Author(s): Kumudu Madhawa Kurugama; So Kazama; Yusuke Hiraga

Linked Author(s):

Keywords: Data assimilation; Hydrological modelling; ENKF; Uncertain observation

Abstract: Accurate flood predictions are essential to reduce socioeconomic losses, but hydrological models often face limitations due to sparse measurements, model imperfections, and uncertainties in input data. Recent studies suggest that incorporating discharge observations can help reduce these uncertainties. This study explores the optimal configuration for discharge assimilation in a spatially distributed hydrological model using the Ensemble Kalman Filter (EnKF) to update state variables in the spatially distributed LISFLOOD model. A process-based flow routing model is used to more accurately represent time delays and flow attenuation. The study involves two synthetic twin experiments in the Fraser River Basin, Canada, and the Po River Basin, Italy. It assesses the effects of different configurations of spatially distributed discharge gauges and varying filtering frequencies on simulated discharge. Results show that higher assimilation frequencies significantly improve model accuracy, particularly in capturing peak discharges during flood events. The 6-hour assimilation interval provided the best results, while longer intervals led to greater forecast errors. Additionally, assimilating spatially distributed observations outperformed single-point assimilation, demonstrating the importance of multi-location updates for a comprehensive watershed representation. Future research should focus on incorporating multiple hydrological state variables, model parameters and optimizing assimilation strategies to sustain forecast improvements over longer lead times. These findings emphasize the need for integrated DA frameworks to enhance real-time flood prediction and support effective flood management strategies.

DOI: https://doi.org/10.64697/978-90-835589-7-4_41WC-P2085-cd

Year: 2025