Hydrological Auditing of LISFLOOD v4.1.1: Impacts of Model Setup on Water Balance Components in the Po River Basin

Abstract. In recent years, large-scale hydrological models have been increasingly used at regional and global scales to support decision making. Their realism in simulating water balance components is crucial for building trust across different use cases. Hydrological models may reproduce streamflow well but misrepresent other fluxes, due to internal fluxes compensations and equifinality. Therefore, alternative setups can benefit specific applications by improving the representation of relevant water balance components. "Hydrological auditing" of models, i.e. a thorough critical review of their realism beyond the calibration targets (usually streamflow), provides useful insights for both practical applications and process understanding. We present one such exercise in a representative European case study using a physically-based hydrological model (LISFLOOD), widely used for flood forecasting and water resources management. We evaluate LISFLOOD v4.1.1's performance in simulating streamflow, evapotranspiration, and overall water balance in the Po River Basin, a complex and highly managed basin in Northern Italy. Six alternative model setups are tested, including different soil layers depths and preferential flow representations. Results show that the model setup currently used in the European Flood Awareness System (EFAS) v.5 performs best in terms of streamflow simulation, particularly at the daily time step, but tends to underestimate evapotranspiration. In turn, this may lead to an overestimation of groundwater recharge and a poor water balance representation. The use of the Budyko framework as a diagnostic tool reveals that model setups without preferential flow better match the expected long-term water balance, but reduce daily streamflow performance. The study highlights the importance of evaluating model performance and auditing alternative parametrizations to ensure accurate simulations of water balance components, crucial for water resources management. We propose criteria to improve the calibration of the LISFLOOD model in a flexible and target-driven way, to better support water resources management in complex river basins.