Improvement of physics-based and data-driven model simulations based on multi-source soil moisture datasets

The upper Jiao River, located in eastern China, is a representative mountainous watershed that experiences frequent flooding caused by saturation-excess runoff. Accurate streamflow simulation is essential for effective water resource management in watersheds, necessitating the use of reliable models and precise data for the simulation and prediction of streamflow. This study integrates multi-source soil moisture datasets into both a process-based model and a data-driven model to explore the potential benefits of calibration and streamflow simulation within a mesoscale watershed in eastern China. The physics-based Distributed-Hydrological-Soil-Vegetable Model (DHSVM), coupled with the multi-objective genetic algorithm ε-NSGA-II, and data-driven Informer model, are chosen and applied to the study area. Six parameter calibration and streamflow simulation schemes are developed, including one traditional simulation scheme and five multi-hydrological-element schemes based on multi-source soil moisture datasets, for both the DHSVM and Informer models, with the purpose of comparing the impacts of different soil moisture datasets on model calibration, streamflow simulation and hydrological signatures. The results show that the incorporation of soil moisture significantly improves the ability of both DHSVM and Informer models to simulate extreme flows, compared to traditional methods relying solely on outlet streamflow and rainfall. Among the datasets, ERA5 and the merged soil moisture datasets provide superior performance in both calibration and simulation. Simulations using the merged datasets help balance the errors of individual datasets, resulting in optimal outcomes for watersheds without observed soil moisture data. These findings offer valuable insights for improving simulations and forecasts of extreme hydrological events. ● Novel integration of multi-source soil moisture in streamflow simulation models. ● Improves extreme streamflow simulation with DHSVM and Informer models. ● ERA5 and merged datasets are identified as superior for model calibration. ● Enhanced simulation accuracy in data-scarce watersheds.