Study on Flood Forecasting Model for Small Pumped Storage Watersheds Based on the LiuxiheModel: A Case Study of the Baiyunxi Watershed

Flood events in small watersheds of pumped storage power stations are characterized by strong suddenness and short duration, and are significantly influenced by human activities such as inter-reservoir water transfer. Accurate forecasting of such floods is crucial for flood control and disaster mitigation, ensuring the safety of dam structures, and supporting operational decision-making for pumped storage power stations. This study focuses on the Baiyunxi River Basin, where the Fujian Yongtai Pumped Storage Power Station is located, and adopts the distributed hydrological model—Liuxihe Model—as a prototype to develop a distributed hydrological model suitable for flood forecasting in small watersheds of pumped storage power stations. Six flood events recorded since the power station’s commissioning were selected for model parameter optimization and validation. The simulation accuracy was evaluated using key indicators including the Nash-Sutcliffe Efficiency (NSE), Kling-Gupta Efficiency (KGE), peak time difference (PT), and absolute peak error (APE). The results demonstrate that the model performs well in the Baiyunxi River Basin. After parameter optimization, significant improvements were observed in both NSE and KGE for the target flood events. During the validation phase, the average NSE reached 0.708, and the average KGE was 0.797. The maximum peak time difference was only 1 hour earlier, and the majority of absolute peak errors were below 10%. Parameters calibrated for a single flood event were successfully applied to other events, indicating strong adaptability to floods of different magnitudes and hydrograph shapes. To address the issue of discharge curve fluctuations caused by deriving flow from observed water levels in the basin, a multi-objective comprehensive function was employed for parameter optimization, effectively enhancing parameter robustness.