Dynamic Prediction of Post-Wildfire Landslide Susceptibility: A Machine Learning Approach

This study presents a novel machine learning approach for predicting post-wildfire landslide susceptibility, with a focus on Los Angeles County (LAC). Using a large dataset encompassing topographical, meteorological, and soil characteristics, we tested several machine learning models. XGBoost emerged as the most effective, achieving $\mathbf{9 8. 6 6 \%}$ accuracy and a 0.9940 ROC AUC score. The model was applied to diverse weather scenarios in LAC, revealing significant spatial and temporal variability in landslide risk. Severe weather events, particularly those involving heavy precipitation, were found to dramatically increase landslide probability across the county. For instance, during a simulated Category 4 hurricane, over $35 \%$ of the county’s area was classified as high-risk, compared to just 5% under normal conditions. This research provides a dynamic tool for disaster preparedness and risk management and emphasizes the importance of weather-specific assessments for effective hazard mitigation.