Climate-informed flood risk mapping using a GAN-based approach (ExGAN)

This study develops a class of robust models for flood risk mapping in highly vulnerable regions by focusing on accurately depicting extreme precipitation patterns aligned with regional climates. By implementing sophisticated hydrodynamics modeling and advanced probabilistic approaches, the present work underscores the efficacy of physical-based methodologies in the flood risk assessment. We propose a machine learning based ExGAN to address the challenge of synthesizing extreme precipitation scenarios which faithfully capture the nuances of local climatology. It is expected that through refined temporal disaggregation, the ExGAN approach exhibits exceptional proficiency in replicating a diverse spectrum of extreme precipitation patterns specific to the vulnerable region under scrutiny. Therefore, using these synthesized scenarios as inputs in a meticulously calibrated hydrological model would enable a comprehensive and detailed flood risk mapping exercise. To demonstrate the robustness of the developed mode, we perform a rigorous testing and validation within the highly susceptible Martil river basin, situated in the northern Mediterranean region of Morocco. The obtained results confirm that extending return periods would provide invaluable insights into the expanding geographical expanse of at-risk areas, clarifying the evolving landscape of vulnerability rather than merely amplifying inherent risk levels. Comparisons against the conventional Monte-Carlo sampling are also carried out in this study and the obtained results highlight significant overestimations within the latter, emphasizing the imperative need to account for diverse uncertainties beyond the basic sampling strategies within the realm of hydrodynamic modeling.