Unsupervised Burned Area Mapping in a Protected Natural Site. An Approach Using SAR Sentinel-1 Data and K-mean Algorithm

De Luca, Giandomenico; Modica, Giuseppe; Fattore, Carmen; Lasaponara, Rosa

This paper is focused on investigating the capabilities of SAR S-1 sensors for burned area mapping. To this aim, we analyzed S-1 data focusing on a fire that occurred on August 10(th,) 2017, in a protected natural site. An unsupervised classification, using a k-mean machine learning algorithm, was carried out, and the choice of an adequate number of clusters was guided by the calculation of the silhouette score. The ΔNBR index calculated from optical S-2 based images was used to evaluate the burned area delimitation accuracy. The fire covered around 38.51 km(2) and also affected areas outside the boundaries of the reserve. S-1 based outputs successfully matched the S-2 burnt mapping.