We evaluated 24-h control forecast products from The International Grand Global Ensemble center over the 10 first-class water resource regions of Mainland China in 2013-18 from the perspective of precipitation processes (continuous) and precipitation events (discrete). We evaluated the forecasts from the China Meteorological Administration (CMA), the Centro de Previsao de Tempo e Estudos Climaticos (CPTEC), the Canadian Meteorological Centre (CMC), the European Centre for Medium-Range Weather Forecasts (ECMWF), the Japan Meteorological Agency (JMA), the Korea Meteorological Administration (KMA), the United Kingdom Met Office (UKMO), and the National Centers for Environmental Prediction (NCEP). We analyzed the differences among the numerical weather prediction (NWP) models in predicting various types of precipitation events and showed the spatial variations in the quantitative precipitation forecast efficiency of the NWP models over Mainland China. Meanwhile, we also combined four hydrological models to conduct meteo-hydrological runoff forecasting in three typical basins and used the Bayesian model averaging (BMA) method to perform the ensemble forecast of different scenarios. Our results showed that the models generally underestimate and overestimate precipitation in northwestern China and southwestern China, respectively. This tendency became increasingly clear as the lead time rose. Each model has a high reliability for the forecast of no-rain and light rain in the next 10 days, whereas the NWP model only has high reliability on the next day for moderate and heavy rain events. In general, each model showed different capabilities of capturing various precipitation events. For example, the CMA and CMC forecasts had a better prediction performance for heavy rain but greater errors for other events. The CPTEC forecast performed well for long lead times for no-rain and light rain but had poor predictability for moderate and heavy rains. The KMA, UKMO, and NCEP forecasts performed better for no-rain and light rain. However, their forecasting ability was average for moderate and heavy rain. Although the JMA model performed better in terms of errors and accuracy, it seriously underestimated heavy rain events. The extreme rainstorm and flood forecast results of the coupled JMA model should be treated with caution. Overall, the ECMWF had the most robust performance. Discrepancies in the forecasting effects of various models on different precipitation events vary with the lead time and region. When coupled with hydrological models, NWP models not only control the accuracy of runoff prediction directly but also increase the difference among the prediction results of different hydrological models with the increase in NWP error significantly. Among all the single models, ECMWF, JMA, and NCEP have better effects than the other models. Moreover, the ensemble forecast based on BMA is more robust than the single model, which can improve the quality of runoff prediction in terms of accuracy and reliability.
