Abstract
Two-dimensional hydrodynamic models are computationally expensive. This drawback can limit their application to solving problems requiring real-time predictions or several simulation runs. To resolve fine-scale physical processes, allowing for local impact assessments, downscaling techniques are essential. Super-resolution is an innovative technique that upscales the resolution of an image and thus enables to reconstruct high-fidelity images from low-resolution data. This study performs super-resolution analysis for spatial downscaling of hydrodynamic data using various deep learning techniques to reconstruct high resolution flow fields from low-resolution flow field data. It increases the spatial resolution of coarsened water depth and unit discharge norm from 4 m to 80 cm. The training data for these models was generated using a physically based hydrodynamic model. To evaluate their performance and accuracy, multiple tests were conducted using synthetic events. Our experiments indicate that these models successfully predicted water depths in the testing flood scenario for the dynamic case but could not preserve the steady states during the reconstruction. Furthermore, these models cannot satisfactorily generalize to flood scenario outside the training datasets with different boundary conditions. The results demonstrate that the proposed models are up to 30 times faster than the hydrodynamic model and promising in terms of accuracy. Therefore, it bridges the gap between detailed flood modelling and real-time applications.