Rapid wildfire burned-area mapping using fused sentinel-1 SAR and sentinel-2 optical imagery

Rapid and reliable burned-area mapping for wildfires remains constrained by cloud/smoke interference and insufficient workflow automation. This study implements an automatic training-sample generation strategy and integrates Sentinel-1 SAR and Sentinel-2 multispectral imagery to extract and map burned areas. First, texture features that capture wildfire-induced changes are used to derive training samples for burned areas; unburned samples are identified and extracted based on the normalized difference vegetation index (NDVI). A multisource feature set—comprising raw spectral bands, spectral indices, and SAR polarimetric features—is then constructed, and classification mapping is performed using an object-based image analysis (OBIA) approach. We applied the method to the wildfire event that occurred in January 2025 in the Pacific Palisades area near Los Angeles, California (Palisades Fire), and evaluated model transferability on the Gifford Fire in August of the same year. The results show that the Random Forest classifier achieved the highest overall accuracy for the Palisades Fire (97.8%); moreover, the model exhibited good transferability to the Gifford Fire, with an overall accuracy of 93.2%.