Satellite-based burn severity mapping and evaluating the transferability of Copernicus EMS data using machine learning approaches

Abiotic and biotic factors in forest ecosystems can all be significantly and immediately impacted by forest fires. Additionally, fires pose a long-term concern because they release greenhouse gases (GHGs) into the atmosphere, damage habitat, cause soil erosion, and affect local and global temperatures. In the absence of sufficient information on the damaged forests, such as location, area, and burn severity, issues in policy decisions for restoration inevitably arise. In this study, burned areas and severity were mapped using eight spectral indices derived from Sentinel 2 MSI images using machine learning approaches (Random Forest (RF) and Support Vector Machine (SVM)). The dataset from Copernicus Emergency Management Service (CEMS) was employed as the reference truth for burned area and severity. Our approaches were tested for two study sites that had a similar meteorological environment (dry season) and species (coniferous vegetation).  This study presents a novel methodology for mapping burned areas and severity using Sentinel-2 MSI data and CEMS data, aiming at achieving mapping accuracy and transferability. RF performed better than SVM when classifying pixels within heterogeneous regions. The Normalized Burn Ratio (NBR) and Green Normalized Difference Vegetation Index (GNDVI) were quite significant in determining the severity of a fire, indicating that they might be useful in identifying senescent plants. The findings also demonstrated that the CEMS dataset can be used as a reference for classifying fire damage in other regions. The use of this approach makes it possible to quickly and accurately map the extent of the damage caused by forest fires and has applicability for other disasters.