Using comprehensive field data for validating large-scale hydrodynamic hindcasting of an extreme flood event

Hydrodynamic modelling is a crucial tool for simulating large-scale floods, yet validation remains challenging due to discrepancies between modelled and observed datasets, especially in urbanized floodplains. The July 2021 Vesdre Valley flood in Belgium provides a unique opportunity for rigorous validation, supported by an extensive post-flood survey with over 16,000 high-water marks recorded by the Service public de Wallonie (SPW). This study evaluates the performance of a high-resolution 2D hydrodynamic model (WOLF) by comparing simulated and observed flood extents and water depths. To address discrepancies arising from structural artifacts in built-up areas, three harmonization approaches—Buffered-Buildings, Filling-of-Holes, and Inverse Adaptation—are tested. Validation using the Jaccard similarity index demonstrates strong spatial agreement, with a domain-average value of 0.85. Water depth comparisons reveal consistent agreement across most sectors, though localized deviations highlight the challenges of reconciling field data and model outputs. Findings suggest that even with extensive field data available, comprehensive validation remains a challenge, thereby necessitating careful consideration and application of appropriate post-processing for effective model validation.