Ecosystem-specific calibration of satellite-derived spectral indices improves the characterization of fire severity

[EN] Accurate estimation of fire severity is critical for integrated post-fire management in Mediterranean ecosystems. While satellite-derived products are widely used, calibration with field data has often lacked ecosystem-type differentiation. This study assessed the performance of global and ecosystem-specific calibration of Sentinel-2 spectral indices in assessing fire severity within a 25,228 ha wildfire in Spain. We established 69 Composite Burn Index (CBI) plots to measure fire-induced ecological impacts at the site, vegetation, and soil level in the dominant ecosystem types (conifer forests, broadleaf forests, and shrublands). We used Ordinary Least Squares (OLS) regression models to evaluate the performance of nine spectral indices derived from Sentinel-2 imagery. Fire severity thresholds were computed for each index using both globally pooled and ecosystem-specific calibrations. We assessed the spatial distribution of fire severity classes to quantify differences in burned area estimates between the two approaches. Fire effects were strongly differentiated between ecosystems, with the highest fire severity observed in conifer forests and shrublands, while the lowest fire severity was found in broadleaf forests. Although the Relativized Burn Ratio (RBR) exhibited a high correlation with CBI at the site level using the entire pooled dataset, the ecosystem-specific calibration consistently improved model performance. Differences in fire severity thresholds between global and specific models led to significant changes in mapped burned areas, particularly in shrublands, where ecosystem-specific calibration enhanced the high-severity detection. These findings underscore the need to account for ecosystem type when assessing fire severity, supporting more accurate and targeted post-fire restoration strategies in heterogeneous Mediterranean landscapes