Uni-Temporal Multispectral Imagery for Burned Area Mapping with Deep Learning

Accurate burned area information is needed to assess the impacts of wildfires on people, communities, and natural ecosystems. Various detection methods have been developed using satellite remote sensing measurements with wide coverage frequent revisits. Our study aims expound capability deep learning (DL) models for automatically mapping areas from uni-temporal multispectral imagery. Specifically, several semantic segmentation network architectures, i.e., U-Net, HRNet, Fast-SCNN, DeepLabv3+, machine (ML) algorithms were applied Sentinel-2 imagery Landsat-8 in three wildfire sites two different local climate zones. The validation results show that DL outperform ML cases compact scars, while seem be more suitable dispersed burn boreal forests. Using images, U-Net HRNet exhibit comparatively identical performance higher kappa (around 0.9) one heterogeneous Mediterranean fire site Greece; Fast-SCNN performs better than others over 0.79 forest various severity Sweden. Furthermore, directly transferring trained corresponding data, dominates test among can preserve high accuracy. demonstrated make full use contextual capture spatial details multiple scales fire-sensitive spectral bands map areas. only a post-fire image, not provide automatic, accurate, bias-free large-scale option cross-sensor applicability, but also potential used onboard processing next Earth observation satellites.