Modeling post-hurricane coarse woody fuels using machine learning and multi-source remote sensing data to support wildfire management in wildland-urban interface

Hurricanes rapidly alter forest fuel structure and composition, with associated effects on wildfire behavior across forest ecosystems. Existing forest-fuel research on disturbance impacts mostly focuses on natural forests and is often limited by temporal and spatial resolution. Wildland-urban interface (WUI) areas, which are expanding rapidly, are highly vulnerable to wildfires. In post-hurricane conditions, rapid assessment of downed woody fuels in WUI areas using high-resolution remote sensing data and advanced machine learning becomes crucial when physical access is limited. We developed a semi-automated and scalable GeoAI framework to support rapid post-disturbance fuel assessment and applied it to a forested WUI landscape centered around Asheville, North Carolina, USA, which was impacted by Hurricane Helene. This framework fuses publicly available datasets (Sentinel-2, L-band SAR, biomass, weather, soil, and LiDAR) to detect fallen trees, estimate biomass loss, and map downed woody fuel accumulation. Results show that hurricane wind intensity was not sufficient to explain forest damage or fuel accumulation. Fuel accumulation was associated with hydroclimatic drivers, forest structural conditions, and topographic conditions, resulting in ∼3.55 million metric tons of downed woody fuel across the forested portion of the study area following Hurricane Helene. High fuel-accumulation hotspots were widespread in cove and oak forests and in WUI areas. However, high fuel-accumulation hotspots were scattered across urban and ruderal forests and were rare in riparian zones, highlighting the importance of spatially targeted debris management actions. The model performed consistently across spatial cross-validation folds, highlighting the framework's potential for regional-scale assessment of hurricane-induced fuel dynamics.