Using Multi-Source DEM Super-Resolution for Improved Inundation Depth Estimation

Kurzfassung

Accurate terrain information is an essential component for reliable flood depth estimation from synthetic aperture radar (SAR) imagery. This study proposes a deep learning-based Digital Elevation Model (DEM) super-resolution approach, integrating remote sensing data to enhance the resolution of globally available bare-earth elevation models. A collaborative loss function is introduced to focus learning on key hydrologic features. Evaluations against real-world flood sites in Germany indicate that using high-resolution super-resolved terrain models can improve the estimation of water depth compared to traditionally interpolated counterparts.