Enhancing hydrological modelling with reanalysis soil moisture: A data-driven approach for optimizing initial conditions through reanalysis integration

Abstract

Hydrological models are fundamental tools for water resource management, flood mitigation, and ecological protection. Soil moisture (SM) critically affects the accuracy and reliability of these models by influencing rainfall infiltration and runoff generation. While previous studies have demonstrated the benefits of incorporating SM observations or products into hydrological simulations, there is still ample room to fully exploit their potential in developing the initial SM conditions and reducing the warm-up process before model calibration. In this study, we develop a novel strategy to enhance the utility of the European Centre for Medium-Range Weather Forecasts Reanalysis v5-Land (ERA5-Land) SM dataset in replacing the default initial SM of the Block-wise use of the TOPMODEL (BTOP) model without warm-up adjustments. This strategy involves establishing robust relationships between BTOP and ERA5-Land SM variables, grounded in their physical definitions, through various curve-fitting functions and Long Short-Term Memory (LSTM) model. The improved ERA5-Land SM series are then applied for the calibration of the BTOP model to assess their effectiveness in substituting initial SM conditions across Japan's Fuji and Shinano River Basins. The results show that the LSTM model outperforms traditional curve fitting in establishing relationships of various SM variable combinations, and the basin-scale LSTM provides a practical advantage for large basins with high computational costs, while still maintaining the reliability of relationship constructed. Furthermore, the proposed strategy for initial SM acquisition exhibits commendable performance in replacing the default initial conditions of the BTOP model, resulting in substantial improvements in hydrological simulations. During the calibration period, the metrics (NSE and KGE’) showed enhancements of up to 30.63 % and 15.03 %, respectively, while in the validation period, these metrics improved by 6.49 % and 25.11 %, further highlighting the effectiveness of the strategy. This satisfactory strategy helps preserve more data for the calibration and validation of hydrological models, particularly in data-scarce basins.