An Unsupervised Burned Area Mapping Approach Using Sentinel-2 Images

The frequency and severity of large, destructive fires have increased in the recent past, with extended impacts on landscape, human population, ecosystems. Earth observations provide a means for frequent, wide coverage accurate monitoring fire impacts. This study describes an unsupervised approach mapping burned areas from Sentinel-2 satellite imagery, which is based multispectral thresholding, introduces adaptive thresholding method. It takes into account localized variability spectral responses two-phase approach. first phase detects that are high probability, while second adaptively adjusts this preliminary by expanding refining its boundaries. resulting classification contains two main classes interest: unburned. latter further classified four (4) impact classes, according to Copernicus Emergency Management Service (CEMS) NASA United States Geological Survey (USGS)’s widely acknowledged nomenclature examples. Three distinct wildfire events assessed, occurred during summers 2020 2021 Greece Portugal. accuracy calculated juxtaposing outputs area validation maps created through photointerpretation very high-resolution (VHR) imagery. corresponding CEMS On-Demand Mapping products also juxtaposed against comparison purposes. assessment showcases closely follows capacity provided (e.g., kappa coefficient—k—of proposed 0.91, 0.83 processed, achieve k 0.94, 0.93 0.8, respectively). considers affected areas’ response; thus, it generalizes well different areas, e.g., characterized land cover types. seems offer effective wildfire-induced changes, can be incorporated used forest management services decision support systems complementary maps.