Spatially distributed calibration of a hydrological model with variational optimization constrained by physiographic maps for flash flood forecasting in France

Abstract. This contribution presents a regionalization approach to estimate spatially distributed hydrologic parameters based on: (i) the SMASH (Spatially distributed Modelling and ASsimilation for Hydrology) hydrological modeling and assimilation platform (Jay-Allemand, 2020; Jay-Allemand et al., 2020) underlying the French national flash flood forecasting system Vigicrues Flash (Javelle et al., 2019); (ii) the variational assimilation algorithm from (Jay-Allemand et al., 2020), adapted to high dimensional inverse problems; (iii) spatial constraints added to the optimization problem, based on masks derived from physiographic maps (e.g., land cover, terrain slope); (iv) multi-site global optimization, which targets multiple independent watersheds. This method gives a regional estimation of the spatially distributed parameters over the whole modeled area. This study uses a distributed rainfall-runoff model with 4 parameters to calibrate, with a spatial resolution of 1×1 km2 and a 15 min time step. Performances of the calibrated hydrological model and the parameters robustness are evaluated on two French study areas with 20 catchments in each, in spatio-temporal extrapolation based on cross-validation experiments over a 12-year period. Several spatial regularization strategies are tested to better constrain the high dimensional optimization problem. The model parameters are calibrated based on the Nash-Sutcliffe Efficiency (NSE) computed for multiple calibration basins in the study area. Results are discussed based on the Nash-Sutcliffe Efficiency and the Kling-Gupta Efficiency criteria obtained on calibration and validation catchments for two subperiods of 6 years. Further work aims to improve the global search of prior parameter sets and to better balance the adjoint sensitivity with respect to the spatial constraints resolution and catchment characteristics. This will ensure a better consistency of simulated fluxes variabilities and enhance the applicability of the regionalization method at higher spatial scales and over larger domains.