Improving Short-Term Runoff Forecasting in a Karst Basin: Integrating ECMWF Precipitation Data with Process-Driven and Machine Learning Models

Reliable short-term runoff forecasting in karst basins remains a major challenge. This study systematically evaluates the performance of three ensemble precipitation forecast products—European Centre for Medium-Range Weather Forecasts (ECMWF), China Meteorological Administration (CMA), and Korea Meteorological Administration (KMA)—combined with two types of models: process-driven models and data-driven models with 24, 48, and 72 h forecast periods for the Chengbi River Basin. The results show that ECMWF consistently outperformed CMA and KMA in precipitation forecasting accuracy. Among hydrological models, the Xin’anjiang model achieved the highest predictive accuracy (R2=0.89). It better captured complex hydrological responses such as groundwater-surface water interactions. In contrast, the Transformer model exhibited larger prediction errors at longer lead times. Overall, this study highlights the critical role of selecting suitable precipitation products and hydrological models for accurate short-term runoff forecasting in karst basins. The findings offer valuable insights for improving flood prediction and water resource management in karst basins.