Assessing the burn severity of wildfires by incorporating vegetation structure information

Abstract

In previous studies, the implementation of vegetation structure information remains underutilized during the assessment of burn severity. A PolSAR-based burn severity assessment model was proposed by incorporating polarimetric decomposition features. Based on coherence matrix, three polarimetric decomposition features (i.e. Entropy, Anisotropy, and Alpha) which might be related with some field variables of burn severity (e.g. percent foliage altered, percent change in cover, percent canopy mortality, and percent tree mortality) were extracted. Then, a sensitivity analysis of 20 PolSAR features was performed and a principal component analysis (PCA) was applied. Finally, the combination of polarimetric decomposition features and new PCA vectors were used as inputs and CBI values as outputs of random forest algorithm to assess burn severity. The Jinyun Mountain in the Chongqing municipality of China was used as study area and the C-band of dual-polarization SAR data acquired from Sentinel-1 satellite were used as remotely sensed data. The sensitivity analysis of PolSAR features showed that H, A, and α features exhibited higher correlations with CBI values, compared to SAR indices. For proposed model, the R was 0.60 and the RMSE was 0.55. This study offered a new research perspective for future investigations on PolSAR-based burn severity assessment of wildfires.

Keywords

• Burn severity
• polarimetric decomposition features
• SAR indices
• PolSAR data
• Wildfires