基于ECMWF细网格模式的短时强降水客观概率预报方法研究

根据陕南643个经质量控制的自动气象站2010—2014年逐小时观测降水，采用百分位法确定陕南短时强降水标准。基于2010—2014年11 824站次短时强降水个例和欧洲中心（ECMWF）间隔6 h的0.25 °×0.25 °再分析格点资料，以空间最近、时间最近前1时次原则:计算并确定陕南汛期5—9月各月短时强降水36种对流参数历史概率分布特征值；考虑对流参数的显著性和适度性指标构建评价方案，利用相对模糊偏差矩阵、标准差系数方法，优选出陕南5—9月各月的15种对流参数及其权重。业务运行以ECMWF细网格模式的基本预报产品，计算优选的对流参数值，结合参数历史概率分布值及其权重，建立陕南分月短时强降水客观概率预报模型。将模型概率预报结果升序排列后80%处对应的数值，且大于0.2作为短时强降水的临界概率，对2015年汛期降水过程进行检验，TS评分为0.59，漏报率为0.18，空报率为0.31。 

STUDY OF THE OBJECTIVE PROBABILITY FORECAST METHOD FOR SHORT-TERM HEAVY RAIN BASED ON ECMWF FINE-MESH MODEL

Using percentile method, the standards of the short-term heavy rain are determined in the south part of Shaanxi province on the base of precipitation data of 643 automatic meteorological station after data quality control from 2010 to 2014. Based on 11 824 cases of short-term heavy rain and 0.25 °×0.25 ° reanalysis data at 6 h intervals of the European Centre for Medium-Range Weather Forecasts(ECMWF) during 2010 to 2014, according to the rule of being within nearby space and near recent time: the characteristic values of 36 kinds of convective parameters probability distribution are obtained in the south part of Shaanxi province from May to Sept during flood season. The evaluate scheme is formulated with the significance and appropriation indexes of 36 kinds of convective parameters, by the methods of the relative deviation fuzzy matrix and standard deviation coefficient, the 15 kinds of convective parameters from May to Sept are selected out, which can well indicate the environmental field for short-term heavy rain, and their weights are given. Based on high-resolution basic forecast products of the ECMWF fine-mesh model, the objective probability forecasting method for short-term heavy rain is established through comprehensive consideration of probability distribution and weights of the selected convective parameters. The value of more than 0.2 corresponding to 80% of the ascending order of probability forecast is taken as critical probability, the TS is 0.59, the rate missing prediction 0.18, and the rate of false predication 0.31 during 2015 flood season.