used to check whether the data sets from each site were derived from the same data population. If the confidence cones overlap each other, the corresponding discontinuity sets can be statistically considered as originating from the same data population (i.e., the two sets represent the same discontinuity). In Figure 7b, there are two instances of cone overlapping, bedding confidence cones and joint set B confidence cones. Statistical Comparison of UAS and Transit Compass Methods Table 1 shows that UAS data reveal three principal discontinuities (bedding and joint sets A and B) for Site 1, whereas the transit compass data reveal the presence of four principal discontinuities (bedding and joint sets A, B, and C). The UAS data contour plot

Table 2. Statistical parameters of the principal discontinuities for both sites. Site 1

Data Collection Method

Principal Discontinuity

Variability Limit*

Conﬁdence Limit*

Fisher’s K

n†

UAS photogrammetry

Bedding Joint set A Joint set B Bedding Joint set A Joint set B Joint set C Bedding Joint set A Joint set B Joint set C Bedding Joint set A Joint set B

7.4° 5.5° 8.4° 9.2° 8.5° 16.2° 12.9° 8.8° 9.4° 21.8° 12.9° 11.8° 12.4° 20.7°

1.0° 0.8° 2.2° 1.4° 1.4° 1.9° 2.4° 1.6° 2.2° 3.4° 2.2° 1.8° 2.3° 3.8°

372.4 665.8 288.3 241.6 280.0 77.7 121.5 263.6 227.8 43.2 122.1 146.5 131.4 47.6

59 49 15 43 35 71 29 31 19 42 34 45 29 30

Transit compass

UAS photogrammetry

Transit compass

* †

95 percent limits. n = population.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427–447

437

Delaney, Shakoor, and Watts

two methods for Site 1 indicate that joint set B in the transit compass data exhibits higher scatter than the other principal discontinuities (Figures 5 and 6). The confidence cone plot for Site 1 (Figure 7a) shows that no centers for any principal discontinuity from the two methods touch or intersect and thus statistically have a poor correlation between the two methods. The Fisher’s K values are mostly high enough to suggest that the principal discontinuities represented by these data sets have tight clusters, and their identification should be reliable (Table 2). Contrary to Site 1, UAS data for Site 2 reveal the presence of four principal discontinuities (bedding plus joint sets A, B, and C) as shown in Table 1, while the transit compass data reveals three principal discontinuities (bedding and joint sets A and B). It is likely that joint set C was not well sampled using the transit compass, and any data points that would belong to joint set C were instead included in joint set B. Joint set C has a higher sampling rate in the UAS data, and it was therefore identified as a separate discontinuity. The differences in mean orientations of bedding and joint set A between the two data sets for Site 2 are smaller than those for Site 1 (Table 1). The variability cone plots for Site 2 (Figures 5 and 6) show that joint set B has a greater amount of scatter for both data sets. The confidence cone plot for Site 2 (Figure 7b) shows that the centers for bedding and joint set B intersect for the two data sets, but the centers for bedding and joint set A do not. This indicates that, statistically, there is a reasonable correlation between the two data collection methods for bedding and for joint set B. The same cannot be said for joint set A. Thus, the UAS results correctly match for bedding and joint set B, but they do not correctly match for joint set A, as the dip direction confidence level is significantly lower. The Fisher’s K values for the principal discontinuities in these data sets are high except for joint set B, indicating bedding and joint set A have tight clusters (Table 2). Thus, the identification of bedding and joint set A is more reliable than that of joint set B. Additional Comparisons of UAS and Transit Compass Methods Based on Stereonets Figure 7. Confidence cone plots for (a) Site 1 and (b) Site 2. Darkblue cones represent UAS data, and green cones represent the transit compass data.

for Site 1 shows no evidence of joint set C (Figure 3). Table 1 shows signiﬁcant diﬀerences in the joint dip and dip direction values for Site 1, suggesting that the UAS data did not suitably identify all principal discontinuities at Site 1. Variability cone plots for the

438

A visual comparison of the stereonet plots for UAS and compass-derived data provides insight into the statistical results, especially within the context of speciﬁc UAS capabilities and individual ﬂight operations. For example, aligning the stereonet pole plots (Figures 3 and 4) on a light table, one on top of the other, provides for direct comparison of results by eye, and for relating those observations to statistical anal-

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427–447

UAS for Collecting Rock Slope Discontinuity Data

yses, as well as to site and aircraft characteristics, as follows:

1. Figures 2a and 2b show that both sites are characterized by a mix of discontinuities. Some discontinuities dominate stability by providing actual slip surfaces, while others assist movement by providing release surfaces. Bedding planes provide the slip surfaces at both sites. They are larger and better exposed than the joint sets. In contrast, joint set exposures at these sites are smaller, produce irregular stairstep patterns, and act as release surfaces around the edges of potential slide blocks. During data collection, the bedding plane exposures are more obvious and easier to measure than the joint sets, both in the real world and in the virtual world of a digital point cloud. Hence, it is not surprising that a visual comparison of stereonets, representing data from both worlds, confirms that the best correlation exists for the well-exposed bedding plane surfaces and that poorer correlations exist for the smaller, more irregular joint set surfaces. 2. Additional influences are at work depending on the data collection methods used. Some are well known, such as inaccuracies introduced when a compass is in close proximity to large metal objects. Lesser-known influences include those inherent to SfM algorithms used to create point clouds from overlapping imagery. For example, the reliability of geotagging aerial images will have essentially no impact on extracted dip angle values but can have a major impact on extracted dip direction values. The effects of this are illustrated in Table 1, which provides a listing of mean values for dip angles and dip directions for discontinuity sets. 3. Bedding planes at Site 1 show a mean dip angle difference of only 2.4° between UAS and compass data. For Site 2, the mean dip angle difference for bedding planes is even less, at just 1.2°. This is a remarkable correlation considering that the accuracy of the transit compass, the standard for these measurements, is judged to be about ±2.0°. The joint set discontinuities, however, show more variability when comparing dip angles, although this is not unexpected because they are more difficult to measure. In a comparison of UAS to compass dip angle results for joint sets A and B, at both sites, variations ranged from as small as 0.5° to as large as 5.7°. In the real world of rock slope engineering, which is heavily dependent on measurements of geologic features that are not perfect, this range of dip angle values for the irregular joints is not unreasonable. It should further be noted that the mean values are being compared here, not values for specific individual

discontinuities. As always, sound engineering judgement must be applied. 4. Reliability of mean dip directions, however, is not so simple. Unlike dip angle values, dip direction values extracted from models generated by SfM software are highly dependent on global positioning system (GPS) readings embedded in the aerial images. Therefore, poor, or absent, satellite connections can negatively impact dip direction values, unless compensated for by including ground-control points (GCPs) into the digital model. At these two different sites with two different aircraft systems, the following results occurred. From the statistics of Table 1, the difference in mean bedding plane dip direction readings between UAS and compass data was 9.0° at Site 1 and 0.6° at Site 2. The difference in mean joint set dip direction readings between UAS and compass data at Site 1 was 7.4° for joint set A and 6.7° for joint set B. At Site 2, the difference in mean joint set dip direction readings was 16.1° for joint set A and 1.0° for joint set B. 5. Comments on the above observations: a. Experience and skill. Over time, rock slope experts develop the ability to recognize rock mass discontinuities that control stability in the field and can translate that ability into recognizing those discontinuities in virtual digital models. This leads to higher correlations between UAS and compass-derived data. b. Nature of exposed discontinuities. The size and inherent irregularities of rock mass discontinuities make some patches easier to identify, enabling greater consistency of measurement. c. Magnetic environment. The magnetic declination at sites, along with any sources of magnetic interference, will impact transit compass readings and should be evaluated for effect. d. Accuracy of GPS readings. Latitude and longitude readings embedded in the UAS images (typically from simultaneous connections to 10 to 12 satellites during geotagging of the images) impact the dip direction values extracted from UAS point clouds, but they have little to no impact on dip angle values. Latitude and longitude readings embedded in the UAS images (typical satellite range of 10 to 12 during geotagging of images in the air) impact the dip direction values extracted from UAS point clouds, but they have little to no impact on dip angle values. This phenomenon results from the fact that discontinuity dip angles are independent of the orientation of the point cloud about its vertical axis, or even of its size and location. Dip direction values, however, depend greatly on having a properly oriented point cloud,

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427–447

439

Delaney, Shakoor, and Watts Table 3. Results of kinematic analysis for plane failure considering principal discontinuity set poles.

Site

Data Collection Method

1 2 1 2

UAS photogrammetry Transit compass

Number of PDS Poles in the Critical Zone Bedding Bedding Bedding Bedding

which is best obtained by having accurate GPS readings. e. Ground-control points (GCPs). The inclusion of surveyed GCPs in the SfM modeling can compensate for the previous two factors, resulting in significant improvements in correlation between UAS and compass-derived data. SfM software provides simple methods for incorporating GCPs into 3D models whenever an accurately surveyed point is visible in a minimum of three images and for a minimum of three points across the project. Image pixels of each survey point are assigned the coordinates, and the entire 3D model is adjusted accordingly. Root mean square error analyses can then be performed within the software to check on the amount of deviation remaining in the model. Kinematic Stability Analysis Results Using Dips software (ROCSCIENCE, 2018a) and PDS data from UAS and transit compass methods, kinematic analyses were performed to evaluate the potential for plane failure, wedge failure, direct/oblique toppling, and flexural toppling for each of the two sites. Dips presents the results of kinematic analyses for plane failure and flexural toppling as a percentage of poles from each principal discontinuity that fall within the critical zone and the results for wedge failure and direct/oblique toppling as a percentage of the total intersections of principal discontinuities falling within the critical zone. The percentages represent the probability of various modes of failure indicated by the data set (not the probability of actual failure) (ROCSCIENCE, 2018h). For the sake of brevity, stereonet plots are shown here for only plane and wedge failures. The plots for direct/oblique toppling and flexural toppling can be found in Delaney (2019). Figure 8 shows the kinematic analysis stereonet plots for plane failure using principal discontinuities derived from UAS data, and Figure 9 shows similar plots for the transit compass data. The results, summarized in Table 3, show that both UAS and transit compass data indicate that a potential for plane failure exists at both sites, occurring along the bedding. The UAS data suggest failure probabilities of 66 percent

440

39 27 23 35

Total Number of Poles

Percent of PDS Poles in the Critical Zone

59 31 43 45

66 87 53 78

and 87 percent for Sites 1 and 2, respectively, compared to 53 percent and 78 percent from the transit compass data. Figures 10 and 11 present the stereonet plots for wedge failure kinematic analysis, using principal discontinuities identiﬁed from UAS and transit compass data, respectively, for both sites. Table 4 summarizes the results of the wedge failure kinematic analysis. UAS data indicate a 33 percent potential for wedge failure between bedding and joint set A at Site 1, and between bedding and joint set B at Site 2. UAS data also indicate a critical intersection between bedding and joint set C for Site 2. Transit compass data indicate no potential for wedge failure at Site 1, and a 33 percent probability of wedge failure at Site 2 between bedding and joint set B. It is important to note that if the assumed basic friction angle value of 27° is decreased even slightly, the Site 1 transit compass data would result in the intersection between bedding and joint set A falling within the primary critical zone and would indicate a 17 percent potential for wedge failure. The wedge failure critical intersections are between the bedding and a joint set in all cases, instead of between orthogonal joint sets. This is because of the steep lines of intersection of the orthogonal joints that do not daylight on the slope face (Figure 2). Although UAS data result in an extra critical intersection between bedding and joint set C, all data sets result in the same probability of wedge failure. The direct and oblique toppling kinematic analysis results indicate that both UAS and transit compass data show potential for base-plane toppling at Site 1. UAS data show no potential for direct/oblique toppling at either of the two sites, but transit compass data show a slight potential for this type of toppling at Site 1 (Delaney, 2019). Only transit compass data indicate a 93 percent probability for ﬂexural toppling at Site 1 (Delaney, 2019). Factor of Safety Results For planar sliding, UAS data resulted in FS values of 0.93 and 0.70 for Sites 1 and 2, whereas the respective values for transit compass data were 1.03 and 0.63 (Delaney, 2019). For wedge sliding, UAS data yielded FS values of 0.95 and 0.70 for Sites 1 and 2,

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427–447

UAS for Collecting Rock Slope Discontinuity Data

Figure 8. Stereonet plot of kinematic analysis for plane failure using UAS data for (a) Site 1 and (b) Site 2.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427â&#x20AC;&#x201C;447

441

Delaney, Shakoor, and Watts

Figure 9. Stereonet plot of kinematic analysis for plane failure using transit compass data for (a) Site 1 and (b) Site 2.

442

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427â&#x20AC;&#x201C;447

UAS for Collecting Rock Slope Discontinuity Data

Figure 10. Stereonet plots of kinematic analysis for wedge failure using UAS data for (a) Site 1 and (b) Site 2.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427â&#x20AC;&#x201C;447

443

Delaney, Shakoor, and Watts

Figure 11. Stereonet plots of kinematic analysis for wedge failure using transit compass data for (a) Site 1 and (b) Site 2.

444

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427â&#x20AC;&#x201C;447

UAS for Collecting Rock Slope Discontinuity Data Table 4. Results of kinematic analysis for wedge failure considering intersections of principal discontinuity great circles.

Site 1 2 1 2

Data Collection Method

Total Number of PDS Intersections

Number of Critical PDS Intersections

Percent of Total PDS Intersections That Are Critical

3 6 6 3

1 2 0 1

33 33 0 33

UAS photogrammetry Transit compass

respectively. Transit compass data indicated potential for wedge failure only for Site 2, with an FS value of 0.69 (Delaney, 2019). These results indicate that the two data sets provide relatively similar FS values. The presence of plane and wedge failures at the two sites (Figure 1) corroborates the above-stated FS values of <1. DISCUSSION Reliability of UAS Data This statistical analysis of UAS and transit compass data indicates that although some discontinuities display a higher degree of scatter, discontinuity set boundaries in both data sets are reliable. Compared to transit compass data, which indicate the presence of four PDS at Site 1, the UAS data indicate only three PDS. Additionally, there are notable differences in the mean orientation values between UAS and transit compass data sets. Therefore, the UAS method provided less reliable data, in some measures, for Site 1. Possible explanations for this discrepancy between the two data include: (a) the flight camera angles may not have been ideal for capturing all discontinuities, (b) for reasons discussed previously, the quality and orientation of the point cloud may have been poor, making discontinuity surfaces difficult to distinguish during the manual creation of patches, and (c) surface exposures for joint set C (<5 in. or 13 cm), missing from UAS data, may have been too small to be properly scanned by UAS. This likely led to under-sampling of joint set C in the UAS data for Site 1. Site 2 UAS scans resulted in a high-quality point cloud, and, therefore, the orientation data derived from the point cloud are relatively more accurate and reliable. For Site 2, the UAS method correctly identified two of three principal discontinuities derived from transit compass data. However, UAS data identified an extra discontinuity (joint set C). This is likely because joint set C was not well sampled using the transit compass, and any data points that would belong to joint set C were instead included in joint set B (Figure 4). Joint set C has a higher sampling rate in UAS data and was, thus, identified as a separate discontinuity.

Effect of the Differences in UAS and Transit Compass Data on Kinematic Analysis Any differences in principal discontinuity orientations, as identified from UAS and transit compass data, consequently affect the results of kinematic analysis. Kinematic analyses indicate that the main mode of failure at both sites is plane failure (Table 3), with sliding along bedding surfaces. Orthogonal joints serve as release surfaces for rock blocks, which then fail by planar sliding. Because of the steep lines of intersection of the orthogonal joints that do not daylight on the slope face (Figure 2), kinematic analysis plots for wedge failure, using both data sets, show critical intersections between bedding and a joint set instead of orthogonal joint sets. UAS and transit compass data indicate the potential for base-plane (bedding) toppling failure at the two sites, whereas only the transit compass data indicate the potential for flexural toppling at Site 1 (Delaney, 2019). Despite the differences in discontinuity orientations between the two data sets, Site 1 UAS data accurately evaluate the potential for plane failure. Site 2 UAS data are accurate for evaluating all modes of failure. Except for transit compass data at Site 1, FS results for plane and wedge failure for both types of data indicate slope instability. It should be noted that any decrease in the assumed friction angle would cause Site 1 transit compass data to indicate potential for wedge failure. Based on the results of this study, the use of UAS data for engineering design without backup transit compass measurements is risky unless surveyors take great care to ensure the highest quality point clouds possible. Slope feature resolution must be excellent. Patch assignment will be inaccurate if surface features are not clearly delineated. Limitations of Research The limitations of this research are as follows: 1. The assumed value of basic friction angle of 27° may not be representative of the two sites. 2. The two UAVs used in this study had different GPS and photo geotagging capabilities. The DJI

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427–447

445

Delaney, Shakoor, and Watts

Phantom 3 Pro was equipped with photo geotagging but was flown under high-voltage electrical transmission lines that may have affected the navigation system. Also, surveyed GCPs were not used for the scans. The 3DR Solo was equipped with a GoPro camera without automatic image geotagging. Hence, the images and point clouds were crudely georeferenced post-flight by synchronizing image time stamps with the aircraft flight logs. 3. As described previously for these sites, the automated patch generation in Split FX (Split Engineering, 2014) resulted in seemingly erroneous discontinuity surfaces. Hence, the option to manually outline discontinuity patches within the point clouds with a mouse was used. Those patches more accurately matched field observations and were then used to extract discontinuity orientation data.

Suggested Protocol for UAS Photogrammetry Data Collection and Use The above discussion suggests that a standard protocol should be followed when using UAS photogrammetry to record discontinuity orientation data. Adherence to several key strategies can reduce error and increase data accuracy. Based on the results of this research, the following guidelines are suggested: 1. An initial reconnaissance survey of sites should always be performed to reduce the risk of missing key geological features or important discontinuity sets. Accessible discontinuity surfaces should be identified and labeled, and a reasonable number of representative transit compass measurements should be taken. This will facilitate UAS flight planning. 2. When a UAS survey is required for large rock faces or those that cannot be accessed by foot, it would be advisable to ground truth at least a portion of the rock face by both the traditional transit compass method and UAS survey in order to validate the UAS results. 3. During UAS scanning, special attention should be paid to different portions of the slope that may contain different structural domains. To ensure accurate mapping of complex features, photos should be taken from multiple UAS positions and angles in good lighting, and the flight should be slow and smooth. If a region is of higher concern, a greater number of photos should be taken. 4. When extracting orientation data from UAS point clouds, patches should be selected that result in consistent discontinuity sampling rates to ensure proper data contouring. If manual assignment of patches is anticipated, reference field pictures, with

446

discontinuity surfaces labeled, will greatly help in maintaining a consistent sampling rate. 5. The resulting orientation data should be visually and/or statistically compared to any transit compass data that were collected. This will help to assesses whether the UAS data are accurate enough for practical application. If the UAS data do not fully characterize the site, a combination of UAS and transit compass data may be more reliable. 6. Ground-truth conﬁrmation is important for all rock mass characteristics derived from remotely sensed data. In rock slope safety and stability studies, other parameters could include discontinuity type, length, continuity, roughness, seepage, and much more. This study focused only on discontinuity orientation.

CONCLUSIONS The following conclusions can be drawn from this study: 1. Transit compass data indicate the presence of four principal discontinuity sets (PDS) for Site 1 and three PDS for Site 2; UAS data indicate three PDS for Site 1 and four PDS for Site 2. Site 1 PDS centers for UAS data are not accurate, but two of the Site 2 PDS centers are accurate. Overall, the UAS data are less reliable for Site 1 than for Site 2. 2. The UAS method of discontinuity orientation data collection may yield different results compared to the transit compass method in cases of low-quality point clouds, poor flight planning, an inability to capture discontinuities with small surface areas, and failure of software to automatically create realistic patches. Higher point cloud quality will lead to higher accuracy of the UAS-derived orientation data. 3. To ensure high-quality UAS point clouds, surveyors should follow a standard protocol that includes ground investigations, discontinuity identification and labeling, collection of representative transit compass measurements, careful planning of the UAS flights, and the inclusion of ground-control points. 4. UAS surveying has the potential to be an economical and rapid method of collecting valuable discontinuity orientation data, especially for inaccessible slope regions. However, great care must be taken, combined with ground-truth confirmation, to ensure the reliability of UAS surveys for discontinuity mapping.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427–447

UAS for Collecting Rock Slope Discontinuity Data

ACKNOWLEDGMENTS The authors would like to thank the three anonymous reviewers for their constructive comments that helped to improve the quality of this paper. We extend our deep appreciation to Dr. Elizabeth McClellan of the Radford University Geology Department, VA, for her help in editing the geologic descriptions of the two sites as well as numerous other edits. We thank Mr. George Stephenson of the Radford University Geology Department for providing support with fieldwork and maintaining the UAS drones. Dr. Martin Woodard of GeoStabilization International also provided fieldwork support. Finally, we are very grateful to Virginia Department of Transportation personnel for logistical support and for coordinating traffic control during UAS flights.

REFERENCES Anderson, C., 2015, Site Scan Drone: Electronic document, available at https://3dr.com/solo-drone/specs/ Bemis, S.; Micklethwaite, S.; Turner, D.; James, M.; Akciz, S.; Thiele, S.; and Bangash, H., 2014, Ground-based and UAVbased photogrammetry: A multi-scale, high-resolution mapping tool for structural geology and paleoseismology: Journal of Structural Geology, Vol. 69, pp. 163–178. Delaney, R. K., 2019, Using an Unmanned Aerial Vehicle (UAV) for Collecting Discontinuity Orientation Data for Slope Stability Analysis: Two Case Studies from Virginia: Unpublished M.S. Thesis, Kent State University, Kent, OH, 143 p. DJI Official, 2017, Phantom 3 Professional—Specs, FAQ, Tutorials, Downloads and DJI GO - DJI: Electronic document, available at http://www.dji.com/phantom-3-pro/info#specs d-maps.com, 2018, Free Maps, Free Blank Maps, Free Outline Maps, Free Base Maps: Electronic document, available at https://dmaps.com/ Fisher, J.; Shakoor, A.; and Watts, C., 2014, Comparing discontinuity orientation data collected by terrestrial LiDAR and transit compass methods: Engineering Geology, Vol. 181, pp. 78–92. Fisher, R., 1953, Dispersion on a sphere: Proceedings Royal Society London, Vol. A217, pp. 295–305. Hammah, R. E. and Curran, J. H., 1998, Fuzzy cluster algorithm for the automatic identiﬁcation of joint sets: International Journal of Rock Mechanics and Mining Sciences, Vol. 35, No. 7, pp. 889–905. Lato, M.; Diederichs, M. S.; Hutchinson, D. J.; and Harrap, R., 2009, Optimization of LiDAR scanning and processing for automated structural evaluation of discontinuities in rock masses: International Journal of Rock Mechanics and Mining Sciences, Vol. 46, pp. 194–199. McCarthy, J., 2014, Multi-image photogrammetry as a practical tool for cultural heritage survey and community engagement: Journal of Archaeological Science, Vol. 43, pp. 175–185.

Niemann, W. L., 2013, Use of digital photogrammetry to monitor change at selected rock slopes. In Marshall Proposal 2011-237: Virginia Center for Transportation Innovation & Research, Charlottesville, VA, p. 14. Piteau, D. R. and Martin, D. C., 1977, Description of detail line engineering geology mapping method. In Rock Slope Engineering, Part G: Federal Highway Administration Reference Manual FHWA-13-97-208, p. 29. Pix4D, 2018, Pix4DMapper Pro (version 3.1) [computer software]: Pix4D, Lausanne, Switzerland, available at pix4d.com. Rader, E. and Wilkes, G., 2001, Geologic Map of the Virginia Portion of the Staunton 30 × 60 Minute Quadrangle: Virginia Division of Mineral Resources Publication 163. Electronic map available at https://ngmdb.usgs.gov/ Prodesc/proddesc_78181.htm ROCSCIENCE, 2018a, Dips (version 7.0): Rocscience Inc., Toronto, Ontario, Canada. Electronic software, available at https://www.rocscience.com ROCSCIENCE, 2018b, RocPlane (version 3.0): Rocscience Inc., Toronto, Ontario, Canada. Electronic software, available at https://www.rocscience.com ROCSCIENCE, 2018c, Swedge (version 6.0): Rocscience Inc., Toronto, Ontario, Canada. Electronic software, available at https://www.rocscience.com ROCSCIENCE, 2018d, Knowledge base overview. Dips help topics: Rocscience Inc., Toronto, Ontario, Canada. Electronic document, available at https://www.rocscience.com/ help/dips/#t=dips%2FDips_FAQs.htm ROCSCIENCE, 2018e, Set statistics. Dips help topics: Rocscience Inc., Toronto, Ontario, Canada. Electronic document, available at https://www.rocscience.com/help/dips/#t=dips% 2FSet_Statistics.htm ROCSCIENCE, 2018f, Sets from cluster analysis. Dips help topics: Rocscience Inc., Toronto, Ontario, Canada. Electronic document, available at https://www.rocscience.com/help/dips/ #t=dips%2FSets_from_Cluster_Analysis.htm ROCSCIENCE, 2018h, Wedge sliding. Dips help topics: Rocscience Inc., Toronto, Ontario, Canada. Electronic document, available at https://www.rocscience.com/help/dips/ #t=dips%2FWedge_Sliding.htm Sang Ho Lee, 2014, GeoID (1.80) [mobile application software]: Electronic software, available at https://itunes.apple.com/us/ app/geoid/id437190196?mt = 8 Split Engineering, 2014, Split-FX Software (version 2.4) [computer software]: Electronic software, available at http://www. spliteng.com/products/split-fx-software/ Sturzenegger, M. and Stead, D., 2009, Close-range terrestrial digital photogrammetry and terrestrial laser scanning for discontinuity characterization on rock cuts: Engineering Geology, Vol. 106, pp. 163–182. Vasuki, Y.; Holden, E.; Kovesi, P.; and Micklethwaite, S., 2014, Semi-automatic mapping of geological structures using UAVbased photogrammetric data: An image analysis approach: Computers & Geosciences, Vol. 69, pp. 22–32. Webb, F., Jr., 1965, Geology of the Big Walker–Crockett Cove Area, Bland, Pulaski, and Wythe Counties, Virginia: Unpublished Ph.D. Dissertation, Department of Geology, Virginia Polytechnic Institute, Blacksburg, VA, 171 p. West, T. and Shakoor, A., 2018, Geology Applied to Engineering, 2nd ed.: Waveland Press, Long Grove, IL, 161 p. Wyllie, D. and Mah, C., 2004, Rock Slope Engineering: Civil and Mining, 4th ed.: Spon Press, New York, pp. 81, 155.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 427–447

447

UAS-Derived Surﬁcial Deformation around the Epicenter of the 2016 Mw 5.8 Pawnee, Oklahoma, USA, Earthquake OLUFEYISAYO ILESANMI* Department of Geosciences and Geological and Petroleum Engineering, Missouri University of Science and Technology, Rolla, MO 65409

XUE LIANG Boone Pickens School of Geology, Noble Research Center, Oklahoma State University, Stillwater, OK 74078

FRANCISCA E. OBOH-IKUENOBE J. DAVID ROGERS Department of Geosciences and Geological and Petroleum Engineering, Missouri University of Science and Technology, Rolla, MO 65409

MOHAMED ABDELSALAM Boone Pickens School of Geology, Noble Research Center, Oklahoma State University, Stillwater, OK 74078

JORDAN FEIGHT Unmanned Systems Research Institute, Oklahoma State University, Stillwater, OK 74078

EMITT C. WITT III ECWHydro LLC, Hydrology and Site Management, Newport, NC 28570

Key Terms: Digital Surface Model, Pawnee Earthquake, En-Echelon Joints, Surface Deformation, Point Cloud

ABSTRACT Unmanned aerial systems (UAS) provide a framework for recording perishable surficial data or information. Open fractures exhibiting regular en-echelon patterns were captured by a 12-megapixel, FL-9 mm camera attached to a Phantom IV UAS over the epicenter of the magnitude (Mw ) 5.8 earthquake of September 3, 2016, 15 months later. The Digital Surface Models (DSMs) and orthoimagery offered a spatial resolution (∼1 cm) sufficient to identify small-scale plastic deformations that appear to be controlled by en-echelon joint sets developed in the underlying formation. The fissure boundaries and intersections are remarkably linear and

*Corresponding author email: obi449@mst.edu

sharp. They appeared to have been recently formed, presumably by seismic swarms believed to have been associated with wastewater injection. The DSMs revealed a series of conjugate patterns suggestive of regional systematic joints with apparent subsidence of inﬁlling up to 50 cm. The earthquakes emanated from the Precambrian metamorphic basement, with epicentral clusters at ∼5- and 8-km depths. Low energy release from depths >1.5 km appears to be locally attenuated by an unconsolidated “soil cap,” which likely formed an impedance contrast. The maximum deformation direction from the cumulative energy of earthquakes correlates with a wrench fault tectonics model that could conceivably produce the observed en-echelon joint sets observed in the orthoimagery and DSMs. These features were observed within 275 m of the reported Mw 5.8 epicenter. The remarkably linear repeating pattern of deformation appears to express ﬁssures that preserve the wrench fault fractures generated by the Mw 5.8 earthquake emanating from discontinuity suites within marine sandstone, shale, and limestone of Pennsylvanian to Permian age.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 449–462

449

Ilesanmi, Liang, Oboh-Ikuenobe, Rogers, Abdelsalam, Feight, and Witt

INTRODUCTION Active faults in Oklahoma have typically exhibited relatively long recurrence intervals (e.g., ∼1,000 years) compared to earthquakes along plate boundaries (Wheeler and Crone, 2001). The ground-shaking intensity from natural and induced earthquakes in Oklahoma (Petersen et al., 2017) vary from very strong (VII) to severe levels of shaking (VIII+) on the Modified Mercalli Intensity Scale. Historically, these earthquakes have caused minor structural damage, depending on the distance from these epicenters. Although, the underground disposal of wastewater into isolated geologic strata began in 1930 (Clark et al., 2005), the seismic records of the northern Oklahoma region, particularly in Pawnee County, spiked between January 2006 and August 2017. According to Oklahoma Geological Survey (OGS) and US Geological Survey (USGS) data, all the local earthquakes greater than magnitude (Mw ) 4.0 during this period occurred at depths of 5.5 km or less. However, the depth of the Pawnee Mw 5.8 earthquake was 5.6 km. Prior to 2016, the largest historic earthquake in the state of Oklahoma was the 1952 Mw 5.0 earthquake in Canadian County (Murphy and Cloud, 1984; Luza, 2008) (Figure 1). Other active ground motions induced by earthquakes in the state between 1977 and 2002 typically ranged from Mw 1.8 to 2.5 with an average focal depth of 4.8 km (Luza, 2008). Before 2012, >Mw 3.0 earthquakes were assumed to be naturally occurring, along with other numerous earthquakes of lesser magnitude. The increase in earthquake frequency was likely activated by wastewater injections, which correlate with increase in magnitudes between 2014 and 2017. The spatiotemporal patterns of these recently occurring earthquakes show a proximal relationship to sites of deep injection of wastewater fluids from hydraulic fracturing (Petersen et al., 2015; Rubinstein and Mahani, 2015). However, the rates have been variable and nonstationary (Petersen et al., 2015; Rubinstein and Mahani, 2015). Consequently, the USGS and OGS records of earthquake occurrences in the state have shown a steady rise in significant earthquakes above 3.0 Mw (ࣙMw 3.0) from 2010 to 2016. The Pawnee County study area is approximately 750 by 1,500 m. The study area occupies 1.125 km2 over the reported epicenter of the Pawnee Mw 5.8 that occurred about 19 km north of Pawnee Township Figure 1. The Mw 5.8 earthquake was associated with triggered rupture along the Sooner Lake Fault (Kolawole et al., 2017). The study area lies within the tectonic influence of the early Paleozoic extension and late Paleozoic compression that resulted in the Southern Oklahoma Aulacogen (Hogan and Gilbert, 1998; Keller and Baldrige, 2006; and Keller, 2014). The

450

northern Oklahoma region, including Pawnee, is characterized by highly weathered Permian dolomitic limestone. This flat-lying region transitions between the red shales and red sandstones (Permian Red Beds), Red Clay Bed Plains, and Northern Limestone Cuesta Plains (Greene, 1928; Curtis et al., 2008; and Johnson, 2008; Figure 1). The seismic influence of the aulacogen provided the structural framework for the reactivation of regional faults from long-range geographic intervals and tempo-spatial tectonic regimes (Green, 1928; Keller and Baldrige, 2006; Johnson, 2008; Chen et al., 2017; and Hincks et al., 2018). The major regional faults have experienced recent movements that contribute to the earthquake hazard potential (Keller and Baldrige, 2006). The crustal unrest during the Pennsylvanian orogeny and basin subsidence in the south resulted in the gentle raising and lowering of the broad areas in the north characterized by a wide variety of local environments, mainly marine shale with beds of sandstone, limestone, conglomerate, and coal (Johnson, 2008). The continuous rise in magnitude and frequency of seismic events in the northern Oklahoma region resulted in an investigation of the possible links between seismicity, wastewater injection (Keranen et al., 2013; Yeck et al., 2017), and estimated variations in channel discharge in the region (Manga et al., 2016). Kolawole et al. (2017) and Yeck et al. (2016) reported the Mw 5.8 event as the strongest injection-induced earthquake in Oklahoma. Surface fault rupture, vibration-induced subsidence, soil liquefaction, slope instability, plastic and permanent deformation (e.g., dilation or lurching) are some of the common surface manifestations of earthquake shaking intensity and duration. Evaluations of the earthquake’s potential deformation field were first performed using near-surface electrical resistivity tomography (Kolawole et al., 2017) and Interferometry synthetic aperture radar (InSAR) time series (Fielding et al., 2017). Additionally, Sentinel-1 InSAR and seismological data (Grandin et al., 2017) and variations in structural slip/offset (Cramer, 2017; Pennington and Chen, 2017; and Politz et al., 2017) triggered by the earthquakes close to the presumed controlling fault systems were also studied. However, the unconsolidated Quaternary-age alluvial deposits were not fully explored. Although the less consolidated alluvial deposits are theoretically susceptible to surface disruption from ground shaking, lurching, or liquefaction, these other surface deformations proximal to the earthquake location were not observed in the study area. Nevertheless, the liquefaction of unconsolidated cohesionless soils near the ground surface has been reported only for >Mw 5.7 earthquakes, exhibiting a maximum ground shaking of >5 seconds (Cloud, 1959).

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 449–462

UAS Mapping on Mw 5.8 Pawnee Earthquake

Figure 1. Map of the study area in Pawnee County, Oklahoma, showing the study location (gray square) (modiďŹ ed from Oklahoma Geologic Survey map database), regional fault lines and the epicenters of all earthquake events. The largest historic earthquake (1952 El Reno, Mw 5.5) in Oklahoma, in Canadian County, is shown in the Oklahoma county map as a red circle in the yellow square.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 449â&#x20AC;&#x201C;462

451

Ilesanmi, Liang, Oboh-Ikuenobe, Rogers, Abdelsalam, Feight, and Witt

The images acquired by the unmanned aerial system (UAS) provided valuable comparison, referred to as “change detection” in the remote sensing industry (Gomez and Purdie, 2016). Point cloud extraction from photogrammetric images provides an elevation component, z, in addition to the traditional x and y planar coordinates (Rosnell and Honkavaara, 2012). When acquired at high densities, point cloud data are usually sufficient to record discrete physical manifestations of the surface deformations that occurred before the deployment of airborne sensors. This information has rarely been documented in rural areas lacking civil infrastructure, such as highways and utilities. Surface deformations are not generally visible for <Mw 5.0 earthquakes (Crone et al., 1997; Wheeler, 2006). Orthorectified point clouds gleaned from aerial platforms are rapidly emerging as useful and inexpensive tools for identifying and tracking surficial deformations. The resolution of UAS point cloud imagery scanned from altitudes of <100 m can be on the order of ±1 cm. The extraction of photographic artifacts is particularly crucial in removing information that obstructs the delineation of ground surface data. Therefore, the resultant DSM can provide a clear view of surficial deformations often hidden by trees. These models can provide a high-resolution image of newly exposed surfaces, including common structures, such as joints, fractures, shears, or faults. This study analyzed the surficial expression of seismic excitation of a small portion of the near-field area that was most impacted by the September 3, 2016, earthquake. The specific goals of this work were to (1) identify seismic impacts on geologic and hydrologic features, (2) ascertain if any surface deformations could serve as proxies or analogs for future events, (3) test the capability of the UAS aerial survey data to detect earthquake-related deformations on the DSMs, and (4) compare them with geospatially correlated epicenters to constrain interpretations. MATERIALS AND METHODS The geospatial correlation of the 10-year seismic data (2006 to 2016) gathered for the Pawnee area prompted a reconnaissance survey conducted 9 months after the Mw 5.8 event in May 2017 to explore the viability of low-cost post-earthquake UAS surveys. Field mapping of the study area was undertaken in June 2017. The UAS data were captured in February 2018 (Figures 2A and 2B). The UAS flight planning used the mission software and data acquisition protocols appropriate for geologic reconnaissance (Appendix 1). The flight path used 23 swaths over the study area, and the flight equipment, a DJI Phantom IV Pro aircraft, was

452

installed with a 12-megapixel camera preset to a speed of 9.7 m/s at an altitude of 30 m. There were nine ground control points (GCPs) established over the study area using Trimble Geo GeoExplorer 7X Rugged GPS Handheld with GNSS Rangefinder TerraFlex. Each GCP determined at a georeferenced point was marked by a white, flat, circular disc with a black cross at its center for visibility at 30-m flight altitude (Figure 2A and B). The positions of the GCPs were limited by our inaccessibility of the adjoining parcels. Consequently, all nine GCPs were established on the section of the land adjacent to the epicenter of the Mw 5.8 earthquake. The photogrammetry images were acquired at approximately 0.2 s per pixel, and they incorporated an 80% forward and 60% side overlap of the scenes. One thousand and thirty-six digital images were processed to produce point clouds, with over 20 million points in the study area. The average density of points was 101.70 points/m². The elevation z ranged from 231.25 to 283.88 m, while the orthophoto mosaic and DSM had a resolution of 0.0006 cm. The map quality of the DSM produced was 1:10,000, and the DSM elevations ranged from 237.10to 273.80 m. Image processing was accomplished using Agisoft PhotoScan Digital Photogrammetric Software for the digital point cloud imagery to produce high-resolution DSMs. The Agisoft processing used the structurefrom-motion (SfM) algorithm to create congruent imagery with more accessible high spatial and temporal resolution DSMs (James and Robson, 2014; Carrivick et al., 2016). The SfM-derived approaches provided satisfactory results that identified small-scale deformations from the high-resolution topography and DSMs (James and Robson, 2014) for further interpretation on a larger scale. RESULTS From the field survey and DSM, we observed subsidence near the epicenter of the Mw 5.8 earthquake, consistent with those documented by Kolawole et al. (2017). The thickness of the Quaternary overburden (mostly clay with sandstone fragments) varied markedly across the study area. Based on field observation, there was little to no overburden near the rock outcrops at the location shown in Figure 3. At the same time, along watercourses, there appeared to be substantive overburden soils (>3 m) due to periodic flooding. Deformation that appeared to be previous slip exposures were also observed within 5 m of the pond closest to the Mw 5.8 epicenter (Kolawole et al., 2017). The observed vertical displacement ranged from 25 cm on the fringe of the subsidence area to ∼50 cm at the midpoint.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 449–462

UAS Mapping on Mw 5.8 Pawnee Earthquake

Figure 2A. Processed photogrammetric image of the study area showing nine ground control point placements collected using a handheld GPS unit (Trimble Geo GeoExplorer 7X Rugged GPS Handheld with GNSS Rangefinder TerraFlex.AS). Images were collected during a flight in February 2019; the red dots indicate the magnitudes of the earthquakes, and the arrows reflect the depths of earthquake epicenters. The colored rectangular box shows the digital surface model reflecting en-echelon fractures and the geospatial relationship of earthquake magnitude and the depth of epicenter.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 449–462

453

Ilesanmi, Liang, Oboh-Ikuenobe, Rogers, Abdelsalam, Feight, and Witt

Figure 2B. Details of locations of deformation structures. (i) En-echelon pattern. (ii) Strike-slip fault pattern from Riedel experiment (after Dooley and Schreurs, 2012). (iii–iv) Sections of the UAS mapped areas with denser vegetation showing conjugate joint sets. (v) Pull-apart wrench fault pattern. (vi) Strike-slip fault at the location of the Mw 5.8 earthquake. The geographic coordinates of each of the photos can be found in Appendix 2.

The UAS aerial images exposed distinct linear patterns after the extraction of tree cover (Figure 2A). However, Figure 2B shows the areas covered by dense vegetation, but after processing, a group of what are believed to be conjugate systematic regional joints were revealed in Figure 3. The tree cover camouﬂaged some of the en-echelon separation pattern (a set of parallel or subparallel, closely spaced, overlapping, or step-like minor structural features) that recently opened up. The en-echelon graben-like features were oriented in a northwest-to-southeast direction, parallel to the regional systematic joints. In addition, Figure 2B illustrates recent strain observed at the ground surface. Figure 3a shows another en-echelon set of graben-like joints spaced at approximately 2.63-m displacements that also appears to be synonymous with the regional systematic joint pattern. Studies associ-

454

ated with wrench-fault tectonics have described similar features using Riedel shear experiment (Wilcox et al., 1973; Moustafa and Abd-Allah, 1992; and Cunningham and Mann, 2007). This type of extension faulting can often occur in fracture zones above offset or lateral spreading centers, and a Mw 5.8 would generate up to 5 to 10 cm of displacement. The DSM image in Figure 3 suggests three waterfilled depressions (or “subsidence ponds”) and an enechelon fault pattern that likely projects from the underlying basement rock. The USGS seismic data for the study area indicate that there were 61 Mw 1.0–2.0 earthquakes, 12 >Mw 2.5 earthquakes, and five >Mw 3.0 earthquakes between 2000 and 2017. According to Hawthorne et al. (2019), low-magnitude earthquakes can deliver slow tremors that produce finite rupture patterns over time. The presence of these finite rup-

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 449–462

UAS Mapping on Mw 5.8 Pawnee Earthquake

Figure 3. (a) The DSM with the vegetation extracted was produced in Agisoft PhotoScan for the Pawnee area, Oklahoma. This exposed a regular fracture pattern typical of en-echelon joints or left-lateral strike-slip fault. (b) Orthoimagery of the same location with shrubs growing in the opening of the en-echelon shear. (c) The processed cloud point of the oblique angle (30°) image processed in ENVI. (d) The cloud points higher than 243 m above sea level were ﬁltered out in ENVI exposing the last ground returns with the same shearing pattern shown in the DSM image.

ture patterns predisposes the higher-magnitude earthquakes to inherit deformation patterns that are likely controlled by pre-existing discontinuities. DISCUSSION The clusters of seismic epicenters in the Pawnee area between 2006 and 2016 exhibited an east-to-west trending pattern that is consistent with the focal strain of the strike-slip fault patterns. The recent spate of earthquakes appears to have originated from depths greater than 1.5 km, with the significant clusters of epicenters emanating from depths between ∼5 and 8 km. The epicentral depths suggest that the earthquakes emanated within the Precambrian metamorphic rocks. While earthquake magnitude size does not correlate with epicenter depth, the increase in focal depth is synonymous with an increase in the number and frequency of earthquakes. Tectonic Deformation Potential The en-echelon fissures identified in Figure 3 illustrate how pre-existing structures tend to channel dynamic energy pulses and more likely to experience plastic (permanent) deformations or sudden changes in effective stress at any depth (Kearey et al. 2009; Zoback et al., 2013). En-echelon and wrench structures are usually associated with strike-slip faulting

(Wilcox et al., 1973). Earthquake shaking can also result in crustal deformation from a possible sudden release of stored elastic strain energy radiating outward, diminishing with the square of the radial distance traveled (Davies et al., 2012; King et al., 2014). Given that the epicenter is the shortest linear distance from the hypocenter to the earth’s surface, the energy released at the epicenter assumed to be the locus of the earliest p-wave arrival. For a given earthquake, the hypocenter expresses the maximum shaking intensity, projecting seismic energy outward based on the geophysical properties and structure of the overlying formations. At the rupture stage, plastic deformation occurs and causes irreversible deformation. The moment magnitude measured by three-motion seismographs records energy released during rupture by seismic wave motion (Miller, 1987). Seismogram inversions characterize earthquakes by diﬀerent origin times, hypocenter (centroid) positions, and second-rank symmetric seismic moment tensors. For larger earthquakes (Mw >5), the magnitude of the surface displacements can occur on the order of a few meters and can be estimated by the energy release on a logarithmic scale of 1moment magnitude (Mw ) = 30-fold increase in energy release. Thus, the strain energy released by earthquake rupture and transmitted by the body waves propagated

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 449–462

455

Ilesanmi, Liang, Oboh-Ikuenobe, Rogers, Abdelsalam, Feight, and Witt

lated joints exhibited linear depressions within the alluvial cover.

Non-Tectonic Deformation Potential

Figure 4. A schematic illustration of the development of visible linear depressions along parallel joint sets after seismic excitation engenders dilation that allows cohesionless ﬁll to consolidate, leaving shallow linear depressions.

by elastic deformation of the rock through which the energy pulses travel is the moment magnitude (Mw ) (Miller, 1987). As these body waves travel through the earth’s interior, seismic excitation engenders dilation in both horizontal and vertical movements, leaving shallow linear depressions that are now filled with cohesionless soils (Figure 4). The UAS aerial images appear to have delineated an extensional conjugate set of regional systematic joints for about 70 m (Figure 3). One possible mechanism is that these linear separations reflect the structure of the underlying metamorphic rocks projecting into the alluvial cover. Given the increasing seismicity over a 10-year period, extensional strain caused the conjugate joint sets to expand. It is likely that as more ground shaking occurred, soil or detritus filled the existing joints. During seismic shaking, repeated cycles of vertical and lateral loading allowed moisture-rich infill between opposing joints to consolidate, leaving linear depressions between expanded joints. This dynamically induced consolidation of the infill material was expressed as shadow cast in the photogrammetry (see Figure 2B). We speculate that during strong shaking, some of these joint sets undergo permanent dilation and deformation that allow soil infilling to subside to unknown depths. Later, the vertical and horizontal shear waves allowed the weathered infilling to consolidate and drop into widened joints during strong shaking cycles (Figure 4). The open joints probably continued to experience additional settlement of infill during aftershocks. After the shaking ceased, the di-

456

Given that the regional patterns of seismic energy decrease as the waves travel away from the hypocenter in non-tectonic regions (Hovius and Meunier, 2012), low-energy release from deep hypocenters can be attenuated by soil moisture and associated pore fluid pressure variations (Ishac and Heidebrecht, 1982) in clay-rich alluvial overburden. This energy decrease might account for lower magnitudes felt along the river courses and highly vegetated areas (Ishac and Heidebrecht, 1982). As a result, the extraction of the vegetation in the DSM images (Figure 3) provided a better view of the en-echelon faults detected in the study area. The seismic surface wave energy dispersion also exhibited zones of overlap and created zones of possible amplification, increasing the potential for localized geomorphic expression (Figure 2B). This result was consistent with the effects of cumulative low-energy faulting, like those that likely generated the jointcontrolled en-echelon patterns imaged in the photogrammetry (Figure 3d and e). In the case of smaller, non-penetrating wrench faults, deformation patterns tend to produce the type of en-echelon joint shown in Figure 3 (Moustafa and Abd-Allah, 1992; Cunningham and Mann, 2007). The Riedel shear experiment documented that smaller enechelon patterns formed at low amplitude and approximately perpendicular to principal stress (σ1) (Dooley and Schreurs, 2012). The behavior of high density (low water content) clay and sandstone in the Riedel experiment provides a suitable analog for the clay-rich beds and sandstone of the Northern Oklahoma Permian Red Reds (Eisenstadt and Sims, 2005; Withjack et al., 2007; and Dooley and Schreurs, 2012). The density of the clay increases the brittleness of fracturing in the upper crust, resulting in discrete and planar patterns similar to the en-echelon patterns in Figure 2B (Arch et al., 1988, Dooley and Schreurs, 2012). It is possible that the observed en-echelon faulting emanated from shallow ground rupture generated from non-tectonic deformation. The Mw 5.8 earthquake occurred as a result of shallow strike-slip faulting from a non-tectonically active origin (USGS, 2016). The main-shock location aligns with the major regional southwest-to-northeast trending fault. At the same time, the focal mechanism of the rupture indicates a left-lateral fault striking east-southeast or a right-lateral fault striking north-northeast. This mechanism was modeled using the classic Riedel shear experiment (Dooley and Schreurs, 2012). Our observa-

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 449–462

UAS Mapping on Mw 5.8 Pawnee Earthquake

tion is closely associated with the strike-slip faults and other strike-slip–associated features that are prevalent in Oklahoma. In their study of the Nemaha Fault in the lower Cherokee platform, Chopra et al. (2018) showed that while regional faults can be of tectonic origin, non-regional faults are less likely to be of nontectonic origin. Using three-dimensional seismic data analysis, they noted that the directions of the maximum and minimum stresses were horizontal, whereas the vertical component exhibited intermediate stress. Therefore, the large regional strike-slip fault they referred to as “wrench fault” branches and step-overs originated from multiple faults. Strike-slip motions have been shown to produce positive and negative ﬂower structures in response to compressional and extensional faults modeled using the Riedel experiment (Wilcox et al., 1973; Dooley and Schreurs, 2012; and Chopra et al., 2018). The initial stages of the Riedel experiment indicated the formation of tension fractures parallel to the principal stress (σ1 ) rotated toward steeper angles with increasing displacement along the basement fault (Dooley and Schreurs, 2012). On the contrary, the empirical relationship between earthquake magnitude and average surface displacement per earthquake event for multiple earthquakes in the range of Mw 5.5 to Mw 5.9 shows that average surface displacement can fall between 5 and10 cm (Wells and Coppersmith, 1994). The regression between magnitude and rupture surface length also shows that rupture can occur between 3.9 and 15 km away from the epicenter with a large standard deviation (Wells and Coppersmith, 1994). The two observations validate the possibility that the surface displacement of 2.6 cm exhibited in the rupture pattern shown in Figures 2B and 3 that occurred within 1.5 km of the Mw 5.8 earthquake can occur from a single event. Therefore, the average displacement and rupture length provide a possible explanation that the ruptures probably also emanated from the single Mw 5.8 event rupture.

vealed structures that appear to be joint-controlled enechelon patterns. Since the presence of woody vegetation tends to camouflage the interconnected joint sets, we suggest the processing of similar high-density imagery in future studies in order to expose patterns of shearing. Low-energy release from 5-km depth was likely attenuated by an unconsolidated “soil cap” that formed an impedance contrast. The DSM revealed areas of permanent deformation that are likely associated with cumulative seismic wave amplification in the study area. The DSM result was consistent with the deformation effects from low-energy earthquake en-echelon joints associated with wrench fault tectonics. These areas of deformation fell within a proximal distance of the Sooner Lake Fault and the earthquake epicenter. Two possible models were considered for the origin of the exposed en-echelon patterns: tectonic deformation and non-tectonic deformation. The tectonic origin was proposed based on the depth of the recent earthquakes and the strong lineation signature expressed on the DSM. However, the analog model of the classic Riedel experiment provided a more plausible explanation of the non-tectonic origin of the closely spaced en-echelon pattern. The Riedel shear experiment on clay models tends to validate the hypothesis that an increase in permanent surface deformation appears to be related to repeated cycles of seismic loading. This study also highlighted the utility of UAS photogrammetry as a sensitive and economical tool for identifying changes associated with active earth processes. Such features may pose significant physical and environmental consequences if they occur in areas crossed by linear infrastructures, such as fences and property lines, wells, highways, culverts, pipelines, transmission lines, embankments, and levees. The seismic signatures of the recent earthquakes emanating from the basement rock complex beneath the study area may be useful in ascertaining the potential extent of wastewater injection impacts.

CONCLUSIONS The increase in the frequency and magnitude of earthquakes in the Pawnee area led to the study of geomorphologic patterns likely expressed as a result of cumulative seismicity. UAS-derived DSMs and SfM technologies oﬀered a markedly higher spatial resolution than existing topographic maps and aerial photogrammetry in identifying the en-echelon fractures exposed in the epicentral area of the Mw 5.8 earthquake. The point clouds enabled the extraction of tree cover and displayed regionally continuous joint sets and fault traces more accurately than would be possible to observe from simple visual reconnaissance. The DSMs generated from the point clouds of the study area re-

ACKNOWLEDGMENTS, SAMPLES, AND DATA This study was made possible through funds provided by the Boone Pickens School of Geology at Oklahoma State University, the Geosciences and Geological and Petroleum Engineering Department at the Missouri University of Science and Technology through the Karl F. Hasselmann Endowment, and the technical contributions of staﬀ of the Aerospace Engineering Department at Oklahoma State University. In particular, we thank Dr. Jacob Jamey for facilitating the scheduling and testing of the point cloud capabilities of the UAS. We also thank Dr. Mathews Adam, Marc Hartmann, and the staﬀ of

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 449–462

457

Ilesanmi, Liang, Oboh-Ikuenobe, Rogers, Abdelsalam, Feight, and Witt

the Unmanned Systems Research Institute at Oklahoma State University for their contributions during the fieldwork, data acquisition, and processing stages of this study. SUPPLEMENTAL MATERIAL Supplemental Material associated with this article can be found online at https://doi.org/10.2113/ EEG-2359. REFERENCES Arch, J.; Maltman, A. J.; and Knipe, R. J., 1988, Shear zone geometries in experimentally deformed clays: The influence of water content, strain rate and primary fabric: Journal Structural Geology, Vol. 10, pp. 91–99. Carrivick, L. J.; Smith M. W.; and Quincey, D. J., 2016, Structure from Motion in the Geosciences: John Wiley & Sons, New York. doi:10.1002/9781118895818. Chen, X.; Nakata, N.; Pennington, C.; Haffener, J.; Chang, C. J.; He, X.; Zhan, Z.; Ni, S.; and Walter, J. I., 2017, The Pawnee earthquake as a result of the interplay among injection, faults, and foreshocks: Scientific Reports, Vol. 7, 4,945 p. doi:10.1038/s41598-017-04992-z. Chopra, S.; Marfurt, K. J.; Kolawole, F.; and Carpenter, B., 2018, Nemaha strike-slip fault expression on 3D seismic data in SCOOP Trend: American Association Petroleum Geologists Explorer, Vol. 39, No. 6, pp. 18–19. Clark, J. E.; Bonura, D. K.; and Voorhees, R. F., 2005, An overview of injection well history in the United States of America. In Tsang, C.-F. and Apps, J. A. (Editors), Underground Injection Science and Technology, Vol. 52: Elsevier Science, Amsterdam, pp. 3–12. https://doi.org/10.1016/S01675648(05)52001-X. Cloud, W. K., 1959, Intensity and Ground Motion of the San Francisco Earthquake of March 22, 1957, San Francisco Earthquake of March 1957: California Division of Mines Special Report 57, 52 p. Cramer, C. H., 2017, Brune stress parameter estimates for the 2016 M 5.8 Pawnee and other Oklahoma earthquakes: Seismological Research Letters, Vol. 88, No.4, pp. 1005–1016. doi:10.1785/0220160224. Crone, A. J.; Machette, M. N.; and Bowman, J. R., 1997, Episodic nature of earthquake activity in stable continental regions revealed by paleoseismicity studies of Australian and North American Quaternary faults: Australian Journal Earth Sciences, Vol. 44, pp. 203–214. doi:10.1080/08120099708728304. Cunningham, W. D. and Mann, P., 2007, Tectonics of strike-slip restraining and releasing bends. In Cunningham, W. D. and P. Mann (Editors), Tectonics of Strike-slip Restraining and Releasing Bends: Special Publications, Vol. 290, Geological Society, London, U.K., pp. 1–12. doi:10.1144/SP290.1. Curtis, N. M., Jr.; Ham, W. E.; and Johnson, H. S., 2008, Geomorphic Provinces of Oklahoma: Oklahoma Geological Survey Educational Publication, Vol. 9, No. 8. Davies, T.; McSaveney, M.; and Boulton, C., 2012, Elastic strain energy release from fragmenting grains: Effects on fault rupture: Journal Structural Geology, Vol. 38, pp. 265–277. doi:10.1016/j.jsg.2011.11.004. Dooley, T. P. and Schreurs, G., 2012, Analogue modelling of intraplate strike-slip tectonics: A review and new ex-

458

perimental results: Tectonophysics, Vol. 574–575, pp. 1–71. https://doi.org/10.1016/j.tecto.2012.05.030. Eisenstadt, G. and Sims, D., 2005, Evaluating sand and clay models: Do rheological diﬀerences matter?: Journal Structural Geology, Vol. 27, pp. 1399–1412. Fielding, E. J.; Sangha, S. S.; Bekaert, D. P. S.; Samsonov, S. V.; and Chang, J. C., 2017, Surface deformation of North-Central Oklahoma related to the 2016 Mw 5.8 Pawnee earthquake from SAR interferometry time series: Seismological Research Letters, Vol. 88, No. 4, pp. 971–982. https://doi.org/10.1785/0220170010. Gomez, C. and Purdie, H., 2016, UAV-based photogrammetry and geocomputing for hazards and disaster risk monitoring—A review: Geoenvironmental Disasters, Vol. 3, No. 23, pp. 1–11. doi:10.1080/19475705.2016.1176605. Grandin, R.; Vallée, M.; and Lacassin, R., 2017, Rupture process of the Mw 5.8 Pawnee, Oklahoma earthquake from Sentinel-1 InSAR and seismological data: Seismological Research Letters, Vol. 88, No. 4, pp. 994–1004. doi:10.1785/0220160226. Green F. C., 1928, Oil and gas in Oklahoma: Geology of Pawnee County: Oklahoma: Geological Survey Bulletin, Vol. 40-CC, 15 p. Hawthorne, J. C.; Thomas, A. M.; and Ampuero, J. P., 2019, The rupture extent of low-frequency earthquakes near Parkﬁeld, CA: Geophysical Journal International, Vol. 216, No. 1, pp. 621–639. doi:10.1093/gji/ggy429. Hincks, T.; Aspinall, W.; Cooke, R.; and Gernon, T., 2018, Oklahoma’s induced seismicity strongly linked to wastewater injection depth: Science, Vol. 259, No. 6381, pp. 1251–1255. doi:10.1126/science.aap7911. Hogan J. P. and Gilbert M. C., 1998, The Southern Oklahoma Aulacogen: A Cambrian analog for Mid-Proterozoic AMCG (Anorthosite-Mangerite-Charnockite-Granite) complexes? In Hogan J. P. and Gilbert M. C. (Editors), Proceedings of the International Conferences on Basement Tectonics, Vol. 6, Basement Tectonics 12: Springer, Dordrecht, Netherlands, pp. 39– 78. https://doi.org/10.1007/978-94-011-5098-9_3. Hovius, N. and Meunier, P., 2012, Earthquake ground motion and patterns of seismically induced landsliding. In Clague, J. and Stead, D. (Editors), Landslides: Types, Mechanisms and Modeling: Cambridge University Press, Cambridge, U.K. pp. 24–36. doi:10.1017/CBO9780511740367.004. Ishac, M. F. and Heidebrecht, A. C., 1982, Energy dissipation and seismic liquefaction in sands: Earthquake Engineering Structural Dynamics, Vol. 10, No. 1, pp. 59–68. https://doi.org/10.1002/eqe.4290100105. James, M. R. and Robson, S., 2014, Mitigating systematic error in topographic models derived from UAV and ground-based image networks: Earth Surface Processes Landforms, Vol. 39, pp. 1413–1420. https://doi.org/10.1002/esp.3609. Johnson, K. S., 2008, Geologic History of Oklahoma: Oklahoma Geological Survey Educational Publication, Vol. 9, pp. 3–7. Kearey, P.; Klepeis, K. A.; and Vine, F. J., 2009, Global Tectonics, 3rd ed.: Blackwell Publishing, Malden, MA. https://doi.org/10.1002/gj.2442. Keller, G. R., 2014. The Southern Oklahoma Aulacogen: It’s a classic. In Neil Suneson, Igneous and Tectonic History of the Southern Oklahoma Aulacogen: Oklahoma Geological Survey Guidebook 38, Part II, Research Paper 15. 405 p. http://ogs.ou.edu/docs/guidebooks/GB38PIIRP15.pdf. Keller, G. R. and Baldridge, W. S., 2006, Chapter 12: The southern Oklahoma Aulacogen: Developments Geotectonics, Vol. 25, pp. 427–436. https://doi.org/10.1016/S0419-0254(06)800200.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 449–462

UAS Mapping on Mw 5.8 Pawnee Earthquake Keranen, K.; Savage, H. M.; Abers, G. A.; and Cochran, E. S., 2013, Potentially induced earthquakes in Oklahoma, USA: Links between wastewater injection and the 2011 Mw 5.7 earthquake sequence: Geology, Vol. 41, No. 6, pp. 699–702. doi:10.1130/G34045.1. King, G. C. P.; Stein, R. S.; and Lin, J., 2014, Static stress changes and the triggering of earthquakes: Bulletin Seismological Society America, Vol. 84, No. 3, pp. 935–953. doi:https://doi. org. Kolawole, F.; Atekwana, E. A.; and Ismail, A., 2017, Near-surface electrical resistivity investigation of coseismic liquefaction-induced ground deformation associated with the 2016 Mw 5.8 Pawnee, Oklahoma earthquake: Seismological Research Letters, Vol. 88, pp. 1017–1023. Küng, O.; Strecha, C.; Beyeler, A.; Zufferey, J. C.; Floreano, D.; Fua, P.; and Gervaix, F., 2011, The accuracy of automatic photogrammetric techniques on ultra-light UAV imagery: International Archives of Photogrammetry, Remote Sensing & Spatial Information Sciences, Vols. XXXVIII– 1/C22, pp. 125–130. Luza, K., 2008, Earthquakes of Oklahoma: Oklahoma Geological Survey Educational Publication, Vol. 9, 9 p. Manga, M.; Wang, C.; and Shirzaei, M., 2016, Increased stream discharge after the 3 September 2016 Mw 5.8 Pawnee, Oklahoma earthquake: Geophysical Research Letters, Vol. 43, No. 22. https://doi.org/10.1002/2016GL071268. Miller, C. D., 1987, Glossary of Selected Volcanic Terms, Earthquakes & Volcanoes, Vol. 19, No. 4: U.S. Geological Survey, Washington, DC, 155 p. Moustafa, A. R. and Abd-Allah, A. M., 1992, Transfer zones with en-echelon faulting at the northern end of the Suez Rift: Tectonics, Vol. 11, No. 3, pp. 499509. https://doi.org/10.1029/91TC03184. Murphy, L. M. and Cloud, W. K., 1984, United States Earthquakes, 1952: U.S. Geological Survey Open-File Report 84-952, pp. 7–9. https://pubs.usgs.gov/of/1984/0952/report.pdf. Pennington, C. and Chen, X., 2017, Coulomb stress interactions during the Mw 5.8 Pawnee Sequence: Seismological Research Letters, Vol. 88, No. 4, pp. 1024–1031. doi:10.1785/0220170011. Petersen, M. D.; Mueller, C. S.; Moschetti, M. P.; Hoover, S. M.; Rubinstein, J. L.; Llenos, A. L.; Michael, A. J.; Ellsworth, W. L.; McGarr, A. F.; Holland, A. A., and Anderson J. G., 2015, Incorporating Induced Seismicity in the 2014 United States National Seismic Hazard Model—Results of 2014 Workshop and Sensitivity Studies: U.S. Geological Survey Open-File Report 2015–1070, 69 p. http://dx.doi.org/10.3133/ofr201510. Petersen, M. D.; Mueller, C. S.; Moschetti, M. P.; Hoover, S. M.; Shumway, A. M.; McNamara, D. E.; Williams, R. A.; Llenos, A. L.; Ellsworth, W. L.; Michael, A. J.; Rubinstein, J. L.; McGarr, A. F.; and Rukstales, K. S., 2017, One-year seismic-hazard forecast for the central and eastern United States from induced and natural earthquakes: Seismological Research Letters, Vol. 88, No. 3, pp. 772–783. doi.org/10.1785/0220170005. Politz, F. F.; Wicks, C.; Schoenball, M.; Ellsworth, W.; and Murray M., 2017, Geodetic slip model of the 3 September 2016 Mw 5.8 Pawnee, Oklahoma earthquake: Evidence for fault-zone collapse: Seismological Research Letters, Vol. 88, No. 4, pp. 983–993. doi:10.1785/0220170002. Rosnell, T. and Honkavaara, E., 2012, Point cloud generation from aerial image data acquired by a quadrocopter type micro unmanned aerial vehicle and a digital still camera: Sensors, Vol. 12, No. 1, pp. 453–480. doi:10.3390/s120100453.

Rubinstein, J. L. and Mahani, AB., 2015, Myths and facts on wastewater injection, hydraulic fracturing, enhanced oil recovery, and induced seismicity: Seismological Research Letters, Vol. 86, No. 4, pp. 1060–1067. doi:10.1785/022015006. U.S. Geological Survey, 2016, Tectonic Summary: https:// earthquake.usgs.gov/archive/product/poster/20160903/us/ 1476475864730/poster.pdf. Wells, D. L. and Coppersmith, K. J., 1994, New empirical relationships among magnitude, rupture length, rupture width, rupture area, and surface displacement: BulletinSeismological Society of America, Vol. 84, No. 4, pp. 974–1002. Wheeler, R. L., 2006, Quaternary tectonic faulting in the eastern United States: Engineering Geology, Vol. 82, No. 3, pp. 165– 186. doi:10.1016/j.enggeo.2005.10.005. Wheeler, R. L. and Crone, A. J., 2001, Known and suggested Quaternary faulting in the Midcontinent United States: Engineering Geology, Vol. 62, pp. 51–78. doi:10.1016/S00137952(01)00050-3. Withjack, M. O.; Schlische, R.; and Henza, A. A., 2007, Scaled experimental models of extension: Dry sand vs. wet clay, Houston: Geological Society Bulletin, Vol. 49, No. 8, pp. 31–49. Wilcox, R. E.; Harding, T. P.; and Seely, D. R., 1973, Basic wrench tectonics: American Association Petroleum Geologists Bulletin, Vol. 57, pp. 74–96. Yeck, W. L.; Hayes, G. P.; McNamara, D. E.; Rubinstein, J. L.; Barnhart, W. D.; Earle, P. S.; and Benz, H. M., 2017, Oklahoma experiences largest earthquake during ongoing regional wastewater injection hazard mitigation eﬀorts: Geophysical Research Letters, Vol. 44, pp. 711–717. doi:10.1002/ 2016GL071685. Yeck, W. L.; Weingarten, M.; Benz, H. M.; McNamara, D. E.; Bergman, E.; Herrmann, R. B.; Rubinstein, J.; and Earle, P. S., 2016, Far-ﬁeld pressurization likely caused one of the largest injection induced earthquakes by reactivating a large pre-existing basement fault structure: Geophysical Research Letters, Vol. 43, No. 10, pp. 10,198–10,207. doi:10.1002/2016GL070861. Zoback, M. L.; Geist, E.; Pallister, J.; Hill, D. P.; Young, S.; and McCausland, W., 2013, Advances in natural hazard science and assessment, 1963–2013. In Bickford, M. E. (Editor), The Impact of the Geological Sciences on Society: Geological Society of America Special Papers, Vol. 501, pp. 81–154. https://doi.org/10.1130/2013.2501(05).

APPENDIX 1 Flight Planning The significant variables considered before developing the flight plan for this study were the aircraft altitude, forward and side overlap, aircraft speed, and access to standard mission planning software. Using the basic principles of photogrammetry for DJI GS Pro and Pix4D capture, a ground sampling distance (GSD) of 2.5 cm/pixel was selected based on the expected scale of the faults within the study area. The resulting flight altitude was determined to be 30 m. The overlap was set to 80% forward and 60% side based on the recommendations in Rosnell and Honkavaara (2012). The aircraft ground speed was 9.2 m/s, calculated from the shutter speed of the camera and GSD.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 449–462

459

Ilesanmi, Liang, Oboh-Ikuenobe, Rogers, Abdelsalam, Feight, and Witt

From these specifications, the optimal pixel size of 1,095 × 730 used along with the GSD to adjust for the actual flight altitude (see Supplemental Material Figure S1). Two camera orientations were considered, nadir (0°, downward facing) and oblique (30° from nadir). These angles were chosen based on the overlap coverage and orientation of the camera with respect to the ground plane. The nadir photographs were the principal set used in our mapping, based on the recommendations of James and Robson (2014), which afforded variability in the camera orientation and reduced the error values. These are shown in Supplemental Material Figure S2. Mission Planning During the mission planning, the baseline flight path parameters were calculated based on the size of the study area (750 × 1,500 m). The next step was to input the parameters into the flight planning software to integrate the flight path instructions into the standard takeoff and landing modalities. To achieve data accuracy on the dual flight operation planned for this study, the DJI GS Pro was flown for the nadir flights, and the Pix4D capture was used for the oblique flights. Typically, the preferred option was to keep the mission planning software consistent between flights; however, each software offered unique utilities. The DJI GS Pro provided ease of implementation and the ability to pause and resume during missions. This reduced the required flight time to complete the task. Pix4Dcapture is more flexible for control of aircraft orientation. The resulting parameters for the survey site correlated with the number of images acquired. Ground Control Point The registration of ground control points (GCPs) enabled the validation of the accuracy of the UASgenerated coordinates. The georeferencing process required the establishment of the fixed GPS locations that were sufficiently visible to be identified within the captured scenes (photos). The establishment of fixed ground objects was accomplished by placing GCP discs in the field. Distinct objects from the surrounding scenery were placed (e.g., a white paper plate with a black center point) before UAS flight and image capture. Invariant landscape features can also be used as GCPs if targets are not available (e.g., a rock outcrop in an agricultural field, the corner of the concrete pad, etc.). During the mission flight, the Global Positioning System (GPS) location of each GCP was recorded with a Trimble GeoExplorer 7X Rugged GPS Handheld

460

with GNSS Rangeﬁnder TerraFlex (Annex 1 spreadsheet of GPS points). The accuracy for tests, including the GCP locations was between 0.05 and 0.2 m. The GCP improved the eﬃciency of the point cloud creation with structure from motion (Küng et al., 2011). The accuracy for tests was based on the geotagged images between 2 and 8 m (Küng et al.. 2011). Although UAS cameras are equipped with GPS receivers for georeferencing, the locations of the sensed images were recorded and stored. The GCPs allowed for the normalization of data and correction of errors from external sources.

Structure-from-Motion Data Processing and Data Products Structure from motion is an approach applied to the creation of a three-dimensional (3D) model of the imagery. The images were processed using Agisoft PhotoScan software based on the following steps: loading captured images into PhotoScan, inspecting loaded images and removing unnecessary images, aligning photos, building a dense point cloud, building a 3D polygonal mesh model, generating texture, building a tiled model, building digital elevation and surface models, compiling an orthomosaic, and exporting the results. The process of building a dense point cloud allows Agisoft to generate and visualize a detailed point cloud model based on the estimated camera positions. The 3D polygonal model was created from the point cloud. The process of building a 3D mesh involves a data reconstruction process that aims to produce an optimal surface as point clouds are resolved back into a continuous image. The mesh is necessary to aid the manual georeferencing effort. PhotoScan employs the 3D mesh surface to tie the GCP targets together. The reconstruction process produced an arbitrary surface model with an accurate representation of the bare ground, rivers, fractures, and trees in the study area. The mesh was created by interpolating for missing data points. The density of the data collection process reduced the number of interpolated data points. Agisoft was used to calculate the surface area within a consistent radius around every dense cloud point. This step ensured that point spaces that were underpopulated were filled. The 3D model can be georeferenced to the realworld location using the GPS coordinates stored in the metadata of each of the capture photos. Direct georeferencing was performed by using the location of the sensor at the moment of data capture to enhance the position of the generated ground data. The GCP locations were identified within the images, and the georeferencing of these points was based on the acquired

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 449–462

UAS Mapping on Mw 5.8 Pawnee Earthquake

coordinates. After the georeferencing of the GCPs, the images were adjusted and reviewed for accuracy. Agisoft PhotoScan was also used to generate and visualize a digital elevation model (DEM) from the dense point cloud and the digital surface model (DSM) from a sparse point cloud extrapolated from the grid of height values. Since the area of interest was the bare earth features, the cloud points were classiﬁed as “ground” and “other.” The classiﬁcation enables the processor to remove all points representing trees, shrubs, buildings, and structures above the ground surface. The DSM values were calculated from the density of point cloud data and converted to a raster image.

Consequently, the generation of a DSM enhanced the visualization of geomorphologic features. The DSM can also be used to perform DEM-based point, distance, area, and volume measurements or to generate cross sections selected by the user. Additionally, contour lines can be calculated for the model and depicted over either DEM or orthoimagery in ArcGIS or ENVI environments. Figure S1. The metadata of UAS image DJI_0125 providing information about the camera used, the location, ﬂight altitude, and quality of the captured data. Figure S2. Impacts of sensor obliquity with camera angle (modiﬁed from James and Robson, 2014).

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 449–462

461

462

2016-09-03T12: 02:44.400Z

2017-12-18T03: 45:00.500Z

2015-11-10T07: 11:48.330Z

2016-09-15T19: 55:51.100Z

2017-01-28T11: 20:54.500Z

2016-09-19T11: 03:15.400Z

2017-06-25T05: 04:21.500Z

2016-09-04T14: 29:23.100Z

2016-06-06T18: 52:16.600Z

2017-05-24T17: 45:37.800Z

36.42

36.43

36.42

36.43

36.42

36.43

5.56 3.81 5 5.17 5.2 6.3 5.32 8.14 8.6 4.71

−96.93 −96.93 −96.93 −96.93 −96.93 −96.92 −96.92 −96.92 −96.92 −96.92

No. Time Latitude Longitude Depth 1 2017-08-21T05: 36.43 −96.93 5.3 06:00.900Z

2.7

3.3

2.9

2.8

2.6

2.7

2.6

5.8

Magnitude 2.5

mwr

mb_lg

mww

—

Magnitude Type nst gap ml — 46

—

0.02

—

0.31

—

dmin —

0.31

0.5

0.41

0.12

0.2

0.34

0.21

0.48

0.21

0.7

rms net 0.07 us

us10008uw3

us200062d7

us10006kc2

us20009pkv

us10006qtk

us10007vxg

us10006px2

us10003w7b

us2000c5i7

us10006jxs

id us2000aa2r

Place 15 km NW of Pawnee, Oklahoma 14 km NW of Pawnee, Oklahoma 14 km NW of Pawnee, Oklahoma 14 km NW of Pawnee, Oklahoma 14 km NW of Pawnee, Oklahoma 14 km NW of Pawnee, Oklahoma 14 km NW of Pawnee, Oklahoma 14 km NW of Pawnee, Oklahoma 14 km NW of Pawnee, Oklahoma 14 km NW of Pawnee, Oklahoma 14 km NW of Pawnee, Oklahoma Earthquake

Earthquake

Type Earthquake

2.1

2.5

1.2

0.8

0.6

1.5

0.9

0.7

1.1

1.2

4.5

1.3

2.7

1.2

6.1

HorizontalError DepthError 1.3 3.1

iii

RefImage vi

DJI_0194.jpg

DJI_0085.jpg

DJI_0125.jpg

DJI_0085.jpg

RefImage Number DJI_0300.jpg

Appendix 2. List of earthquakes within the study area shown in Figure 2A with the UAS reference image numbers and reference image identity signature in Figure 2B.

Ilesanmi, Liang, Oboh-Ikuenobe, Rogers, Abdelsalam, Feight, and Witt

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 449–462

Sources of Perennial Water Supporting Critical Ecosystems, San Pedro Valley, Arizona CHRISTOPHER J. EASTOE1, * Department of Geosciences, University of Arizona, Tucson, AZ 85719

Key Terms: Geohydrology, Base Flow, Groundwater, Stable Isotopes, Tritium, Arizona ABSTRACT Stable O and H isotope data distinguish three sources for base flow in five reaches of the San Pedro River: (A) base flow and sub-flow from upstream reaches of the river; (B) bank storage derived from summer monsoon floodwater; and (C) water from the mountainous flanks of the river catchment. A and C support base flow in the sub-basin upstream of Sierra Vista. A, B, and C combine to support base flow near St. David. Source C in this area is ancient deep-basin groundwater. Source C dominates in Cascabel near Benson Narrows, with downstream additions from A. In Cascabel near Gamez Road, sources A and C combined to support base flow that had disappeared by 2019. Near Redington, source C appears to have operated through a limestone aquifer vulnerable to short-term drought. Groundwater sub-basins separated by impermeable sills in the riverbed are evolving into hydrologically separate subbasins as base flow across the sills decreases. The decrease in base flow partly reflects regional long-term drought, which has been exacerbated by pumping. Additional groundwater demand from urban growth upstream of Benson is likely to cause further decline of base flow near St. David and Sierra Vista. INTRODUCTION The San Pedro River (SPR) rises in the Huachuca Mountains of southern Arizona, ﬂows into northern Sonora, Mexico, returns to Arizona, and then ﬂows north about 200 km to join the Gila River (Figure 1). Surface and subsurface water in the riparian zone supports a semi-continuous ribbon of riparian vegetation through the surrounding semiarid-toarid landscape. Discontinuous reaches with perennial water furnish small but crucial amounts of water to local and migrating fauna. Since documentation began in 2007, most of the perennial reaches have declined in 1 Retired

*Corresponding author email: eastoe@email.arizona.edu

length (The Nature Conservancy, 2020). The decline has occurred concurrently with a drought of decadal timescale (Abatzoglu et al., 2017; WestWide Drought Tracker, 2019), but it probably also reflects overdraft of groundwater (Cordova et al., 2015; Gungle et al., 2017). The river valley constitutes a corridor between forested terrains of the Sierra Madre Occidental in Mexico and the southern edge of the Colorado Plateau in the United States and lies close to the boundary between the Sonoran and Chihuahuan Deserts. Its location, in combination with the availability of water, has engendered rich faunal biodiversity. Vertebrates in the Upper San Pedro Valley (SPV) include 61 to 87 species of mammals, 100 species of breeding birds, 200 species of migrating birds, and 55 recorded reptiles and amphibians (Brand et al., 2009; Rosen, 2009; and Soycan et al., 2009). Maricopa Audubon Society (2020) lists “over 375” bird species recorded in the San Pedro Riparian National Conservation Area. Avian density in the riparian Populus forests is high: 431 ± 22 birds per 40 ha in the breeding season, and 1468 ± 92 birds per 40 ha during the spring migration, with the latter figure exceeding comparable avian density estimates on the Rio Grande and Colorado River by factors of 3.1 and 8.5, respectively (Brand et al., 2009). In the lesssettled area north of Benson, AZ, the SPV is crossed by major wildlife corridors linking the “sky islands” of flanking mountain ranges. The SPV is therefore a highpriority area for conservation, a major aspect of which is preservation of surface and subsurface water in the riparian zone. From the city of Sierra Vista to the Mexico-U.S. border (Figure 1), the valley has undergone high human population growth. Pressure for urban development between Sierra Vista and Benson is increasing (Arizona Daily Star, 2016). Such development inevitably leads to competition between human settlements and riparian plant and animal communities for scarce water resources. Preservation of surface water requires a clear understanding of sources of the perennial water. Stable O and H isotope effects such as altitude effects and the damping of ranges of δ18 O and δ2 H in groundwater relative to precipitation provide means of identifying sources of base flow, locating

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

463

Eastoe

land (Tetzlaff and Soulsby, 2008), but identified baseflow sources in higher-permeability strata of catchments in Luxemburg (Pfister et al., 2017) and Oregon, United States (Nickolas et al., 2017). In the Willamette River catchment, Oregon, base flow in dry years is sustained by groundwater stored in high-permeability volcanic strata of the Cascade Range (Tague and Grant, 2004; Brooks et al., 2012). Larger catchments, >105 km2 , include rivers that flow from well-watered source regions into desert, where isotope changes are controlled by evaporation, e.g., the Barwon-Darling River, Australia (Hughes et al., 2012), and the Colorado River, United States (Guay et al., 2006). In the upper Rio Grande catchment, United States, discharge of groundwater into the river was not detected in isotope data where the river exits alluvium-filled grabens, but it was visible as stepwise increases in Cl− concentration and Cl− /Br− ratios (Phillips et al., 2003; Hogan et al., 2012). The SPR catchment, about 1.2 × 104 km2 , differs from catchments of comparable size mentioned above in that the river is ephemeral over much of its length (The Nature Conservancy, 2020). Between December and June, source points of base flow are readily identified at the heads of reaches with perennial water. The sources of base flow and the relationship of perennial water in successive reaches are matters of relevance to management of the river. This article reviews current understandings of the geology and hydrology of the SPV between the Mexico-U.S. border and Redington, Arizona (Figure 1), and examines new and existing isotope data with the aim of providing an improved understanding of water sources for future management of the SPV. BACKGROUND Study Area

Figure 1. Location map showing study areas 1 to 5.

groundwater discharge zones, and relating sources and discharge to geology at a variety of scales. In smallarea catchments, <102 km2 , isotope signals related to base ﬂow may be masked by seasonal variation in δ18 O and δ2 H (Soulsby et al., 2000; Blumstock et al., 2015). In other cases, sources of base ﬂow can be related to topographic features (Singh et al., 2016), deep groundwater in volcanic strata (Fujimoto et al., 2016), and storage in colluvium (Segura et al., 2019). In mesoscale catchments, 102 to 104 km2 , isotope evidence yielded little insight in the Dee River watershed, Scot-

464

This study considered a 130 km length of the SPV between the U.S.-Mexico border (1410 meters above sea level [masl]) and Redington, Arizona (980 masl). Within this interval, five river segments (areas 1 to 5 of Figures 1 and 2) have had perennial water, i.e., base flow in the driest seasons, for part or all of the interval 2000–2019. Watershed altitudes extend to 2700 masl in flanking mountain ranges. The climate is semiarid, except above 2000 masl. Average annual rainfall at representative stations is 355 mm at Sierra Vista, 288 mm at Benson, 338 mm at Cascabel, and 345 mm at Redington, while at 2345 masl in the Santa Catalina Mountains, the average is 762 mm (Western Regional Climate Center, 2019). Precipitation totals vary greatly from year to year. Two wet seasons occur: a season of orographic rain or snow

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

Sources of Perennial Water, San Pedro Valley, Arizona

by Quercus, Juniperus, and Arctostaphylos, and by Pinus at higher elevations. Geology and Hydrogeology

Figure 2. Detailed maps of study areas 2 to 5, indicated by bold numerals. Site name abbreviations in area 3 map: TLC = Three Links Crossing; HSC = Heaven Sent Ranch; UTW = upstream of Teran Wash; DTW = downstream of Teran Wash.

from Paciﬁc fronts during winter and spring, and a season of convective precipitation from the North American Monsoon between late June and September. In some years, tropical depressions provide heavy rains in September and October. June–October precipitation makes up about 70 percent of precipitation south of Sierra Vista, and about 60 percent of that north of Benson (Western Regional Climate Center, 2019). Along the river channel, riparian forest, mainly Populus, Salix, Baccharis, and invasive Tamarix, forms a discontinuous strip up to a few hundred meters wide; its presence and width depend on availability of shallow groundwater. Mesquite (Prosopis) woodland occupies higher terraces, commonly forming bands tens to hundreds of meters wide. Much mesquite woodland has been cleared for agriculture. Dry basin slopes are occupied by Sonoran Desert vegetation, commonly stunted Prosopis, Larrea, Acacia, Opuntia, Cylindropuntia, and Yucca, along with forbs, grasses, and annuals. Vegetation on lower mountain slopes is dominated

The SPV is a faulted trough within the extensional Basin and Range Province (Fenneman, 1931). Tectonic extension began about 20 Ma, leading initially to the deposition of granitic detritus forming the San Manuel Formation in the northern part of the study area (Dickinson, 1991, 2002). From 11 Ma, deep extensional basins opened within the study area. Geologic evolution of the SPV differs north and south of the Benson Narrows (Figure 2), a valley-wide block of Proterozoic granite. To the north, internal drainage prevailed between 11 and 5 Ma, resulting in the accumulation of alluvial, fluvial, and lacustrine/evaporitic facies of the Quiburis Formation (Dickinson, 1998, 2003). South of the Benson Narrows, exposed basin fill in the Benson–St. David area (Figure 2) consists of alluvial/fluvial siltstone and mudstone with lesser lacustrine facies, named the St. David Formation of Pliocene to Pleistocene age (Gray, 1965). Beneath the St. David Formation, an up to 200 m section of lacustrine or wetland clay-rich sediment overlies about 200 m of coarser clastics, both of undetermined age (Dickinson et al., 2010a, 2010b). Clay-rich sediment also accumulated in the basin upstream of Sierra Vista (Pool and Dickinson, 2007). Dickinson (1998) proposed that an external, integrated drainage system comprising the ancestral San Pedro and Gila Rivers developed north of the Benson Narrows by 5 Ma, leading to subsequent erosion of much of the Quiburis Formation. Drainage from south of the Benson Narrows into the ancestral SPR had developed by the mid-Pliocene. The rate of erosion of the St. David Formation in the Benson area increased greatly during the Pleistocene. Except where it crosses impermeable sills, the present riverbed lies within a band of Holocene fluvial and wetland sediments up to 1500 m wide, partly concealed by younger alluvium (Cook et al., 2010). The sediments occupy a trench cut into basinfill sediments since about 20 ka (Huckleberry et al., 2009). Starting possibly in the 1850s in the Cascabel area (Hereford and Betancourt, 2009), the river excavated an entrenched channel up to 6 m deep in the trench sediments. Late nineteenth-century entrenchment was a regional phenomenon in Arizona and neighboring areas (Bryan, 1925), possibly resulting from climate change or from human-related activities such as removal of beavers and overgrazing (Hereford and Betancourt, 2009). Six synchronous cycles of entrenchment and valley filling have been documented across southeastern Arizona in the last

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

465

Eastoe

4,000 years (Waters and Haynes, 2001), suggesting a natural, cyclic cause. Basin-fill sediments constitute the productive aquifers of the SPV. Principal hydrogeological features of the basin are as follows. (1) The well-watered mountain areas, where winter snow may accumulate, provide runoff that recharges alluvium along the basin flanks and at times reaches the SPR. (2) Coarse clastic sediments along the basin flanks, locally fining towards the basin axis, make up a regional, unconfined aquifer where no thick clay unit is present (e.g., Baillie et al., 2007; Hopkins et al., 2014). (3) Thick clay-rich units near Benson and south of Sierra Vista confine groundwater, locally under flowing artesian conditions, in underlying coarse clastic sediments. In the Benson area, confinedaquifer groundwater is ancient; 19 samples for which data are available contained 1 to 33 pMC. Of 16 samples with tritium analyses, 14 contained less than 1 TU (Hopkins et al., 2014). (4) The post–20 ka trench sediments, including clastic units, are less consolidated and more permeable than the sediment into which the trench was cut. (5) Impermeable sills in the riverbed (Figure 1) consist of Cretaceous igneous and sedimentary rock in area 1 near Sierra Vista, Proterozoic granite in area 3 at the Benson Narrows, and an argillaceous unit of the San Manuel Formation in area 4 (Figure 2). Little or no trench sediment is present at the sills, but unconsolidated Holocene fluvial sediment occurs beneath the river. (6) Parallel paleochannels act as groundwater conduits within the trench sediments. Such channels were identified near Sierra Vista (Waters and Haynes, 2001) and were inferred near Cascabel (Eastoe and Clark, 2018). The post–20 ka trench sediments constitute the most productive aquifers north of the Benson Narrows, where clay-rich lenses confine groundwater within underlying clastic sediment of the trench (Eastoe and Clark, 2018). South of the Benson Narrows, much groundwater is extracted from the coarse clastic basinfill alluvium beneath broader bodies of clay-rich sediment (Hopkins et al., 2014; Pool and Dickinson, 2007). In both areas, local shallow riparian aquifers are perched atop the clay-rich sediments (Baillie et al., 2007; Huckleberry et al., 2009; MacNish et al., 2009; and Eastoe and Clark, 2018). Recharge to basin-fill alluvium is probably focused along major tributary washes, occurring mainly from mountain runoff at or near the mountain fronts (Wahi et al., 2008; Hopkins et al., 2014). Local recharge

466

occurs from the SPR to shallow riparian aquifers (Baillie et al., 2007; Hopkins et al., 2014; and Eastoe and Clark, 2018). Near Sierra Vista, diffuse recharge may not reach the regional aquifer from areas between streambeds (Glenn et al., 2015). Previous Water Isotope Studies In the SPV near Sierra Vista, measurements of stable O and H isotopes and tritium in precipitation were made from 2000 to 2003 (Coes and Pool, 2007), and in 2004–2005 (Baillie et al., 2007). Coes and Pool (2007) used tritium data to determine infiltration rates beneath tributary washes near Sierra Vista. Wahi et al. (2008) and Baillie et al. (2007) used stable water isotopes, tritium, 14 C, and major ion concentrations to establish the relationships among mountainblock groundwater in the Huachuca Mountains, basin groundwater, and discharge into the SPR. Gungle et al. (2017) summarized long-term water isotope data sets collected at five U.S. Geological Survey stream gauge sites on the SPR south of Tombstone, AZ (Figure 1). Their data documented: (1) an increase in δ18 O in base flow from 2000 to 2012, consistent with a decreasing fraction of local groundwater; and (2) discharge of mountain-derived groundwater into certain reaches of the SPR under base-flow conditions. Near Benson, Hopkins et al. (2014) used stable water isotopes, tritium, 14 C, and major ion concentrations to determine groundwater types and residence times in shallow riparian, confined sub-clay, and unconfined alluvial aquifers. In the Cascabel area, Eastoe and Clark (2018) used stable water isotopes and tritium to distinguish groundwater of different sources and residence times. METHODS Water samples for isotope measurement were collected from surface flows, springs, and wells in continual use. New isotope data (Supplemental Table S1) were measured at the Environmental Isotope Laboratory, University of Arizona. Stable O and H isotope measurements were made on a Finnigan Delta S dualinlet mass spectrometer with an automated CO2 equilibrator (for O) and an automated Cr-reduction furnace (for H). Results are expressed in the usual delta notation, e.g., R (sample) 18 2 δ O or δ H = 1000 − 1 %0 , (1) R (standard ) where R = 18 O/16 O or 2 H/1 H. Analytical precisions (1σ) are 0.08‰ for O and 0.9‰ for H. Tritium was measured by liquid scintillation counting in a Quantulus 1220 spectrometer. Samples of

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

Sources of Perennial Water, San Pedro Valley, Arizona Table 1. Simplified water budgets in three sub-basins of the San Pedro Valley after Cordova et al. (2015) and Gungle et al. (2017). Sub-Basin

Year

Inﬂow (hm3 /yr)

ET (hm3 /yr)

HD (hm3 /yr)

ET/Inﬂow (%)

HD/Inﬂow (%)

Sierra Vista* Benson# Cascabel to Redington§

2012 2010 2010

20.3 20.0 13.8

15.0 17.1 12.2

15.0 9.1 2.8

74 86 88

74 46 20

ET = evapotranspiration; HD = human demand. *Sierra Vista, AZ, i.e., area 1 of this study. # Charleston to Benson Narrows, including area 2 of this study. § Benson Narrows to Redington, including areas 3 and 4 of this study.

0.19 L were enriched by electrolysis and measured with a detection limit of 0.6 TU. Results are expressed in tritium units (TU); 1 TU corresponds to 1 tritium atom per 1018 hydrogen atoms. For calibration, international standards Vienna standard mean ocean water (VSMOW), standard light Antarctic precipitation (SLAP), and NBS-4361C were used. Values of δ18 O and δ2 H are expressed relative to VSMOW. RESULTS SPR Floodwater Stable O and H isotope data from four gauge sites in area 1 (Gungle et al., 2017) and from area 3 (Table 1; see locations in Figures 1 and 2) are shown in Figure 3; the data do not represent all ﬂood events during the intervals indicated. The ranges at both sites are broad, reﬂecting the isotopic heterogeneity of individual rain events in the region (Eastoe and Dettman, 2016). Low values of δ18 O and δ2 H from area 3 between Septem-

ber 19 and October 4, 2014, followed heavy rain generated by hurricane Odile; in Tucson, associated rain had (δ18 O, δ2 H) values of (−15.2‰, −125‰), evolving to (−9.3‰, −73‰). Shallow Riparian Groundwater Data for area 1 form a linear trend with a slope of 5.6, with two clusters corresponding to gaining and losing reaches (Figure 4), and these data were interpreted to indicate mixing between summer runoﬀ that dominates recharge in losing reaches and basin groundwater containing winter recharge from the Huachuca Mountains in gaining reaches (Baillie et al., 2007). Data for areas 2 and 3, including Benson (Hopkins et al., 2014) and the riverbed near the Apache Nitrogen factory (Beilke, 2006), are from losing reaches of the river and plot with the losing-reach data of area 1. The data form a linear trend having a slope of 3.4 (omitting one outlier that probably represents

Figure 3. δ2 H vs. δ18 O values for ﬂoodwater in the San Pedro River. Data for area 3, 2014, are for ﬂoodwater following rain associated with Hurricane Odile. GMWL = global meteoric water line.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

467

Eastoe

tween the Huachuca Mountains and the SPR (Wahi et al., 2008). The trend with highest slope, 5.9, is for the Lewis Springs site, and it represents the river reach upstream of that site. The slope is like that of the mixing trend for SRGW in area 1 (Figure 4) and indicates mixing between basin groundwater and summer runoff. The lowest slope, 3.7, is for the Tombstone site, and it is generated by evaporation of base flow from the Charleston site, 14 km upstream. At the Hereford and Charleston sites, intermediate slopes suggest combinations of mixing and evaporation in the reaches upstream of those sites. Area 2 Figure 4. δ2 H vs. δ18 O values for shallow riparian groundwater (SRGW) in the floodplain of the San Pedro River. Also shown are estimated seasonal weighted mean isotope data for precipitation (Ppt), based on data for Tucson Basin with altitude corrections (Eastoe and Dettman, 2016). “Ppt 1400 m all” signifies means for all precipitation at an altitude of 1400 masl. “Ppt 1400 m wettest 30%” signifies means for precipitation in the wettest 30% of months at an altitude of 1400 masl. Data for area 1 are distinguished as belonging to gaining and losing reaches of the river, following Baillie et al. (2007). GMWL = global meteoric water line.

recharge from a particular flood event) and appear to define an evaporation trend originating near the summer end member of a modified local meteoric water line (LMWL) defined by amount-weighted means for precipitation falling during the wettest 30 percent of months. Recharge to basin fill in southern Arizona occurs mainly during the wettest months (Eastoe and Towne, 2018). The modified LMWL was calculated for an altitude of 1400 masl using precipitation data and isotope lapse rates for Tucson Basin (Eastoe and Dettman, 2016; Eastoe and Wright, 2019). Modified LMWLs up to 1700 masl differ little from the LMWL for 1400 masl. Shallow riparian groundwater (SRGW) in losing reaches of areas 1, 2, and 3 represents isotopically integrated bank storage of summer floodwater, consistent with the near absence of winter floods (Figure 7 of Cordova et al., 2015), and it has undergone evaporation since integration, suggesting repeated discharge and infiltration in the river channel. Area 1 The data (Figure 5) are from Gungle et al. (2017), and they represent samples collected between December and June (in order to avoid monsoon runoff) at four U.S. Geological Survey stream gauging stations: Hereford, Lewis Springs, Charleston, and Tombstone (Figure 1). For each site, the (δ18 O, δ2 H) values define a statistically significant linear trend, which is compared with data for groundwater in the basin alluvium be-

468

Along the axis of the SPV in area 2, a unit of impermeable clay-rich sediment 200 m thick, locally termed the Benson Clay, separates a shallow riparian aquifer from a deep, confined aquifer (Hopkins et al., 2014). Warm (23–26°C), flowing artesian groundwater discharges through the Benson Clay at the uncapped Dunlevy well and within the St. David Cienega (Figure 2). Springs in the bed of the SPR up to 100 m upstream of Escalante Crossing (Seeps A in Figure 2) apparently discharge water that moves upward through the Benson Clay. A second set of seeps (Seeps B, active in 2014 and 2015, but dry in February 2018) discharges water 1–2 m above the riverbed from a low cliff face about 500 m upstream of Escalante Crossing. This water flows along the top surface of the Benson Clay, probably in a buried paleochannel separate from the present river course. A perennial reach commonly begins at these seeps and at times continues beyond the Highway 80 bridge. Confined-aquifer groundwater in the Benson–St. David area ranges in isotope composition (Figure 6) from (δ18 O, δ2 H) = (−7.4‰, −49‰), similar to shallow riparian groundwater, to (−11.8‰, −85‰). Higher (δ18 O, δ2 H) values are observed near the SPR (Hopkins et al., 2014). The least-evaporated samples from St. David Cienega, near (−8.3‰, −60‰), differ from confined groundwater near the river. Groundwater from atop the Benson Clay (Seeps B) has (δ18 O, δ2 H) values consistent with base flow measured at the Tombstone gauge (Figure 5D). Groundwater from Seeps A changed in (δ18 O, δ2 H) values between 2014 and 2018. In 2018, three seeps were discharging water that plots as shallow riparian groundwater. In 2015, a sample from one of the seeps, marked by growth of orange algae, yielded (−8.9‰, −64‰), and in 2014, a nearby seep was discharging water with (−7.5‰, −55‰). A riverbed seep sampled near Curtis Station in 2006 gave (−8.9‰, −66‰). Four samples of base flow from the river plot in a linear array between shallow riparian groundwater

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

Sources of Perennial Water, San Pedro Valley, Arizona

Figure 5. δ2 H vs. δ18 O values for base ﬂow in area 1, corresponding to sampling between December and June over several years at four stream gauge sites (Gungle et al., 2017), and groundwater in the central part of the alluvial basin (Baillie et al., 2007; Wahi et al., 2008). GMWL = global meteoric water line. P < 0.0001 for all correlation coeﬃcients R.

and Tombstone gauge base flow. At least four sources of water contribute to base flow in area 2: (1) SRGW matching the evaporation trend of Figure 4; (2) base flow matching that from the Tombstone gauge (Figure 5); (3) confined-aquifer discharge with low values of (δ18 O, δ2 H), e.g., (−8.9‰, −64‰); and (4) apparent confined-aquifer discharge with isotope composition close to that of bank storage. The contributions of (2) and (3) cannot be distinguished in Figure 6 because the data are close to collinear. Discharge of type (4) is likely to be masked by bank storage. The change in isotope composition at Seeps A is discussed below. The confined aquifer appears to be compartmentalized in isotope composition beneath area 2, yielding water of different isotope composition at neighboring sites (Seeps A, St. David Cienega, Dunlevy Wells, and a number of supply wells near St. David). Type (4) water from wells near the SPR south of St. David (Hopkins et al., 2014; data reproduced in Supplemental Table S1) contains 10 to 20 pMC. It resembles confined-aquifer groundwater to the east of the SPR (Hopkins et al., 2014) and was probably recharged at the eastern flanks of the basin. It is not river water drawn recently into the confined aquifer as a result of pumping near St. David.

Area 3 The sub-basin in this area is separated from the sub-basin upstream by the granite sill at the Benson Narrows (Figure 1). Basin-fill sediment in area 3 consists mainly of semi-consolidated conglomerate of the Quiburis Formation. The river channel is confined to the post–20 ka trench, which is filled to a depth of at least 40 m with fluvial and clay-rich wetland or lacustrine sediments (Eastoe and Clark, 2018). The clay-rich sediments are present in the post–20 ka trench throughout area 3 (Cordova et al., 2015), forming impermeable lenses that separate a thin layer of active fluvial deposits from deeper gravel and sand deposits. A shallow riparian aquifer, possibly discontinuous, is perched above the clay-rich units, and a principal aquifer, continuous with a less-productive regional aquifer in the Quiburis Formation, occurs below the clay-rich unit. Recharge in area 3 consists of about-equal amounts of winter and summer precipitation from the wettest 30 percent (approximately) of months (Eastoe and Towne, 2018). Base flow originates consistently near the confluence of Red Rock Creek with the SPR. In early 2007, base flow extended north to site DTW (Figure 2). Since

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

469

Eastoe

Figure 6. δ2 H vs. δ18 O values for area 2, showing data for: base flow in the San Pedro River (SPR); groundwater seeping into the riverbed; groundwater from the confined aquifer in basin alluvium, sampled from water supply wells (Hopkins et al., 2014) and at St. David Cienega (this study). Linear trends are from Figure 3 (shallow riparian groundwater, SRGW) and Figure 5 (base flow, BF, at Tombstone). CS = Curtis Station. GMWL = global meteoric water line. Locations are shown in Figure 2.

spring 2007, base flow has not extended beyond site UTW, and by June 2018, it had contracted to about 3 km beginning at Red Rock Creek (The Nature Conservancy, 2020). From 2007 to 2014, base flow at site TLC (location shown in Figure 2) commonly had (δ18 O, δ2 H) near (−8.2‰, −59‰), and a tritium content of 1.0–1.7 TU (five of six samples; see Supplemental Table S1). In 2007–2009, base flow at downstream sites HSR, UTW, and DTW had a broader range of (δ18 O, δ2 H). Values lower than those at TLC may result from local discharge of groundwater. Samples with higher values plot in an array that diverges from the global meteoric water line with increasing distance downstream and incorporate some evaporated irrigation reflux (originally groundwater of isotope composition like that of base flow at TLC). Samples from site DTW form a linear trend of slope 2.7, which could be an evaporation trend, but the slope is lower than usual for southern Arizona (Eastoe and Towne, 2018). A more nuanced explanation might involve progressive evaporation of base flow from TLC and mixing with SRGW (Figure 4). The downstream increase of tritium to 2.1– 2.4 TU at site HSR (Table 1) is consistent with the latter suggestion.

470

Base flow at site TLC resembles neither SRGW nor confined groundwater immediately upstream of the Benson Narrows (Figure 7). The (δ18 O, δ2 H) values of the base flow resemble those in groundwater associated with nearby catchments of similar elevation on the east flank of the sub-basin (Eastoe and Clark, 2018). Area 4 Volcanic-derived conglomerate of the San Manuel Formation is faulted against fluvial/alluvial conglomerate of the Quiburis Formation in this area. Post– 20 ka sediments fill a trench 100–300 m wide incised into the San Manuel Formation near the fault (Cook et al., 2010), except where an outcrop of impermeable, argillaceous beds of the San Manuel Formation occurs in the riverbed (32.3395°N, 110.4176°W). A cluster of seeps occurs 200–300 m upstream of that outcrop. Base flow, here termed the Gamez Road interval, extended about 200 m from the seeps in early 2014, but by June 2019, it had disappeared (The Nature Conservancy, 2020). Eastoe and Clark (2018) provided evidence for parallel streams of groundwater of distinctive isotope

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

Sources of Perennial Water, San Pedro Valley, Arizona

Figure 7. δ2 H vs. δ18 O values for area 3. Inset shows detail of Three Links Crossing (TLC) samples. BF = base ﬂow, GW = groundwater, GMWL = global meteoric water line, SRGW = shallow riparian groundwater. Site name abbreviations (cf. Figure 2, area 3): TLC = Three Links Crossing; HSC = Heaven Sent Ranch; UTW = upstream of Teran Wash; DTW = downstream of Teran Wash.

composition, probably localized within paleochannels of the river between Hot Springs Canyon and the Gamez Road base-ﬂow interval. To the east of the SPR, samples from numerous wells delineate an aquifer in which groundwater is derived from Hot Springs Canyon. Close to the river channel, a stream of groundwater like that in area 3 has been traced as shown in Figure 2; no well samples are available further downstream. Groundwater from Paige Canyon is a third distinctive type (Figure 8) that has not been

traced beyond the end of Paige Canyon. Of three seep samples taken in March 2014 (“Unnamed Spring” in Supplemental Table S1), two resemble groundwater from Hot Springs Canyon, and one may be a mixture of groundwater from Hot Springs Canyon and area 3 (Figure 8). Base ﬂow sampled at its downstream limit in March 2014 consisted mainly of groundwater from area 3. In November 2014, the seeps were discharging bank storage from the large ﬂood generated by Hurricane Odile in September 2014.

Figure 8. δ2 H vs. δ18 O values for area 4. GW = groundwater, GMWL = global meteoric water line, SRGW = shallow riparian groundwater.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

471

Eastoe

Area 5 Holocene fluvial sediments in the post–20 ka trench broaden to a width of 1 km in this section of the SPV and are flanked on both sides by a further 2 km of Pleistocene alluvium (Cook et al., 2010) overlying lacustrine facies of the Quiburis Formation (Dickinson, 2003). Edgar and Buehman Canyons, both draining high elevations of the Santa Catalina Mountains, join the SPR near Redington, and upstream of Bingham Cienega, there is a spring-fed wetland 0.4 km north of the confluence of Edgar Canyon. Lower reaches of both canyons intersect the Buehman Canyon tilt block (Dickinson, 2003), consisting largely of eastdipping late Paleozoic limestone and Cretaceous shale and phyllite (Bykerk-Kaufmann, 1990). Base flow was observed in the SPR adjacent to Bingham Cienega between 1998 and 2001; since 2001, base flow has been absent. In Figure 9, values of (δ18 O, δ2 H) from Bingham Cienega and base flow in the adjacent SPR are compared with data for groundwater from areas 3 and 4 (which supports base flow in area 4), and for surface water and groundwater from Edgar and Buehman Canyons. The following observations are offered: (1) base flow and spring water at Bingham Cienega are closely similar in isotope composition; (2) base flow has lower values of (δ18 O, δ2 H) than groundwater from area 3 and from Hot Springs Canyon in area 4; (3) water in Edgar Canyon is isotopically similar to ground-

water associated with Paige Canyon; (4) three samples of water in Buehman Canyon on average have higher (δ18 O, δ2 H) values and appear to be more subject to evaporation than water from Edgar Canyon. Samples from base flow, Bingham Cienega, and Buehman Canyon contained 4.5, 4.0, and 5.0 TU, respectively, in June 2001. Bingham Cienega and adjacent base flow appear to share a common source of recharge that was at most 5 years old in 2001 relative to average Tucson rainwater containing 5.3 TU (Eastoe et al., 2011). Water from the seeps at Gamez Road (area 4) cannot account for base flow at Bingham Cienega without addition of water with lower (δ18 O, δ2 H). Such water could originate from Edgar, Buehman, or Paige Canyons. Of these, Edgar and Buehman Canyons are the more likely because they are closer to the cienega. Water from Paige Canyon resembles that in Edgar Canyon because of source areas at altitudes above 2000 masl in each case. A mixture of water from Edgar and Buehman Canyons could account for water in base flow and Bingham Cienega. DISCUSSION Sills and Sub-Basins Sills of impermeable rock near Sierra Vista, at the Benson Narrows, and near Gamez Road divide the study area into four hydrologic sub-basins. Prior to

Figure 9. δ2 H vs. δ18 O values for area 5. GW = groundwater, SW = surface water, GMWL = global meteoric water line. Precipitation 2420 masl points show amount-weighted mean winter (W) and summer (S) means for an altitude of 2420 masl in the Santa Catalina Mountains (Eastoe and Wright, 2019).

472

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

Sources of Perennial Water, San Pedro Valley, Arizona

Figure 10. δ2 H vs. δ18 O summary diagram for base ﬂow throughout the study area. Bold black arrows indicate isotope evolution corresponding to cases A, B, and C as discussed in the text. BF = base ﬂow, GW = groundwater, GMWL = global meteoric water line, SRGW = shallow riparian groundwater. Ppt 2420 points show amount-weighted mean winter (W) and summer (S) means for an altitude of 2420 masl in the Santa Catalina Mountains (Eastoe and Wright, 2019).

development, groundwater would have crossed each sill as base flow or sub-flow through fluvial sediment in the post–20 ka trench. Observations near the Benson Narrows (“Shallow Riparian Groundwater” section) indicate no base flow for many years, consistent with overdraft of groundwater in the Benson–St. David area upstream of the sill (Cordova et al., 2015). At the Charleston gauge (Figure 1), base flow in 2012 was less than half the typical amount in the 1930s (Gungle et al., 2017, Figure 31). At Gamez Road, base flow had almost vanished by 2018 and was absent in 2019 (The Nature Conservancy, 2020). The SPV in the study area is becoming a set of sub-basins for which the only significant hydrological connection is summer monsoon floodwater. Separation appears to be complete at the Benson Narrows, close to complete at Gamez Road, and well advanced near Sierra Vista.

Isotope Evolution in Base Flow Figure 10 is a summary of the isotope compositions in base flow from area 1 to area 5. Three possibilities for isotope evolution relating base flow in the five areas can be postulated. Case A consists in sub-flow originating upstream of each area and forced to the surface above each impermeable rock sill. Progressive downstream increases in (δ18 O, δ2 H) along an evaporation trend of slope near 4 may occur if flow is at the surface. Case B consists in mixing of sub-flow from upstream with shallow riparian groundwater recharged from summer floods, possibly resulting in downstream increases in (δ18 O, δ2 H), as appears to occur in area

2. Case C is characterized by downstream decreases in (δ18 O, δ2 H) as a result of additions of low–(δ18 O, δ2 H) water originating from the higher, well-watered mountain blocks flanking the river basin. All three mechanisms are likely to operate; isotope data allow the identification of the dominant mechanism in a particular area. Case A is dominant between the Charleston and Tombstone gauges in area 1, resulting in the evaporation trend observed at the Tombstone gauge (Figure 2). Case B is observed in area 2, where base flow plots between the trends for base flow at the Tombstone gauge and for shallow riparian groundwater (Figures 5 and 10). Case C applies in areas 3 and 5 (Figure 10) and between the Hereford and Lewis Springs gauges in area 1 (Figure 5). Between the Lewis Springs and Charleston gauges, cases A and C appear to operate together. In area 4, case B applies to water contributed from Hot Springs Canyon, and case A applies to groundwater from area 3, but without an isotope evaporation signature (Figure 8). Mountain-Derived Water Baillie et al. (2007) and Gungle et al. (2017) determined that shallow riparian groundwater and base flow in gaining reaches of area 1 contain a large fraction of groundwater that can be traced to the Huachuca Mountains on the west flank of the river valley in that area (Figures 1 and 5). In area 5, a simple interpretation might identify base flow and groundwater emerging in Bingham Cienega as evaporated amount-weighted mean winter precipitation from near

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

473

Eastoe

2420 masl (Figure 10). Such an interpretation is oversimplified, however, because observed values in tributary surface water (Figure 9), base flow, and water in the cienega are likely to be mixtures of water from Edgar and Buehman Canyons. Water in Edgar Canyon is evaporated relative to the LMWL defined by the amount-weighted seasonal means for 2420 masl, and it would originate from precipitation with (δ18 O, δ2 H) values lower than those of mean winter precipitation. The data for base flow in Edgar Canyon represent the years 1998, 1999, 2001, and 2012 and are therefore unlikely to be influenced by individual seasons with low– (δ18 O, δ2 H) precipitation. Cunningham et al. (1998) observed comparable low (δ18 O, δ2 H) values in the discharge of springs between 1400 and 2000 masl. The reason for preferential infiltration of water with (δ18 O, δ2 H) values lower than those of average winter precipitation is not understood; it is not an isotope effect related to the amount or monthly intensity of highelevation precipitation (Eastoe and Wright, 2019). Area 1 Base flow in area 1 is dependent partly on discharge from a sub-basin in Mexico, south of Palominas (Pool and Dickinson, 2007, Figure 2), and partly on discharge from the alluvial aquifer between the Huachuca Mountains and the SPR, focused between Hereford and Lewis Springs (Figure 5). The latter groundwater flows east from the Huachuca Mountains (Pool and Dickinson, 2007, Figure 5) and originates largely as precipitation in the mountains (Wahi et al., 2008). Area 2 Sources of base flow in area 2 are: (1) bank storage, mainly recharge from summer floodwater, (2) discharge from area 1, (3) and (4) discharge from the confined aquifer beneath the Benson Clay (Figure 6). The relative importance of the sources cannot be established from the isotope data. Two riverbed discharge zones of confined groundwater have been found; others may exist in the area, or they may have existed prior to intensive pumping of the confined aquifer at St. David. The temporal change in isotope content at Seeps A from confined-aquifer to bank-storage discharge (Figure 6) indicates that the head driving discharge from the confined aquifer is at times exceeded by the head of water in the river during summer floods, leading to recharge rather than discharge. Area 3 Base flow in area 3 might also be interpreted in three alternative ways: (1) as evaporated winter precipita-

474

tion from 2420 masl (indicated by the line of slope 4 in Figure 10), (2) as a mixture of confined and shallow riparian groundwater from upstream of the Benson Narrows (Figure 7), or (3) as groundwater originating from the watershed of Red Rock Creek. Alternative (1) is not reasonable because the watershed of Red Rock Creek is mainly at altitudes of 1150–1400 masl, with a small area near 1600 masl. Alternative (2) would require about-equal proportions of the confined and shallow riparian groundwater, but it lacks a means to transfer water across the granite sill separating the sub-basins. No base flow was observed in the riverbed from 2006 to 2009 when a stream gauge was operated at the sill (U.S. Geological Survey, 2019) or since (personal observation). Haney and Lombard (2005) proposed that groundwater from the sub-basin upstream of the granite sill flows through alluvium to the east and west of the granite hills in the area. Drewes (1974) constructed a cross section across the SPV at the Benson Narrows, showing a granite barrier extending continuously across the valley, locally capped by alluvium deposited on a surface that lies slightly higher than the riverbed. If this is correct in detail, confined and shallow riparian groundwater leaves the upstream sub-basin along the river channel, and only unconfined groundwater from alluvium along the basin flanks could flow around the granite sill. Alternative (3) is favored by the spatial association of base flow with Red Rock Creek and the close isotopic similarity between the base flow at site TLC and other groundwater of local origin in the sub-basin below the Benson Narrows (Eastoe and Clark, 2018). Red Rock Creek is unusual in generating a perennial reach of the river. Kelsey and Teran Washes, which drain watersheds of comparable size and elevation, do not generate perennial flow in the river, even though impermeable clay is present in the riverbed near their junctions. This suggests the presence of an unusual groundwater storage in the watershed of Red Rock Creek. Such a storage may be associated with sediment deposited by a major tributary that was located near Red Rock Creek at the time of deposition of the Quiburis Formation. The tributary deposited a large alluvial fan, of which the eroded remnant is visible in the Quiburis Formation (Figure 2). Area 4 Hot Springs Canyon also supplies enough water to feed seeps in the riverbed. In this instance, much of the sub-flow originating in the canyon derives from a set of hot springs in the upper watershed (Eastoe and Clark, 2018). Groundwater supplied from Hot Springs Canyon is subject to pumping for irrigation

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

Sources of Perennial Water, San Pedro Valley, Arizona

Figure 11. Drought status at four stations in southeastern Arizona, from 1896 to present. Shading indicates drought conditions, and white indicates non-drought conditions, as indicated by annual average precipitation (upper figure) and the Palmer Drought Severity Index (lower figure), both averaged over the time interval indicated by each box. Long-term average precipitation is given below each site name. Data are from WestWide Drought Tracker, 2019, 2–10). Site locations are: Tombstone (31.7147°N, 110.0659°W), Pearce (31.9410°N, 109.8363°W), Safford (32.8149°N, 109.6804°W), Tucson (32.1242°N, 110.9403°W).

immediately up-gradient of the seeps; prior to development, this aquifer may have supplied a larger fraction of base flow in area 4. Area 5 At Bingham Cienega, the following hypothesis (illustrated in Figure 2) is proposed for the origin of groundwater and former base flow in the adjacent SPR. Both Edgar and Buehman Canyons cross east-dipping limestone of the Pennsylvanian–Permian Naco Group. The limestone is overlain unconformably by Cretaceous Bisbee Group strata, largely shale and phyllite, in the area near Edgar Canyon (BykerkKaufmann, 1990). Recharge into limestone occurs beneath both canyons, and groundwater flows east through the tilted limestone strata. Over much of the area, the groundwater is confined beneath the Bisbee Group. At the fault bounding the relatively impermeable lacustrine basin-fill sediments of the SPV, the groundwater flows upwards along the fault, and discharge is focused near Bingham Cienega. Flow through the karst aquifer is rapid, as indicated by the tritium measurements in 2001 (Supplemental Table S1). The aquifer supplying Bingham Cienega and perennial flow in the nearby reach of the SPR are therefore vulnerable to drought at decadal or shorter timescales. Discharge was plentiful in the cienega prior to the early 2000s, and base flow was present in the river. Subsequently, discharge has been largely absent. The change probably reflects a regional change from relatively wet years to drought in the late 1990s (Figure 11).

Natural or Human Causes At Charleston, a steady decline in base flow has been documented over the period 1934–2012 (Gungle et al., 2017). Long-term drought conditions began in the 1940s at the nearby climate station in Tombstone; between 1941 and 2019, negative Palmer Drought Severity Index values indicative of drought applied in 56 of 78 years (Figure 11; data from WestWide Drought Tracker, 2019). Intervals of drought in neighboring areas of southeastern Arizona have not corresponded exactly to those in Tombstone and may not correspond in other parts of the SPV. Further indications of longterm drought in the SPV are found in the alluvial aquifer in Hot Springs Canyon, Cascabel (Figure 2), where pumping demands are small, and where longterm declines in groundwater levels have been documented since 1993 (Eastoe and Clark, 2018). Similar declines have been observed in several, but not all, mountain-front wells in the area near Benson and Cascabel (Cordova et al., 2015). Climate in southwest North America was wetter during the Little Ice Age. At the peak of the Little Ice Age, extensive standing water was present in Lake Palomas, Chihuahua, in a basin that is typically dry at present (Castiglia and Fawcett, 2006). Silver Lake in the Mojave River drainage, southern California system, was flooded at 390 ± 90 years before present but is at present typically dry (Enzel et al., 1989). Large floods were more common than at present in Arizona streams during an interval of wet climate beginning about A.D. 1400 (Ely, 1997), and ending since A.D. 1650 at a time that cannot be constrained

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

475

Eastoe

precisely from calibrated radiocarbon dates because of cyclic fluctuations in the production rate of 14 C between A.D. 1650 and 1950 (Stuiver and Becker, 1993). Nonetheless, a drying trend is likely since A.D. 1800. Natural drying at decadal to centennial timescales in the SPV has been exacerbated by human exploitation of groundwater, and it may be further affected by human-caused climate change. Demand for groundwater declined between 1970 and 2010 across the SPV as a result of retirement of irrigated agricultural land and increased water-use efficiency in Sierra Vista; nonetheless, estimated outflow exceeded inflow in the Sierra Vista, Benson, and Cascabel-Redington sub-basins of the SPV (for sub-basins as defined in their works, see Cordova et al., 2015; Gungle et al., 2017). In all three sub-basins, human demand takes up a large fraction of estimated inflow (Table 1). Errors in the water budgets are large, especially for the evapotranspiration component of outflow. In the case of Sierra Vista sub-basin, the imbalance between inflow and outflow in 2012 was estimated with an error at −6.2 ± 6.8 hm3 /yr. Errors notwithstanding, human demand in the Sierra Vista sub-basin is very likely unsustainable at present, and it was clearly unsustainable in 2002, when the imbalance was −13.3 ± 6.8 hm3 /yr (Gungle et al., 2017). Declining base flow in the SPR is therefore most likely an effect of persistent drought combined with human demand for groundwater, even as human demand has decreased. Implications for Management of the SPV The sub-basins of the SPV will become increasingly reliant on local water sources, i.e., surface water and groundwater supplied by tributaries, as their hydrological separation develops. Groundwater in areas 3 and 4, within the sub-basin downstream of the Benson Narrows, will be little affected by additional pumping in area 2, upstream of Benson Narrows. Declining water levels in area 2 have already separated the two subbasins. Within areas 3 and 4, the future of perennial flow in the SPR and groundwater supply will be dominantly affected by climate and by local pumping. Monsoon floodwater appears to have little role in recharge in these areas. In area 1, delayed effects of pumping to date are likely to cause capture of some of the river flow, regardless of future measures to manage groundwater (Gungle et al., 2017). Artificial recharge of stormwater from the City of Sierra Vista is an attempt at capturing the abundant summer runoff that otherwise contributes little to recharge beyond losing reaches of the riparian zone (Figure 4; Baillie et al., 2007). Lowdensity urban development south of Sierra Vista re-

476

lies on private supply wells for water pumped from the regional aquifer, competing for water that would otherwise discharge into the reach between Hereford and Lewis Springs. Increased density of urban development south of Sierra Vista will combine with the effects of drought to reduce the volume of base flow in the river. In area 2, any lessening of the confined-aquifer head as a result of increased pumping for proposed urban development near Benson risks permanently reversing the discharge of confined-aquifer groundwater in and near the SPR near St. David. Over-pumping causes long-lasting changes in artesian aquifers, as in the case of the Roswell Basin, New Mexico, where over-exploitation of groundwater has decreased the head of formerly artesian groundwater by tens of meters (Havenor, 1996). In area 2, likely effects of over-exploitation include reversal of groundwater flow from the confined aquifer, draining of shallow riparian groundwater from the riverbed alluvium, reduction or elimination of perennial flow in in the river, and drying of the St. David Cienega. A change in head of 1–2 m (typical floodwater depth) may be enough to reverse the direction of groundwater flow, on the indication of changes in isotope composition of the seeps at Escalante Crossing. Loss of perennial flow in this area would eliminate the only dependable source of surface water between the Tombstone gauge and Red Rock Creek, a distance of 65 km. In areas 3 and 4, long-term decline in base flow reflects a combination of pumping and drought effects as observed in Hot Springs Canyon (Eastoe and Clark, 2018). Drought-related drainage of groundwater above the level of the SPR in tributary washes is likely to be a regional phenomenon. The retirement of irrigated land in area 3 (Haney and Lombard, 2005) has no doubt slowed the shortening of the perennial reach in that area but cannot address drought as a factor contributing to the decline of groundwater discharge. In area 4, the effects of droughtrelated drainage from Hot Springs Canyon will probably take several decades to reach the declining perennial reach, on the indication of tritium data (Eastoe and Clark, 2018). Perennial flow in this reach will also depend on the supply of groundwater from area 3, with a similar lag time relative to drought and pumping demand. In area 5, both the SPR and Bingham Cienega are expected respond to climate fluctuations. If the interpretation of groundwater origin in area 5 is correct, the presence of base flow in the SPR and groundwater discharge to Bingham Cienega will depend on climate changes at decadal or shorter timescales, and will be little affected by pumping for irrigation along the SPR upstream of Bingham Cienega.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

Sources of Perennial Water, San Pedro Valley, Arizona

Future Work Area 1 has been studied in greater detail than areas 2 to 5 because of its ecological importance and because of its high degree of urban development. Further study of areas 2 to 5 would improve the interpretations presented here. Area 2 is of particular importance because of proposals for urban development equal to that in area 1. CONCLUSIONS 1. Four reaches of the SPR in the study area currently have ecologically important base flow. A fifth reach, near Redington, had base flow until the early 2000s. Base flow in the SPR is declining, both in volume and in the lengths of river reaches with perennial flow. 2. The SPV in the study area comprises four subbasins separated by impermeable rock sills at the level of the river. Base flow crosses sills near Sierra Vista and north of Cascabel near Gamez Road, but it has almost disappeared in the latter case. No base flow crosses the Benson Narrows sill at present. 3. Recharge in losing reaches of the SPR from summer floodwater results in bank storage and shallow riparian groundwater with stable O and H isotope data forming an evaporation trend. 4. In area 1, south of Sierra Vista, stable O and H isotope data indicate addition of alluvial basin groundwater originating in the Huachuca Mountains to SPR base flow. 5. In area 2, near St. David, base flow is a combination of water from area 1, local bank storage, and deep-basin groundwater. Small changes of head in the deep-basin groundwater in this area will result in reversal of groundwater flow and recharge to the deep-basin aquifer from the riverbed. 6. In area 3, north of the Benson Narrows, stable O and H isotope and tritium data suggest that base flow originates from a local tributary drainage rather than upstream in the SPV. 7. In area 4, in Cascabel near Gamez Road, a single base-flow sample indicated a water origin mainly in the SPV alluvial aquifer immediately upstream. Riverbed seeps also contribute water from Hot Springs Canyon. 8. In area 5, near Redington, former base flow originated as precipitation from neighboring high mountains, and it was apparently supplied to the riverbed through a limestone aquifer. Tritium data are consistent with transit times of a few years and indicate the vulnerability of this reach and nearby wetlands to short-term drought. 9. High human demand for groundwater in the SPV in recent decades has coincided with a period of dry

climate at decadal to centennial timescale. Lessening agricultural demand for groundwater since 1970 has not yet resulted in observable recovery of base flow in the SPR. Large additional human demands for groundwater will exacerbate the decline of base flow. SUPPLEMENTAL MATERIAL Supplemental Material associated with this article can be found online at https://doi.org/10.2113/EEGxxxx. ACKNOWLEDGMENTS The author expresses his gratitude to: Pima County Regional Flood Control District and Pima Association of Governments for permission to use their data from the Bingham Cienega area, and to Pima County employees Jennifer Becker, Brian Powell, Julia Fonseca, and Gregory Hess for assistance in sampling; Apache Nitrogen Products, Inc., employees Greg Hall for assistance with sampling and Pamela Beilke for permission to use data; Mary McCool and the Community Watershed Alliance for organizing sampling at St. David Cienega; the Cascabel community for access to their property and private wells; Robert Rogers and Barbara Clark of The Nature Conservancy for access to a river sampling site and assistance with sampling in Cascabel; Bureau of Land Management employees David Murray and Benjamin Lomeli for access to the San Pedro Riparian National Conservation Area; and Lynwood Hume for collecting floodwater samples in Cascabel when the author was unable to be present. The input of two anonymous reviewers helped greatly in improving the quality of the manuscript. REFERENCES Abatzoglou, J. T.; McEvoy, D. J.; and Redmond, K. T., 2017, The West Wide Drought Tracker: Drought monitoring at fine spatial scales: Bulletin American Meteorological Society, Vol. 98, No. 9, pp. 1815–1820. Arizona Daily Star, 2016, Mega-Project in Benson Casting Shadow over Kartchner Caverns Water Supply: Electronic document, available at https://azdailysun.com/news/local/ mega - project - in - benson - casting - shadow - over - kartchner caverns-water/article_4b88974e-753e-52e0-ad8a-bebef18e0a12. html Baillie, M. N.; Hogan, J. F.; Ekwurzel, B.; Wahi, A. K.; and Eastoe, C. J., 2007, Quantifying water sources to a semiarid riparian ecosystem, San Pedro River, Arizona: Journal Geophysical Research, Vol. 112, G03S02. doi:10.1029/2006JG000263. Beilke, P., Apache Nitrogen Products Inc., Benson, AZ, personal communication. Blumstock, M.; Tetzlaff, D.; Malcolm, I. A.; Nuetzmann, G.; and Soulsby, C., 2015, Baseflow dynamics: Multi-tracer surveys to assess variable groundwater contributions to mon-

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

477

Eastoe tane streams under low flows: Journal Hydrology, Vol. 527, pp. 1021–1033. Brand, L. A.; Cerasale, D. J.; Rich, T. O.; and Krueper, D. J., 2009, Breeding and migratory birds: Patterns and processes. In Stromberg, J. C. and Tellman, B. (Editors), Ecology and Conservation of the San Pedro River: The University of Arizona Press, Tucson, AZ, pp. 153–174. Brooks, J. R.; Wigington, J.; Phillips, D. L.; Comeleo, R.; and Coulombe, R., 2012, Willamette River Basin surface water isoscape (delta O-18 and delta H-2): Temporal changes of source water within the river: Ecosphere, Vol. 3, No. 5, article 39. doi:10.1890/es11-00338.1. Bryan, K., 1925, Date of channel entrenching (arroyo cutting) in the arid southwest: Science, Vol. 62, No. 1607, pp. 338–344. Bykerk-Kaufmann, A., 1990, Kinematic Analysis of Deformation at the Margin of a Regional Shear Zone, Buehman Canyon Area, Santa Catalina Mountains, Arizona: Unpublished M.S. Thesis, Department of Geosciences, University of Arizona, Tucson, AZ, 79 p. Castiglia, P. J. and Fawcett, P. F., 2006, Large Holocene lakes and climate change in the Chihuahuan desert: Geology, Vol. 34, No. 2, pp. 113–116. Coes, A. L. and Pool, D. R., 2007, Ephemeral-stream channel and basin-floor infiltration and recharge in the Sierra Vista subwatershed of the Upper San Pedro Basin, southeastern Arizona. In Stonestrom, D. A.; Constantz, J.; Ferre, T.; and Leake, S. A. (Editors), Ground-Water Recharge in the Arid and Semiarid Southwestern United States: U.S. Geological Survey Professional Paper 1703-J, pp. 253–311. Cook, J. P.; Youberg, A.; Pearthree, P. A.; Onken, J. A.; MacFarlane, B. J.; Hadad, D. E.; Bigio, E. R.; and Kowler, A. L., 2010, Mapping of the Holocene River Alluvium Along the San Pedro River, Aravaipa Creek, and Babocomari River, Southeastern Arizona: Arizona Geological Survey Digital Map DM-RM-1, 76 p. Cordova, J. T.; Dickinson, J. E.; Beisner, K. R.; Hopkins, C. B.; Kennedy, J. R.; Pool, D. R.; Glenn, E. P.; Nagler, P. L.; and Thomas, B. E., 2015, Hydrology of the Middle San Pedro Watershed, Southeastern Arizona: U.S. Geological Survey Scientific Investigations Report 2013-5040, 77 p. Cunningham, E. E. B.; Long, A.; Eastoe, C. J.; and Bassett, R. L., 1998, Migration of recharge waters downgradient from the Santa Catalina Mountains into the Tucson Basin aquifer: Hydrogeology Journal, Vol. 6, No. 1, pp. 94–103. Dickinson, J. E.; Kennedy, J. R.; Pool, D. R.; Cordova, J. T.; Parker, J. T. C.; Macy, J. P.; and Thomas, B. E., 2010a, Hydrogeologic Framework of the Middle San Pedro Watershed, Southeastern Arizona: U.S. Geological Survey Scientific Investigations Report 2010–5126, 36 p. Dickinson, J. E.; Pool, D. R.; Groom, R. W.; and Davis, L. J., 2010b, Inference of lithologic distributions in an alluvial aquifer using airborne transient electromagnetic surveys: Geophysics, Vol. 75, No. 4, p. WA149–WA161. Dickinson, W. R., 1991, Tectonic Setting of Faulted Tertiary Strata Associated with the Catalina Core Complex in Southern Arizona: Special Paper 264, Geological Society of America, Boulder, CO, 106 p. Dickinson, W. R., 1998, Facies Map of Post–Mid-Miocene Quiburis Formation, San Pedro Trough, Pinal, Pima, Gila, Graham, and Cochise Counties, Arizona: Arizona Geological Survey Contributed Map CM-98-A, map scale 1:24,000, 10 map sheets. Dickinson, W. R., 2002, The Basin and Range Province as a composite extensional domain: International Geology Review, Vol. 44, No. 1, pp. 1–38.

478

Dickinson, W. R., 2003, Depositional facies of the Quiburis Formation, basin fill of the San Pedro trough, southeastern Arizona Basin and Range Province. In Raynolds, R. G. and Flores, R. M. (Editors), Cenozoic Systems of the Rocky Mountain Region: Society for Sedimentary Geology, Denver, CO, pp. 157–181. Drewes, H., 1974, Geologic Map and Sections of the Happy Valley Quadrangle, Cochise County, Arizona: U.S. Geological Survey Miscellaneous Investigations Series Map I-832, 1 sheet, scale 1:48,000. Eastoe, C. J. and Clark, B., 2018, Understanding the water resources of a small rural community: Citizen science in Cascabel, Arizona: Journal Contemporary Water Research and Education, Vol. 164, pp. 19–41. Eastoe, C. J. and Dettman, D. L., 2016, Isotope amount effects in hydrologic and climate reconstructions of monsoon climates: Implications of some long-term data sets for precipitation: Chemical Geology, Vol. 430, pp. 76–89. Eastoe, C. J. and Towne, D., 2018, Regional zonation of groundwater recharge mechanisms in alluvial basins of Arizona: Interpretation of isotope mapping: Journal Geochemical Exploration, Vol. 194, pp. 134–145. Eastoe, C. J.; Watts, C. J.; Ploughe, M.; and Wright, W. E., 2011, Future use of tritium in mapping pre-bomb groundwater volumes: Ground Water, Vol. 50, pp. 87–93. Eastoe, C. J.; and Wright, W. E., 2019, Hydrology of mountain blocks in Arizona and New Mexico as revealed by isotopes in groundwater and precipitation: Geosciences, Vol. 9, article 461. doi:10.3390/geosciences9110461. Ely, L. L., 1997, Response of extreme floods in the southwestern United States to climatic variations in the late Holocene: Geomorphology, Vol. 19, No. 3, pp. 175–201. Enzel, Y.; Cayant, D. R.; Anderson, R. Y.; and Wels, S. G., 1989, Atmospheric circulation during Holocene lake stands in the Mojave Desert: Evidence of regional climate change: Nature, Vol. 341, No. 6237, pp. 44–47. Fenneman, N. M., 1931, Physiography of Western United States, 1st ed.: McGraw-Hill, New York, NY, 534 p. Fujimoto, M.; Ohte, N.; Kawasaki, M.; Osaka, K.; Itoh, M.; Ohtsuka, I.; and Itoh, M., 2016, Influence of bedrock groundwater on streamflow characteristics in a volcanic catchment: Hydrological Processes, Vol. 30, No. 4, pp. 558–572. Glenn, E. P.; Scott, R. L.; Nguyen, U.; and Nagler, P. L., 2015, Wide-area ratios of evapotranspiration to precipitation in monsoon-dependent semiarid vegetation communities: Journal Arid Environments, Vol. 117, pp. 84–95. Gray, R. S., 1965, Late Cenozoic Sediments in the San Pedro Valley near St. David, Arizona: Unpublished M.S. Thesis, University of Arizona, Tucson, AZ, 198 p. Guay, B. E.; Eastoe, C. J.; Bassett, R. L.; and Long, A., 2006, Sources of surface and ground water adjoining the Lower Colorado River inferred by δ18 O, δD and 3 H: Hydrogeology Journal, Vol. 14, No. 4, pp. 146–188. Gungle, B. G.; Callegary, J. B.; Paretti, N. V.; Kennedy, J. R.; Eastoe, C. J.; Turner, D. S.; Dickinson, J. E.; Levick, L. R.; and Sugg, Z. P., 2017, Hydrological Conditions and Evaluation of Sustainable Groundwater Use in the Sierra Vista Subwatershed, Upper San Pedro Basin, Southeastern Arizona: U.S. Geological Survey Scientific Investigation Report 2016–5114, 90 p. Haney, J. and Lombard, J., 2005, Interbasin groundwater flow at the Benson Narrows, Arizona: Southwest Hydrology, March/April issue, pp. 8–9. Havenor, K. C., 1996, The Hydrogeologic Framework of the Roswell Groundwater Basin, Chaves, Eddy, Lincoln, and Otero

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

Sources of Perennial Water, San Pedro Valley, Arizona Counties, New Mexico: Unpublished Ph.D. thesis, Department of Geosciences, University of Arizona, Tucson, AZ, 274 p. Hereford, R. and Betancourt, J., 2009, Historic geomorphology of the San Pedro River. In Stromberg, J. C. and Tellman, B. (Editors), Ecology and Conservation of the San Pedro River: The University of Arizona Press, Tucson, AZ, pp. 232–250. Hogan, J.; Phillips, F.; Eastoe, C.; Lacey, H.; Mills, S.; and Oelsner, G., 2012, Isotopic tracing of hydrological processes and water quality along the upper Rio Grande, USA. In Monitoring Isotopes in Rivers: Creation of the Global Network of Isotopes in Rivers (GNIR): Tecdoc 1673, International Atomic Energy Agency, Vienna, Austria, pp. 111–136. Hopkins, C.; McIntosh, J.; Eastoe, C.; Dickinson, J. E.; and Meixner, T., 2014, Evaluation of the importance of clay confining units on groundwater flow in alluvial basins using solute and isotope tracers: The case of Middle San Pedro Basin in southeastern Arizona (USA): Hydrogeology Journal, Vol. 22, No. 4, pp. 829–849. doi:10.1007/s10040-013-1090-0. Huckleberry, G.; Lite, S. T.; Katz, G.; and Pearthree, P., 2009, Fluvial geomorphology. In Stromberg, J. C. and Tellman, B. (Editors), Ecology and Conservation of the San Pedro River: The University of Arizona Press, Tucson, AZ, pp. 251–267. Hughes, C. E.; Stone, D. J. M.; Gibson, J. J.; Meredith, K. T.; Sadek, M. A.; Cendon, D. I.; Hankin, S. I.; Hollins, S. E.; and Morrison, T. N., 2012, Stable water isotope investigation of the Barwon–Darling River system, Australia. In Monitoring Isotopes in Rivers: Creation of the Global Network of Isotopes in Rivers (GNIR): Tecdoc 1673, International Atomic Energy Agency, Vienna, Austria, pp. 95–110. MacNish, R.; Baird, K. J.; and Maddock, T., 2009, Groundwater hydrology of the San Pedro River Basin. In Stromberg, J. C. and Tellman, B. (Editors), Ecology and Conservation of the San Pedro River: The University of Arizona Press, Tucson, AZ, pp. 285–299. Maricopa Audubon Society, 2020, Birds of San Pedro Riparian National Conservation Area: Electronic document, available at https://www.maricopaaudubon.org/bird-checklistfor-san-pedro-riparian-national-conservation-area Nickolas, L. B.; Segura, C.; and Brooks, J. R., 2017, The influence of lithology on surface water sources: Hydrological Processes, Vol. 31, No. 10, pp. 1913–1925. Pfister, L.; Martinez-Carreras, N.; Hissler, C.; Klaus, J.; Carrer, G. E.; Stewart, M. K.; and McDonnell, J. J., 2017, Bedrock geology controls on catchment storage, mixing, and release: A comparative analysis of 16 nested catchments: Hydrological Processes, Vol. 31, No. 10, pp. 1828–1845. Phillips, F. M.; Mills, S.; Hendrickx, M. H.; and Hogan, J., 2003, Environmental tracers applied to quantifying causes of salinity in arid-region rivers: Results from the Rio Grande Basin, southwestern USA: Developments in Water Science, Vol. 50, pp. 327–334. Pool, D. and Dickinson, J. E., 2007, Ground-Water Flow Model of the Sierra Vista Subwatershed and Sonoran Portions of the Upper San Pedro Basin, Southeastern Ari-

zona, United States, and Northern Sonora, Mexico: U.S. Geological Survey Scientific Investigations Report 2006-5228, 48 p. Rosen, P. C., 2009, Reptiles and amphibians. In Stromberg, J. C. and Tellman, B. (Editors), Ecology and Conservation of the San Pedro River: The University of Arizona Press, Tucson, AZ, pp. 175–191. Segura, C.; Noone, D.; Warren, D.; Jones, J.; Tenney, J.; and Ganio, L., 2019, Climate, landforms, and geology affect baseflow sources in a mountain catchment: Water Resources Research, Vol. 55, No. 7, pp. 5238–5254. Singh, N. K.; Emanuel, R. E.; and McGlynn, B. L., 2016, Variability in isotopic composition of base flow in two headwater streams of the southern Appalachians: Water Resources Research, Vol. 52, No. 6, pp. 4264–4279. Soulsby, C.; Malcolm, R.; Helliwell, R.; Ferrier, R. C.; and Jenkins, A., 2000, Isotope hydrology of the Allt a’ Mharcaidh catchment, Cairngorms, Scotland: Implications for hydrological pathways and residence times: Hydrological Processes, Vol. 14, No. 4, pp. 747–762. Soycan, C. U.; Brand, L. A.; and Saba, J. I., 2009, Causes and consequences of mammal species richness. In Stromberg, J. C. and Tellman, B. (Editors), Ecology and Conservation of the San Pedro River: The University of Arizona Press, Tucson, AZ, pp. 107–126. Stuiver, M. and Becker, B., 1993, High-precision decadal calibration of the radiocarbon time scale, AD 1950–6000 BC: Radiocarbon, Vol. 35, No. 1, pp. 35–66. Tague, C. and Grant, G. E., 2004, A geological framework for interpreting the low-flow regimes of Cascade streams, Willamette River Basin, Oregon: Water Resources Research, Vol. 40, No. 4, W04303. Tetzlaff, D. and Soulsby, C., 2008, Sources of baseflow in larger catchments—Using tracers to develop a holistic understanding of runoff generation: Journal Hydrology, Vol. 359, pp. 287–302. The Nature Conservancy, 2020, San Pedro River Wet-Dry Maps: Electronic document, available at http://azconservation. org/downloads/san_pedro_wet_dry_mapping U.S. Geological Survey, 2019, National Water Information System, Web Interface, USGS 09471800 San Pedro River near Benson, Ariz.: Electronic document, available at https:// nwis.waterdata.usgs.gov/az/nwis/uv?cb_00060=on&format =gif_default&site_no=09471800& Wahi, A. K.; Hogan, J. F.; Ekwurzel, B.; Baillie, M. N.; and Eastoe, C. J., 2008, Geochemical quantification of semiarid mountain recharge: Ground Water, Vol. 46, pp. 414–425. Waters, M. R. and Haynes, C. V., 2001, Late Quaternary arroyo formation and climate change in the American Southwest: Geology, Vol. 29, No. 5, pp. 399–402. Western Regional Climate Center, 2019, Coop Sites Arizona: Electronic document, available at http://www.wrcc.dri. edu/summary/Climsmaz.html WestWide Drought Tracker, 2019, WestWide DroughtTracker: Electronic document, available at https://wrcc.dri.edu/ wwdt/time/

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 463–479

479

A Robust and Eﬃcient Method of Designing Piles for Landslide Stabilization YANG YU XINGMIN LI Ocean College, Zhejiang University, Zhoushan 316021, China

XIAOHUA PAN* School of Earth Science and Engineering, Nanjing University, Nanjing 210023, China

QING LÜ College of Civil Engineering and Architecture, Zhejiang University, Hangzhou 310058, China

Key Terms: Landslide, Stabilizing Piles, Factor of Safety, Robust Geotechnical Design, Uncertainty ABSTRACT Stabilizing pile is a widely used method to reduce the development of large-scale landslides. Optimizing the pile geometry is a great challenge in the design of stabilizing piles with the purpose of cost-effectiveness, especially for soil strength parameters with large uncertainty. The objective of this study is to propose a robust and efficient method of designing piles for landslide stabilization with the consideration of the safety of slope, uncertainty of soil parameters, and cost of stabilizing piles. A new response surface, which incorporates soil parameters and stabilizing force into a quadratic polynomial function, is first proposed. Unknown coefficients of the quadratic polynomial function are solved with a numerical method at typical sampling points. Based on the solved quadratic polynomial function, the mean and standard deviation of factor of safety (FOS) of the pile-stabilized slope as well as the signal-to-noise factor are then calculated in order to evaluate the design robustness. A framework based on the concept of robust geotechnical design is presented, and its feasibility is illustrated by two cases of soil slopes. The results indicate that the proposed robust geotechnical design method could be used to optimize the design of landslide-stabilizing piles.

*Corresponding author email: xhpan@ntu.edu.sg

INTRODUCTION Landslides are one of the most common natural hazards that can cause economic and human life losses every year (Huang and Fan, 2013). Numerous stabilization methods, such as stabilizing pile, drainage, anchors, geometry modiﬁcation, etc., have been adopted in practice to reduce the development of the landslides (Hassiotis et al., 1997; Sun et al., 2010; Gong et al., 2014; Yu et al., 2018, 2019b; and Wang et al., 2020). For large-scale landslides, stabilizing piles are the most popular method because they can provide larger forces to stabilize the slope (Chow, 1996; Ausilio et al., 2001; Ashour and Ardalan, 2012; Pulko et al., 2014; Li et al., 2015; Li et al., 2017; and Xiao et al., 2017). The magnitude of the stabilizing force is controlled by the geometric parameters of the stabilizing piles, such as the diameter, spacing, and length, and the soil strength parameters of the sliding mass (i.e., cohesion and friction angle) (Ito and Matsui, 1975; He et al., 2015). Optimizing the pile geometric and soil strength parameters is thus very important in the design of stabilizing piles for cost saving. For the pile geometric parameters, deterministic methods, such as analytical and numerical methods, are normally used for the optimization of design. Ito and Matsui (1975) established an analytical model that can be used to predict the lateral force of the stabilizing pile and the optimum spacing and diameter of the stabilizing piles. Nian et al. (2008) proposed a kinematic approach to evaluate the safety of piles, stabilizing non-homogeneous landslides with an optimized pile position. Li and Liang (2014a) developed a program by incorporating the limit equilibrium method and soil arching eﬀects to optimize multiple rows of stabilizing piles. Wu et al. (2017) optimized the spacing of stabilizing piles with trapezoid cross section using an

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 481–492

481

Yu, Li, Pan, and Lü

analytical method. However, the values of soil strength parameter vary greatly in practice, making it difficult to find a unique set of optimal geometric parameters for the stabilizing piles based on deterministic methods (Phoon and Kulhawy, 1999). As a result, reliabilitybased methods have been widely used to solve this problem. Li and Liang (2014b) investigated the failure probability of a pile-stabilized slope along a given slip surface. Zhang et al. (2017) established a new response surface to assess the factor of safety (FOS) of the stabilized landslides using a quadratic polynomial function, in which the stabilizing force was considered as a linear term of an uncertain variable. Nevertheless, the cost of stabilizing piles has not been considered during the optimization process corresponding to the pile geometric parameters. To overcome these concerns, robust geotechnical design (RGD) might be a promising method (Juang and Wang, 2013; Wang et al., 2013). The RGD originated from the well-established robust design methodology proposed by Taguchi (1986) in the field of industrial engineering. It aims to make the product of a process insensitive to (or robust against) “hard-tocontrol” input parameters (called noise factors), by adjusting “easy-to-control” input parameters (called design parameters). In the concept of RGD, the values of design parameters can be assigned by the designer, such as the geometric parameters of the stabilizing pile. The noise factors are the ones with significant uncertainties, such as the soil parameters. Compared with the deterministic and reliability-based methods, the RGD can not only deal with the uncertainty of the soil parameters, but it also can consider the cost of a design. The RGD has been used to optimize the design of extensive geotechnical systems, such as pure soil slopes, foundations, and tunnels (Gong et al., 2014, 2015; Khoshnevisan et al., 2014; Peng et al., 2017; and Shen et al., 2018). The purpose of the RGD is to find a design that is insensitive to the variation of the noise factors and satisfies the safety requirements of a geotechnical system at the same time. This purpose coincides with the design principles of the pile-stabilized slopes that have significant uncertainties in soil parameters. The objective of this study is to propose a robust and efficient method of designing piles for landslide stabilization with the consideration of the uncertainty of soil parameters and cost of stabilizing piles. The rest of this study is organized as follows: A deterministic method for the evaluation of the FOS of a pilestabilized slope is established. The methodology and framework for the RGD of the stabilizing pile are proposed. Verifications of the proposed RGD methodology and framework are conducted based on two slope cases. Finally, some conclusions are obtained.

482

METHODOLOGY FOR AUTOMATIC CALCULATION OF THE FOS OF A PILE-STABILIZED SLOPE The automatic calculation method is established based on a two-dimensional (2D) slope model (Figure 1) using Slope/w (GeoStudio, 2007). In addition to the soil parameters (unit weight, friction angle, and cohesion) of each formation, the stabilizing force and center-to-center spacing of the stabilizing piles are the other two types of input parameters in the 2D model. Soil parameters are directly obtained from the site investigation. The stabilizing force could be determined based on the theory of pile-soil interaction after the geometric parameters of the stabilizing piles (i.e., center-to-center spacing) are selected by the designers. In this paper, the theory proposed by Ito and Matsui (1975) is adopted to calculate the stabilizing force. The calculation process is shown as follows. As shown in Figure 1, the soil around stabilizing piles is assumed to be in a plastic state, and the limiting earth pressure at depth z can be expressed as (Ito and Matsui, 1975),

1/2 D1 N1 tanϕ+N1 −1 p(z) = cD1 D2 π ϕ 1 D1 − D2 × N1 tanϕtan exp + N1 tanϕ D2 8 4 2tanϕ + 2N 1/2 + N −1/2 1 1 − 2N11/2 tanϕ − 1 + N11/2 tanϕ + N1 − 1

2tanϕ + 2N11/2 + N1−1/2 −1/2 − 2D2 N1 − c D1 N11/2 tanϕ + N1 − 1

1/2 D1 N1 tanϕ+N1 −1 γz D1 + N1 D2

π ϕ D1 − D2 − D2 , × exp N1 tanϕtan + D2 8 4 (1) where p(z) is the limiting earth pressure at depth z, N1 = tan2 (π/4 + ϕ/2), D is the diameter of the piles, D1 is the center-to-center spacing between piles in a row, D2 is the net spacing between piles, and γ, c, and ϕ are the unit weight, cohesion, and friction angle of the soil around the pile, respectively. For a slope with a pre-existing slip surface, if the pile row position is determined, then the pile length in sliding mass can be obtained. In other words, the pile length in sliding mass is a function of pile row position. With the pile row position being assigned by the designer, the stabilizing force provided by a pile could be determined by integrating p(z) over the pile length

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 481–492

Robust Geotechnical Design of Pile

Figure 1. Conﬁguration of a pile-stabilized slope.

in the sliding mass,

Vs =

p (z)dz,

(2)

where Vs is the stabilizing force provided by the pile, and h1 is the pile length in the sliding mass as shown in Figure 1. The FOS is also aﬀected by the failure mode of the stabilizing pile, which is normally dominated by the pile length in a stable layer. According to the research of Viggiani (1981), there are three failure modes for a stabilizing pile in clay with the variation of the ratio of h2 to h1 (h2 /h1 ), where h2 is the pile length in the stable layer. For the ﬁrst type of mode, piles slide with the sliding mass when the pile length in the stable layer is too short. The earth pressure around the pile cannot reach the ultimate value. For the second and third types, the pile is long enough such that only soil failure occurs in the sliding mass, which coincides with the assumption of Eq. 1. In addition, the recent research by Pirone and Urciuoli (2018) indicated that the second and third types of failure modes can normally be observed in “ci-phi” soil (c ࣔ 0 and ϕ ࣔ 0) when the value of h2 /h1 is larger than 1.0. Thus, in this paper, h2 /h1 is assumed to be 1.0, and the total length of the stabilizing pile, H, is equal to 2h1 . After the pile row position, the center-to-center spacing, and stabilizing force are determined, the FOS of the pile-stabilized slope can be calculated us-

ing the Morgenstern-Price (M-P) method in Slope/w (Morgenstern and Price, 1965), which is suitable for both circular and non-circular slip surfaces. An automatic calculation method for the FOS of a pile-stabilized slope is also developed. The Slope/w model is saved as a calculation file (with the suffix is .xml), which can be automatically edited via coding (e.g., Matlab) and saved as a new calculation file. The new calculation file is then reread by Slope/w for further evaluation of the FOS of the slope with another pile design. This procedure is automatically executed by Winbatch. METHODOLOGY FOR ROBUST DESIGN OF THE STABILIZING PILES Robust Design Concept and Parameter Setting For the proposed deterministic method, the calculation of the FOS for the pile-stabilized slope is mainly dominated by the geometric parameters of the stabilizing pile and the shear strength parameters of the sliding mass and slip surface. In practice, the geometric parameters, which are the diameter (D), pile spacing (D1 or D2 ), and length of the stabilizing pile (H), are determined by the designers. Thus, they are classified as design parameters in the framework of RGD. The shear strength parameters of the slip surface, i.e., cohesion of soil on the slip surface (cs ) and friction angle of the slip surface (ϕs ), show significant uncertainty and

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 481–492

483

Yu, Li, Pan, and LĂź

are regarded as noise factors. In addition, the stabilizing force (Vs ) is the third noise factor considering the uncertainty of the shear strength parameters (c and Ď&#x2022;) of the sliding mass. Due to the existence of signiďŹ cant uncertainties in the noise factors (Vs , cs , and Ď&#x2022;s ), the calculated FOS from the deterministic method has high uncertainty too. This causes diďŹ&#x192;culty for designers to determine the optimal geometric parameters. Thus, RGD is introduced to select geometric parameters for the stabilizing piles, making sure the FOS of the pile-stabilized slope is insensitive to the variation in the noise factors. Meanwhile, it is also important to note that the FOS of the pile-stabilized slope should be no less than a target value (safety requirement).

FOS (Î¸) â&#x2030;&#x2C6; a0 +

Evaluation of Design Robustness and Cost

where ÎźFOS and Ď&#x192;FOS are the mean and standard deviation of the FOS of the pile-stabilized slope, respectively. A higher SNR means the FOS of the pile-stabilized slope has a smaller variation about its mean, and thus it is less sensitive to noise factors (Yu et al., 2019a). Because of the complexity of pile-slope system, the explicit expression between the FOS and noise factors is diďŹ&#x192;cult to be obtained. To overcome this diďŹ&#x192;culty, a quadratic polynomial function without cross terms, as shown in Eq. 4, is often used as a response surface to approximate the FOS of a slope (Zhang et al., 2011; Ji and Low, 2012), FOS (Î¸) â&#x2030;&#x2C6; a0 +

i=1

ai Î¸i +

ai Î¸2i ,

(4)

i=n+1

where FOS(Î¸) is the factor of safety for the pilestabilized slope, and it is a function of Î¸, where Î¸ is the vector of the uncertain variables, Î¸i is the ith element 484

nâ&#x2C6;&#x2019;1

ai Î¸i + anVs +

i=1

Based upon the reliability analysis, the RGD introduces robustness and cost in a geotechnical design. In order to achieve an optimal geotechnical design, the design robustness and cost should be quantiďŹ ed. Indexes such as signal-to-noise ratio (SNR), feasibility robustness, variation of response, and the gradient of the system response have been used to quantify the design robustness (e.g., Khoshnevisan et al., 2014). Among these indexes, SNR is adopted in this paper because it has been widely used to evaluate the design robustness of slope engineering (Gong et al., 2015; Yu et al., 2019b). The SNR can be expressed by the mean and standard deviation of the FOS of the pile-stabilized slope, 2 ÎźFOS , (3) SNR = 10log10 2 Ď&#x192;FOS

of Î¸, n is the number of uncertain variables (n = 3 in this paper), and ai (i = 0, 1, â&#x20AC;Ś, 2n) is the ith unknown coeďŹ&#x192;cient to be determined. In Eq. 4, however, only the uncertainties of the soil parameters are considered, and the uncertainty of the stabilizing force is not included. For a pilestabilized slope, the stabilizing force is related to the shear strength of the sliding mass (Ito and Matsui, 1975), and thus the stabilizing force will also have uncertainty. In this paper, the stabilizing force is considered as one of the noise factors and is introduced into Eq. 4. A new response surface represented by the following quadratic polynomial function is then obtained, 2nâ&#x2C6;&#x2019;1

ai Î¸2i + a2nVs2 ,

i=n+1

(5) where Vs stands for the stabilizing force provided by the piles. There are (2n + 1) unknown coeďŹ&#x192;cients in Eq. 5. These unknown coeďŹ&#x192;cients can be evaluated at the following (2n + 1) points using the deterministic method: {ÎźÎ¸1 , ÎźÎ¸2 , â&#x20AC;Ś, ÎźÎ¸n }, {ÎźÎ¸1 Âą mĎ&#x192;Î¸1 , ÎźÎ¸2 , â&#x20AC;Ś, ÎźÎ¸n }, {ÎźÎ¸1 , ÎźÎ¸2 Âą mĎ&#x192;Î¸2 , â&#x20AC;Ś, ÎźÎ¸n }, â&#x20AC;Ś, and {ÎźÎ¸1 , ÎźÎ¸2 , â&#x20AC;Ś, ÎźÎ¸n Âą mĎ&#x192;Î¸n }, where ÎźÎ¸i and Ď&#x192;Î¸i are the mean and the standard deviation of Î¸i , respectively, and m is a parameter representing the distance among calibration points. In this study, m = 2 was adopted (Zhang et al., 2017). The mean and standard deviation of soil parameters are evaluated based on the data from site investigation and/or local experience. The mean and standard deviation of the stabilizing force are evaluated using Eq. 1 and Eq. 2 based on ďŹ rst-order second moment (FOSM). By equating the values of the FOS obtained by Slope/w at the (2n + 1) points with those calculated using Eq. 5, the unknown coeďŹ&#x192;cients are determined. The mean and standard deviation of the FOS of the pile-stabilized slope are determined based on the solved quadratic polynomial function, ÎźFOS â&#x2030;&#x2C6; FOS Î¸Ě&#x201E; ; (6)

Ď&#x192;FOS

n 2 â&#x2C6;&#x201A;FOS (Î¸) Ď&#x192;Î¸ , â&#x2030;&#x2C6; i â&#x2C6;&#x201A;Î¸i Î¸Ě&#x201E;

(7)

i=1

where Î¸Ě&#x201E; is the mean vector of uncertain variables, and Ď&#x192;Î¸i is the standard deviation of the ith uncertain variable. If the mean value of the FOS of a design is no less than the target FOS (safety requirement), it means this design is a feasible design that satisďŹ es the safety requirement. By substituting Eq. 6 and Eq. 7 into Eq. 3,

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 481â&#x20AC;&#x201C;492

Robust Geotechnical Design of Pile

the SNR of the feasible design can be calculated. The SNR of each stabilizing pile design can be obtained by repeating the preceding computational procedure. The cost of stabilizing piles is mainly composed of the material and labor costs, which can be expressed as a function of the total volume of the stabilizing piles used in the project. For illustrative purpose, the volume of the stabilizing piles per unit center-to-center spacing is used as the index for cost, C=

πD2 H , 4D1

(8)

where H is the total length of the stabilizing pile, which is equal to the sum of pile length in the sliding mass (h1 ) and that in the stable layer (h2 ). In this paper, it is equal to 2h1 .

0 to 1 by the following function, fNi (d) =

fi (d) − [ fi (d)]min [ fi (d)]max − [ fi (d)]min

(9)

where [fi (d)]max and [fi (d)]min are the maximum and minimum values of the ith objective function fi (d) among all the feasible designs. It is noted that the coordinates of the normalized utopia point are all equal to 0 or 1. When the cost and SNR are the two objective functions, the coordinates of the utopia point are (0, 1). With the minimum distance approach, the distance from the normalized utopia point to the normalized objective functions for each feasible design is computed. The design yielding the minimum distance is taken as the knee point, which is a feasible design that is closest to the utopia point. FRAMEWORK FOR SELECTING THE MOST PREFERRED DESIGN OF STABILIZING PILES

Obtain Optimal Designs Based on Pareto Front and Knee Point After the design robustness and cost of all feasible designs are obtained, a Pareto front, which represents a set of non-dominated designs, can be determined according to the simplified procedure proposed by Khoshnevisan et al. (2014). This procedure can be summarized into four steps (Yu et al., 2019a). First, the least-cost design and the most robust design are determined based on all feasible designs. Second, the cost interval is divided (the cost between the least-cost design and the most robust design) into a series of cost levels. Third, the design with highest robustness is determined within each cost level. Fourth, the design robustness is plotted versus the cost for all the designs obtained in step 3, which yields the Pareto front. Based on the Pareto front, designers can determine the preferred design according to the project budget (cost) or robustness requirements. If no budget or robustness requirement is available, a single best design can be obtained at the knee point. Recall that the design robustness and cost are the two objectives in the optimization of stabilizing piles, and these two objectives are in conflict with each other. Thus, a single best design, which is optimal with respect to both objectives (the lowest cost and the highest design robustness), is not attainable. In fact, such a design is termed a utopia point in the context of robust design optimization. The knee point (Deb and Gupta, 2011), which is the best compromise design with respect to the dual objectives (cost and design robustness), may be obtained based on the minimum distance with respect to the utopia point (Khoshnevisan et al., 2014). To find the knee point, the design robustness and cost are first normalized into a value ranging from

The framework for RGD of the stabilizing piles is presented in Figure 2, which includes five steps: Step 1: Define the pile-slope system and classify all input parameters into design parameters and noise factors. The design parameters for stabilizing piles include the diameter (D), spacing (D1 or D2 ), and length (H = 2h1 ) of the stabilizing pile. The noise factors are the cohesion (cs ) and friction angle (ϕs ) of the slip surface, and the stabilizing force (Vs ), which means n = 3. Step 2: Determine the design space of the stabilizing pile. Discrete numbers of design parameters are selected based upon their typical ranges. The number of candidate designs in the design space is denoted as M. Step 3: Characterize the uncertainty of noise factors. The uncertainty of noise factors related to soil (the cohesion and friction angle of the sliding mass and the slip surface) can be evaluated based on the data from a geological survey and geotechnical tests, augmented with literature data or local experiences. The mean and standard deviation of the stabilizing force for a specific design are evaluated using Eq. 1 and Eq. 2 based on the FOSM method. Step 4: For each of the M designs in the design space, compute the cost and design robustness that satisfy the safety requirement. Based on the characterized uncertainty of noise factors, the unknown coefficients in Eq. 5 are then determined. The mean and standard deviation of FOS of the pile-stabilized slope are then evaluated using Eq. 6 and Eq. 7, respectively. If the mean of the FOS is greater than the target FOS, the design of the stabilizing pile is a feasible design. Following up, the SNR (design robustness) and cost are calculated based on Eq. 3 and Eq. 8, respectively. Repeat this computational procedure for M times until

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 481–492

485

Yu, Li, Pan, and Lü

Figure 2. Flowchart for the proposed eﬃcient robust geotechnical design method of landslide-stabilizing pile.

all the designs in the design space are completed. The procedure in this step is automatically performed by Matlab, Winbatch, and Slope/w (GeoStudio, 2007), as shown in Figure 3. Step 5: Obtain the optimal design and most preferred design according to the Pareto front and knee point. The application procedure and feasibility of the proposed framework for the RGD method are illustrated

and veriﬁed by the following two cases of soil slopes, respectively. ILLUSTRATIVE CASE A homogeneous soil slope is adopted to illustrate the framework proposed here. As shown in Figure 4, the length and height of the slope are 16 m and 12 m, respectively. The unit weight of the soil is 19 kN/m3 .

Figure 3. Procedure for auto-calculation of the FOS.

486

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 481–492

Robust Geotechnical Design of Pile

Figure 4. Schematic diagram of the slope used as an illustration case.

The mean values of cohesion and friction angle of the soil are 20 kPa and 10°, respectively. The coefficients of variation (COVs) of cohesion and friction angle are assumed to be 0.4 and 0.1, respectively (Phoon and Kulhawy, 1999). Based on the unit weight and the mean values of cohesion and friction angle, a numerical model is established with Slope/w, in which a critical slip surface with the FOS = 1.0 is determined with the Morgenstern-Price method (Morgenstern and Price, 1965). The critical slip surface is defined as the preexisting slip surface of the slope. Because the slope is assumed to be a homogeneous soil slope, the mean and standard deviation of the cohesion and friction angle of the pre-existing slip surface are the same as the soil. Stabilizing piles are installed in the slope to achieve a target FOS of 1.3. The pile position is defined by the horizontal distance between the pile and the slope toe, which is denoted as s. For construction convenience, the stabilizing piles are normally installed in the lower part of the slope (Yu et al., 2015). Thus, in this case, s is assumed to be in the range of 1.0 m to 8.0 m with an increment of 1.0 m, corresponding to the pile length in sliding mass h1 = {1.26, 2.17, 3.01, 3.79, 4.53, 5.20, 5.78, 6.30} m. It is known previously that h1 is equal to h2 , and the total pile length H would be {2.52, 4.34, 6.02, 7.58, 9.06, 10.4, 11.56, 12.6} m. Assuming the pile diameter (D) ranges from 0.3 m to 2.0 m with an increment of 0.1 m, the net spacing (D2 ) ranges from 0.6 m to 8 m with an increment of 0.1 m. To ensure the soil arching effect between neighboring piles, the value of D2 /D is limited between 2.0 and 4.0 (Gelagoti et al., 2011). Thus, there are a total of 3,440 designs in the design space.

During the calculation procedure, the stabilizing force (Vs ), cohesion (cs ), and friction angle (ϕs ) of the slip surface are the noise factors. For each of the 3,440 designs, the mean and standard deviation of Vs are first evaluated. Then, the unknown coefficients in Eq. 5 are solved at seven sampling points, as described earlier herein. Based on the solved quadratic polynomial function, the cost and SNR of the designs are calculated with the mean of the FOS ࣙ 1.3. Thus, 400 feasible designs are obtained. In order to verify the accuracy of Eq. 5 in predicting the FOS of the pile-stabilized slope, eight designs are randomly selected from the 400 feasible designs. For each selected design, 20 sets of Vs , cs , and ϕs , which are different from the designs at the sampling points, are randomly generated. Then, the predicted FOS using Eq. 5 and the simulated FOS using Slope/w are compared. The comparison results shown in Figure 5 indicate that the predicted results of FOS for the pile-stabilized slope using Eq. 5 have a high accuracy, with R2 ranging from 0.9981 to 0.9998. Following the procedure proposed by Deb and Gupta (2011) and Khoshnevisan et al. (2014), the Pareto front and knee point are obtained and presented in Figure 6. It can be observed from the Pareto front that SNR increases as the cost of the stabilizing piles increases. However, such a benefit is not significant when the cost further increases. On the Pareto front, none of the designs is better than the others on both SNR and cost. Take design A and design B on the Pareto front in Figure 6 as examples. Although the SNR of design A is better than that of design B, the cost of design A is higher than that of design B. The designer can select an optimal design based on the perspective of the project (i.e., better SNR or lower cost).

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 481–492

487

Yu, Li, Pan, and Lü

Figure 5. FOS of illustration case predicted from response surfaces and M-P method.

If the perspectives of SNR and/or cost are not designated, the knee point could be used to ﬁnd the most preferred design, which can ﬁnd the balance between the SNR and cost. All the designs on the Pareto front are listed in Table 1. Among these designs, the one with design parameters D = 0.5 m, D2 = 1.1 m, and H = 10.4 m (s = 6 m) is the most preferred design (knee point). VERIFICATION CASE Figure 7 shows a pile-stabilized slope case reported by Ito et al. (1981). This case is adopted to verify the feasibility of the proposed framework. Based on this

488

case, the effect of COV of the soil shear strength parameters on the selection of optimal design is also discussed. The landslide is initially triggered along a fixed slip surface by the excavation of the slope toe in order to construct a road. As shown in Figure 7, the length and width of the landslide are 65 m and 95 m, respectively. The sliding mass is composed of clayey soil with a unit weight of 16.2 kN/m3 . The mean calculated value of the cohesion of the sliding mass is 59 kPa, based on the unconfined compressive strength qu . As the friction angle cannot be zero in Eq. 1, a small value of 5° is assigned to the sliding mass. The cohesion and the friction angle of the slip surface are 9.8 kPa and 5°,

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 481–492

Robust Geotechnical Design of Pile

Figure 6. Feasible designs, Pareto front, and knee point of the illustration case.

Figure 7. Cross section of the landslide used as a veriﬁcation case (after Ito et al. (1981).

Table 1. The designs on the Pareto front of the illustration case. No. 1 2 3 4 5 6 7 8 9

D (m)

D2 (m)

H (m)

S (m)

Cost (m2 )

SNR

0.30 0.30 0.30 0.40 0.50 0.50 0.90 1.30 1.70

0.80 0.60 0.70 1.00 1.30 1.10 2.00 2.90 3.81

6.30 5.20 5.78 5.78 5.78 5.20 5.20 5.20 5.20

8 6 7 7 7 6 6 6 6

0.4048 0.4084 0.4086 0.5188 0.6305 0.6381 1.1407 1.6434 2.1460

12.0249 12.0293 12.0312 12.0582 12.0757 12.0829 12.0889 12.0907 12.0917

respectively. COVs of the cohesion and friction angle of the sliding mass and slip surface are assumed to be 0.4 and 0.1, respectively (Phoon and Kulhawy, 1999). As shown in Figure 8, the calculated FOS of the landslide after geometric modiﬁcation is 1.0, which is consistent with the calculated FOS (0.92) in the previous study (Ito et al., 1981). In order to increase the FOS of the landslide to 1.3, an extra row of stabilizing piles is installed. According to the literature of Ito et al. (1981), the diameter of the stabilizing piles ranges from 0.5 m to 0.9 m with an increment of 0.1 m. The net spacing increases from 1.0 m to 3.6 m with an increment of 0.1 m. The range of variation of pile length in the sliding mass, h1 , is between

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 481–492

489

Yu, Li, Pan, and Lü

Figure 8. FOS of the veriﬁcation landslide without piles.

1.0 m and 5.0 m with an increment of 1.0 m. Thus, the total length of the stabilizing pile ranges from 2.0 m to 10.0 m with an increment of 2.0 m. The value of D2 /D is limited between 2.0 and 4.0, and the value of D2 /D1 is limited between 0.3 and 0.9. Thus, there is a total of 775 designs in the design space. Using the same computational procedure as before, the Pareto front, knee point, and 279 feasible designs are obtained and presented in Figure 9. The most preferred design corresponding to the knee point is D = 0.5 m, D2 = 1.9 m, and H = 10 m (h1 = 5 m, h2 = 5 m). In order to investigate the inﬂuence of COVs on the optimal designs, the knee points with diﬀerent COVs

of the cohesion and friction angle are obtained and presented in Table 2. According to these results, the following conclusions can be obtained: (1) The most preferred designs at knee points are not inﬂuenced by the variation of the COV; (2) a smaller COV results in better design robustness; and (3) variation in the COV of cohesion has a larger inﬂuence on the design robustness than variation in the COV of the friction angle. CONCLUSIONS A method and framework for the selection of the most preferred design of stabilizing piles are proposed

Figure 9. Feasible designs, Pareto front, and knee point of the veriﬁcation case.

490

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 481–492

Robust Geotechnical Design of Pile

REFERENCES

Table 2. The knee point with diﬀerent COVs. COV c 0.4 0.4 0.4 0.3 0.5

D (m)

D2 (m)

H (m)

Cost (m2 )

SNR

0.3 0.2 0.1 0.1 0.1

0.5 0.5 0.5 0.5 0.5

1.9 1.9 1.9 1.9 1.9

10 10 10 10 10

0.82 0.82 0.82 0.82 0.82

9.3838 9.4527 9.4945 11.9913 7.5849

based on the concept of RGD and considering significant uncertainties in the soil parameters. The proposed method can simultaneously consider safety, cost, and design robustness of the pile-stabilized slope. According to the illustration and verification results, the following conclusions are obtained: (1) A deterministic method for evaluating the FOS of a pile-stabilized slope is established based on limit equilibrium theory, soil arching effects, and soil plastic theory. Combined with Matlab, Winbatch, and Slope/w, the method can automatically calculate the FOS of a pile-stabilized slope with the consideration of pile geometric parameters, such as diameter, spacing, and installation position (or length). (2) A new characterization method, consisting of a response surface, for efficient calculation of design robustness of a stabilizing pile is proposed by considering both the linear and quadratic terms of the stabilizing force as uncertain variables in the quadratic polynomial function without cross terms. (3) The verification results indicate that the most preferred design of the stabilizing pile can be obtained using the proposed method and framework with high accuracy. (4) Results also show that smaller COV values indicate better design robustness, and the variation in COV of the cohesion has a larger influence on the design robustness than variation in the COV of the friction angle. ACKNOWLEDGMENTS The authors wish to acknowledge the financial support of the National Natural Science Foundation of China (41807224), the Natural Science Foundation of Zhejiang Province (LQ17D020001), and Science and Technology Plan of Zhejiang Provincial Department of Transportation (2020051) for this study. The first author greatly wishes to thank Dr. C. Hsein Juang for his kindly help on the research of robust geotechnical design.

Ashour, M. and Ardalan, H., 2012, Analysis of pile stabilized slopes based on soil–pile interaction: Computers and Geotechnics, Vol. 39, pp. 85–97. Ausilio, E.; Conte, E.; and Dente, G., 2001, Stability analysis of slopes reinforced with piles: Computers and Geotechnics, Vol. 28, No. 5, pp. 591–611. Chow, Y. K., 1996, Analysis of piles used for slope stabilization: International Journal for Numerical and Analytical Methods in Geomechanics, Vol. 20, No. 9, pp. 635–646. Deb, K. and Gupta, S., 2011, Understanding knee points in bicriteria problems and their implications as preferred solution principles: Engineering Optimization, Vol. 43, No. 11, pp. 1175–1204. Gelagoti, F.; Gazetas, G.; Anastasopoulos, I.; and Kourkoulis, R., 2011, Slope stabilizing piles and pile-groups: Parametric study and design insights: Journal of Geotechnical and Geoenvironmental Engineering, Vol. 137, No. 7, pp. 663–677. GeoStudio, 2007, User’s Guide: Geo-Slope International, Ltd., Calgary, Alberta, Canada. Electronic document, available at https://www.geoslope.support Gong, W.; Wang, L.; Juang, C. H.; Zhang, J.; and Huang, H., 2014, Robust geotechnical design of shield-driven tunnels: Computers and Geotechnics, Vol. 56, pp. 191–201. Gong, W.; Wang, L.; Khoshnevisan, S.; Juang, C. H.; Huang, H.; and Zhang, J., 2015, Robust geotechnical design of earth slopes using fuzzy sets: Journal of Geotechnical and Geoenvironmental Engineering, Vol. 141, No. 1, pp. 04014084. Hassiotis, S.; Chameau, J. L.; and Gunaratne M., 1997, Design method for stabilization of slopes with piles: Journal of Geotechnical and Geoenvironmental Engineering, Vol. 123, No. 4, pp. 314–323. He, Y.; Hazarika, H.; Yasufuku, N.; Teng, J.; Jiang, Z.; and Han, Z., 2015, Estimation of lateral force acting on piles to stabilize landslides: Natural Hazards, Vol. 79, No. 3, pp. 1981– 2003. Huang, R. Q. and Fan, X., 2013, The landslide story: Nature Geoscience, Vol. 6, No. 5, pp. 325–326. Ito, T. and Matsui, T., 1975, Methods to estimate lateral force acting on stabilizing piles: Soils and Foundations, Vol. 15, No. 4, pp. 43–59. Ito, T.; Matsui, T.; and Hong, W. P., 1981, Design method for stabilizing piles against landslide-one row piles: Soils and Foundations, Vol. 21, No. 1, pp. 21–37. Ji, J. and Low, B. K., 2012, Stratiﬁed response surfaces for system probabilistic evaluation of slopes: Journal of Geotechnical and Geoenvironmental Engineering, Vol. 138, No. 11, pp. 1398– 1406. Juang, C. H. and Wang, L., 2013, Reliability-based robust geotechnical design of spread foundations using multiobjective genetic algorithm: Computers and Geotechnics, Vol. 48, pp. 96–106. Khoshnevisan, S.; Gong, W.; Wang, L.; and Juang, C. H., 2014, Robust design in geotechnical engineering—An update: Georisk: Assessment and Management of Risk for Engineered Systems and Geohazards, Vol. 8, No. 4, pp. 217–234. Li, C.; Wang, X.; Tang, H.; Lei, G.; Yan, J.; and Zhang, Y., 2017, A preliminary study on the location of the stabilizing piles for colluvial landslides with interbedding hard and soft bedrocks: Engineering Geology, Vol. 224, pp. 15–28. Li, C. D.; Wu, J. J.; Tang, H. M.; Wang, J.; Chen, F.; and Liang, D. M., 2015, A novel optimal plane arrangement of stabilizing piles based on soil arching eﬀect and stability limit for 3D colluvial landslides: Engineering Geology, Vol. 195, pp. 236–247.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 481–492

491

Yu, Li, Pan, and Lü Li, L. and Liang, R. Y., 2014a, Limit equilibrium based design approach for slope stabilization using multiple rows of drilled shafts: Computers and Geotechnics, Vol. 59, pp. 67–74. Li, L. and Liang, R. Y., 2014b, Reliability-based design for slopes reinforced with a row of drilled shafts: International Journal for Numerical and Analytical Methods in Geomechanics, Vol. 38, No. 2, pp. 202–220. Morgenstern, N. R. and Price, V. E., 1965, The analysis of the stability of general slip surfaces: Geotechnique, Vol. 15, No. 1, pp. 79–93. Nian, T. K.; Chen, G. Q.; Luan, M. T.; Yang, Q.; and Zheng, D. F., 2008, Limit analysis of the stability of slopes reinforced with piles against landslide in nonhomogeneous and anisotropic soils: Canadian Geotechnical Journal, Vol. 45, No. 8, pp. 1092–1103. Peng, X.; Li, D.; Cao, Z.; Gong, W.; and Juang, C. H., 2017, Reliability-based robust geotechnical design using Monte Carlo simulation: Bulletin of Engineering Geology and the Environment, Vol. 76, No. 3, pp. 1217–1227. Phoon, K. and Kulhawy, F. H., 1999, Evaluation of geotechnical property variability: Canadian Geotechnical Journal, Vol. 36, No. 4, pp. 625–639. Pirone, M. and Urciuoli, G., 2018, Analysis of slope-stabilising piles with the shear strength reduction technique: Computers and Geotechnics, Vol. 102, pp. 238–251. Pulko, B.; Majes, B.; and Mikoš, M., 2014, Reinforced concrete shafts for the structural mitigation of large deep-seated landslides: An experience from the Macesnik and the Slano blato landslides (Slovenia): Landslides, Vol. 11, No. 1, pp. 81–91. Shen, M.; Khoshnevisan, S.; Tan, X.; Zhang, Y.; and Juang, C. H., 2018, Assessing characteristic value selection methods for design with LRFD—A design robustness perspective: Canadian Geotechnical Journal, Vol. 56, No. 10, pp. 1475– 1485. Sun, H. Y.; Wong, L. N. Y.; Shang, Y. Q.; Shen, Y.; and Lü, Q., 2010, Evaluation of drainage tunnel eﬀectiveness in landslide control: Landslides, Vol. 7, No. 4, pp. 445–454. Taguchi, G., 1986, Introduction to Quality Engineering: Designing Quality into Products and Processes: Quality Resources, New York.

492

Viggiani, C., 1981, Ultimate lateral load on piles used to stabilize landslides. In Proceedings of the 10th International Conference on Soil Mechanics and Foundation Engineering, A. A. Balkema, Rotterdam, Stockholm, Vol. 3, pp. 555–560. Wang, L.; Hwang, J. H.; Juang, C. H.; and Atamturktur, S., 2013, Reliability-based design of rock slopes—A new perspective on design robustness: Engineering Geology, Vol. 154, pp. 56–63. Wang, Z.; Yu, Y.; Sun, H. Y.; Lü, Q.; and Shang, Y. Q., 2020, Robust optimization of the constructional time delay in the design of double-row stabilizing piles: Bulletin of Engineering Geology and the Environment, Vol. 79, No. 1, pp. 53–67. Wu, J. J.; Li, C. D.; Liu, Q. T.; and Fan, F. S., 2017, Optimal isosceles trapezoid cross section of laterally loaded piles based on friction soil arching: KSCE Journal of Civil Engineering, Vol. 21, No. 7, pp. 2655–2664. Xiao, S. G.; Zeng, J. X.; and Yan, Y. P., 2017, A rational layout of double-row stabilizing piles for large-scale landslide control: Bulletin of Engineering Geology and the Environment, Vol. 76, No. 1, pp. 309–321. Yu, Y.; Liu, X. L.; Wang, E. Z.; Fang, K.; and Huang, L., 2018, Dam safety evaluation based on multiple linear regression and numerical simulation: Rock Mechanics and Rock Engineering, Vol. 51, pp. 2451–2467. Yu, Y.; Shang, Y. Q.; Sun, H. Y.; and Wang, E. Z., 2015, Displacement evolution of a creeping landslide stabilized with piles: Natural Hazards, Vol. 75, No. 2, pp. 1959–1976. Yu, Y.; Shen, M.; and Juang, C. H., 2019a, Assessing initial stiﬀness models for laterally loaded piles in undrained clay: Robust design perspective: Journal of Geotechnical and Geoenvironmental Engineering, Vol. 145, No. 10, pp. 04019073. Yu, Y.; Shen, M.; Sun, H. Y.; and Shang, Y. Q., 2019b, Robust design of siphon drainage method for stabilizing rainfall-induced landslides: Engineering Geology, Vol. 249, pp. 186–197. Zhang, J.; Wang, H.; Huang, H. W.; and Chen, L. H., 2017, System reliability analysis of soil slopes stabilized with piles: Engineering Geology, Vol. 229, pp. 45–52. Zhang, J.; Zhang, L. M.; and Tang, W. H., 2011, New methods for system reliability analysis of soil slopes: Canadian Geotechnical Journal, Vol. 48, No. 7, pp. 1138–1148.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 481–492

The Eﬀectiveness of Reactive Materials for Contaminant Removal in the Process of Coal Conversion JACEK GRABOWSKI* Department of Environmental Monitoring, Central Mining Institute, 40-166 Katowice, Poland

ALEKSANDRA TOKARZ Department of Energy Saving and Air Protection, Central Mining Institute, 40-166 Katowice, Poland

Key Terms: Georeactor, Underground Coal Gasification, Reactive Materials, Permeable Reactive Barriers ABSTRACT The technology of permeable reactive barriers (PRB) is one of the most frequently developed methods for protecting soil and water from pollution. These barriers are zones filled with reactive material in which contaminants are immobilized and/or their concentration is reduced to the limit values during the flow of contaminated groundwater. This article presents a study on the efficiency of the removal of contaminants from the post-processing water from the underground coal gasification (UCG) process. The tests were carried out in a laboratory using a flow-through reactor design. The post-processing water came from a UCG experiment carried out in the Barbara mine, Mikołów, Poland. Activated coal, zeolite, and nano-iron were used as the reactive materials in the experiment. The obtained results were compared to tests carried out with reference water (artificial) with strictly defined characteristics. Research has shown that activated carbon is the most effective material used in the reaction zone for removing organic contaminants from groundwater generated during the coal conversion process. A new feature is the use of PRB in a georeactor zone during the UCG process to limit the potential risk of contamination spreading in the case of uncontrolled and unpredictable operation, in emergency situations related to gas leaks into the environment, during underground fires, and for water polluted by high-toxicity substances. INTRODUCTION The implementation of underground coal gasification technology on an industrial scale can cause damage to the environment, both during the process and many years after its completion. Underground *Corresponding author email: jgrabowski@gig.eu

coal gasification (UCG) involves the risk of groundwater contamination through gas migrating into the surrounding permeable rock and through gasification residues leached by the natural water inflow (Kapusta et al., 2010; Wiatowski et al., 2019). The risk of groundwater contamination related to underground gasification depends on the possibility of the migration of contaminants out of the protection zone. The transportation of liquid contaminants depends on the permeability of the surrounding rocks, the geological environment of the georeactor, and the hydrogeological conditions of the area. A rock mass should constitute a basic barrier for the georeactor, but in Polish geological conditions, there is a need to strengthen and seal the rock mass. The most favorable conditions for the sealing and strengthening of a rock mass and headings are in opened-out headings, such as in excavation shafts and caverns, drilled into compact rocks. In the UCG process, it is important to identify pollutants containing organic and inorganic substances. During the process, organic contaminants are formed, including benzene, toluene, ethylbenzene, xylene (BTEX), phenols, and polycyclic aromatic hydrocarbons (PAHs). Additionally, considerable amounts of heavy metals from the coal and the formation of ashes can be released. Dangerous inorganic contaminants are formed, including ammonia and cyanide. As a result of the high temperature, there is an increase in contaminant solubility in the waters and their possible migration to water-bearing layers. Water generated during UCG trials showed values exceeding acceptable limits for biochemical oxygen demand (BOD), chemical oxygen demand (COD), total nitrogen, cyanides, phenols, volatile organic compounds (VOCs), total organic carbon (TOC), and iron. Particular attention should be paid to the high rates of BTEX and PAHs (Smoliński et al., 2013). The introduction of this wastewater to groundwater or surface water, without purification, can cause severe aquatic environment pollution. Pollution control can be carried out by reducing pressure, by placing hydraulic bar-

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 493–505

493

Grabowski and Tokarz

riers, and by pumping contaminated water to the surface for treatment. An additional element of environmental protection involves securing the locations that are critical in terms of contaminant migration, where selected components will be removed (i.e., those most dangerous to humans and the environment). The techniques used to remove contaminants from groundwater, which are based on traditional methods of soil treatment, are usually characterized by high energy consumption and the risk of new environmental problems emerging. Therefore, more efficient and economical techniques of contaminated soil and groundwater remediation are needed. The simplest approach seems to be the method of purification called “pump-andtreat,” which removes contaminated groundwater out of the ground by pumping and treats the sewage on the surface. However, the main disadvantage of this method is the disruption of the groundwater flow system (Xenidis et al., 2002; Meggyes et al., 2007). The technology of permeable reactive barriers (PRB), in recent years, has become one of the most frequently developed methods of protecting soil and water from pollution. These are zones filled with reactive material in which contaminants are immobilized and/or their concentration is reduced to the limit values during the flow of contaminated groundwater. The main goal of the reactive material is the selective retention and neutralization of chemicals contained in a stream. This reactive bed acts like a surface or underground absorber reactor (Gavaskar et al., 2000, 2003; Simon and Meggyes, 2000; Simon et al., 2000). These methods do not require any mechanical energy to be supplied to the system. The flow of a contaminated stream through the bed is driven by the natural pressure gradients. The only maintenance action required is the periodic regeneration of the reaction medium, especially when a rapid decrease in a barrier’s activity, due to blockage or coating of an active surface caused by sediments or bacteria, is observed. In some cases, when the protective capacity is exceeded, contaminants that were previously absorbed can be leached from a barrier. Therefore, the continuous monitoring of the outlet water quality is necessary. Recently, attempts have been made to use a reactive barrier in the UCG process to minimize the spread of liquid pollutants in emergency situations, in cases of uncontrolled and unpredictable operation during the process, and to limit the leaching of pollutants after the deposit exploitation is finished (Tokarz et al., 2016). MATERIALS AND METHOD Selection of Reactive Materials The process of treating contaminants with reactive media can be described by sorption processes (type of

494

reaction, including ion exchange), biochemical reactions (biodegradation of an organic compound, precipitation of heavy metals by sulfate-reducing bacteria, redox reactions, precipitation of heavy metals, pH control), and the effect of pH (Powell et al., 1998; Roehl et al., 2005). The selection of a reactive medium is based on the composition, concentration, and type of water, gas, and mixed contaminants, and the recognition of bed retention mechanisms (sorption, precipitation, decomposition). Material selection should be based on the following criteria: reactivity, stability, availability, and price (Simon et al., 2001). A thorough understanding of system hydrogeology and plume boundaries is needed before using a reactive material, due to the need for the plume to passively flow through the reaction zone of the bed. The hydrogeological characterization must also yield information suitable for determining the rate of groundwater flow through the reactive zone. This is necessary to establish the groundwater/contaminant residence time per unit of thickness of reactive media, which, when combined with the contaminant transformation rate as it passes through the media, determines the total thickness of reactive media required. During the initiation of the UCG process, the reactive media must be made accessible to the contaminant through the use of an emplacement method, and, as with most remediation technologies, this becomes increasingly difficult at greater contaminant depth or for contaminants in fractured rock. The reaction zone should be carefully monitored for both compliance and performance. Compliance must be monitored to ensure that regulatory contamination goals are being met, and performance must be monitored in order to assess whether the reactive bed emplacement is meeting its design criteria and longevity expectations. As for any remediation technology, it is important to fully understand the factors that can result in either successful implementation and remediation or failure to achieve the remediation design goals. Complete site characterization (surface features, structures, buried services, the plume location, extent, groundwater flow direction, velocity, contaminant concentrations, stratigraphic variations in permeability, fracturing, and aqueous geochemistry) is of critical importance for the design and installation of a reactive material and must be accurate in order to achieve the required performance. The plume must not pass over, under, or around the reactive material, and the reactive zone must reduce the contaminant to concentration goals without becoming rapidly plugged with precipitates or becoming passivated. The reactive zone design, location, emplacement methodology, and estimated life expectancy are based on the site characterization

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 493–505

Reactive Materials in Underground Coal Gasiﬁcation

Figure 1. Diagram of the reaction zone model.

information; therefore, insufficient or faulty information could threaten the remediation efforts. The following are issues of importance when designing the reaction zone: the selection of the reactive medium (chemical composition, grain size distribution, and proportions of additives creating the structure of the bed), estimation of the flow time of a stream through the bed, the size and geometry of the reaction zone necessary for the assumed lifetime of the material, adoption of rational expectations in terms of the quality of water in the outlet of the reaction zone and changes in bed performance during operation. The reactive materials were taken into account in the study of the effectiveness of removing contaminants generated during the UCG experiment. Three reactive materials were selected for the study (activated carbon, zeolite, and nano-iron). Activated carbon removes organic chemicals, such as BTEX, phenols, PAHs, volatile organic compounds (VOCs), cresol (C7 H8 O), pyridine (C5 H5 N), and trichloroethylene (TCE) (Matejkova et al., 2015). For the experiment, activated carbon particulate forms, such as powders or fine granules, were selected for the tests, and these had a grain size of 4–8 mm and an average diameter between 0.15 and 0.25 mm. Zeolites have very high ion-exchange capacities, and this makes them potentially useful as treatment materials. Zeolites also remove organic chemicals, such as TCE (Muir et al., 2017). For this study, clinoptilolite (solid solution composition [Ca, Mg, Na2 , K2 ] [Al2 Si10 O24 • 8H2 O]) was used. In the trial, the zeolite fractions were 0.5–1 mm and 1–2.5 mm. The larger grain size of 1–2.5 mm reactive material (and thus more hydraulic conductivity in the reactive material) was chosen to increase filtration efficiency and responsiveness, given the time and the area of contact with dirt. The particle size of 1–2.5 mm corresponds to the size of the grains for loamy sand/coarse sand. Nano-iron enables the removal of organic chemicals such as ammonia nitrogen (NH4 + ), TCE (C2 HCl3 ), and inorganic chemicals such

as heavy metals and non-metals, e.g., aluminum (Al), barium (Ba), cadmium (Cd), manganese (Mn), mercury (Hg), nickel (Ni), uranium (U), strontium (Sr), arsenic (As), chromium (Cr), lead (Pb), and selenium (Se) (Patil et al., 2015). For the experiment, nano-iron was used as nano-particles with special surface modification, which was based on the combination of a biodegradable organic and inorganic stabilizer. Due to the narrow size distribution of the nanoparticles and their sophisticated stabilization process, the nano-iron exhibits high reactivity with a large scale of pollutants. Composition of Fe(0) was: Fe (14–18 percent), Fe2 O4 (6–2 percent), C (0–1 percent), H2 O (80 percent). Zeolite and activated carbon samples were tested to determine the specific surface area, total pore volume, and pore size distribution. The specific surface area was determined by means of the BrunnauerEmmett-Teller (BET) method. Samples were degassed overnight under reduced pressure and at a high temperature using an Autosorb iQ instrument. Through the analysis of the data, it can be concluded that the activated carbon has good sorption capacity resulting from the highly developed specific surface area of 671.2 m2 /g, as compared to 27.6 m2 /g for zeolite.

Methodology The results of previous activities, including the experimental results of batch and column experiments, were used to design the flow-through reactor. The reaction zone was designed to allow the use of solutions with highly variable pollutants, changes in flow velocity, and reaction zones, and to change simulations for different geological and hydrogeochemical conditions. The flow-through reactor was made of Perspex® and synthetic materials, which enabled visual control, easy manipulation, and facile sampling, as well as the variability of flows and temperature measurements (Figure 1).

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 493–505

495

Grabowski and Tokarz

Figure 2. Flow-through reactor.

Figure 3. Filling the ex situ reaction zone with reactive material.

The construction of the reaction zone was as follows (Figures 2 and 3): dimensions = 120 × 50 × 30 cm, volume = 180 L, easily removable partitions across flow, six sections for reactive materials, six research holes for collecting samples or dose materials or gases, three vent inlets and three outlets for dosage solutions, removable internal walls to allow the resizing of individual sections and a peristaltic pump for adjusting the flow rate. During the experiment, each reactive material was tested using the reference liquid, and then it was tested using the post-processing water (condensate) from the in situ experiment in the experimental Barbara mine in Mikołów, Poland. The composition of the reference water (water made in a laboratory to compare results with condensate) and post-processing water was measured at the flow-through reactor inlet. The flow rate for both reference water and post-processing water through the reactor was constant and was 40 mL/min for each reactive material. The experiment was carried out so that the reactor (filtration columns) was prepared and filled with reactive materials and distilled water and then fed with reference water at a constant rate of 40 mL/min. The reactive zone in the first trial was made with activated carbon, the next trial was made with zeolite, and the last was made with nanoiron. Samples were taken at the entrance to the reactive zone and the exit from it at the same time every 4 hours. The experiments lasted 52 hours each. The same steps were made with post-processing water. Six experiments were performed (the first with reference water, and the others with post-processing water for each reactive material). The testing of pH, temperature, redox

Figure 4. Phenol: results of the removal of contaminants from reference water using reactive beds.

496

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 493–505

Reactive Materials in Underground Coal Gasiﬁcation

Figure 5. Phenol: results of the removal of contaminants from post-processing water using reactive beds.

potential, and electrolytic conductivity of water was carried out continuously. Figure 2 shows the flow-through reactor together with the measuring installation, and Figure 3 shows the construction of the reaction zone in the flowthrough reactor. RESULTS AND DISCUSSION The removal efficiency of contaminants generated during UCG was tested using the selected reactive materials (activated carbon, zeolite, and nano-iron). Parameters of other materials filling the flow-through reactor (chemical composition, size distribution, and

the proportion of additives forming the bed structure) were selected. The flow time through the reactive bed as well as the size and geometry of the reaction zone were evaluated before testing. This data set was necessary to calculate the assumed working time, check the quality of the water leaving the reaction zone, and change the effectiveness of the reactive bed during operation (the time after which the bed breaks down, i.e., reduce the sorption capacity of the material used). Detailed results of the efficiency of the removal of contaminants from reference water and postprocessing water using different reactive materials are shown in the following graphs (Figures 4 to 13).

Figure 6. Benzene: results of the removal of contaminants from reference water using reactive beds.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 493–505

497

Grabowski and Tokarz

Figure 7. Benzene: results of the removal of contaminants from post-processing water using reactive beds.

The experiments showed significant differences in the efficiency of contaminant removal by reactive materials in the ratio of reference water to post-processing water (condensate). Activated carbon proved to be more effective than zeolite for all measurements. In the case of phenols, activated carbon removed the contamination from the reference water and post-processing water, while zeolite proved to be effective only for the reference water. Nano-iron slightly contributed to the removal of these contaminants from the reference water and did not show any effect on the post-processing water at all. As can be seen from the graphs, activated carbon removed BTEX from both reference wa-

ter and post-processing water, while zeolite and nanoiron were not effective in removing these contaminants. Examples of the selected results of physicochemical parameters (pH, redox potential, conductivity) are shown in Figures 14 to 19. The conductivity for individual reactive materials in the reference water did not show significant changes during the removal of pollutants. However, unlike reference water, the conductivity measured in the postprocessing water showed a significant increase in the zeolite, activated carbon, and nano-iron experiments. Analysis of the samples in the tests showed that the penetration of contaminants in the reference water at

Figure 8. Toluene: results of the removal of contaminants from reference water using reactive beds.

498

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 493–505

Reactive Materials in Underground Coal Gasiﬁcation

Figure 9. Toluene: results of the removal of contaminants from post-processing water using reactive beds.

a constant flow of 40 mL/min through the nano-iron bed occurred at the beginning of the experiment, and penetration of contaminants from the post-processing water did not show any effectiveness at all. Zeolite showed poor effectiveness of adsorption of pollutants from the reference water and post-processing water, and activated carbon turned out to be the most effective material used in all measurements for the tested pollutants. Tables 1 to 3 show the amount of contaminants removed from the water during the tests with different reactive materials.

As shown in Table 1 to 3, activated carbon demonstrated the greatest efficiency in removing contaminants in the case of the reference water and postprocessing water. Activated carbon efficiently removed phenols and BTEX, both in the reference water and the post-processing water, whereas there was no effect on the conductivity or redox potential. Zeolites removed phenols and BTEX only in the reference water. Nano-iron also only removed phenols and BTEX in the reference water, but it caused an increase in pH, while the potential redox decreased. The effectiveness of the removal of contaminants by the reactive

Figure 10. Ethylobenzene: results of the removal of contaminants from reference water using reactive beds.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 493–505

499

Grabowski and Tokarz

Figure 11. Ethylobenzene: results of the removal of contaminants from post-processing water using reactive beds.

Figure 12. Xylene (m,p,o-xylene): results of the removal of contaminants from reference water using reactive beds.

Figure 13. Xylene (m,p,o-xylene): results of the removal of contaminants from post-processing water using reactive beds.

500

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 493â&#x20AC;&#x201C;505

Reactive Materials in Underground Coal GasiďŹ cation

Figure 14. Comparison of pH values for reactive materials in reference water.

Figure 15. Comparison of pH values for reactive materials in post-processing water.

Figure 16. Comparison of redox potential for reactive materials in reference water.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 493â&#x20AC;&#x201C;505

501

Grabowski and Tokarz

Figure 17. Comparison of redox potential for reactive materials in post-processing water.

Figure 18. Comparison of conductivity for reactive materials in reference water.

Figure 19. Comparison of conductivity for reactive materials in post-processing water.

502

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 493â&#x20AC;&#x201C;505

Reactive Materials in Underground Coal Gasiﬁcation Table 1. Eﬀectiveness of removing contaminants using reactive materials from reference water. Water before Reactive Material Determination of the Sample in Reference Water Activated carbon conductivity (mS/cm) pH phenol (mg/L) benzene (mg/L) toluene (mg/L) ethylobenzene (mg/L) xylene (mg/L) BTEX (mg/L) Zeolite conductivity (mS/cm) pH phenol (mg/L) benzene (mg/L) toluene (mg/L) ethylobenzene (mg/L) xylene (mg/L) BTEX (mg/L) Nano-iron conductivity (mS/cm) pH phenol (mg/L) benzene (mg/L) toluene (mg/L) ethylobenzene (mg/L) xylene (mg/L) BTEX (mg/L)

Water after Reactive Material

Water before Reactive Material

24 Hours

Water after Reactive Material

48 Hours

— — 54.40 0.038 0.028 0.005 0.015 0.086

0.260 8.920 1.400 0.000 0.000 0.000 0.000 0.000

— — 92.59 0.055 0.035 0.007 0.021 0.118

0.253 8.360 16.19 0.008 0.004 0.000 0.000 0.012

— — 34.13 0.048 0.279 0.008 0.024 0.359

0.238 7.990 5.110 0.048 0.005 0.004 0.010 0.067

— — 9.800 0.076 0.093 0.011 0.033 0.213

0.229 7.880 12.79 0.083 0.044 0.010 0.031 0.168

— — 16.10 0.015 0.014 0.004 0.013 0.046

0.228 7.480 5.000 0.008 0.007 0.002 0.010 0.027

— — 58.16 0.058 0.017 0.018 0.037 0.130

0.228 7.390 34.25 0.035 0.006 0.012 0.026 0.079

material was affected by structural changes occurring during the reaction (Tokarz and Grabowski, 2019). CONCLUSIONS The introduction of pollutants generated in the UCG process into groundwater without treatment can lead to serious pollution of the aquatic environment. Pollution control can be carried out by reducing pressure, placing hydraulic barriers, and pumping contaminated water to the surface for processing. An additional element of environmental protection is the protection of places that are critical in terms of the migration of pollutants into the reactive deposit, in order to remove selected components that are most dangerous to people and the environment. The main goal of this study was to test the effectiveness of contaminant removal from water by applying a reactive bed. During the tests, we appropriately assessed the time of flow through the reactive bed and the size and geometry of the reaction zone, which were necessary for calculating the assumed operating time, checking the quality of water leaving the reaction zone, changing the effectiveness of the reactive bed during operation, and evaluating the time to sorption capac-

ity exhaustion followed by a breakthrough. Occasionally, the protective capacity of the bed was exceeded, and the contaminants already absorbed by the reactive material could be leached out again. Therefore, during operation, the systematic monitoring of the composition of the water leaving the bed was essential. This article presents the results of the use of activated carbon, zeolite, and nano-iron as reactive materials used to remove contaminants from water generated during the UCG process. In the case of reference water and post-processing water, the activated carbon was a suitable material for removing phenols and BTEX from water. Activated carbon did not affect water temperature, redox potential, or electrolytic conductivity. Zeolites removed phenols to some extent for the reference water, but they were completely ineffective for post-processing water. Zeolites, however, did not affect the temperature and caused a significant decrease in the pH value and the electrolytic conductivity of the water. Water flowing through the zeolite bed was enriched with dissolved oxygen, which was reflected in higher values of redox potential at subsequent water intake points. The last material analyzed was nano-iron. It had a weak effect on the removal of contaminants due to the rapid reactivity

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 493–505

503

Grabowski and Tokarz Table 2. Eﬀectiveness of removing contaminants using reactive materials from post-processing water. Water before Reactive Material Determination of the Sample in Post-Processing Water Activated carbon conductivity (mS/cm) pH phenol (mg/L) benzene (mg/L) toluene (mg/L) ethylobenzene (mg/L) xylene (mg/L) BTEX (mg/L) Zeolite conductivity (mS/cm) pH phenol (mg/L) benzene (mg/L) toluene (mg/L) ethylobenzene (mg/L) xylene (mg/L) BTEX (mg/L) Nano-iron conductivity (mS/cm) pH phenol (mg/L) benzene (mg/L) toluene (mg/L) ethylobenzene (mg/L) xylene (mg/L) BTEX (mg/L)

Water after Reactive Material

Water before Reactive Material

24 Hours

Water after Reactive Material

48 Hours

— — 6.940 0.032 0.015 0.001 0.001 0.049

0.435 9.290 1.744 0.000 0.000 0.000 0.000 0.000

— — 192.7 0.176 0.000 0.000 0.000 0.176

1.784 8.109 90.04 0.034 0.000 0.000 0.000 0.034

— — 25.26 0.112 0.117 0.011 0.052 0.292

0.853 7.700 18.66 0.093 0.108 0.017 0.052 0.270

— — 0.000 0.000 0.000 0.000 0.000 0.000

1.295 7.950 0.000 0.000 0.000 0.000 0.000 0.000

— — 4.000 0.007 0.009 0.001 0.010 0.027

0.748 7.770 15.60 0.019 0.024 0.003 0.027 0.073

— — 157.2 0.557 0.177 0.033 0.166 0.933

8.116 8.210 227.6 0.576 0.221 0.054 0.210 1.061

of nano-iron in the case of a highly contaminated solution and its leaching. Nano-iron is more suited to high-speed operations with a particular plume or frequent applications than it is for a continuous removal process. Studies have shown that this material does

not aﬀect the concentration of organic compounds in waters in any way. Also, nano-iron caused an increase in temperature and pH value as well as a signiﬁcant decrease in redox potential and dissolved oxygen concentration.

Table 3. The composition of water used during the experiment in the ﬂow-through reactor. Phenol (mg/L)

Reactive Material Reference water at the entrance to the flow-through reactor activated carbon zeolites nano-iron Post-processing water at the entrance to the flow-through reactor activated carbon zeolites nano-iron Reference water at the exit of the flow-through reactor activated carbon zeolites nano-iron Post-processing water at the exit of the flow-through reactor activated carbon zeolites nano-iron

504

Benzene (mg/L)

Toluene (mg/L)

Ethylobenzene (mg/L)

Xylene (mg/L)

100 100 100

0.1 0.1 0.1

140.51 189.41 178.60

0.285 0.519 0.345

0.339 0.614 0.473

0.054 0.110 0.097

0.400 0.580 0.569

16.19 12.79 34.25

0.008 0.083 0.035

0.004 0.044 0.006

0.000 0.010 0.012

0.000 0.031 0.026

90.04 0.000 227.6

0.034 0.000 0.576

0.000 0.000 0.221

0.000 0.000 0.210

0.000 0.000 0.054

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 493–505

Reactive Materials in Underground Coal Gasiﬁcation

The obtained results also showed that tests on the efficiency of removing contaminants generated during the gasification process should be carried out on real post-processing water (condensate) due to the presence of solid particles, such as tar and ash, which are not present in the reference water. When considering the use of these reactive materials in PRB technology for the removal of UCG products, it is necessary to account for the limited sorption capacity of activated carbon and zeolites and the need to reactivate them. These materials should be replaced periodically so as to avoid becoming a secondary source of environmental pollution, while nano-iron can be used without the need of being replaced or without having an adverse effect on the environment and groundwater. This fact and the great technological difficulties associated with the installation of the material and its replacement are challenges, and further analysis and work in this area are required. ACKNOWLEDGMENTS The publication was realized within the framework of statutory research GIG No. 11157038 Central Mining Institute (GIG). REFERENCES Gavaskar, A.; Gupta, N.; Sass, B.; Janosy, R. J.; and Hicks, J., 2000, Design Guidance for Application of Permeable Reactive Barriers for Groundwater Remediation: Battelle Columbus Operations, Columbus, Ohio. Gavaskar, A.; Sass, B.; Gupta, N.; Drescher, E.; Yoon, W. S.; Sminchak, J.; Hicks, J.; and Condit, W., 2003, Evaluating the Longevity and Hydraulic Performance of Permeable Reactive Barriers at Department of Defense Sites: Battelle Columbus Operations, Columbus, Ohio. Kapusta, K.; Stanczyk, K.; Korczak, K.; Pankiewicz, M.; and Wiatowski, M., 2010, Wybrane Aspekty Oddziaływania Procesu Podziemnego Zgazowania Wegla ˛ na Środowisko Wodne: Prace Naukowe GIG Gowisko Wodnego Zgazoa (Research Reports of Central Mining Institute—Mining & Environment), 4/2010. (Some aspects of impact of underground coal gasification process on water environment): Prace Naukowe GIG. Grnictwo i rodowisko. (Research Reports of Central Mining Institute. Mining & Environment), No. 4, pp 17–27. Matejkova, M.; Soukup, K.; Kastanek, F.; Capek, P.; Grabowski, J.; Stańczyk, K.; and Solcova, O., 2015, Application of sorbents for industrial waste water purification: Chemical Engineering & Technology, Vol. 38, No. 4, pp. 667–674. doi:10.1002/ceat.201400638. Meggyes, T.; Debreczeni, A.; Reddy, K.; Karri, M.; Riddler, G.; Roehl, K.; Czurda, K.; Sarsby, R.; Simon, F.; and Biermann, V., 2007, Sustainable Environmental Protection:

Federal Institute for Materials Research and Testing (BAM) Forschungsbericht 280. Muir, B.; Wołowiec, M.; Bajda, T.; Nowak, P.; and Czupryński, P., 2017, The removal of organic compounds by natural and synthetic surface-functionalized zeolites: A mini-review: Mineralogia, Vol. 48, No. 1–4, pp. 145–156. doi:10.1515/mipo-2017-0017. Patil, S.; Shedbalkar, U.; Truskewycz, A.; Chopade, B.; and Ball, A., 2015, Nanoparticles for environmental clean-up: A review of potential risks and emerging solutions: Environmental Technology & Innovation, Vol. 5, pp. 10–21. doi:10.1016/j.eti.2015.11.001. Powell, R. M.; Puls, R. W.; Blowes, D. W.; Gillham, R. W.; Schultz, D.; Sivavec, T.; Vogan, J. L.; Powell, P. D.; and Landis, R., 1998, Permeable Reactive Barrier Technologies for Contaminant Remediation: Environmental Protection Agency Report EPA/600/R-98/125. Roehl, K. E.; Simon, F. G.; Meggyes, T.; and Stewart, D. I., 2005, Long-term performance of permeable reactive barriers. In Roehl, K. E.; Meggyes, T.; Simon, F. G.; and Stewart, D. I. (Editors), Trace Metals and Other Contaminants in the Environment, Vol. 7: Elsevier, Amsterdam. Simon, F. G. and Meggyes, T., 2000, Removal of organic and inorganic pollutants from groundwater using permeable reactive barriers. Part 1. Treatment processes for pollutants: Land Contamination & Reclamation, Vol. 8, No. 2, pp. 103–116. Simon, F. G.; Meggyes, T.; and Debreczeni, E., 2000, New developments in reactive barrier technology. The exploitation of natural resources and the consequences. In Sarsby, R. W. and Meggyes, T. (Editors), The Exploitation of Natural Resources and the Consequences: Proceedings of the Third International Symposium on Geotechnics Related to the European Environment: Thomas Telford, London, U.K., pp. 474–483. Simon, F. G.; Meggyes, T.; and Tunnermeier, T., 2001, Evaluation of Long-Term Aspects of Passive Groundwater Remediation Techniques: Federal Institute for Materials Research and Testing, Division Waste Treatment and Remedial Engineering, Berlin, Germany. Smoliński, A.; Stańczyk, K.; Kapusta, K.; and Howaniec, N., 2013, Analysis of the organic contaminants in the condensate produced in the in situ underground coal gasification process: Water Science & Technology, Vol. 67, No. 3, pp. 644–650. Tokarz, A. and Grabowski, J., 2019, Surface morphology and structure of reactive materials used in the removal of pollutants generated during the process of coal conversion in the rock mass: Environmental Earth Sciences, Vol. 78, pp. 424. https://doi.org/10.1007/s12665-019-8425-7. Tokarz, A.; Grabowski, J.; and Nowak, D., 2016, Improvement of the UCG process safety in a georeactor surroundings under emergency conditions. 16th International Multidisciplinary Scientific GeoConference SGEM 2016, Book 1, Vol. 2, pp. 473–480. doi:10.5593/SGEM2016/B12/S03.062. Wiatowski, M.; Kapusta, K.; and Stańczyk, K., 2019, Efficiency assessment of underground gasification of ortho- and metalignite: High-pressure ex situ experimental simulations. Fuel, Vol. 236, pp. 221–227. doi:10.1016/j.fuel.2018.08.143. Xenidis, A.; Moirou, A.; and Paspaliaris, I., 2002, Reactive materials and attenuation processes for permeable reactive barriers: Mineral Wealth, Vol. 123, pp. 35–49.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 493–505

505

Eﬀects of Laumontite Hydration/Dehydration on Swelling Deformation and Slake Durability of Altered Granodiorite JUNSONG YAN JUNHUI SHEN* KAIZHEN ZHANG State Key Laboratory of Geohazard Prevention and Geoenvironment Protection, Chengdu University of Technology, Chengdu 610059, China, and College of Environment and Civil Engineering, Chengdu University of Technology, Chengdu 610059, China

JIANJUN XU WEIFENG DUAN RICHANG YANG PowerChina Huadong Engineering Corporation Limited, Hangzhou 311122, China

Key Terms: Altered Granodiorite, Laumontite, Hydration/Dehydration, Swelling Deformation, Slake Durability, Dry-Wet Cycle ABSTRACT The mineral laumontite can undergo hydration/dehydration reactions at room temperature. The hydration/dehydration produces a 3 to 6 percent volume change in the unit cell. The effects of laumontite hydration/dehydration on swelling and slake durability were investigated using altered granodiorite containing laumontite from the dam foundation of Yangfanggou Hydro Power Station, Sichuan, China. The occurrence of laumontite in altered rocks was first determined by petrological analysis. Typical samples were then collected for laboratory X-ray diffraction (XRD) analyses, free swelling tests, and slake durability index (SDI) tests. The test results were analyzed to determine the quantitative relationships between laumontite content, maximum axial strain, and slake durability index. We found that hydration of laumontite led to rock swelling. As laumontite content increased, maximum axial strain increased linearly; if water penetrated the rock quickly, swelling occurred over a short period. The hydration/dehydration of laumontite decreased slake durability of the rock; the SDI decreased approximately linearly as laumontite content increased.

*Corresponding author email: shenjunhui@cdut.cn

INTRODUCTION From an engineering perspective, the volume expansion and contraction of minerals due to changes in ambient humidity adversely affect the macroscopic properties of rocks; the most obvious effects are rock swelling and slaking. Minerals that swell/contract are mainly hydrophilic clay minerals, such as montmorillonite. Other minerals, such as anhydrite and laumontite, are also prone to hydration/dehydration reactions accompanied by volume changes in the unit cell. Extensive studies have been conducted on rock swelling and slaking caused by hydrophilic clay minerals (e.g., Einstein, 1996; Gökceoğlu et al., 2000; Tu et al., 2005; Doostmohammadi et al., 2008; Erguler and Ulusay, 2009; Vergara and Triantafyllidis, 2015; Gautam and Shakoor, 2017; and Vlastelica et al., 2018). However, the related work on other minerals is relatively scarce, and the literature mainly focuses on swelling of anhydrite-containing rocks (e.g., Butscher et al., 2016, 2018; Ramon and Alonso, 2018). Laumontite hydrates or dehydrates at room temperature (Yamazaki et al., 1991; Fridriksson et al., 2003; and Comboni et al., 2018). The hydration/dehydration results in a 3 to 6 percent volume change in the unit cell (Erlin and Jana, 2003), which can affect the overall swelling and slake durability of laumontite-containing rocks. Laumontite is one of the most common zeolites. It is mostly distributed in low-grade metamorphosed sedimentary rocks and pyroclastic rocks, but it is also found in altered intermediate or basic igneous rocks and deep metamorphic rocks that have undergone low-temperature hydrothermal alterations (Coombs, 1952; Vincent and Ehlig, 1988; Bell and Haskins,

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 507–516

507

Yan, Shen, Zhang, Xu, Duan, and Yang

Figure 1. The crystal structure of fully hydrated laumontite (from the c-axis perspective); blue spheres represent water molecules, and black spheres are Ca2+ ; the water positions W1, W2, W5, and W8 are labeled 1, 2, 5, and 8, respectively; the silicon–oxygen tetrahedrons are light gray, and the aluminum–oxygen tetrahedrons are black; a and b are the a axis and b axis of the crystal (Fridriksson et al., 2003).

1997; Utada, 2001a, 2001b; and Bravo et al., 2017). A laumontite unit cell consists of a ring structure of silicon–/aluminum–oxygen tetrahedrons (Fridriksson et al., 2003; Comboni et al., 2018). The unit cell of fully hydrated laumontite contains 18 water molecules (Ca4 Al8 Si16 O48 ·18H2 O) (Yamazaki et al., 1991). Twelve of the water molecules are associated with Ca2+ , occupying the W2 (four molecules) and W8 (eight molecules) positions of bound water in the center of the ring structure. The other six water molecules occupy the W1 (four molecules) and W5 (two molecules) positions of bound water with stable hydrogen bonds (Figure 1) (Ståhl et al., 1996; Fridriksson et al., 2003). The hydration/dehydration of laumontite is mainly governed by ambient humidity and temperature (Yamazaki et al., 1991; Ståhl et al., 1996; and Comboni et al., 2018). At room temperature, water molecules in the unit cell are successively removed from the W1 and W5 positions to form leonhardite (Ca4 Al8 Si16 O48 ·12–14H2 O) as ambient humidity decreases (Yamazaki et al., 1991; Fridriksson et al., 2003). This reaction is reversible and is accompanied by changes in the unit cell parameters (Yamazaki et al., 1991). An increase in ambient temperature will dehydrate laumontite. Ståhl et al. (1996) found from a real-time X-ray diﬀrac-

508

tion (XRD) study that as ambient temperature increased, even when the laumontite remained immersed in water, most water molecules at the W1 position were removed at 76°C, and 80 percent of those at the W5 position were removed at approximately 100°C. Early in the 1920s, the hydration/dehydration of laumontite was found to adversely affect engineering construction; laumontite was referred to “concrete poison” by Pearson and Loughlin (1923). It is noted in ASTM C 294 (2001) that the cyclic conversion of laumontite to leonhardite can cause concrete distress. Erlin and Jana (2003) emphasized that, although the volume of laumontite changed only slightly during hydration/dehydration, accumulated damage could lead to concrete slaking or “crumbling.” Examples of engineering rock mass deterioration caused by hydration/dehydration of laumontite are also common. Bell and Haskins (1997) and Sumner et al. (2009) suggested that the volume change of zeolites (including laumontite) during hydration/dehydration was responsible for the deterioration and appearance of cracks in the basaltic rocks of the transfer tunnel in the Lesotho Highlands Water Project. Bravo et al. (2017) attributed shotcrete detachment in a transfer tunnel in central Chile to the swelling and contraction of laumontitecontaining surrounding rocks. Apart from the paper by Bravo et al. (2017), there is little experimental data for swelling deformation and slake durability of laumontite-containing rocks. Bravo et al. (2017), using a non-standard slaking test on laumontite-containing andesitic rocks, found that specimens slaked more significantly under dry-wet cycling than when directly immersed in water, but they did not provide a quantitative description of slake durability. They also conducted a swelling stress test, but the swelling capacity of the rock could not be accurately determined due to the use of reconstituted samples (Bravo et al., 2017). This study aimed to obtain quantitative indicators of swelling deformation and slake durability for laumontite-containing rocks and to determine the effects of laumontite hydration/dehydration on swelling deformation and slake durability. We investigated altered granodiorite containing laumontite from the dam foundation of Yangfanggou Hydro Power Station, Sichuan, China. Through petrological analysis, the occurrence of laumontite in altered rocks was first determined. Typical samples were then taken for laboratory XRD analyses, free swelling tests, and slake durability index (SDI) tests. Quantitative relationships between laumontite content, maximum axial strain, and slake durability index were developed through regression analysis of the test results.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 507–516

Laumontite Hydration/Dehydration in Granodiorite

CURRENT RESEARCH AND THE OCCURRENCE OF LAUMONTITE The Yangfanggou Hydro Power Station is located in the middle reaches of the Yalong River in Liangshan Prefecture, Sichuan Province, China. The water is retained by a hyperbolic concrete arch dam with a height of 155 m. The rock mass of the dam foundation is Late Triassic intrusive granodiorite (U-Pb age 208–212 Ma) (Yuan et al., 2011), with principal mineral constituents of quartz (20 to 30 percent), plagioclase (30 to 40 percent), potassium feldspar (10 to 20 percent), biotite (5 to 10 percent), and amphibole (10 to 15 percent). Accessory mineral constituents are mainly sphene, apatite, zircon, and Fe-Ti oxides (Yuan et al., 2011). Drilling and excavation revealed that a mediumscale alteration zone had developed in the rock mass of the left bank dam foundation. The alteration zone is distributed along a fault, and the dam foundation is obliquely cut at an elevation of 1947–2000 m, with a total width of 1.5–3.9 m. Altered rocks are gray-green, fragmented, and interspersed with many fine white and gray-white veins of different densities (Figure 2a). Petrological research showed that altered rocks are characterized mainly by chloritization of biotite, amphibole, and plagioclase, and strong laumontization had occurred in some locations. The altered rocks generally retained their original granitic texture, but brittle minerals such as quartz and feldspar were more severely broken, and micro-fractures and pores were more developed. Laumontite is unevenly distributed in the altered rocks. It is found principally in three forms. One is fine white and gray-white veins distributed in altered rocks; the second form dispersively fills the micro-fissures and pores near the veins (Figure 2b); and the third form is a feldspar alteration product distributed as flakes near the fine veins in altered rocks. When observed under a single-polarized light microscope, laumontite is a colorless transparent mineral that develops two groups of perfect cleavage, i.e., [010] and [110] (Gottardi and Galli, 1985). Its refractive index is between 1.502 and 1.525, which is much lower than that of quartz (1.544– 1.553) or Ca-rich plagioclase (1.554–1.590), by which it can be distinguished from the other two minerals (Taylor et al., 1990). The birefringence of laumontite is between 0.009 and 0.013 (Gottardi and Galli, 1985). Under an orthogonal polarizer, the interference color of laumontite produced in the form of fine veins or filled-in micro-fractures is usually first-order gray (Figure 2c), while that formed by feldspar alteration can reach first-order yellow (Figure 2d). Occasionally, sporadic solitary residues of feldspar are encased in laumontite crystals (Figure 2d). The single grain size of

laumontite is extremely small, generally not more than 50 μm, but the single grain size formed by feldspar alteration can reach 400 μm. We separated the fine vein material and performed XRD analyses to determine the presence of laumontite. Figure 3 shows that the XRD pattern of the fine vein material matches the standard XRD pattern of laumontite, partially hydrated (JCPDS, 1985). Characteristic peaks appeared at 9.4424Å, 6.8254Å, 4.1519Å, and 3.5040Å, with intensities of 100, 51.8, 50.1, and 66.9 (the characteristic peak at 3.3558Å is due to quartz incorporation). This result shows that the molecular formula of laumontite in the fine veins is Ca4 Al8 Si16 O48 ·13.2H2 O. MATERIALS AND METHODS Based on our knowledge of the occurrence of laumontite, we collected lump-like altered granodiorite samples from different locations in the alteration zone for laboratory analysis and testing. Six samples were collected of diameter 20–40 cm. Samples I and III–VI were taken from densely veined rock, and sample II was taken from sparsely veined rock. The samples were sealed in plastic film immediately after collection and carefully transported to the laboratory. XRD was used for quantitative analysis of the mineral composition of the samples, with a particular focus on laumontite content. Each rock sample was tested four or five times to obtain an average value for laumontite content because of sample heterogeneity. Powder samples passing the 200 mesh were prepared and tested using an Ultima IV diffractometer (Rigaku Corporation, Japan). The operating parameters were 40 kV, 40 mA, CuKα radiation, 0.01°/0.24 s scanning speed, and 5° to 60° scanning range. Quantitative analysis of the XRD data was performed using whole pattern fitting with Rietveld refinement in Jade 6 software. A short introduction to this quantitative phase analysis method is given in the Appendix. Free swelling tests were performed on samples I– IV (12 specimens) to determine the axial swelling strain under unconfined conditions. We followed the International Society for Rock Mechanics (ISRM) free swelling test protocols published in 1989 (ISRM, 1989), but we made adjustments according to the conditions observed using the following procedure. Cube specimens (three specimens per sample), with side length of 5 cm, were prepared in dry conditions (Figure 4). Each specimen was dried to a constant weight at 105°C and then placed in a desiccator. The specimen was cooled to room temperature (20 ± 2°C), and its dimensions were measured. By this time, any laumontite that may have been contained in the specimen was sufficiently dehydrated and had been converted into

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 507–516

509

Yan, Shen, Zhang, Xu, Duan, and Yang

Figure 2. (a) Field photos of laumontite-altered granodiorite; (b) hand-sized specimen of laumontite-altered granodiorite; (c) laumontite crystals ﬁlling in micro-fractures (orthogonal polarization); (d) laumontite crystals formed by feldspar alteration (orthogonal polarization). Lmt = laumontite.

leonhardite (Ståhl et al., 1996). The specimen was placed in the expansion apparatus, which was initialized, and the initial readings of the dials were recorded. The specimen was immersed in distilled water, and measurement of the axial strain commenced. The shape of the specimen was observed in real time. Dial readings were recorded at 10 minute intervals during the ﬁrst hour and then at 1 hour intervals until the dial reading was stable. The specimen was immersed in water for at least 48 hours. The axial swelling strain (εax ) of the specimen was calculated by: εax =

δax , h0

(1)

where δax is the axial swelling displacement (mm), and h0 is the initial height of the specimen (mm). 510

When the swelling test of each specimen was completed, the specimen was dried, crushed, and ground to powder, and XRD analyses were performed for further investigation of the relationship between laumontite content and maximum axial strain. The slake durability index (SDI) quantiﬁes and indicates the resistance of a rock to physical weathering (Franklin, 1972; Erguler and Shakoor, 2009). This index, which can be used to indicate the eﬀects of laumontite hydration/dehydration on rock, is measured over a number of dry-wet cycles. The SDI tests were performed on samples I–VI according to the ISRM (1979) protocol. Each set of tests required the preparation of ten 40–60 g spherical specimens (prepared without contact with water). Five standard disintegration cycles were performed. The SDI for selected cycle

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 507–516

Laumontite Hydration/Dehydration in Granodiorite Table 1. Results (%) of XRD analyses.

Figure 3. XRD pattern of the ﬁne vein material and standard XRD pattern of laumontite.

i (Idi ) was calculated as the ratio (percentage) of ﬁnal to initial dry weights of the specimens in the drum. RESULTS AND DISCUSSION XRD Analyses Table 1 shows the main results of the XRD analyses. More experimental details are given in the Appendix. The altered granodiorite is composed of primary rock-forming minerals and secondary alteration minerals. The primary constituents are quartz (14.8 to 32.6 percent), plagioclase (16.5 to 34.6 percent), K-feldspar (10.4 to 26.5 percent), and amphibole (4.0 to 11.8 percent); secondary constituents are chlorite (6.3 to 23.8 percent), laumontite (0 to 22.0 percent), and kaolinite (0.2 to 5.0 percent). No swelling clay minerals such as montmorillonite were detected in any sample. When compared with the parent granodiorite, the altered rock contained no biotite, and the contents of plagioclase and amphibole were relatively low, suggesting that alterations due to chloritization, laumon-

Sample No.

Qtz

Kfs

Chl

Lmt

Kln

I-1 I-2 I-3 I-4 Mean II-1 II-2 II-3 II-4 Mean III-1 III-2 III-3 III-4 Mean IV-1 IV-2 IV-3 IV-4 IV-5 Mean V-1 V-2 V-3 V-4 V-5 Mean VI-1 VI-2 VI-3 VI-4 Mean

25.3 21.8 22.9 22.3

21.9 19.1 16.5 23.6

10.4 17.0 17.2 18.9

7.3 7.9 8.9 6.2

11.3 12.0 13.6 6.3

3.1 3.1 3.4 0.7

23.2 27.3 19.3 27.2

27.7 23.8 31.4 34.6

14.3 18.3 24.0 16.9

6.0 4.2 6.2 10.3

23.8 21.5 17.5 10.1

20.8 14.8 21.6 16.2

22.0 31.2 29.6 34.5

19.9 20.1 22.9 17.9

7.8 7.8 4.2 4.0

11.2 11.4 6.3 8.3

21.6 25.0 24.7 22.3 22.9

32.3 20.3 32.5 31.0 27.8

14.4 13.3 14.5 22.5 21.0

9.6 9.9 7.5 11.4 11.8

11.7 12.7 8.4 7.4 8.7

20.8 24.9 26.2 21.6 21.4

24.9 23.8 28.5 19.6 31.6

12.6 14.0 15.6 24.9 17.0

8.7 7.7 9.3 9.2 9.0

11.0 10.4 8.7 11.8 10.9

20.4 32.4 29.7 32.6

27.0 18.7 17.5 23.2

19.5 13.2 17.8 12.3

7.1 7.9 5.6 6.8

9.3 12.5 12.3 12.5

20.6 19.0 17.7 22.0 19.8 0.0 0.0 0.0 0.0 0.0 16.8 13.7 14.8 18.1 15.9 9.2 17.0 11.1 5.2 6.8 9.9 19.3 16.6 9.8 11.0 7.5 12.8 12.6 10.7 12.0 7.8 10.8

5.0 4.8 1.5 0.8 1.6 0.9 0.6 0.9 1.2 1.8 1.3 0.2 1.0 2.7 2.7 1.9 1.9 2.7 4.1 4.6 5.0 4.8

Qtz = quartz; Pl = plagioclase; Kfs = K-feldspar; Am = amphibole; Chl = chlorite; Lmt = laumontite; Kln = kaolinite.

tization, and kaolinization may be consuming these minerals. All specimens, except those from sample II, contained laumontite. This agrees with our general understanding of laumontite occurrence. In terms of average content, sample I contained the highest amount of laumontite (19.8 percent), and sample IV contained the lowest (9.9 percent). According to Bell and Haskins (1997) and Sumner et al. (2009), the average content of laumontite in altered basalt in Lesotho was 11 percent, and the laumontite caused cracking deformation in the rock surrounding a tunnel. The heterogeneity of laumontite distribution was most obvious in sample IV. The laumontite content in specimen IV-2 was 17.0 percent, while that in IV-4 was only 5.2 percent. Free Swelling Tests

Figure 4. Specimen cubes for free swelling tests.

The free swelling test results are shown in Table 2 and Figure 5. No obvious shape change or cracking, except the detachment of few sand grain-sized particles, occurred on the specimen cubes during

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 507–516

511

Yan, Shen, Zhang, Xu, Duan, and Yang Table 2. Results of free swelling tests. Specimen I-1 I-2 I-3 II-1 II-2 II-3 III-1 III-2 III-3 IV-1 IV-2 IV-3

Maximum εax (%)

Time to Stabilize (minutes)

0.404 0.286 0.310 0.010 0.026 0.012 0.331 0.199 0.187 0.129 0.360 0.237

60 480 120 30 30 30 120 40 50 60 120 60 Figure 6. Free swelling curves of the specimens (ﬁrst 480 minutes).

immersion in water. The range of maximum axial strain for all specimens was 0.010 to 0.404 percent, which is small. As the laumontite content increased, maximum axial strain increased signiﬁcantly, indicating increased swelling of the specimen. Bivariate regression analysis was performed using the SPSS software; the results showed a good linear relationship between laumontite content and maximum axial strain (R2 = 0.907), given by the equation: εax = 0.008 + 0.017CLmt ,

(2)

where εax is the maximum axial strain (percent), and CLmt is the laumontite content (percent). The level of signiﬁcance for the Fisher test was 0.000, i.e., less than the standard 0.05 level, which indicates that the regression model was statistically signiﬁcant (Moradizadeh et al., 2016). According to Einstein (1996), there are three typical rock swelling mechanisms: mechanical swelling triggered by stress release or rebound; osmotic swelling related to the ionic double layer on the clay particle surface; and intra-crystalline swelling caused by hy-

Figure 5. Relationship between maximum axial strain and laumontite content.

512

dration of montmorillonite or inter-layer clay minerals, anhydrite, pyrite, and other minerals. The relationship shown in Figure 5 indicates that intra-crystalline swelling due to laumontite hydration is an important cause of specimen swelling. Figure 6 shows the free swelling curves of the specimens. All the laumontite-containing specimens except specimen I-2 showed a similar swelling pattern: The axial strain rapidly increased after immersion in water and then quickly converged to an equilibrium in less than 120 minutes (Table 2). Specimen I-2 reached equilibrium at 480 minutes. There are two principal reasons for the rapid swelling rate: the reaction rate of laumontite hydration is rapid (Comboni et al., 2018); and the micro-fissures and pores in the specimen are well developed, allowing water to quickly penetrate and hydrate the laumontite (Pejon and Zuquette, 2002; Vergara and Triantafyllidis, 2015). Slake Durability Index Tests The conventional SDI test measures only the SDI of the second cycle of the sample (Id2 ) (Franklin, 1972; ISRM, 1979, 2007). However, many researchers have observed that the SDI test over multiple cycles (>2 times) is more suitable for accurate evaluation of slake durability (Gökceoğlu et al., 2000; Erguler and Shakoor, 2009; Yagiz, 2011; and Momeni et al., 2017). Thus, five dry-wet cycles were performed for each group of tests. The test results are shown in Table 3 and Figure 7. Id5 for all specimens was in the range of 40 to 87.20 percent, and it decreased significantly as average laumontite content increased (Table 2), indicating that slake durability of the sample decreased. Bivariate regression analysis showed that the relationship between the two is approximately linear (R2 = 0.751), and it is defined by the equation: Id5 = 85.511 − 2.103ACLmt ,

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 507–516

(3)

Laumontite Hydration/Dehydration in Granodiorite Table 3. Results (%) of slake durability index tests. Sample No. I II III IV V VI

Id1

Id2

Id3

Id4

Id5

85.80 95.40 84.80 94.60 90.64 83.04

74.00 92.50 68.60 84.20 79.63 74.49

64.30 90.60 57.10 78.10 71.22 67.17

57.50 89.00 47.60 73.10 66.11 62.24

52.40 87.20 40.00 70.60 62.24 55.21

where Id5 is the SDI for the fifth cycle (percent), and ACLmt is the average laumontite content (percent) (Figure 7). The level of significance was 0.026, i.e., less than the standard 0.05 level, which indicates that the regression model was statistically significant. Figure 7 shows that the hydration/dehydration reaction of laumontite has an adverse effect on slake durability, which we explain as follows. Laumontite often fills micro-fissures. As the rock dries, the laumontite dehydrates and contracts, and tensile stress is applied to the fissure wall, loosening mineral particles around the fissure. Conversely, when the rock absorbs water, the laumontite hydrates, swells, and exerts stress on the fissure wall, forcing the micro-fissures apart. The damage accumulates as the number of dry-wet cycles increases, eventually leading to slaking. The slake durability of the sample can be inferred from the morphology of the residual specimen after the SDI test is completed (Erguler and Shakoor, 2009; Gautam and Shakoor, 2016, 2017). Figure 8 shows the specimens before and after SDI testing. After five drywet cycles, samples I and III slaked more, the residual specimens were chipped or lumpy, and the particle size was significantly reduced. Samples V and VI slaked less, and the residual specimens were mainly fragments. Sample II slaked the least, and the residual specimens were almost intact. Sample IV slaked extremely unevenly; some specimens slaked completely, and the rest remained almost intact; the unevenness seemed to be

Figure 8. Comparisons of specimen morphology before and after SDI tests.

related to the uneven distribution of laumontite. Thus, XRD analyses were performed on residual specimens of sample IV having varying degrees of fragmentation (Figure 8). The results showed that the laumontite content of the slaked residual samples reached 12.1 percent and that the unslaked specimen did not contain laumontite (Table 4). These results show that the hydration/dehydration reactivity of laumontite directly determined the slake durability of the altered granodiorite. CONCLUSIONS Altered granodiorite containing laumontite from the dam foundation of Yangfanggou Hydro Power Station was investigated to determine the eﬀects of laumontite hydration/dehydration on swelling deformation and slake durability. Based on the petrological research, XRD analyses, free swelling tests, and Table 4. Quantitative mineralogy (%) of the residual specimen.

Figure 7. Relationship between the ﬁfth-cycle SDI (Id5 ) and average laumontite content.

Sample No.

Qtz

Kfs

Chl

Lmt

Kln

IV-s IV-z

24.7 28.4

20.3 30.3

26.5 21.3

6.8 6.8

6.6 10.4

12.7 0.0

2.40 2.70

Qtz = quartz; Pl = plagioclase; Kfs = K-feldspar; Am = amphibole; Chl = chlorite; Lmt = laumontite; Kln = kaolinite.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 507–516

513

Yan, Shen, Zhang, Xu, Duan, and Yang

slake durability index tests were performed. The main conclusions are as follows. (1) Laumontite is unevenly distributed in altered rocks. It mostly occurs in the form of fine veins, or it dispersively fills the micro-fissures and large pores near the veins, or it is distributed as feldspar alteration products near the veins. Laumontite that fills the micro-fissures is an important factor in rock disintegration. (2) Laumontite hydration causes rock swelling. As laumontite content increased, maximum axial strain increased linearly. If water penetrated the rock quickly, swelling deformation of the laumontite-containing rock occurred over a short period. (3) The hydration/dehydration of laumontite had an adverse effect on the slake durability of surrounding rocks. As laumontite content increased, the SDI decreased approximately linearly. In this study, we conducted only free swelling tests on laumontite-containing rocks, but in engineering construction, swelling stress is often a more useful indicator than maximum axial strain. The effect of laumontite hydration on rock swelling stress needs to be further studied. Laumontite-containing rocks are sensitive to dry-wet cycles; if such rocks are exposed in engineering construction, they should be covered as soon as possible and waterproofed. ACKNOWLEDGMENTS We acknowledge the support received from the National Natural Science Foundation of China (No. 41572308). REFERENCES ASTM C 294, 2001, Standard Descriptive Nomenclature for Constituents of Concrete Aggregates: ASTM International, West Conshohocken, PA. Bell, F. G. and Haskins, D. R., 1997, A geotechnical overview of Katse Dam and Transfer Tunnel, Lesotho, with a note on basalt durability: Engineering Geology, Vol. 46, No. 2, pp. 175–198. Bish, D. L. and Howard, S. A., 1988, Quantitative phase analysis using the Rietveld method: Journal of Applied Crystallography, Vol. 21, pp. 86–91. Bravo, A.; Jerez, O.; Kelm, U.; and Poblete, M., 2017, Dehydration-hydration reactivity of laumontite: Analyses and tests for easy detection: Clay Minerals, Vol. 52, pp. 315–327. Butscher, C.; Breuer, S.; and Blum, P., 2018, Swelling laws for clay-sulfate rocks revisited: Bulletin of Engineering Geology and Environment, Vol. 77, No. 1, pp. 399–408. Butscher, C.; Mutschler, T.; and Blum, P., 2016, Swelling of clay-sulfate rocks: A review of processes and controls: Rock Mechanics and Rock Engineering, Vol. 49, No. 4, pp. 1533– 1549.

514

Comboni, D.; Gatta, G. D.; Lotti, P.; Merlini, M.; and Hanfland, M., 2018, Crystal-fluid interactions in laumontite: Microporous and Mesoporous Materials, Vol. 263, pp. 86–95. Coombs, D. S., 1952, Cell size, optical properties and chemical composition of laumontite and leonhardite: With a note on regional occurrences in New Zealand: American Mineralogist, Vol. 37, No. 9–10, pp. 812–830. Doostmohammadi, R.; Moosavi, M.; Mutschler, T.; and Osan, C., 2008, Influence of cyclic wetting and drying on swelling behavior of mudstone in south west of Iran: Environmental Geology, Vol. 58, No. 5, pp. 999–1009. Einstein, H. H., 1996, Tunnelling in difficult ground—Swelling behaviour and identification of swelling rocks: Rock Mechanics and Rock Engineering, Vol. 29, No. 3, pp. 113–124. Erguler, Z. A. and Shakoor, A., 2009, Quantification of fragment size distribution of clay-bearing rocks after slake durability testing: Environmental and Engineering Geoscience, Vol. 15, No. 2, pp. 81–89. Erguler, Z. A. and Ulusay, R., 2009, Assessment of physical disintegration characteristics of clay-bearing rocks: Disintegration index test and a new durability classification chart: Engineering Geology, Vol. 105, No. 1, pp. 11–19. Erlin, B. and Jana, D., 2003, Forces of hydration that can cause havoc in concrete: Concrete International, Vol. 25, pp. 51–57. Franklin, J. A., 1972, The slake-durability test: International Journal of Rock Mechanics Mining Sciences & Geomechanics Abstracts, Vol. 9, No. 3, pp. 325–328. Fridriksson, T. I.; Bish, D. L.; and Bird, D. K., 2003, Hydrogenbonded water in laumontite I: X-ray powder diffraction study of water site occupancy and structural changes in laumontite during room-temperature isothermal hydration/dehydration: American Mineralogist, Vol. 88, No. 2–3, pp. 277–287. Fu, W.; Peng, Z.; Zeng, X.; Qin, J.; Li, X.; Lai, S.; Li, X.; and Zhang, Y., 2018, Quantitative analysis of mineral composition in granite regolith based on XRD-Rietveld full-spectrum fitting method: Spectroscopy and Spectral Analysis, Vol. 38, No. 7, pp. 2290–2295 (in Chinese with English abstract). Gautam, T. P. and Shakoor, A., 2016, Comparing the slaking of clay-bearing rocks under laboratory conditions to slaking under natural climatic conditions: Rock Mechanics and Rock Engineering, Vol. 49, No. 1, pp. 19–31. Gautam, T. P. and Shakoor, A., 2017, A durability classification of clay-bearing rocks based on particle size distribution of slaked material: Environmental and Engineering Geoscience, Vol. 23, No. 2, pp. 125–136. Gökceoğlu, C.; Ulusay, R.; and Sönmez, H., 2000, Factors affecting the durability of selected weak and clay-bearing rocks from Turkey, with particular emphasis on the influence of the number of drying and wetting cycles: Engineering Geology, Vol. 57, No. 3, pp. 215–237. Gottardi, G. and Galli, E., 1985, Natural Zeolites: Springer, Berlin. ISRM, 1979, Suggested methods for determining water content, porosity, density, absorption and related properties and swelling and slake-durability index properties: Part 2: Suggested methods for determining swelling and slake durability index properties: International Journal of Rock Mechanics and Mining Sciences & Geomechanics Abstracts, Vol. 16, No. 2, pp. 151–156. ISRM, 1989, International Society for Rock Mechanics Commission on Swelling Rock and Working Group on Swelling Rock of the Commission on Testing Methods: Suggested methods for laboratory testing of argillaceous swelling rocks: International Journal of Rock Mechanics and Mining Sciences & Geomechanics Abstracts, Vol. 26, No. 5, pp. 415–426.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 507–516

Laumontite Hydration/Dehydration in Granodiorite ISRM, 2007, The Complete ISRM Suggested Methods for Rock Characterization, Testing and Monitoring: 1974–2006: Commission on Testing Methods, International Society for Rock Mechanics (ISRM), Turkish National Group, Ankara, Turkey. Joint Committee on Powder Diffraction Standards (JCPDS), 1985, Mineral Powder Diffraction File: JCPDS, Swarthmore, PA. Momeni, A.; Hashemi, S. S.; Khanlari, G. R.; and Heidari, M., 2017, The effect of weathering on durability and deformability properties of granitoid rocks: Bulletin of Engineering Geology and Environment, Vol. 76, No. 3, pp. 1037–1049. Moradizadeh, M.; Cheshomi, A.; Ghafoori, M.; and Trighazali, S., 2016, Correlation of equivalent quartz content, slake durability index and Is50 with Cerchar abrasiveness index for different types of rock: International Journal of Rock Mechanics and Mining Sciences, Vol. 86, pp. 42–47. Pal, H.; Chanda, A.; and De, M., 1998, Characterisation of microstructure of isothermal martensite in Fe–23Ni–3.8Mn by Rietveld method: Journal of Alloys and Compounds, Vol. 278, No. 1, pp. 209–215. Pearson, J. C. and Loughlin, G. F., 1923, An interesting case of dangerous aggregate: ACI Journal Proceedings, Vol. 19, No. 1, pp. 142–154. Pejon, O. J. and Zuquette, L. V., 2002, Analysis of cyclic swelling of mudrocks: Engineering Geology, Vol. 67, No. 1, pp. 97–108. Ramon, A. and Alonso, E. E., 2018, Heave of a building induced by swelling of an anhydritic Triassic claystone: Rock Mechanics and Rock Engineering, Vol. 51, No. 9, pp. 2881–2894. Ståhl, K.; Artioli, G.; and Hanson, J. C., 1996, The dehydration process in the zeolite laumontite: A real-time synchrotron X-ray powder diffraction study: Physics and Chemistry of Minerals, Vol. 23, No. 6, pp. 328–336. Sumner, P. D.; Hall, K. J.; Rooy, J. L. V.; and Meiklejohn, K. I., 2009, Rock weathering on the eastern mountains of southern Africa: Review and insights from case studies: Journal of African Earth Sciences, Vol. 55, pp. 236–244. Taylor, K.; Graham, R. C.; and Ervin, J. O., 1990, Laumontite in soils of the San Gabriel Mountains, California: Soil Science Society of America Journal, Vol. 54, No. 5, pp. 1483. Tu, X. B.; Jian, B.; Wang, S. J.; Bian, H. Y.; Wang, J.; and Li, S. G., 2005, Swelling behavior induced by alteration in granite and its implications on underground excavation: Tunneling and Underground Space Technology, Vol. 20, No. 4, pp. 378–389. Utada, M., 2001a, Zeolites in burial diagenesis and low-grade metamorphic rocks: Reviews in Mineralogy and Geochemistry, Vol. 45, No. 1, pp. 277–304. Utada, M., 2001b, Zeolites in hydrothermally altered rocks: Reviews in Mineralogy and Geochemistry, Vol. 45, No. 1, pp. 305–322. Vergara, M. R. and Triantafyllidis, T., 2015, Swelling behavior of volcanic rocks under cyclic wetting and drying: International Journal of Rock Mechanics and Mining Sciences, Vol. 80, pp. 231–240. Vincent, M. W. and Ehlig, P. L., 1988, Laumontite mineralization in rocks exposed north of San Andreas Fault at Cajon Pass, southern California: Geophysical Research Letters, Vol. 15, No. 9, pp. 977–980. Vlastelica, G.; Miščević, P.; and Štambuk Cvitanović, N., 2018, Durability of soft rocks in Eocene flysch formation (Dalmatia, Croatia): Engineering Geology, Vol. 245, pp. 207–217. Yagiz, S., 2011, Correlation between slake durability and rock properties for some carbonate rocks: Bulletin of

Engineering Geology and Environment, Vol. 70, No. 3, pp. 377–383. Yamazaki, A.; Shiraki, T.; Nishido, H.; and Otsuka, R., 1991, Phase change of laumontite under relative humiditycontrolled conditions: Clay Science, Vol. 8, No. 2, pp. 79–86. Yuan, J.; Xiao, L.; Wan, C.; and Gao, R., 2011, Petrogenesis of Fangmaping-Sanyanlong granites in southern SongpanGarze Fold Belt and its tectonic implication: Acta Geologica Sinica, Vol. 85, No. 2, pp. 195–206 (in Chinese with English abstract).

APPENDIX The quantitative mineral composition of the samples was determined using the Rietveld method. This quantitative phase analysis method, based on whole pattern fitting, has been commonly used since it was developed by Bish and Howard (1988). Three parameters are used to evaluate the quality and reliability of fitting (Pal et al., 1998), i.e., the weighted residual error (Rw ), the expected error (Rexp ), and the goodness of fit (GoF), where GoF = Rw /Rexp . A smaller value of GoF indicates a better fitting, and thereby a more reliable analysis result. According to Fu et al. (2018),

Table A1. Parameters (%) evaluating the quality of ﬁtting. Sample No. I-1 I-2 I-3 I-4 II-1 II-2 II-3 II-4 III-1 III-2 III-3 III-4 IV-1 IV-2 IV-3 IV-4 IV-5 V-1 V-2 V-3 V-4 V-5 VI-1 VI-2 VI-3 VI-4 IV-s IV-z

Rw (%)

Rexp (%)

GoF

10.44 9.28 9.96 14.26 11.53 11.60 10.99 11.12 10.05 11.00 11.83 11.12 9.97 9.82 12.09 9.29 11.30 10.89 11.80 12.10 11.71 12.11 12.96 10.34 9.25 10.85 12.96 10.34

5.32 4.23 4.26 4.22 4.12 4.78 3.32 4.20 4.14 3.48 3.53 4.20 5.46 4.20 3.42 4.20 3.55 3.46 3.44 3.55 3.50 3.51 3.52 3.44 3.55 3.48 3.52 3.44

1.96 2.19 2.34 3.38 2.80 2.43 3.31 2.65 2.43 3.16 3.35 2.65 1.83 2.34 3.54 2.21 3.18 3.15 3.43 3.41 3.35 3.45 3.68 3.01 2.61 3.12 3.68 3.01

Rw = weighted residual error, Rexp = expected error, GoF = goodness of ﬁt.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 507–516

515

Yan, Shen, Zhang, Xu, Duan, and Yang

when meeting Rw < 15 percent and GoF < 5, the analysis result is considered to be reliable. As shown in Table A1, the Rw values of all the XRD samples varied in the range of 9.25 to 14.26 percent,

516

while the GoF values were in the range of 1.83 to 3.68. Thus, according to the criterion suggested by Fu et al. (2018), the results obtained in the present study were reliable.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, pp. 507â&#x20AC;&#x201C;516

Notice of Retraction

The editors of Environmental & Engineering Geoscience journal would like to inform the readers of the journal that the paper entitled “Geological and Geotechnical Characterization of the Terrateig Dam in Valencia, Spain”, coauthored by Francisco Javier Torrijo, Santiago Alija, Julio Garzón-Roca, and Mario Quinta-Ferreira, published in the February 2019 issue, Vol. 25, No. 1, pages 1-14, has been retracted from the journal because signiﬁcant portions of the text and

ﬁgures were taken from Ana Santiago Loriente’s master’s thesis without attribution.

REFERENCE Ana Santiago Loriente, 2012, Estudio geológico-geotécnico de la presa de Terrateig en el Río Vernissa (Gandía): unpublished MS thesis. Universitat Politècnica de València.

Environmental & Engineering Geoscience, Vol. XXVI, No. 4, November 2020, p. 517

517

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

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

SUBSCRIPTIONS:

Environmental & Engineering Geoscience November 2020 VOLUME XXVI, NUMBER 4

SUBMISSION OF MANUSCRIPTS

Single copies are $75.00 each. Requests for single copies should be sent to AEG, 3053 Nationwide Parkway, Brunswick, OH 44212.

THIS PUBLICATION IS PRINTED ON ACID-FREE PAPER EDITORS

Brian G. Katz Environmental Consultant Tallahassee, FL 32309 eegeditorbkatz@gmail.com

Volume XXVI, Number 4, November 2020

Abdul Shakoor Department of Geology Kent State University Kent, OH 44242 330-672-2968 ashakoor@kent.edu

ENVIRONMENTAL & ENGINEERING GEOSCIENCE