FireRisk-Multi: A Dynamic Multimodal Fusion Framework with Geographic Feature-Driven Weighting for High-Precision Wildfire Risk

This study presents an integrated wildfire risk assessment framework combining multi-source geospatial data (NAIP imagery, SRTM topography, ERA5 meteorology, and MODIS NDVI) with machine learning. Our methodology features a novel weighted fusion approach for risk modeling, validated across 49 US regions. The system achieves 87% prediction accuracy with SHAP-interpretable feature importance, offering actionable insights for land management and emergency planning.