Advancing forest fire burn severity and vegetation recovery assessments using remote sensing and machine learning approaches

Burn severity assessment and initial post-fire vegetative response dynamics are critical for understanding the ecological impacts of wildfires and informing restoration efforts. This study investigates the March 2024 Forest fire in Yajiang County, Sichuan Province, China, using Sentinel-2 spectral indices and Random Forest (RF) modeling to enhance burn severity classification and quantify early post-fire vegetative response dynamics. Indices sensitive to fire-induced vegetation and soil changes, including the Differenced Normalized Burn Ratio (dNBR), Burn Area Index for Sentinel-2 (dBAIS2), Relativized Burn Ratio (RBR), and Relative differenced Normalized Burn Ratio (RdNBR), were used to assess burn severity. Vegetation status and early-stage regrowth were monitored using the Normalized Difference Vegetation Index (NDVI) and its kernel-based variant (kNDVI). The RF model achieved a validation accuracy of 90.03%, outperforming traditional threshold-based methods by reducing misclassification rates by 3.03% and capturing complex, non-linear interactions in fire severity classification. Key findings indicate that dNBR identified large-scale vegetation loss in high severity burned zones, while dBAIS2 effectively captured soil exposure and initial regrowth signals. RdNBR excelled in monitoring early post-fire vegetation response, particularly in moderate-severity zones. Despite signs of early regrowth, high-severity areas exhibited slow regrowth, highlighting the need for targeted restoration strategies. This study underscores the advantage of integrating spectral indices with machine learning for more accurate and interpretable burn severity assessments, offering a scalable framework for adaptive fire management. Recommendations include leveraging high-resolution data, incorporating ancillary environmental variables, and adopting long-term monitoring frameworks to support global ecological resilience in fire-prone landscapes.