Caractérisation des rugosités par données satellitaires pour les calculs d’inondations

Flood simulation commonly relies on two-dimensional (2D) hydrodynamic models, which must be carefully calibrated and validated using past flood events to ensure accurate forecasting. A critical component of this calibration process is the spatial distribution of the roughness parameter, which significantly influences simulated water depths and velocities. In the minor bed, the hydrodynamic models are calibrated using time-series data from stage-gauging stations. For floodplains, few data are available, and the roughness spatial distribution is often poorly known and typically inferred from land-use classifications using empirical values drawn from expert judgment. Hydrodynamic models' calibration could benefit from incorporating the maximum available observational data.In this context, satellite remote sensing can be leveraged as a valuable source of information. Synthetic Aperture Radar (SAR) imagery, in particular, is capable of detecting surface water extents even under cloud cover, making it highly suitable for flood monitoring. This thesis aims to develop a methodology to calibrate floodplain roughness coefficients in 2D hydrodynamic models, leveraging satellite-derived observations.The first step aims to recover information from the satellite observations to calibrate the flood models. Thus, this thesis proposed an evaluation of the extraction of flood extent maps and water depth fields from SAR imagery, along with an evaluation of the associated uncertainties arising from method choices and hyperparameter settings. The flood extent maps should be compared with the hydrodynamic model to calibrate it. Thus, in this thesis, we developed a step-by-step methodology to guide the selection of appropriate performance measures to compare simulated and observed flood extent maps. Then, the uncertain SAR-derived flood observations are integrated into the hydrodynamic modeling framework through data assimilation techniques to enhance the calibration of spatially distributed roughness coefficients on floodplains.By combining remote sensing data, uncertainty quantification, and hydrodynamic modeling, this thesis proposes an approach to enhance flood model calibration, with the potential to improve flood forecasting and risk management practices.