Satellite Based Burn Severity Mapping Using Machine Learning Approaches

This study, the burned areas severity were mapped using eight spectral indices that were computed from Sentinel 2 satellite data and machine learning approaches, Random Forest (RF) and Support Vector Machine (SVM). Two study sites with similar climatic conditions (dry season) and species (coniferous vegetation) were investigated, and the Copernicus Emergency Management Service (CEMS) dataset (EMSR448) was chosen as the ground truth. Pixels from classes with similar features could be classified more accurately by RF than by SVM. The findings also demonstrated the transferability of the CEMS dataset as an acceptable for classifying fire damage in different regions. This approach can be used for other disasters and allows for the quick and precise mapping of the extent and intensity of severe damage caused by forest fires.