Predicting post-wildfire debris flow onset using machine learning models on multi-parameter experimental data

Abstract

The increasing frequency of wildfires, particularly in wildland-urban interface zones, has elevated the hazard posed by post-wildfire debris flows. Following combustion, surface and near-surface soils often become hydrophobic, a phenomenon that occurs predominantly on sand-based hillslopes. Rainwater and eroded soil accumulate downslope, frequently evolving into destructive debris flows. Soil hydrophobicity exacerbates erosion, distinguishing post-wildfire debris flows from natural debris flows in terms of intensity, duration, and destructive potential. Therefore, understanding the timing and conditions under which such debris flows initiate is critical. This initiation results from coupled effects among several key parameters: rainfall intensity (RI), slope gradient (δ), water entry value (

{\mathrm{\Psi }}_{\text{wev}}

\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$${\Psi }_{\text{wev}}$$\end{document}

), and grain size (D₅₀). Machine Learning (ML) techniques have become increasingly valuable in geotechnical engineering due to their ability to model complex systems without predefined assumptions. This study applied multiple ML algorithms: multiple linear regression (MLR), logistic regression (LR), support vector classifier (SVC), K-means clustering, and principal component analysis (PCA) to predict and classify outcomes from laboratory experiments that model field conditions using a rain device on various soils in sloped flumes. The results indicate that while MLR performs adequately in predicting total discharge, its accuracy in predicting erosion is limited, particularly for coarse sand. In contrast, LR and SVC achieve satisfactory accuracy in classifying failure outcomes, supported by clustering and dimensionality reduction techniques. Sensitivity analysis reveals that fine sand is highly susceptible to erosion, especially under low-intensity, prolonged rainfall. Furthermore, the first ten minutes of high-intensity rainfall emerge as the most critical period for discharge generation and slope failure. Collectively, these findings underscore the potential of machine learning to enhance post-wildfire hazard assessment and inform emergency response planning.

Keywords

Post-wildfire debris flow / Erosion prediction / Machine learning

References

[1] Addison P, Oommen T. Post-fire debris flow modeling analyses: Case study of the post-Thomas Fire event in California. Natural Hazards, 2020, 100(1): 329-343

[2] Bhattacharya B, Solomatine DP. Machine learning in soil classification. Neural Networks, 2006, 19(2): 186-195

[3] Cannon SH, Boldt EM, Laber JL, Kean JW, Staley DM. Rainfall intensity-duration thresholds for postfire debris-flow emergency-response planning. Natural Hazards, 2011, 59(1): 209-236

[4] Cannon SH, Gartner JE, Rupert MG, Michael JA, Rea AH, Parrett C. Predicting the probability and volume of postwildfire debris flows in the intermountain western United States. Bulletin, 2010, 122(1-2): 127-144

[5] Chen HX, Wang JD. Regression analyses for the minimum intensity-duration conditions of continuous rainfall for debris flows triggering in Yan'an, northern Shaanxi (China). Bulletin of Engineering Geology and the Environment, 2014, 73(4): 917-928

[6] Cui Y, Cheng D, Chan D. Investigation of post-fire debris flows in Montecito. ISPRS International Journal of Geo-Information, 2018, 8(1): 5

[7] Cui Y, Cheng D, Chan D. Investigation of post-fire debris flows in Montecito. ISPRS International Journal of Geo-Information, 2018, 8: 5

[8] DeBano, L. F. (1981). Water repellent soils: A state-of-the-art (Gen. Tech. Rep. PSW-GTR-46). U.S. Department of Agriculture, Forest Service, Pacific Southwest Forest and Range Experiment Station.

[9] Di Napoli M, Marsiglia P, Di Martire D, Ramondini M, Ullo SL, Calcaterra D. Landslide susceptibility assessment of wildfire burnt areas through earth-observation techniques and a machine learning-based approach. Remote Sensing, 2020, 12(15): 2505

[10] Doerr SH, Shakesby RA, Walsh RPD. Soil water repellency: Its causes, characteristics and hydro-geomorphological significance. Earth-Science Reviews, 2000, 51(1–4): 33-65

[11] El Naqa I, Li R, Murphy MJMachine learning in radiation oncology, 2015Springer

[12] Erzin Y, Çetin T. The prediction of the critical factor of safety of homogeneous finite slopes using neural networks and multiple regressions. Computers & Geosciences, 2013, 51: 305-313

[13] Friedman EQ, Santi PM. Debris-flow hazard assessment and validation following the Medano Fire, Great Sand Dunes National Park and Preserve, Colorado. Landslides, 2014, 11(6): 1093-1113

[14] Goh AT. Neural networks for evaluating CPT calibration chamber test data. Computer-Aided Civil and Infrastructure Engineering, 1995, 10(2): 147-151

[15] Goh ATC, Zhang WG. An improvement to MLR model for predicting liquefaction-induced lateral spread using multivariate adaptive regression splines. Engineering Geology, 2014, 170: 1-10

[16] Goh ATC, Zhang W, Zhang Y, Xiao Y, Xiang Y. Determination of earth pressure balance tunnel-related maximum surface settlement: A multivariate adaptive regression splines approach. Bulletin of Engineering Geology and the Environment, 2018, 77(2): 489-500

[17] James G, Witten D, Hastie T, Tibshirani RAn introduction to statistical learning, 2013Springer

[18] Kean JW, Staley DM, Cannon SH. 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 of Geophysical Research, 2011, 116(4): 1-21

[19] Kern AN, Addison P, Oommen T, Salazar SE, Coffman RA. Machine learning based predictive modeling of debris in the Intermountain Western United States. Mathematical Geosciences, 2017, 49(6): 717-735

[20] Leelamanie DAL, Karube J, Yoshida A. Characterizing water repellency indices: Contact angle and water drop penetration time of hydrophobized sand. Soil Science and Plant Nutrition, 2008, 54(2): 179-187

[21] Martin DA, Moody JA. Comparison of soil infiltration rates in burned and unburned mountainous watersheds. Hydrological Processes, 2001, 15(15): 2893-2903

[22] Movasat MMicro to macro investigation of post-wildfire debris flow initiation mechanisms, 2022University of California

[23] Movasat M, Tomac I. Kinematics of post-wildfire debris flow initiation mechanism: Impact of hydrophobic layer spatial variability and particle size. Canadian Geotechnical Journal, 2024, 62: 1-18

[24] Nikolopoulos EI, Destro E, Bhuiyan MAE, Borga M, Anagnostou EN. Evaluation of predictive models for post-fire debris flow occurrence in the western United States. Natural Hazards and Earth System Sciences, 2018, 18(9): 2331-2343

[25] Nyman P, Sheridan GJ, Smith HG, Lane PNJ. Evidence of debris flow occurrence after wildfire in upland catchments of south-east Australia. Geomorphology, 2011, 125(3): 383-401

[26] Parsons, A., Robichaud, P. R., Lewis, S. A., Napper, C., & Clark, J. T. (2010). Field Guide for Mapping Post-Fire Soil Burn Severity. Technical Report RMRS-GTR-243. US Department of Agriculture, Forest Service, Rocky Mountain Research Station, Fort Collins, CO

[27] Puri, N., & Jain, A. (2015, March). Correlation between California bearing ratio and index properties of silt and clay of low compressibility. In Proc. Fifth Indian Young Geotechnical Engineers Conference, Vadodara

[28] Puri N, Prasad DH, Jain A. Prediction of geotechnical parameters using machine learning techniques. Procedia Computer Science, 2018, 125: 509-517

[29] Puri N, Jain A. Correlation Between California Bearing Ratio and Index Properties of Silt and Clay of Low Compressibility . 2015;(March 2020).

[30] Qi C, Tang X. Slope stability prediction using integrated metaheuristic and machine learning approaches: A comparative study. Computers & Industrial Engineering, 2018, 118: 112-122

[31] Ritsema CJ, Dekker LWSoil water repellency: Occurrence, consequences, and amelioration, 2012Elsevier

[32] Staley, B. D. M., Negri, J. A., Kean, J. W., Laber, J. M., Tillery, A. C., Youberg, A. M., et al. (2016). Updated logistic regression equations for the calculation of post-fire debris-flow likelihood in the western United States. U.S. Geological Survey

[33] Staley DM, Negri JA, Kean JW, Laber JL, Tillery AC, Youberg AM. Prediction of spatially explicit rainfall intensity–duration thresholds for post-fire debris-flow generation in the western United States. Geomorphology, 2017, 278: 149-162

[34] Tien Bui D, Moayedi H, Gör M, Jaafari A, Foong LK. Predicting slope stability failure through machine learning paradigms. ISPRS International Journal of Geo-Information, 2019, 8(9): 395

[35] Wang L, Wu C, Gu X, Liu H, Mei G, Zhang W. Probabilistic stability analysis of earth dam slope under transient seepage using multivariate adaptive regression splines. Bulletin of Engineering Geology and the Environment, 2020, 79: 2763-2775

[36] Westerling AL, Hidalgo HG, Cayan DR, Swetnam TW. Warming and earlier spring increase Western U.S. forest wildfire activity. Science (Washington, d.c.), 2006, 313(5789): 940-943

[37] Wilder BA, Lancaster JT, Cafferata PH, Coe DB, Swanson BJ, Lindsay DN, et al.. An analytical solution for rapidly predicting post-fire peak streamflow for small watersheds in southern California. Hydrological Processes, 2021, 35(1): e13976

[38] Zhang W, Li H, Li Y, Liu H, Chen Y. Application of deep learning algorithms in geotechnical engineering: A short critical review. Artificial Intelligence Review, 2021, 54: 5633-5673

[39] Zhang W, Wu C, Li Y, Wang L, Samui P. Assessment of pile drivability using random forest regression and multivariate adaptive regression splines. Georisk: Assessment and Management of Risk for Engineered Systems and Geohazards, 2019

[40] Zhang Y, Ge T, Tian W, Liou YA. Debris‑flow susceptibility mapping using machine‑learning techniques in the Shigatse area, China. Remote Sensing, 2019, 11(23): 2801

[41] Cannon SH, Gartner JE, Wilson RC, Bowers JC, Labe JL. Storm rainfall conditions for fl oods and debris fl ows fromrecently burned areas in southwestern Colorado and southern California. Geomorphology, 2008, 96: 3-4

[42] Debano LF. The role of fire and soil heating on water repellency in wildland environments: a review. J Hydrol#[Internet]. 2000;231–232(0):195–206. Available from:http://www.sciencedirect.com/science/article/pii/S0022169400001943

Funding

Hellman Family Foundation

University of California

RIGHTS & PERMISSIONS

The Author(s)