Performance Improvement of Hydrological Models Using Unscented Kalman Filter-type Data Assimilation and Data Fusion

Improving the performance of hydrological models and reducing predictive uncertainty remain major challenges in water resources management and planning. Although numerous modeling approaches and enhancement techniques have been proposed, data fusion methods such as simple averaging possess valuable yet often underexplored potential. In addition, various filtering approaches have been employed to address predictive uncertainty. The Unscented Kalman Filter (UKF), a nonlinear Kalman type data assimilation method, has been applied in a limited number of hydrological studies; however, its application in arid and semi-arid regions has not been adequately investigated. This study introduces and evaluates novel data assimilation strategies for enhancing the predictive capability of hydrological models in semi-arid environments. Three models with distinct spatial structures, namely a lumped model (HBV), a semi distributed model (SWAT), and a fully distributed model (WetSpa), were employed to improve runoff simulations using both UKF based assimilation and data fusion techniques. Results demonstrated that UKF application significantly enhanced model performance, yielding up to a 29.8% increase in the Nash–Sutcliffe efficiency (NSE) for HBV during the validation phase. The greatest relative improvement was observed in the lumped model and the smallest in the distributed model. Both UKF and data fusion methods improved runoff predictions, with UKF consistently outperforming data fusion across calibration and validation stages. These findings highlight the potential of UKF based data assimilation as a robust tool for hydrological modeling in data scarce arid and semi-arid regions.