Improving real-time operational streamflow simulations using discharge data to update state variables of a distributed hydrological model

Reducing errors in streamflow simulations is one of the main issues for a reliable forecast system aimed to manage floods and water resources. Data assimilation is a powerful tool to reduce model errors. Unfortunately, its use in operational chains with distributed and physically based models is a challenging issue since many methodologies require computational times that are hardly compatible with operational needs. The implemented methodology corrects modelled water level in channels and root-zone soil moisture using real-time water level gauge stations. Model's variables are corrected locally, then the updates are propagated upstream with a simple approach that accounts for sub-basins' contributions. The overfitting issue, which arises when updating a spatially distributed model with sparse streamflow data, is hence here addressed in the context of a large-scale operational implementation working in real time thanks to the simplicity of the strategy. To test the method, a hindcast of daily simulations covering 18 months was performed on the Italian Tevere basin, and the modelling results with and without assimilation were compared. The setup was that currently in place in the operational framework in both cases. The analysis evidences a clear overall benefit of applying the proposed method even out of the assimilation time window.