Data driven model benchmark reveals key factors in wildfire debris flow prediction
Towards Explainable Benchmarking for Data-driven Post-Wildfire Debris Flow Prediction
Machine Learning
Summary
Wildfires can lead to dangerous debris flows when rain hits burned areas, causing landslides that threaten homes and roads. The authors note that existing computer models to predict these flows vary a lot, making it hard to compare or understand what really causes them. They created a standard way to test and compare many prediction methods fairly. They also developed a technique to find which environmental factors, like weather or soil, are most important for causing debris flows in different places. Their tools and data are shared online for others to use.
What this means in practice
- •For disaster response teams: Use the benchmarked models to improve forecasting of debris flows and prioritize emergency actions in wildfire-affected regions.
- •For environmental engineers: Integrate the feature selection method to identify critical risk factors for designing more effective debris flow mitigation measures.
Authors
Zhisheng Qi, Li Zhu, Utkarsh Sahu, Douglas Tommey, Josh Roering, Yu Wang
Abstract
Post-wildfire debris flows (PFDFs) are destructive sediment-laden hazards triggered when intense rainfall strikes recently burned terrain, destabilizing hillslopes and threatening infrastructure, local economies, and community safety. Data-driven methods have been proposed to learn predictive patterns directly from historical PFDF observations. However, the current research landscape of data-driven PFDF prediction remains highly fragmented across feature spaces, model architectures, and evaluation protocols, making rigorous comparison and the derivation of scientific insights difficult. Moreover, existing studies lack a systematic investigation into the relative importance of heterogeneous factors (e.g., meteorological conditions, terrain characteristics, soil properties, and burn severity) in triggering PFDF. To address these limitations, we present a unified benchmark for data-driven PFDF prediction, enabling fair and comprehensive evaluation across diverse models and feature configurations. Furthermore, to better understand the underlying drivers of PFDF formation, we propose a reinforcement learning-based feature selection framework that identifies factors whose perturbations render positive and negative events indistinguishable, thereby discovering the regional underlying mechanisms of PFDF occurrence across regions. Our code and benchmark are publicly available at https://github.com/KINDLab-Fly/PFDF-Benchmark.