Fusion of SAR and Multispectral Satellite Images Using Multiscale Analysis and Dempster-Shafer Theory for Flood Extent Extraction

Monitoring flood extent by means of Synthetic Aperture Radar (SAR) images has become a very common practice among decision makers and planners in disaster management as these images provide wide area coverage in extreme weather conditions. However, due to the satellite revisit time, their availability hinders their efficient use in disaster management. To capitalize on SAR images characteristics, this work considers both SAR and optical multispectral (MS) images, and proposes a novel method for SAR and optical image fusion in application to flood extent monitoring, which is based on two main steps: 1- Extraction of water pixels from the pre- and post-flooding images using a Modified Water Index (MWI) for water bodies identification from optical MS images and the Structural Feature Set (SFS) texture measurement for homogeneous areas extraction from SAR images, and 2- Applying the Max-Tree structure to estimate mass functions based on the multiscale and the multishape analysis of the input features map which are subsequently incorporated into the fusion module using Dempster-Shafer theory (DST). The results obtained in the evaluation of the proposed fusion method for three flood events characterized by different satellite image scenarios demonstrate the benefits of the multiscale DST fusion strategy in terms of chosen metrics in the classification of water body and monitoring of flood extent.