新一代全球集合预报（global?ensemble?forecast?system?version?12，GEFSv12）是美国国家环境预报中心 （National?Centers?for?Environmental?Prediction，NCEP）发布的最新的再预报数据集，对该数据集中的降水数据在淮 河流域的适用性进行评估。由于原始预报存在系统偏差，所以应用贝叶斯联合概率模型（Bayesian?joint?probability， BJP），对原始 GEFSv12 再预报数据进行统计后处理。预报验证的评价指标包括均方根误差（ERMS）、Brier 技巧评 分（EBSS）、连续等级概率技巧评分（ECRPSS）、α 指数及可靠性框图，注重从预报偏差、精度以及可靠性 3 方面评价原 始预报及经过后处理的预报。结果表明：GEFSv12 降水再预报数据对淮河流域具有较好的适用性，在预见期 1~7?d 具有预报技能；BJP 统计后处理方法能有效降低预报系统性偏差，增加预报精度以及预报可靠性，经过后处理的降 水集合预报将有利于进一步的水文集合预报应用。

Due to the complexity of atmospheric physical processes, numerical weather prediction often suffers from uncertainty. Ensemble forecasts can be used to represent the uncertainty and provide uncertainty information for risk-based decision-making. In recent years, hydrometeorological ensemble forecasting has become widely applied, which uses the ensemble outputs from the numerical weather prediction models to drive the hydrological models. Bias-free, accurate and reliable ensemble precipitation forecasts are important for producing accurate streamflow forecasts. Forecast verification is the process of assessing the quality of forecasts, which involves the investigation of the properties of the joint distribution of forecasts and observations. Objective evaluations of forecast quality can help monitor operational forecasts, assess the specific strengths and weaknesses of forecast systems and support decision-making. Global Ensemble Forecast System, version 12 (GEFSv12), is a new generation of reforecast dataset generated by the National Centers for Environmental Prediction (NCEP). Reforecast dataset is generated by retrospectively re-forecasting weather for previous years using the current dynamical model. It is an important dataset for improving weather predictions. The performance of GEFSv12 precipitation forecasts has not been discussed in the Huaihe River basin, so we evaluated the performance of GEFSv12 precipitation forecasts in the Huaihe River basin in this research. Due to the systematic bias of the raw forecasts, the Bayesian Joint Probability (BJP) model was used to perform statistical post-processing on the raw GEFSv12 reforecasts. The quality of the raw and post-processed GEFSv12 ensemble reforecasts was evaluated in three aspects, i. e. , bias, accuracy and reliability. Verification metrics included Root Mean Squared Error (ERMS), Brier Skill Score ( EBSS), Continuous Ranked Probability Skill Score (ECRPSS), αindex and Reliability Diagram. The main results are as follows:(1) In terms of forecast bias, according to the map of the spatial variation of precipitation and the results of ERMS, raw GEFSv12 precipitation forecasts suffer from a large bias, and have difficulty in reproducing the spatial variability of precipitation within the Huaihe River basin. Statistical post-processing approach can effectively reduce the systematic bias of the raw forecasts. (2) In terms of forecast accuracy, the accuracy of raw GEFSv12 precipitation forecasts is better than that of climatology at lead times of 1-7 days. The accuracy of the raw and post-processed forecasts decreases with lead times. According to the EBSS,when the lead times are up to 6 days, the accuracy of raw GEFSv12 precipitation forecasts for light rain events at some grid cells in the Huaihe River basin is lower than that of climatology. The statistical post-processing approach can improve the forecast accuracy of raw forecasts for light rain at long lead times(≥6 days). According to the ECRPSS, when the lead times are up to 7 days, the accuracy of raw GEFSv12 precipitation forecasts at some grid cells in the Huaihe River basin is lower than that of climatology. The statistical post-processing approach can improve the forecast accuracy of raw forecasts at long lead times(≥7 days). (3) In terms of forecast reliability, according to the αindex and reliability diagram, the raw GEFSv12 precipitation forecasts suffer from overestimation for light and moderate rain events. Statistical post-processing approach can significantly improve forecast reliability.The results show that the raw GEFSv12 forecasts perform well in the Huaihe River basin at lead times of 1-7 days. BJP post-processing approach can effectively reduce the systematic bias of the raw forecasts and increase forecast accuracy and reliability. Post-processed precipitation ensemble forecasts can be applied in further applications such as hydrological ensemble forecasting.