Forest Burned Area Mapping using Bi-Temporal Sentinel-2 Imagery Based on Convolutional Nueral Network (Case Study: Golestan’s Forest).

Forest areas are profoundly important for the planet Earth due to the considerable advantages they offer. Therefore, it is essential that the forest areas are closely monitored, but unfortunately, in the past decades we have witnessed some forest fires that have led to missing some parts of forest areas. Mapping and estimation of the burned forest areas are critical to the next decision makings. In this case, remote sensing can be of great help. This paper presents a method to estimate burned areas on the Sentinel-2 imagery using CNN algorithm. This framework touches change detection using pre/post-fire datasets. The proposed CNN architecture has four convolution layers that are able to extract deep features. We have investigated the performance of the proposed method by visual and numerical analysis. The case study of this research is Golestan’s forest which is located in north of Iran. The result of the `burned area detection shows that the proposed method produces a performance which is more than 91.35% by Overall Accuracy.