CNN-based burned area mapping using radar and optical data

In this paper, we present an in-depth analysis of the use convolutional neural networks (CNN), a deep learning method widely applied in remote sensing-based studies recent years, for burned area (BA) mapping combining radar and optical datasets acquired by Sentinel-1 Sentinel-2 on-board sensors, respectively. Combining active passive into seamless wall-to-wall cloud cover independent algorithm significantly improves existing methods based on either sensor type. Five areas were used to determine optimum model settings sensors integration, whereas five additional ones utilised validate results. The CNN dimension data normalisation conditioned observed land class type (i.e., or radar). Increasing network complexity number hidden layers) only resulted rising computing time without any accuracy enhancement when BA. optimally defined within joint active/passive combination allowed (i) BA with similar slightly higher those achieved previous approaches (Dice coefficient, DC 0.57) (DC 0.7) (ii) eliminating information gaps due cover, typically optical-based algorithms.