Multi-feature fusion of L-band polarimetric SAR for accurate burned-area mapping: a case study of the 2020 LNU complex fire

Accurate burned-area mapping is critical for wildfire monitoring and post-event impact assessment, yet previous SAR-based studies have often focused on individual descriptors or limited feature combinations, leaving the complementarity and relative efficacy of L-band polarimetric feature subsets largely unexplored. To address this gap, this study develops a multidimensional SAR-based framework for delineating forest burned area using L-band fully polarimetric SAR observations. Five categories of pre-/post-fire features are systematically examined, including backscatter intensity, interferometric coherence, burn-related indices, polarimetric decomposition parameters, and grey-level co-occurrence matrix (GLCM) texture metrics. A random forest classifier is employed to quantify the contribution of individual features and their combinations, with burned-area masks derived from Sentinel-2 multispectral imagery used as the validation reference. Results demonstrate that coherence provides the strongest single-feature performance, whereas multi-feature fusion further enhances mapping robustness. In particular, the combination of coherence and backscatter achieves the highest extraction accuracy (86.07%), whereas the integration of coherence and GLCM features significantly reduces omission errors by 5.8% and improves boundary completeness (IoU: 80.06%) by resolving spatial heterogeneity. Overall, this study clarifies the complementarity between SAR structural stability and texture features, and proposes an effective multi-feature fusion strategy for fine-scale, all-weather burned-area mapping under complex terrain and persistent obscuration.