Advances in fire severity assessment in forest ecosystems: remote sensing metrics with physical-basis versus spectral indices

Abstract Wildfires are becoming more frequent and severe in Mediterranean ecosystems, underscoring the urgent need for forest managers to rely on rapid and reliable methodologies to map fire-induced impacts and support post-fire management efforts. This study evaluates the accuracy and transferability of conventional spectral indices and physically-based remote sensing metrics, both retrieved from pre- and/or post-fire Sentinel-2 Level-2A scenes, for assessing photointerpreted fire severity (n = 646) through high-resolution satellite imagery (SPOT 6/7). The study sites are forest ecosystems located across wide environmental gradients and affected by four large wildfires (> 10,000 ha) that occurred in Spain during summer 2022. The physically-based metrics included the ratio of the fractional vegetation cover (FCOVER ratio) retrieved from the inversion of radiative transfer models (RTMs) and the canopy fraction burned (CFB) retrieved from multiple endmember spectral mixture analysis (MESMA). The calculated spectral indices were the dNBR, RBR, and the recently-proposed dNBR-EVI. A Random Forest classification algorithm was implemented to estimate categorized fire ecological impact levels (low, moderate and high) in a fivefold nested cross-validation approach from physically-based metrics and spectral indices. Findings highlighted the enhanced effectiveness of physically-based metrics, along with the dNBR-EVI index, as compared to conventional spectral indices in classifying fire severity across broadleaf and conifer forests, as well as with pooled data from both forest types. Particularly, the CFB (overall accuracy = 84.40%; Kappa = 0.77) minimized the underestimation of the high-severity class and showed high accuracy in classifying moderate fire severity, a category where traditional spectral indices showed significant limitations. The findings underscore the potential value of physically-based metrics and new spectral indices combined with photointerpretation of high-resolution satellite and aerial imagery as a rapid, scalable, and ecologically relevant approach to fire severity assessment.