考察贝叶斯模式平均（BMA）对第二代气候预报系统（CFSv2）在东江流域汛期月降雨量预报的订正效果，同时引入标准化异常度（SA）指标识别异常的降雨值，分别进行SA的确定性预报以及集合预报，通过建立SA和BMA结果之间的联系构建一个较为完整且精度较好的降雨预报模型，提高东江流域中长期降雨预报的精度。主要结论如下：①BMA50％以下的分位数不具有预报作用，75％分位数具有最优的预报效果。但BMA还存在不足之处，常表现为对极端降雨的低估；②利用CFSv2集合平均值进行SA计算时，SA严重偏小，可能说明CFSv2存在系统性误差。对CFSv2原始预报分别进行伽玛函数订正以及多项式订正后，降水预报成功指数（Ts）和异常值报对的次数有明显地提高，但预报偏差（Bs）也相应地增大；③ SA与BMA之间大致可建立如下的关系，即当SA判断会出现异常值时可选择95％分位数的预报值，相反则选择75％分位数预报值。%Bayesian Model Averaging (BMA)is applied to monthly precipitation forecasting in the flood season over the Dongjiang basin to correct the bias of Climate Forecast System version2 (CFSv2).In the meantime,Standardized Anomaly (SA)is used to quantify the precipitation abnormality and incorporated into the deterministic and ensemble forecasting.A better precipitation forecasting model is then estab-lished by the combination of BMA and SA to improve accuracy of long-term precipitation forecasting in the Dongjiang basin.Conclusions are drawn as follows:① The 50th percentile and below of ensemble fore-casting have poor skill ,whereas the 75 th percentile is usually in agreement with observations.However, BMA has disadvantage in that it underestimates precipitation amount when extreme events occur.②The value of SA based on the ensemble average of CFSv2 is too small,indicating a systematic bias of CFSv2. When the CFSv2 raw forecasting is corrected by gamma function and multinomial,both Threat Score and the number of greatly increases but Bias Score increases in the meanwhile;③ The relationship between SA and BMA can be expressed as follows:the 95 th percentile of ensemble forecasting is used when SA indicates an abnormal precipitation,otherwise the 75 th percentile is used.
展开▼
