Augmenting observation network design and assimilation frequency in distributed hydrological models: insights from the LISFLOOD-based hydrological data assimilation framework

This study presents the development and application of a spatially distributed hydrological data assimilation framework (LISFLOOD-HDAF) to investigate the influence of assimilation frequency, gauge density, and spatial distribution on prediction performance. The framework couples the process-based LISFLOOD model with an Ensemble Kalman Filter (EnKF). It was systematically evaluated in the upper Po River Basin, Italy, through synthetic twin experiments covering six representative wet and dry events, three assimilation frequencies (6, 12, and 24 h intervals), and eight contrasting observation network configurations. The results indicate that EnKF assimilation consistently enhanced the simulation performance relative to the open-loop run, with the most pronounced improvements occurring during wet periods and at the downstream outlet. In these cases, the normalized Kling–Gupta efficiency assimilation index (KGEAI) attained values of 0.80–0.90, whereas the root mean square error (RMSE) reductions exceeded 50 %. The impact of assimilation frequency was found to be non-monotonic, with 12 h-interval yielding the most robust improvements at midstream and downstream locations. In contrast, the upstream headwaters benefited from 6 h-interval updates in wet conditions and 24 h-interval updates in dry conditions. Assimilation performance depended strongly on gauge density and placement, with dense or strategically distributed networks performing best, although with diminishing returns. Five gauges with 12 h-interval updates provided near-optimal skill in wet regimes, whereas only three to four gauges sufficed in dry conditions with 24 h-interval updates. Overall, LISFLOOD-HDAF provides a transparent and transferable framework for a cost-effective gauge network design, offering practical guidance for operational flood forecasting.