Combined Use of Sentinel-1 and Sentinel-2 for Burn Severity Mapping in a Mediterranean Region.

The present study is focused on investigating the capabilities of the combined use of synthetic aperture radar (SAR) Sentinel-1 (S1) and optical Sentinel-2 (S2) for burn severity mapping. For this purpose, a fire that occurred in August 2018 in southern Portugal was analyzed. The composite burn index (CBI) was used to visually classify geo-referenced photographs in the field and create the training data for image classification. A supervised classification was carried out using the machine learning random forests (RF) algorithm, on which the optimization of the parameters setting was carried out through an exhaustive grid search approach. In order to assess the advantages of combining optical and SAR data, and the importance of each band, the approach was tested separately on three data combinations (S1, S2 and S1 + S2) and feature importance was computed to evaluate the contribution of each input layer. The multi-class F-score, used to assess the accuracy of the map, reached a value of 0.844 when both the datasets were combined (S1 + S2), compared with the values 0.514 and 0.805 achieved by only SAR (S1) and only optical (S2), respectively.