Multi-grained wildfire damage estimation from satellite vegetative scenario by fuzzy decision tree

Among climate effects, fires are reported to be the most catastrophic events both economically and environmentally, in fact, recent statistics report about 340,000 hectares (ha) burnt in countries of the European Community during 2020, corresponding to an area 30% larger than Luxembourg. In worst cases, fire-affected areas will be permanently damaged. Then, there is a need for solutions to help institutions and researchers to keep the environment under monitoring, assess the damage severity and help planning burned area recovery. To this purpose, this paper presents a smart monitoring framework that bridges spectral image clustering with soft computing techniques to describe the effective fire damages in the monitored area. The approach collects images from Sentinel-2 satellite, assesses Spectral Indices (SIs) from them, then by clustering, divides the area into different sub-regions according to their vegetative features. Then these sub-regions are parsed by a fuzzy decision tree, able to interpret the damage of fire severity in each sub-region. Experiences show that by dividing the area into many sub-regions, fire damage can be detected at different levels of granularity, providing a detailed map of the most damaged sub-areas to plan, for example, ad-hoc recovery interventions.