Overcoming cloud obstruction: Fast forest-damage assessment in post-tropical cyclone optical remote sensing

Timely mapping of damaged forests is critical for disaster assessment. However, remote sensing data immediately after natural hazards is always scarce and susceptible to cloud contamination, hindering holistic assessment of damaged forests in a timely manner. Herein, we propose a novel method to map damaged forests obscured by clouds in post-hazard images by taking the September 2024 typhoon Yagi in Hainan Island, China as an example. Our approach uniquely integrates observed forest damage in cloud-free pixels with its influencing factors (the maximum wind speed and cumulative rainfall during the typhoon, terrain (elevation, slope and aspect), and canopy height) to interpolate the relationship into cloud-covered pixels by using three mainstream machine learning models (XGBoost, artificial neural networks and random forest). We found severe forest damage in the Northeast Hainan and the total area of the typhoon-damaged forests accounts for 12.8 %–15.5 % of the island's forest cover. This method can also be used for fast mapping of forest damage in partially available remote sensing images after other major natural hazards such as wildfires and landslides.