Multi-source meteorological data assessment on daily runoff simulation in the upper reaches of the Hei River, Northwest China

Abstract

Study region

The upper reaches of the Hei River Basin, northwest China

Study focus

To improve the accuracy and physical consistency of runoff simulations, as well as to compare the applicability of meteorological data obtained from multiple sources, this study integrates physical mechanisms with deep learning methods to construct a coupled model, HIMS-LSTM. Considering the impact of meteorological data on runoff simulation and prediction, meteorological station observation data, ERA5 data and CFSv2 data were obtained for runoff simulation and prediction. This approach enables the assessment of the applicability of meteorological data obtained from three different sources.

New hydrological insights for the region

The HIMS-LSTM model leverages physical mechanisms to compensate for the lack of physical knowledge in data-driven models. Consequently, the accuracy and physical consistency of runoff simulation results are significantly improved compared to the single models HIMS and LSTM. Furthermore, a comparative assessment of simulation results based on multi-source meteorological data demonstrates that daily runoff simulations using meteorological station observation data yield the best results, indicating the highest applicability of this data source. The constructed coupled HIMS-LSTM model provides some insight into the simulation and prediction of daily runoff. Furthermore, this study provides a valuable reference for selecting suitable meteorological data sources for runoff simulation and prediction.