143-1 Machine Learning Framework for Enhanced Post-Wildfire Debris Flow Prediction Across California Fire Events: Integrating Temporal Dynamics and Advanced Geomorphological Parameters

Session: A Showcase of Student Research in Geoinformatics and Data Science (Posters)

Presenting Author:

Authors:

Abstract:

Wildfire, when followed by large precipitation events, has caused debris flows in slanted terrain of the US west coast in the post-wildfire debris flow is a critical hazard, especially in the US west coastal area, where the interface between wildland-urban areas is expanding. Current USGS logistic regression models demonstrate limited predictive capability, constraining effective hazard mitigation efforts. This study develops a comprehensive ensemble machine learning framework to enhance debris flow probability prediction across diverse Pacific Coast settings.

The enhanced parameter framework incorporates temporal dynamics through time-since-fire decay functions and vegetation recovery trajectories, capturing the evolving nature of post-fire hazard conditions. Advanced burn severity metrics including Relativized Burn Ratio (RBR) and Relativized difference Normalized Burn Ratio (RdNBR) replace standard differenced Normalized Burn Ratio (dNBR) measurements. Comprehensive terrain characterization utilizes Topographic Wetness Index (TWI) and curvature metrics beyond conventional slope gradients, while physics-based soil infiltration capacity is derived from detailed Soil Survey Geographic Database (SSURGO) properties. Multi-temporal rainfall analysis encompasses antecedent moisture conditions and sub-hourly intensity metrics to better capture precipitation patterns that trigger debris flows.

Multiple machine learning approaches including Random Forest, XGBoost, and LSTM networks were evaluated to identify the most effective prediction method. Model performance was assessed through cross-validation and threat score analysis, enabling direct comparison with existing USGS frameworks. Feature importance analysis identified temporal and spatial predictors most critical for debris flow initiation across West Coast environments.

Results demonstrate substantial improvements over current methods, with enhanced burn severity indices providing significant advancement in prediction capability. Temporal decay functions effectively capture the dynamic nature of post-fire hazard evolution, while advanced terrain metrics reveal critical hydrological processes governing debris flow initiation in steep coastal watersheds. This research provides emergency managers and land-use planners with significantly improved tools for post-wildfire risk assessment across the West Coast region.

Geological Society of America Abstracts with Program. Vol. 57, No. 6, 2025

doi: 10.1130/abs/2025AM-5769

© Copyright 2025 The Geological Society of America (GSA), all rights reserved.