Sentinel-2 RGB and NIR bands simulation after fire events using a multi-temporal conditional generative adversarial network

In remote sensing applications, optical images are widely used to monitor land changes. However, clouds, haze, or smoke hide the area below and, therefore, limit the use of optical data to favorable weather conditions. Since the SAR signal can penetrate through clouds, haze, or smoke, it has been recently proposed to combine optical images and SAR data to overcome this limitation. In this report, we investigate a deep-learning model based on a multi-temporal conditional generative adversarial neural network that generates optical images from SAR data, based on optical cloudfree images and SAR data previously acquired. Quantitative and qualitative results over the region of Goulburn, Australia, are also provided to evaluate the effectiveness of this multi-temporal approach in monitoring vegetation changes after fire events. Software code is publicly available