E&EG Journal Volume XXVII, Number 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. SUBSCRIPTIONS: Member subscriptions: AEG members automatically receive digital access to the journal as part of their AEG membership dues. Members may order print subscriptions for $75 per year. GSA members who are not members of AEG may order for $60 per year on their annual GSA dues statement or by contacting GSA. Nonmember subscriptions are $310 and may be ordered from the subscription department of either organization. A postage differential of $10 may apply to nonmember subscribers outside the United States, Canada, and Pan America. Contact AEG at 844-331-7867; contact GSA Subscription Services, Geological Society of America, P.O. Box 9140, Boulder, CO 80301. Single copies are $75.00 each. Requests for single copies should be sent to AEG, 3053 Nationwide Parkway, Brunswick, OH 44212. © 2021 by the Association of Environmental and Engineering Geologists

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

EDITORS

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

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

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

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

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

Environmental & Engineering Geoscience November 2021 VOLUME XXVII, NUMBER 4 Special Issue on Slope Stability in Memory of Jerome (Jerry) V. De Graff: Part 1 Guest Editors: Dennis Staley, Jeremy Lancaster, Alan Gallegos, Thad Wasklewicz

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

Cover photo In memory of Jerome (Jerry) V. De Graff, 1945-2020, U.S. Forest Service, geologist, colleague, friend. Photo courtesy of the De Graff family.

Volume XXVII, Number 4, November 2021

All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or by any information storage and retrieval system, without permission in writing from AEG.

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

ENVIRONMENTAL & ENGINEERING GEOSCIENCE

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

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

Environmental & Engineering Geoscience Volume 27, Number 4, November 2021 Special Issue on Slope Stability in Memory of Jerome (Jerry) V. De Graff: Part 1 Guest Editors: Dennis Staley, Jeremy Lancaster, Alan Gallegos, Thad Wasklewicz Table of Contents 375

Introduction to Special Issue on Slope Stability in Memory of Jerome (Jerry) De Graff: Part 1 Dennis Staley, Jeremy Lancaster, Alan Gallegos, and Thad Wasklewicz

377

The Stillwater Scarp, Central Nevada, USA; Coseismic Gravitational Failure on a 1.200-M-High Range-Front Escarpment James P. McCalpin and Leon C. A. Jones

395

Magnitude and Timing of the Tiltill Rockslide in Yosemite National Park, California Christopher J. Pluhar, Kiersti R. Ford, Greg M. Stock, John O. Stone and Susan R. Zimmerman

409

Sixty Years of Post-Fire Assessment and Monitoring on Non-Federal Lands in California: What Have We Learned? Peter H. Cafferata, Drew B. R. Coe, and William R. Short

423

Assessment of a Post-Fire Debris Flow Impacting El Capitan Watershed, Santa Barbara County, California, U.S.A. Jonathan Yonni Schwartz, Nina S. Oakley, and Paul Alessio

439

Rainfall Thresholds for Post-Fire Debris-Flow Generation, Western Sierra Nevada, CA Chad K. Neptune, Jerome V. Degraff, Christopher J. Pluhar, Jeremy T. Lancaster, and Dennis M. Staley

455

Runout Number: A New Metric for Landslide Runout Characterization Cory S. Wallace and Paul M. Santi

471

Comparison of Two Logistic Regression Models for Landslide Susceptibility Analysis Through a Case Study Lauren Southerland and Wendy Zhou

Introduction to Special Issue on Slope Stability in Memory of Jerome (Jerry) De Graﬀ: Part 1 DENNIS STALEY* U.S. Geological Survey, 4230 University Drive, Anchorage, AK 99508

JEREMY LANCASTER California Geological Survey, 801 K Street, Suite 1200, Sacramento, CA 95814

ALAN GALLEGOS U.S. Forest Service (retired), 28270 Burrough Valley Rd, Tollhouse, CA 93667

THAD WASKLEWICZ Stantec Consulting Services Inc., 3325 South Timberline Rd Ste 150, Fort Collins, CO 80525

The environmental and engineering geoscience community is indebted to the career of Jerome (Jerry) De Graff, whose publications, teachings, and presentations spanned a diversity of physiographic environments, geologic and geomorphic processes, and applied scientific topics that focused on understanding and assessing natural hazards, reducing risk, improving life safety, and providing scientifically informed guidance for establishing public policy. Results of his scientific pursuits continue to be influential in the hazard arenas of North America and abroad. Jerry’s publications and lectures inspired us, entertained us, enlightened us, and challenged us to be better scientists, engineers, public servants, and global citizens. Jerry passed away on March 26, 2020, and is survived by his

*Corresponding author email: dstaley@usgs.gov

wife Sandy, sons Nicholas and Mark, and grandson Liam. Jerry was raised in the small western New York agricultural community of Honeoye, and attended the nearby State University of New York, College at Geneseo. Geneseo was a small public college specializing in education and nestled on the glacially sculpted hillslopes above the Genesee River. Here, Jerry received formal education toward his passions of education and geology and received a Baccalaureate in Science in Earth Science Education in 1967. After his undergraduate studies, Jerry served as a junior high school science teacher in Pittsford, NY, and as an instructor for the Strasenburgh Planetarium at the Rochester Museum of Science and Nature in Rochester, NY. The lure of exposed bedrock and easily seen evidence of active geomorphologic processes in the western United States led Jerry to Utah, where he pursued his Master of Science degree in Geology at Utah State University and completed a thesis related to the Quaternary geomorphology of the Bear River mountain range in northcentral Utah. Jerry spent 2 years at Utah State as a research technician after completing his graduate studies. In 1977, he began his long and productive career in the U.S. Forest Service, spending several years in Utah before moving to California, where he worked on a variety of projects related to geologic and hydrologic hazards in the western United States and abroad. Our goal for this special volume of Environmental and Engineering Geoscience is to publish papers that captured the essence of Jerry’s legacy, including research related to landslide, rockfall, and debrisﬂow processes and susceptibility; the geomorphology

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 375–376

375

Staley, Lancaster, Gallegos, and Wasklewicz

of wildfire-affected landscapes; geological hazards and hazard assessment; and the general advancement of the disciplines of environmental and engineering geology in the context of population growth and a changing planet. In this issue, the first of two in this special volume, we have included seven papers that span a range of topics representative of Jerry’s research interests. Three papers, first-authored by his U.S. Forest Service colleague Jonathan (Yonni) Schwartz, Pete Cafferata from the California Geological Survey, and Chad Neptune, his most recent graduate student, focus on post-fire debris flows in the western United States. Post-fire landslides and debris flows were a topic particularly important to Jerry since at least 1987, when he wrote his first fire-related report, titled “Using Past Landslide Activity to Guide Post-Wildfire Mitigation,” and spent the rest of his career working on post-fire debris flow, flooding, and landslide research, hazard assessment, and mitigation. Landslide and rockfall processes were a topic that Jerry actively researched throughout his career. His first publication related to landslide processes and haz-

376

ard assessment in Utah appeared in the journal Environmental Geology in 1978. He continued to publish numerous rockfall and landslide papers and make presentations at conferences throughout his career. Two excellent papers in this issue, first-authored by James McCalpin, engineering geologist with GEO-HAZ Inc. and Chris Pluhar, Jerry’s colleague at Fresno State University, testify to Jerry’s impact on research of rockfall and landslide processes in the western United States. Jerry also had a passion for the development and application of robust science and innovative methodology to assess hazard potential, inform policy decisions, and reduce public risk to these hazards. We are pleased to include two papers, authored by Cory Wallace of Yeh & Associates, Inc. and Wendy Zhou from the Colorado School of Mines, that focus on those objectives in this issue. As co-editors, we hope you find this diverse collection of manuscripts interesting and informative, a worthy addition to the environmental and engineering geoscience literature, and a testament to the lasting legacy of Jerry De Graff.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 375–376

The Stillwater Scarp, Central Nevada, USA; Coseismic Gravitational Failure on a 1.200-M-High Range-Front Escarpment JAMES P. MCCALPIN* GEO-HAZ Consulting, Inc., P.O. Box 837, Crestone, CO 81131

LEON C. A. JONES Independent geologist, P.O. Box 827, Ogden, UT 84402

Key Terms: Range Front, Landslide, Gravitational Spreading, Sackung, Triggered Faulting, Trench ABSTRACT The Stillwater scarp bounds one side of a Quaternary range-crest graben in the northern Stillwater Range, central Nevada. The scarp was reactivated in the 1915 M7.3 Pleasant Valley and 1954 M6.8 Dixie Valley earthquakes, the only such occurrence known in the Central Nevada Seismic Belt. The scarp is a partial reactivation of an older, north-striking, west-dipping Neogene normal fault that crops out at the range crest. This section of the east-facing range-front escarpment had previously failed in a 2.8 km2 dipslope landslide. Our field mapping extended the known length of the Stillwater scarp from 1.5 to 3.8 km and identified additional landslide elements in the range-front escarpment below. We dug a 3.2-m-deep trench across the Stillwater scarp where it offsets and dams the graben drainage outlet. The trench exposed a series of six stacked sag pond deposits, each overlain by a coarsening-upward package of alluvium, all in normal fault contact with Triassic limestone of the range crest. Based on hanging-wall sedimentology and a series of four nested fissure fills along the fault zone, we infer four prehistoric displacement events between 1954 CE and 35–45 ka, with vertical displacements ranging from 40 to 120 cm (mean 71 cm). We conclude that periodic late Quaternary earthquakes on nearby active faults triggered additional ridge-crest spreading and incipient slope failures on the escarpment, and future occurrences should be expected. Conversely, the rest of the range does not have the same lithology, structural attitudes, and topography favorable for ridgecrest spreading, so we do not anticipate spreading hazards there.

*Corresponding author email: mccalpin@geohaz.com

INTRODUCTION The Stillwater Range and adjacent Dixie Valley (Nevada) lie in middle of the Central Nevada Seismic Belt (CNSB; Bell et al., 2004), a 275-km-long zone of historic surface ruptures in the Basin and Range extensional province (Figure 1). During both the 1915 Pleasant Valley (Mw7.3; Wallace, 1984) and the 1954 Dixie Valley-Fairview Peak (Mw 6.8 and 7.2; Slemmons, 1957; Caskey et al., 1996) earthquakes, a graben-bounding scarp (the Stillwater scarp [SS]) was reactivated at the crest of the Stillwater Range more than 1,000 m above the adjacent floor of Dixie Valley (Figure 2; Wallace, 1984). This reactivated graben is unusual because the range-front ruptures of 1915 and 1954 terminated, respectively, 29 km north of and 7 km south of the range-crest scarp and thus did not rupture the master normal fault downslope of the graben. The 36-km-long gap between the 1915 and 1954 ruptures was named the Stillwater Seismic Gap (Wallace, 1978; Wallace and Whitney, 1984), reflecting the seismic gap hypothesis popular in the late 1970s (e.g., McCann et al., 1979). The SS is of interest for two reasons. First, it is the only known ridge-crest scarp reactivated during a historic earthquake in the Basin and Range province. Second, it lies in a section of the range that did not experience historic surface rupture on its range-front normal fault. From this, Wallace (1984) speculated that the scarp formed due to earthquake shaking and gravitational failure rather tectonic faulting. He reported, “When B. M. Page (personal communication, 1975) visited the site in 1933, he was told by a rancher that the scarp was formed in 1915. The scarp was reactivated in the early 1950’s, probably during the earthquake of 1954, during which scarps formed in Dixie Valley along the southeast flank of the Stillwater Range... [the SS] could be a secondary structure related to lateral spreading [see Figure 1b in this article]... Zischinsky (1969) suggests the name ‘sackungen’ for such structures related to lateral spreading.”

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

377

McCalpin and Jones

Figure 1. (a) Location map of the Stillwater scarp within the Stillwater seismic gap and historic earthquake surface ruptures in the Central Nevada Seismic Belt. Dates of ruptures are given with earthquake name abbreviations: PV = Pleasant Valley; RM = Rainbow Mountain; DV = Dixie Valley; W = Wonder; FP = Fairview Peak; CM = Cedar Mountain; EM = Excelsior Mountain; 1869? = Olinghouse. Mࣙ7 ruptures shown in black, smaller earthquakes in gray. Modiﬁed from Wallace and Whitney, 1984. (b) Schematic cross section of Stillwater Range showing Wallace’s (1984, p. A18) interpretation that the Stillwater scarp is “a secondary structure related to lateral spreading.”

If Wallace’s inference is correct, the SS is a sackung (or deep-seated gravitational slope deformation [DSGSD]), on which slip dies out downward (Figure 1b). But that interpretation raises two questions: (1) Why did such deformation occur only here and nowhere else among all the historic fault ruptures in the CNSB? (2) Does the SS and its adjacent graben ﬁll contain a record of previous fault reactivations similar to those of 1915 and 1954, and do their ages correspond to the ages of nearby paleoearthquakes? Answering these questions required a paleoseismic trenching

378

Figure 2. Legacy images from Wallace (1984). (a) Small sketch map of the 1915 Stillwater scarp (SS) from Wallace’s plate 1. (b) Oblique aerial photograph looking northeast at the northern part of the graben, with fault-dammed meadow at lower right; SS is between white arrows. Black arrow shows trench site of this study (40.0447°N, 177.8450°W). Photo by Robert E. Wallace, September 19, 1974. U.S. Geological Survey Photographic Collection, Roll 1, frame 9, Library ID: wreb0603.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

Stillwater Scarp

investigation of the SS, combined with a geomorphic and structural analysis of the range-front escarpment as reported herein. METHODS Our field study involved 21 days of scarp mapping, profile measuring, and paleoseismic-style trenching and sampling (e.g., McCalpin, 2009), supported by later numerical dating of faulted and unfaulted deposits. Mapping The SS was mapped in the field over a length of ∼850 m, with scarp strike measured by the Bruntonand-pace method and fault scarp profiles measured by stadia rod and Abney level. Near the trench, we mapped scarp micro-topography over a 1,760-m2 area using a Leitz total station. Vertical surface offset was measured as the vertical distance between the downthrown surface and the upthrown surface projected across the scarp. Where both surfaces have gentle gradients, such as at the drainage outlet, error is on the order of 10 cm. If the faulted surfaces are steep or the upthrown surface is steeper than the downthrown, error can increase to several decimeters. Trenching We excavated a single trench (0.6 m wide, 5.2 m long, 3.2. m deep) with picks and shovels across the southern part of the SS where the scarp blocked the drainage outlet of the graben. Due to pervasive carbonate cementation, the trench was difficult to excavate, but walls were extremely stable. The south trench wall was cleaned and gridded with string, after which we drew a field log on graph paper at a scale of 1:15. The log was then digitized for this article. Unconsolidated stratigraphic units were defined on the basis of color, texture, and sedimentary structures. Soil horizons developed on deposits (parent materials) were also identified and labelled according to the A/B/C horizon terminology used in the United States (e.g., Soil Survey Staff, 2014). Geochronology The trench contained both datable organics and silty sediments amenable to luminescence dating. Radiocarbon samples were detrital charcoal or in situ burn layers, dated by accelerator mass spectrometry either at the Arizona NSF Radiocarbon Facility (Tucson, AZ) or at Beta Analytic Inc. (Coral Gables, FL). We used the calibration curve of Reimer et al. (2020) to convert radiocarbon years to calendar years.

Fine-grained sediments were dated by the infraredstimulated luminescence method on (4- to 11-μm multi-mineral silt grains) using pretreatment with dilute HCl and H2 O2 and gravity settling according to the Stokes method (all at the Luminescence Laboratory at Dalhousie University, Halifax, NS, Canada). X-ray diffraction confirmed that both feldspar and quartz were present in the extracts. In addition, a pure quartz extract was obtained for one sample (SRL963) by treating the 4- to 11-μm multi-mineral extract with silica-saturated fluorosilicic acid for 3 days, during which time the acid was refreshed daily. The 4- to 11-μm grains were deposited onto 0.98-cm Al disks by settling from an acetone suspension at a concentration of 1 mg/cm3 , which yielded disk coverage of ∼1 mg/cm2 . Multiple aliquot additive dose analysis was used for all extracts. A reference background value for the growth curves was provided by a set of disks naturally bleached by a 2-hour exposure to ordinary daylight. Pre-irradiation, short-shine normalization was applied to all data. Groups of six disks were irradiated in an AECL 220 Gamma-Cell, which had a dose rate of ∼1.6 Gy/min at the time the irradiations were performed.

GEOLOGY AND GEOMORPHOLOGY OF THE EASTERN STILLWATER RANGE FRONT The Stillwater range-front escarpment rises 1,180 m from the floor of Dixie Valley (∼1,150 m above sea level [asl]) to the range crest (2,330 m asl), at an average angle 32°–35°. Escarpment geology is admittedly complex (Figure 3) but can be understood by first considering the autochthonous rocks beneath the FencemakerBoyer thrusts. Rocks in the autochthon (all Triassic) are, from oldest to youngest, the Koipato Group of andesitic volcanics; Star Peak Group of thin- to mediumbedded platform carbonates (Nichols and Silberling, 1977), and post–Star Peak mixed carbonate and clastic rocks, all broadly folded in the south and more tightly folded in the north, closer to the thrusts, expressed as footwall [FW] synclines and drag folds, respectively (Speed, 1976). In late Jurassic times, the autochthon was overridden by the Fencemaker thrust system, here composed of the Fencemaker and Boyer thrusts. Thrust sheets are subhorizontal in this part of the range and composed of Triassic pelite and sandstone (TR s), Boyer Ranch Formation (Jb, limestone and quartzite), and mafic intrusives of the Humboldt lopolith. The latter are slices of the Jurassic sea floor emplaced here by tens of kilometers of thrust movement from the west. PostFencemaker rocks lie unconformably over the thrust sheets and are composed of Oligocene to Miocene

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

379

McCalpin and Jones

Figure 4. Oblique view to north-northeast of range crest and escarpment, showing landslide deposits (yellow) and landslide headscarps (green around largest slides, purple elsewhere). Area shown in this ﬁgure is outlined in Figure 3. The Stillwater scarp is the red hachured line at the range crest. Multiple curved headscarps in center indicate incipient failure of Star Peak Group dipslopes.

Figure 3. Geologic map of the Stillwater range front near the Stillwater scarp (yellow, at range crest). Geologic units, thrust faults (FT = Fencemaker Thrust; BT = Boyer Thrust), and most highangle faults from Speed (1976). Additional faults (black) and dikes (green) from Smith et al. (2001); orange faults in southwest quadrant from Plank (1998). Dotted rectangle shows area of Figure 4. Sections A–A and B–B (red lines) are shown in Figure 5.

rhyolitic ash-flow tuffs (Tvs) and Miocene to Pleistocene basalts (QTb). In the Neogene, steep north- and northeast-trending normal faults formed throughout the range, offsetting the autochthon, the allochthon, and even the Tertiary rocks. The high-angle faults have pervasively disrupted the two thrusts so badly that it is difficult to trace them in Figure 3. Movement of the fault blocks appears to have been mainly vertical, with little block rotation, leaving the autochthonous rocks with their original, moderate dips. Most of the Star Peak Group, for example, dips east to east-southeast at a gradient similar to the current topography, making the entire Star Peak Group (Trsp) in the center of Figure 3 a large dipslope. Geomorphology The range-front geomorphology downslope of the SS is dominated by a large (2.5 × 2.5 km) section 380

of depressed, irregular topography coincident with the Star Peak Group (Figure 3). The northern half of this area is occupied by a 2-km2 landslide deposit mapped by Speed (1976; largest yellow polygon “Qs” in Figure 4), with its steep bedrock headscarp (30°–38°) lying directly downslope of the SS. The entire landslide (scarp and deposit) is at least 2.8 km long (L) in the transport direction, but the toe may be truncated by the range-front fault. Vertical relief (H) between headscarp and toe is 850–910 m. Dimensions of the landslide deposit are as follows: L, 2.3 km; width, 2.3 km; area, 2.05 km2 ; thickness, 10–15 m at edges, up to 50 m in center (estimated mean thickness 20 m); and volume, 0.041 km3 , or 41 × 106 m3 . The “Fahrböschung” (energy line) slope of the entire landslide (H/L; after Hsü, 1978) ranges from 0.23 to 0.31 (12.8°–17.0°). The mass of the slide can be appreciated from its axial topographic proﬁle, compared with that of the unfailed escarpment farther north (Figure 5). The slide has a 200-m elevation deﬁcit in its evacuation area, matched by a 200-m elevation surplus in the toe area. Near the toe, Wallace (1984) mapped arcuate scarps (red lines in Figure 4) that appeared to have formed during the 1915 earthquake. In vertical section, the toe is composed of the landslide deposit

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

Stillwater Scarp

Figure 5. Topographic profiles down the range front at the Qs landslide (B–B , dashed gray line) compared to the profile of the unfailed range front 1.25 km farther northeast (A–A , solid black line). Location of profiles shown in Figure 3.

(∼20 m thick) unconformably overlying steeply eastdipping bedrock, which is then cut off by the Dixie Valley fault (Figure 4). We assume that, when deposited, the bottom of the landslide deposit was graded to valley-floor level. That contact is now perched 100 m above the valley floor. This suggests 100 m of vertical uplift across the Dixie Valley fault since slide emplacement. At an uplift rate of 0.25–0.5 m/kyr (Bell et al., 2004), it implies slide emplacement at ∼200–400 ka. Remnants of an even older slide are expressed as discontinuous levee-like deposits on the north flank of the slide, 15–25 m above the main slide. The southernmost 1.5 km of the slide toe is oriented approximately parallel to the range front and is separated from the bedrock escarpment by a 2.5-km-long fault mapped by Smith et al. (2001; see Figure 3). This fault crosses the main slide body and creates a topographic saddle before continuing north into Triassic bedrock. This saddle is the site of 1915 landslide (?) scarps mapped by Wallace (1984). The fault coincides with a gap in fault scarps at the toe of the range front. This pattern suggests that recent slip on the Dixie Valley fault has followed this fault, offset the Qs landslide, and then stepped eastward to the toe of the northern range front. The southern half of the depressed area on the range front is occupied by dipslope topography developed on a hard (limestone?) bed in the Star Peak Group (Figure 3). There are numerous curved (concave downslope) scarps and benches in this terrain, along with uphill-facing and downhill-facing scarps, which do not appear to be the result of differential bedrock erosion. At its uphill end, this zone contains several short antislope scarps (red in Figure 4), some reactivated in the two historic earthquakes (according to Wallace, 1984). The overall impression is that this dipslope is fragmenting and extending.

RESULTS Geomorphology of the Stillwater Graben The SS bounds the eastern side of a 40-ha range crest half-graben, with its master fault on the east. The graben depocenter is a sandy meadow dominated by sagebrush (Figure 6a) in contrast to surrounding bedrock slopes covered by pinyon-juniper forest. Discounting the narrow stream outlet, the graben has a topographic closure of about 20 m. The depocenter meadow is a single aggradational surface, lacking multiple late Quaternary geomorphic surfaces such as found on the valley floor (Bell and Katzer, 1990). Starting with Wallace’s (1984) sketch map of the SS (Figure 2), we mapped the tallest part of his scarp in the field for a distance of ∼850 m and measured 17 scarp profiles (Figure 6b). The goal of this work was threefold: (1) to locate the best site for trenching, (2) to compile measurements of vertical surface offset along strike, and (3) to compare variations in offset to local geomorphology. Along the 850 m of our scarp traverses, vertical surface offset ranged from 0.4 to 1.8 m, with a bimodal distribution (Figure 7). Within and just south of the drainage outlet channel, surface offset varied from 0.4 to 0.65 m. Outside of that area, surface offsets were much larger and formed an arc-shaped pattern with the highest offsets lying just north of the drainage outlet (1.4–1.8 m) and then tapering down to the north and south to about 1 m. Evidence from the trench (discussed later) indicates that the smaller scarps represent only the 1915 + 1954 ruptures, whereas the larger scarps resulted from those two ruptures plus an unknown number of prehistoric ruptures. Wallace (1984) had mapped a short (300 m) antislope scarp about 1.3 km south of the main 1.5-km-long

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

381

McCalpin and Jones

Figure 6. Fault scarp profiles on the Stillwater scarp. (a) Location of the graben depocenter (meadow, in green) and trench in relation to the range crest (dotted blue line) and range-front landslide. (b) Location of 17 fault scarp profiles (thin black lines) along the central 850 m of the scarp. Profiles are numbered outward from the trench site; red numbers above each profile show vertical surface offset (m); (c) 10-cm contour map of the scarp face (gray) where it crosses the drainage outlet and dams it. Dots are total station elevation points. In the outlet channel, the scarp represents only the 1915 and 1954 displacements; outside the channel, it includes the latest prehistoric event.

SS (see Figure 2, scarp in section 18). Our mapping revealed additional antislope scarps between Wallace’s two mapped scarps, which nicely filled the gap between them. In addition, we mapped antislope scarps an additional ∼600 m south of his southern scarp. Together with the extended scarp length mapped at the northern end of our field traverse, it appears that the 1915/1954 fault scarps were created by down-to-the-west slip on a single fault structure ∼3.8 km long (Figure 3). Geomorphology of the SS Trench Site As shown in Figure 6c, our trench was located where the SS offsets and blocks the narrow outlet chan-

nel of the half-graben. This channel drains the entire half-graben, and over time its eastward ﬂow eroded a narrow slot through bedrock to reach the top of the range-front escarpment. The present 0.6-m-high scarp faces west (upstream) and dams the outlet stream (Figure 8b). To the north and south of the channel, scarp heights are 1.7–1.8 m on a colluvial slope of unknown age (Figure 8a), but the scarp abruptly decreases in height to 0.6 m across the channel. This relationship suggests that the stream periodically eradicates the scarp across the channel (by either FW erosion or hanging wall [HW] aggradation) and that today’s 0.6-m-high scarp represents only the latest displacement events (1915 and 1954).

Figure 7. Plot of vertical surface offset (VSO) of the Stillwater scarp along strike at the locations of the 17 scarp profiles measured in the field (locations shown on Figure 6b). Black bars show total VSO of the historic + prehistoric scarp; gray bars show VSO of the historic (1915 + 1954) scarp.

382

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

Stillwater Scarp

Figure 8. Photos of the 1915/1954 scarp at the trench site. (a) The 1.8-m-high scarp directly north of the trench site, looking north. Part of the trench spoil pile appears in the right foreground. (b) Looking southeast through the graben drainage outlet gap with the Clan Alpine Mountains (on the east side of Dixie Valley) in the far distance. The 1915/1954 scarp (∼0.6 m high) crosses the foreground of the photo. The head of our trench (in shadow at lower left) was dug into the scarp face and fault footwall.

Trench Structure and Stratigraphy The main fault plane is exposed beneath the scarp for the full 3.2-m depth of the trench (Figure 9). The fault dips ∼70° northwest and juxtaposes altered Triassic limestone (unit 7) against Quaternary graben-fill sediments (units 1–6) of the HW. Unaltered FW limestone (unit 7d) is very dark gray micrite in massive beds ∼50 cm thick. Within 50 cm of the fault plane, the limestone has been pervasively altered to a green to white stony clay (unit 7b), possibly by hydrothermal processes. The upper half of the fault zone displays a 5- to 10-cm-thick zone of fault gouge (unit 7c) composed of thin bands of white to green to gray gravelly clay containing many small (<2.5 cm) angular rock chips. In the HW, the most distinctive structures are a series of antithetic normal faults that splay upward off the main fault and define wedge-shaped fissures or grabens. These fissure-grabens ascend up the fault plane with the higher (younger) fissures inset into and crosscutting the lower (older) ones. On the HW, we defined six major alluvial sequences, further subdivided into 33 subunits (Figure 10). The most distinctive deposits are thin (7–30 cm), silty to sandy, clast-free, sag-pond sediments (units 1b, 2a, 3d, and 4g; yellow in Figure 9). These sag pond deposits overlie coseismic fissures filled with scarp-derived rubble (units 1d, 2d, 3e, and 4h). Each sag pond deposit (except the youngest at the modern ground surface) is overlain by a coarsening-upward sequence of alluvium. The thickest alluvial deposits (units 3a–c and 6c) are 60- to 80-cm-thick gravelly sands (10%– 40% clasts), poorly to moderately stratified. Sedimen-

tary structures indicate this alluvium was deposited by stream flow toward the scarp. The sandy units maintain a uniform thickness and sedimentology right up to the fault zone, indicating they once extended across the fault zone and were truncated rather than ponded against the scarp. In contrast, the coarsest alluvium occurs in channels trending parallel to the scarp (units 4c, 6a [debris flow], 6b, and 6d). These channels are eroded into the thicker sandy alluvium. Trench Soil Profiles The trench wall exposed two soil profiles: the weak modern surface soil and a better-developed buried soil representing a long depositional hiatus. The surface soil consists of two parts. On the scarp face, a 5-cmthick, grayish-brown A horizon is developed on units 1c and 3f (pebbly sand). On the HW, this soil profile has no A horizon and is represented only by a deeper, diffuse zone of calcite precipitation on clasts bottoms (stages 1–1+; Gile et al., 1966). The calcite precipitation zone extends from unit 1c down through unit 3e. The buried soil lies 1 m below the present ground surface and consists of an Av horizon (unit 4A, 10– 12 cm thick) and a Bk horizon (Unit 4Bk, 15–30 cm thick). The Bk horizon contains stage II+ carbonate coatings and hard subangular blocky ped structures, suggestive of several thousands of years of soil development time. Directly underlying the Bk horizon is a lens of very clean, white silt to fine sand (unit 4d) that we interpret as volcanic ash. Cumulative soil development above the ash is similar to that in soils overlying the Mazama Ash described by Adams and

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

383

McCalpin and Jones

Figure 9. Log of the southwest wall of the Stillwater scarp trench. Polygon patterns reﬂect deposit grain size (lithologic patterns at left). Colors represent diﬀerent coarsening-upward sequences deposited after each paleoearthquake, except that all sag pond deposits at the base of sequences are colored light yellow. Unconformities created by post-earthquake deposition are labeled E1 through E6, with E6 representing the oldest event exposed.

Wesnousky (1999; Bk horizons 30–50 cm thick). The ash may be the Mazama Ash (age 7,627 ± 150 cal yr BP; Egan et al., 2015) or could correlate with other late Pleistocene ashes younger than ∼28 ka (see the next section). QUATERNARY GEOCHRONOLOGY Age control for deformation events was derived from radiocarbon and infrared-stimulated lumines-

384

cence dates of HW sediments (Table 1). The sampling strategy followed that used in paleoseismic studies of normal faults (e.g., McCalpin, 2009, chapter 3). Sample locations are shown on the trench log (Figure 9). Six of our nine samples yielded optically-stimulated luminescence (OSL) dates on sag pond sediments. These ﬂuvial-lacustrine sediments might not be as well zeroed initially as eolian sediments, so we performed the following test on the youngest of them (unit 1b), which accumulated after the latest historic earthquake.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

Stillwater Scarp

Figure 10. Explanation of trench units on hanging wall. Each of the six coarsening-upward alluvial sequences are shown in a diﬀerent color. Textural patterns shown on log are omitted for clarity. Table 1. Descriptions and ages of radiocarbon (C-14) samples and optically stimulated luminescence (OSL) sampled from the Stillwater trench.

Unit

Sample No.

SRL96-11

3a2

AA-28786

3d 4A 4g

SRL96-7 SRC-14-1 (β-103343) SRC-14-1 (AA-23753) SRL96-3

SRL96-9 SRL96-1

Material Sag pond ﬁne sand, polyminerallic Weak soil horizon on sandy alluvium, with stage 1+ CaCO3 Sag pond silt, polyminerallic Buried A horizon (organic silt) Sag pond silt, polyminerallic quartz separate Sag pond silt, polyminerallic Sag pond silt, polyminerallic

Dating Method and Laboratory

Calendar Age (cal yr BP), 2σ1

IR, Dalhousie

0.11 ± 0.02 ka

AMS-NSF

890 ± 60 C-14 yr BP 4.6 ± 0.4 ka 4,560 ± 285 3,693–3,987 44.0 ± 12.1 ka <28.7 ± 3.7 ka 117.0 ± 19.5 ka 34.7 ± 3.4 ka

IR, Dalhousie Beta Analytic AMS-NSF IR, Dalhousie GL, Dalhousie IR, Dalhousie IR, Dalhousie?

1 Calendar-year calibration from the IntCal20 curve (Reimer et al., 2020). Assumes dated samples contain a 50-year average of C-14. IR = infrared stimulated luminescence; AMS = radiocarbon, accelerator mass spectrometry; NSF = NSF AMS Facility, Tucson, Arizona; Beta = Beta Analytic, Miami, Florida; GL = green-light stimulated luminescence; Dalhousie = Luminescence Geochronology Laboratory, Dalhousie University, Halifax, NS, Canada.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

385

McCalpin and Jones

If all sediment grains were completely zeroed during deposition, the OSL sample should yield an age of ∼60 years. But if unit 1b yielded an OSL age of hundreds or thousands of years BP, it would indicate incomplete zeroing of not only unit 1b but also, by extension, all the other dated sag pond sediments (units 3d, 4g, and 6e). Unit 1 dated at 0.11 ± 0.02 ka (110 years), which represents an offset of its true age by about +50 years. This +50 years of inherited luminescence signal amounts to ∼50% of the apparent age in this very young sample. Another check on the polyminerallic silt ages was made on sample SRL96-3, for which a separate analysis was made on sand-sized quartz grains (Table 1) illuminated by green light. By removing the feldspar grains, the effect of anomalous fading (high in feldspars but low in quartz) is diminished, which in theory will decrease the analytical uncertainty. However, the green light analysis does yield smaller uncertainty than the infrared analysis (3.7 ka versus 12.1 ka) and also yields a younger mean age (28.7 ka vs. 44 ka). We prefer the ∼28-ka age for unit 4g, not only on analytical grounds but also because it is consistent with the 34.7-ka age of underlying unit 6e. INTERPRETED SEQUENCE OF LATE QUATERNARY DEFORMATION The trench log was subjected to rigid-block retrodeformation analysis to determine the minimum number of displacement required to create the present geometry (Figure 11). General-case retrodeformation rules for normal faults were assumed (e.g., rigid blocks are translocated or rotated but no ductile deformation; McCalpin, 2009, pp. 260–266). Uphill-facing scarps such as the SS pose some a special-case conditions, so we make three additional assumptions: 1. HW fluvial units that are abruptly truncated at the fault plane originally extended across the fault plane onto the FW (i.e., they buried the uphillfacing scarp). 2. Intact blocks in fissure fills are pieces of FW deposits that fell from the free face. 3. HW sag pond silts ponded against the scarp but did not cross it. The lithology of each fissure-fill block is distinct enough that it can be correlated with specific HW fluvial units, which at one time must have existed also on the FW. In addition, if a fissure fill block is composed of altered bedrock, it means that the fault free face had to be high enough to expose altered bedrock. Thus, it is possible to crudely estimate the height of the free face from each faulting event, even though correlative alluvial deposits may no longer preserved on both HW and FW. Based on stratigraphic superposition, cross-

386

cutting relationships, and the occurrence of two paleosols, we interpret a sequence of six late Quaternary deformation events at Trench 1. Most of the six alluvial units are cleanly truncated at the fault zone, but only one (unit 3) stills exists on the FW (green in Figures 9 –11). This geometry requires periodic erosion of the other units off the upthrown block. This is reasonable since the trench is located in the axis of a channel that drains the entire ridgetop graben. Also note that between steps 10 and 11 in Figure 11, not only the fluvial strata were eroded, but so was an additional 1 m of altered bedrock from the FW. Also note that the time span between the stages is not uniform. For example, of the 35 kyr represented by HW stratigraphy, half of it (15–20 kyr) occurred between stages 11 and 12, which was a long soil-forming interval. No faulting events occurred in this interval. HW deposits occurred in six discrete packages that led us to adopt the following depositional model (after McCalpin, 2005). After each displacement event, an uphill-facing fault scarp was produced and blocked the graben drainage outlet. Most scarp-derived debris rapidly fell into large tension fissures at the scarp base, so colluvial wedges are absent or have small volume. Following the first significant runoff event, thin sag pond silts buried the fissure fill and/or colluvium. As the scarp was progressively buried by alluviation over thousands of years, HW deposits became coarser, indicating a more through-going outlet stream with greater bedload competence. Finally, the scarp was completely buried by HW aggradation, the stream channel flowed unimpeded from the HW over the FW, and channel-facies alluvium was deposited. This coarsening-upward package is often capped by a soil if sufficient time elapses between faulting events. In this last phase, channel scouring can occur on the FW, removing older fluvial deposits and paleosols. However, correlative deposits and paleosols were preserved on the down-dropped HW. DISCUSSION Slippage on the SS during 1915 and 1954 shaking was encouraged by several different factors: (1) the lack of topographic buttressing of the ridge crest to the east in the headscarp area of the Quaternary landslides, (2) topographic amplification of seismic waves on the thinned ridge crest, and (3) incipient downslope movement of the toe of the Quaternary landslide mass (as mapped by Wallace, 1984), which further debuttressed the upper slopes. These factors exist in the outcrop area of the Star Peak Group. Notably, large failures abruptly end at the thrust sheets of the Fencemaker allochthon, which carry the relatively strong igneous rocks of the Humboldt lopolith. In a mechanical

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

Stillwater Scarp

Figure 11. Retrodeformation sequence of south trench wall. The diagrams were constructed by starting with the present geometry (stage 21) and successively removing post-faulting deposits (scarp-derived and other), reversing displacement on the fault, and restoring pre-faulting deposits to their original geometry. Reproducing the present trench wall relationships requires six displacement events, from event U at 35–45 ka to event Z in 1954.

sense, the strong thrust sheets act like geotextile layers in an artificial fill, stiffening the range-front rock mass and preventing large through-going failure planes. Late Quaternary Displacement History of the SS The present SS scarp was formed by the latest three displacement events exposed in our trench. Reconstructed slip on the SS was 50 cm during the 1915 Mw 7.3 earthquake (at a distance of 7 km from the south end of that rupture), compared to 25 cm of slip from the 1954 Mw 6.9 earthquake (at a distance of 30 km from the north end of that rupture). It is not surprising that the larger, closer earthquake induced more slip on the SS. But another factor may be rupture directivity. The 1915 rupture was bilateral, with 30% of its rupture length south of the epicenter. Seismic waves in that section headed toward the SS and were focused in the direction of rupture. In contrast, the 1954 rupture was unilateral, with its epicenter at the north end

of the rupture. All the seismic focusing from 1954 rupture directivity was focused southward, away from the SS. In this light, the 67 cm of slip in event X (∼4.6 ka) was presumably caused by an earthquake either closer, stronger, or more directed than the 1915 event. There are at least nine Quaternary faults within 40 km of the SS that could be responsible (Figure 12). To the south, the Fairview fault; and the Dixie Valley fault (1954 earthquake section); to the east, the eastern Dixie Valley fault and the Clan Alpine fault; to the north, the Stillwater seismic gap section of the Dixie Valley fault and the Sou Hills and Tobin sections of the 1915 Pleasant valley rupture; and to the west, the Rainbow Mountain faults and the West Stillwater fault. Table 2 summarizes Holocene paleoearthquakes known on the closest faults. Two paleoearthquakes span the 4.6-ka time period of trench event X. The closest (only 2 km away) is the Holocene event on the

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

387

McCalpin and Jones Table 2. Holocene earthquake surface ruptures within 40 km of the Stillwater scarp (SS) that might have triggered displacement events on the SS.

Fault Dixie Valley

Dixie Valley

Section Stillwater Seismic Gap, north of DVGF1 gap in Holocene rupture Stillwater Seismic Gap, south of DVGF gap in Holocene rupture (the “Bend event”)

Closest Distance to SS (km)

Age of EQ (cal ka)

Length (km)

Estimated Mw

2.0

4–6 ka?

∼28 km

6.8

9.0

2.2–3.7

7.1–7.3

2.0–2.5

“West Stillwater”

Dixie Valley

Informal name for a 40-km-long part of the Buena Vista Valley fault of U.S. Geological Survey

The Bend south to IXL Canyon (the “IXL event”)

9.2

2.0–2.5

3.7–7.64

<5.6

5.6–7.66

13–35.4

>25

∼75

∼6.97

Reference

Displacement Event on SS That Might Correlate

Wesnousky et al. (2002, p. 8)

Lutz et al. (2002, p. 2)

No SS events recognized in this age range

Wesnousky et al. (2002, p. 8) Bell et al. (2004, p. 1232) Lutz et al. (2002, ﬁgure 7)

Lutz et al. (2002, p. 2) Bell et al. (2004, p. 1240),Adams et al. 2000) Bell and Katzer (1990), Bell et al. (2004, p. 1232)

V or W

Dixie Valley Geothermal Field; there is a 10-km-long gap in Holocene faulting at the range front, centered on the DVGF (Wesnousky et al., 2002). 2 “Several thousand years older than [scarps] south of DVGF” (Wesnousky et al., 2002, p. 8). 3 The distance measured from 4 km north of the DVGF to the north end of faults suped “Dixie Valley fault, Stillwater Seismic Gap section” on Quaternary Fault and Fold Database of the United States ((U.S. Geological Survey 2021). 4 This age range is bracketed by a 3,740 ± 40 C-14 yr BP date and by the age of the Mazama Ash (cited in the references as 6.85 ka) but is now known to be 7.63 ka (Egan et al., 2015). 5 Based on scarp heights of 1–3 m. 6 Event is younger than Mazama Ash, but the 5.6-ka age is deﬁned here as a minimum age, whereas Lutz et al. (2002) deﬁne it as a maximum age. 7 Said to be “comparable in length and vertical displacement to the 1954 rupture” (Bell and Katzer, 1990, p. 624).

northern Stillwater Seismic Gap, north of the Dixie Valley geothermal ﬁeld. Its inferred rupture length of 28 km suggests a moment magnitude of M 6.8 (Wells and Coppersmith, 1994). The second candidate is the latest event on the West Stillwater fault, 9.2 km west of the SS. That rupture was 40 km long with scarps 1–3 m high (assumed mean 1.5 m), suggesting moment magnitudes of M7.0 and M6.9, respectively. Displacement events on the SS are very irregular in time, ranging from two events in 39 years (1915–1954) to no events between 4.5 and 20–25 ka years (a 15to 20-ka soil-forming interval). Such recurrence variability would be very unusual for ruptures on a sin-

388

gle tectonic fault but not necessarily for strong shaking from multiple area faults or fault segments. Large earthquakes on the nine nearby active faults, if they are independent, quasi-periodic seismic sources, could be temporally in phase, out of phase, or anywhere in between. However, Wallace (1987) and Bell and Katzer (1990) proposed that earthquake ruptures in the CNSB occur in short, belt-like temporal clusters followed by long periods of inactivity when rupture clusters shift to parallel belts to the east or west. If true, then the hiatus between 4.6 and 20–25 ka observed on the SS could reﬂect a period of inactivity in this part of the CNSB.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

Stillwater Scarp

Figure 12. Active source faults in the vicinity of the Stillwater scarp (red line with hachures). The “West Stillwater fault” is a local name for a prominent strand of the 76-km-long Buena Vista Valley fault zone (Adams et al., 2000). From Lutz et al. (2002).

Other Examples of Coseismic Ridgetop Spreading Coseismic ridge-crest graben have been reactivated in several well-described historic earthquakes: the 1972 M6.6 San Fernando, California, earthquake (Barrows, 1974; Barrows et al., 1974); the 1989 M6.9 Loma Prieta, California, earthquake (Cotton et al., 1990; Ponti and Wells, 1991); the 2002 M7.9 Denali, Alaska, earthquake (Jibson et al., 2004; Haeussler, 2009); and several M>6 Italian earthquakes (Moro et al., 2007, 2009, 2011, 2012; Aringoli et al., 2016). After the Loma Prieta earthquake, trenching documented that many 1989 scarps had also experienced prehistoric vertical displacements, with slip style and displacement amounts similar to or greater than the 1989 displacements (Nolan and Weber, 1992). Similar trenching studies elsewhere on coseismic gravitational scarps (e.g., McCalpin and Hart, 2003; Tibaldi et al., 2004; Moro et al., 2012; Gori et al., 2014; Mariotto and Tibaldi, 2016; and Komura et al., 2020) demonstrated that an interpretable displacement-event stratigraphy does exist in ridgetop grabens. However, all these studies focused on the failure trigger mechanism. Few discussed why gravitational failures occurred only in small sections of selected ridges when ground shaking of the same strength aﬀected dozens of ridges. Triggered Faulting versus DSGSD and Surface Rupture Hazards If the SS had formed in a valley rather than on a ridge crest, it would have been considered “triggered

faulting” rather than an expression of DSGSD. Triggered faulting is broadly thought to be earthquakeinduced movement on a preexisting fault that is either (1) not physically connected in the subsurface to the seismogenic fault or (2) not even in the zone affected by elastic rebound during the earthquake. Confirming either of these origins is often not possible due to a lack of subsurface or geodetic data. In probabilistic fault displacement hazard analysis, fault ruptures created during strike-slip ruptures that lie farther than 3 km from the seismogenic fault are assumed to be “triggered faults” (Petersen et al., 2011). McCalpin (cited in Baize et al., 2016, p. 37) suggested that only surface ruptures outside the area of elastic rebound qualify as triggered faults, which can mean up to 20 km from the seismogenic fault. Triggered faults are assumed to slip to crustal depths, which means in theory that they could generate smaller, secondary earthquakes, depending on the amount of slip. In contrast, DSGSD faults slip only to accommodate stresses created by steep surface topography (Figure 13). These stresses are assumed to die out rapidly below the base level of the mountain range, so the deeper parts of the preexisting fault do not slip and thus cannot generate secondary earthquakes. We do not have the definitive subsurface data to confirm if the SS is a triggered fault or a DSGSD fault. However, we can evaluate the ratio of scarp height to length for tectonic versus gravitational scarps and compare them to the SS. If the 1.5- to 3.8-kmlong SS were a tectonic surface rupture, it should have maximum displacements in the range of 0.015–0.05 m (Wells and Coppersmith, 1994). However, the measured per-event displacements in the SS trench of 0.25– 1.2 m are nearly an order of magnitude larger than this. McCalpin (2003) noted that the displacement on a sackung scarp of a given length is about eight times larger than that of a tectonic fault of the same length. However, ridge-crest scarps pose a surface rupture hazard to infrastructure regardless of whether they are tectonic or whether the surface slips continue to great depths. This has been recognized since the publication of Hart (2003). Large Range-Front Landslide in the Basin and Range Province Large range-front landslides have been common in the Basin and Range province since the Miocene, when the present horsts began to rise above the grabens. The Nevada Bureau of Mines and Geology (2020) landslide database (https://gisweb.unr.edu/ MyHAZARDS; see also Sturmer and Micander, 2020) contains ∼700 landslides larger than 1 km2 , most of which slid off range fronts. However, most large

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

389

McCalpin and Jones

Figure 13. Comparison of slip distribution on an earthquake-triggered fault (left) to that on a coseismic deep-seated gravitational slope deformation fault (right). In the latter, topography, escarpment slope angles, and slip amounts were derived from the Stillwater scarp.

landslides originate in Tertiary volcanic or volcaniclastic deposits, which in Nevada are commonly weakened by pervasive low-temperature alteration to clay minerals (e.g., Whitebread, 1976; Nash et al., 1995) or near faults and hot springs by hydrothermal alteration. Only six landslides in the database (0.9%) originate in Triassic limestones, such as the Star Peak Group, making the Stillwater slide a rare occurrence based on available landslide data. The low percentage reflects two factors: (1) Mesozoic sedimentary rocks are not abundant in Nevada horsts (e.g., Crafford, 2007), and (2) the rocks must dip toward the range-front escarpment at an angle steep enough to slide (>15°) but less steep than the average topographic slope (<35°). Such dips are found on the west limb of a syncline in the footwall of a thrust fault at our site, but such structures may be uncommon in horsts statewide. In addition to landslides still exposed at the surface (on either FW or HW), buried landslide megabreccia deposits are commonly interbedded within the Neogene graben fill (e.g., Longwell, 1951; Burchfiel, 1966; Krieger, 1977; and French and Guth, 2016). In most cases, the source slides are so old (Miocene to Pliocene) that any corresponding FW deposits have been removed by erosion on steep range fronts. In contrast, our Qs landslide is young enough that its deposits are still preserved on the range-front escarpment. Its estimated H/L ratio and volume are similar to other famous rockslide avalanches (e.g., the Goldau, Austria, rockslide of 1806), with a volume of 30–40 × 106 m3 and an “energy line” slope of 0.21 (12°) (see Hsü, 1978, table 1). With an energy line slope of 0.23–0.31, the Qs landslide qualifies as a long-runout landslide (H/L < 390

0.6). A similar but smaller rockslide was recently described by Shaller et al. (2020) in Eureka Valley, California (deposit volume 5 × 106 m3 , age ∼100 ka). That range-front fault has a vertical slip rate nearly identical to the Dixie Valley fault (∼0.5 mm/yr), and there is even a sackung scarp above its headscarp.

CONCLUSIONS The SS was reactivated by earthquake shaking in 1915 and 1954 on a section of range front that had previously failed in a >2-km2 -long-runout landslide in the late middle Pleistocene. That landslide’s evacuation zone left the range crest as a narrow oversteepened ridge that contained an older (middle Tertiary?) westdipping normal fault. Our mapping and trenching evidence indicates that over the past ∼35–45 ka, the normal fault had slipped down to the west on six occasions with the latest two events occurring in 1915 and 1954. Displacements ranged from 40 to 120 cm per event (average 71 cm), with a long-term average recurrence interval of 7–9 ka (average slip rate 0.01–0.12 mm/yr). We know that the latest two displacement events (Z and Y) were associated with earthquakes to the south and north of the SS, respectively. However, for prehistoric displacement events X, W, and U, there are multiple faults that could have triggered ridgetop spreading. For event X, there are two nearby faults with documented paleoearthquakes in the appropriate age range (Table 2). For older events U, V, and W, the paleoseismic history of area faults is insuﬃciently known to correlate with SS events.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

Stillwater Scarp

Spatially, coseismic DSGSD requires a combination of strong ground motion, favorably oriented weak rocks, and possibly topographically amplified triggers. Areas possessing these characteristics are rare, as shown by the scarcity of coseismic ridgetop scarps in Nevada. However, sites susceptible to future coseismic DSGSD can be identified today on steep range fronts by locating where such lithologic, structural, and topographic constraints coincide. ACKNOWLEDGMENTS Jocasta Champion and Jason Amato assisted in mapping the SS, measuring scarp profiles, digging and logging the trench, and removing snakes. Dorothy Godfrey-Smith (Dalhousie University) processed the luminescence dating samples. John Caskey (San Francisco State University) helped us understand the local neotectonic setting. Richard P. Smith (Smith Geologic and Photographic Services, Nathrop, Colorado) provided digital geologic mapping of range-front structures and his extensive knowledge of the Dixie Valley fault system and geothermal field. Field support was provided by National Science Foundation grant EAR9506371 to principal investigator Peter Birkeland, University of Colorado, Boulder. REFERENCES Adams, K.; Sawyer, T. L.; and Anderson, R. E. (Compilers), 2000, Fault number 1638, Buena Vista Valley fault zone. In Quaternary Fault and Fold Database of the United States: Electronic document, available at https://earthquakes.usgs.gov/ hazards/qfaults Adams, K. D. and Wesnousky, S. G., 1999, The Lake Lahontan highstand: Age, surficial characteristics, soil development, and regional shoreline correlation: Geomorphology, Vol. 30, pp. 357–392. Aringoli, D.; Farabollini, P.; Giacopetti, M.; Materazzi, M.; Paggi, S.; Pambianchi, G.; Pierantoni, P. P.; Pistolesi, E.; Pitts, A.; and Tondi, E., 2016, The August 24th 2016 Accumoli earthquake: Surface faulting and deep-seated gravitational slope deformation (DSGSD) in the Monte Vettore area: Annals Geophysics, Vol. 59, pp. 1–8. Baize, S.; Cinti, F.; Costa, C.; Dawson, T.; Elliott, A.; Guerreri, L.; McCalpin, J.; Okumura, K.; Scotto, O.; Takao, M.; Villamor, P.; and Walker, R., 2016, Surface rupture hazard database for seismic hazard assessment: Institut de radioprotection et de sûreté nucléaire Report RT/PRPDGE/2016-00022, 40 p. Barrows, A. G., 1974, Surface effects and related geology of the San Fernando earthquake in the foothill region between Little Tujunga and Wilson canyons. In Oakeshott, G. B. (Editor), California Division of Mines and Geology Bulletin 196, pp. 97–117. Barrows, A. G.; Kahle, J. E.; Weber, F. H., Jr.; Saul, R. B.; and Morton, D. M., 1974, Surface effects maps of the San Fernando earthquake area. In Oakeshott, G. B. (Editor), California Division of Mines and Geology Bulletin 196, Plate 2, scale 1:18,000.

Bell, J. W.; Caskey, S. J.; Ramelli, A. R.; and Guerreri, L., 2004, Pattern and rates of faulting in the Central Nevada Seismic Belt, and paleoseismic evidence for prior beltlike behavior: Bulletin Seismological Society America, Vol. 94, No. 4, pp. 1229–1254. Bell, J. W. and Katzer, T., 1990, Timing of late Quaternary faulting in the 1954 Dixie Valley earthquake area, central Nevada: Geology, Vol. 18: pp. 622–625. Burchfiel, B. C., 1966, Tin Mountain landslide, southeastern California, and the origin of megabreccia: Geological Society America Bulletin, Vol. 77, No. 1, pp. 95–99. Caskey, S. J.; Wesnousky, S.; Zhang, P.; and Slemmons, D. B., 1996, Surface faulting of the 1954 Fairview Peak (Ms7.2) and Dixie Valley (Ms6.8) earthquakes, central Nevada: Bulletin Seismological Society America, Vol. 86, No. 3, pp. 761–787. Cotton, W. R.; Fowler, W. L.; and Van Velsor, J. E., 1990, Coseismic bedding plane faults and ground ﬁssures associated with the Loma Prieta earthquake of 17 October 1989. In McNutt, S. R. and Sydnor, R. H. (Editors), The Loma Prieta (Santa Cruz Mountains), California, Earthquake of 17 October 1989: California Division of Mines and Geology, Special Publication 104, pp. 95–103. Crafford, A. E. J., 2007, Geologic Map of Nevada: U.S. Geological Survey Data Series 249, 46 p. Egan, J.; Staff, R.; and Blackford, J., 2015, A high-precision age estimate of the Holocene Plinian eruption of Mount Mazama, Oregon: Holocene, Vol. 25, pp. 1054–1067. French, D. E. and Guth, P. L., 2016, Megabreccias of the Sheep Range, Clark County, Nevada: Nevada Petroleum and Geothermal Society 2016 Field Trip Guidebook, NPS26, 50 p. Gile, L. H.; Peterson, F. F.; and Grossman, R. B., 1966, Morphological and genetic sequences of carbonate accumulation in desert soils: Soil Science, Vol. 101, pp. 347–360. Gori, S.; Falcucci, E.; Dramis, F.; Galadini, F.; Galli, P.; Giaccio, B.; Messina, P.; Pizzi, A.; Sposato, A.; and Cosentino, D., 2014, Deep-seated gravitational slope deformation, largescale rock failure, and active normal faulting along Mt. Morrone (Sulmona basin, Central Italy): Geomorphological and paleoseismological analyses: Geomorphology, Vol. 208, pp. 88–101. Hart, E. W. (Editor), 2003, Ridge-Top Spreading in California; Contributions toward Understanding a Significant Seismic Hazard: California Geological Survey, CD 2003-05, 2 CD-ROMs. Haeussler, P. J., 2009. Surface Rupture Map of the 2002 M7.9 Denali Fault Earthquake, Alaska; Digital Data: U.S. Geological Survey Data Series 422, 9 p., available at https://doi.org/10.3133/DS422 Hsü, K. J., 1978, Albert Heim: Observations on landslides and relevance to modern interpretations. In Voight, B. (Editor), Rockslides and Avalanches, Vol. 1: Elsevier, Amsterdam, Netherlands, pp. 70–93. Jibson, R. W.; Harp, E. L.; Schulz, W.; and Keefer, D. K., 2004, Landslides triggered by the 2002 Denali fault, Alaska, earthquake and the inferred nature of strong shaking: Earthquake Spectra, Vol. 2, No. 3, pp. 669–691. Komura, K.; Kaneda, H.; Tanaka, T.; Kojima, S.; Inoue, T.; and Nishio, T., 2020, Synchronized gravitational slope deformation and active faulting; A case study on and around the Neodani fault, central Japan: Geomorphology, Vol. 365, pp. 107241-1–107241-22. Krieger, M. H., 1977, Large Landslides, Composed of Megabreccia, Interbedded with Miocene Basin Deposits, Southeastern Arizona: U.S. Geological Survey Professional Paper 1008, 25 p.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

391

McCalpin and Jones Longwell, C.R., 1951, Megabreccia developed downslope from large faults [Ariz.-Nev.]: American Journal Science, Vol. 249, No. 5, pp. 343–355. Lutz, S. J.; Caskey, S. J.; Mildenhall, D. D.; Browne, P. R. L.; and Johnson, S. D., 2002, Dating sinter deposits in northern Dixie Valley, Nevada—The paleoseismic record and implications for the Dixie Valley geothermal system. In Proceedings of 27th Workshop on Geothermal Reservoir Engineering, SGP-TR-171: Stanford University, Stanford, CA, pp. 284–297. Mariotto, F. P. and Tibaldi, A., 2016, Inversion kinematics at deep-seated gravity slope deformations revealed by trenching techniques: Natural Hazards Earth System Science, Vol. 16, pp. 663–674. McCalpin, J. P., 2003, Criteria for determining the seismic significance of sackungen and other scarplike landforms in mountainous regions. In Hart, E. W. (Editor), Ridge-Top Spreading in California; Contributions toward Understanding a Significant Seismic Hazard: California Geological Survey, CD 2003-05, 2 CD-ROMs. McCalpin, J. P., 2005, Late Quaternary activity of the Pajarito fault, Rio Grande rift of northern New Mexico, USA: Tectonophysics, Vol. 408, No. 1–4, pp. 213–236. McCalpin, J. P. (Editor), 2009, Paleoseismology, 2nd ed.: Elsevier, Amsterdam, Netherlands, 608 p. McCalpin, J. P. and Hart, E. W., 2003, Ridge-top spreading features and relationship to earthquakes, San Gabriel Mountains Region, Southern California—Part B: Paleoseismic investigations of ridge-top depressions. In Hart, E. W. (Editor), RidgeTop Spreading in California: Contributions toward Understanding a Significant Seismic Hazard: California Geological Survey, CD 2003-05, 2 CD-ROMs. McCann, W. R.; Nishenko, S. P.; Sykes, L. R.; and Krause, J., 1979, Seismic gaps and plate tectonics: Pure Applied Geophysics, Vol. 117, pp. 1082–1147. Moro, M.; Chini, M.; Saroli, M.; Atzori, S.; Stramondo, S.; and Salvi, S., 2011, Analysis of large, seismically induced, gravitational deformations imaged by high resolution COSMOSkyMed SAR: Geology, Vol. 39, pp. 527–530. Moro, M.; Saroli, M.; Gori, S.; Falcucci, F.; and Messina, P., 2012, The interaction between active normal faulting and large scale gravitational mass movements revealed by paleoseismological techniques: A case study from central Italy: Geomorphology, Vol. 151–152, pp. 164–174. Moro, M.; Saroli, M.; Salvi, S.; Stramondo, S.; and Doumaz, F., 2007, The relationship between seismic deformation and deep-seated gravitational movements during the 1997 Umbria–Marche (Central Italy) earthquakes: Geomorphology, Vol. 89, pp. 297–307. Moro, M.; Saroli, M.; Tolomei, C.; and Salvi, S., 2009, Insights on the kinematics of deep seated gravitational slope deformations along the 1915 Avezzano earthquake fault (Central Italy), from time-series DInSAR: Geomorphology, Vol. 112, pp. 261–276. Nash, T.; Utterback, W. C.; and Trudel, W. S., 1995, Geology and Geochemistry of Tertiary Volcanic Host Rocks, Sleeper Gold-Silver Deposit, Humboldt County, Nevada: U.S. Geological Survey Bulletin 2090, 63 p. Nevada Bureau of Mines and Geology, 2020, MyHAZARDS Web Application. In Geohazards, available at http://datanmbg.opendata.arcgis.com/pages/geohazards Nichols, K. M. and Silberling, N. J., 1977, Stratigraphy and Depositional History of the Star Peak Group (Triassic), Northwestern Nevada: Geological Society of America Special Paper 178.

392

Nolan, J. M.; and Weber, G. E., 1992, Evaluation of ground cracking caused by the 1989 Loma Prieta earthquake, Santa Cruz County, California: Case histories. In Sharma, S. (Editor), Proceedings of the 28th Symposium on Engineering Geology and Geotechnical Engineering: University of Idaho, Moscow, ID, pp. 272–286. Petersen, M. D.; Dawson, T. E.; Chen, R.; Cao, T.; Wills, C. J.; Schwartz, D. P.; and Frankel, A. D., 2011, Fault displacement hazard for strike-slip faults: Bulletin Seismological Society America, Vol. 101, No. 2, pp. 805–825. Plank, G.R., 1998, Structure, Stratigraphy, and Tectonics of a Part of the Stillwater Escarpment and Implications for the Dixie Valley Geothermal System: Unpublished M.S. Thesis, University of Nevada, Reno, 153 p. Ponti, D. J. and Wells, R. E., 1991, Oﬀ-fault ground ruptures in the Santa Cruz Mountains, California: Ridge-top spreading versus tectonic extension during the 1989 Loma Prieta earthquake: Bulletin Seismological Society America, Vol. 81, No. 5, p. 1480–1510. Reimer, P. J. and 41 others, 2020, The IntCal20 Northern Hemisphere radiocarbon age calibration curve (0-55 cal kBP): Radiocarbon, Vol. 62, No. 4, pp. 725–757. Shaller, P. J.; Doroudian, M.; and Hart, M. W., 2020, The Eureka Valley landslide: Evidence of a dual failure mechanism for a long-runout landslide: Lithosphere, Vol. 2020, No. 1, pp. 1–26. Slemmons, D. B., 1957, Geological eﬀects of the Dixie ValleyFairview Peak, Nevada, earthquake of December 16, 1954: Seismological Society America Bulletin, Vol. 47, No. 4, pp. 353–375. Smith, R. P.; Wisian, K. W.; and Blackwell, D. D., 2001, Geologic and geophysical evidence for intra-basin and footwall faulting at Dixie Valley, Nevada: Geothermal Resources Council Transactions, Vol. 25, pp. 323–326. Soil Survey Staff, 2014, Keys to Soil Taxonomy, 12th ed.: U.S. Department Agriculture, Natural Resources Conservation Service, Washington, DC, 360 p. Speed, R. C., 1976, Geologic Map of the Humboldt Lopolith and Surrounding Terrain, Nevada: Geological Society America Map Chart Series MC-14. Sturmer, D. M. and Micander, R. E., 2020, A new database of large-scale bedrock landslides in Nevada, USA: Geological Society America Abstracts Programs, Vol. 52, No. 6, available at https://gsa.confex.com/gsa/2020AM/meetingapp.cgi/ Paper/356734 Tibaldi, A.; Rovida, A.; and Corazzato, C., 2004, A giant deepseated slope deformation in the Italian Alps studied by paleoseismological and morphometric techniques: Geomorphology, Vol. 58, pp. 27–47. U.S. Geological Survey, 2021, Quaternary Fault and Fold Database for the United States: Electronic document, available at https://www.usgs.gov/natural-hazards/earthquakehazards/faults Wallace, R. E., 1978, Patterns of faulting and seismic gaps in the Great Basin Province. In Isaacks, B. L. and Plafker, G. (Editors), Proceedings of Conference VI, Methodology for Identifying Seismic Gaps and Soon-to-Break Gaps: U.S. Geological Survey Open-File Report 78-943, pp. 857–868. Wallace, R. E., 1984, Fault Scarps Formed during the Earthquakes of October 2, 1915 in Pleasant Valley, Nevada, and Some Tectonic Implications: U.S. Geological Survey Professional Paper 1274-A, 33 p. Wallace, R. E., 1987, Grouping and migration of surface faulting and variations in slip rates on faults in the Great Basin

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

Stillwater Scarp province: Bulletin Seismological Society America, Vol. 77, pp. 868–876. Wallace, R. E. and Whitney, R. A., 1984, Late Quaternary history of the Stillwater Seismic Gap, Nevada: Bulletin Seismological Society America, Vol. 74, No. 1, pp. 301–314. Wells, D. L. and Coppersmith, K. J., 1994, New empirical relationships among magnitude, rupture length, rupture width, rupture area, and surface displacement: Bulletin Seismological Society America, Vol. 84, No. 4, pp. 974–1002.

Wesnousky, S. G.; Caskey, S. J.; and Bell, J. W., 2002, Recency of faulting and neotectonic framework in the Dixie Valley Geothermal Field and other geothermal ﬁelds of the Basin and Range: Unpublished Final Technical Report submitted to U.S. Department of Energy, Contract DE-FG07-98ID13620, 26 p. Whitebread, D. H., 1976, Alteration and Geochemistry of Tertiary Volcanic Rocks in Parts of the Virginia City Quadrangle, Nevada: U.S. Geological Survey Professional Paper 936, 43 p.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 377–393

393

Magnitude and Timing of the Tiltill Rockslide in Yosemite National Park, California CHRISTOPHER J. PLUHAR* KIERSTI R. FORD Department of Earth and Environmental Sciences, California State University, Fresno, 2576 East San Ramon Ave M/S ST24, Fresno, CA 93740

GREG M. STOCK National Park Service, Yosemite National Park, Resources Management and Science, 5083 Foresta Road, P.O. Box 700, El Portal, CA 95318

JOHN O. STONE Department of Earth and Space Sciences and Quaternary Research Center, University of Washington, Box 351310, 70 Johnson Hall, Seattle, WA 98195-1310

SUSAN R. ZIMMERMAN Lawrence Livermore National Laboratory Center for Accelerator Mass Spectrometry, 7000 East Avenue L-397, Livermore, CA 94550

Key Terms: Rockslide, 10 Be Cosmogenic Nuclide Exposure Dating, Rock Slide, Rock Avalanche, Long Runout Landslide, Seismic Trigger ABSTRACT Yosemite National Park, California, is one of the best-documented sites of historical rockfalls and other rock slope failures; however, past work shows that this record does not capture the infrequent largest occurrences, prehistoric events orders of magnitude larger than the largest historic ones. These large prehistoric events are evident as voluminous bouldery landslide deposits, permitting volume and age quantification to better understand local volume–frequency relationships, potential triggering mechanisms, and the hazard such events might pose. The Tiltill rockslide in northern Yosemite is one such example, consisting of 2.1 × 106 m3 ± 1.6 × 106 m3 of talus (1.5 × 106 m3 original volume of rock mass) that slid across the floor of Tiltill Valley, partially damming Tiltill Creek to create a seasonal pond that drains through and around the rockslide mass. This volume and the rockslide’s effective coefficient of friction, 0.47, place it near the boundary between long-runout landslides and ordinary Coulomb failure. Although the rockslide superficially appears to consist of two separate lobes, statistically indistinguishable 10 Be exposure dates from eight samples indicate

*Corresponding author email: cpluhar@csufresno.edu

a single event that occurred at 13.0 ± 0.8 ka. The age of the Tiltill rockslide and its relatively low elevation compared to equilibrium line altitudes at this place and time make glacial debutressing a highly unlikely triggering mechanism. Seismic shaking associated with fault rupture along the eastern Sierra Nevada is shown to be a plausible but unverified trigger. INTRODUCTION The Sierra Nevada mountains of central California frequently experience rock slope failures such as rockfalls and rockslides (Harp et al., 1984; Bull et al., 1994; Wieczorek and Jäger, 1996; Wieczorek et al., 2000; Wieczorek, 2002; Jibson, 2007; Stock et al., 2012; Zimmer et al., 2012; and Stock et al., 2013), as well as less frequent rock avalanches (Wieczorek, 2002; Stock and Uhrhammer, 2010; Cordes et al., 2013; and Pacheco et al., 2020). The glacially steepened cliffs and exfoliation-jointed granitic bedrock that comprise most of the Sierra Nevada promote such slope failures (Keefer, 1984b; Stock et al., 2012 and references therein). These events pose significant hazards to people and infrastructure, but also reveal fundamental aspects of these geologic processes. An extensive database of historic rock slope failures in Yosemite National Park in the central part of the range includes events as early as AD 1857 and consists of nearly 1,500 events (Stock et al., 2013, and subsequent unpublished data). These historical slope failures have sometimes caused considerable damage to

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 395–407

395

Pluhar, Ford, Stock, Stone, and Zimmerman

structures, especially in developed areas of Yosemite Valley, and numerous triggers contribute to their occurrence, including precipitation, snowmelt, thermal stresses (Collins and Stock, 2016), seismicity, and others (Wiezcorek and Jäger, 1996; Stock et al., 2013). Precipitation and snow melt frequently trigger rockfalls, whereas seismic triggering occurs much less frequently but tends to result in larger-volume slope failures than the aforementioned processes (Wieczorek and Jäger, 1996). For example, the Yosemite database includes events related to the 1980 Mammoth Lakes earthquakes, M 6.2, M 5.9, and M 6.2 (McJunkin and Bedrossian, 1980), which triggered thousands of rockfalls in the Sierra Nevada (Harp et al., 1984), including nine significant rockfalls and rockslides in Yosemite Valley (Stock et al., 2013). The database provides important insight into process, but could be enhanced in key ways. Previous researchers have analyzed the existing Yosemite historical rockfall database, revealing the inverse power law governing rockfall volume–frequency relationships for this location for mid-sized rock slope failures, those with volumes 100–100,000 m3 (Wieczorek et al., 1999; Dussauge et al., 2003; Guzzetti et al., 2003; and Guerin et al., 2020). However, their negative power law distribution is not well constrained for the very small and very large events (Guzzetti et al., 2003) due to sampling bias. Very small events (ࣘ1 m3 ) may not be noticed or easily measured, whereas the very large events occur less often than the historic record is long. Additional frequency–volume data from both the small (Guerin et al., 2020) and very large events would aid in constraining this power law. This is significant because small changes in the power law exponent can have profound effects on overall hazard. Also, large historical failures (>500 m3 ) are relatively rare, but they rapidly move massive amounts of material that frequent, smaller rockfalls may spend millennia accomplishing (Guzzetti et al., 2003). Thus, the larger infrequent slope failures are potentially more hazardous and definitively more effective as agents of landscape evolution (Guerin et al., 2020). Furthermore, the present assumption that relatively small rockfalls and larger rock avalanche volume–frequency relationships are described by the same function remains untested. For example, Keefer (1984a) showed that there are different earthquake triggering-threshold magnitudes for rockfalls, rockslides, and rock avalanches, suggesting that the governing volume–frequency power laws could differ between those processes. Because the largest historical slope failure in Yosemite is ∼100,000 m3 , the especially large (ࣙ106 m3 ) and most hazardous rock slope failures are not well represented in the event database (Guzzetti et al.,

396

Figure 1. Location of Yosemite National Park (YNP) and geologic map of the study area around Tiltill rockslide, adapted from Huber et al. (1989). The inset map of California and Nevada shows the study area’s location relative to major seismicity sources; CNSB = Central Nevada Seismic Belt, SAF = San Andreas Fault, and WL = Walker Lane.

2003; Stock et al., 2013; and Guerin et al., 2020). Relict deposits in Yosemite indicate that these much larger slope failures occurred prehistorically (Wieczorek et al., 1999). Studies of these prehistoric events are necessary to rectify this data gap, provide insight into long-term landscape evolution rates, and enhance understanding of mass wasting processes. Consequently, recent eﬀort has focused on investigating deposits resulting from the largest of these prehistoric slope failures in Yosemite (Stock and Uhrhammer, 2010; Cordes et al., 2013; and Pacheco et al., 2020). Here, we add to the dataset by investigating the runout characteristics, volume, age, and potential triggering mechanisms of the Tiltill rockslide. GEOLOGIC SETTING The Tiltill rockslide lies in the northern portion of Yosemite National Park, on the central Sierra Nevada’s windward, i.e., high-precipitation, western slope in the middle elevations of the range at ∼1,900– 2,500 m. The deposit lies within the southwestdraining Tiltill Valley (Figure 1), a tributary of the westward-ﬂowing Tuolumne River. Nearby Hetch Hetchy Valley experiences 92 cm average annual rainfall equivalent, with much of that falling as 162 cm average annual snowfall (Western Regional Climate Center, 2021); given that the study site lies 800 m higher than Hetch Hetchy Valley, Tiltill Valley probably experiences somewhat higher amounts of precipitation.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 395–407

Magnitude and Timing of the Tiltill Rockslide

The asymmetric Sierra Nevada displays a broad western slope and a steep, fault-bounded eastern escarpment. The range originated as a continental volcanic arc during the Mesozoic (see Dickinson, 2006, and the references therein), followed by a period of Paleogene erosion. Although the detailed elevation history of the range is debated (e.g., Cassel et al., 2009), it is likely that at least some surface uplift occurred during the last 10 million years, as the range tilted westward with normal faulting on its eastern side (Huber, 1990; Wakabayashi, 2013; and Martel et al., 2014). Early Paleozoic to Cretaceous metamorphic, extensive Mesozoic plutonic, and Neogene volcanic and volcaniclastic rocks dominate the Sierra Nevada, and are thinly mantled in some places by Quaternary surficial deposits (Bateman et al., 1966). Cretaceous granite to quartz diorite composes the majority of Yosemite National Park (Huber et al., 1989). Deep canyons such as Yosemite and Hetch Hetchy Valleys result from rock uplift of the range, which drove bedrock incision by westward-flowing rivers (e.g., Stock et al., 2005). Pleistocene glaciation of the range deepened and widened these canyons into U-shaped valleys (Blackwelder, 1931). The most recent significant glaciation occurred during the Last Glacial Maximum (LGM), locally termed the Tioga glaciation, which further incised these canyons into the U-shaped valleys as they appear today (Matthes, 1930; Wahrhaftig et al., 2019). Ice of the Tioga glaciation (approximately 15,000– 20,000 years ago) completely filled Hetch Hetchy Valley, and almost filled Tiltill Valley, whereas Yosemite Valley was last full during the Sherwin glaciation (approximately 780,000 years ago) (Huber et al., 1989; Wahrhaftig et al., 2019). After Tioga glaciation, the Recess Peak glacial advance, occurring sometime between 14 and 12 ka, expanded glacial cover slightly compared to interglacial conditions (Clark and Gillespie, 1997; Phillips 2016, 2017), but is only known to have affected the Sierran crest, at an equilibrium line altitude of 3,500 m, and not the lower elevations near Tiltill rockslide at ∼2,000 m elevation. The study area’s deeply entrenched glacial valleys with steep cliffs are now dominantly eroded by rockfall, rockslides, and rock avalanches. THE TILTILL ROCKSLIDE Although not extensively studied before now (the brief summary by Wieczorek, 2002 constituting the only prior study), the Tiltill rockslide is one of the largest rock slope failure deposits in Yosemite National Park. The rockslide originated from the southeastern slope of 2,560-m–elevation Mount Gibson (Figures 1–3; Wieczorek, 2002), the summit of which projected above the valley glacier occupying Tiltill

Figure 2. Overviews of the Tiltill rockslide. (a) Oblique aerial view looking north showing key features of the Tiltill rockslide. From a source area just below the Last Glacial Maximum glacial trimline (blue dashed line) on the southeastern slope of Mount Gibson, rock debris extends as far as 900 m across Tiltill Valley in two distinct lobes (north and south lobes). The larger south lobe extends all the way across the valley, damming Tiltill Creek and forming a seasonal pond. Runout of the smaller north lobe was constrained by a prominent bedrock knob. (b) Panoramic view looking southwest from the bedrock knob in (a) showing the rockslide source area, north and south depositional lobes, and seasonal pond.

Valley during the LGM (Tioga) (Wahrhaftig et al., 2019). Consequently, the Tioga glaciation left weathered rock in place on the upper slopes of Mount Gibson, whereas the middle and lower slopes, including the rockslide source area, were below the glacier trimline and thus scoured by glacial erosion (Figure 2). Our investigation indicates that the source area consists of a weathered, undulating but relatively planar surface oriented subparallel to the overall slope of Mount Gibson in this area. The plane projects into the sky at the head and is bounded by a major scarp to the northeast and a smaller one to the southwest. This geometry, coupled with visual evidence of slope-parallel discontinuities along the northeastern edge of the scar, suggests that failure initiated as a block slide along surface-subparallel exfoliation joints. The slide disaggregated into a bouldery mass during runout. The highest point on the headscarp of the Tiltill rockslide source area lies at 2,467 m elevation, and the

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 395–407

397

Pluhar, Ford, Stock, Stone, and Zimmerman

Figure 3. An aerial image showing locations of samples for cosmogenic 10 Be exposure dating on the north and south lobes of the Tiltill rockslide and locations of representative profiles through the rockslide. Profiles A-A’ (profile taken transverse to Tiltill Valley and parallel to the rockslide direction of motion), B-B’ (profile taken subparallel to Tiltill Valley and across the rockslide deposit), and C-C’ (Tiltill Creek profile taken as rockslide base reference) are representative profiles used to estimate the volume of the rockslide.

lowest point of the deposit sits at 1,952 m. Thus, the slide debris dropped a maximum of 515 m vertically while running out a maximum of 1,095 m horizontally (Figure 3), a fall-height–to–runout-length ratio (H/L, also known as effective coefficient of friction) of 0.47, at the lower end of ordinary values exhibited during Coulomb failure, 0.5–0.8 (Collins and Melosh, 2003).This value and the volume of the Tiltill rockslide put it at the lower end of events classified as long runout landslides or sturzstroms (Collins and Melosh, 2003). Figure 4 shows the rockslide volume vs. relative runout ratio (L/H, the inverse of effective coefficient of friction), compared to some previously documented long-runout landslides (Dade and Hubbert, 1998). On this graph, Coulomb L/H ranges from 1.25 to 2, while the value for Tiltill rockslide lies above that at 2.13. The graphed examples have been used to suggest that long-runout landslide characteristics are largely independent of atmospheric effects or variations in gravity, i.e., they exhibit similar distributions on this graph for both Earth and extraterrestrial cases. Instead, a nonCoulomb process such as acoustic fluidization could promote long-runout landslides (Collins and Melosh, 2003). Since our data fall in the long-runout landslide data array on Figure 4, it suggests that the Tiltill rockslide could have experienced such a non-Coulomb transport process. Clasts within the Tiltill rockslide are mostly composed of coarse-grained, dark-gray, hornblende-

398

biotite, pyroxene-bearing Quartz Diorite of Mount Gibson, part of the Intrusive Suite of Jack Main Canyon (Kistler, 1973; Huber et al., 1989). This rock type composes the entire Tiltill Valley at this location and is thus not diagnostic for demonstrating the exact source location for the rockslide. However, the rockslide scar is evident in the topography (Figure 2) as previously described.

Figure 4. Volume vs. relative runout ratio (length/height, L/H) for the Tiltill rockslide compared to a database of long runout landslides. Ordinary Coulomb-failure–dominated landslides occur at L/H ratios of 1.25–2. Above certain volume, landslides depart from this range. Tiltill Slide L/H is 2.13, slightly higher than the range for Coulomb failure and within the array of long-runout landslides. Adapted from Dade and Hubbert (1998).

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 395–407

Magnitude and Timing of the Tiltill Rockslide

The Tiltill rockslide deposit consists of two lobes, coarse boulder fields exhibiting hummocky topography, along with some vegetated areas. We demarcate the lobes by the boundary between bouldery talus that does not support significant vegetation and talus areas that do support vegetation. The deposit extends from the base of the rockslide scar on Mount Gibson approximately 900 m to the south-southeast across Tiltill Valley (Figures 2 and 3). Individual boulder volumes range from tenths of a cubic meter up to 60 m3 . Initially, we surmised that the two unvegetated coarse talus exposures (Figures 2 and 3) were separate mass wasting deposits, here referred to as the north and south lobes. In the distal portion of the southernmost lobe (south lobe), the width of the deposit ranges from 170 to 250 m, whereas the distal northernmost lobe (north lobe) ranges in width from 70 to 100 m. The north lobe of the Tiltill rockslide abuts a bedrock knob that likely prevented the northeastern edge from running out farther (Figure 2). The larger south lobe of the deposit extends across Tiltill Valley, creating a landslide-dammed pond that Wieczorek (2002) stated is 15 m deep, but is probably less than 5 m deep when full. During late summer the pond is often dry, which enabled our estimate of its maximum depth, but in winter and spring it fills and drains through and around the rockslide deposit, allowing Tiltill Creek to continue downstream to Hetch Hetchy Reservoir. The pond has accumulated an unknown thickness of sandy sediment since emplacement, some of which is likely derived from the rockslide deposit. Wieczorek (2002) also suggested that the distal portion of the Tiltill rockslide deposit turns southwest, parallel to Tiltill Valley, but we did not find clear evidence of that. Since deposition of the Tiltill rockslide, the surrounding forest has encroached upon parts of the debris field (Figures 2 and 3). Aerial photography and GIS analysis enabled an estimate of the rockslide volume and an evaluation of the possibility that it consisted of more than one deposit. Topography visible in aerial imagery and in topographic profiles derived from 3-m–resolution airborne LiDAR data demonstrate no topographic low between the two lobes that might indicate separate emplacement. Instead, the two lobes and the vegetated area between them all appear to be part of a single rockslide landform, with the vegetated area being composed of finer-grained rock debris, thus more readily supporting plant life (Figure 3). VOLUME ESTIMATION A key element of understanding rock slope failure triggering conditions and assessing the hazard posed by large failures is determining event sizes. Typically,

the volume of a rock slope failure deposit is the measurable quantity representing size. However, volume estimation can be challenging; Wieczorek (2002) addressed the difficulty of estimating the volume of the Tiltill rockslide, “The thickness of the deposit in the bottom of the canyon is difficult to estimate because of the irregular stepped topography in other parts of Tiltill Canyon; consequently, volume of the rockslide deposit was not estimated.” In our study, to approximate the volume of the Tiltill rockslide deposit, we used ArcGIS (version 10.1) to calculate the difference between the current land surface of the deposit and an inferred pre-slide land surface. The current surface elevation consisted of bare-earth–filtered, 3-m LiDAR data derived from the Airborne Snow Observatories program (Airborne Snow Observatories, Inc, 2021). We mapped the rockslide deposit based on field observations and the expected surface expression of such deposits. The rockslide deposit forms a prominent knickpoint in the topographic profile of Tiltill Creek (Figure 3), which permitted inference of the likely pre-slide stream profile. We assumed a constant gradient for the pre-rockslide stream rather than assuming a concavity or convexity for which we had no evidence. Then, we constructed a series of five valley-transverse, rockslide-parallel cross sections using the crossing point of the inferred pre-slide stream profile as the minimum elevation of the pre-slide ground surface for any given cross section (Ford, 2014). We inferred subsurface rockslide–bedrock contacts that were consistent with the current bedrock surfaces around the slide and the geomorphology of a glaciated valley. We then created eight valley-parallel cross sections across the rockslide mass that were consistent with the valleytransverse cross sections (Ford, 2014). The net of cross sections was adjusted where necessary for consistency between them and the surrounding topography in order to minimize unexplained inflection points and maximize smoothly varying pre-slide topography. The cross sections were then converted to a 3-D surface, a pre-rockslide ground-surface-elevation raster, using nearest-neighbor interpolation (Ford, 2014). We used the ArcGIS “cut and fill” tool in 3-D analyst to subtract the pre-rockslide surface from present-day topography, yielding the approximate rockslide volume. In order to eliminate rockfall talus younger than Tiltill rockslide from the calculation, some sourceproximal parts of the Tiltill rockslide were excluded from our calculation. This upper part of the Tiltill rockslide deposit is likely thin, so omitting it is unlikely to have a significant effect on the overall volume of the deposit. The planimetric surface area of the entire Tiltill rockslide deposit is as much as 2.3 × 105 m2 (Figure 2) and reaches up to 30 m thick (Figure 3).

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 395–407

399

400

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 395–407

Be/9 Be ratios normalized to a value of 2.851 × 10−12 for the KNSTD-Be-01-5-4 standard (Nishiizumi et al., 2007). This is equivalent to the 07KNSTD normalization used in the CRONUS calculator (Balco et al., 2008). Uncertainty is 1σ analytical uncertainty. b Exposure ages calculated using the CRONUS calculator (Balco et al., 2008) assuming a 10 Be half-life of 1.387 ± 0.012 Myr (Chmeleﬀ et al., 2010; Korschinek et al., 2010), attenuation length of 160 g cm−2 , erosion rate of 0.0006 cm/yr, and rock density of 2.70 g cm−3 . Ages are based on the LSDn time-dependent production rate scaling scheme (Lifton et al., 2014). Ages based on the St and Lm production models provided by the CRONUS calculator are ∼300–600 years younger. a10

993 854 998 953 986 882 521 218 505 462 478 437 13,344 13,049 13,555 13,154 13,578 12,161 0.9754 0.9758 0.9735 0.9726 0.9775 0.9783 –119.69022° –119.69005° –119.69087° –119.69025° –119.69085° –119.69147° 37.987933° 37.988100° 37.987683° 37.988417° 37.987800° 37.990000°

1985 1989 1984 1994 1986 1957

19.0984 14.081 28.1596 16.6927 27.4625 12.0038

2.220 ± 0.050 1.643 ± 0.022 3.508 ± 0.070 2.038 ± 0.033 3.455 ± 0.057 1.365 ± 0.025

197067 ± 4739 194461 ± 3035 202711 ± 4349 198733 ± 3615 204141 ± 3713 183797 ± 3634

923 830 951 448 415 498 12,770 11,447 12,820 195464 ± 3567 182112 ± 3741 198053 ± 4735 0.9751 0.9751 0.9747 1.770 ± 0.029 2.641 ± 0.050 2.947 ± 0.066 16.0435 23.7481 25.6313 2003 2003 2009 –119.69165° –119.69165° –119.69218° 37.990233° 37.990233° 37.990300°

Shielding Longitude Latititude Sample

North lobe KRF-TIL-N01A KRF-TIL-N01B KRF-TIL-N03 South lobe KRF-TIL-S01 KRF-TIL-S02 KRF-TIL-S03 KRF-TIL-S04 KRF-TIL-S05 KRF-TIL-S06

Internal Error (yr) Be (atom/g qtz)

Age (yr)

LSDn Scalingb

Be/9 Bea (×10–13 ) 10

Quartz Mass (g)

Wieczorek (2002) estimated that the Tiltill rockslide was relatively old, based on soil development and forest encroachment onto the deposit, lichen growth, and “severe weathering of individual large rock blocks that can be pulled apart along once-intact joints.” However, Wieczorek (2002) did not speculate further on the age of the deposit. We employed cosmogenic 10 Be exposure dating to determine the age of the Tiltill rockslide, collecting nine samples from the distal portions of the slide deposit (Figures 3 and 5 and Table 1). We collected three samples from the north lobe of the Tiltill rockslide and six from the south lobe, a distribution that mirrored the relative areal extent of each lobe. To assess reproducibility, we collected two samples from the same boulder. We sampled large ﬂat boulders with minimal topographic shielding that typically stand slightly to substantially higher (up to 2 m) than the surrounding rockslide debris. Since the morphology of long runout landslides, such as Tiltill rockslide, tends to be much ﬂatter than that of ordinary rockfall talus, ordinary rockfalls occurring from the Tiltill rockslide source area after the main failure would come to rest on the proximal Tiltill rockslide surface, far from the distal edge. Therefore, we sampled the distal portion of the Tiltill rockslide to avoid sampling more recent rockfall clasts. After eliminating an outlier sample (see below) cosmogenic exposure dating yielded an error-weighted mean age of 12.8 ± 0.9 ka and 13.1 ± 0.8 ka (1σ external uncertainties) for the north and south lobe, respectively (Table 2). To obtain this result we used

Elevation (m, WGS84)

AGE OF THE TILTILL ROCKSLIDE

Table 1. Cosmogenic beryllium-10 data for the Tiltill rockslide.

Our approach yielded a volume of 2.1 × 106 m3 for the Tilltill rockslide deposit. We estimated the error of this value as follows. The minimum possible volume is the planimetric surface area multiplied by 2 m, the average depth that one can see between clasts into the slide mass, yielding 4.6 × 105 m3 . The maximum possible volume can be estimated by assuming that the slide ﬁlled a knickpoint concavity in Tiltill Valley, increasing the volume by ∼80 percent, and yielding 3.8 × 106 m3 . Though these margins of error are not standard deviations, the uncertainty can be stated most simply as 2.1 × 106 ± 1.6 × 106 m3 . Assuming porosity of 30 percent (Sass and Wollny, 2001; Moore et al., 2009; and Zimmer et al., 2012), the original volume of intact rock shed from Mount Gibson to produce this deposit was approximately 1.5 × 106 m3 . This derived from a 4.7 × 104 m2 source area (Figure 2), suggesting failure of a rock slab averaging 32 m thick, generally consistent with the headscarp scar geometry and volume.

External Error (yr)

Pluhar, Ford, Stock, Stone, and Zimmerman

Magnitude and Timing of the Tiltill Rockslide Table 2. Exposure ages for the Tiltill rockslide.

Accepted landform ages North lobe excluding outlier South lobe all samples North + south lobes excluding outlier Rejected landform ages North lobe all samples North + south lobes all samples

Mean Age (yr)

Standard Deviation (yr)

χ2 p-value

Error-Weighted Mean Age

Internal Error (yr)

External Error (yr)

12,795 13,140 13,054

35 524 471

0.9407 0.2447 0.3983

12,792 13,077 13,028

333 153 139

874 842 836

12,346 12,875

779 693

0.0409 0.0091

— —

Lifton-Sato-Dunai (LSDn) scaling (Balco et al., 2008; Phillips et al., 2016) and assumed a boulder erosion rate of 0.0006 cm/yr, derived from 10 Be measurements of pre-LGM boulders in Yosemite (Wahrhaftig et al., 2019). Slower erosion rates would decrease the expo-

Figure 5. (a) View of the south lobe of the Tiltill rockslide, looking southeast, showing the bouldery texture and hummocky topography of the debris field. (b) Characteristic cosmogenic 10 Be sample (KRF-TIL-N01A) taken from the top of a flat boulder projecting above the mean debris field surface.

sure ages. Topographic shielding at each sampling site was minimal; the most significant shielding factor was 0.973 (i.e., a maximum of 2.7 percent of cosmic rays shielded by obstacles), making the effect of shielding corrections smaller than error due to analytical uncertainties on 10 Be concentrations and production rate. Due to uncertainty in the magnitude of snow shielding at the study site, we neglected its effect in our calculation, although we acknowledge that snow shielding has been shown to reduce cosmogenic nuclide production by as much as 14 percent in mid-latitude mountainous regions (Schildgen et al., 2005). However, because we sampled from boulders standing higher than surrounding slide debris, snowpack on these surfaces would be less than on flat ground at this site, and thus the effect would probably be less than the stated maximum of Schildgen et al. (2005). Taking snow shielding into account would somewhat increase the age of the Tiltill rockslide. The nine exposure ages show very close agreement within error, based on the external uncertainty, except for one outlying sample: KRF-TIL-N01B at 11.4 ± 0.8 ka (Figure 6 and Table 1). This, the youngest sample, was collected from the same boulder as KRFTIL-N01A, yet there is a ∼9 percent age difference (Table 1). Sample KRF-TIL-N01A groups well with the remaining sample set. This substantial difference between samples KRF-TIL-N01A and KRF-TILN01B suggests that the latter experienced some process that reduced its inventory of 10 Be, such as localized spallation of the exposed boulder surface. With KRF-TIL-N01B included, the entire sample population exhibits a chi-squared p-value of 0.009, and the north lobe, from which it derives, yields 0.04. On the other hand, when KRF-TIL-N01B is excluded, the chi-squared p-value for the entire rockslide rises to 0.40, and the north lobe value rises to 0.94. Without measured erosion rates on the deposit, we cannot independently test localized spallation (i.e., erosion differences) as the cause for the younger apparent age of KRF-TIL-N01B, but the weight of evidence supports pruning it from the sample population.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 395–407

401

Pluhar, Ford, Stock, Stone, and Zimmerman

Figure 6. Individual (gray lines) and cumulative (dotted and solid black lines) probability density distributions for cosmogenic 10 Be exposure ages from boulders within the Tiltill rockslide. The youngest sample is an outlier and the solid black cumulative probability excludes that anomalous value. These data indicate an exposure age for the rockslide deposit of 13.0 ± 0.8 kyr, using LSDn scaling, an erosion rate of 0.0006 cm/yr, and neglecting snow shielding. The Tiltill rockslide exposure age overlaps with a rupture on the Inyo Mountain Fault (Bacon et al., 2005), a probable rupture on Truckee Fault (Melody et al., 2012), and a turbidite deposit in Lake Tahoe thought to be seismically triggered (Smith et al., 2013).

The remaining 10 Be results suggest a single event that created both the north and south lobes. With KRF-TIL-N01B excluded, the north lobe exposure age, 12.8 ± 0.9 years ago (two samples), and south lobe age, 13.1 ± 0.8 (six samples), are statistically indistinguishable. With other criteria also arguing for the Tiltill rockslide being a single landform, the eight retained samples yield a combined error-weighted mean age of 13.0 ± 0.8 ka (Figure 6 and Table 2). Alone, the dating does not eliminate the possibility that the two lobes of the Tiltill rockslide occurred separately, within a few hundred years of each other, although 402

the rockslide surface morphology suggests that this is unlikely. POTENTIAL TRIGGERING MECHANISMS A key aspect of understanding the hazard posed by large bedrock slope failures is identifying the triggering mechanisms responsible for their failure. However, identifying triggers for prehistoric events is inherently speculative, since detailed information regarding the environmental conditions present at the exact moment of failure is always lacking. In general, most studies

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 395–407

Magnitude and Timing of the Tiltill Rockslide

explore temporal coincidences between the slope failure and environmental conditions, such as debuttressing eﬀects associated with deglaciation (e.g., Cossart et al., 2008; Shroder Jr., et al., 2011), identiﬁable climate shifts (e.g., Hormes et al., 2008; Dortch et al., 2009; and Ivy-Ochs et al., 2009), or paleoseismicity (e.g., Mitchell et al., 2007; Hermanns and Schellenberger, 2008; Stock and Uhrhammer, 2010; Hermanns and Niedermann, 2011; Hewitt et al., 2011; Barth, 2014; and Nagelisen et al., 2015), and that is the approach that we take here. In some documented cases (e.g., Cossart et al., 2008), glacial retreat altered the stress state of bedrock valley walls, triggering slope failures. Glacial erosion certainly steepened the walls of Tiltill Valley and may have contributed to weakening of the slope during glacial retreat. However, this region was last glaciated at least 16 kyr ago (Wahrhaftig et al., 2019), well before the Tiltill rockslide occurred. The Tiltill rockslide did occur around the time of the Sierran Recess Peak glacial advance (14–12 ka), coincident with the Younger Dryas event, but this advance only expanded glaciers in the very highest cirques of the Sierra Nevada crest, with a drop in equilibrium line altitudes of only 100–200 m to ∼3,500 m elevation (Phillips, 2017). This would not have led to glacial advance into the vicinity of the Tiltill rockslide at 1,900–2,500 m elevation. Consequently, although glacial debuttressing or other (peri)glacial processes could have contributed to slope instability (Ballantyne et al., 2014), they do not appear to have acted as an immediate trigger. The Sierra Nevada lies in a seismically active region (Figure 1), and large-magnitude earthquakes are known slope failure triggers here (Harp et al., 1984; Keefer, 1994; Bull et al., 1994; Wieczorek and Jäger, 1996; and Stock et al., 2013), as they are elsewhere (e.g., Keefer 1984a,b). M 7+ earthquakes occur both on the nearby Sierra Nevada Frontal and Walker Lane Belt faults (e.g., Hough and Hutton, 2008), as well as on more distant faults of the Central Nevada Seismic Belt (e.g., Bell et al., 2004) and San Andreas Fault system (e.g., Song et al., 2008). Major earthquakes can trigger mass wasting 200+ km away (Keefer, 1984a). The seismic triggering threshold for rockslides is distance dependent but as low as ML 4 (Richter local magnitude; Keefer, 1984a), so the seismicity environment around Yosemite constitutes adequate potential for triggering. Moreover, the Sierra Nevada exhibits all of the needed ingredients for seismic triggering of large rock slope failures (Keefer, 1984b). These include slopes steeper than 25° and higher than 150 m, slopes undercut by active or geologically recent erosion, slopes composed of intensely fractured rocks (e.g., exfoliation-jointed), and other

geological instability indicators such as daylighted planes of weakness (joints) and evidence of prior mass wasting. Activity along the Sierra Nevada Frontal Fault system and the larger Walker Lane Belt probably dominate seismic shaking in the study area (e.g., Bormann et al., 2016) due to their proximity (Figure 1), although the San Andreas Fault system is also demonstrably important to mass wasting in some parts of the Sierra (Bull et al., 1994). The Walker Lane extends northward from the Garlock Fault in the southernmost Sierra, encompassing a zone of right-lateral faults and leftstepping normal faults (Figure 1). The Owens Valley Fault, part of the Walker Lane, produced one of the largest earthquakes in California’s recorded history, rupturing in a magnitude ∼7.5 event in 1872 (Beanland and Clark, 1994; Lee et al., 2001) and triggering many rockfalls in Yosemite (Stock et al., 2013). On the other hand, the Yosemite database lacks triggered rockfall occurrences for the 1906 San Francisco earthquake (Stock et al., 2013), indicating either an absence of significant associated slope failures or, less likely, poor record keeping at the time. Similarly, the 2019 Mw 7.1 Ridgecrest earthquake generated rockfall within 80 km of the epicenter (Brandenberg at al., 2019), whereas the 2020 Mw 5.8 Lone Pine earthquake produced rockfall up to 20 km away (Hauksson et al., 2021), but neither triggered mass wasting in Yosemite. This demonstrates that seismic triggering of rock slope failures is very sensitive to epicenter distance, earthquake magnitude (Keefer, 1984a), and perhaps directivity. Coincidence between the timing of rock slope failures and paleoseismic events can be suggestive of seismic triggering. For example, the coincidence of ages between the El Capitan rock avalanche and the late Holocene rupture of the White Mountain Fault and/or the Owens Valley Fault between 3.3 and 3.8 ka (Stock and Uhrhammer, 2010) suggests causality. Trenching across the southern Inyo Mountain Fault, 210–235 km from the Tiltill rockslide, revealed a rupture between 13.3 and 10.6 ka, overlapping in time with the Tiltill rockslide (Bacon et al., 2005). Dextral offset on the latest earthquake on this fault was 2.3 ± 0.8 m, big enough to produce an M 7.2 (based on Wells and Coppersmith, 1994). Elsewhere, Melody et al. (2012) infer that an ∼30 cm fault scarp formed >13,100 years ago on the Truckee Fault. This could produce an M 6.5 earthquake (based on Wells and Coppersmith, 1994), lying ∼160 km from Tiltill rockslide. Finally, Sangani (2017) outlines evidence for paleoearthquake(s) on the northwest side of Mono Lake, about 51 km from Tiltill rockslide, between 8 and 14 kyr ago, but does not provide any explicit measures of per-event offset from which to infer

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 395–407

403

Pluhar, Ford, Stock, Stone, and Zimmerman

paleoearthquake size. It is also well documented that the San Andreas Fault, 200+ km from the site (Figure 1), experiences large earthquakes with recurrence between 78 and 450 years, though no event of Tilltill rockslide age is documented (Sieh, 1978; Sieh and Jahns, 1984; Niemi and Hall, 1992; and Grant and Sieh, 1994). Keefer (1984a) showed that ML 7.9 earthquakes can cause rockslides up to ∼300 km from the rupture zone, decreasing to ∼200 km for ML 7.2, ∼140 for ML 6.5, ∼90 km for ML 6.0, and 15 km for ML 5.0. Therefore, the Inyo Mountain and Truckee Fault earthquakes may be too distant for their magnitudes to be triggers, although we cannot rule out those possibilities. Given the forgoing, beyond about 90 km from Tiltill rockslide, a candidate triggering earthquake would probably have produced a surface rupture and thus a paleoseismic record. However, paleoseismic records for the region are sparse, especially for events as old as the Tiltill rockslide. Numerous proximal faults remain unstudied or with no data for the time period in question: Robinson Creek, Silver Lake, Hilton Creek, Round Valley, Southern Sierra Nevada Frontal, Kern Canyon, Hartley Springs, numerous faults of Long Valley Caldera and Mono Lake, Wheeler Crest, Owens Gorge, Long Valley, Fish Slough, Independence, and Foothills Fault systems. Despite this, an important paleoseismic time series is available for the region. Sediments of Lake Tahoe, about 100 km north of the study site, preserve turbidites thought to be triggered by earthquakes, many of them without a known correlated fault paleorupture (Smith et al., 2013). Lake Tahoe turbidite “deposit O” (2σ model age 12.5– 11.2 ka) overlaps the age of the Tiltill rockslide, though event O is only weakly linked to seismic triggering (Smith et al., 2013). Still, existing historical and paleoseismic evidence is consistent with the possibility of seismic triggering of the Tiltill rockslide, but no definitive candidate paleoearthquakes exist. The dozens of understudied proximal faults of the eastern Sierra Nevada remain as possible Tiltill rockslide triggers, provided that they are capable of generating earthquakes of sufficient size for their distance from the site, as defined by Keefer (1984a) and outlined above. In summary, we cannot identify a trigger for the Tiltill rockslide with certainty. Without highresolution paleoclimate records for the study area, weather or climate-related conditions (e.g., a prolonged period of precipitation, or even one unusually powerful storm) cannot be tested and remain potential triggers. Given the volume of the Tiltill rockslide, the history of seismic triggering of rockfalls in Yosemite, and the lack of available paleoseismic data, we consider seismicity to be a plausible—but as yet unverifiable—trigger for the Tiltill rockslide.

404

CONCLUSIONS Yosemite National Park hosts an outstanding record of historical and pre-historic rock slope failures generated from a glacially oversteepened bedrock landscape. Rockfall, rockslides, and rock avalanches are the dominant form of mass wasting in this region, making it an ideal natural laboratory for studying these processes with few confounding effects. The present study quantifies characteristics of one large ancient slope failure, the Tiltill rockslide. Although the remote setting of the Tiltill rockslide dictates low risk, it is similar in timing and magnitude to slope failures in Yosemite Valley, a popular tourist destination where future slope failures of this magnitude would present very high risk. Our investigation of the Tiltill rockslide, in concert with multiple other recent and ongoing studies (e.g., Stock and Uhrhammer, 2010; Stock et al., 2014; and Pacheco et al., 2020) informs hazard and risk in nearby settings by populating the database of prehistoric rock slope failure age and volume and, in some cases, constraining potential trigger mechanisms. Cosmogenic 10 Be exposure dating reveals that the Tiltill rockslide occurred at 13.0 ± 0.8 ka, likely as one discrete event. GIS analysis demonstrates that an estimated 1.5 × 106 m3 of intact rock slid along slope-parallel fractures on the east face of Mount Gibson and traveled into and across Tiltill Valley, leaving the inferred 2.1 × 106 m3 ± 1.6 × 106 m3 of bouldery debris now blocking Tiltill Creek. The Tiltill rockslide exhibits a vertical drop to horizontal runout distance ratio, i.e., the effective coefficient of friction, of 0.47, and lies at the upper end of the volume range for ordinary slides that obey Coulomb failure mechanisms and the lower end of long-runout landslides. The age of the Tiltill rockslide, postdating glacial retreat by several thousand years, suggests that loss of glacier buttressing was not an immediate trigger. Although the relatively large volume of the Tiltill rockslide suggests a seismic trigger, and proximal faults do support the potential for earthquakes of sufficient magnitude to trigger the rockslide, we cannot exclude weather or climate-related triggering mechanisms. Quantification of the key metrics of the Tiltill rockslide expands the database of known bedrock slope failures in Yosemite, improving our overall understanding of the hazards posed by these events. ACKNOWLEDGMENTS This paper is dedicated to the memory of Jerry DeGraff, a valued colleague and mentor who helped provide guidance on Ford’s thesis research. Early versions of the manuscript benefited from review by John Wakabayashi, while the final manuscript greatly ben-

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 395–407

Magnitude and Timing of the Tiltill Rockslide

eﬁted from scrutiny by three anonymous reviewers. Ford’s work was supported by the Fresno Gem and Mineral Society, Geological Society of America Graduate Student Research Grant, Association of Environmental and Engineering Geologists (AEG) Martin L. Stout Scholarship, the AEG Sacramento Chapter Graduate Scholarship, and CSU Fresno’s Faculty Sponsored Student Research Award and Ken Schmidt Hydrogeology Scholarship. The authors are indebted to Shelby Jones and Yvan Mendoza for sample collection assistance. This is LLNL-JRNL-820168. REFERENCES Airborne Snow Observatories, Inc., 2021, available at https://www.airbornesnowobservatories.com Bacon, S. N.; Jayko, A. S.; and McGeehin, J. P., 2005, Holocene and latest Pleistocene oblique dextral faulting on the southern Inyo Mountains fault, Owens Lake Basin, California: Bulletin Seismological Society America, Vol. 95, No. 6, pp. 2472–2485. Balco, G.; Stone, J. O.; Lifton, N. A.; and Dunai, T. J., 2008, A complete and easily accessible means of calculating surface exposure ages or erosion rates from 10 Be and 26 Al measurements: Quaternary Geochronology, Vol. 3, No. 3, pp. 174–195. Ballantyne, C. K.; Wilson, P.; Gheorghiu, D.; and Rodés, À., 2014, Enhanced rock-slope failure following ice-sheet deglaciation: Timing and causes: Earth Surface Processes Landforms, Vol. 39, No. 7, pp. 900–913. Barth, N., 2014, The Cascade rock avalanche: Implications of a very large Alpine Fault-triggered failure, New Zealand: Landslides, Vol. 11, No. 3, pp. 327–341. Bateman, P. C.; Wahrhaftig, C.; and Bailey, E. H., 1966, Geology of the Sierra Nevada: Geology of northern California: California Division Mines Geology Bulletin, Vol. 190, pp. 107–172. Beanland, S. and Clark, M. M., 1994, The Owens Valley fault zone, eastern California, and surface faulting associated with the 1872 earthquake: U.S. Geological Survey Bulletin, Vol. 1982, 29 p. Bell, J. W.; Caskey, S. J.; Ramelli, A. R.; and Guerrieri, L., 2004, Pattern and rates of faulting in the central Nevada seismic belt, and paleoseismic evidence for prior beltlike behavior: Bulletin Seismological Society America, Vol. 94, No. 4, pp. 1229–1254. Blackwelder, E., 1931, Pleistocene glaciation in the Sierra Nevada and Basin ranges: Bulletin Geological Society America, Vol. 42, No. 4, pp. 865–922. Bormann, J. M.; Hammond, W. C.; Kreemer, C.; and Blewitt, G., 2016, Accommodation of missing shear strain in the Central Walker Lane, western North America: Constraints from dense GPS measurements: Earth Planetary Science Letters, Vol. 440, pp. 169–177. Brandenberg, S. J.; Wang, P.; Nweke, C. C.; Hudson, K.; Mazzoni, S.; Bozorgnia, Y.; Hudnut, K. W.; Davis, C. A.; Ahdi, S. K.; and Zareian, F., 2019, Preliminary report on engineering and geological eﬀects of the July 2019 Ridgecrest earthquake sequence: Geotechnical Extreme Event Reconnaissance Association, 69 p. Bull, W. B.; King, J.; Kong, F.; Moutoux, T.; and Phillips, W.M., 1994, Lichen dating of coseismic landslide hazards in alpine mountains: Geomorphology,Vol. 10, No. 1-4, pp. 253–264.

Cassel, E. J.; Graham, S. A.; and Chamberlain, C. P., 2009, Cenozoic tectonic and topographic evolution of the northern Sierra Nevada, California, through stable isotope paleoaltimetry in volcanic glass: Geology, Vol. 37, No. 6, pp. 547–550. Chmeleff, J.; von Blanckenburg, F.; Kossert, K.; and Jakob, D., 2010, Determination of the 10 Be half-life by multicollector ICP-MS and liquid scintillation counting: Nuclear Instruments Methods Physics Research Section B: Beam Interactions Materials Atoms, Vol. 268, No. 2, pp. 192–199. Clark, D.; Gillespie, A.; Clark, M.; Burke, R.; and Easterbrook, D., 2003, Mountain glaciations of the Sierra Nevada. In Easterbrook, D. (Editor), Quaternary Geology of the United States: INQUA 2003 Field Guide Volume, pp. 287–311. Clark, D. H. and Gillespie, A. R., 1997, Timing and significance of late-glacial and Holocene cirque glaciation in the Sierra Nevada, California: Quaternary International, Vol. 38, pp. 21–38. Collins, B. D. and Stock, G. M., 2016, Rockfall triggering by cyclic thermal stressing of exfoliation fractures: Nature Geoscience, Vol. 9, No. 5, pp. 395–400. Collins, G. S. and Melosh, H. J., 2003, Acoustic fluidization and the extraordinary mobility of sturzstroms: Journal Geophysical Research: Solid Earth, Vol. 108, No. B10, pp. EPM 4-1–4-14. Cordes, S. E.; Stock, G. M.; Schwab, B. E.; and Glazner, A. F., 2013, Supporting evidence for a 9.6±1 ka rock fall originating from Glacier Point in Yosemite Valley, California: Environmental Engineering Geoscience, Vol. 19, No. 4, pp. 345–361. Cossart, E.; Braucher, R.; Fort, M.; Bourlès, D.; and Carcaillet, J., 2008, Slope instability in relation to glacial debuttressing in alpine areas (Upper Durance catchment, southeastern France): Evidence from field data and 10 Be cosmic ray exposure ages: Geomorphology, Vol. 95, No. 1–2, pp. 3–26. Dade, B. W. and Huppert, H. E., 1998, Long-runout rockfalls: Geology, Vol. 26, No. 9, pp. 803–806. Dickinson, W. R., 2006, Geotectonic Evolution of the Great Basin: Geosphere, Vol. 2, No. 7, pp. 353–368. Dortch, J. M.; Owen, L. A.; Haneberg, W. C.; Caffee, M. W.; Dietsch, C.; and Kamp, U., 2009, Nature and timing of large landslides in the Himalaya and Transhimalaya of northern India: Quaternary Science Reviews, Vol. 28, No. 11–12, pp. 1037–1054. Dussauge, C.; Grasso, J. R.; and Helmstetter, A., 2003, Statistical analysis of rockfall volume distributions: Implications for rockfall dynamics: Journal Geophysical Research: Solid Earth, Vol. 108, No. B6, pp. ETG 2-1–2-11. Ford, K. R., 2014, Cosmogenic beryllium-10 nuclide exposure age dating of the Tiltill rock avalanche in Yosemite National Park: M.S. Thesis, California State University, Fresno, 117 p. Grant, L. B. and Sieh, K., 1994, Paleoseismic evidence of clustered earthquakes on the San Andreas fault in the Carrizo Plain, California: Journal Geophysical Research: Solid Earth, Vol. 99, No. B4, pp. 6819–6841. Guerin, A.; Stock, G. M.; Radue, M. J.; Jaboyedoff, M.; Collins, B. D.; Matasci, B.; Avdievitch, N.; and Derron, M.-H., 2020, Quantifying 40 years of rockfall activity in Yosemite Valley with historical structure-from-motion photogrammetry and terrestrial laser scanning: Geomorphology, Vol. 356, No. 107069, pp. 1–18. Guzzetti, F.; Reichenbach, P.; and Wieczorek, G., 2003, Rockfall hazard and risk assessment in the Yosemite Valley, California, USA: Natural Hazards Earth System Sciences, Vol. 3, No. 6, pp. 491–503. Harp, E.; Tanaka, K.; Sarmiento, J.; and Keefer, D., 1984, Landslides from the May 25–27, 1980, Mammoth Lakes, California,

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 395–407

405

Pluhar, Ford, Stock, Stone, and Zimmerman earthquake sequence: U. S. Geological Survey Miscellaneous Investigation Series Map I-1612. Hauksson, E.; Olson, B.; Grant, A.; Andrews, J. R.; Chung, A. I.; Hough, S. E.; Kanamori, H.; McBride, S. K.; Michael, A. J.; and Page, M., 2021, The normal-faulting 2020 M w 5.8 Lone Pine, Eastern California, earthquake sequence: Seismological Society America, Vol. 92, No. 2A, pp. 679–698. Hermanns, R.L. and Niedermann, S., 2011, Late Pleistocene– early Holocene paleoseismicity deduced from lake sediment deformation and coeval landsliding in the Calchaquíes valleys, NW Argentina. In Audemard, F.A.M. and Michetti, A.M. (Editors), Geological Criteria for Evaluating Seismicity Revisited: Forty Years of Paleoseismic Investigations and the Natural Record of Past Earthquakes. Geological Society of America, Boulder, Special Paper 479, pp.181–194. Hermanns, R. L. and Schellenberger, A., 2008, Quaternary tephrochronology helps deﬁne conditioning factors and triggering mechanisms of rock avalanches in NW Argentina: Quaternary International, Vol. 178, No. 1, pp. 261–275. Hewitt, K.; Gosse, J.; and Clague, J. J., 2011, Rock avalanches and the pace of late Quaternary development of river valleys in the Karakoram Himalaya: Bulletin, Vol. 123, No. 9–10, pp. 1836–1850. Hormes, A.; Ivy-Ochs, S.; Kubik, P. W.; Ferreli, L.; and Michetti, A. M., 2008, 10 Be exposure ages of a rock avalanche and a late glacial moraine in Alta Valtellina, Italian Alps: Quaternary International, Vol. 190, No. 1, pp. 136–145. Hough, S. E. and Hutton, K., 2008, Revisiting the 1872 Owens Valley, California, earthquake: Bulletin Seismological Society America, Vol. 98, No. 2, pp. 931–949. Huber, N. K., 1990, The late Cenozoic evolution of the Tuolumne River, central Sierra Nevada, California: Geological Society America Bulletin, Vol. 102, No. 1, pp. 102–115. Huber, N. K.; Bateman, P. C.; and Wahrhaftig, C., 1989, Geologic Map of Yosemite National Park and Vicinity, California: U.S. Geological Survey Map I-1874. Ivy-Ochs, S.; Poschinger, A.; Synal, H.-A.; and Maisch, M., 2009, Surface exposure dating of the Flims landslide, Graubünden, Switzerland: Geomorphology, Vol. 103, No. 1, pp. 104–112. Jibson, R. W., 2007, Regression models for estimating coseismic landslide displacement: Engineering Geology, Vol. 91, No. 2– 4, pp. 209–218. Keefer, D. K., 1984a, Landslides caused by earthquakes: Geological Society America Bulletin, Vol. 95, No. 4, pp. 406–421. Keefer, D. K., 1984b, Rock avalanches caused by earthquakes: Source characteristics: Science, Vol. 223, No. 4642, pp. 1288– 1290. Keefer, D. K., 1994, The importance of earthquake-induced landslides to long-term slope erosion and slope-failure hazards in seismically active regions: Geomorphology, Vol. 10, No. 1-4, pp. 265–284. Kistler, R. W., 1973, Geologic Map of the Hetch Hetchy Reservoir Quadrangle, Yosemite National Park, California: U.S. Geological Survey Map GQ-1112. Korschinek, G.; Bergmaier, A.; Faestermann, T.; Gerstmann, U.; Knie, K.; Rugel, G.; Wallner, A.; Dillmann, I.; Dollinger, G.; and Von Gostomski, C.L., 2010, A new value for the half-life of 10 Be by heavy-ion elastic recoil detection and liquid scintillation counting: Nuclear Instruments Methods Physics Research Section B: Beam Interactions Materials Atoms, Vol. 268, No. 2, pp. 187–191. Lee, J.; Spencer, J.; and Owen, L., 2001, Holocene slip rates along the Owens Valley fault, California: Implications for the recent

406

evolution of the Eastern California Shear Zone: Geology, Vol. 29, No. 9, pp. 819–822. Lifton, N.; Sato, T.; and Dunai, T. J., 2014, Scaling in situ cosmogenic nuclide production rates using analytical approximations to atmospheric cosmic-ray fluxes: Earth Planetary Science Letters, Vol. 386, pp. 149–160. Martel, S. J.; Stock, G. M.; and Ito, G., 2014, Mechanics of relative and absolute displacements across normal faults, and implications for uplift and subsidence along the eastern escarpment of the Sierra Nevada, California: Geosphere, Vol. 10, No. 2, pp. 243–263. Matthes, F. E., 1930, Geologic History of the Yosemite Valley, U.S. Geological Survey Professional Paper 160, 137 p. McJunkin, R. D. and Bedrossian, G. T. L., 1980, Mammoth Lakes earthquakes, May 25–27, 1980 Mono County, California: California Geology, Vol. 33, No. 9, pp. 194–201. Melody, A. D.; Whitney, B. B.; and Slack, C. G., 2012, Late Pleistocene and Holocene Faulting in the Western Truckee Basin North of Truckee, California: Bulletin Seismological Society America, Vol. 102, No. 5, pp. 2219–2224. Mitchell, W.; McSaveney, M.; Zondervan, A.; Kim, K.; Dunning, S.; and Taylor, P., 2007, The Keylong Serai rock avalanche, NW Indian Himalaya: Geomorphology and palaeoseismic implications: Landslides, Vol. 4, No. 3, pp. 245–254. Moore, J. R.; Sanders, J. W.; Dietrich, W. E.; and Glaser, S. D., 2009, Influence of rock mass strength on the erosion rate of alpine cliffs: Earth Surface Processes Landforms, Vol. 34, No. 10, pp. 1339–1352. Nagelisen, J.; Moore, J. R.; Vockenhuber, C.; and Ivy-Ochs, S., 2015, Post-glacial rock avalanches in the Obersee valley, Glarner Alps, Switzerland: Geomorphology, Vol. 238, pp. 94–111. Niemi, T. M. and Hall, N. T., 1992, Late Holocene slip rate and recurrence of great earthquakes on the San Andreas fault in northern California: Geology, Vol. 20, No. 3, pp. 195–198. Nishiizumi, K.; Imamura, M.; Caffee, M. W.; Southon, J. R.; Finkel, R. C.; and McAninch, J., 2007, Absolute calibration of 10 Be AMS standards: Nuclear Instruments Methods Physics Research Section B: Beam Interactions Materials Atoms, Vol. 258, No. 2, pp. 403–413. Pacheco, M.; Plattner, A. M.; Stock, G. M.; Rood, D. H.; and Pluhar, C. J., 2020, Surface exposure dating and geophysical tomography of the Royal Arches Meadow rock avalanche, Yosemite Valley, California: Rock Avalanches. Frontiers in Earth Science, Vol. 8, No. 372, pp. 1-12. Phillips, F., 2017, Glacial chronology of the Sierra Nevada, California, from the last glacial maximum to the Holocene: Cuadernos de investigación geográfica/Geographical Research Letters, Vol. 43, No. 2, pp. 527–552. Phillips, F. M., 2016, Cosmogenic nuclide data sets from the Sierra Nevada, California, for assessment of nuclide production models: I. Late Pleistocene glacial chronology: Quaternary Geochronology, Vol. 35, pp. 119–129. Phillips, F. M.; Argento, D. C.; Balco, G.; Caffee, M. W.; Clem, J.; Dunai, T. J.; Finkel, R.; Goehring, B.; Gosse, J. C.; and Hudson, A. M., 2016, The CRONUS-Earth project: A synthesis: Quaternary Geochronology, Vol. 31, pp. 119–154. Sangani, R. C., 2017, Structure of the Mina Deflection in Mono Lake, CA: Inferences from Paleoseismology, M.S. Thesis, State University of New York at Buffalo, 64 p. Sass, O. and Wollny, K., 2001, Investigations regarding Alpine talus slopes using ground-penetrating radar (GPR) in the Bavarian Alps, Germany: Earth Surface Processes Landforms:

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 395–407

Magnitude and Timing of the Tiltill Rockslide Journal British Geomorphological Research Group, Vol. 26, No. 10, pp. 1071–1086. Schildgen, T.; Phillips, W.; and Purves, R., 2005, Simulation of snow shielding corrections for cosmogenic nuclide surface exposure studies: Geomorphology, Vol. 64, No. 1–2, pp. 67–85. Shroder Jr., J. F.; Owen, L. A.; Seong, Y. B.; Bishop, M. P.; Bush, A.; Caffee, M. W.; Copland, L.; Finkel, R. C.; and Kamp, U., 2011, The role of mass movements on landscape evolution in the Central Karakoram: Discussion and speculation: Quaternary International, Vol. 236, No. 1–2, pp. 34–47. Sieh, K. E., 1978, Prehistoric large earthquakes produced by slip on the San Andreas Fault at Pallett Creek, California: Journal Geophysical Research: Solid Earth, Vol. 83, No. B8, pp. 3907– 3939. Sieh, K. E. and Jahns, R. H., 1984, Holocene activity of the San Andreas fault at Wallace creek, California: Geological Society America Bulletin, Vol. 95, No. 8, pp. 883–896. Smith, S. B.; Karlin, R. E.; Kent, G. M.; Seitz, G. G.; and Driscoll, N. W., 2013, Holocene subaqueous paleoseismology of Lake Tahoe: Bulletin, Vol. 125, No. 5–6, pp. 691–708. Song, S. G.; Beroza, G. C.; and Segall, P., 2008, A uniﬁed source model for the 1906 San Francisco earthquake: Bulletin Seismological Society America, Vol. 98, No. 2, pp. 823–831. Stock, G. M.; Anderson, R. S.; and Finkel, R. C., 2005, Rates of erosion and topographic evolution of the Sierra Nevada, California, inferred from cosmogenic 26 Al and 10 Be concentrations: Earth Surface Processes Landforms: Journal British Geomorphological Research Group, Vol. 30, No. 8, pp. 985–1006. Stock, G. M.; Collins, B. D.; Santaniello, D. J.; Zimmer, V. L.; Wieczorek, G. F.; and Snyder, J. B., 2013, Historical Rock Falls in Yosemite National Park, California (1857–2011): U.S. Geological Survey Data Series 746, 17 p. Stock, G. M.; Luco, N.; Collins, B. D.; Harp, E. L.; Reichenbach, P.; and Frankel, K. L., 2014, Quantitative rock-fall hazard and risk assessment for Yosemite Valley, Yosemite National Park, California: U.S. Geological Survey Scientiﬁc Investigations Report, Vol. 5129, No. 2014, 52 p. Stock, G. M.; Martel, S. J.; Collins, B. D.; and Harp, E. L., 2012, Progressive failure of sheeted rock slopes: the 2009– 2010 Rhombus Wall rock falls in Yosemite Valley, California, USA: Earth Surface Processes Landforms, Vol. 37, No. 5, pp. 546–561.

Stock, G. M. and Uhrhammer, R. A., 2010, Catastrophic rock avalanche 3600 years BP from El Capitan, Yosemite Valley, California: Earth Surface Processes Landforms, Vol. 35, No. 8, pp. 941–951. Wahrhaftig, C.; Stock, G. M.; McCracken, R. G.; Sasnett, P.; and Cyr, A. J., 2019, Extent of the Last Glacial Maximum (Tioga) Glaciation in Yosemite National Park and Vicinity, California: U.S. Geological Survey Scientific Investigations Series Map, Vol. 3414. Wakabayashi, J., 2013, Paleochannels, stream incision, erosion, topographic evolution, and alternative explanations of paleoaltimetry, Sierra Nevada, California: Geosphere, Vol. 9, No. 2, pp. 191–215. Wells, D. L. and Coppersmith, K. J., 1994, New empirical relationships among magnitude, rupture length, rupture width, rupture area, and surface displacement: Bulletin Seismological Society of America, Vol. 84, No. 4, pp. 974–1002. Western Regional Climate Center, 2021, Hetch Hetchy, California (043939): available at https://wrcc.dri.edu/cgibin/cliMAIN.pl?cahetc+nca Wieczorek, G. F., 2002, Catastrophic rockfalls and rockslides in the Sierra Nevada, USA. In Evans, S. G., and DeGraff, J. V. (Editors), Catastrophic Landslides: Effects, Occurrence, and Mechanisms: Geological Society of America Reviews in Engineering Geology, Boulder, CO, pp. 165–190. Wieczorek, G. F. and Jäger, S., 1996, Triggering mechanisms and depositional rates of postglacial slope-movement processes in the Yosemite Valley, California: Geomorphology, Vol. 15, No. 1, pp. 17–31. Wieczorek, G. F.; Morrissey, M. M.; Iovine, G.; and Godt, J., 1999, Rock-fall potential in the Yosemite Valley, California: U.S. Geological Survey Open-File Report, Vol. 99, No. 578, p. 1. Wieczorek, G. F.; Snyder, J. B.; Waitt, R. B.; Morrissey, M. M.; Uhrhammer, R. A.; Harp, E. L.; Norris, R. D.; Bursik, M. I.; and Finewood, L. G., 2000, Unusual July 10, 1996, rock fall at Happy Isles, Yosemite National Park, California: Geological Society America Bulletin, Vol. 112, No. 1, pp. 75–85. Zimmer, V. L.; Collins, B. D.; Stock, G. M.; and Sitar, N., 2012, Rock fall dynamics and deposition: an integrated analysis of the 2009 Ahwiyah Point rock fall, Yosemite National Park, USA: Earth Surface Processes Landforms, Vol. 37, No. 6, pp. 680–691.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 395–407

407

Sixty Years of Post-Fire Assessment and Monitoring on Non-Federal Lands in California: What Have We Learned? PETER H. CAFFERATA* California Department of Forestry and Fire Protection, P.O. Box 944246, Sacramento, CA 94244-2460

DREW B. R. COE California Department of Forestry and Fire Protection, 6105 Airport Road, Redding, CA 96002

WILLIAM R. SHORT California Department of Conservation - California Geological Survey, 801 K Street, Sacramento, CA 95820

Key Terms: Post-Fire Assessment, Flooding, Debris Flows, Wildfire, Watershed Hazards, Post-Fire Rehabilitation, Emergency Response, Public Safety ABSTRACT California has a long history of damage from postfire flooding and debris flows, often described as the fire-flood sequence. Post-wildfire assessment has been conducted on California’s non-federal lands for more than 60 years. From 1956 to 1999, the focus was on aerial grass seeding for emergency revegetation. Later, hillslope treatments including straw mulch and hydromulch were shown to be more effective in reducing erosion, but their high cost make them generally infeasible. In the 1990s the state moved away from aerial grass seeding and embraced a dual strategy for minimizing post-fire impacts. First, fire suppression repair work became consistently applied to all fires. Typical repairs include installing waterbreaks on firelines, grading roads, removing soil from crossings, and mulching mechanically disturbed areas near streams. Second, there was greater use of interdisciplinary teams to evaluate lifesafety and property threats from debris flows, flooding, and rockfall on non-federal lands. The state process was pioneered by the California Geological Survey (CGS) in 1993 and has benefited from advances in spatially explicit modeling since 2000. Currently, state teams are co-led by the California Department of Forestry and Fire Protection and CGS, with evaluations only conducted on fires with significant threats. This approach is effective for rapidly notifying emergency management

*Corresponding author email: Pete.Caﬀerata@ﬁre.ca.gov

agencies of hazards, helping to protect lives and property. This process will benefit from further technological advances, monitoring, and improved modeling. Strategies for improving post-fire hazard evaluation and better emergency preparedness through pre-fire hazard mitigation planning are provided.

INTRODUCTION California has a long history of human casualties and property damage resulting from increased watershed hazards following wildfire (i.e., the “fireflood sequence”), particularly in the southern and central parts of the state. The fire-flood sequence became a recognized phenomenon after the deadly La Crescenta-Montrose New Year’s Day Flood of 1934 (Troxell and Peterson, 1937), with periodic deadly and damaging flood/flow events following fires in the ensuing decades despite increases in the understanding of the post-fire environment (e.g., Cleveland, 1973, 1977; Spittler, 1995, 2005; Cannon et al., 2010; and Kean et al., 2011, 2019). Post-fire watershed hazards are a consequence of wildfire alterations of hydrologic and geomorphic processes at the hillslope and watershed scales. Highseverity wildfires consume ground cover, decrease surface roughness, and can produce a water repellent layer in the mineral soil, often referred to as a hydrophobic layer (DeBano, 1981; Neary et al., 2005). Intense fire also reduces soil structure and incinerates binding shallow roots, which results in the loss of mechanical support, the formation of raveling, and an increase in the supply of readily erodible soil. The changes in vegetation, litter, and soils lead to a much

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 409–422

409

Caﬀerata, Coe, and Short

lower capacity for the soil to absorb rainfall and a much greater potential for flooding, erosion, and debris flows. Values-at-risk (VARs), including residential developments and critical infrastructure located near stream channels in low lying areas and on depositional landforms (e.g., floodplains and alluvial fans), are at greatest risk from these post-fire hazards. In addition to life-safety and property/infrastructure threats, increased post-fire sediment yields adversely impact reservoir storage capacity, water quality, water treatment costs, power generation facilities, and aquatic species. Post-wildfire flooding, erosion, and debris flow generation are highly variable in both space and time. The magnitude of post-fire erosion depends on site conditions; the severity of the wildfire; and the intensity, size, and number of storm events during the first few winters after the fire. Site conditions such as slope, watershed morphology, tectonic regime, lithology, and soil type are key factors in predicting susceptibility to postfire erosion. Runoff and erosion rates are commonly increased by one or more orders of magnitude, especially for fires burned at high and/or moderate soil burn severity. Conversely, the potential for increasing peak flows and erosion rates is relatively small when areas are burned at low or very low soil burn severity (Robichaud et al., 2010b). To address potential watershed hazards following wildfire, the California Department of Forestry and Fire Protection (CAL FIRE, known as CDF until 2006), along with the California Geological Survey (CGS) after 1990, has conducted post-fire evaluations on private and state lands, implemented treatments, and initiated and/or funded monitoring and applied research on burned areas for more than 60 years. The evolution in how post-fire watershed hazards are mitigated is an example of the application of emerging science and technology in emergency response planning. This paper provides a brief historical perspective on post-fire assessment on non-federal lands in California, key technical and procedural lessons learned to date, and the current understanding regarding appropriate procedures to use when addressing post-fire watershed hazards. STATE POST-FIRE ASSESSMENT AND MONITORING FROM 1956 TO 1999: THE AERIAL GRASS SEEDING PERIOD The initial approach to reduce post-fire hazards in California was to aerially apply grass seed. This work began in the late 1930s using biplanes to sow mustard seed in southern California. By the 1950s annual ryegrass had replaced mustard seed as the main choice for post-fire erosion control (Griffith, 1998;

410

Wohlgemuth et al., 2009). The advantages of annual ryegrass included its rapid germination, dense fibrous root system, low cost, and limited perseverance on the landscape. The CDF Emergency Watershed Protection program began in 1956 and utilized aerial grass seeding for rapid revegetation on non-federal lands in California until the late 1990s. Seeding was often recommended after large fires with watersheds upstream of high-risk values (e.g., reservoirs, homes) that had been at least 40 percent burned. Annual CDF reports from 1957 to 1972 show that approximately 25 percent of the total state responsibility area (SRA), or 573,000 acres, was seeded out of nearly 2.4 million acres burned. The largest seeding project was for the 1970 Laguna Fire in San Diego County (Figure 1). By the mid-1970s the value of annual ryegrass seeding was questioned, but the practice continued into the 1980s. CDF monitored the establishment of ryegrass in post-fire emergency revegetation projects from 1957 to 1972 and reported poor results in southern California (Blanford and Gunter, 1972). Additionally, studies conducted by the U.S. Forest Service Pacific Southwest Research Station in the late 1980s and early 1990s in southern California showed that aerial grass seeding was unlikely to reduce post-fire erosion during the first winter, which is when most of the impacts occur (e.g., Barro and Conard, 1987; Conard et al., 1991; and Wohlgemuth et al., 1998). Aerial grass seeding was also shown to be detrimental to native species recovery and conifer establishment. Based on these study results, aerial grass seeding decreased considerably in the 1990s, and when used, native species or non-persistent non-native species such as cereal grains were generally selected (Griffith, 1998). By 2000 CAL FIRE had stopped aerial seeding, while the last aerial seeding by the U.S. Forest Service on federal lands in California took place in 2002 on the Biscuit Fire. Studies conducted in the early 2000s showed that straw/wood mulch and hydromulch are much more effective in preventing post-fire erosion because they provide cover prior to the first fall rains (Wagenbrenner et al., 2006; Foltz and Copeland, 2009; and Robichaud et al., 2010a). Unfortunately, the high cost of these treatments precludes their use over large percentages of burned areas. In addition, these methods are not effective on steep source areas that produce the most sediment entrained in debris flows and sedimentladen floods. Hillslope application of these treatments is only used on gentle to moderate slopes by federal agencies in California for very specific areas with high VARs downstream, by state and county public works departments along highways, and by private landowners near homes.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 409–422

Post-Fire Assessment and Monitoring: What Have We Learned?

Figure 1. Summary of number of CDF emergency revegetation projects and acreage seeded from 1956 through 1972 (data from CDF, 1972).

SHIFT FROM EMERGENCY REVEGETATION TO FIRE SUPPRESSION REPAIR AND LIFE-SAFETY AND PROPERTY THREAT EVALUATIONS Two types of state-led post-fire evaluations began to replace widespread aerial grass seeding in the 1990s. First, assessments to identify repairs needed to reduce potential damage resulting from fire suppression efforts became consistently applied to all fires. Typical repairs include installing drainage structures (i.e., waterbreaks or waterbars) on tractor and handconstructed firelines, grading and restoring drainage structures on roads, removing soil and debris pushed into watercourses during fireline construction, and seeding and mulching fireline approaches to watercourse crossings and other mechanically disturbed areas near streams. These field assessments are conducted by CAL FIRE foresters, and the work is implemented as part of the fire incident by CAL FIRE equipment operators, private contractors, and Conservation Camp inmate crews. Robichaud et al. (2000) describe road treatments such as installing drainage structures (e.g., rolling dips and waterbreaks) as an important component of post-fire rehabilitation work.

The second type of state-led evaluation was the increased use of post-fire interdisciplinary teams to assess life-safety, property, and infrastructure threats from debris flows, flooding, rockfall, and hillslope erosion. This process began in 1993 with geologists from CGS identifying the geologic and geomorphic factors that can lead to post-fire debris flows and using these factors to identify which fires posed significant lifesafety risks after a fire siege in southern California (Spittler, 1993, 1995; and Spittler et al., 1994). Technology for field work was limited in the 1990s, and field teams relied on helicopter and ground surveys to assess soil burn severity within the fire perimeter based on the degree of vegetation consumption, remaining ground cover, and soil hydrophobicity. Hand-drawn maps were developed to rate areas as high, moderate, or low soil burn severity (DeGraff, 2014). Significant technological advances after 2000 have greatly aided post-fire evaluations. The first state use of satellite-derived burned area reflectance classification (BARC) maps for preliminary soil burn severity ratings (Parsons et al., 2010) took place in 2003, when state agency teams conducted rapid post-fire assessments for watershed hazards during a large southern California fire siege. Field-verified BARC maps and satellite imagery to locate structures potentially at risk, along

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 409–422

411

Caﬀerata, Coe, and Short

with a new burn site evaluation form, were used to help locate and document high hazard areas and warn residents of debris flow and flooding hazards in San Bernardino County. Destructive debris flows occurred after the 2003 Old and Grand Prix fires (Cannon et al., 2011), with 16 fatalities occurring on both private and federal lands despite in-person and written warnings and facility evacuations and closures. These fire assessments made it evident that accurate site evaluation is only the first step in mitigating hazard and that warnings must be clearly communicated to the public and heeded to minimize the potential for injuries and fatalities. Additional advances were still needed as well, because U.S. Geological Survey (USGS) debris flow hazard reports and maps displaying modeling results, while valuable to local agencies and the public, were not available until after the state field team evaluations were completed (e.g., Cannon et al., 2003). USGSpublished post-fire debris flow hazard maps typically took 15 to 45 days to produce prior to 2014 (Staley, 2020). The scope of state-led post-fire evaluations changed considerably later in the 2000s. Numerous state multiagency post-fire assessment teams were assembled for a large fire siege in southern California in 2007 and for a statewide fire siege in 2008. These large-scale efforts produced detailed assessments addressing geology, hydrology, engineering, wildlife, and cultural resources, as well as spatially explicit Hazard Awareness Maps. These efforts were generally successful in identifying high hazard areas, and floods, sediment-laden flows, debris flows, and rockfall did occur in several of these fire areas. However, the size of the teams and the broad scope of the post-fire assessments meant that this process was slow to propose specific emergency protection measures. This protracted process undercut the rapid life-safety assessment needs driven by the short time frame between wildfires and fall rains in California. THE CURRENT POST-FIRE ASSESSMENT PROCESS From 2008 to 2014 there were only limited statesponsored post-fire evaluations and monitoring due to shifts in CAL FIRE priorities and the small number of large fires on SRA. However, in 2015 two large fires mostly on SRA catalyzed a new state postfire evaluation process that could take advantage of the technological advances that had occurred since 2008. State Post-Fire Watershed Emergency Response Teams (SPFWERT, later shortened to WERT in 2016) were able to rapidly evaluate hazards to life-safety and property risks, develop emergency protection measures for the identified sites, and communicate the findings in a 2-week period. These rapid assessments were fa-

412

cilitated by the availability of debris flow modeling results from the USGS Landslide Hazards Program in less than 24 hours. Debris flow probability, potential volume, and combined hazard are predicted for each basin outlet within the fire perimeter, as well as along the upstream drainage networks (Cannon et al., 2010; Gartner et al., 2014; and Staley et al., 2016). In 2016, the WERT process was formalized with defined procedures for evaluating post-fire debris flows, flooding, rockfall, and erosion risks on non-federal lands. Screening criteria are implemented so that postfire evaluations are conducted on fires with (1) significant threats to lives and property from post-fire debris flows, flooding, and rockfall, and (2) a significant percentage of non-federal lands (CAL FIRE and CGS, 2021). The WERT evaluation process includes the following steps: (1) gather existing hazard and remote sensing data, including LiDAR-based topographic information where it is available, to assist in field evaluations; (2) determine soil burn severity using fieldvalidated and adjusted BARC data; (3) model watershed hazards, including USGS debris flow modeling for predicting debris flow probability and volumetric magnitude (Staley et al., 2016), post-fire flood flow modeling, and surface erosion modeling; (4) conduct geomorphic interpretation by licensed professionals; (5) identify VARs by field checking locations identified as potentially at risk; (6) develop emergency protection measures for the identified VARs; (7) assist with rainfall threshold determination for evacuation planning; and (8) communicate findings by rapidly disseminating VAR spatial data and attributes to local agency representatives within approximately 1 to 2 weeks from the beginning of field deployment and compiling a concise final report, reducing risk to life-safety, property, and infrastructure (Figure 2). Typical WERT emergency protection measures include (1) the use of early warning systems (including cell phone watches/advisories/warnings using National Weather Service [NWS] radar forecasts and/or data from telemetered precipitation gages) for evacuation planning; (2) monitoring and maintaining road drainage infrastructure during strong storms; (3) structure protection using flood control and/or deflection structures (K-rails, sandbags, Muscle Wall® portable barrier, etc.), (4) channel clearance of woody debris and vegetation near watercourse crossings; (5) debris removal from debris basins; (6) temporary signage in areas of potential hazard (e.g., low-water crossings, rockfall along public roads), and (7) closure of high-risk areas (e.g., campgrounds, parks, hiking trails). Other types of recommendations are made depending on the site-specific conditions that are present. Along with the NWS and the USGS, WERT has become increasingly involved in assisting with

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 409–422

Post-Fire Assessment and Monitoring: What Have We Learned?

Figure 3. A Montecito home destroyed by debris ﬂows on January 9, 2018. This is an example of the many structures destroyed by these events.

Figure 2. A ﬂowchart illustrating the simpliﬁed WERT process.

developing rainfall thresholds for triggering early warning and storm response planning that are uniquely tailored to each individual ﬁre or critical burned watershed area. This is done by carefully comparing the locations of high-threat VARs relative to the rainfall threshold predictions of the USGS debris ﬂow

model, while also considering the magnitude of the potential post-fire hazard, local empirical data on postfire rainfall response thresholds, and flood/landslide history. The WERT procedures have been used 35 times for fires or multiple fires/complexes from 2016 through 2020. There has been a steady increase in use over time, particularly during the severe fire seasons of 2017, 2018, and 2020. As with earlier fires, when the burned areas included extensive federal lands along with the SRA lands, the WERT collaborated closely with the federal Burned Area Emergency Response (BAER) teams from the U.S. Forest Service or the U.S. Department of Interior, sharing knowledge and modeling results and ensuring that critical portions of the fire were evaluated. This cooperative effort occurred for nearly half the fires assessed by state teams. Although several of these 35 WERT-assessed fire areas produced debris flows and hyper-concentrated flood flows, the 2017 Thomas Fire was the most severe test of the WERT approach. A very high-intensity storm in January 2018 generated numerous destructive debris flows that killed 23 people in Montecito, damaged or destroyed over 500 homes, and closed U.S. Route 101 for 13 days (Oakley et al., 2018; Kean et al., 2019; Lukashov et al., 2019; and Lancaster et al., 2021; Figure 3). The storm event was much more severe than expected and occurred only 48 hours after the field team had the detailed topographic data needed to more accurately predict potential flowpaths on developed portions of the alluvial fan landform. The WERT identified areas of highest risk prior to the storm and warned Santa Barbara County officials of the possible consequences of the impending storm event. This led to widespread evacuations and the saving of

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 409–422

413

Caﬀerata, Coe, and Short

Figure 4. The number of post-fire assessments that have overwintered, number of high life-safety VARs, and number of assessments with documented debris flow, flooding, or rockfall events by year (the number of events is limited to one per assessment).

numerous lives. A secondary CGS and USGS-led postdebris flow WERT mapped the extent of the debris flows and produced a decision matrix for different triggering rainfall events to be used for future evacuation planning. This secondary, more focused process led by CGS has been used for several incidents after the Thomas Fire. SUMMARY OF STATE POST-FIRE ASSESSMENTS CONDUCTED FROM 2003 THROUGH 2020 Post-fire watershed response varies by hydrogeomorphic regime, with different regimes having unique topography, soil properties, and sediment supply (Moody et al., 2013). California’s geomorphic provinces (CGS, 2002) provide a proxy for these hydrogeomorphic regimes. Of the 64 fires or fire complexes evaluated by state-led post-fire assessment teams from 2003 through 2020, the majority have taken place in the Coast Ranges and Transverse Ranges geomorphic provinces of California (∼60 percent). For all 64 fires with storm events from at least one winter, 27 (42 percent) had some level of documented debris flow, flooding, and/or rockfall events. The number of postfire assessments; high-threat, life-safety VARs; and debris flow, flooding, or rockfall events by year are displayed in Figure 4. The frequency of fires with a documented response suggest a north-south gradient in post-fire watershed response, with areas in the Transverse Ranges, Peninsular Ranges, and southern Coast Ranges having a higher potential likelihood for post-

414

fire response than the northern Coast Ranges and Klamath Mountains (Figure 5). The areas with state post-fire assessments conducted in the Sierra Nevada have had mostly minor events, with the exception of major flooding and debris flows following the 2007 Piute Fire in the southern part of the range (DeGraff et al., 2011. Post-fire debris flow risk is high in southern California due to steep topography, high-intensity rainfall, and high fire frequency (Kean and Staley, 2021). In general, areas outside of southern California are more subject to post-fire flood hazards than debris flow hazards. Short-duration, high-intensity rainfall during the first two post-fire winters is a key factor in determining whether debris flows and flooding occur in burned areas (e.g., Moody, 2012; Staley et al., 2016). Rainfall intensity data are available for several of the debris flow and flooding events that have occurred in areas with state post-fire assessments from 2003 through 2020 (Table 1). For southern California, median rainfall intensities required for debris flow initiation in the Transverse Ranges are approximately 23 mm/hr, 18 mm/hr, and 14 mm/hr for 15-, 30-, and 60-minute durations, respectively (Staley et al., 2020; Figure 6). These events are mostly associated with atmospheric rivers (Oakley and Lancaster, 2018) and generally have a recurrence interval of 2 years or less (Cannon et al., 2008; Staley et al., 2020). To date, limited data are available for other key geomorphic provinces in California. However, rainfall intensities that trigger debris flows appear to be generally higher in northern California when compared to those in southern

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 409–422

Post-Fire Assessment and Monitoring: What Have We Learned?

Figure 5. Post-fire assessments conducted from 2003 through 2020 by geomorphic province, along with the number with documented debris flow, flooding, or rockfall events (note that two evaluations took place in multiple provinces, and the number of events is limited to one per assessment).

California (Oakley and Lancaster, 2018; Neptune et al., in press). INCREASING THE EFFICACY OF POST-FIRE HAZARD MITIGATION THROUGH MONITORING AND RESEARCH Substantial progress in post-fire assessment has been made in the past 30 years, and without concurrent technological and analytical advances, post-fire evaluation would be much more time consuming, less spatially resolved, and much less accurate than it is today (DeGraff, 2014). For example, technological advances such as satellite-derived BARC data enable a reasonably accurate and spatially explicit indexing of soil burn severity. The widespread use of these data after 2000 has greatly increased the efficiency of post-fire assessments by allowing the state post-fire assessment teams to prioritize evaluation of areas with the highest likelihood of hazard. Field-validated BARC maps provide critical input data (e.g., burn severity or differenced normalized burn ratio) for most of the models used by WERT, including the USGS debris flow models, the Forest Service Watershed Erosion Predic-

tion Project (WEPP) derivative models for post-fire surface erosion (e.g., Erosion Risk Management Tool [ERMiT]), and rapid post-fire flood flow prediction methods (Kinoshita et al., 2013, 2014). Considerably more detailed hydrologic and hydraulic models can be used where appropriate (Silver Jackets, 2020), but there is uncertainty in calibrating them for post-fire conditions. Post-fire debris flow models are the most useful analytical model used by state assessment teams. Debris flow modeling was initiated in southern California by the USGS first in 2003 with debris flow likelihood estimates (Cannon et al., 2003), followed by incorporation of debris flow volume estimations in 2007 (Staley, 2020). The USGS has continued to refine its suite of models, with periodic updates to the debris flow likelihood model (Cannon et al., 2010; Staley et al., 2016) and the volumetric model (Gartner et al., 2014). The USGS debris flow models are very useful in southern California, since the dataset used to develop the model largely came from this area. However, based on the limited data collected to date, these models tend to overpredict debris flow likelihood and volumetric magnitude in geomorphic provinces such as the Sierra

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 409–422

415

Cafferata, Coe, and Short Table 1. State post-fire assessments with precipitation data and debris flow, flooding, and/or rockfall events from 2003 through March 2021 (m = missing data).1

Fire Year

Fire Name

2003

Old and Grand Prix

2003 2007

Simi Poomacha

2007 2008

Cannon et al., 2011 Longstreth, 2012

2008

Santiago Basin Complex/ Indians Piute

2008 2015

Sayre Butte

2016

Soberanes

Oakley et al., 2017 Oakley and Lancaster, 2018 Oakley and Lancaster, 2018

2016

Loma

WERT/Cal OES, 2017

2017

Thomas

2018 2020

Holy River

Lukashov et al., 2019; Oakley et al., 2018; Lancaster et al., 2021 Swanson, 2020 Preliminary USGS data

2020

Carmel

Preliminary USGS data

2020

CZU Lightning Complex Bond

Preliminary USGS data

2020

Reference Cannon et al., 2011; DeGraﬀ, 2014 Oakley et al., 2017 Cannon et al., 2011

DeGraﬀ et al., 2011

Preliminary NWS data

Brief Event Description 68 debris flows, 16 deaths, 52 homes damaged 2 debris flows 20 debris flows/flood flows 5 debris flows Debris flow damage to State Park Flooding/major debris flows 2 debris flows Small debris flows impacting infrastructure Small debris flows; hyper-concentrated flood flows Debris flows/ landslides >20 debris flows, >500 homes damaged or destroyed, 23 deaths Flooding and debris flows Hyper-concentrated flood flows; 25 homes damaged, one injury Small debris flow Small debris flows and sediment-laden flows Debris flows, flooding; minor impacts to homes

Geomorphic Province

15-min Rainfall Rate (mm/hr)

Maximum 60-min Rainfall Rate (mm/hr)

Transverse Ranges

28–88

15–28

Transverse Ranges Peninsular Ranges

m 16–28

6–14 9–18

Peninsular Ranges Southern Coast Ranges Sierra Nevada

44 m

m 21

16–20

Transverse Ranges Sierra Nevada

m 23–43

5–13 6–17

Southern Coast Ranges

19–31

16–25

5–22

72–104

17–39

24–40 11–18

14–26 8

32–57

17–19

Southern Coast Ranges Transverse Ranges and Southern Coast Ranges Peninsular Ranges Southern Coast Ranges Southern Coast Ranges Southern Coast Ranges Peninsular Ranges

1 Note that the rainfall rates provide a range based on the peak reported rainfall at the gage site, sometimes for events producing debris ﬂows on diﬀerent days. In many cases these peak values were used as an approximation of the triggering rainfall rate; the actual triggering rainfall rate may have been above or below the peak.

Nevada, Coast Ranges, and Klamath Mountains. The ability of the USGS debris model to integrate the key driving variables into spatially explicit predictions of post-fire response still make it extremely useful for prioritizing field work in these provinces, especially when under the threat of high-intensity storms. Similarly, Forest Service WEPP derivatives (e.g., Disturbed WEPP and/or ERMiT) have been widely used in burned area evaluations to estimate surface erosion since the early 2000s. These models are used to communicate relative threat to VARs such as damage to water supply/treatment infrastructure, public road networks, and aquatic resources. The development and widespread application of these models by post-fire assessment teams are a testament to their utility, as well as the longstanding commitment and investment on

416

the part of agencies such as the U.S. Forest Service and the USGS. Despite the success of these models, it is clear that a new or reﬁned generation of analytical tools are needed to expand the scope of predictive capability outside of southern California and to reﬁne hazard predictions in general. Examples of potential improvements include:

r Existing model refinement and calibration based upon additional monitoring data to predict the likelihood, volumetric magnitude, and threshold rainfall intensities for post-fire debris flows in regions outside of southern California; r Enhanced ability to discriminate where different post-fire processes (i.e., sediment-laden floods

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 409–422

Post-Fire Assessment and Monitoring: What Have We Learned?

Figure 6. A box and whisker plot showing maximum rainfall intensity for 15-, 30-, and 60-minute durations for debris ﬂows in the Transverse Ranges of southern California (data from Staley et al., 2020). Horizontal lines represent the median, boxes represent the interquartile range, whiskers represent 1.5 times the interquartile range. Median values are displayed in the ﬁgure.

versus debris flows) will operate under different rainfall intensities; r Modeling to predict the spatial extent (i.e., runout distance), density, depth, and velocities from postfire debris flows and sediment-laden floods; r Replacement of outdated and unvalidated rapid post-fire flood hydrology methods (e.g., southern California look-up tables in Rowe et al., 1949; the flow modifier approach described in Foltz et al., 2009) with new updated models (e.g., Moody, 2012; Wilder et al., 2020), as well as better ability to rapidly predict bulked peak flows (i.e., hyper-concentrated flows) and runoff hydrographs throughout the entire state of California; r Better ability to estimate surface erosion and account for sediment supply and debris flow volumes after a wildfire, since currently available models have a tendency to predict inaccurate erosion rates across many portions of California;

r Ability to more accurately predict post-fire rockfall (DeGraff and Gallegos, 2012; DeGraff et al., 2015);

r Ability to better predict potential ash and contaminant transport following wildfire (Doerr et al., 2019); r Expanded use of unmanned aircraft systems to assist with post-fire emergency response planning and provide baseline topographic and aerial imagery for post-storm monitoring (Cress et al., 2014), and r Ability to better identify post-wildfire sediment transport and topographic change due to debris flows and flooding, for instance by using repeat LiDAR surveys (Corsini et al., 2009; Anderson and Pitlick, 2014). These improvements are dependent upon the collection of field data for model development, calibration, and/or validation. Additionally, more LiDAR-based topographic information across the state is necessary

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 409–422

417

Caﬀerata, Coe, and Short

for more accurate modeled and field identified hazards. In particular, a move toward more distributed, physically based models (e.g., hydrologic and hydraulic models) requires extensive parameterization and calibration, which in turn requires more field data. Ideally, physically based models should be detailed enough to represent the dominant hydrologic and geomorphic processes following wildfire, while still being simple enough to calibrate and run in a rapid timeframe. Although substantial progress has been made in terms of increasing the understanding of the post-fire hydrologic and geomorphic response, further advancement will involve a significant investment of time and resources, as well as the ability to move quickly to instrument burned watersheds so that critical data from the first storms can be recorded. Hence, it is essential to increase monitoring and research capacity so that advancement in post-fire hazard prediction can continue well into the future, with increased efficacy for post-fire hazard evaluation. This is particularly important, as numerous published papers suggest that climate change will increase the length of the fire season and the number of large, severe fires in California and the other western states (e.g., Keyser and Westerling, 2019; Williams et al., 2019; and Goss et al., 2020). These fires are expected to have greater impacts on California’s wildland-urban interface (WUI), and they have prompted a renewed emphasis on pre-fire planning to reduce fuel hazard in and around WUI areas. A consequence of increased fire activity is expected to be greater post-fire erosion (Bladon, 2018), with the likelihood and magnitude of post-fire watershed hazards anticipated to increase over time and with larger impacts to populations exposed to these potential hazards. As such, this necessitates an explicit and proactive pre-fire planning approach that recognizes the co-benefits of reducing fuel hazard while simultaneously increasing the resilience of WUI areas to post-fire watershed hazards. STRATEGIES FOR BETTER EMERGENCY PREPAREDNESS WITH PRE-FIRE HAZARD MITIGATION PLANNING Currently, maps of post-wildfire debris and flood flow risk generally do not exist until after a post-fire hazard assessment is conducted. Under the existing process, local hazard mitigation strategies and projects are difficult to prepare and implement in the short time between fire and rain, especially after late-season wildfires that occur close to or during the wet season. In addition, when these maps are prepared late in the season, communities have little time to develop an understanding and awareness of the risk, thus limiting their effectiveness as a means to support evacuation plan-

418

ning. However, the same methods used for post-fire hazard evaluation can also be utilized to simulate watershed response under a variety of different post-fire scenarios. To promote community resilience, a pre-fire lifesafety hazard assessment program that simulates potential post-fire impacts under a range of wildfire and rainfall scenarios in a changing climate is necessary in both underserved and higher risk counties. This prefire effort would produce post-fire hazard data and maps in advance of wildfire so that long-term planning, risk communication, and mitigation measures can be implemented. In the event of multiple fires, it can be used to prioritize post-fire hazard assessments when staffing is limited. Additionally, with this prefire hazard information in hand prior to a fire, a postfire hazard assessment can rapidly confirm those hazards and identify if additional emergency mitigation measures are necessary to protect lives, property, and infrastructure. This pre-fire planning approach can be used to inform local agencies of their risk prior to devastating post-fire effects and prepare local communities for post-fire hazards (Staley et al., 2018). For example, Lancaster et al. (2014) used existing regionalscale predictive models to assess the effects of postfire runoff from a pre-fire planning perspective for 20 southern California watersheds in Santa Barbara, Ventura, Los Angeles, San Bernardino, and Riverside counties situated near populated areas with residential and commercial infrastructure VARs. Multiple predictive models enabled hazard identification and mapping, highlighting watersheds prone to potentially dangerous post-fire events. The models used by Lancaster et al. (2014) and more current approaches can be used for watersheds covering large geographic areas in California and other western states (Kean and Staley, 2021). Also, since WUI-to-wildland ignitions are common (Syphard and Keeley, 2015), it can help to guide targeted fuel treatments that reduce the exposure of the most hazard-prone watersheds to human ignitions. More broadly, California counties have adopted allhazard mitigation plans that include wildfire response, typically in a general way. With post-fire life-safety hazard information available prior to a devastating wildfire, detailed hazard mitigation plans can specify how a county will address the threat of postfire debris flows and flooding. Additionally, long-term risk communication and infrastructure projects to reduce post-fire risk can be implemented that are consistent with the recommendations of the Association of State Floodplain Managers (ASFPM Foundation, 2019) and information gaps identified in the California Statewide Hazard Mitigation Plan (Cal OES, 2018).

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 409–422

Post-Fire Assessment and Monitoring: What Have We Learned? Table 2. Summary of post-ﬁre research and observations.

Topic

Research Result/Observations

Post-ﬁre grass seeding

Straw mulching, helimulching, hydromulching, wood shred mulching Fire suppression repair work

State post-ﬁre evaluation teams

Post-fire debris flows and flooding events Rainfall intensity thresholds for debris flows and flooding events Pre-fire hazard mitigation planning

Aerial application not eﬀective for highly erodible sites with a Mediterranean climate; limits native species recovery Eﬀective for reducing erosion since cover provided before fall rains

Effective; road treatments found to be an important component of post-fire rehabilitation work Licensed professionals can assess post-fire watershed hazards and rapidly inform local agencies regarding emergency protection measures Short-duration, high-intensity rainfall (e.g., <30 minutes) is a key factor for triggering events Reliable data exist for southern California; more data are required for geomorphic regions in northern and central California Better emergency preparedness is possible in higher risk counties with a pre-fire hazard mapping and assessment program

CONCLUSIONS The state of California’s approach to addressing post-fire watershed hazards has evolved substantially in 60 years. Despite initial reliance on aerial grass seeding to minimize post-fire impacts, studies conducted in California have shown that aerial grass seeding is not effective for reducing erosion, especially in the southern part of the state (Peppin et al., 2010). Where aerial seeding is effective, it often limits native species recovery and conifer regeneration success. Effective hillslope treatments for surface erosion control are those that provide cover prior to the first storm events. These types of treatments include straw mulching, helimulching, hydromulching, and wood shred mulching, which are expensive and, thus, can only be applied to limited areas above identified VARs due to their cost. Overall, the amount of erosion reduction provided by any hillslope treatment for a large wildfire is small, but these types of treatments can be effective depending upon the objectives of the landowner or local agency. Fire suppression repair work, including erosion control on firelines, removing soil and debris at fireline stream crossings, and implementing erosion control best management practices (i.e., drainage) on roads, should be accomplished on all wildland fires. After an initial screening for potential threat, highly technical interdisciplinary teams can (1) rapidly assess post-fire threats to life-safety, property, and infrastruc-

Management Strategy/Recommendations Limit to ground application to mechanically disturbed sites; use with mulching Can be eﬀective for limited areas above key values-at-risk; large scale application is cost prohibitive in most situations Apply to all ﬁres regardless of size

Screen ﬁres and utilize state teams only where necessary and appropriate

Utilize National Weather Service watches and warnings based on regional radar for cell phone early warning alerts Increase post-ﬁre monitoring and research capacity to improve hazard prediction Model watershed hazards covering large geographic areas with high life-safety exposure

ture from debris flows, flood flows, rockfall, and erosion and (2) quickly inform local agencies regarding appropriate emergency protection measures. The current state post-fire evaluation process is an effective method to identify post-fire hazards and notify local agencies of post-fire hazards and risks. When counties need further assistance (i.e., counties without extensive public works or flood control departments), additional evaluations can be conducted to refine assessments based on the needs of the local jurisdictions. When there is mixed federal/state/private ownership, close cooperation with federal BAER teams during the field evaluation and with local entities when findings are disseminated is critical to ensure that postfire risks are reduced. Continual improvement of postfire analytical methods for more accurate and spatially resolute hazard evaluation is necessary to achieve risk reduction in a hydro-geomorphically diverse area such as California. Additionally, increased post-fire monitoring and research is required to improve the conceptual understanding of how fire changes physical processes and for developing new analytical methods, especially with changing climate where the likelihood and magnitude of post-fire watershed hazards is expected to increase. Monitoring and research data are not only useful for the advancement of the science, but they also increase the capacity to rapidly provide actionable intelligence to emergency response planners and first responders.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 409–422

419

Caﬀerata, Coe, and Short

Finally, emergency preparedness, including pre-fire hazard mitigation planning for post-fire flooding and debris flows, is needed in populated, higher risk areas such as southern California. Improved public education regarding post-fire threats from flooding and debris flows and post-fire recovery is also required (e.g., Silver Jackets, 2019). Increased public awareness and understanding will likely result in better adherence to evacuation orders and, along with improved analytical methods, reduce the risk from post-fire watershed hazards. A brief synopsis of these summary points is provided in Table 2. ACKNOWLEDGMENTS We thank the following people for providing valuable comments on an expanded report version of this paper: Dr. Joe Wagenbrenner, USFS PSW; Dr. Lee MacDonald, Colorado State University; Tom Spittler, CGS (retired); Jeremy Lancaster, CGS; Dave Longstreth, CGS; Brian Swanson, CGS; Kevin Doherty, CGS; Dr. Dennis Staley, USGS; Dave Young, USFS; and John Munn, CAL FIRE (retired). REFERENCES Anderson, S. and Pitlick, J., 2014, Using repeat lidar to estimate sediment transport in a steep stream: Journal Geophysical Research Earth Surface, Vol. 119, No. 3, pp. 621–643. Association of State Floodplain Managers (ASFPM) Foundation, 2019, California Floodplain Risk Management Symposium Results and Recommendations White Paper: Floodplain Management Association (FMA) and ASFPM Foundation, Madison, WI, 15 p.: Electronic document, available at https://www.asfpmfoundation.org/ace-images/ WhitePaper_20191206__FMA-California.pdf Barro, S. C. and Conard, S. G., 1987, Use of Ryegrass Seeding as an Emergency Revegetation Measure in Chaparral Ecosystems: Pacific Southwest Forest and Range Experiment Station, Forest Service, U.S. Department of Agriculture, Berkeley, CA, Gen. Tech. Rep. PSW-102, 12 p. Bladon, K. D., 2018, Rethinking wildfires and forest watersheds: Science, Vol. 359, No. 6379, pp. 1001–1002. Blanford, R. H. and Gunter, L. E., 1972, Emergency Revegetation-A Review of Project Evaluations: California Department of Forestry, Forest, Range and Watershed Management Section, Sacramento, CA, 21 p. California Department of Forestry (CAL FIRE) and Fire Protection and California Geological Survey (CGS), 2021, Procedural Guide for Watershed Emergency Response Teams: Sacramento, CA, 69 p. California Division of Forestry (CDF), 1972, Emergency Revegetation of Burned Watersheds: Annual report, Sacramento, CA, 14 p. California Geological Survey (CGS), 2002, California Geomorphic Provinces: Note No. 36, Sacramento, CA, 4 p. California Governor’s Office of Emergency Services (Cal OES), 2018, California State Hazard Mitigation Plan: Mather, CA, 801 p.: Electronic document, available at https://www.

420

caloes.ca.gov / cal-oes-divisions / hazard-mitigation / hazardmitigation-planning/state-hazard-mitigation-plan Cannon, S. H.; Boldt, E. M.; Kean, J. W.; Laber, L. J.; and Staley, D. M., 2010, Relations Between Rainfall and Postfire Debris-Flow and Flood Magnitudes for Emergency-Response Planning, San Gabriel Mountains, Southern California: U.S. Geological Survey Open-File Report 2010-1039, 31 p. Cannon, S. H.; Boldt, E. M.; Laber, J. L.; Kean, J. W.; and Staley, D. M., 2011, Rainfall intensity-duration thresholds for postfire debris-flow emergency-response planning: Natural Hazards, Vol. 59, pp. 209–236. Cannon, S. H.; Gartner, J. E.; Rupert, M. G.; and Michael, J. A., 2003a, Emergency Assessment of Debris-Flow Hazards from Basins Burned by the Grand Prix and Old Fires of 2003, Southern California: U.S. Geological Survey Open-File Report OF-03-475, 10 p. plus maps. Cannon, S. H., Gartner, J. E.; Wilson, R. C.; Bowers, J. C.; and Laber, J. L., 2008, Storm rainfall conditions for floods and debris flows from recently burned areas in southwestern Colorado and southern California: Geomorphology, Vol. 96, pp. 250–269. Cleveland, G. B., 1973, Fire + rain = mudflows, Big Sur, 1972: California Geology, Vol. 26, No. 6, pp. 127–135. Cleveland, G. B., 1977. Analysis of Erosion Following the Marble Cone Fire, Big Sur Basin, Monterey County, California: California Division of Mines and Geology Open File Report 7712, Sacramento, CA, 13 p. Conard, S. G.; Regelbrugge, J. C.; and Wills, R., 1991, Preliminary effects of ryegrass seeding on postfire establishment of natural vegetation in two California ecosystems. In Andrews, P. L. and Potts, D. F. (Editors), Proceedings of the 11th Conference on Fire and Forest Meteorology: Society of American Foresters, Bethesda, MD, pp. 314–321. Corsini, A.; Borgatti, L.; Cervi, F.; Dahne, A.; Ronchetti, F.; and Sterzai, P., 2009, Estimating mass-wasting processes in active earth slides–earth flows with time-series of highresolution DEMs from photogrammetry and airborne LiDAR: Natural Hazards and Earth System Sciences, Vol. 9, pp. 433–439. Cress, J.; Hutt, M.; Sloan, J.; Bauer, M.; Feller, M.; and Goplen, S., 2014, U.S. Geological Survey Unmanned Aircraft Systems (UAS) Roadmap 2014: U.S. Geological Survey OpenFire Report 2015-1032, 60 p. DeBano, L. F., 1981, Water Repellent Soils: A State-of-the-Art: U.S. Department of Agriculture, Forest Service, Pacific Southwest Forest and Range Experiment Station Gen. Tech. Rep. PSW-GTR-46, Berkeley, CA, 21 p. DeGraff, J. and Gallegos, A., 2012, The challenge of improving identification of rockfall hazard after wildfires: Environmental Engineering Geoscience, Vol. 18, No. 4, pp. 389–397. DeGraff, J.; Shelmerdine, B.; Gallegos, A.; and Annis, D., 2015, Uncertainty associated with evaluating rockfall hazard to roads in burned areas: Environmental Engineering Geoscience, Vol. 21, No. 1, pp. 21–33. DeGraff, J.; Wagner, D.; Gallegos, A.; DeRose, M.; Shannon, C.; and Ellsworth, T., 2011, The remarkable occurrence of large rainfall-induced debris flows at two different locations on July 12, 2008, southern Sierra Nevada, CA, USA: Landslides, Vol. 8, pp. 343–353. DeGraff, J. V., 2014, Improvement in quantifying debris flow risk for post-wildfire emergency response: Geoenvironmental Disasters, Vol. 1, No. 5, doi:10.1186/s40677-014-0005-2. Doerr, S.; Neris, J.; Elliot, W. J.; Robichaud, P. R.; Lew, R.; Santin, C.; and Sheridan, G. J., 2019, Incorporating Water Contamination Risk from Wildfire Ash into the

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 409–422

Post-Fire Assessment and Monitoring: What Have We Learned? Decision-Making Process: A New Online Tool for Researchers and End-Users: American Geophysical Union (AGU) Fall Meeting Abstracts: Electronic document, available at https://ui.adsabs.harvard.edu/abs/2019AGUFM.H31H..06D/ abstract Foltz, R. B., and Copeland, N. S., 2009, Evaluating the efficacy of wood shreds for mitigating erosion: Journal of Environmental Management, Vol. 90, pp. 779–785. Foltz, R. B.; Robichaud, P. R.; and Rhee, H., 2009, A Synthesis of Postfire Road Treatments for BAER Teams: Methods, Treatment Effectiveness, and Decision Making Tools for Rehabilitation: U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station Gen. Tech. Rep. RMRS-GTR228, Fort Collins, CO, 152 p. Gartner J. E.; Cannon, S. H.; and Santi, P. M., 2014, Empirical models for predicting volumes of sediment deposited by debris flows and sediment-laden floods in the transverse ranges of southern California: Engineering Geology, Vol. 176, pp. 45–56. Goss, M.; Swain, D. L.; Abatzoglou, J. T.; Sarhadi, A.; Kolden, C. A.; Williams, A. P.; and Diffenbaugh, N. S., 2020, Climate change is increasing the likelihood of extreme autumn wildfire conditions across California: Environmental Research Letters, Vol. 15, No. 9, 14 p. Griffith, R. W., 1998, Burned area emergency rehabilitation. In Proceedings of the 19th Annual Forest Vegetation Management Conference: Wildfire Rehabilitation. Forest Vegetation Management Conference, Redding, CA, pp. 4–7. Kean, J. W. and Staley, D. M., 2021, Forecasting the frequency and magnitude of postfire debris flows across southern California: Earth’s Future, Vol. 9, No. 3, e2020EF001735, https://doi.org/10.1029/2020EF001735. Kean, J. W.; Staley, D. M.; and Cannon, S. H., 2011, In situ measurements of post-fire debris flows in southern California: Comparisons of the timing and magnitude of 24 debris-flow events with rainfall and soil moisture conditions: Journal Geophysical Research, Vol. 116, No. F4, F04019, 21 p. Kean, J. W.; Staley, D. M.; Lancaster, J. T.; Rengers, F. K.; Swanson, B. J.; Coe, J. A.; Hernandez, J. L.; Sigman, A. J.; Allstadt, K. E.; and Lindsay, D. N., 2019, Inundation, flow dynamics, and damage in the 9 January 2018 Montecito debris-flow event, California, USA: Opportunities and challenges for post-wildfire risk assessment: Geosphere, Vol. 15, No. 4, pp. 1140–1163, https://doi.org/10.1130/GES02048.1. Keyser, A. R. and Westerling, A. L., 2019, Predicting increasing high severity area burned for three forested regions in the western United States using extreme value theory: Forest Ecology and Management, Vol. 432, pp. 694–706. Kinoshita, A.; Hogue, T. S.; and Napper, C., 2013, A Guide for Pre- and Post-Fire Modeling and Application in the Western United States: USDA Forest Service, National Technology and Development Program, 1325 1802—SDTDC, 62 p. Kinoshita, A. M.; Hogue, T. S.; and Napper, C., 2014, Evaluating pre- and post-fire peak discharge predictions across western U.S. watersheds: Journal American Water Resources Association, Vol. 50, No. 6, pp. 1540–1557, doi:10.1111/jawr.12226. Lancaster, J. T.; McCrea, S. E.; and Short, W. R., 2014, Assessment of Post-Fire Runoff Hazards for Pre-Fire Hazard Mitigation Planning—Southern California: California Geological Survey Special Report 234, California Department of Conservation, Sacramento, CA, 66 p. plus appendices. Lancaster, J. T.; Swanson, B. J.; Lukashov, S.; Oakley, N.; Lee, J. B.; Spangler, E.; Hernandez, J. L.; Olson, B. P. E.; DeFrisco, M. J.; Lindsay, D. J.; Schwartz, Y. J.; McCrea, S. E.; Roffers, P. D.; and Tran, C. M., 2021, Observations and

analyses of the 9 January 2018 debris flow disaster, Santa Barbara County: Environmental Engineering Geoscience, Vol 27, No. 1, pp. 3–27. Longstreth, D. L., 2012, Assessing Post Fire Landsliding and Flooding along Highway Corridors, an Example from California State Route 1, Monterey County, California, 2008: Poster, 63rd Annual Highway Geology Symposium, Redding, CA. Lukashov, S. G.; Lancaster, J. T.; Oakley, N. S.; and Swanson B. J., 2019, Post-fire debris flows of 9 January 2018, Thomas Fire, southern California: Initiation areas, precipitation and impacts. In Kean, J. W.; Coe, J. A.; Santi, P. M.; and Guillen, B. K. (Editors), Debris-Flow Hazards Mitigation: Mechanics, Monitoring, Modeling, and Assessment: Proceedings of the Seventh International Conference on Debris-Flow Hazards Mitigation. Association of Environmental and Engineering Geologists Special Publication 28, Golden, CO, pp. 774–781. Moody, J. A., 2012, An Analytical Method for Predicting Postwildfire Peak Discharges: U.S. Geological Survey Scientific Investigations Report 2011-5236, 36 p. Moody, J. A.; Shakesby, R. A.; Robichaud, P. R.; Cannon, S. H.; and Martin, D. A., 2013, Current research issues related to post-wildfire runoff and erosion processes: Earth-Science Reviews, Vol. 122, pp. 10–37. Neary, D. G.; Ryan, K. C.; and DeBano, L. F., 2005, Wildland Fire in Ecosystems: Effects of Fire on Soils and Water: U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station Gen. Tech. Rep. RMRS-GTR-42, Vol. 4, Ogden, UT, 250 p. Neptune, C. K.; DeGraff, J. V.; Pluhar, C. J.; Lancaster, J. T.; and Staley, D. M., 2021, Rainfall thresholds for post-fire debris flow generation, western Sierra Nevada, CA. Environmental Engineering Geoscience, Vol. 27, pp. 439–453. Oakley, N. S.; Cannon, F.; Munroe, R.; Lancaster, J. T.; Gomberg, D.; and Ralph, F. M., 2018, Brief communication: meteorological and climatological conditions associated with the 9 January 2018 post-fire debris flows in Montecito and Carpinteria, California, USA: Natural Hazards and Earth System Sciences, Vol.18, pp. 3037–3043. Oakley, N. S. and Lancaster, J. T., 2018, Post-Fire Debris Flows in California: An Atmospheric Perspective: PowerPoint presentation, 27th Flood Warning Systems Training Conference and Exposition, April 16–20, 2018, Ventura, CA: Electronic document, available at https://www.alertsystems. org/presentations/Conf2018/Session7-Forecasting_Weather/ Oakley_FORx.pdf Oakley, N. S.; Lancaster, J. T., Kaplan, M. L.; and Ralph, F. M., 2017, Synoptic conditions associated with cool season post-fire debris flows in the Transverse Ranges of southern California: Nat. Hazards, Vol. 88, pp. 327–354, doi:10.1007/s11069-017-2867-6. Parsons, A.; Robichaud, P. R.; Lewis, S. A.; Napper, C.; and Clark, J. T., 2010, Field Guide for Mapping Post-Fire Soil Burn Severity: U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station Gen. Tech. Rep. RMRSGTR-243, Fort Collins, CO, 49 p. Peppin, D.; Fuléa, 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, No. 5, pp. 573–586. Robichaud, P. R.; Ashmun, L. E.; and Sims B. D., 2010a, PostFire Treatment Effectiveness for Hillslope Stabilization: U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station Gen. Tech. Rep. RMRS-GTR-240, Fort Collins, CO, 62 p.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 409–422

421

Cafferata, Coe, and Short Robichaud, P. R.; Beyers, J. L.; and Neary D. G., 2000, Evaluating the Effectiveness of Postfire Rehabilitation Treatments: U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station Gen. Tech. Rep. RMRS-GTR-63, Fort Collins, CO, 85 p. Robichaud, P. R.; MacDonald, L. H.; and Foltz, R. B., 2010b, Chapter 5: Fuel management and erosion. In Elliot, W. J., Miller, I. S., Audin, L. (Editors), Cumulative Watershed Effects of Fuel Management in the Western United States: U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station Gen. Tech. Rep. RMRS-GTR-231, Fort Collins, CO, pp. 79–100. Rowe, P. B.; Countryman, C. M.; and Storey H. C., 1949, Probable Peak Discharges and Erosion Rates from Southern California Watersheds as Influenced by Fire: U.S. Department of Agriculture, Forest Service, California Forest and Range Experiment Station, Berkeley, CA, 305 p. Silver Jackets, 2019, After Wildfire—A Guide for California Communities: U.S. Army Corps of Engineers, Sacramento District, Sacramento, CA, 44 p.: Electronic document, available at http://www.readyforwildfire.org/wp-content/uploads/AfterWildfire-Guide-10JUNE2019_draft_final-ADA-compliant. pdf Silver Jackets, 2020, Flood after Fire California Toolkit: A Resource for Technical Specialists to Assess Flood and Debris Flow Risk after a Wildfire: U.S. Army Corps of Engineers, Sacramento District., Sacramento, CA, 69 p.: Electronic document, available at http://www.readyforwildfire.org/wp-content/ uploads/Flood-After-Fire_California - Toolkit_September2020-2.pdf Spittler, T. E., 1993, Emergency landslide hazard evaluation following the Tunnel Fire, October 19-23, 1991: California Geology, Vol. 46, pp. 174–179. Spittler, T. E., 1995, Fire and the debris flow potential of winter storms. In Keeley, J. E. and Scott, T. (Editors), Brushfields in California Wildlands: Ecology and Resource Management: International Association of Wildland Fire, Fairfield, WA, pp. 113–120. Spittler, T. E., 2005, California Fires, Floods and Landslides: Paper presented at the Disaster Resistant California Conference, California Governor’s Office of Emergency Services, May 15– 18, 2005, Sacramento, CA, 10 p. Spittler, T. E.; Barrows, A. G.; Tan, S.; Irvine, P.; and Treiman, J., 1994, Debris flow potential following the 1993 southern California fire storms: Geological Society America, Abstracts Programs, Vol. 26, No. 2, p. 95. Staley, D. M., 2020, personal communication, USGS, Geologic Hazards Science Center, Denver, CO. Staley, D. M.; Kean, J. W.; Rengers, F. K., 2020, The recurrence interval of post-fire debris-flow generating rainfall in the southwestern United States: Geomorphology, Vol. 370, No. 107392, 10 p. Staley, D. M.; Negri, J. A.; Kean, J. W.; Tillery, A. C.; and Youberg, A. M., 2016, Updated Logistic Regression Equations for the Calculation of Post-Fire Debris-Flow Likelihood in the

422

Western United States: U.S. Geological Survey Open-File Report 2016-1106, 13 p. Staley, D. M.; Tillery, A. C.; Kean, J. W.; McGuire, L. A.; Pauling, H. E.; and Rengers, F. K., 2018, Estimating postfire debris-flow hazards prior to wildfire using a statistical analysis of historical distributions of fire severity from remote sensing data: International Journal Wildland Fire, Vol. 27, No. 9, pp. 595–608, doi:10.1071/WF17122. Swanson, B., 2020, Post-Fire Watershed Response and Impacts – Lessons Learned from the 2018 Holy Fire: PowerPoint presentation, 2020 Watershed University SoCal Summit. Session: The Juggling Act: Managing Multiple Watershed Hazards, April 21, 2020: Electronic document, available at https://silverjackets.nfrmp.us/State-Teams/California/2020Watershed-University-Summit Syphard, A. D. and Keeley, J. E., 2015, Location, timing and extent of wildfire vary by cause of ignition: International Journal Wildland Fire, Vol. 24, No. 1, pp. 37–47. Troxell, H. C. and Peterson, J. Q., 1937, Flood in La Canada Valley, California, January 1, 1934: U.S. Geological Survey Water Supply Paper 796-C, pp. 53–98. Wagenbrenner, J. W.; MacDonald, L. H.; and Rough, D., 2006, Effectiveness of three post-fire rehabilitation treatments in the Colorado Front Range: Hydrological Processes, Vol. 20, No. 14, pp. 2989–3006. Watershed Emergency Response Team/California Governor’s Office of Emergency Services (WERT/Cal OES), 2017, Loma Fire Watershed Emergency Response Team Report Supplement-Storm Damage Response: Cal OES Mission Task 2017-Coastal-29698, report sent to the Santa Clara County Office of Emergency Services, San Jose, CA, dated February 6, 2017, 10 p. plus appendices. Wilder, B. A.; Lancaster, J. T.; Cafferata, P. H.; Coe, D. B. R.; Swanson, B. J.; Lindsay, D. N.; Short, W. R.; and Kinoshita A. M., 2020, An analytical solution for rapidly predicting post-fire peak streamflow for small watersheds in southern California: Hydrological Processes, pp. 1–14, doi:10.1002/hyp.13976. Williams, A. P.; Abatzoglou, J. T.; Gershunov, A.; GuzmanMorales, J.; Bishop, D. A.; Balch, J. K.; and Lettenmaier D., 2019, Observed impacts of anthropogenic climate change on wildfire in California: Earth’s Future, Vol. 7, No. 8, pp. 892–910. Wohlgemuth, P. M.; Beyers, J. L.; and Hubbert, K. R., 2009, Rehabilitation strategies after fire: the California, USA experience. In Cerdá, A.; Robichaud, P. R. (Editors), Fire Effects on Soils and Restoration Strategies: Science Publishers, Enfield, NH, pp. 511–536. Wohlgemuth, P. M.; Beyers, J. L.; Wakeman, C. D.; and Conard S. G., 1998, Effects of grass seeding on soil erosion in southern California chaparral. In Proceedings of the 19th Annual Forest Vegetation Management Conference: Wildfire Rehabilitation. Forest Vegetation Management Conference, Redding, CA, pp. 41–51.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 409–422

Assessment of a Post-Fire Debris Flow Impacting El Capitan Watershed, Santa Barbara County, California, U.S.A. JONATHAN YONNI SCHWARTZ* Minerals & Geology, Los Padres National Forest, U.S. Forest Service, 1190 East Ojai Avenue, Ojai, CA 93023

NINA S. OAKLEY Center for Western Weather and Water Extremes, Scripps Institution of Oceanography, 9500 Gilman Drive, La Jolla, CA 92093

PAUL ALESSIO Department of Earth Science, 1006 Webb Hall, University of California, Santa Barbara, CA 93106

Key Terms: Debris Flows, Post-Fire, Sherpa Fire, Wildfire, Transverse Ranges, Santa Ynez Mountains

protocols to help reduce the threat to life, property, and infrastructure in downstream communities.

ABSTRACT

INTRODUCTION

In the summer of 2016, the Sherpa Fire burned 30.2 km² in steep terrain in western Santa Barbara County, California. Rainfall events in the subsequent wet season produced damaging post-wildfire flooding and debris flows. This paper presents a case study along a watershed within the burned area, El Capitan Creek, that (1) describes the events and conditions that led to the post-wildfire flooding and debris flow events, and (2) documents the debris flow deposits and inundation zone impacted by the events. Observations compiled after three post-wildfire precipitation events indicate that three distinct flow processes impacted El Capitan Creek between 19 and 22 January 2017. These flow processes included watery flows, hyper-concentrated flows, and debris flows. The velocity and concentrated nature of these flows caused overbanking and channel avulsions that resulted in damaged roads, bridges, pipelines, and major infrastructure damage to the El Capitan Canyon Resort. These events occurred only 1 year prior to the devastating Montecito debris flows of 2018 and call attention to the conditions that produced these impactful flows and highlight the timing and conditions that generate postwildfire debris flows. Information from case studies such as this can guide decision makers and emergency managers to understand the hazards and risks that floods and debris flows pose on communities below steep mountain drainages and support the development of sound

On the morning of 20 January 2017 at approximately 8:55 a.m. PST an intense rainstorm with a peak 15-minute rainfall intensity rate of 76 mm/hr initiated debris flows and sediment-laden flooding in several watersheds recently burned by the Sherpa Fire: El Capitan Creek, Cañada del Corral Creek, and Las Flores Canyon, Santa Barbara County, California, United States (Figure 1). The flooding and debris flows damaged buildings, property, and infrastructure in the El Capitan Canyon Resort, El Capitan State Beach, and an oil and gas facility located in Cañada Del Corral (Figure 2). There is a history of post-fire flooding and debris flows in this region (e.g., U.S. Army Corps of Engineers, 1974; Kean et al., 2019). As the climate warms and dries, there is potential for more frequent and/or severe events in Santa Barbara County due to increased fire size and frequency (Barbero et al., 2015; Westerling, 2018) and precipitation intensification (e.g., Prein et al., 2017). Given the potential for future destructive post-fire debris flows to occur along the Santa Ynez Range and other mountainous regions of the southwestern United States, it is valuable to document these types of events with a focus on the conditions that led to their initiation and damage potential to inform the design and development of infrastructure (e.g., culvert size or bridge design), building codes, and land use planning. Case studies also assist decision makers and emergency managers at local, state, and federal levels in understanding and

*Corresponding author email: jonathan.schwartz@usda.gov

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

423

Schwartz, Oakley, and Alessio

Figure 1. A vicinity and geology map of the watersheds that burned in the Sherpa Fire and were impacted by the 20 January 2017 rainstorm event. Measurement locations and geographic features that are referenced in the paper are denoted on the map.

424

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

Debris Flow Assessment, El Capitan Watershed, Santa Barbara County, California

Figure 2. A Sherpa Fire soil burn severity map showing the watersheds impacted by the 20 January 2017 rainstorm event, Santa Barbara County, CA.

mitigating the impacts presented by post-fire flooding and debris flows. Infrastructure design and emergency management protocols that are informed by past events will reduce the threat to life, property, and infrastructure in downstream communities. Addition-

ally, the information presented here can provide input for modeling efforts that will help improve future rapid federal and state post-fire assessments and inform future research on post-fire hazards, including longitudinal studies that look at meteorological

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

425

Schwartz, Oakley, and Alessio

triggering events, basin characteristics, soil coverage, and vegetation types across a wide range of events and rock types (e.g., Oakley et al., 2017; Staley et al., 2017; and Dibiase and Lamb, 2020). Post-Fire Debris Flows Wildfires dramatically change the hydrological response of watersheds. Higher runoff potential and vegetation loss increase the susceptibility to erosion, flooding, and debris flows (Cannon et al., 2003a, 2003b; Cannon and De Graff, 2009; Kean et al., 2011; De Graff et al., 2015; and Staley et al., 2020). The increase in the hydrological response and susceptibility to erosion is partially related to the soil burn severity (SBS). SBS is classified based on the degree of physical and biological changes to soil surface characteristics, such as char depth, organic matter loss, altered color and structure, and reduced infiltration (Ryan and Noste, 1985; DeBano et al., 1998; Lentile et al., 2006; and Parsons et al., 2010). Wilder et al. (2020) found that other important factors influencing post-fire peak flows include peak rainfall intensity, watershed size, total burned area, and time after fire containment. Burned watersheds with steep slopes and first order channels that contain significant volumes of stored sediment are likely to experience increases in runoff and erosion from a lack of protective vegetation cover, soil hydrophobicity, and loss in cohesive root strength, which provide the potential to generate debris flows (Kean et al., 2011; Parise and Cannon, 2012; and Kean et al., 2019). In semi-arid landscapes like the southwestern United States, runoff-generated debris flows are the most common and initiate as result of progressive bulking or accumulation of slurry in stream channels (Cannon, 2000, 2001; Cannon et al., 2001a). There are many different mechanisms by which slurry is generated, such as intense rilling on hillslopes, (Meyer and Wells, 1997; Cannon et al., 2001b), saturation and failure of channel bed sediment (Kean et al., 2013), and/or mobilization and mixing of dry ravel accumulations in first-order channels (Dibiase and Lamb, 2020). The increase in slurry production from enhanced runoff is also related to the proportion of the burned area at high and moderate SBS. Runoff generated slurry typically has high sediment concentrations (40–65 percent) and can scour colluvial and fluvial stream deposits. The flow can then progressively grow in size as it moves downstream by recruiting boulders and woody debris, resulting in destructive debris flows (Iverson, 1997). As debris flows progress down mountain channels at speeds of 30–50 km/h boulders, woody debris, and saturated materials impact drainage infrastructure by clogging culverts, bridges, and underpasses, which can result in

426

a diversion (avulsion) of the flow and destruction of infrastructure built across and along its banks (e.g., Kean et al., 2019; Lukashov et al., 2019; and Lancaster et al., 2020). Other post-fire hydrologic processes, such as hyper-concentrated flooding and debris flows, threaten life, property, and infrastructure. They can destroy houses, block or carve out sections of roads and cause transportation impacts, sever pipelines and damage utilities that cause business disruptions, and add large quantities of sediment to stream channels that impact water resources. Event Setting The Sherpa Fire was ignited on 15 June 2016 in the Los Padres National Forest, Santa Barbara County, California, during a strong sundowner wind event. With the strong northerly winds, the fire spread rapidly southward and downslope, propagating over 6 km in 24 hours (Smith et al., 2018). The fire burned a total of 30.2 km², out of which 10.5 km² burned on National Forest lands, 12.6 km² burned on State lands, and 7.4 km² burned on private lands. Within the fire perimeter, 4 percent of the area burned at a high SBS, 60 percent burned at a moderate SBS, 28 percent burned at a low SBS, and 8 percent was either unburned or burned at a very low SBS (Schwartz, 2016) (Figure 2). The Sherpa Fire occurred on the south slopes of the Santa Ynez Mountains within the east–west-oriented Transverse Ranges of Southern California. The Santa Ynez Mountains parallel the south coast of Santa Barbara County and extend eastward into Ventura County. The Transverse Ranges are some of the most tectonically active mountains in the United States, and with uplift rates of 1–2 mm/yr, they are growing faster than they are eroding (Dibble, 1982; Melosh and Keller, 2013). Unlike the San Gabriel and San Bernardino mountains to the east, this section of the Transverse Ranges is composed almost entirely of un-metamorphosed, mostly marine sedimentary rocks (Figure 1) of Cenozoic age, where shales have a continuous soil mantle and sandstones have low–moderate colluvial coverage (Dibblee, 1988; Keller et al., 2015). Soils in this area are typically shallow and rocky, containing variable soil textures based on age and parent material. Vegetation in the Sherpa Fire burn area is dominated by chaparral with oak woodlands. Conifers exist in small patches along ridgetops and on northfacing slopes. At the bottom of the drainages, narrow riparian corridors contrast sharply with the otherwise dry landscape. Based on U.S. Forest Service fire history data, the last fire that burned this footprint was the Refugio Fire in September 1955. The physiography of the area that was impacted by the

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

Debris Flow Assessment, El Capitan Watershed, Santa Barbara County, California

Sherpa Fire is dominated by extremely steep and rugged slopes (30°–45°) with elevations ranging from 60 m above mean sea level (AMSL) by US Highway 101 (Hwy 101) to approximately 760 m AMSL at the northern boundary of the fire. The burned area is drained by five major creeks flowing south to the Pacific Ocean: Cañada del Refugio, Cañada del Venadito, and Las Flores Canyon, which flows into Cañada del Corral and Cañada del Capitan. In the steep, geologically young Santa Ynez Mountains, various forms of storm-driven landsliding, rockfalls, and surface erosion are frequent and occur in unburned settings as well (Alessio, 2019). The most significant impacts of the 20 January 2017 debris flow event were observed in the Cañada del Capitan watershed. This watershed is situated at the east end of the area burned by the Sherpa Fire, adjacent to and east of the Cañada del Corral watershed. The watersheds impacted by the Sherpa Fire are underlain by alternating sedimentary rock formations of shales and sandstones with some conglomerate, ranging in age from late Eocene to late Miocene, and overlain by Quaternary alluvial, landslide rubble, and surficial sediments (Figure 1); strata of older rock types typically dips (inclines) steeply toward the south or southwest (Dibblee, 1988). Since this area is experiencing rapid uplift, the upper parts of the watershed present deeply incised canyons, which tend to transport high-energy flows, creating steep canyon walls and producing rocky colluvial slopes and stream channel deposits of gravel, sand, and silt. The area drained by the Cañada del Capitan watershed is 15.8 km², out of which 6.42 km² (41 percent) were burned in the Sherpa Fire. Of the area burned, 3.6 percent (0.23 km²) burned at high SBS and 67.9 percent (4.36 km²) burned at moderate SBS, amounting to 71.5 percent of moderate and high SBS (Figure 2). During the post-fire Burned Area Emergency Response (BAER) assessment conducted by the U.S. Forest Service, a majority of the soils were identified as soil Hydrologic Group D, indicating soils that are shallow and prone to runoff, especially when vegetation is removed. In addition, hydrophobicity testing in numerous plots revealed subsurface hydrophobic layers 2–5 mm thick with moderate to strong hydrophobic severity, lasting 1–5 minutes (Nicita, 2016). El Capitan State Beach and State Park are located at the mouth of this watershed on the south side of Hwy 101. The El Capitan Canyon Resort is situated above the State Beach, on the north side of Hwy 101, along the floodplain of El Capitan Creek. At the time of the 20 January 2017 post-fire debris flow event, the resort consisted of approximately 160 cabins, yurts, and tent sites, all surrounded by El Capitan State Park lands. To the north and above the El

Capitan State Park are the Los Padres National Forest lands. Based on pre-flooding aerial and ground surveys, it was evident that in the past, mass wasting has occurred in areas dominated by the Sespe Sandstone, Gaviota Sandstone, and Sacate Shale Formations, loading stream channels with boulders, loose debris, and debris flow deposits in the mountain channels. Based on reports from El Capitan Canyon Resort personnel and State Park staff, the unnamed road that traverses through El Capitan Canyon Resort and into the El Capitan State Park lands has been historically impacted multiple times by mass wasting as rockfall and debris slides, even in the absence of wildfires. Meteorological Triggering Event for Debris Flow On the morning of 20 January 2017, a storm featuring a weak-to-moderate atmospheric river (Figure 3a) produced rainfall in the Santa Ynez Mountains. This was a very dynamically active and complex storm resulting in the development of several narrow bands of intense rainfall in California, one of which moved over the Sherpa Fire burned area and was enhanced by orographic forcing along the Santa Ynez Mountains (Figure 3b). Similar intense bands of rainfall have previously been associated with post-wildfire debris flows in Southern California (Oakley et al. 2017, 2018). Rainfall in this storm event began at the Refugio Pass rain gage (Figure 1), which best represents the upper El Capitan watershed, at 0:45 a.m. PST. By 11:55 a.m. PST, 98 mm of rain had fallen, with a maximum 1-hour rainfall accumulation of 52 mm occurring between 8:10 and 9:10 a.m. PST and a peak 15-minute rainfall accumulation of 19 mm occurring between 8:53 and 9:08 a.m. PST, representing a 76 mm/hr rate (Figure 4). Based on NOAA Atlas 14 (NOAA, 2016) using the methods described by Staley et al. (2020), the peak 15-minute accumulation was calculated to an average precipitation return interval frequency of 15.8 years. Kean et al. (2011) and Staley et al. (2013) found that rainfall intensities measured over durations of 60 minutes or less are best correlated with post-fire debris flow initiation. Moreover, the 15-minute duration provides the most accurate prediction of post-fire debris flow generation (Staley et al., 2017). This triggering event was the second rainfall event in 2 days at this location (Figure 5), following the 19 January rainstorm event that produced 32 mm over a period of 2:45 hours at the Refugio Pass rain gage. The 20 January rainstorm event started 22 hours after the 19 January rainstorm ended. The average annual rainfall (1958–2020) at the Refugio Pass rain gage is 711 mm/yr. From the start of Water Year 2017 (October 1), this rain gage had received 470 mm of

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

427

Schwartz, Oakley, and Alessio

Figure 3. (a) Integrated water vapor transport (IVT; color ﬁll and vectors) and sea level pressure (black contours) for 1000 PST 20 January 2017. Darker/redder colors indicate greater IVT, a measure combining total atmospheric water vapor and wind strength. Data are from the ERA5 atmospheric reanalysis (C3S 2017). (b) Radar reﬂectivity (dBZ) at 9:15 PST shows intense rainfall (>50 dBZ) over the Sherpa Fire burn area. Intense rain is indicated by yellow-to-red colors and less intense rain is indicated by blue-to-green colors.

precipitation, 66.1 percent of the annual average. Thus, the soil and channel bed moisture were likely elevated, potentially contributing to an increase in surface runoﬀ and mobility of materials (Iverson et al., 2011; Reid et al., 2011; and Kean et al., 2013). OBSERVATIONS AND DATA COLLECTION This investigation is based on ﬁeld observations and data collected during and immediately after the

Sherpa Fire, as well as multiple post-flooding field assessments. Field observations and data collected during and immediately after the fire included ground surveys, flight reconnaissance, and aerial photography throughout the burned area. Post-flooding and debris flow event data collection began on 24 January 2017, after the third consecutive rainstorm in 4 days, and was focused on the El Capitan drainage inundation zone. Field work included documentation and measurements of the debris flow inundation extent,

Figure 4. Rainfall accumulation (blue line) and 15-minute intensities (green line) for the storm on 20 January 2017. The peak 15-minute intensity was 76 mm/hr.

428

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

Debris Flow Assessment, El Capitan Watershed, Santa Barbara County, California

Figure 5. Cumulative rainfall accumulations for the three rainfall events that occurred over a 4-day period, 19–23 January 2017 at Refugio Pass Raingage Station. The El Capitan debris ﬂows occurred during the second consecutive storm on 20 January 2017.

debris flow deposit depth, the type and nature of sediment delivered by the debris flow, scour depths, and the extent of damage caused by the debris flow. In addition to field assessments, observations are supported by eyewitness accounts of the debris flow and flooding events, including photos and videos, which supported reconstruction of these events.

El Capitan Drainage El Capitan Debris Flow Source Area Since this post-debris flow investigation focused on the inundation zone, ground assessment and documentation were minimal in the source area. In the limited post-flooding field assessment of the upper portions of the impacted watersheds, rill development and surface erosion were noted on the slopes above the El Capitan Creek and its tributaries channels, but no detailed measurements were taken to document the extent of surface erosion and quantity material delivered from the source area. On 9 July 2017, 6 months after this flooding and debris flow event, the Whittier Fire ignited and burned on both the north and south facing slopes of the Santa Ynez Mountains, adjacent to and overlapping with portions of the Sherpa Fire. Aerial flight reconnaissance conducted during the BAER assessment of the Whittier Fire revealed widespread rilling and surface erosion throughout the steep burned slopes, especially in the shale units. Aerial photography shows the post-fire soil conditions for shale and sandstone units in the Whittier Fire burn area and reveal that many gullies and first-order channels in the Sherpa Fire burn area had been scoured to bedrock (Figure 6).

El Capitan Debris Flow Inundation Zone The inundation zone associated with the El Capitan debris flow occurred along the El Capitan Creek and floodplain, starting approximately 2 km above and north of Hwy 101 and extending to the ocean. The channel gradients in the upper parts of the watershed where debris flow material was transported ranges from 30° to 10° and decreases to 2°–3° by the upper end of the El Capitan Canyon Resort where debris flow deposits were identified. Field observations along the stretch of El Capitan Creek between the State Park boundary (upper end of the El Capitan Canyon Resort) and the Mid-Canyon Bridge (Figure 7) revealed that some segments of the stream experienced channel incision and scour depths ranging from 1.2 to 1.5 m. Other segments of the creek, mostly below the Mid-Canyon Bridge, accrued boulders and finer sediment deposition up to 4 m in depth. Lateral bank erosion occurred mostly on the outer banks at channel bends, exposing unsorted, nonstratified, matrix-supported alluvial channel and older debris flow deposits (Figure 8a). Since the field observation for this study began 24 January 2017 after the third consecutive rainstorm in a 4-day period (Figure 5), it is difficult to determine the exact time the scour and deposition took place along these segments of the stream channel. Along segments of the creek that experienced deposition, elongated boulder fields ranging in size from ∼60 m to 140 m long, 10 m to 20 m wide were documented in the channel and were comprised of sub-angular to sub-rounded sandstone boulders ranging from 0.3 m to more than 2.5 m in diameter (Figure 9). Based on eyewitness reports, the debris flow reached the El Capitan Canyon Resort at approximately

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

429

Schwartz, Oakley, and Alessio

Figure 6. (a) Burned areas of the Sherpa and Whittier fires. This photo was taken a year after the Sherpa Fire, 6 months after the rainstorm and debris flow events, and 2 weeks after the Whittier Fire. Gullies and first-order channels in the Sherpa Fire burn area were scoured to bedrock (photo taken by Kevin Cooper, USFS). (b) Shale hillslopes mantled with thick, continuous fine-grained soil, ash, and charred organic matter. Vegetation is completely incinerated. (c) Sandstone hillslopes composed of thin, patchy soils; bedrock outcrop; ash; charred organic matter; and silt-rich colluvium with grain sizes up to boulders.

9:20 a.m. PST, about 13 minutes after the period of highest 15-minute rainfall intensity. Field evidence and eyewitness reports suggest the debris flow surge front included coarse woody debris that plugged foot and vehicle bridges, culverts, and the Hwy 101 underpass, similar to the 9 January 2018 Montecito debris flows that followed the 2017 Thomas Fire. This coarse woody debris included burned tree limbs and other vegetation transported down the creek from the upper portions of the watershed as well as riparian vegetation that survived the fire but was stripped and transported downstream by the powerful flooding and debris flow. The first bridge to be impacted by the debris flow was the Mid-Canyon Bridge, a large wood deck bridge, approximately 5 m wide and 15 m long. This bridge was supported on both sides of the creek by concrete abutments (Figure 10a). Attached to the top of these abutments and supporting a wooden deck were three metal I-beams 45.72 cm high, 19.05 cm wide, 0.95 cm thick, and 15 m long. The debris flow surge front demolished the bridge, knocking it off its abutments. The metal I-

430

beams were carried downstream and twisted around large standing trees (Figure 10b). Similar to observations made by Lancaster et al. (2020) in the 9 January 2018 Montecito debris flow, overbank flows and avulsions occurred where the overall flow height exceeded the capacity of the channel bank or at locations where channel constrictions as bridges, bridge abutments, or culverts were present. Ground assessments revealed evidence of dispersed overbank flows as well as debris flow inundation characteristics. At locations where dispersed overbank flows took place, such as below the Mid-Canyon Bridge, high watermarks revealed evidence of up to 1.2 m of flooding that impacted the floodplain above the bankfull discharge. In some locations this flooding impacted an area extending 30 m up each side of the channel. Maximum debris flow inundation depth was determined from mud patinas on structures and tree trunks, broken tree limbs, and fresh scarring on the upstream side of tree trunks. Some structures located on the floodplain were marked on their upstream side by

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

Debris Flow Assessment, El Capitan Watershed, Santa Barbara County, California

Figure 7. A map showing the El Capitan Canyon Resort and El Capitan State Beach locations with scale.

mud patinas to heights of up to 2.5 m above the bankfull discharge (Figure 10c). Similarly, along the channel bed, large oak and sycamore trees surviving the ﬂow path were marked by mud patinas and scarring on their upstream side to heights of 2.5 m above bank-

full discharge. Boulder deposition in the floodplain consisted of isolated rocks with few boulder fields. In the area below the Mid-Canyon Bridge where avulsions took place, boulder and other debris impacted many cabins located on the floodplain, causing varying

Figure 8. (a) Lateral bank erosion occurred mostly on the outer banks at channel bends, exposing unsorted, non-stratified, matrix-supported, paleo debris flow deposits. (b) Paleo debris flow levees composed of weathered sub-angular boulders were observed on the lateral margins of the active channel in El Capitan Creek.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

431

Schwartz, Oakley, and Alessio

Figure 9. (a) Large elongated boulder ﬁelds that were deposited in the active channel. Boulder ﬁelds ranged in size from ∼60 m to 140 m long and ∼10 m to 20 m wide, and contained (b) sub-angular to sub-rounded sandstone boulders ranging in size from 0.3 m to more than 2.5 m in diameter.

levels of damage (Figure 10d). Nine cabins were forced oﬀ their foundations and swept downstream and as many as 21 vehicles were carried away and crushed, some reaching the ocean.

As the debris ﬂow surge front reached this lower area of the El Capitan Canyon Resort, large amounts of coarse woody debris plugged the foot bridge leading to the Canyon Market and the El Capitan Canyon

Figure 10. (a) Damage to the Mid-Canyon Bridge, supported on both sides of the creek by concrete abutments. (b) Twisted metal I-beams that supported the Mid-Canyon Bridge prior to the debris ﬂow event. (c) Structure on the ﬂood plain marked on its up-stream side by mud patinas to a height of up to 2.5 m. (d) Damaged cabins at the El Capitan Canyon Resort.

432

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

Debris Flow Assessment, El Capitan Watershed, Santa Barbara County, California

Figure 11. (a) Large soil-surfaced parking lot near the entrance of the El Capitan Canyon Resort turned into a debris catchment basin as debris flows and flooding impacted the El Capitan watershed and plugged the Hwy 101 underpass. (b) Once the Hwy 101 underpass was clear of debris, the high level of water in the soil-surfaced parking lot area subsided and much of the debris (including cars and cabins) was deposited in the parking lot. (c) In total, four cars, one truck, and one tractor were transported out to the ocean at the mouth of El Capitan Creek. (d) In the Canada Del Corral watershed, woody debris clogged the Hwy 101 underpass, which created a backup that caused mud and boulders to flow overbank onto the floodplain, resulting in damaged historic adobe structures located north of Hwy 101.

Resort Entrée Bridge. As these bridges were completely plugged, the flood was diverted out of the channel and into a large soil-surfaced parking lot (approximately 100 m by 60 m) located below and to the west of the El Capitan Canyon Resort Entrée Bridge and adjacent to Hwy 101 (Figure 7). As debris and flows avulsed the channel, flooding the parking lot area, flows continued toward the Hwy 101 underpass. At this stage, in addition to large amounts of woody debris, mud, rocks and boulders, the debris included cars and cabins that were swept away by the flows. As a result of debris plugging the Hwy 101 concrete underpass (3.6 m wide × 4.6 m high), the large soil-surfaced parking lot turned into a debris catchment basin (Figure 11a). Based on the size of the parking lot and depth of debris estimated from photo documentation of the parking lot during the peak of the event, it is estimated

that the volume of debris retained in this “temporary debris basin” ranged from 12,000 m³ to 16,000 m³ between 9:20 and 9:45 a.m. PST. Based on eyewitness reports, at ∼9:45 a.m. PST, when the pressure was high enough, the debris plugging the Hwy 101 underpass broke through and large amounts of debris (including four cars, one truck, and one tractor) were transported out to the ocean (Figure 11c). The estimated distances these vehicles were transported by the event ranged from 1 to2 km depending on their starting point within the resort. Once the Hwy 101 underpass was clear of debris, the water level in the dirt parking lot subsided, depositing much of the debris (including cars and cabins) at this location (Figure 11b). According to Santa Barbara County Fire Department oﬃcials, no fatalities or injuries occurred in this event, although 22 people were

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

433

Schwartz, Oakley, and Alessio

trapped in cabins in the resort and required immediate rescue from first responders. El Capitan Canyon State Beach El Capitan Creek flows under Hwy 101 and continues ∼0.75 km before reaching its mouth at El Capitan Beach. On its way to the ocean, the creek flows under a large arch culvert (4.5 m wide × 3.6 m high) located under the entrance road to the State Beach (Figure 7). Reports from staff occupying the State Beach entrance kiosk indicate that at approximately 9:45 a.m. PST, as the surge front of debris came down the creek, large amounts of debris plugged the arch culvert, causing the flows to avulse and divert down the road toward the kiosk. This diversion of the flow persisted for approximately 10 to 15 minutes. Once pressure on the debris plugging the culvert broke through, flows and debris resumed their natural flow path down the creek to the ocean. Cañada Del Corral and Las Flores Canyon From eyewitness reports and security camera videos at the Exxon-Mobile oil and gas facility located at the confluence of Cañada Del Corral and Las Flores Canyon, it appears that at the same time El Capitan Creek was impacted by flooding and debris flows, Cañada Del Corral and Las Flores Canyon were experiencing similar events. Video footage demonstrates that the event impacting these drainages was primarily sediment-laden flooding with large amounts of woody debris and smaller quantities of large boulders as compared to the event in El Capitan Creek. The surge front of this sediment-laden flooding reached the oil and gas facility at 9:20 a.m. PST and continued for approximately 25–30 minutes. By 9:50 a.m. PST the high flows in both drainages subsided. As in El Capitan Creek, once the woody debris flowing down Cañada Del Corral reached the Hwy 101 underpass, it clogged the underpass and inundated the floodplain, damaging historic adobe structures located north of Hwy 101 (Figure 11d). DISCUSSION The 20 January 2017 flooding and debris flow events that impacted the Cañada del Capitan, Cañada Del Corral, and Las Flores Canyons originated in recently burned watersheds with 64 percent moderate and high SBS, as observed during a post-fire assessment conducted by the U.S. Forest Service BAER Team. The burn area had steep, colluvial-mantled slopes and channels loaded with unsorted, unconsolidated materials, including colluvium and older de-

434

bris flow deposits, available to be transported(Figure 8b). The parent materials of these watersheds are, for the most part, alternating sandstone and shale units, where shale units provide fine-grained colluvium for slurry production and sandstone units provide coarser colluvium and boulders (Figure 6b and c). In addition, immediately after the Sherpa Fire, ground surveys revealed widespread dry ravel further loading the channels (Schwartz, 2016). Analysis of aerial photography taken from a flight reconnaissance conducted during the BAER assessment of the Whittier Fire (9 July 2017) revealed surface erosion in the form of widespread rilling and sheetwash throughout the steep burned slopes of the Sherpa Fire, especially in the shale units, and first-order channels were scoured to bedrock (Figure 6a). Based on the steep terrain, the underlying geological units, the extensive moderate and high SBS, and the high-intensity rainstorm that triggered this event, we concluded that the steep unstable slopes and channels in the upper portions of these watersheds functioned as major sediment sources for the 20 January 2017 El Capitan flooding and debris flow event. Observations compiled from the 19–22 January 2017 post-fire storm events indicate that three distinct flow processes occurred in El Capitan Creek throughout these storm events. The three basic flow processes include water flows, hyper-concentrated flows, and debris flows. These flow processes represent a continuum where boundaries between the flow types are not sharp, such that any one flow event may exhibit different flow types at different times and points along the flow path (Pierson, 2005; Wagner et al., 2012). Eyewitness accounts and ground-based observations suggest that the surge front that caused most of the damage in the El Capitan Canyon Resort during the 20 January 2017 storm was related to a debris flow. The velocity and concentrated nature of these flows caused overbanking and channel avulsions that resulted in damaged roads, bridges, pipelines, infrastructure, and cabins in the El Capitan Canyon Resort. Based on observations and analysis done by the U.S. Forest Service BAER Team following the Sherpa Fire, this type of damage was predicted (Schwartz, 2016). Furthermore, in their final report, the U.S. Forest Service BAER Team recommended that during rainstorm events meeting or exceeding 28 mm/hr in the first wet season following the fire, the El Capitan Canyon Resort and El Capitan State Beach should be evacuated. The physical setting and soil conditions of the source area for the 20 January 2017 El Capitan flooding and debris flow events are similar to those that led to the 9 January 2018 debris flows in Montecito, California. In Montecito, large, channel-clearing debris flows were generated ∼3 weeks after the Thomas

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

Debris Flow Assessment, El Capitan Watershed, Santa Barbara County, California

Fire that burned in the Santa Ynez Mountains of southern California (Kean et al., 2019). The debris flows were triggered by a high-intensity rainstorm (Oakley et al., 2018) that extensively stripped soil from hillslopes via rilling and scoured boulders and paleodebris flow deposits from six adjacent mountain catchments, which significantly impacted the community below (Kean et al., 2019). The economic impact from commercial and private property damage alone was ∼$400 million, not including lost wages and impacts from the Hwy 101 closure (RDN, Inc., 2018). In both settings, the source area included steep south-facing slopes along the Santa Ynez Range. In both locations, the underlying geological units consist, for the most part, of alternating sandstone and shale units (Dibble, 1982), and the source area for the debris flows experienced moderate and high SBS (Young et al., 2018), dry ravel, and rilling of hillslopes. Additionally, the storms that initiated these debris flows were shortduration, high-intensity rainstorms in the form of a north–south-oriented narrow band. Though both events had a similar setup, there are several major differences between the El Capitan and Montecito events that help to explain the differences in their magnitude and impacts. First, the peak 5-minute and 15-minute rainfall intensities were much higher in the Montecito event. For the 20 January 2017 El Capitan debris flows, the peak 5-minute rainfall intensity was 84 mm/hr and the peak 15-minute rainfall intensity was 76 mm/hr, recorded at the Refugio Pass gauge. In the 9 January 2018 Montecito event, the peak 5-minute rainfall intensity was 180 mm/hr and the peak 15-minute rainfall intensity was 106 mm/hr at the Doulton Tunnel ALERT gauge. During the 9 January 2018 event, the most intense rainfall was focused on the eastern half of Santa Barbara County. The Sherpa Fire burn area received much less intense rainfall than the Thomas Fire burn area, with a peak 15-minute rainfall intensity of 20 mm/hr at the Refugio Pass gauge. No impactful hydrologic response was reported on the Sherpa Fire burn area in this event. Second, Montecito had a greater fraction and total area of the watersheds that burned at moderate and high SBS. In the El Capitan, only 41 percent (6.42 km2 ) of the watershed was burned by the Sherpa Fire and the fire did not extend up to the ridge crest, whereas over 90 percent of the area above Montecito and Carpinteria was burned by the Thomas Fire and extended up to the ridge crest. Lastly, the amount and type of development below the mountain front was distinct between the two events. The El Capitan debris flow impacted mostly “soft” structures consisting of about 160 cabins, yurts, and tent sites. In contrast, for the Montecito debris flows, the inundation zone was

largely urbanized with houses, paved roads, and essential infrastructure. Additional key differences between the two events also exist in the physical setting of the inundation zone. In the case of the El Capitan event, the inundation zone impacted by the debris flow and flooding event consists of a narrow floodplain, approximately 100–150 m wide and 2 km long, situated along the El Capitan Creek valley floor. In Montecito, the inundation zone consisted of steeply sloping wide debris flow and alluvial fans, bisected by a fault. The total area inundated in Carpinteria and Montecito in the 9 January 2018 event was 5.56 km² (Lancaster et al., 2020), whereas the area inundated in the 2017 El Capitan event was ∼0.12 km². CONCLUSIONS The Montecito and El Capitan events occurred 1 year apart and ∼35 km away from one another, with similar physical and climatological settings. Observations presented here in the El Capitan Creek and adjacent watersheds, as well as prior research in other drainages in the Santa Ynez Mountains, demonstrate that hillslopes have ample colluvium to produce slurry from rilling and load channels with dry ravel, channels have ample boulders and coarse material to be mobilized by slurry, and the lithology (alternating sandstone and shale units) produces a favorable setup for post-wildfire debris flow generation. A comparison of the El Capitan and the Montecito debris flow events highlights the potential for future destructive post-wildfire debris flows in this region, especially in a warming climate with projected increases in wildfire activity and precipitation intensity. Information from case studies such as this support decision makers and emergency managers in understanding the hazards and risks that floods and debris flows pose on communities below steep, recently burned drainages and inform the development of sound protocols to reduce the threat to life, property, and infrastructure in downstream communities. ACKNOWLEDGMENTS We would like to start by acknowledging Jerome (Jerry) De Graff who passed away this last year. Over the years, Jerry was a colleague, a mentor, and a friend who shared his knowledge and experience with us, helping us better understand landform processes. We would like to thank Dennis Staley (USGS) for reviewing this manuscript and providing feedback. We would also like to thank Shawn Johnson (Santa Barbara County, Water Resources Division, Flood Control District) for providing precipitation data in

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

435

Schwartz, Oakley, and Alessio

addition to some technical help, Patrick Lieske (USFS) for help with data collection, and David Patterson and Marilyn Porter (USFS) for help with GIS. REFERENCES Alessio, P., 2019, Spatial variability of saturated hydraulic conductivity and measurement-based intensity duration thresholds for slope stability, Santa Ynez Valley, CA: Geomorphology, Vol. 342, pp. 103–116. Barbero, R.; Abatzoglou, J. T.; Larkin, N. K.; Kolden, C. A.; and Stocks, B., 2015, Climate change presents increased potential for very large fires in the contiguous United States: International Journal Wildland Fire, Vol. 24, No. 7, pp. 892–899. Cannon, S. H., 2000, Debris-flow response of southern California watersheds recently burned by wildfire. In Wieczorek, G. F., and Naeser, N. D. (Editors), Debris-Flow Hazards Mitigation, Mechanics, Prediction, and Assessment, Proceedings of the Second International Conference on Debris-Flow Hazards Mitigation, Taipei, Taiwan, 16–18 August 2000: A. A. Balkema, Rotterdam, Netherlands, pp. 45–52. Cannon, S. H., 2001, Debris-flow generation from recently burned watersheds: Environmental Engineering Geoscience, Vol. 7, pp. 321–341. Cannon, S. H.; Bigio, E. R.; and Mine, E., 2001a, A process for fire-related debris flow initiation, Cerro Grande fire, New Mexico: Hydrological Processes, Vol. 15, No. 15, pp. 3011–3023. Cannon, S. H. and De Graff, J., 2009, The increasing wildfire and post-fire debris-flows threat in Western USA, and implications for consequences of climate change. In Sassa, K. and Canuti, P. (Editors), Landslides – Disaster Risk Reduction: Springer, Berlin, Germany, pp. 177–190. Cannon, S., J. Gartner, C. Parret, and M. Parise (2003), Wildfirerelated debris-flow generation through episodic progressive sediment bulking processes, western USA, in Debris-Flow Hazards Mitigation: Mechanics, Prediction, and Assessment, Proceedings of the Third International Conference on DebrisFlow Hazards Mitigation, pp. 71–82, Mill press, Rotterdam, Netherlands. Cannon, S. H.; Gartner, J. E.; Rupert, M. G.; and Michael, J. A., 2003b, Emergency Assessment of Debris-Flow Hazards from Basins Burned by the Piru, Simi, and Verdale Fires of 2003, Southern California. U.S. Geological Survey Open-File Report OF-03-481. Cannon, S. H.; Kirkham, R. M.; and Parise, M., 2001b, Wildfirerelated debris-flow initiation processes, Storm King Mountain, Colorado: Geomorphology, Vol. 39, No. 3–4, pp. 171–188. DeBano, L. F.; Neary, D.; and Folliott, P., 1998, Fire’s Effects on Ecosystems: John Wiley and Sons Inc., New York, 333 p. De Graff, J.; Cannon, S. H.; and Gartner, J. E., 2015, The timing of susceptibility to post-fire debris flows in Western United States: Environmental Engineering Geoscience, Vol. 21, No. 4, pp. 277–292. http://doi.org/10.2113/gseegeosci.21.4.277. Dibblee, T. W., Jr., 1982, Geology of the Santa Ynez - Topatopa Mountains, Southern California. In Fife, D. L., and Minch, J. A., (Editors), Geology and Minerals Wealth of the California Transverse Ranges: Mason Hill Volume: Santa Ana, California: South Coast Geological Society, pp. 41–56. Dibblee, T. W., Jr., 1988, Geology Map of the Santa Ynez / Tajiguas Quadrangles, Santa Barbara County, California: Helmut, E. (Editor), Published by the Thomas W. Dibblee, Jr. Geological Foundation, Map scale 1:24,000.

436

DiBiase, R. A. and Lamb M. P., 2020, Dry sediment loading of headwater channels fuels post-wildfire debris flows in bedrock landscapes: Geology, Vol. 48, No. 2, pp. 189–193. Iverson, R. M., 1997, The physics of debris flows: Reviews Geophysics, Vol. 35, pp. 245–296. Iverson, R. M.; Reid, M. E.; Logan, M.; LaHusen, R. G.; Godt, J. W.; and Griswold, J. P., 2011, Positive feedback and momentum growth during debris-flow entrainment of wet bed sediment: Nature Geoscience, Vol. 4, No, 2, pp. 116–121, available at http://www.nature.com/ngeo/journal/ v4/n2/abs/ngeo1040.html#supplementary-information Kean, J. W.; Staley, D. M.; and Cannon, S. H., 2011, In situ measurements of post-fire debris flows in southern California and soil moisture conditions: Journal Geophysical Research, Vol. 116, No. F4, F04019. http://doi.org/10.1029/2011jf002005. Kean, J. W.; Staley, D. M.; Lancaster, J. T.; Rangers, F. K.; Swanson, B. J.; Coe, J. A.; Hernandez, J. I.; Sigman, A. J.; Allstadt, K. E.; and Lindsay, D. N., 2019, Inundation, flow dynamics, and damage in the 9 January 2018 Montecito debris-flow event, California, USA; opportunities and challenges for post-wildfire risk assessment: Geosphere, Vol. 15, No. 4, pp. 1140–1163. https://doi.org/10.1130/ges02048.1. Kean, J. W.; Tucker, G. E.; Staley, D. M.; and Coe, J. A., 2013, Runoff-generated debris flows: Observations and modeling of surge initiation, magnitude and frequency: Journal Geophysical Research: Earth Surfaces, 2013JF002796. Keller, E. A.; Bean, G.; and Best, D., 2015, Fluvial geomorphology of a boulder-bed, debris-flow—Dominated channel in an active tectonic environment: Geomorphology, Vol. 243, pp. 14–26. Lancaster, J. T.; Swanson, B. J.; Lukashov, S. G.; Oakley, N. S.; Lee, J. B.; Spangler, E. R.; Hernandez, J. L.; Olson, B. P. E.; Defrisco, M. J.; Lindsay, D. N.; Schwartz, J. Y.; Mccrea, S. E.; Roffers, P. D.; and Tran, C. M., 2020, Observations and analysis of the 9 January 2018 debris-flow disaster, Santa Barbara County, California: Environmental Engineering Geoscience, Vol. XXVI, No. 4, pp. 1–25. Lentile, L. B.; Holden, Z. A.; Smith, A. M. S.; Falkowski, M. J.; Hudak, A. T.; Morgan, P.; Lewis, S. A.; Gessler, P. E.; and Benson, N. C., 2006, Remote sensing techniques to assess active fire characteristics and post-fire effects: International Journal of Wildland Fire, Vol. 15, pp. 319–345. Lukashov, S. G.; Lancaster, J. T.; Oakley, N. S.; and Swanson, B. J., 2019, Post fire debris flows of 9 January 2018, Thomas Fire, southern California: Initiation areas, precipitation and impacts. In Kean, J. W.; Coe, J. A.; Santi, P. M.; and Guillen, B. K. (Editors), Debris-Flow Hazards Mitigation: Mechanics, Monitoring, Modeling, and Assessment: Proceedings of the Seventh International Conference on DebrisFlow Hazards Mitigation, June 10–13, 2019: Association of Environmental and Engineering, Geologists, Golden, CO, pp. 774–781. Melosh, B. L. and Edward A. Keller, E. A., 2013, Effects of active folding and reverse faulting on stream channel evolution, Santa Barbara Fold Belt, California: Geomorphology, Vol. 186, pp. 119–135. Meyer, G. A. and Wells, S. G., 1997, Fire-related sedimentation events on alluvial fans, Yellowstone National Park, USA: Journal Sedimentary Research, Vol. 67, No. 5, 776–791. Nicita, E., 2016, Sherpa Fire - Soil Resource Assessment - Burned Area Emergency Response (BAER) Report. USDA Forest Service internal report, 11 p. NOAA Atlas 14, 2016, Hydrometeorological Designs Study Center Precipitation Frequency Data Server (PFDS): available at http://hdsc.nws.noaa.gov/hdsc/pfds/index.html

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

Debris Flow Assessment, El Capitan Watershed, Santa Barbara County, California Oakley, N. S.; Cannon, F.; Munroe, R.; Lancaster, J. T.; Gomberg, D.; and Ralph, F. M., 2018, Brief communication: Meteorological and climatological conditions associated with the 9 January 2018 post-fire debris flows in Montecito and Carpinteria, California, USA: Natural Hazards Earth System Sciences, Vol. 18, No. 11, 3037–3043. Oakley, N. S.; Lancaster, J. T.; Kaplan, M. L.; and Ralph, F. M., 2017, Synoptic conditions associated with cool season post-fire debris flows in the Transverse Ranges of southern California: Natural Hazards, Vol. 88, No. 1, pp. 327–354. Parise, M. and Cannon, S. H., 2012, Wildfire impacts on the processes that generate debris flows in burned watersheds: Natural Hazards, Vol. 61, No. 1, pp. 217–227. Parsons, A.; Robichaud, P.; Lewis, S.; Napper, C.; and Clark, J. T., 2010, Field Guide for Mapping Post-fire Soil Burn Severity: Gen. Tech. Rep. RMRS-GTR-243, Fort Collins, CO, U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station. 49 p. Pierson, T. C., 2005, Distinguishing between Debris Flows and Floods from Field Evidence in Small Watersheds: U.S. Geological Survey Fact Sheet 2004–3142, available at http://volcanoes.usgs.gov/ Prein, A. F.; Rasmussen, R. M.; Ikeda, K.; Liu, C.; Clark, M. P.; and Holland, G. J., 2017, The future intensification of hourly precipitation extremes: Nature Climate Change, Vol. 7, No. 1, pp. 48–52. Robert D. Niehaus, Inc. (RDN, Inc.), 2018, The Economic Impacts of the Montecito Mudslides: A Preliminary Assessment: Electronic document, available at http://www. rdniehaus.com / rdn / wp-content / uploads / 2018 /03/RDN_ Montecito_Mudslides_Impacts-1.pdf. Reid, M. E.; Iverson, R. M.; Logan, M.; LaHusen, R. G.; Godt, J. W.; and Griswold, J. P., 2011, Entrainment of bed sediment by debris flows: Results from large-scale experiments. In: Genevois, R.; Hamilton, D. L.; and Prestininzi A. (Eds.), Proceedings of the 5th International Conference on Debris-Flow Hazards: Mitigation, Mechanics, Prediction and Assessment. Italian Journal of Engineering Geology and Environment and Casa Editrice Universita La Sapienza, Rome, pp. 367–374. Ryan, K. C. and Noste, N. V., 1985, Evaluating prescribed fires; 15–18 November 1983; Missoula, MT. In: Lotan, J. E.; Kilgore, B.M. Fischer, W.C., Mutch, R.W. (Editors), Proceedings of the Symposium and Workshop on Wilderness Fire, Gen. Tech. Rep. INT-GTR-182. Ogden, UT, U.S. Department of Agriculture, Forest Service, Intermountain Forest and Range Experiment Station: p. 230–238.

Schwartz, J. Y., 2016, Sherpa Fire - Burn Area Emergency Response (BAER), Geology Report: USDA Forest Service internal report, 23 p. Smith, C. M.; Hatchett, B. J.; and Kaplan, M. L., 2018, Characteristics of sundowner winds near Santa Barbara, California, from a dynamically downscaled climatology: Environment and effects near the surface: Journal Applied Meteorology Climatology, Vol. 57, No. 3, pp. 589–606, available at https://journals.ametsoc.org/view/journals/apme/57/3/ jamc-d-17-0162.1.xml Staley, D. M.; Kean, J. W.; Cannon, S. H.; Schmidt, K. M.; and Laber, J. L., 2013, Objective definition of rainfall intensity-duration thresholds for the initiation of post-fire debris flows in southern California: Landslides, Vol.10, pp. 547– 562. https://doi.org/10.1007/s10346-012-0341-9 Staley, D. M.; Kean, J. W.; and Rangers, F. K., 2020, The recurrence of post-fire debris-flow generating rainfall in the southwestern United States: Geomorphology, Vol. 370, pp. 1–10. Staley, D. M.; Negri, J. A.; Kean, J. W.; Laber, J. L.; Tillery, A. C.; and Youberg, A. M., 2017, Prediction of spatially explicit rainfall intensity–duration thresholds for post-fire debris-flow generation in the western United States: Geomorphology, Vol. 278, pp. 149–162. U.S. Army Corps of Engineers, 1974, Flood Plain Information, Montecito Streams, Vicinity of Montecito, Santa Barbara County, California: Electronic document, available at http:// www.countyofsb.org / uploadedFiles / pwd / Content / Water / MontStreamsRpt1974.pdf Wagner, D. L.; Lancaster, J. T.; and DeRose, M. B., 2012, The Oak Creek Post Fire Debris and Hyperconcentrated Flows of July 12, 2008, Inyo County, California: A Geologic Investigation: Special Report 225, Version 1.0, California Geological Survey, California Department of Conservation, Sacramento, California, 44 p. Westerling, A. L., 2018, Wildfire Simulations for the Fourth California Climate Assessment: Projecting Changes in Extreme Wildfire Events with a Warming Climate: Technical report for California’s Fourth Climate Change Assessment, Publication No. CCCA4-CEC-2018-014, California Energy Commission, Sacramento, CA. Wilder BA, Lancaster JT, Cafferata PH, et al. An analytical solution for rapidly predicting post-fire peak streamflow for small watersheds in southern California. Hydrological Processes. 2020;1–14. https://doi.org/10.1002/hyp.13976 Young, D.; Courtney, A.; and Nicita, E., 2018, Thomas Fire – Burned Area Emergency Response (BAER), Soil Resource Report: USDA Forest Service internal report, 19 p.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 423–437

437

Rainfall Thresholds for Post-Fire Debris-Flow Generation, Western Sierra Nevada, CA CHAD K. NEPTUNE* JEROME V. DEGRAFF CHRISTOPHER J. PLUHAR California State University, 2576 East San Ramon Avenue M/S ST24, Fresno, CA 93740

JEREMY T. LANCASTER California Geological Survey, 801 K Street, Suite 1200, Sacramento, CA 95814

DENNIS M. STALEY U.S. Geological Survey, Alaska Volcano Observatory, 4230 University Drive, Anchorage, AK 99645

Key Terms: Post-Fire Debris Flow, Wildfire, IntensityDuration Thresholds, Ferguson Fire, Briceburg Fire ABSTRACT Wildfires have become more frequent, intense, and widespread in the western United States in the era of global climate change. Concurrently, development along wildland-urban interfaces has increased. Thus, quantification of post-fire hazards, such as debris flows and their triggering rainfall conditions, is critical for effective emergency response. We monitored the hydrologic and geomorphic responses of 10 recently burned watersheds in the areas impacted by the 2018 Ferguson Fire and the 2019 Briceburg Fire in the lower Merced River Canyon near Yosemite National Park, CA. During our monitoring, which spanned the period between November 2018 and May 2020, our study basins produced 26 debris flows in response to 60 rainstorms. The corresponding peak rainfall intensity data were used to establish empirically derived rainfall intensity-duration thresholds for debris-flow initiation for these burned areas. First-year data revealed objectively defined 15-minute-duration rainfall intensity thresholds of approximately 31 and 35 mm/hr for the Ferguson and Briceburg fires, respectively. Given the small sample size and the bimodal distribution of 15-minute-duration peak rainfall intensities observed, the objectively defined threshold may be only partially constrained. When compared to modeled estimates of spatially explicit thresholds, the objectively defined debris-flow thresholds in our analysis were higher. When compared to empirically derived thresholds from other regions, our empirical values fell in the *Corresponding author email: chadkneptune@gmail.com

middle of the range. This study suggests that model refinement or development of a regionally specific model may be warranted to provide more accurate information for use in emergency management decisions, potentially reducing unnecessary impacts to the public. INTRODUCTION The probability of rainfall-induced debris-flow initiation increases significantly following a wildfire (Cannon et al., 2001; Malkinson and Wittenberg, 2011; Parise and Cannon, 2012; Degraff et al., 2015). Wildfires can temporarily change drainage basin hydrologic characteristics, including possible increases in soil hydrophobicity, loss of protective vegetation and litter, decreases in surface roughness, and changes in the geotechnical characteristics of the soil (Neary et al., 1999; DeBano, 2000; Huffman et al., 2001; Shakesby and Doerr, 2006; Moody and Ebel, 2012; Parise and Cannon, 2012). These wildfire-caused changes contribute to an overall increase in soil susceptibility to erosion and enhance runoff generation (Neary et al., 1999; DeBano, 2000; Huffman et al., 2001; Shakesby and Doerr, 2006; Moody and Ebel, 2012; Parise and Cannon, 2012), which all affect drainage basin hydrology and therefore can increase the likelihood of debris-flow initiation. Debris flows generally initiate in two discrete modes: (1) runoff-induced surficial entrainment of progressively coarser sediment, causing progressive bulking of runoff and localized sediment capacitors that generate surges (i.e., a “snowball” effect), and (2) saturation-induced landslide initiation of a discrete mass and subsequent fluidization of the water-saturated soil mass or mixing with surface water (Wells, 1987; Parise and Cannon, 2012; Kean

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

439

Neptune, DeGraﬀ, Pluhar, Lancaster, Staley

et al., 2013). In the southwestern United States, the first mode predominates for rainfall-induced postfire debris flows, particularly within the first 2 years following the fire, when changes in surface hydrology and soil mechanical characteristics from wildfire result in increased surficial sediment entrainment and progressive runoff bulking (Wells, 1987; Cannon, 2001; Degraff et al., 2015). Thereafter, tree and shrub root decay enhances the loss of root-induced pseudocohesion, and burned-area debris-flow initiation becomes dominated by the second mode (Degraff et al., 2015). This second mode typically occurs later in the post-wildfire recovery time line (Degraff et al., 2015). Post-fire debris flows pose a risk to people and infrastructure, particularly when wildfires burn near the wildland-urban interface (Williams et al., 2019), and while this interface is an important part of life-safety impacts, infrastructure is also at risk in non-urbanized burn areas. Wildfires have increased in frequency and size since the 1970s, likely linked to anthropogenic climate change, while increased urban development has progressively encroached on more wildlands, increasing the amount of at-risk infrastructure and population (Abatzoglou and Williams, 2016; Westerling, 2016; Williams et al., 2019). Following wildfire, regional, empirically derived peak rainfall intensity-duration (ID) thresholds or, in the absence of a regional threshold, statistical or deterministic models are used to inform emergency management and emergency response decisions, including roadway closures and evacuations. As such, accurate predictions of debris-flow initiation and triggering rainfall conditions are necessary for effective risk mitigation. Models often produce false-positive and/or falsenegative predictions (Staley et al., 2013, 2016, 2017). The goal of model improvement is to reduce false predictions. False positives, which over-predict the likelihood of debris-flow occurrence, can lead to closures and evacuations when no debris flows occur, which may cause loss of the public’s faith in, and future public disregard of, warnings. False negatives underestimate likelihood, which could leave the public unprotected from debris-flow events. Thus, critical emergency response is improved by effective modeling and post-wildfire hazard prediction. With more frequent and larger wildfires in central and northern California, established tools to predict rainfall ID thresholds rapidly and accurately for debris-flow occurrence are essential for effective burned area emergency management. Currently, logistic regression analyses are used for rapid estimation of rainfall-initiated debris-flow likelihood using widely available public data sources, including basin morphology, soil burn severity, soil properties, and predicted storm intensity (Gartner et al., 2008;

440

Cannon et al., 2010; Staley et al., 2013, 2016, 2017). Existing rainfall-induced post-fire debris-flow likelihood models, including the Intermountain West and Southern California models (Cannon et al., 2010; Staley et al., 2016, 2017), were created and calibrated using empirical rainfall and debris-flow occurrence data from wildfires that occurred in those regions, which have different combinations of climatic, geologic, geomorphologic, and vegetation characteristics than California’s western Sierra Nevada. The U.S. Geologic Survey (USGS) currently uses the Staley et al. (2016, 2017) M1 model to estimate rainfall-induced debris-flow initiation likelihood and estimate initial guidance of rainfall ID thresholds for post-fire areas in the southwestern United States. However, rainfall ID thresholds can vary significantly between regions (Cannon et al., 2008; Staley et al., 2013). Therefore, additional testing of predictive models is warranted in areas outside those included in the calibration and test data sets of Staley et al. (2017). Our study objective was to define regionally specific rainfall ID thresholds for the western Sierra Nevada to assess whether the development of a separate region-specific model is needed in the western Sierra Nevada, a region not covered by the original model calibration. BACKGROUND Study Area The study area (Figure 1), composed of portions of the Ferguson Fire and Briceburg Fire burned areas, is located on the western flank of the central Sierra Nevada just west of Yosemite National Park, CA, in the El Portal 15-minute quadrangle. The burned areas are bounded primarily by Bear Creek to the west, California State Route 41 to the east, Jerseydale, an unincorporated community in Mariposa County, to the south, and the Merced River to the north, though a small portion of the burned area extends across the river (Figure 2). The region generally drains southwestwards from ∼2130 m above mean sea level in the quadrangle’s northeast corner to ∼335 m above mean sea level in the southwest. The Merced River has produced rugged topography by deeply incising through the area’s Paleozoic to early Mesozoic, weakly metamorphosed, strongly deformed metasediments and metavolcanics and Cretaceous intrusions, with a thin overlying veneer of Quaternary alluvial sediments and soils in the valley bottoms and on the hillslopes (Bateman and Krauskopf, 1987) (Figure 3). The local geology is predominantly metamorphosed stratified sedimentary rock units that trend N30°W, with parallel-striking, steeply dipping beds exhibiting localized complex folding. Units generally decrease in age

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

Rainfall Thresholds in California

Figure 1. Regional setting. The Ferguson Fire (2018) and Briceburg Fire (2019) occurred in Mariposa County, CA. The ﬁres fall within the Sierra Nevada Geomorphic Province. Fire perimeters are displayed with red and yellow borders.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

441

Neptune, DeGraﬀ, Pluhar, Lancaster, Staley

Figure 2. Site map. Basin and stream segment model estimated debris-ﬂow likelihood is based on a 24 mm/hr peak rainfall intensity-duration design storm.

442

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

Rainfall Thresholds in California

Figure 3. Site bedrock geology. Bedrock geologic data were sourced from California Geological Survey 1:250,000 Geologic Atlas Map.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

443

Neptune, DeGraﬀ, Pluhar, Lancaster, Staley

toward the southwest and contain scattered outcrops of metavolcanics with foliation and cleavage generally parallel to the surrounding units (Bateman and Krauskopf, 1987). The most common lithologies are phyllite, quartzite, chert, limestone, basaltic andesite, and diamictite. The rocks are primarily greenschist metamorphic facies, with occasional hornblendehornfels facies adjacent to intrusions (Bateman and Krauskopf, 1987). The soils in the study area primarily derive from a combination of metavolcanics and metasedimentary rocks. Soils in the Briceburg burned area primarily derive from Paleozoic to Late Jurassic marine deposits and consist of Auburn, Mariposa, and Maymen series gravelly silt loam to rocky loam with shallow soil horizons that generally give way to unweathered bedrock at depths between 400 and 500 mm (USDA, 2015) (Figure 4). Soils in the Ferguson burned area primarily derive from Paleozoic marine deposits and consist of Inceptisols in the suborder Xerepts with areas of Maymen and Mariposa series soils of gravelly loam to rocky loam with shallow soil horizons that generally extend to depths from 760 to 1500 mm. According to Schwartz and Alexander (1995), the study area’s soil KF factor, also known as the soil erodibility index of the ﬁne fraction of the soil, ranges from 0.21 to 0.50, indicating moderate to high erodibility. Chaparral shrubland and oak woodland (southwest lowlands) and mixed conifer forest (northeast highlands) cover the study area, with principal vegetation types including ponderosa pine, chamise shrubs, interior live oak, canyon live oak, and lower montane mixed chaparral (USDA Forest Service, 2018). The climate in the study area is typiﬁed by warm and dry summers and cool and wet winters, according to the Western Regional Climate Center. According to the National Oceanic and Atmospheric Administration’s ATLAS 14 Point Precipitation Frequency Estimate database (Perica et al., 2013), the 1-year, 2-year, and 10-year recurrence intervals for 15-minute rainfall accumulation for the Yosemite National Park headquarters within the study area are 7 mm, 9 mm, and 14 mm, respectively. The 1-year, 2-year, and 10-year recurrence intervals for 30-minute rainfall accumulation for the study area are 9 mm, 12 mm, and 18 mm, respectively. The 1-year, 2-year, and 10-year recurrence intervals for 60-minute rainfall accumulation for the study area are 12 mm, 16 mm, and 24 mm, respectively. According to climate summary data from 1908 to 1984 for the Dudley’s and Mariposa weather stations from the Western Regional Climate Center’s Cooperative Climatological Data Summaries Database (Western Regional Climate Center, 2021), the average total annual snowfall for the study area ranges from ∼10 cm to ∼100 cm. 444

Table 1. Logistic regression variables and coeﬃcient values for link function of the Staley et al. (2016, 2017) M1 model. Coeﬃcient or Variablea β C1 X1

C2 X2 C3 X3

Model M1 Values (15-, 30-, 60-Minute Intervals) −3.63, −3.61, −3.21 0.41, 0.26, 0.17 T (proportion of upslope area with moderate to high dNBR values and gradients ࣙ23°) × Rainfall accumulation (mm) 0.67, 0.39, 0.20 F (dNBR ÷ 1,000) × Rainfall accumulation (mm) 0.70, 0.50, 0.22 S (soil KF factor) × Rainfall accumulation (mm)

a β = intercept, Cn = coeﬃcient, Xn = independent variable, dNBR = diﬀerence-normalized burn ratio.

Estimating Debris-Flow–Generating Rainfall Rates Establishing the rainfall conditions that are likely to generate debris flows is critical for reducing risk to downstream infrastructure and populations. Regionspecific, empirically derived ID thresholds have been used to help inform emergency managers when intervention is warranted (e.g., Cannon et al., 2011). The thresholds estimate the peak rainfall intensity for a given duration that is likely to produce a debris flow in the burned area. Establishing these thresholds requires a large collection of rainfall and hydrologic response data, which is typically not available in most regions. In order to more rapidly inform emergency response managers about the hazards of burned areas where regional thresholds have not been established, so they can make better decisions regarding post-fire debris-flow events, logistic regression models have been used to estimate rainfall-initiated debrisflow generation likelihood and estimate debris-flow– initiating peak rainfall intensities (Staley et al., 2016, 2017). Existing rainfall-induced post-fire debris-flow probability models, including the Intermountain West model (Cannon et al., 2010), Southern California model, and M1 model, which is currently used by the USGS (Staley et al., 2016, 2017), were developed in regions with geologic, geomorphic, climatic, and botanic characteristics that overlap with but do not match the precise combination of characteristics of our study area. The models are based on debris-flow occurrence, rainfall intensity and duration, and geospatial data (including basin morphology, soil burn severity, soil properties, and rainfall characteristics) and use logistic regression analysis to produce an equation that generates an estimate of debris-flow likelihood with a given rainfall intensity (Gartner et al., 2008; Cannon et al., 2010; Staley et al., 2016) (Table 1). These models are effective at estimating the rainfall-induced debris-flow

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

Rainfall Thresholds in California

Figure 4. Site soil classiﬁcations. Soil classiﬁcation data were sourced from Schwartz and Alexander (1995).

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

445

Neptune, DeGraﬀ, Pluhar, Lancaster, Staley

activity likelihood during the first 2 years following a wildfire in Southern California and other locations (Staley et al., 2016, 2017). The models are based on a logistic regression model that estimates the statistical likelihood of a binary response (debris flow or no debris flow) as: P=

ex 1+ex

(1)

where P is the likelihood of debris-ﬂow occurrence with a value ranging from 0 to 1, where increasing values indicate increasing likelihood, and x is a link function deﬁned as x = β + C1 X1 + C2 X2 · · · + Cn Xn ,

(2)

where β and Cn are coefficients, and Xn values represent independent variables derived from geospatial data and rainfall accumulation records, which influence the likelihood of occurrence (Table 1). The models use soil burn severity as an aggregate proxy for the basin hydrologic response changes caused by the fire (USDA Forest Service, 2006). The models are used to estimate the likelihood of debris-flow occurrence for each location for a range of rainfall IDs. By inverting the M1 logistic regression model to solve for the rainfall that results in a likelihood = 0.5 (for the first year following wildfire), an estimated rainfall ID threshold for each observed basin can be calculated (Staley et al., 2017). These predicted thresholds can be used as surrogates for the empirically determined rainfall ID thresholds, which require direct observation to establish and are usually not available for most burned areas in the western United States. Geomorphic and Stratigraphic Characteristics of Debris-Flow Deposits This study required accurate characterization of the hydrologic and geomorphic responses of the monitored watersheds to the storms occurring during the study period. Here, we relied upon evidence-based methods using geomorphic and stratigraphic features at watershed outlets after each storm to determine whether a watershed produced a flood or hyperconcentrated flow, a debris flow, or no response. These flow types exist on a sediment-water ratio continuum, each with distinct stratigraphic and geomorphic characteristics (Pierson and Costa, 1987). Flow type is a function of discharge rate, velocity, channel width and depth, and the size, shape, cohesion, and concentration of transported sediment (Pierson and Costa, 1987). All flow types are common during rainstorms in postfire environments and even may occur during the same event. Stratigraphic and geomorphic characteristics of storm-deposited sediment can be used to infer the

446

flow types that occurred after the storm has passed and channel flow has returned to the pre-storm stage. Debris-flow deposits are characterized by very poorly to extremely poorly sorted, matrix-supported sediment with no stratification within the flow body and inverse grading at the base and normal grading at the top, weak to no imbrication, and the presence of an extreme range of particle sizes, which are poorly sorted boulder fields, boulder levees, marginal levees, and terminal lobes. These stratigraphic characteristics are a result of the entire flow mixture coming to rest en masse (Pierson and Costa, 1987; Cannon, 1995; Giraud, 2005). Hyperconcentrated flow is characterized by clast-supported sediment deposits, which are poorly sorted and weakly stratified to massive, with thin gravel lenses and both normal and reverse grading (Pierson and Costa, 1987; Giraud, 2005). Stream-flow deposits are well to moderately sorted, and stratified to massive, with weak to strong imbrication and no grading to normal grading (Pierson and Costa, 1987; Giraud, 2005). We used these characteristics to define the geomorphic and hydrologic responses of each watershed after the analyzed storms. This permitted evaluation of the response using a binary classifier model (i.e., rainfall ID thresholds), where the binary outcomes are “debris flow” and “no debris flow.” METHODS Field Observations Our primary data consisted of study area precipitation records and debris-flow occurrence observations from a total of 10 first-order drainage basins ranging in size from 0.033 km2 to 1.86 km2 , seven basins in the Ferguson Fire burned area, and three basins in the Briceburg Fire burned area. In total, 32 and 20 storm events were observed during the first and second years following the Ferguson Fire, respectively, and 28 storm events were observed the first year following the Briceburg Fire. Observations of watershed response were collected by the first author or local collaborators (residents of El Portal, CA, and California Department of Transportation [Caltrans] maintenance crews) at monitored drainage termini (outlets) within the Ferguson and Briceburg burned areas following rainstorms that generated measurable rainfall of greater than 3 mm in a 24-hour period. Watershed response (debris flow or no debris flow) was determined by observation and interpretation of sediment deposits. Local precipitation accumulation was recorded by tipping bucket rain gauges, which reported rainfall in time-stamped equal-volume increments. To ensure that measured rainfall rates were representative of those occurring at the moni-

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

Rainfall Thresholds in California Table 2. Observational data-quality summary.

Fire Briceburg Ferguson

Total Debris-Flow Observations

Total Observations with Certain Conﬁdence

Total Observations with Probable Conﬁdence

Total Observations with Questionable Conﬁdence

Observers

3 23

3 8

0 9

0 6

Author and CalTrans crews Author, CalTrans crews, local residents

toring sites, all observed basins were situated within a 2-km radius of a rain gauge (Figure 2). This study defined individual storms as intervals of precipitation bounded by at least 8 hours without recorded rainfall, an inter-storm period consistent with other post-fire debris-flow threshold analyses (e.g., Staley et al., 2013). Gauge data were converted into rainfall accumulation and peak storm intensity following the backward differencing scheme of Kean et al. (2011) for durations of 15, 30, and 60 minutes. The analyzed durations were chosen to match the rainfall-accumulation intervals for the M1 model (Staley et al., 2016, 2017).

marginal levee, and terminal lobes). Data considered “probable” were based on collaborator observations, which were not directly verified by the first author witnessing undisturbed sediment deposits, but where disturbed sediment with compositions consistent with debris-flow activity was observed by the first author (e.g., matrix-supported deposit, extreme range of particle sizes, etc.). Data considered “questionable” were provided by collaborators who are not trained professionals familiar with evaluating flow mechanics, and no corroboration by any type of field verification was completed by the first author.

Debris-Flow Occurrence

Rainfall-Intensity Thresholds

Efforts were made to conduct observations following each storm; however, on several occasions, multiple closely spaced storms passed before field observations were made. In order to corroborate observations recorded by Caltrans crews, the first author directly observed basin termini for residual lithologic evidence of debris-flow activity and observed depositional characteristics of disturbed materials removed from roadways by Caltrans crews. Table 2 provides an observation data-quality evaluation summary for the entire observation data set found in Supplemental Material Table S1. Positive debris-flow determinations were based on geomorphic and stratigraphic characteristics observed in the field that indicated debris-flow or hyperconcentrated-flow activity. Negative debrisflow determinations were based on observations that lacked evidence of recent substantial sediment deposition or where sediment deposits were well to moderately sorted, stratified to massive, with normal grading. Supplemental Material Table S2 provides a photo log of representative field photos that document geomorphic and stratigraphic characteristics of sediments interpreted to be deposited by debris flows. Table 2 shows the number of observations that fall into three categories of data quality (certain, probable, and questionable). Data considered “certain” were based on field-verified evidence of debris-flow occurrence (e.g., the deposit is matrix supported and has an extreme range of poorly sorted particle sizes, no stratification, and boulder fields, boulder levee,

Peak rainfall intensities for each recorded storm were plotted on a storm maximum rainfall intensity versus time-interval graph with each data point depicted by a symbol dependent on observed flow type (i.e., no debris flow or debris flow), with 15-, 30-, and 60-minute time intervals in accordance with Staley et al. (2013). Since each rain gauge was associated with multiple drainages, a given storm produced multiple drainage responses for a given rainfall intensity—all negative (no debris-flow generation), a mix, or in some cases all positive (all observed basins generated debris flows). Objectively defined debris-flow rainfall intensity thresholds were then established in accordance with the methods of Staley et al. (2013). Model-estimated rainfall ID thresholds were generated using the M1 model and study basin data sets in accordance with the methods of Staley et al. (2017). The predicted thresholds were calculated using a rearranged M1 model equation: Rp =

p −β ln 1−p C1 T +C2 F +C3 S

(3)

where Rp represents the rainfall accumulation in millimeters in the given time duration, p is the probability threshold for debris-flow occurrence (a standard value of 0.5 was used for the first year following the fire, and a value of 0.75 was used for the second year to account for watershed recovery, as is commonly used among post-fire watershed emergency response teams and in accordance with Staley et al., 2017),

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

447

Neptune, DeGraﬀ, Pluhar, Lancaster, Staley

T represents the proportion of upslope area burned at high or moderate severity and gradients ࣙ23°, F represents the average upslope difference-normalized burn ratio divided by one thousand (dNBR/1,000), S represents the soil KF factor, and β and Cn are as previously defined. Rp was then converted to rainfall intensity (mm/hr). The data for this study were sourced from the Burned Area Emergency Response (BAER) report for the Ferguson Fire (USDA Forest Service, 2018) and the Watershed Emergency Response Team (WERT) report for the Briceburg Fire (USDA Forest Service, 2018; CAL FIRE, 2019), as well as dNBR data provided at https://fsapps.nwcg.gov/baer/baerimagery-support-data-download and soils data from Schwartz and Alexander (1995). The median modelpredicted rainfall intensity threshold for each fire was compared to the objectively defined threshold within each fire data set.

Figure 5. Receiver operating characteristic (ROC) analysis binary confusion matrix demonstrating four possible outcomes from the combination of the binary classiﬁer model output and the observed watershed response. Threat scores (TS) were used to measure performance of binary classiﬁer model.

Logistic Regression Model Performance Evaluation The rainfall accumulation in 15-, 30-, and 60-minute intervals and debris-flow occurrence data from the two burned areas were used to evaluate the USGS model efficacy as two separate data sets and as a combined data set. Model performance was tested using model validation methods applied by Staley et al. (2016, 2017). USGS likelihood model predictions were calculated for each observed basin and storm event combination using basin-specific morphologic and hydrologic properties (sourced from USGS, 2018, 2019) as input variables and measured maximum rainfall accumulation values for each storm at 15-, 30-, and 60minute reporting intervals. Model likelihood outputs, which are a continuous variable between 0 and 1, were converted into a binary variable using a classification threshold of 0.50 for the first year and 0.75 for the second year. A binary confusion matrix comparing the binary predictions with the observed flow type was developed for each rainfall accumulation reporting interval (Swets, 1988; Fawcett, 2006) (Figure 5). The resulting output of the confusion matrix categorized the combined data into four classes (true positive, true negative, false positive, and false negative). The confusion matrix output was subjected to receiver operating characteristics (ROC) analysis using the threat score (TS) metric, where a perfect model would have a threat score of 1, and each false prediction (false positive and false negative) reduces the TS value (Swets, 1988; Fawcett, 2006; Staley et al., 2016, 2017). Finally, Tjur’s discrimination coefficient, or Tjur’s R2 (a form of pseudo R2 ), was calculated (Tjur, 2009) for each burned area and the combined data set for each rainfall accumulation interval, i.e., nine cases.

448

Tjur’s R2 calculates the difference between the average model probability output when the event of interest did not occur and the average model probability output when the observed event of interest did occur. Tjur’s R2 values will range from 0 (an indication of a model that does not discriminate between positive and negative real-world outcomes) to 1 (an indication of a model that discriminates perfectly). RESULTS AND DISCUSSION In this study, we observed a total of 10 basins across 60 storm events, for a total of 282 observations, during which a total of 26 suspected debris flows occurred, resulting in road blockage and closure of the California State Route 140 roadway and damage to several culverts, which required replacement (Figure 2 and Supplemental Material Table S1). All of the debris flows observed in this study occurred in the Paleozoic to Late Jurassic marine or Paleozoic marine geologic units (Figure 3), which are overlain by Maymen gravelly loam within the Briceburg Fire study area and by Humic Haploxerepts in the Ferguson Fire study area (Figure 4). Both of these soils are composed of fine- to coarse-grained soil with angular to subangular gravel, cobbles, and occasional boulders, which are necessary to form true debris flows. Figure 6 graphically displays the objectively defined peak rainfall ID thresholds, while Table 3 reports the objectively defined rainfall intensity thresholds and the median model-predicted thresholds for each year following fire occurrence. The objectively defined 15-minute, 30-minute, and 60-minute thresholds for the first year after the Ferguson Fire and Briceburg Fire exceeded

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

Rainfall Thresholds in California Table 3. Empirical and model-predicted rainfall intensity thresholds results and comparisons.

Fire

Year after Fire

Ferguson

Briceburg Combined Cannon et al. (2008), Southern California, Piru Fire Cannon et al. (2008), Southern California, Old and Grand Prix Staley et al. (2013), San Gabriel, San Bernardino and San Jacinto, Santa Ana Mountains, CA McGuire and Youberg (2020), New Mexico, Buzzard Fire

Objectively defined year 1 Median model estimated year 1 Objectively defined year 2a Model estimated year 2 Objectively defined year 1 Median model estimated year 1 Objectively defined year 1 Median model estimated year 1 1 1

15-Minute Threshold (mm/hr)

30-Minute Threshold (mm/hr)

60-Minute Threshold (mm/hr)

31 20 >32 26 35 26 31 17 22 13

18 16 >26 21 25 17 18 14 16 10

17 13 >17 17 16 15 16 11 13 7

a Due to no debris-ﬂow occurrences during the second year following the Ferguson Fire, objectively deﬁned thresholds could not be established. However, it is likely that the threshold would be greater than the largest peak storm intensity recorded.

the model-predicted threshold by 11 mm/hr, 2 mm/hr, and 4 mm/hr, respectively, for the Ferguson Fire, and by 9 mm/hr, 8 mm/hr, and 1 mm/hr, respectively, for the Briceburg Fire. Also, though objectively defined thresholds could not be defined for the second year after the Ferguson Fire, the highest recorded 15- and 30-minute peak rainfall intensities for the second year at the Ferguson Fire exceeded the corresponding median model-predicted thresholds by 6 mm/hr and 5 mm/hr, respectively. No debris-flow activity was observed during the second year following the Briceburg Fire, which would preclude objective definition of thresholds for the second year. Rainfall intensity data were not retrieved from field instruments at the time of the publishing of this document and thus are not included. It should be noted that the objectively defined thresholds are based on peak rainfall thresholds, not triggering rainfall thresholds. Thus, rainfall intensities sufficient to initiate debris flows may have been lower. Also, the bimodal distribution of 15-minute-duration peak rainfall intensities observed during the first year following both the Ferguson and Briceburg fires (Figure 6A and 6B) leaves some uncertainty regarding the 15-minute objectively defined thresholds. As compared with other published data for post-fire debris flows, the empirical rainfall ID thresholds established in this study fall within the middle of the range when compared to Southern California and New Mexico thresholds (Cannon et al., 2008; Staley et al., 2013; McGuire and Youberg, 2020) (Table 3). The modelestimated thresholds are conservative (i.e., estimate lower rainfall thresholds) compared to objectively de-

fined thresholds derived from our study. This may result in an increased number of unnecessary evacuations and road closures. This result suggests that additional model refinement or regionally specific model development may be warranted. Improving the model performance and reducing false positives would improve prediction credibility with the public and reduce evacuation fatigue, both of which are vital to effective emergency response. Statistical and ROC model performance measures are reported in Table 4. Three model equations were evaluated, one for each rainfall accumulation duration considered (M1_15, M1_30, and M1_60). Model performance was evaluated for each burned area individually and as a combined data set. For the combined data set, the M1_30 model produced the least false prediction (false positive and false negative) results (M1_15 = 36, M1_30 = 23, and M1_60 = 26). The TS metric indicated M1_30 performed best on the Ferguson Fire data set, M1_15 and MW_30 performed equally well on the Briceburg Fire data set, and M1_30 performed best on the combined fire data set. The bimodal distribution of the 15-minute-duration peak rainfall intensities observed may explain why the TS indicated that the M1_30 equation performed best instead of the M1_15 model, in contrast to the findings of Staley et al. (2016, 2017). Overall, the M1 model TS metrics indicated that the model performed comparably well or better on the current study’s data set than the model performed with the training data set used by Staley et al. (2016) to establish the model. The Tjur R2 metric indicated M1_15 performed the best among the current study’s data set. Overall, the Tjur R2 metric

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

449

Neptune, DeGraﬀ, Pluhar, Lancaster, Staley Table 4. Model performance evaluation results. Equations for the M1 model are based on rainfall accumulation durations of 15, 30, and 60 minutes.

Data Set

Model

Ferguson

M1_15 M1_30 M1_60 Average M1_15 M1_30 M1_60 Average M1_15 M1_30 M1_60 Average M1_15 M1_30 M1_60 Average M1_15 M1_30 M1_60 Average

Briceburg

Combined

Staley et al. (2016) Training and testing data model performance Training data subset sensitivity test

Number Number of of Debris Observations Flows 224

308

1243

316

546

110

was substantially higher for the current study’s data set (average 0.61) versus the combined data set used by Staley et al. (2016) to develop the model (average 0.33). The results of the statistical and ROC analysis indicate that the prediction model presented by Staley et al. (2016, 2017) performs relatively well in the study area when compared to the original training data set. This may indicate that a regionally specific model is not warranted, and instead further model refinement with data from throughout the western United States may be the best approach to improving prediction accuracy. Given the limited scope of this study and limited variability between basins, an evaluation of which basin characteristics had the greatest effect on debrisflow probability was not conducted. Several study limitations may have impacted the reliability or applicability of the results. Biased drainage basin selection may explain why model performance metrics indicated better model performance in this study than in the study by Staley et al. (2016). First, this study predominantly focused on drainage basins that the model predicted had a high debrisflow likelihood. It is possible that model performance in drainage basins falling within the moderate to low predicted likelihood would be less reliably modeled than those that we chose. The analysis of independent variable inputs for basins observed in this study indicated that basin selection was biased toward high-risk basins. Second, likely as a result of the first limitation presented, independent variable values (T, F, and S)

450

True Negative (TN)

False Negative (FN)

False Positive (FP)

True Positive (TP)

Threat Score (TS)

Tjur’s R2

174 187 189 183 75 75 72 74 249 262 261 257 766 836 888 830 337 335 334 335

3 3 5 4 0 0 0 0 3 3 5 4 125 215 273 204 99 94 100 98

27 14 12 18 6 6 9 7 33 20 21 25 161 101 43 102 22 24 25 24

20 20 18 19 3 3 3 3 23 23 21 22 191 101 43 112 88 93 87 89

0.40 0.54 0.51 0.48 0.33 0.33 0.25 0.31 0.39 0.50 0.45 0.45 0.4 0.38 0.34 0.37 0.42 0.44 0.41 0.42

0.67 0.62 0.52 0.60 0.74 0.67 0.44 0.62 0.68 0.62 0.50 0.60 0.35 0.34 0.30 0.33 0.33 0.34 0.31 0.33

for the drainage basins studied did not vary greatly; in particular, F (difference-normalized burn ratio ÷ 1,000) and S (soil KF factor) had little to no variability. The limited variability of the independent variables precluded assessment of correlations between independent variable value and debris-flow occurrences or determination of the variables that were most influential to model prediction in the study area. Third, it is possible that the study area geomorphic characteristics and soil KF factor fall in a range in which the model performs exceptionally well, which may not be representative of the overall western Sierra Nevada of California. The geographic extent, lithologic variability, and vegetation of the study area are limited. Consequently, the study results may not be applicable to other geographic regions, lithologies, and vegetation regimes present within the greater Sierra Nevada region. For comparison, peak rainfall data from debris-flow events that occurred in nearby burned areas that have been studied for post-fire debris-flow incidence can be considered. These burned areas include the 2017 Detwiler Fire (Mariposa County) burned area, which generated debris flows that killed two people during a March 22, 2018, storm with a peak rainfall intensity of 46.7 mm/hr over a 30-minute interval observed, and the 2015 Butte Fire (Amador County) burned area, which generated debris flows during a March 6, 2016, storm with a peak rainfall intensity of 38.1 mm/hr observed (Oakley and Lancaster, 2018). The rainfall intensities observed in these nearby burned areas are

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

Rainfall Thresholds in California

higher than the thresholds established in this study. However, they may exceed the minimum rainfall intensity required to generate a debris flow. Finally, uncertainties in the post-fire debris-flow hazard emergency assessment input data may have affected the results of the model performance analysis. Variability in the quality and/or representativeness of the dNBR data and interpretation of the dNBR data to establish soil burn severity for the T and F variables and/or variability in the quality and/or representativeness of the publicly available soil KF factor data sets may explain the better model performance. Additional data collection from more widely varying drainage basins and differing geographies, lithologies, and vegetation regimes in the Sierra Nevada is warranted, and collection of more precise debris-flow initiation timing data would greatly improve the assessment of model performance. CONCLUSIONS In this study, we established preliminary, objectively defined rainfall ID thresholds for post-fire debris flows in the area burned by the Ferguson and Briceburg fires west of Yosemite Nation Park, CA. The current USGS M1 model performed well; the first-year model estimated rainfall ID thresholds were generally equivalent to somewhat conservative when compared to objectively defined thresholds. The M1 30-minuteduration model appears to have performed better compared to the 15-minute and 60-minute durations. Given that the model-estimated rainfall ID thresholds probably overestimate debris-flow hazard, additional model refinement for this region may be warranted. This study focused on high-likelihood drainage basins (where likelihood of debris flows is higher), and, for these basins, the model performed well. It is possible that model performance is especially optimized toward high-likelihood drainage basins and may not perform as well in moderate- to low-likelihood drainage basins. Further research is recommended, including model performance evaluation across a greater portion of the Sierra Nevada region and a larger diversity of predicted basin debris-flow likelihoods. ACKNOWLEDGMENTS

Figure 6. Watershed response data and objectively defined rainfall intensity-duration thresholds for the first year after the Briceburg Fire (A) and the first 2 years following Ferguson Fire (B and C). Blue X’s represent peak storm intensities for rainstorms that did not produce debris flows. Red dots represent peak storm intensities for rainstorms that did produce observed debris flows. The black line represents the objectively defined rainfall intensity-duration thresholds calculated for each duration.

This study was supported with field equipment donated by the U.S. Geological Survey and the California Department of Water Resources, and funding for additional field equipment was provided by a California State University–Fresno Faculty Sponsored Student Research grant. The authors would like to acknowledge the contributions supporting the field effort from U.S. Forest Service staff, California Geological Survey staff, Caltrans road maintenance crews and engineer-

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

451

Neptune, DeGraﬀ, Pluhar, Lancaster, Staley

ing staff, as well as several local residents from El Portal and Jerseydale, who acted as eyes on the ground. The authors would like to thank Jerome DeGraff for recommending this study. We are thankful for his mentorship and friendship and the countless hours he invested in getting this study off the ground. We would also like to thank his wife, Sandy DeGraff, and the entire DeGraff family. SUPPLEMENTAL MATERIAL Supplemental Material associated with this article can be found online at https://aegweb.org/e-egsupplements. REFERENCES Abatzoglou, J. T. and Williams, A. P., 2016, Impact of anthropogenic climate change on wildfire across western US forests: Proceedings of the National Academy of Sciences of the United States of America, Vol. 113, pp. 11770–11775, https://doi.org/10.1073/pnas.1607171113. Bateman, P. C. and Krauskopf, K. B., 1987, Geologic Map of the El Portal Quadrangle, West-Central Sierra Nevada, California: U.S. Geological Survey Miscellaneous Field Studies Map 1998, https://doi.org/10.3133/mf1998. California Department of Forestry and Fire Protection (CAL FIRE), 2019, Watershed Emergency Response Team Evaluation, Briceburg Fire Limited Scope Summary: CAL FIRE CA-MMU-021257. Cannon, S. H.; Powers, P. S.; Pihl, R. A., and Rogers, W. P. 1995. Preliminary Evaluation of the Fire-Related Debris Flows on Storm King Mountain, Glenwood Springs, Colorado, https://doi.org/10.3133/ofr95508. Cannon, S. H., 2001, Debris-flow generation from recently burned watersheds: Environmental & Engingeering Geoscience, Vol. 7, pp. 321–341, https://doi.org/10.2113/gseegeosci.7.4.321. Cannon, S. H.; Boldt, E. M.; Laber, J. L.; Kean, J. W., and Staley, D. M., 2011, Rainfall intensity-duration thresholds for postfire debris-flow emergency-response planning: Natural Hazards, Vol. 59, pp. 209–236, https://doi.org/ 10.1007/s11069-011-9747-2. Cannon, S. H.; Gartner, J. E.; Rupert, M. G.; Michael, J. A.; Rea, A. H., and Parrett, C., 2010, Predicting the probability and volume of postwildfire debris flows in the intermountain western United States: Geological Society of America Bulletin, Vol. 122, pp. 127–144, https://doi.org/10.1130/B26459.1. Cannon, S. H.; Gartner, J. E.; Wilson, R. C.; Bowers, J. C., and Laber, J. L., 2008, Storm rainfall conditions for floods and debris flows from recently burned areas in southwestern Colorado and southern California: Geomorphology, Vol. 96, pp. 250–269, https://doi.org/10.1016/j.geomorph.2007.03.019. Cannon, S. H.; Kirkham, R. M., and Parise, M., 2001, Wildfire-related debris-flow initiation processes, Storm King Mountain, Colorado: Geomorphology, Vol. 39, pp. 171–188, https://doi.org/10.1016/S0169-555X(00)00108-2. DeBano, L. F., 2000, The role of fire and soil heating on water repellency. In Ritsema, C. J. and Dekker, L. W. (Editors), Soil Water Repellency: Occurrence, Consequences, and Amelioration: Elsevier, Amsterdam, pp. 193–202, https://doi.org/10.1016/B978-0-444-51269-7.50020-5.

452

Degraff, J. V.; Cannon, Q. S., and Gartner, Q. J. E., 2015, The timing of susceptibility to post-fire debris flows in the western USA: Environmental & Engineering Geoscience, Vol, 21, No. 4, pp. 277–292. Fawcett, T., 2006, An introduction to ROC analysis: Pattern Recognition Letters, Vol. 27, pp. 861–874, https://doi.org/10.1016/j.patrec.2005.10.010. Gartner, J. E.; Cannon, S. H.; Santi, P. M., and Dewolfe, V. G., 2008, Empirical models to predict the volumes of debris flows generated by recently burned basins in the western U.S.: Geomorphology, Vol. 96, pp. 339–354, https://doi.org/10.1016/j.geomorph.2007.02.033. Giraud, R. E., 2005, Guidelines for the Geologic Evaluation of Debris-Flow Hazards on Alluvial Fans in Utah: Utah Geological Survey Miscellaneous Publication 05-6, https://doi.org/10.34191/mp-05-6. Huffman, E. L.; MacDonald, L. H., and Stednick, J. D., 2001, Strength and persistence of fire-induced soil hydrophobicity under ponderosa and lodgepole pine, Colorado Front Range: Hydrological Processes, Vol. 15, pp. 2877–2892, https://doi.org/10.1002/hyp.379. Kean, J. W.; McCoy, S. W.; Tucker, G. E.; Staley, D. M., and Coe, J. A., 2013, Runoff-generated debris flows: Observations and modeling of surge initiation, magnitude, and frequency: Journal of Geophysical Research–Earth Surface, Vol. 118, pp. 2190–2207, https://doi.org/10.1002/jgrf.20148. Kean, J. W.; Staley, D. M., and Cannon, S. H., 2011, In situ measurements of post-fire debris flows in Southern California: Comparisons of the timing and magnitude of 24 debrisflow events with rainfall and soil moisture conditions: Journal of Geophysical Research–Earth Surface, Vol. 116, pp. 1–21, https://doi.org/10.1029/2011JF002005. Malkinson, D. and Wittenberg, L., 2011, Post fire induced soil water repellency—Modeling short and longterm processes: Geomorphology, Vol. 125, pp. 186–192, https://doi.org/10.1016/j.geomorph.2010.09.014. McGuire, L. A. and Youberg, A. M., 2020, What drives spatial variability in rainfall intensity-duration thresholds for post-wildfire debris flows? Insights from the 2018 Buzzard Fire, NM, USA: Landslides, Vol. 17, pp. 2385–2399, https://doi.org/10.1007/s10346-020-01470-y. Moody, J. A. and Ebel, B. A., 2012, Hyper-dry conditions provide new insights into the cause of extreme floods after wildfire: Catena, Vol. 93, pp. 58–63, https://doi.org/10.1016/j.catena.2012.01.006. Neary, D. G.; Klopatek, C. C.; DeBano, L. F., and Ffolliott, P. F., 1999, Fire effects on belowground sustainability: A review and synthesis: Forest Ecology and Management, Vol. 122, pp. 51–71, https://doi.org/10.1016/S0378-1127(99)00032-8. Oakley, N. and Lancaster, J. T., 2018, Post-Fire Debris Flows in California: an Atmospheric Perspective. In: ALERT User Group Conference, Ventura, CA 2018 (Power Point Presentation), https://www.alertsystems.org/ presentations/Conf2018/Session7-Forecasting_Weather/ Oakley_FORx.pdf. Parise, M. and Cannon, S. H., 2012, Wildfire impacts on the processes that generate debris flows in burned watersheds: Natural Hazards, Vol. 61, pp. 217–227, https://doi.org/10.1007/s11069-011-9769-9. Perica, S.; Martin, D.; Pavlovic, S.; Roy, I.; Laurent, M. St.; Trypaluk, C.; Unruh, D.; Yekta, M., and Bonnin, G., 2013, NOAA Atlas 14—PrecipitationFrequency Atlas of the United States 8, Ver. 2.0: National Oceanic and Atmospheric Administration, Boulder, CO.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

Rainfall Thresholds in California Pierson, T. C. and Costa, J. E., 1987, A rheologic classification of subaerial sediment-water flows: GSA Reviews in Engineering Geology, Vol. 7, pp. 1–12, https://doi.org/10.1130/ REG7-p1. Schwartz, G. E. and Alexander, R. B., 1995, Soils Data for the Conterminous United States Derived from the NRCS State Soil Geographic (STATSGO) Database: U.S. Geological Survery Open-File Report 95-449. Electronic document, available at http://water.usgs.gov/lookup/getspatial?/ussoils (accessed 1.10.21) Shakesby, R. A. and Doerr, S. H., 2006, Wildfire as a hydrological and geomorphological agent: Earth-Science Reviews, Vol. 74, pp. 269–307, https://doi.org/10.1016/j.earscirev.2005. 10.006. Staley, D. M.; Kean, J. W.; Cannon, S. H.; Schmidt, K. M., and Laber, J. L., 2013, Objective definition of rainfall intensityduration thresholds for the initiation of post-fire debris flows in southern California: Landslides, Vol. 10, pp. 547–562, https://doi.org/10.1007/s10346-012-0341-9. Staley, D. M.; Negri, J. A.; Kean, J. W.; Laber, J. L.; Tillery, A. C., and Youberg, A. M., 2016, Updated Logistic Regression Equations for the Calculation of Post-Fire Debris-Flow Likelihood in the Western United States: U.S. Geological Survery Open-File Report 2016-1106, https://doi.org/10.3133/ofr20161106. Staley, D. M.; Negri, J. A.; Kean, J. W.; Laber, J. L.; Tillery, A. C., and Youberg, A. M., 2017, Prediction of spatially explicit rainfall intensity–duration thresholds for post-fire debris-flow generation in the western United States: Geomorphology, Vol. 278, pp. 149–162, https://doi.org/10.1016/j.geomorph.2016.10.019. Strand, R. G., 1967. Geologic Map of California: Mariposa Sheet. Swets, J. A., 1988, Measuring the accuracy of diagnostic systems: Science, Vol. 240, pp. 1285–1293. Tjur, T., 2009, Coefficients of determination in logistic regression models—A new proposal: The coefficient of discrimination: American Statistician, Vol. 63, pp. 366–372, https://doi.org/10.1198/tast.2009.08210.

U.S. Department of Agriculture (USDA), 2015, Web Soil Survey: Natural Resources Conservation Service, U.S. Department of Agriculture. Electronic document, available at https://websoilsurvey.sc.egov.usda.gov/App/HomePage.htm U.S. Department of Agriculture (USDA) Forest Service, 2006, Burned Area Emergency Response Treatments Catalog: USDA Forest Service. Electronic document, available at https://www.fs.fed.us/t-d/pubs/pdf/BAERCAT/lo_ res/06251801L.pdf U.S. Department of Agriculture (USDA) Forest Service, 2018, Ferguson Fire—Burned Area Report: USDA Forest Service. U.S. Geological Survey, 2018, Emergency Assessment of PostFire Debris-Flow Hazards—Ferguson Fire: Landslides Hazards Program Online Database. Electronic document, available at https://landslides.usgs.gov/hazards/postfire_debrisflow/ detail.php?objectid=209 U.S. Geological Survey, 2019, Emergency Assessment of Post-Fire Debris-Flow Hazards—Briceburg Fire: Landslides Hazards Program Online Database. Electronic document, available at https://landslides.usgs.gov/hazards/postfire_debrisflow/ detail.php?objectid=262 Wells, W. G., 1987, The effects of fire on the generation of debris flows in Southern California: GSA Reviews in Engineering Geology, Vol. 7, pp. 105–114, https://doi.org/10.1130/REG7p105. Westerling, A. L. R., 2016, Increasing western US forest wildfire activity: Sensitivity to changes in the timing of spring: Philosophical Transactions of the Royal Society B: Biological Science, Vol. 371, https://doi.org/10.1098/rstb.2015.0178. Western Regional Climate Center, 2021, Cooperative Climatological Data Summaries: Western Regional Climate Center. Electronic document, available at https://wrcc.dri.edu/Climate/west_coop_summaries.php Williams, A. P.; Abatzoglou, J. T.; Gershunov, A.; GuzmanMorales, J.; Bishop, D. A.; Balch, J. K., and Lettenmaier, D. P., 2019, Observed impacts of anthropogenic climate change on wildfire in California: Earth’s Future, Vol. 7, pp. 892–910, https://doi.org/10.1029/2019EF001210.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 439–453

453

Runout Number: A New Metric for Landslide Runout Characterization CORY S. WALLACE* Yeh and Associates, Inc., 2000 Clay Street, Suite 200, Denver, CO 80211 and Colorado School of Mines, Department of Geology and Geological Engineering, 1516 Illinois Street, Golden, CO 80401

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

Key Terms: Landslides, landslide runout, landslide mobility, GIS, geologic hazards, engineering geology ABSTRACT Landslide runout has traditionally been quantified by the height-to-length ratio, H/L, which, in many cases, is strongly influenced by the slope of the runout path. In this study, we propose an alternative mobility measure, the unitless Runout Number, measured as the landslide length divided by the square root of the landslide area, which characterizes landslide shape in terms of elongation. We used a database of 158 landslides of varying runout distances from locations in northern California, Oregon, and Washington state to compare the two runout measurement methods and explore their predictability using parameters that can be measured or estimated using geographic information systems. The Runout Number better describes the overall runout for several landslide and slope geometries. The two mobility measures show very little correlation to each other, indicating that the two parameters describe different landslide mobility mechanisms. When compared to predictive parameters shown by prior research to relate to landslide runout, the two runout measurement methods show different correlations. H/L correlates more strongly to initial slope angle, upslope contributing area, landslide area, and grain size distribution (percent clay, silt, total fines, and sand). The Runout Number correlates more strongly to planimetric curvature, upslope contributing area normalized by landslide area, and percent sand. Although these correlations are not necessarily strong enough for prediction, they indicate the validity of both runout measurement methods and the benefit of including both numbers when characterizing landslide mobility.

*Corresponding author email: cswallace@alumni.mines.edu

INTRODUCTION Landslides threaten communities and infrastructure throughout the United States, causing over $1 billion in damages and several fatalities each year (U.S. Geological Survey, 2005). Landslides can be extremely destructive, depending on the volume and mobility of the displaced material. Researchers have made great progress toward understanding the factors contributing to landslide susceptibility and recognizing when slope failures tend to occur. Far less work has focused on understanding the mobility (i.e., the travel distance and velocity) of landslides following the initiation of movement. With the exception of a few notable examples (Horton et al., 2013; Reid et al., 2016; and Melo et al., 2018), landslide hazard assessments rarely incorporate landslide runout or post-failure mobility as components of the overall hazard and instead tend to be based on landslide susceptibility, or the tendency for landslides to occur (Guzzetti et al., 1999), and landslide size (i.e., area or volume). This is problematic because smaller landslides with high mobility can have greater impacts than larger landslides with low mobility because they can travel at higher velocities and impact areas farther from their sources. In most published landslide runout studies, runout is quantiﬁed by the height-to-length ratio, H/L, of the landslide (e.g., Corominas, 1996; Hunter and Fell, 2003; and Hungr et al., 2005). This ratio describes the overall longitudinal geometry of a landslide from the crown of the source area to the distal end of the deposit (Hunter and Fell, 2003), with smaller H/L values theoretically corresponding to longer runout events. A major shortcoming of this mobility measure is that it is inﬂuenced greatly by the slope of the runout path; for example, on uniform slopes, the H/L value simply equals the slope gradient, regardless of the travel distance of the material involved. Researchers have also proposed measures that quantify mobility in terms of landslide elongation (e.g.,

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

455

Wallace and Santi

Lockyear, 2018; Taylor et al., 2018) rather than longitudinal geometry, such as length-to-width (L/W) and length-to-area (L/A) ratios. These methods are less likely to be dominated by slope gradient, so we propose a new mobility measure, the ratio of the length of a landslide to the square root of its total area (L/A1/2 ), as an alternative elongation parameter with fewer practical limitations than H/L, L/W, and L/A. This study seeks to evaluate the effectiveness of L/A1/2 , a unitless number that we refer to as the “Runout Number,” for landslide runout characterization. Regional-scale (i.e., 1:8,000 to 1:12,000) landslide inventories are used to test whether several relevant factors influence landslide runout, using L/A1/2 as a mobility measure. The selected inventories for this study are located in the relatively humid, mountainous, western portions of northern California, Oregon, and Washington state; thus, the results of this study are expected to apply in environments that are geographically and climatologically similar to the Pacific Northwest of the United States. In general, the Runout Number L/A1/2 is expected to be a useful measure of landslide mobility because (1) the parameters L and A are relatively easy to obtain from aerial photographs, digital terrain data, and/or field measurements; (2) the values are normalized to the size of the landslide, so mobility can be evaluated independent of landslide volume; and (3) the values describe the shape of the deposit (elongation), which relates directly to the mobility of the materials involved, rather than the longitudinal geometry of the runout path (i.e., H/L). BACKGROUND Landslide Mobility and Long-Runout Landslides Landslide hazard severity is often controlled by landslide mobility, which is defined by the travel distance and velocity that a landslide reaches after failure (Iverson et al., 2015). Because direct measurement of landslide velocity is rare, velocity is usually estimated based on scour marks or other evidence of runup at bends in the runout path (Hungr et al., 2005; Hungr, 2007). Due to the fleeting nature of this type of evidence and the difficulty in acquiring it, many landslide mobility studies instead focus on landslide travel distance, or runout. The defining characteristic of long-runout landslides is that they travel much farther than expected according to basic frictional sliding models (Legros, 2002). In such a model, a landslide is treated as a sliding block, which is expected to come to rest at a distance from the source governed by the coefficient of friction between the block and the underlying sur-

456

face (Scheidegger, 1973). For long-runout landslides, there are flow mechanisms at play that allow the material to travel farther than expected, such as fluidization by high pore-water pressure (i.e., liquefaction; Iverson and Denlinger, 2001; Legros, 2002), trapped interstitial air (Kent, 1966), dispersion of particles from strong acoustic waves generated by rapid shearing (i.e., acoustic fluidization; Melosh, 1986), lubrication of the basal shear surface of the landslide by pore water (i.e., basal liquefaction; Collins and Reid, 2019), or lubrication by trapped and compressed air between the sliding mass and the underlying ground (Shreve, 1968). A recent example of long-runout behavior is the 2014 West Salt Creek landslide in Mesa County, Colorado, where a mass of Eocene Green River Formation detached from its source on the upper slopes of Grand Mesa (a broad plateau capped by basaltic volcanic rocks) and mobilized into a rapid debris avalanche with a volume of approximately 29 × 106 m3 . This mass traveled over 4 km, completely filling the valley of West Salt Creek (White et al., 2015). Three people were buried in the debris, and the deposit came within meters of nearby active gas production wellheads. This landslide exhibited high mobility with H/L and L/A1/2 values of 0.14 (White et al., 2015) and 2.8 (calculated based on approximate length and area measurements), respectively. In this case, the mobility of the sliding material was attributed to basal liquefaction (Coe et al., 2016). Another recent example is the 2014 Oso Landslide in the Stillaguamish River valley of northwestern Washington state, where a block of Quaternary glacial sediments with a volume of approximately 8 × 106 m3 mobilized into an extremely rapid, flow-like landslide that traveled over 1 km across a broad, low-relief river valley (Iverson et al., 2015; Wartman et al., 2016). This landslide also exhibited high mobility with H/L and L/A1/2 values of 0.105 (Iverson et al., 2015) and 1.6 (calculated based on approximate length and area measurements), respectively. The mobility of this landslide was also due to basal liquefaction (Collins and Reid, 2019). This landslide inundated nearly an entire residential community, killing 43 people. Landslide Mobility Measures Direct Measurement of Runout Length Landslide runout, R, is best described as the length of the depositional zone of a landslide, or the length of the deposit beyond its original source zone (Rickenmann, 2005). Because the location delineating the transition between the source zone and the depositional zone tends to be obscured or destroyed by the landslide itself, the parameter R is generally difficult to

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

A Runout Number for Landslide Characterization

Figure 1. Schematic diagram showing the parameters H, L, R, and the travel distance angle. After Hunter and Fell (2003).

measure (Stark and Guzzetti, 2009; Taylor et al., 2018) and is rarely used. As an alternative to the parameter R, the total length of the landslide deposit, L, has been considered as a mobility measure (Rickenmann, 1999; Legros, 2002; and Guthrie et al., 2010). This parameter depends greatly on the size (i.e., volume) of a landslide, so L will tend to be relatively high for very large landslides and relatively low for very small landslides, regardless of mobility. As such, the utility of this parameter is limited. The parameters L and R are shown in Figure 1. Height-to-Length Ratio, H/L Because of the limitations associated with absolute measures of landslide length, runout is often quantiﬁed indirectly using relative mobility measures, the most common of which is the H/L ratio of the landslide (Corominas, 1996; Hunter and Fell, 2003; and Hungr et al., 2005). The inverse tangent of this ratio gives the “travel distance angle” (Hunter and Fell, 2003), also known as the “angle of reach” (Corominas, 1996) or the “Fahrböschung” (Heim, 1932). The parameter H/L generally describes the overall longitudinal geometry of a landslide from the crown of the source area to the distal end of the deposit (Hunter and Fell, 2003). Smaller values of H/L theoretically correspond to longer runout events. The parameter H/L has been described as an estimate of the friction coefﬁcient at the interface between the landslide material and the underlying ground (Scheidegger, 1973). Figure 1 schematically summarizes H/L and the travel distance angle. Although H/L is widely accepted, it has several shortcomings. In cases of uniform slopes, the H/L

value mimics the gradient of the terrain over which the landslide travels (Roback et al., 2018). Additionally, in order for a highly mobile landslide to generate a very low H/L value, it must pass over an area of low relief so that the change in height across the deposit can become very small relative to the horizontal length of the deposit. For a highly mobile landslide that does not reach a flat valley bottom, H/L will not identify the landslide as having high mobility. Many studies have shown good negative correlation between H/L and landslide volume, which suggests that large landslides are likely to be more mobile (Scheidegger, 1973; Nicoletti and Sorriso-Valvo, 1991; Corominas, 1996; Legros, 2002; Hunter and Fell, 2003; and Roback et al., 2018). However, all else being equal, large landslides are more likely to reach valley bottoms (breaks in slope) than are small landslides due to their areal extent, which makes them more likely to exhibit lower H/L values. Accordingly, H/L has limited ability to describe landslide mobility independent of the slope of the runout path (Hsü, 1975; Roback et al., 2018). Mobility Measures Derived from H/L Other researchers have proposed mobility measures derived from H/L to provide more realistic assessments of landslide mobility. Because H/L is based on total vertical and horizontal distances from the extreme distal ends of a deposit, some researchers have suggested that this parameter overestimates landslide mobility and that the vertical and horizontal displacements of the center of mass of the landslide (i.e., Hcm /Lcm ) should be considered instead (Davies, 1982; Legros, 2002). Although Hcm /Lcm may provide a more physically meaningful measure of mobility in terms of energy transfer, the centers of mass of the source and deposit zones are difficult to identify, which means that the use of this parameter is often impractical. Hsü (1975) proposed a mobility measure called the “excess travel distance,” Le , which is the distance traveled by a landslide beyond that of a typical sliding block with a basal coefficient of friction equal to tan 32°. This model represents the expected behavior of a typical landslide with normal mobility. The parameter Le scales with landslide size like L and R. To minimize size dependence, Nicoletti and Sorriso-Valvo (1991) suggest the normalization of the excess travel distance to the total length of the landslide, L, to obtain the “relative excess travel distance,” Le /L. Davies and McSaveney (1999) show experimentally that the simple sliding block model with a friction coefficient of tan 32° is too simplistic and does not provide a good approximation of the expected travel distance. Therefore, the utility of these mobility measures is questionable.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

457

Wallace and Santi

Spreading Parameter, As /Af More recently, the ratio of the landslide source area, As , to the full mapped landslide area, Af , has been suggested as a mobility measure to describe the degree of spreading of the deposit (Roback et al., 2018). This parameter seems to capture an important aspect of landslide mobility: spreading and thinning of material away from its source. The limitation of this parameter is that the source area must be unobscured. For studies of recent landslides with fresh features in which the source and deposition zones can be mapped in detail, this parameter would be simple to calculate. However, in many cases, especially when working at the scale of a regional landslide inventory, landslide features are not mapped in great enough detail to accurately estimate source areas separately from deposits. For the purposes of this study, the use of As /Af as a mobility measure is considered impractical because the required area measurements are not typically available (Stark and Guzzetti, 2009; Taylor et al., 2018). Elongation Parameters Other mobility measures have been proposed to quantify the elongation of landslide deposits. A landslide that is highly elongated is theoretically highly mobile, as the sliding material must travel far from its source to produce an elongated shape. Lockyear (2018) evaluated the parameter L/A as a measure of elongation, where L and A are the total length and total area of the landslide, respectively. For two landslides of the same size (equal areas), the landslide that is more elongated will have a higher L/A. Lockyear (2018) noted that L/A has limited utility because as the size of a landslide increases, area increases more rapidly than length, so L/A values are highly scale dependent. Therefore, Lockyear (2018) suggested exploring the use of the mobility measure L/A1/2 as an alternative. Taking the square root of the area normalizes landslide length to landslide area while minimizing the size dependence of the parameter and creating a unitless ratio we term the “Runout Number.” To our knowledge, this study explores the mobility measure L/A1/2 for the first time. L/W (i.e., the aspect ratio) is geometrically similar to L/A1/2 and has also been considered as a measure of elongation (Taylor et al., 2018). However, because the width of a landslide is often variable along its profile, it is difficult to consistently measure width. Factors that Potentially Influence Landslide Mobility Several parameters have been identified in the technical literature as potential influences on landslide mo-

458

bility and, therefore, should correlate to a valid measure of landslide mobility. For this study, the following parameters were evaluated against H/L and L/A1/2 . Landslide Size—As noted earlier, large landslides are likely to exhibit higher mobility than small landslides (Scheidegger, 1973; Nicoletti and Sorriso-Valvo, 1991; Corominas, 1996; Legros, 2002; Hunter and Fell, 2003; and Roback et al., 2018). Because larger landslides are more likely to extend into low-gradient areas, the correlation between volume and H/L may result from factors independent of the landslide material itself and it may not directly demonstrate the influence of volume on mobility. Also, in many cases, landslide volume is estimated by multiplying measured landslide area by an estimated depth or average thickness of the slide mass, a process that introduces a source of error. To minimize error and variability from the thickness estimate, landslide area, rather than volume, was used to represent landslide size in this study. Material Characteristics—Two primary material types influence landslide runout: liquefiable sands (Hunter and Fell, 2003; Iverson et al., 2015) and sensitive clays (Thakur et al., 2017). For most regional landslide inventories, it is unlikely that these specific materials will be widespread or mapped in detail. However, the influence of sand and clay content can be evaluated using grain size distribution data (i.e., percent sand, silt, and clay). Influence of Water—Water is an important trigger for all types of landslides (Varnes, 1958) and can contribute to the fluidization of granular materials (Legros, 2002; Hunter and Fell, 2003; and Iverson et al., 2015). However, long-runout landslides do not always consist of highly saturated materials; for example, rock and debris avalanches can be extremely mobile but are typically relatively dry (Davies and McSaveney, 1999; White et al., 2015). McKenna et al. (2012) suggest that landslides that occur in topographic hollows, as opposed to on open slopes, are more likely to mobilize into flows because of increased availability of water to the sliding mass. Moreover, the size of the drainage basin area upstream of a landslide theoretically influences the availability of water to the sliding mass. Additionally, rainfall-triggered landslides exhibit higher mobility than gravity-driven or earthquake-triggered landslides (Wang and Sassa, 2003; Zou et al., 2017). Therefore, there is value in evaluating the influence of water on landslide runout, so we used upslope contributing area, or the drainage basin area upslope of a landslide polygon, as a proxy for overall water availability. We did not consider rainfall triggering as a potential influencing factor because

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

A Runout Number for Landslide Characterization

the triggering mechanisms for the landslides used in this study are unknown. Slope Curvature—Concave slopes correspond to “convergent” topography, which tends to accumulate surface water drainage and weathering products; convex slopes correspond to “divergent” topography, which tends to shed these materials (Bierman and Montgomery, 2014). Additionally, landslides forming in concave topography tend to be more confined, which promotes channelization (McKenna et al., 2012). We expect that channelization along landslide paths leads to longer total runout due to greater elongation. Thus, we used slope curvature to represent the availability of soil and surface water to the landslide and to represent the confining effect of the terrain on the sliding mass. Slope Angle—The slope of the terrain may also influence landslide mobility. Lockyear (2018) noted a correlation between the slope of the landslide source zone, or the initial slope angle (ISA), and H/L. Therefore, we evaluated ISA as a potential influence on landslide mobility using both mobility measures. Previous Movements—Granular material that is disturbed through landsliding is likely to be looser than its native state, so landslides that form in existing landslide deposits are theoretically more likely to contract upon shearing, causing a positive pore pressure response and subsequent liquefaction and acceleration (Hunter and Fell, 2003; Iverson et al., 2015). Therefore, we considered previous landslide movements as a potential influence on mobility.

STUDY AREAS AND DATA SOURCES We selected several regional-scale landslide inventories throughout the Paciﬁc Northwest of the United States to develop a dataset for this study using the following criteria:

r The landslide inventories must contain shapeﬁles showing outlines, movement directions, and movement type for at least 30 individual landslide events at a minimum scale of 1:12,000. r Soil survey data and a digital elevation model (DEM) of at least 10-m resolution must be available for each study area. r A range of landslide sizes (i.e., volumes or areas) must be represented within each study area. r Both short- and long-runout events must be represented, with at least several landslides per study area that appear to be signiﬁcantly elongated.

Figure 2. Map of the study area locations.

This study focused on landslides that occur by sliding and/or flowing rather than falling or toppling; therefore, polygons identified as “falls” or “topples” were removed from the dataset. We also focused on landslides consisting of “earth” or “debris” as defined by Varnes (1978), so landslides primarily consisting of “rock” (e.g., rock falls, rockslides) were removed from the dataset. We filtered the dataset by the character of movement and material type to limit the number of physical processes at play, which helped narrow the list of factors potentially influencing runout behavior. To remove as much random “noise” from the data as possible, landslides with “low” confidence, as assigned by the authors of the landslide inventories, or that are smaller than approximately 2,000 m2 were removed from the dataset. Landslide polygons whose geometries are obscured or truncated by cross cutting of other landslides were also removed from the dataset, as were complexes consisting of multiple coalescing landslides. Based on these criteria, study areas were selected in northern California, Oregon, and Washington (Figure 2). All three study areas are in regions of rolling to mountainous terrain with humid, temper-

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

459

Wallace and Santi

ate climates and rainy winter seasons. Selecting study areas distributed throughout the Pacific Northwest of the United States provides a robust dataset representative of a relatively large geographic region. Results of this study are expected to apply throughout this region and in those that are geographically and climatologically similar. Thus, we considered the three study areas together as one dataset throughout this study. The California dataset consists of several hundred mapped landslides within the American River Canyon along US Highway 50 between the communities of Riverton and Kyburz in northern California. At least 50 of the landslides moved during the winter of 1996– 1997, although the study area contains both prehistoric and historic landslides primarily triggered by heavy precipitation and/or rapid snowmelt (Wagner and Spittler, 1997). Out of approximately 640 inventoried landslides, 65 met the requirements for this study and were retained for analysis. The study area is located in a narrow canyon with steep walls. Bedrock consists of an assemblage of Paleozoic ultramafic and metamorphic rocks, Mesozoic granitic rocks, and Tertiary volcanic and volcaniclastic rocks (Wagner et al., 1981; Wagner and Spittler, 1997). The Oregon dataset consists of a regional landslide inventory of the Eugene-Springfield metropolitan area and parts of the surrounding unincorporated county land. The landslide inventory was compiled by the Oregon Department of Geology and Mineral Industries based on existing landslide inventories and recent lidar datasets (Calhoun et al., 2018). The terrain is generally rolling to mountainous with gentle to steep slopes. Bedrock consists of Miocene to Eocene volcanic and sedimentary rocks overlain by late Pliocene and Quaternary sediments (Walker and Duncan, 1989; Calhoun et al., 2018). Primary triggering mechanisms for landslides in this study area are heavy precipitation and large earthquakes (Calhoun et al., 2018). The inventory contains both prehistoric and historic landslides, and out of 634 inventoried landslides, 56 were retained for analysis. The Washington dataset consists of a regional landslide inventory of the western Columbia River gorge in Skamania County, Washington, on the border with Oregon. The inventory was compiled by the U.S. Geological Survey based on recent lidar datasets and high-resolution aerial imagery (Pierson et al., 2016). Bedrock consists of Oligocene to Pleistocene volcanic and volcaniclastic rocks interbedded locally with clastic sedimentary rocks (Korosec, 1987; Phillips, 1987; and Pierson et al., 2016). Primary triggering mechanisms for landslides in this study area are heavy precipitation, large earthquakes, and toe erosion by the Columbia River (Pierson et al., 2016). The inventory contains prehistoric and historic landslides; at least 12

460

Figure 3. Histogram showing the range of landslide areas for the 158 events used in this study.

of the recent landslides were moving at the time of the study or within the two decades prior (Pierson et al., 2016). Out of 215 inventoried landslides, 37 were retained for analysis. Data for this study were derived from regional-scale, publicly available, governmental data sources and can be displayed and analyzed using geographic information systems (GIS). The factors selected for use include landslide area, planimetric curvature, grain size percentages (clay, silt, ﬁnes, and sand), upslope contributing area, initial slope angle, and previous landslide movement at the location. The 158 landslides included in this study show a wide range in area, from ∼2,000 m2 to ∼20 million m2 , with most values <500,000 m2 and a median of ∼18,000 m2 (Figure 3). Landslide inventories are from the publications noted previously; elevation data are from U.S. Geological Survey National Elevation Dataset 1/3 arc-second DEMs (U.S. Geological Survey, 2017), and soils data are from U.S. Department of Agriculture (USDA) Natural Resources Conservation Service Soil Surveys (Mitchell and Silverman, 1985; Patching, 1987; and Haagen, 1990).

METHODS OF DATA COLLECTION AND ANALYSIS The following sections describe the methods of data collection and processing implemented in this study. All the data used in this study were evaluated in ArcGIS 10.7 (ESRI, 2019) and analyzed in MATLAB (Mathworks, 2019).

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

A Runout Number for Landslide Characterization

Calculating Mobility Measures

Upslope Contributing Area

Both L/A1/2 and H/L values were calculated to compare their effectiveness. Landslide area, A, was measured directly in ArcGIS using a simple geometrical calculation. The height, H, of each landslide polygon was measured by calculating the change in elevation across each polygon based on zonal statistics of elevation data within each polygon. The length of each landslide, L, was estimated by generating minimum bounding rectangles around each landslide polygon. The minimum bounding rectangles record the lengths of the long and short axes of each polygon. Given the direction of movement of each landslide, the length of the landslide can be approximated by the dimension of the minimum bounding rectangle that most closely parallels the direction of movement. The effectiveness of this approach would be limited for cases in which the direction of movement changes significantly along the length of a landslide (e.g., there are sharp curves or bends in the landslide path); however, there were no landslides in the dataset that were significantly affected by this limitation. For both mobility measures, A, H, and L were calculated using the full post-failure landslide geometry (i.e., the source and runout zones were not differentiated).

Upslope contributing area is deﬁned by the drainage basin area contributing to a landslide area. Drainage basin areas were calculated using a watershed calculator in ArcGIS, which took as inputs a ﬂow direction raster (calculated based on the DEM for the study area) and a zone from which to calculate a drainage basin area (landslide polygon). The watershed calculator calculated the drainage basin area upslope of each entire landslide polygon.

Planimetric Curvature Planimetric curvature represents the degree of concavity or convexity of a slope. Negative and positive planimetric curvature values represent concave (convergent) and convex (divergent) slopes, respectively. A planimetric curvature raster was generated for each study area based on the DEM, and the mean curvature value over the full area of the landslide polygon was taken to summarize the curvature of the terrain. The full landslide area, rather than the source area, was used because runout behavior can be aﬀected by terrain inﬂuences both at the source and along the runout path (McKenna et al., 2012; Ng et al., 2013).

Initial Slope Angle The initial slope angle represents the slope angle of the source area. Because source areas are rarely mapped in detail at the regional scale, we estimated the source zone as the upper 25 percent of the landslide length based on observations of landslide geometries and typical observed topographic inflection points for landslides within our dataset. This estimation provided an approximate, systematic approach to source area delineation that we considered sufficient for a regionalscale study given the mapping resolution of the landslide inventories. The length and height differences were calculated from the upper 25 percent of the minimum bounding rectangle that was used to measure the total landslide length. The slope of the source zone was calculated as the ratio of these differences. Previous Movements The Washington dataset identifies previous movement if a landslide overlaps other older landslides (Pierson et al., 2016). The landslides in this dataset were categorized as follows:

r First-generation landslides (no previous movement) r Landslides that have reactivated within terrain that has failed previously

r Reactivated slides involving one older landslide r Reactivated slides involving two or more older landslides

Grain Size Percentages Surﬁcial soil data were imported into ArcGIS from the USDA Web Soil Survey Soil Data Viewer (U.S. Department of Agriculture, 2019). Representative grain size percentages (clay, silt, sand, and “ﬁnes” calculated as the sum of the clay and silt contents) were added as layers to the map document. Where landslide polygons overlapped two or more map units, a weighted average (mean) of representative grain size percentages was calculated for each landslide based on the percentage of landslide area in each map unit.

The California and Oregon datasets do not explicitly indicate whether each landslide experienced previous movements, so each landslide was examined individually using an automated process in GIS that identiﬁed those that overlap other landslides as having experienced previous movements. In some cases, this distinction was unclear, such as for landslides that partially overlap other landslides that appear roughly the same age or older. In such cases, if the upper 25 percent of the landslide length (the estimated source zone) fully overlaps the other landslide, then it was considered to have experienced a previous movement.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

461

Wallace and Santi

Identifying Factors that Influence Runout Behavior For each continuous variable, scatter plots were generated showing the mobility measures L/A1/2 and H/L as a function of each variable. Least-squares linear regressions were performed to identify statistically significant linear trends. For variables that appeared to be log-normally distributed, the natural logarithms of the data values were plotted so that the relationships could be interpreted in terms of a straight-line equation. Other non-linear trends were explored where appropriate. Least-squares regression analyses in MATLAB generated several key statistics for interpreting the relationship between variables. The primary output for evaluating whether a statistically significant trend exists between a variable and a mobility measure is a p-value associated with the estimated slope of the best-fit line. This study relied on the commonly used criterion that p-values less than a significance level of 0.05 indicate that there is a statistically significant trend in the data. The coefficient of determination (R2 ) was used to estimate how much variation in the data was accounted for by the linear model. Therefore, the p-values suggest whether there is a significant relationship, and R2 suggests whether the model has predictive capability. This study used simple linear regression rather than multiple linear regression because (1) it is relatively simple to compare how each variable influences each of the two mobility measures when the correlations are evaluated separately and (2) the calculated variable values exist in very different scales, so performing simple linear regressions allows the results to be interpreted without needing to standardize the values across scales. RESULTS OF DATA COLLECTION AND ANALYSIS The following sections briefly summarize the mobility measures and input parameter data collected for this study. The full dataset is available upon request from the authors. Calculating Mobility Measures Figure 4 shows histograms of the Runout Number L/A1/2 and H/L values for the full dataset, with mean values of 1.9 and 0.33, respectively. Because these values were calculated using approximate length measurements based on the lengths of minimum bounding rectangles, there may be some error in these values, which Wallace (2020) estimates at approximately 5 to 7 percent. The histograms are plotted on the margins of a scatter plot showing L/A1/2 vs. H/L for the full

462

Figure 4. Scatter plot of L/A1/2 and H/L values from this study with a least-squares linear ﬁt. L/A1/2 and H/L values for the 2014 West Salt Creek and 2014 Oso landslides are also shown for comparison. Marginal histograms on the vertical and horizontal axes show the distributions of L/A1/2 and H/L values, respectively.

dataset. A least-squares linear fit is also shown to illustrate the nature of the correlation between mobility measures. Mobility measure values for the two recent long-runout landslides introduced earlier in this paper (Oso and West Salt Creek) are shown on the scatter plot for comparison with our dataset. The H/L values for both landslides plot within the bottom ∼10 percent of values within our dataset, indicating that they exhibit high mobility. The L/A1/2 value for West Salt Creek plots within in the top ∼90 percent of the dataset, whereas Oso only plots in the top ∼35 percent of the dataset, indicating that the landslides are differently elongated and may exhibit different mobility. Factors that Influence Runout Behavior The factors that potentially influence runout identified earlier were compared to the mobility measures L/A1/2 and H/L to evaluate statistically significant correlations for the dataset of 158 landslides. Table 1 summarizes the results of least-squares linear regressions between the variables and the mobility measures. The values for landslide area (A) and upslope contributing area (UCA), appear to be log-normally distributed, so the analysis was performed on natural logarithm transformed data. Because UCA scales with landslide size, correlations were also calculated against UCA values normalized to landslide area (nUCA). All of the factors show some correlation to landslide runout, although neither runout measurement method relates to every factor. Runout measured by H/L strongly correlates to landslide area, upslope contributing area, and initial slope angle. Runout mea-

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

A Runout Number for Landslide Characterization Table 1. Results of least-squares regressions between landslide mobility measures L/A1/2 and H/L and factors that may correlate to landslide runout. Factor

L/A1/2

H/L

Interpretation

Landslide area (ln A) p-value 0.49 1.50E-13 Correlates to H/L (negative) but not L/A1/2 . 2 0.0031 0.307 R Planimetric curvature p-value 0.005 0.29 Correlates to L/A1/2 (negative) but R2 is very low. No correlation with H/L. 0.051 0.0074 R2 Upslope contributing area (ln UCA) p-value 0.53 1.40E-14 Correlates to H/L (negative) but not L/A1/2 . 2 R 0.0026 0.33 Upslope contributing area normalized by landslide area (ln nUCA) p-value 0.030 0.44 Correlates to L/A1/2 (positive) but R2 is very low. No correlation with H/L. 0.030 3.97E-03 R2 Initial slope angle (ISA) p-value 9.27E-04 2.68E-13 Correlates to L/A1/2 (negative) but R2 is very low. Correlates to H/L (positive). 0.069 0.30 R2

sured by L/A1/2 correlates to planimetric curvature (although the predictive capacity is low, as evidenced by a low R2 value), upslope contributing area when normalized by landslide area, and initial slope angle (also with low predictive capacity). Although landslide area is included as a component of L/A1/2 , the two variables do not correlate, probably due to the squareroot transformation in the denominator of L/A1/2 (by comparison, the untransformed parameter L/A correlates strongly with landslide area within our dataset). The direction of correlation (positive or negative slope of the best-fit line, as indicated in Table 1) generally matches expectations. We would expect H/L (with smaller values indicating longer runout) to show a negative correlation with both landslide area and UCA, indicating longer runout as landslide area and UCA increase. Additionally, we would expect L/A1/2 (with larger values indicating longer runout) to show a negative correlation with planimetric curvature (due to channelization) and a positive correlation with nUCA. Unexpectedly, both runout measurements suggest that shallower slopes correlate with longer runout. A positive correlation between ISA and H/L is intuitive because both parameters contain very similar information (i.e., the gradient of the terrain); however, the positive correlation between ISA and L/A1/2 is not intuitive. A possible explanation is that lower-gradient slopes tend to have thicker soil mantles that could be more prone to long-runout behavior due to their physical characteristics. Otherwise, this finding would seem to imply that there may be another factor collinear with ISA that is also influencing runout. The most obvious factors, landslide area, UCA, and nUCA, did not show any correlation to ISA, so any potential collinearities remain unresolved. Future studies could explore these potential collinearities us-

ing multiple linear regression with a higher-resolution landslide inventory dataset. Table 2 presents the results of least-squares linear regressions for each mobility measure against the various grain size percentages. In general, both L/A1/2 and H/L correlate negatively with fine-grained soil constituents (clay, silt, and fines) and positively with coarse-grained constituents (sand). However, the correlations involving H/L are stronger, with R2 values in the range of 0.28 to 0.35 and p-values less than 1E-10. This pattern of correlations implies that, for H/L, long runout is associated with higher fines and lower sand content. For L/A1/2 it implies the opposite: Long runout is associated with lower fines and higher sand content. Scatter plots of L/A1/2 vs. grain size percentages indicate that the trends appear to be influenced by clusters of values at the lower and upper ends of the distributions of fines and sand contents, respectively. Therefore, in a general sense, L/A1/2 is sensitive to overall fines and sand content, as shown in Figure 5, which compares L/A1/2 values for landslides with relatively high and low sand contents, illustrating how runout tends to be longer with higher sand contents. It is likely that the strong correlations between grain size percentages and H/L are influenced by the relationship between grain size percentages and slope angle, as grain size tends to decrease downslope (Bierman and Montgomery, 2014). Table 3 summarizes the results of two-sample ttests that evaluated if groups of landslides, sorted by whether they experienced previous movements, have significantly different runouts. Based on the results for this dataset, there does not appear to be a significant difference (at the 0.05 significance level) in H/L between landslides that experienced previous movements and those that have not. Unexpectedly,

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

463

Wallace and Santi Table 2. Results of least-squares linear regressions for L/A1/2 and H/L as a function of grain size percentages. L/A1/2 Parameter Coeﬃcient (slope) p-value R2

H/L

Clay

Silt

Fines

Sand

Clay

Silt

Fines

Sand

–0.011 0.027 0.035

–0.015 0.022 0.037

–0.0071 0.016 0.041

0.0069 0.020 0.038

–0.0072 1.33E-11 0.28

–0.011 6.93E-14 0.33

–0.0049 1.86E-14 0.34

0.0049 1.26E-14 0.35

the inﬂuence of previous movements on landslide runout remains unclear. DISCUSSION Runout Thresholds

Figure 5. Histograms of L/A1/2 values sorted by sand contents above and below 65 percent. Histogram bars with black outlines represent sand contents ࣙ65 percent; histogram bars with gray outlines represent sand contents <65 percent. Dashed and solid lines are probability density functions representing the distributions of L/A1/2 values for sand contents above and below 65 percent, respectively.

L/A1/2 is signiﬁcantly higher for landslides that have not experienced previous movements. This contrasts with conclusions from earlier studies (e.g., Lockyear, 2018) indicating that previous landslide movement tends to produce longer runout landslides. Therefore,

An important question that has not been fully answered is “What constitutes a long-runout landslide?”, or put another way, “What is the threshold value between long and short runout landslides?” For H/L, the “excess travel distance” model proposed by Hsü (1975) points to a long-runout threshold of tan 32° = 0.62, which is probably too high (for example, from our dataset in Figure 4, this would categorize 97 percent of the landslides as long runout). Scheevel (2017) suggests a cutoff H/L value of 0.1 (5.7°), which we consider too low (only 1 percent of our dataset would be identified as long runout). We suggest a threshold of 0.2 (11°), for which 26 percent of our dataset would be identified long runout. For L/A1/2 , we used selected landslides from the California dataset (Wagner and Spittler, 1997) representing a wide range of L/A1/2 values to identify short–long thresholds for L/A1/2 , as shown in Figure 6. We also included a “medium” runout category to provide better separation between short and long runout, so we delineated short–medium and medium–long runout thresholds as well. The selected landslides are shown without scale because L/A1/2 de-

Table 3. Results of t-tests (0.05 significance level) indicating whether landslides with indications of previous movement have longer runout than those without. L/A1/2 Parameter Mean Standard deviation N* Degrees of freedom tcritical t-score p-value *

H/L

Previous Movements

No Previous Movements

Previous Movements

No Previous Movements

1.7 0.61 65

2.1 0.78 91

0.32 0.14 60

0.34 0.20 91

154 ±1.98 –3.29 1.20E-03

149 ±1.98 –0.5 0.62

N values diﬀer between mobility measures due to errors in the automated evaluation of previous movements in GIS.

464

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

A Runout Number for Landslide Characterization

Figure 6. Landslides with varying runout from the California dataset. Runout measurements for each landslide are shown in Table 4. Where more than one landslide is shown, the landslide of interest is indicated with an arrow. Hillshade basemap is generated from 1/3 arc-second DEM (U.S. Geological Survey, 2017). Scale not shown because L/A1/2 measurement is independent of scale.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

465

Wallace and Santi Table 4. Calculated mobility measures for selected landslides from the California dataset shown in Figure 6.

Table 5. Comparison of mobility measures for various hypothetical slope geometries shown in Figure 7.

Mobility Measure Landslide A B C D E F G H I

Mobility Measure

L/A1/2

H/L

Scenario and Case

6.4 4.1 2.5 2.2 2.0 1.8 1.7 1.4 0.9

0.32 0.34 0.32 0.37 0.31 0.25 0.38 0.91 0.40

Scenario A Short Long Scenario B High Low Scenario C Small Large Scenario D Scenario E Steeper angle Shallower angle Scenario F

scribes landslide shape independent of landslide size. Table 4 summarizes the mobility measures calculated for each landslide. Note that when using H/L as the mobility measure, none of the examples shown were identified as long-runout landslides because the steep terrain does not have any slope breaks, so H/L merely measures the slope angle. From visual inspection, Landslides A and B are highly elongated and clearly represent long-runout landslides. These events appear to have been highly mobile flows and have L/A1/2 values greater than 4.0. Landslide C is also quite elongated, with an L/A1/2 value of 2.5. Landslides D and E are noticeably elongated (values of 2.2 and 2.0, respectively), although less so than C. Finally, Landslides F, G, H, and I are roughly ellipsoidal or even circular in shape and do not appear to be significantly elongated (values ranging from 0.9 to 1.8). From these examples, it appears that medium and long runout events could be divided by an L/A1/2 threshold of 2.5 (with 18 percent of our dataset in Figure 4 falling in the long runout category), and short and medium events could be divided by a threshold of 1.5 (with 27 percent of our dataset in the short runout category, leaving 55 percent in the medium runout category). If only two categories are considered, we suggest an L/A1/2 threshold of 2.0 to divide short and long runout events (categorizing 39 percent of our dataset as long runout and 61 percent as short runout). Comparison of L/A1/2 and H/L The parameters L/A1/2 and H/L were compared to evaluate whether there is correlation between them or they are independent measurements. Figure 4 shows a scatter plot of both values with a least-squares linear fit. There is a very weak negative trend between L/A1/2 and H/L, which is the expected result because L/A1/2 values increase with longer runout whereas H/L values decrease with longer runout. The p-value of 0.058 and R2 of 0.024 indicate little true correlation and that

466

H/L

L/A1/2

0.47 0.47

2.0 2.5

0.47 0.32

2.2 2.2

0.47 0.32 0.25

2.2 2.2 1.7

0.53 0.36 0.47

2.2 2.2 2.4

the two parameters describe different landslide mobility mechanisms and generally contain very different information. H/L tends to describe the longitudinal geometry of the runout path, highlighting landslides that travel over relatively flat terrain, whereas L/A1/2 describes the planform shape of landslides, highlighting those that travel far from their sources or over consistently sloped terrain. Applicability of Mobility Measures to Different Slope Geometries To demonstrate how L/A1/2 performs against H/L for a variety of landslide and slope geometries, we compared the mobility measures for six idealized landslide scenarios, summarized in Table 5. Scenario A (Figure 7A) shows two landslides with the same source area and volume traveling over a uniform slope. Here, the landslide labeled “long” runs out farther than the landslide labeled “short.” Because the landslides form on uniform slopes with the same gradient, the measured H/L for both cases simply matches the slope gradient. However, the L/A1/2 value for the long case is greater than that of the short case. This scenario demonstrates that L/A1/2 tends to perform better when characterizing landslide runout on uniform slopes. Scenario B (Figure 7B) shows two landslides that have the same volume and character of movement, but one landslide is positioned high on a uniform slope and the other is positioned lower, where it crosses a slope break. The H/L value for the landslide positioned close to the slope break is lower than that of the landslide positioned far from the slope break, but the L/A1/2 values are the same. Here, H/L correctly identifies the shallower case as having longer runout,

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

A Runout Number for Landslide Characterization

Figure 7. Idealized landslide cross sections showing longitudinal and planimetric geometry used in the calculation of mobility measures H/L and L/A1/2 . Planimetric geometry is represented by rectangles projected above each cross section. Mobility measures for each scenario are shown in Table 5. (A) Two landslides with the same source area and volume, but different mobility, travel over a uniform slope. (B) Two similar landslides—one crosses a slope break. (C) The larger of two similar but differently sized landslides crosses a slope break. (D) A relatively wide landslide reaches a break in slope and runs out over flat terrain. (E) Two similar landslides form on slopes with different gradients. (F) A landslide initiates in a steep valley headwall and runs out a long distance over a relatively low-gradient slope.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

467

Wallace and Santi

since the landslide achieved the same total distance traveling over shallower terrain. However, this scenario also illustrates that H/L can only identify long-runout landslides when they reach a break in slope or otherwise travel from steeper terrain into shallower terrain. Scenario C (Figure 7C) is similar to Scenario B, but one landslide is significantly larger than the other landslide. Because of its size, the larger landslide reaches a break in slope and has a lower H/L value (correctly indicating longer runout), whereas the L/A1/2 values for both cases are the same. This scenario illustrates the correlation between H/L and landslide size, as larger events are more likely to cross breaks in slope. In Scenario D (Figure 7D), a relatively wide landslide reaches a break in slope and runs out a significant distance from its source over flat terrain. In this case, because the landslide is relatively wide, it does not appear to be significantly elongated, and the calculated L/A1/2 value is only 1.7, which suggests short to medium runout. However, because the landslide reaches a break in slope and runs out a significant distance over flat terrain, the H/L value of 0.25 correctly suggests medium to long runout. The 2014 Oso landslide in Washington state is an example of this type of landslide, in which the landslide ran out over a relatively flat river valley, so the H/L value was very low, but because the landslide was relatively wide, the corresponding L/A1/2 value was only about 1.6. The Oso landslide is shown on the scatter plot in Figure 4. This scenario shows that, in general, H/L may perform better for landslides whose initial sliding masses are at least as wide as they are long. Scenario E (Figure 7E) shows two similar landslides that form on slopes with different gradients. Here, H/L is lower (indicating longer runout) for the shallower case, whereas L/A1/2 is the same for both cases. Although H/L simply matches the slope gradient, it correctly identifies the shallower case as having longer runout since the landslide was able to move the same distance over shallower terrain. In this case, H/L describes mobility better than L/A1/2 because it highlights landslide movement over shallower terrain, whereas L/A1/2 does not. However, this scenario also emphasizes the dependence of the mobility measure H/L on slope gradient. In Scenario F (Figure 7F), a landslide initiates in a steep valley headwall and runs out a long distance over a relatively low-gradient slope below the headwall. Because the steep headwall adds a significant amount of height to the overall path, the H/L value (0.47) does not identify this as a long-runout landslide, whereas the L/A1/2 value (2.4) correctly identifies long runout. This scenario emphasizes that L/A1/2 often characterizes long runout better than H/L when the landslide source area consists of a steep valley headwall.

468

L/A1/2 does not necessarily perform better in every scenario, but it appears to perform well where H/L is less useful, such as in Scenarios A and F. The strength of L/A1/2 as a mobility measure is also demonstrated in Figure 6, where, because the landslides occur on the steep slopes of the deeply incised American River canyon, the slope gradient strongly influences the H/L values. The H/L values for these landslides are between 0.3 and 0.4, which does not seem to indicate long runout. Conversely, the calculated L/A1/2 values show a broad range, correctly identifying landslides with both short and long runout. L/A1/2 is especially useful for characterizing runout when the sliding mass becomes channelized (i.e., the landslide path follows convergent topography). The correlation between planimetric curvature and L/A1/2 is clear due to the tendency of convergent topography to channelize landslides (McKenna et al., 2012). This relationship is less clear with H/L, whose values may be unaffected by channelization. The West Salt Creek landslide, in which the sliding mass was confined to the West Salt Creek valley, is an example of channelization promoting long runout behavior (White et al., 2015). In this case, the H/L value of 0.14 characterizes the landslide as having reasonably long runout, but the L/A1/2 value of 2.8 stands out as having very long runout. Thus, L/A1/2 is likely a more useful measure of mobility when landslides occur in channelizing (convergent) terrain, whereas H/L may be more useful when the terrain is divergent or generally flat (as in the case of the Oso landslide). In general, the performance of a mobility measure is based on how well it quantifies mobility, not necessarily by how well it correlates with other geomorphological factors. Many of the correlations between mobility measures and input parameters are stronger for H/L than for L/A1/2 , but this should not be interpreted to mean that H/L is a better measure of mobility. For example, sand content correlates significantly better with H/L than with L/A1/2 , but this is most likely due to the relationship between slope gradient and grain size, where grain size tends to decrease downslope (Bierman and Montgomery, 2014). Similarly, landslide area correlates much better with H/L than with L/A1/2 , which is probably because as landslides increase in size, they are more likely to reach a break in slope or otherwise extend into lower gradient terrain, such as by reaching a valley bottom (Scenario C). Many researchers have discussed this relationship (e.g., Scheidegger, 1973; Nicoletti and Sorriso-Valvo, 1991; Corominas, 1996; Legros, 2002; Hunter and Fell, 2003; and Roback et al., 2018), but the importance of slope breaks has not been carefully investigated previously.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

A Runout Number for Landslide Characterization

Predicting Landslide Runout

REFERENCES

Based on the analyses of each mobility measure, it appears that efforts to predict the runout of an individual landslide could benefit by including both Runout Number and H/L, although Runout Number requires some significant assumptions in order to be used in a predictive sense. For areas where landslide boundaries are obscured, or for areas where a landslide has not yet occurred, these runout measures are more difficult (or impossible) to quantify. Landslide area A, total length L, and total fall height H must all be estimated. Scheevel (2017) successfully used limited local landslide inventories to bracket the range of these values within specific areas, assuming that previous local landslides demonstrate the typical behavior of an area. This assumption could be extended to infer that new landslides will also fall within the same range of behavior, so landslide runout prediction can be estimated from an inventory of previous landslides in the area. The reliability of each mobility measure depends on the site-specific topography, as indicated in Figure 7.

Bierman, P. R. and Montgomery, D. R., 2014, Key Concepts in Geomorphology: W. H. Freeman and Company Publishers, New York, 552 p. Calhoun, N.; Burns, W.; Franczyk, J.; and Monteverde, G., 2018, Landslide Hazard and Risk Study of Eugene-Springfield and Lane County, Oregon: Oregon Department of Geology and Mineral Industries, Interpretive Map Series 60, 230 p. Coe, J. A.; Baum, R. L.; Allstadt, K. E.; Kochevar Jr., B. F.; Schmitt, R. G.; Morgan, M. L.; White, J. L.; Stratton, B. T.; Hayashi, T. A.; and Kean, J. W., 2016, Rockavalanche dynamics revealed by large-scale field mapping and seismic signals at a highly mobile avalanche in the West Salt Creek valley, western Colorado: Geosphere, Vol. 12, No. 2, pp. 607–631. Collins, B. D. and Reid, M. E., 2019, Enhanced landslide mobility by basal liquefaction: The 2014 State Route 530 (Oso), Washington, landslide: GSA Bulletin, Vol. 132, No. 3–4, pp. 451–476. Corominas, J., 1996, The angle of reach as a mobility index for small and large landslides: Canadian Geotechnical Journal, Vol. 33, No. 2, pp. 260–271. Davies, T. R., 1982, Spreading of rock avalanche debris by mechanical fluidization: Rock Mechanics, Vol. 15, No. 1, pp. 9–24. Davies, T. R. and McSaveney, M. J., 1999, Runout of dry granular avalanches: Canadian Geotechnical Journal, Vol. 36, No. 2, pp. 313–320. ESRI, 2019, ArcGIS 10.7: Redlands, CA. Guthrie, R. H.; Hockin, A.; Colquhoun, L.; Nagy, T.; Evans, S. G.; and Ayles, C., 2010, An examination of controls on debris flow mobility: Evidence from coastal British Columbia: Geomorphology, Vol. 114, No. 4, pp. 601–613. 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. Haagen, E., 1990, Soil Survey of Skamania County Area, Washington: U.S. Department of Agriculture, Soil Conservation Service (WA659), 247 p. Heim, A., 1932, Bergsturz und Menschenleben: Fretz und Wasmuth Verlag, Zürich, Switzerland, 218 p. Horton, P.; Jaboyedoff, M.; Rudaz, B.; and Zimmermann, M., 2013, Flow-R, a model for susceptibility mapping of debris flows and other gravitational hazards at a regional scale: Natural Hazards and Earth System Sciences, Vol. 13, No. 4, pp. 869–885. Hsü, K. J., 1975, Catastrophic debris streams (Sturzstroms) generated by rockfalls: GSA Bulletin, Vol. 86, No. 1, pp. 129–140. Hungr, O., 2007, Dynamics of rapid landslides. In Sassa, K.; Fukuoka, H.; Wang, F.; and Wang, G. (Editors), Progress in Landslide Science: Springer Berlin Heidelberg, Berlin, Germany, pp. 47–57. Hungr, O.; Corominas, J.; and Eberhardt, E., 2005, Estimating landslide motion mechanism, travel distance and velocity. In Hungr, O.; Fell, R.; Couture, R.; and Eberhardt, E. (Editors), International Conference on Landslide Risk Management: Joint Technical Committee on Landslides and Engineered Slopes, London, pp. 99–128. Hunter, G. and Fell, R., 2003, Travel distance angle for “rapid” landslides in constructed and natural soil slopes: Canadian Geotechnical Journal, Vol. 40, No. 6, pp. 1123–1141. Iverson, R. M.; George, D. L.; Allstadt, K.; Reid, M. E.; Collins, B. D.; Vallance, J. W.; Schilling, S. P.; Godt, J. W.; Cannon, C. M.; Magirl, C. S.; Baum, R. L.; Coe, J. A.;

CONCLUSIONS In this study, the effectiveness of the landslide mobility measure L/A1/2 , referred to as the Runout Number, was investigated and compared to the commonly used mobility measure H/L. The Runout Number is generally an effective measure of landslide mobility that may be sensitive to different controlling parameters than H/L and in some cases indicates long runout correctly whereas H/L does not. In other cases, H/L correctly indicates long runout whereas Runout Number does not. The Runout Number effectively characterizes the elongation of landslides, which is good for identifying landslides that mobilize into flows and deposit material significant distances away from their source zones. In many cases, the mobility measure H/L only reflects surface slope gradients, so L/A1/2 better characterizes runout on uniform slopes. However, for cases where relatively wide landslides run out over flat terrain and do not become significantly elongated (such as the 2014 Oso landslide), H/L tends to be a more effective mobility measure than L/A1/2 . The Runout Number has statistically significant correlations with planimetric curvature, grain size percentages (clay, silt, fines, and sand contents) of surficial soils, and nUCA. These correlations make physical sense, and the mechanisms driving these correlations are well documented in the technical literature. Based on these observations, we suggest that landslide mobility should be characterized using both L/A1/2 and H/L, with recognition that the slope geometry influences the effectiveness of each method for individual cases.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

469

Wallace and Santi Schulz, W. H.; and Bower, J. B., 2015, Landslide mobility and hazards: implications of the 2014 Oso disaster: Earth and Planetary Science Letters, Vol. 412, pp. 197–208. Iverson, R. M. and Denlinger, R. P., 2001, Flow of variably fluidized granular masses across three-dimensional terrain: 1. Coulomb mixture theory: Journal Geophysical Research: Solid Earth, Vol. 106, No. B1, pp. 537–552. Kent, P. E., 1966, The transport mechanism in catastrophic rock falls: Journal Geology, Vol. 74, No. 1, pp. 79–83. Korosec, M. A., 1987, Geologic Map of the Hood River Quadrangle, Washington and Oregon: Washington Division of Geology and Earth Resources Open File Report 87-6, 61 p. Legros, F., 2002, The mobility of long-runout landslides: Engineering Geology, Vol. 63, No. 3–4, pp. 301–331. Lockyear, R. A., 2018, Identification of Parameters for Predicting Long-Runout Landslides in the Western United States: Unpublished M.S. Thesis, Department of Geology and Geological Engineering, Colorado School of Mines, 126 p. Mathworks, 2019, MATLAB R2019b: Natick, MA. McKenna, J. P.; Santi, P. M.; Amblard, X.; and Negri, J., 2012, Effects of soil-engineering properties on the failure mode of shallow landslides: Landslides, Vol. 9, No. 2, pp. 215–228. Melo, R.; van Asch, T.; and Zêzere, J. L., 2018, Debris flow runout simulation and analysis using a dynamic model: Natural Hazards and Earth System Sciences, Vol. 18, No. 2, pp. 555– 570. Melosh, H. J., 1986, The physics of very large landslides: Acta Mechanica, Vol. 64, No. 1–2, pp. 89–99. Mitchell, C. R. and Silverman, L. J., 1985, Soil Survey of Eldorado National Forest Area, California: U.S. Department of Agriculture, Soil Conservation Service (CA724), 329 p. Ng, C. W. W.; Choi, C. E.; and Law, R. P. H., 2013, Longitudinal spreading of granular flow in trapezoidal channels: Geomorphology, Vol. 194, pp. 84–93. Nicoletti, P. G. and Sorriso-Valvo, M., 1991, Geomorphic controls of the shape and mobility of rock avalanches: GSA Bulletin, Vol. 103, No. 10, pp. 1365–1373. Patching, W. R., 1987, Soil Survey of Lane County Area, Oregon: U.S. Department of Agriculture, Soil Conservation Service (OR637), 237 p. Phillips, W. M., 1987, Geologic Map of the Vancouver Quadrangle, Washington: Washington Division of Geology and Earth Resources Open File Report 87-10, 27 p. Pierson, T. C.; Evarts, R. C.; and Bard, J. A., 2016, Landslides in the Western Columbia Gorge, Skamania County, Washington: U.S. Geological Survey Scientific Investigations Map 3358, 25 p. Reid, M. E.; Coe, J. A.; and Brien, D. L., 2016, Forecasting inundation from debris flows that grow volumetrically during travel, with application to the Oregon Coast Range, USA: Geomorphology, Vol. 273, pp. 396–411. Rickenmann, D., 1999, Empirical relationships for debris flows: Natural Hazards, Vol. 19, No. 1, pp. 47–77. Rickenmann, D., 2005, Runout prediction methods. In Jakob, M. and Hungr, O. (Editors), Debris-flow Hazards and Related Phenomena: Springer Berlin Heidelberg, Berlin, Germany, pp. 305–324. Roback, K.; Clark, M. K.; West, A. J.; Zekkos, D.; Li, G.; Gallen, S. F.; Chamlagain, D.; and Godt, J. W., 2018, The size, distribution, and mobility of landslides caused by the 2015 M w 7.8 Gorkha earthquake, Nepal: Geomorphology, Vol. 301, pp. 121–138. Scheevel, C. R., 2017, Predicting Landslide Stability, Runout, and Failure Velocity at Cook Lake Landslide, Wyoming: Unpub-

470

lished M.S. Thesis, Department of Geology and Geological Engineering, Colorado School of Mines, 87 p. Scheidegger, A. E., 1973, On the prediction of the reach and velocity of catastrophic landslides: Rock Mechanics Rock Engineering, Vol. 5, No. 4, pp. 231–236. Shreve, R. L., 1968, Leakage and fluidization in air-layer lubricated avalanches: GSA Bulletin, Vol. 79, No. 5, pp. 653–658. Stark, C. P. and Guzzetti, F., 2009, Landslide rupture and the probability distribution of mobilized debris volumes: Journal Geophysical Research: Earth Surface, Vol. 114, No. F00A02, pp. 1–16. Taylor, F. E.; Malamud, B. D.; Witt, A.; and Guzzetti, F., 2018, Landslide shape, ellipticity and length-to-width ratios: Earth Surface Processes Landforms, Vol. 43, No. 15, pp. 3164– 3189. Thakur, V.; L’Heureux, J.-S.; and Locat, A., 2017, Landslide in Sensitive Clays – From Research to Implementation: Springer International Publishing, Cham, Switzerland, 1–11 p. U.S. Department of Agriculture, 2019, Soil Data Viewer 6.2: Washington, D.C. U.S. Geological Survey, 2005, Landslide Hazards - A National Threat: U.S. Geological Survey Fact Sheet 2005-3156, 2 p. U.S. Geological Survey, 2017, 1/3rd arc-second Digital Elevation Models (DEMs) - USGS National Map 3DEP Downloadable Data Collection: Online data source, available at https://viewer.nationalmap.gov/basic/. Varnes, D. J., 1958, Landslide types and processes: Landslides Engineering Practice, Vol. 24, pp. 20–47. Varnes, D. J., 1978, Slope movement types and processes. In Schuster, R. L. and Krizek, R. J. (Editors), Landslides: Analysis and Control: Transportation Research Board Special Report No. 176, pp. 11–33. Wagner, D. L.; Jennings, C. W.; Bedrossian, T. L.; and Bortugno, E. J., 1981, Geologic Map of the Sacramento Quadrangle: California Division of Mines and Geology Regional Geologic Map 1A, 4 p. Wagner, D. L. and Spittler, T. E., 1997, Landsliding Along the Highway 50 Corridor: Geology and Slope Stability of the American River Canyon between Riverton and Strawberry, California: California Department of Conservation, Open-File Report 9722, 25 p. Walker, G. W. and Duncan, R. A., 1989, Geologic Map of the Salem 1 Degree by 2 Degree Quadrangle, Western Oregon: U.S. Geological Survey Miscellaneous Investigations Series Map I1893, 1 p. Wallace, C. S., 2020, A New Mobility Measure and Scoring System for Predicting Long-Runout Landslides: Unpublished M.S. Thesis, Department of Geology and Geological Engineering, Colorado School of Mines, 103 p. Wang, G. and Sassa, K., 2003, Pore-pressure generation and movement of rainfall-induced landslides: Effects of grain size and fine-particle content: Engineering Geology, Vol. 69, No. 1, pp. 109–125. Wartman, J.; Montgomery, D. R.; Anderson, S. A.; Keaton, J. R.; Benoît, J.; dela Chapelle, J.; and Gilbert, R., 2016, The 22 March 2014 Oso landslide, Washington, USA: Geomorphology, Vol. 253, pp. 275–288. White, J. L.; Morgan, M. L.; and Berry, K. A., 2015, The West Salt Creek Landslide: A Catastrophic Rockslide and Rock/Debris Avalanche in Mesa County, Colorado: Colorado Geological Survey Bulletin 55, 45 p. Zou, Z.; Xiong, C.; Tang, H.; Criss, R. E.; Su, A.; and Liu, X., 2017, Prediction of landslide runout based on influencing factor analysis: Environmental Earth Sciences, Vol. 76, No. 21, pp. 723.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 455–470

Comparison of Two Logistic Regression Models for Landslide Susceptibility Analysis Through a Case Study LAUREN SOUTHERLAND WENDY ZHOU* Department of Geology and Geological Engineering, Colorado School of Mines, 1516 Illinois Street, Golden, CO 80401

Key Terms: Logistic Regression, Logit, Geology-Slope, Borehole-Slope, Landslide Susceptibility ABSTRACT The effectiveness of a geology-slope logistic regression (logit) model versus a borehole-slope logit model was compared and evaluated to determine the most suitable method for use in creating landslide susceptibility maps in a 15.54-km2 (6-mi2 ) area southwest of Colorado Springs. The inputs of the geology-slope logit model include readily available geology and topography data. The borehole-slope logit model uses data collected in the field and produced through laboratory experiments; these inputs include unit weight, cohesion, friction angle, and groundwater depths in addition to topography data. The results of both models were compared and evaluated for their accuracy using the area under the receiver operating characteristics (ROC) curve method. Landslide susceptibility maps were created specifically to failure mechanisms present in the study area, including circular failures within the colluvium deposits, circular failures within the weathered shale, and planar failures within the weathered shale. The results show that the geologyslope model yields an area under the ROC curve ranging from 53 percent to 62 percent. The area under the ROC curve for the borehole-slope model ranges from 52 percent to 79 percent. The borehole-slope model outperformed the geology-slope model for circular failures in the colluvium deposits and planar failures within the weathered shale, while the geology-slope model slightly outperformed the borehole-slope model for circular failures within the weathered shale. This study provides insight into the effectiveness of using existing geology and topography data for a rapid assessment of landslide susceptibility. INTRODUCTION Mass-wasting processes, such as landslides, can be destructive to existing and future developments. Land*Corresponding author email: wzhou@mines.edu

slide susceptibility maps present the likelihood of landslides and can assist in providing information to land developers and planners to avoid future landslide damage. Many researchers conducted landslide susceptibility studies in conjunction with numerical or analytical modeling and Geographic Information System (GIS) techniques (e.g., Luo et al., 2004, 2009; Ayalew and Yamagishi, 2005; Choi et al., 2012; Schicker and Moon, 2012; Devkota et al., 2013; Althuwaynee et al., 2014; Nourani et al., 2014; Regmi et al., 2014; Umar et al., 2014; Chen et al., 2016, 2017, 2018; Zhang et al., 2016; Zezere et al., 2017; Southerland, 2019; Wu et al. 2019, 2020; Nie et al., 2021; Zhou, 2021; and Zhou et al., 2021). However, very few (Chen et al., 2016; Southerland, 2019) performed the analysis on a failure mechanism–specific basis. This article presents a study using logistic regression analysis and GIS technique to assess landslide susceptibility on a failure mechanism–specific basis (i.e., landslide susceptibility was investigated one failure mechanism at a time based on the landslide failure modes present in the study area). Logistic regression, also known as logit, is a commonly used statistical model for estimating an event’s probability in response to a group of predictor variables. Logit can be used to identify areas susceptible to landslides based on a combination of input parameters (i.e., predictor variables). The input parameters can vary depending on the failure mechanisms of the landslides (Chen et al., 2016; Southerland, 2019) and the availability of data, as well as the potential redundancy of the parameters. This study evaluated the effectiveness of a geology-slope logit model to predict landslide-susceptible locations in a 15.54-km2 (6-mi2 ) area of southwestern Colorado Springs. This model consisted of two predictor variables to landslides: geology and topography. These two input parameters are readily available for all Colorado Springs and, therefore, can be used to assess landslides rapidly. In order to evaluate the effectiveness of this geology-slope logit model, a borehole-slope model with a broader range of input parameters was also developed to compare performance in predicting existing landslide locations. The borehole-slope model includes inputs

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

471

Southerland and Zhou

of unit weight, cohesion, friction angle, groundwater depth, and slope angle. Both models were applied to create failure mechanism–specific landslide susceptibility maps, including circular failures within the colluvium deposits, circular failures within weathered shale, and planar failures within weathered shale. The results from these two models were compared quantitatively using a receiver operating characteristics (ROC) curve method. Potentially, the geology-slope logit model can be useful for rapid landslide susceptibility studies because of the availability of the data compared to the data needed for the borehole-slope logit model. This study provides insight into the effectiveness of using existing geological and topographic data for a rapid assessment of landslide susceptibility. The selected study area of southwestern Colorado Springs has been burdened by mass-wasting processes, which have impacted existing developed residential areas and are likely to affect future developments (White and Wait, 2003a; FEMA, 2015). Colorado Springs was selected for this study for a few reasons: (1) the frequency of landslide occurrences, (2) the technical limitations of previous studies, and (3) availability of geotechnical data from previous engineering studies in the area. The Colorado Geological Survey (CGS) produced landslide susceptibility maps based on the geology, geomorphology, and topography associated with existing landslides (White and Wait, 2003a). This publication includes three map plates covering the entirety of the El Paso County that present uniformly rated landslide-susceptible “zones” as well as landslide extents mapped by CGS, the United States Geological Survey (USGS, 2004), and various geotechnical consultants. An additional effort to identify landslide susceptibility in Colorado Springs includes research conducted by Garrett (2011) assessing landslide susceptibility by using a “factor of safety” approach. Garrett’s study included an infinite slope analysis for planar failures, as well as a simplified Bishop’s analysis of landslides for circular failures. Three hypothetical groundwater conditions (dry, partially saturated, and fully saturated) were assumed to create the factor of safety maps for each groundwater condition and failure mechanism. While the previous studies provided the information needed at those times, the limitations of these studies include, but are not limited to, the following: (1) the CGS maps are binary landslide susceptibility maps, rather than landslide susceptibility ratings; (2) Garrett’s (2011) study assumed one failure mechanism across the entire study area; (3) neither of these studies conducted landslide analysis based on specific failure mechanism (failure-specific); and (4) neither of these studies evaluated the accuracy of the landslide susceptibility maps.

472

STUDY AREA The 15.54-km2 (6-mi2 ) study area for this project is located in southwest Colorado Springs, CO, at the base of Cheyenne Mountain, west of Colorado State Highway 115, south of West Cheyenne Boulevard, and north of Norad Road (Figure 1). Geologic Background The geology of Colorado Springs includes various colluvial and alluvial deposits, generally overlying the Cretaceous age Pierre Shale bedrock. The Pierre Shale bedrock extends across the Colorado Springs area and most of eastern Colorado (Carroll and Crawford, 2000; Rowley, 2003). According to the geologic map of the Colorado Springs quadrangle (Carroll and Crawford, 2000), much of the area also contains both older and recent landslide deposits. The study area near the Cheyenne Mountain base is known to have unstable geology, which produces hazardous ground movement, including landslide events (Rowley, 2003; White and Wait, 2003a, 2003b). Mapped Quaternary age deposits in southwest Colorado Springs, as shown in Figure 2a, include older landslide deposits (Qls, Holocene, and Pleistocene) as well as more recent landslide deposits (Qlsr, Late Holocene). Other colluvium descriptions include old fan and rockfall deposits (Qfro, Late to Middle Pleistocene) and young alluvial fan deposits (Qfy, Holocene). Typically, most of these deposits contain ﬁne-grained material that can contribute to slope instability. The Pierre Shale (Kp) is easily weathered and contains overly consolidated claystone, expansive clays, and evaporite minerals and salt deposits, including sulfates such as gypsum. The Pierre Shale is susceptible to slope instability (Carroll and Crawford, 2000). A cross section depicting the local stratigraphy is shown in Figure 2b. Climate Colorado Springs is located in a semiarid climate with an average rainfall of approximately 40.64 cm (16 in.) per year. Saturation of overlying materials is generally minimal throughout the late summer to winter but increases with the spring snowmelt. Intermittent heavy rainfall has historically triggered landslide events in Colorado Springs, such as in May and August of 1999, which triggered new landslides and reactivation of existing landslides (Carroll and Crawford, 2000). A monthly average of precipitation from 1990 to 2018, according to the City of Colorado Springs Municipal Airport Weather Station (KCOS, latitude: 38.81°N, longitude: 104.69°W, Elevation, 1,885 m [6,186 ft]), ranges from 10.2 to 91.4 mm (0.4 to 3.6 in.), with the wettest months from May to August and driest months

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

Landslide Susceptibility Analysis

Figure 1. Study area location of southwestern Colorado Springs, CO.

from November to January (National Oceanic and Atmospheric Administration [NOAA], 2018). METHODOLOGY The workﬂow for this project is shown in Figure 3. This section provides a detailed description of data acquisition, logistic regression modeling, implementation of the geology-slope and borehole-slope models, and accuracy assessment. Data Acquisition The City of Colorado Springs, the CGS, and Colorado Springs Utilities provided the information necessary to complete this study. The City of Colorado Springs, (n.d.) has provided an openly available database of information containing geotechnical reports for the study area. The data from 74 geotechnical reports include geotechnical borehole information and slope stability analyses. This information was used for inputs into the borehole-slope model as well as for the classiﬁcation of existing landslides by failure mechanism. The CGS provided the geologic maps of the study area in a digital geospatial vector format (Carroll and

Crawford, 2000; Rowley, 2003). The geologic maps were used as one of the two inputs for the geologyslope model. In addition, the “Potential Areas of Landslide Susceptibility of Colorado Springs” map was made available in a GIS raster format; this map includes binary landslide-susceptible zones as well as landslide extents mapped by CGS, the USGS, and geotechnical consultants (such as Colorado Engineering & Geotechnical Group; CTL Thompson; Earth Engineering Consultants, Inc.; Entech Engineering, Inc.; etc.). A total of 25 landslide extents were digitized from this data set and used as the landslide inventory for this study. Colorado Springs Utilities (2011) provided a light detection and ranging (LiDAR) digital elevation model (DEM) acquired in 2011 with a spatial resolution of approximately 5 ft and a vertical accuracy of 1.5 ft. The LiDAR DEM was used to derive the slope map, an input parameter in both the geology-slope and borehole-slope models. Borehole information was delineated from geotechnical reports. The depths of geologic formation contacts and geologic unit descriptions were extracted from boring logs and assigned to each borehole point. A total of 198 boreholes were digitized in point shapeﬁle format in ArcGIS and included only those related

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

473

Southerland and Zhou

Figure 2. (a) Geologic maps depicting the northern (digitized from Carroll and Crawford, 2000) and southern (digitized from Rowley, 2003) portions of the study area; (b) Cross section, west-east orientation, of underlying geology encompassing Colorado Springs. Note the Kp (or Pierre Shale unit in light green) prominently present (Carroll and Crawford, 2000).

474

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

Landslide Susceptibility Analysis

Figure 4. Generalized stratigraphic column for the study area.

tions in this area, a discrepancy in the interpretation of the geology is expected. The explanations in Figure 4 illustrate the discrepancy in geologic descriptions from diﬀerent geotechnical consultants. Depths of geologic units vary, depending on location in the study area. With this generalized stratigraphic column, the process of determining the appropriate failure mechanism for each slope stability model becomes simpler. This was done by noting the type of failure plane that was modeled and the deepest geologic layer at which the plane was located. A total of 99 consultant-modeled slope stability models were digitized based on maps from the geotechnical reports that could be georeferenced. Geotechnical boreholes, landslide extents, and slope stability lines were digitized in ArcGIS, as shown in Figure 5. This data set was used to classify the landslides by failure mechanism, as shown in Table 1. Figure 3. Workﬂow followed for this project.

to an existing slope stability analysis to create part of the input parameters for the borehole-slope model. Landslide Inventory Classiﬁcation This study requires landslide failure mechanism– speciﬁc data, and therefore a procedure had to be developed to classify the existing landslide inventory. A generalized stratigraphic column was created to simplify this process based on the review of the geotechnical reports and corresponding borehole logs (Figure 4). Three geologic units of varying depths were generalized based on the knowledge that the Pierre Shale was exposed prior to the Quaternary landslide events, creating a layer of weathered shale above the Pierre Shale bedrock (Carroll and Crawford, 2000). The generalized stratigraphic column was used to identify major geologic units in the borehole logs by assuming lateral consistency for certain geologic units. Since multiple consultants have conducted investiga-

General Workﬂow Upon completion of data acquisition, the input parameters (i.e., predictors) for the logit models were prepared as maps in ArcGIS. The models were then constructed based on the inputs and the response of those inputs to a binary parameter (Reed and Berkson, 1929). In this case, the binary parameter is deﬁned as either a landslide area or a non– landslide area. The input parameters, which are typically causative factors of landslides, were derived from various sources, such as soil shear strength and unit weight from geotechnical reports and slope angle from Table 1. Summary of the number of landslides by failure mechanism based on evaluation of slope stability cross sections available in the consultant geotechnical reports. Failure Mechanism

No. of Modeled Landslides

Colluvium circular Weathered shale, circular Weathered shale, planar Shale, circular

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

9 9 15 1

475

Southerland and Zhou

Figure 5. Landslide inventory map displaying slope stability cross sections (black lines), landslides (red polygons), and geotechnical borehole locations (red circles).

LiDAR DEM, when using ArcGIS. Inputs can be assigned into the model as continuous or categorical variables, which are determined based on the nature of the data (Hosmer and Lemeshow, 2000). Continuous assignment of predictors is appropriate if the input parameter should be evaluated in terms of a one-unit interval. Categorical assignment of predictors is appropriate for inputs in which the evaluation of a one-unit interval is not the objective and for which larger variances in values should be evaluated. For the purpose of this study, all inputs were assigned to be continuous. The probability of an event is calculated based on the statistical signiﬁcance of the input parameters determined in the logit model associated with the landslide locations (Eqs. 1 and 2): z = β0 + β1V1 + β2V2 + . . . + βnVn

(1)

P = 1/ 1 + e−z

(2)

where z = output significance equation; β0 = the calculated coefficient not associated with any input feature; β1 , β2 , … βn = calculated coefficients associated with each input parameter; V1 , V2 , … Vn = input parameters; and P = model probability equation. 476

The procedures for the logit modeling for both geology-slope and borehole-slope inputs are the same as depicted in Figure 6. Input parameters for each model were created as ArcGIS maps or image coverages based on the data sources available for the study area. Values of the input parameters as well as the landslide inventory locations were transferred to 4.57 m × 4.57 m (15 ft × 15 ft) mesh, or an equally sized grid, in ArcGIS. A subset of the mesh was randomly extracted to train the model, and the rest of the mesh was used to assess the model’s accuracy. The model was trained using a subset of the data to output a probability equation that was then applied to the remaining non-training area. Training areas were the same per failure mechanism for both the geology-slope and borehole-slope models. Since there was only one landslide classiﬁed by a geotechnical consulting ﬁrm as a circular failure within the shale bedrock (Table 1), this mechanism was not used for either the geologyslope or borehole-slope model. A subset of the mesh was randomly extracted to train the model, and the rest of the mesh was used to assess the model’s accuracy. The model was trained by using a subset of the data to output a probability equation that was then applied to the remaining

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

Landslide Susceptibility Analysis

from low to high susceptibility. The Jenks Natural Breaks Classiﬁcation method optimizes the arrangement of a set of values into natural classes by minimizing the variance within classes and maximizing the variance between classes (Jenks, 1967). The logit models were failure mechanism speciﬁc (i.e., a susceptibility map was created for each failure mechanism one at a time). Geology-Slope Logit Model

Figure 6. Workﬂow followed for a failure-speciﬁc logistical regression model.

non-training area. Training areas were the same per failure mechanism for both the geology-slope and borehole-slope models. The training areas were determined by choosing two landslides per failure mechanism randomly through a stratified sampling technique, which allowed the sample pool to be divided into subgroups. These subgroups were then randomly sampled to minimize subjectivity in the data pool (Daniel, 2011). For this study, subgroups of “east” and “west” locations were created per failure mechanism. This subgroup type was chosen for the study area because of the geologic and soil property changes from east to west in the region. After the subgroups were identified, a buffer was created surrounding each of the chosen landslides. This buffer area incorporates both non-landslide and landslide areas. The buffer size is equal to one-half the landslide length, such that there are approximately balanced amounts of landslide and non-landslide training cells. Once the output model probability equation was applied to the non-training area, a susceptibility map was then created by classifying the calculated probabilities of the mesh into different susceptibility ranges. The Jenks natural breaks classification method was chosen to classify the probability values into five categories,

For the geology-slope approach, the geologic maps for the study area were used (Carroll and Crawford, 2002; Rowley, 2003), as was the LiDAR slope raster (Colorado Springs Utilities, 2011). These input parameters are readily available throughout Colorado Springs and are easy to prepare. Both geology and slope were determined to be continuous predictors in the logit model so as to evaluate the inputs on a oneunit interval. A review of multiple geotechnical reports for the study area has provided insight into geology’s impact on slope instability in Colorado Springs. Geotechnical consultants analyzed landslides in varying geologic units depending on subsurface investigation findings. The landslide susceptibility map published by the CGS has also provided insight based on geologists’ interpretations of landslide hazards associated with geology, geomorphology, and topography (White and Wait, 2003a). Therefore, the surficial geology should be considered as a causative factor for landslides in Colorado Springs. The CGS has provided digital surficial geologic maps for the Colorado Springs (Carroll and Crawford, 2000) and Cheyenne Mountain (Rowley, 2003) quadrangles. In order to convert the geologic map into a model input parameter, numerical values based on groupings of similar geology were assigned to each geologic unit (Table 2). These numerical values are related to the relationship between the geology and existing landslides, for which the higher the value, the more likely the geology is to influence landslides. Exposed bedrock shale outcrops were considered to have the greatest impact on landslide failures based on historical and recent landslide developments throughout the study area (Carroll and Crawford, 2000; White and Wait, 2003a). Therefore, shale surficial geologic units were given a rank of 6 in the logistic regression model. The landslide deposits also experience landslide failures through reactivation of material during and after copious rainfall events (Scott and Wobus, 1973; White and Wait, 2003a), so younger and older landslide deposits were given rankings of 5 and 4, respectively. Next, surficial deposits containing mostly granular material, including colluvium and other granular

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

477

Southerland and Zhou Table 2. Summary of geologic unit integer values for use as an input parameter in the geology-slope logistical regression model. Integer Value (1 = Most Problematic for Landslides)

Geologic Unit Group Kp (Pierre Shale), Kcgg (Carlile Shale/Greenhorn Limestone/Graneros Shale, undivided) Qlsr (recent landslide deposits) Qls (older landslide deposits) Qc (colluvium deposits) af, Qfo, Qfro, Qfy, Qg2, Qsw, Qt (artiﬁcial ﬁll, older fan deposits, older rockfall deposits, younger fan deposits, pediment gravels, sheetwash, and terrace gravels) Kn, Xgd, Cs, and Kdp (Niobrara Formation, Granodiorite, Sawatch sandstone, and Dakota sandstone/Purgatoire Formation, undivided)

5 4 3 2

The surﬁcial slope is represented in 1.52 m × 1.52 m (5 ft × 5 ft) pixels derived from the LiDAR DEM. Borehole-Slope Logit Model In order to determine the eﬀectiveness of the geology-slope model, a borehole-slope model was created to compare the same objective. The boreholeslope model includes soil strength parameters, groundwater depth, and slope, which are important to consider because of their direct relationships to landslide failures (Duncan, 1996). The slope map used for the borehole-slope model is the same as that used for the geology-slope model. All input parameters for the borehole-slope model were used as continuous predictors in the logit modeling process. Soil Strength Parameters

deposits, were given rankings of 3 and 2, respectively. Finally, geologic units consisting of granite and sandstone bedrock were given the lowest ranking of 1 given that they carried the least likelihood of a landslide failure within these competent bedrock units. At the boundaries of the Cheyenne Mountain and Colorado Springs quadrangles, the geologic units do not agree in some places because of the edge eﬀect caused by differences among the authors in terms of their mapping techniques. Slope Surface slope is an important consideration for the Colorado Springs area, based on previous landslide studies. Landslides have occurred in Colorado Springs, where slope angles were as low as 12 percent (or 7°), and landslides in this area can be sensitive to slope topography (Scott and Wobus, 1973; Hill, 1974; and Gruntfest and Huber, 1985). Therefore, slope angles greater than 12° contribute even more to instability in the area. The slope map was derived from the 2011 LiDAR DEM acquired by Colorado Springs Utilities.

Soil strength parameters considered for the borehole-slope model include cohesion, friction angle, and unit weight. Strength parameters were extracted from each slope stability analysis in geotechnical reports that used either laboratory-tested values or inferred values. Laboratory testing for soil strength parameters includes results from direct shear and unconﬁned compressive testing, depending on the geotechnical consultant. In order to determine whether the inferred soil strength values used in the slope stability analyses were reasonable, the medians of laboratory-tested values were extracted and compared against the medians of inferred values for similar soil types (Table 3). Median values were chosen as a comparison strategy for outlier values to compare the range of data values for laboratory values versus inferred values. The observed landslides used as training sites in this methodology have already succumbed to movement; therefore, residual cohesion and friction angle values were used. This consideration is expected to yield conservative results in the study area. Cohesion, friction angle, and unit weight values were calculated per

Table 3. Summary of median values determined from laboratory and inferred values of cohesion, friction angle, and unit weight. Cohesion, c (lb/ft2 ) Laboratory tests Colluvium Weathered Shale Shale Inferred values Colluvium Weathered Shale Shale

478

Friction Angle, ϕ (°)

Unit Weight, γ (lb/ft3 )

n-Value (No. of Reports)

0 395 1,000

30 15 18

127 125 130

18 16 13

0 150 1,000

25 13 15.5

125 125 125

13 15 12

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

Landslide Susceptibility Analysis

Figure 7. Example calculations for friction angle, cohesion, and unit weight per borehole location.

borehole location by use of a weighted mean method to create a single value at each borehole location. More speciﬁcally, these values were calculated by measuring slope stability lengths and cross-sectional area for each geologic unit. The lengths were used as weights per value of cohesion, and friction angle and crosssectional area were used as weights per value of unit weight (Figure 7). By using the weighted mean values, cohesion, friction angle, and unit weight were better represented per borehole location. Properties associated with borehole locations were interpolated using the inverse distance weighting (IDW) interpolation method in ArcGIS to create the maps of the input parameters for the borehole-slope model. IDW is a deterministic non-linear interpolation method that allows for weighted average calculations for locations for which information is not known, based on adjacent locations for which information is known (Shepard, 1968). This method assumes that the closer a known value is to an unknown value, the more likely these locations are similar, and therefore, more

weight is given to points with closer proximity. Because this is a deterministic method, there is no assessment of error associated with the interpolated product (Li and Heap, 2008). An example of this calculation is shown in Eq. 3. n zi i=1 di 1 i=1 di

z (x) = n

(3)

where z(x) = calculated value for unknown point location; zi = value at known point location; and di = distance between the point of known value to the point of unknown value. Groundwater Depth Groundwater is a causative factor with landslides, and typically a higher groundwater table represents a higher risk, as it pertains to landslides (Wieczorek, 1996). Groundwater depth data were obtained from water level readings at the borehole locations, as

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

479

Southerland and Zhou

detailed in borehole logs found in the geotechnical reports. According to NOAA, significant rainfall has occurred in the months of April through September for the past 28 years (NOAA, 2018). Because of the significance of higher rainfall during these months, water level readings were extracted per borehole location corresponding to these months. As with soil strength parameters, the groundwater depth distribution map was interpolated from borehole locations using the IDW method. Map Verification The ROC curve was used to assess the performance of the landslide susceptibility maps produced from the logit models with geology-slope and borehole-slope inputs. The true positive rate (sensitivity) and false positive rate (i.e., 1-specificity) are required inputs for plotting the ROC curves; these inputs were calculated using Eqs. 4 and 5, respectively. The sensitivity and 1-specificity are based on the model’s performance at a probability cutoff value and determine how well the model predicts landslide and non-landslide locations for a specific probability value. The ROC curve is constructed by plotting sensitivity and 1-specificity for cutoff values equaling 0 percent to 100 percent (Fawcett, 2006; Gorsevski et al., 2006), as follows: Sensitivity =

TP TP + FN

1 − speci f icity = 1 −

TN TN + FP

(4)

(5)

where TN = true negatives (correctly identified nonlandslide cells); FP = false positives (incorrectly identified landslide cells); TP = true positive (correctly identified landslide cells); and FN = false negative (incorrectly identified non-landslide cells). The logit model graph will appear farther in the positive direction away from the random model’s curve if the logit model is working better than a random model. The area under the curve (AUC) is significant because it represents the accuracy of the model. The area under a random curve is always 50 percent, whereas logit models are typically within a range of 50 percent to 100 percent, with “1” equaling 100 percent accuracy. In the literature, logit models used in landslide susceptibility analyses have been validated via the ROC curve method, with accuracies reported up to 75–90 percent (Choi et al., 2012; Schicker and Moon, 2012; Devkota et al., 2013; Althuwaynee et al., 2014; Nourani et al., 2014; Regmi et al., 2014; Umar et al., 2014; Chen et al., 2016; Zhang et al., 2016; and Zezere et al., 2017).

480

Table 4. Comparison of ROC curve accuracy of geology-slope versus borehole-slope susceptibility models. AUC of ROC Curve Accuracy Failure Mechanism Colluvium circular Weathered shale Circular weathered shale planar

Geology Slope (%)

Borehole Slope (%)

62.1 53.1 62.8

79.9 52.4 68.1

The output logit equations were evaluated to determine the model’s prioritization of the inputs in MATLAB© to examine both models’ results further. Each output equation has a coefficient or weight assigned to each input (Eq. 1). This coefficient can be directly related to its prioritization. However, without a standardization technique, the coefficients cannot be correctly prioritized. This standardization subtracts the mean of all input data sets and divides it by the standard deviation of all input data sets. This process is standard when trying to compare inputs of varying ranges and units (Pampel, 2000). A standardized coefficient output option was selected in MATLAB© as part of the output for the map verification process. RESULTS Both logit models, with geology-slope and boreholeslope inputs, were assessed for each landslide failure mechanism (colluvium circular, weathered shale circular, and weathered shale planar) present in the study area. Failure mechanism–specific landslide susceptibility maps were produced by both the geology-slope and borehole-slope models. The geology-slope model makes use of readily available data and can produce rapid results, while the borehole-slope model is tedious and involves using geostatistical analysis in order to produce the strength parameter and groundwater depth map. Both models were used to produce the AUC value from their corresponding ROC curve, shown in Table 4. The AUC values demonstrate the accuracy at which both models performed for each failure mechanism. For the geology-slope model, the colluvium circular susceptibility map performed at a ROC curve accuracy of 62.1 percent. The weathered shale circular susceptibility map performed at a ROC curve accuracy of 53.1 percent, while the weathered shale planar susceptibility map performed at a ROC curve accuracy of 62.8 percent. All geology-slope models prioritized geology over the slope in the logit equations. For the borehole-slope model, the colluvium circular susceptibility map performed at a ROC curve accuracy of 79.9 percent and prioritized cohesion in the logit equation. The weathered shale circular

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

Landslide Susceptibility Analysis

Figure 8. Susceptibility map results for planar failures within the weathered shale geology, geology-slope model.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

481

Southerland and Zhou

Figure 9. Susceptibility map results for planar failures within the weathered shale geology, borehole-slope model.

482

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

Landslide Susceptibility Analysis

Figure 10. Comparison of the results from the two models (Figures 8 and 9) pixel by pixel through a Minus operation. This resulting discrepancy map has normalized to the discrepancies to a 100 percent scale.

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

483

Southerland and Zhou

Figure 11. The histogram of the discrepancies between the two models (Figure 10). The horizontal axis represents the discrepancy in percentage, and the vertical bar represents the number of pixels corresponding to a particular discrepancy range value.

susceptibility map performed at a ROC curve accuracy of 52.4 percent and prioritized friction angle in the logit equation. The weathered shale planar susceptibility map performed at a ROC curve accuracy of 68.1 percent and prioritized cohesion in the logit equation. Overall, the geology-slope model performed relatively well for the weathered shale circular and weathered shale planar failure mechanisms, and it even slightly outperformed the borehole-slope model for the weathered shale circular failure mechanism. However, the borehole-slope model signiﬁcantly outperformed the geology-slope model for the colluvium circular failure mechanism. Figures 8 and 9 show the weathered shale planar failure mechanism–speciﬁc landslide susceptibility maps created using both the geology-slope and borehole-slope logit models, respectively. We further compared the results from the geologyslope and borehole-slope logit models (Figures 8 and 9) for planar failures within the weathered shale. A Minus operation was performed, and the resulting map (Figure 10) shows the discrepancy, pixel by pixel, of the two models. The discrepancies were normalized to 100 percent. Compared with the borehole-slope model, a negative percentage indicates that the geology-slope model underestimated the landslide potential, while a positive percentage indicates that the geology-slope model overestimated the landslide potential. The majority of the map shows an error of less than 25 percent. The blue and red pixels are where the two models show the maximum discrepancy, although these areas are small. The blue area appears in two places that coincide with steep slopes, and the red pixels appear to

484

be scattered along the west and south sides of the study area. Figure 11 displays the discrepancies between the two models, which reveal a nearly normal distribution. It also reveals that the majority of the discrepancy is less than 25 percent, (i.e., in the range of −25 percent to 25 percent). However, it is noticed that the left tail (i.e., the blue pixels in Figure 10) is thicker than the right tail (i.e., the red pixels in Figure 10). This can be further interpreted to indicate that the geology-slope model has relatively underestimated the landslide susceptibility more than it has overestimated it. DISCUSSION AND CONCLUSION This study created failure mechanism–specific landslide susceptibility maps by identifying failure mechanisms present in the study area and using established slope stability models. The logit models developed in this study were used to create failure mechanism– specific landslide susceptibility maps, which has not been done in previous studies of the Colorado Springs area. The landslide susceptibility maps produced in this study provide insights for prioritizing areas of additional field investigation. In addition, the geologyslope model can be easily and rapidly applied to other study areas, since both geologic and topographic information are readily available. Based on the comparison of ROC curve accuracy of each failure mechanism–specific model (Table 4) and the spatial comparison of the results from the geology-slope and borehole-slope logit models for pla-

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

Landslide Susceptibility Analysis

nar failures within the weathered shale pixel by pixel (Figures 10 and 11), we concluded the following: The borehole-slope model significantly outperformed the geology-slope model for circular failures within colluvium. In addition, the geology-slope model performance level is relatively close to that of the boreholeslope model for the weathered shale circular and weathered shale planar failure mechanisms; in fact, it even slightly outperformed the borehole-slope model for the weathered shale circular failure mechanism. This outperformance is important to note because the geology-slope model could potentially be more practical as a result of the more readily available data, simple input parameters, and capacity for rapid analysis associated with this model. Our study overcame the limitations of previous studies (e.g., White and Wait, 2003a, 2003b; Garret, 2011). White and Wait (2003a, 2003b) interpreted the landslide susceptibility map based on geology, geomorphology, and topography knowledge. The resulting maps are binary (i.e., there would be a landslide, or there would not be). Our maps display the landslide susceptibility in the form of probability (i.e., the likelihood of a landslide). In addition to geology and topology, our borehole-slope models consider soil properties and depth to groundwater. Compared with Garrett’s study (Garrett, 2011), our models consider different failure mechanisms for different regions based on the inventory data, while Garrett (2011) assumed one failure mechanism across the study area for each susceptibility map. Borehole data and soil strength laboratory tests are costly and time consuming to obtain. On the other hand, the slope map can be easily derived from DEM, which is available globally, and geology maps are readily available throughout the United States. Therefore, the geology-slope model can be easily and rapidly applied to other study areas since geologic and topographic information is readily available. However, our study still has some limitations, including the following: (1) the potential misclassification of landslide failure mechanisms and (2) the selection of representative borehole-slope model input parameters. The slope stability models used by consulting firms in the area to determine failure mechanisms were all assumed to be correct and representative of subsurface landslide conditions in the study area. As a result, more than one failure mechanism may have been modeled for a given landslide, which may not represent the actual landslide conditions. Additionally, the landslide inventory delineated for this project was categorized based on existing slope stability models conducted by multiple geotechnical consultants. Our models’ accuracies can be improved if the inventory data sets are more extensive and if the data-

collecting methods are more consistent with different consulting companies. REFERENCES Althuwaynee, O. F.; Pradhan, B.; Park, H. J.; and Lee, J. H., 2014, A novel ensemble bivariate statistical evidential belief function with knowledge-based analytical hierarchy process and multivariate statistical logistic regression for landslide susceptibility mapping: Catena, Vol. 114, pp. 21–36. https://doi.org/10.1016/j.catena.2013.10.011. Ayalew, L. and Yamagishi, H., 2005, The application of GISbased logistic regression for landslide susceptibility mapping in the Kakuda–Yahiko Mountains, Central Japan: Geomorphology, Vol. 65, pp. 15–31. Carroll, C. J. and Crawford, T. A., 2000, Geologic Map of the Colorado Springs Quadrangle, El Paso County, Colorado. 1:24,000 Map: Colorado Geological Survey. Chen, J.; Li, Y.; Zhou, W.; Iqbal, J.; and Cui, Z., 2017, Debrisflow susceptibility assessment model and its application in semiarid mountainous areas of the Southeastern Tibetan Plateau: Natural Hazards Review, Vol. 18, No. 2, pp. 1–15. doi:10.1061/(ASCE)NH.1527-6996.0000229, 05016005. Chen, J.; Zhou, W.; Cui, Z.; Li, W.; Ma, J.; and Wu, S., 2018, Formation process of a large paleolandslide-dammed lake at Xuelongnang in the upper Jinsha River, SE Tibetan Plateau: Constraints from OSL and 14C dating: Landslides, Vol. 15, pp. 2399–2412. Chen, T.; Niu, R.; and Jia, X., 2016, A comparison of information value and logistic regression models in landslide susceptibility mapping by using GIS: Environmental Earth Sciences, Vol. 75, No. 10. doi:10.1007/s12665-016-5317-y. 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, No. 1, pp. 12–23. https://doi.org/10.1016/ j.enggeo.2011.09.011. City of Colorado Springs, n.d., Subdivision Document Reviewer: Electronic document, available at https://web1. coloradosprings.gov/subdivview/ Received 3/20/2017. Colorado Springs Utilities, 2011, LiDAR DEM, 5ft spatial resolutions and 1.5 ft vertical accuracy. Received March 23, 2017. Daniel, J., 2011, Sampling Essentials: Practical Guidelines for Making Sampling Choices: SAGE Publications, Inc. Thousand Oaks, California USA. Devkota, K. C.; Regmi, A. D.; Pourghasemi, H. R.; Yoshida, K.; Pradhan, B.; Ryu, I. C.; 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. https://doi.org/10.1007/s11069-012-0347-6. Duncan, J. M., 1996, Soil slope stability analysis. In Turner, A. K. and Schuster, R. L. (Editors), Landslides. Investigation and Mitigation, Special Report 247: National Academy Press, Washington, DC. Fawcett, T., 2006, An introduction to ROC analysis, 27, 861–874: Electronic document, available at https://doi.org/ 10.1016/j.patrec.2005.10.010 Federal Emergency Management Agency (FEMA), 2015, Major Disaster Declaration “Colorado Severe Storms, Tornadoes, Flooding, Landslides, and Mudslides (DR-4229-CO)” on July 16, 2015. https://www.fema.gov/disaster/4229

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

485

Southerland and Zhou Garrett, J., 2011, GIS Based Landslide Susceptibility Analysis of Southwestern Colorado Springs, El Paso County, Colorado: Master’s Thesis. Colorado School of Mines. 70 p. Gorsevski, P. V.; Gessler, P. E.; Foltz, R. B.; and Elliot, W. J., 2006, Spatial prediction of landslide hazard using logistic regression and ROC analysis: Transactions GIS, Vol.10, pp. 395– 415. http://dx. doi.org/10.1111/j.1467-9671.2006.01004.x. Gruntfest, E. and Huber, T., 1985, Environmental Hazards: Colorado Springs, Colorado: Department of Geography and Environmental Studies, University of Colorado, Colorado Springs, 54 p. Hill, J. J., 1974, Environmental Resource Study for Teller and El Paso Counties, Colorado, Part B: Geology, Pikes Peak Area Council of Governments, 73 p. Hosmer, D. W. and Lemeshow, S., 2000, Applied Logistic Regression, 2nd ed.: Wiley-Interscience Publication. JOHN WILEY & SONS, INC. New York, 373 p. Jenks, G. F., 1967, The data model concept in statistical mapping: International Yearbook Cartography, Vol. 7, pp. 186-190. Li, J. and Heap, A. D., 2008, A Review of Spatial Interpolation Methods for Environmental Scientists: Geoscience Australia. Geocat #68229. 17 p. Luo, H. Y.; Zhou, W.; Huang, S. L.; and Chen, G., 2004, Earthquake-induced stability analysis of Las Colinas Landslide in El Salvador: International Journal Rock Mechanics Mining Sciences, Vol. 41, No. 1, pp. 617–622. https://doi.org/10.1016/j.ijrmms.2004.03.109. Luo, H. Y.; Zhou, W.; Huang, S. L., and Chen, G., 2009, GIS-based approaches to earthquake-induced landslide hazard zonation. In Proceedings of the 2009 International Symposium on Rock Mechanics: Rock Characterization, Modelling and Engineering Design Methods. National Oceanic and Atmospheric Association (NOAA), 2018, Record Precipitation Data – Colorado Springs: Electronic document, available at https://www.weather.gov/pub/ climateCosPrecipitationRecords Nie, Y.; Li, X.; Zhou, W.; and Xu, R., 2021, Dynamic hazard assessment of group-occurring debris ﬂows based on a coupled model: Natural Hazards, https://doi.org/10.1007/s11069021-04558-3. 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. https://doi.org/10.1007/s11069-013-0932-3 Pampel, F. C., 2000, Logistic Regression: A Primer: Sage Publications, Inc. Thousand Oaks, California, USA. 96 p. Reed, L. J. and Berkson, J., 1929, The application of the logistic function to experimental data: Journal Physical Chemistry, Vol. 33, pp. 760–779. 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. https://doi.org/10.1007/s10346012-0380-2 Rowley, P. D., 2003, Geologic Map of the Cheyenne Mountain quadrangle, El Paso County, Colorado. 1:24,000 Map: Colorado Geological Survey.

486

Schicker, R. and Moon, V., 2012, Comparison of bivariate and multivariate statistical approaches in landslide susceptibility mapping at a regional scale: Geomorphology, Vol. 161– 162, pp. 40–57. https://doi.org/10.1016/j.geomorph.2012.03. 036 Scott, G. R. and Wobus, R. A., 1973, Reconnaissance Geologic Map of Colorado Springs and Vicinity, Colorado: United States Geological Survey MF-482, scale 1:62,500. Shepard, D., 1968, A two-dimensional interpolation function for irregularly-spaced data: In Proceedings of the 1968 ACM National Conference. Southerland, L., 2019, Landslide Susceptibility Mapping Using a Logistic Regression Approach: A Case Study in Colorado Springs, El Paso County, Colorado: Master’s thesis. Umar, Z.; Pradhan, B.; Ahmad, A.; Jebur, M. N.; and Tehrany, M. S., 2014, Earthquake induced landslide susceptibility mapping using an integrated ensemble frequency ratio and logistic regression models in West Sumatera Province, Indonesia: Catena, Vol. 118, pp. 124–135. https://doi.org/ 10.1016/j.catena.2014.02.005 U.S. Geological Survey (USGS), 2004, Landslide Types of Processes. Fact Sheet 2004-3072: Electronic document, available at https://pubs.usgs.gov/fs/2004/3072/fs-2004-3072.html White, J. L. and Wait, T. C., 2003a, Potential Areas of Landslide Susceptibility of Colorado Springs, El Paso County, Colorado, Map Series 42. 1:24,000 Map, 3 pages: Colorado Geological Survey. White, J. L. and Wait, T. C., 2003b, Report of Map Series 42 Colorado Springs Landslide Susceptibility Map. El Paso County, Colorado: Colorado Geological Survey, 45 p. Wieczorek, G. F., 1996, Landslides: Investigation and Mitigation, Chapter 4: Landslide Triggering Mechanisms: Transportation Research Board Special Report 247. Wu, S.; Chen, J.; Xu, C.; Zhou, W.; Yao, L.; Yue, W.; and Cui, Z., 2020, Susceptibility assessments and validations of debrisflow events in meizoseismal areas: A case study in China’s Longxi River Watershed: Natural Hazards Review, Vol. 21, No. 1, 05019005. Wu, S.; Chen, J.; Zhou, W.; Iqbal, J.; and Yao, L., 2019, A modified logit model for assessment and validation of debrisflow susceptibility: Bulletin Engineering Geology Environment, https://doi.org/10.1007/s10064-018-1412-5, 1-18 Zezere, J. L.; Pereira, S.; Melo, R.; Oliveira, S. C.; and Garcia, R. A. C., 2017, Mapping landslide susceptibility using data-driven methods: Science Total Environment, Vol. 589, pp. 250–267. https://doi.org/10.1016/j.scitotenv.2017.02. 188 Zhang, M.; Cao, X.; Peng, L.; and Niu, R., 2016, Landslide susceptibility mapping based on global and local logistic regression models in Three Gorges Reservoir area, China: Environmental Earth Sciences, Vol. 75, No. 11, pp. 1–11. https://doi.org/10.1007/s12665-016-5764-5 Zhou, W., 2021, GIS for Earth sciences. In Alderton, D. and Elias, S. A. (Editors). Encyclopedia of Geology, 2nd ed. Vol. 6: Academic Press, Oxford, U.K., pp. 281–293. Zhou, W.; Minnick, M.; Chen, J.; Garret, J.; and Acikalin, E., 2021, GIS-based landslide susceptibility analysis methods suitable for different sizes of study area: Natural Hazards Review, doi:10.1061/(ASCE)NH.1527-6996.0000485

Environmental & Engineering Geoscience, Vol. XXVII, No. 4, November 2021, pp. 471–486

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

EDITORS

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

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

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

Cover photo In memory of Jerome (Jerry) V. De Graff, 1945-2020, U.S. Forest Service, geologist, colleague, friend. Photo courtesy of the De Graff family.

Volume XXVII, Number 4, November 2021

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

ENVIRONMENTAL & ENGINEERING GEOSCIENCE