Evaluating the Spatial Generalizability Across Sub-Basins and Temporal Stability of a Reanalysis-Forced SWAT Model: A Case Study of the Poyang Lake Basin

Hydrological simulation in uncalibrated sub-basins faces common challenges including limited forcing data quality, low reliability of parameter spatial transfer, and poor long-term simulation stability. Taking the Poyang Lake Basin (approximately 162,200 km2) as the study area, this study aims to systematically evaluate the spatiotemporal generalization performance of a coupled hydrological model. A SWAT distributed hydrological model was constructed driven by the high-resolution ERA5-Land meteorological dataset. Model calibration was conducted at the Waizhou station, and validation was performed across three sub-basin outlets not used in calibration for spatial transfer testing, as well as over a 30-year period for temporal stability assessment. The results indicate that the model achieved high accuracy at the calibrated plain station, with NSE and R2 consistently above 0.86. Spatial transfer to the three uncalibrated sub-basins yielded NSE and R2 both above 0.7, demonstrating good spatial generalization capability. The 30-year long-term simulation maintained NSE and R2 generally above 0.8, confirming reliable temporal stability. However, systematic runoff underestimation was observed in long-term runs, with a maximum negative PBIAS of −20.68% during 2005–2014, mainly attributable to rainfall resolution limits, static land use assumption, and single-objective calibration trade-offs. Overall, the ERA5-Land–SWAT model shows dependable spatiotemporal extrapolation capability in the Poyang Lake Basin, and can be further improved with higher-resolution precipitation and dynamic land use data for water resources management applications.