由于全球气象模型产生的原始预报通常包含复杂的误差，为评估新一代的GFDL-SPEAR模型对我国各二级水资源区的适用性，构建伯努利-伽马-高斯模型开展统计订正的对比实验，从逐月与累计降水2个方面评估预报的相关性、系统偏差、可靠性以及预报精度，从而辨析原始预报的误差并分析预报订正的作用。结果表明：GFDL-SPEAR原始预报与观测呈现良好的相关关系，但包含?20%到50%的系统偏差，导致预报可靠性与预报精度较低；伯努利-伽马-高斯模型能够有效订正系统偏差，生成可靠的预报时间序列，使逐月与累计降水的预报精度分别提高约25%和45%；相比总量订正，逐月订正能够进一步提高预报精度。整体上，订正后的GFDL-SPEAR降水预报可为流域水资源调控与防洪抗旱提供6个月乃至1年预见期的重要信息。

Global climate models, formulating complex processes in the atmosphere, oceans, land, and sea ice, serve as effective tools to provide meteorological forecasts at the global scale for water resources management. Precipitation is one of the most important variables in hydrological processes. The predictive performances of streamflow, soil moisture, and other hydrological variables can be improved by utilizing precipitation forecasts from global climate models to drive hydrological models. In recent years, operational climate centers have developed different global climate models and produced global precipitation forecasts. Since 2020, the Geophysical Fluid Dynamics Laboratory (GFDL) has been operating the Seamless System for Prediction and EArth System Research (SPEAR) to generate monthly climate forecasts with lead times of up to twelve months. Monthly precipitation forecasts can be used to facilitate hydrological forecasting for wet season (May to October) and dry season (November to April of the subsequent year) in China. Three steps were taken to evaluate the predictive performances of GFDL-SPEAR monthly precipitation forecasts in China. Firstly, the time-series forecasts are extracted from global datasets for second-level water resources regionalization in China. The start times are set to be the beginning of May and November from 1991 to 2020 so that the obtained forecasts with twelve lead times can span the subsequent wet and dry seasons. Secondly, the Bernoulli-Gamma-Gaussian model is formulated to calibrate raw forecasts. Comparative experiments are devised to investigate the effectiveness of calibration on monthly amounts and accumulated totals. Thirdly, the Pearson correlation coefficient (EPCC), relative bias (ERB), alpha index (Eα) and continuous ranked probability skill score (ECRPSS) are computed to verify the association, bias, reliability, and skill of raw and calibrated forecasts in terms of monthly and accumulated precipitation. The results show that the predictive performance of the GFDL-SPEAR model tends to vary by season and region. Raw forecasts can be reasonably correlated with observations at a short lead time while the correlation tends to decrease with lead time. In particular, a reasonable correlation can be observed in the Yellow River, Yangtze River, and Southeast Rivers regions even at the one-year lead time. Raw forecasts usually suffer from substantial biases and random errors. Specifically, the values of ERB range from –20% to 50% and the values of Eα fall between 0.4 to 0.8, primarily owing to positive biases and too-narrow ensemble spreads in raw forecasts. In the meantime, forecast errors can accumulate with lead time and then deteriorate predictive performances of accumulated precipitation. The Bernoulli-Gamma-Gaussian model is effective in removing systematic biases and generating reliable time-series forecasts. For calibrated forecasts, it is shown that the values of ERB tend to be nearly zero and that the values of Eα are mostly larger than 0.8 across all lead times. While raw forecasts can be negatively skillful due to impacts of systematic and random errors, the values of ECRPSS for calibrated forecasts are overall improved by 25% and 45% for monthly and accumulated precipitation, respectively. Compared with the calibration on accumulated totals, the calibration on monthly amounts can further improve forecast skill, leading to higher ECRPSS for accumulated precipitation.The applicability of the GFDL-SPEAR monthly precipitation forecasts over second-level water resources regionalization in China was evaluated. The GFDL-SPEAR provides informative forecasts but suffers from systematic and random errors. The Bernoulli-Gamma-Gaussian model can effectively correct bias and generate reliable calibrated forecasts. The calibration on monthly amounts tends to outperform the calibration on accumulated totals by effectively leveraging forecasts across different lead times. Overall, the GFDL-SPEAR monthly precipitation forecasts calibrated by the Bernoulli-Gamma-Gaussian model can be used to inform hydrological forecasting and water resource management at a one-year lead time in China.