信息矩阵方法在面雨量预报集成中的应用

Application of information matrix method to consensus forecast of areal rainfall

对多种数值天气预报降水产品进行集成应用,是克服单一预报模式缺点、提高预报准确率的有效途径。任何预报集成方法其本质是建立输入-输出关系,在数学上可以用函数关系表达。给出了用信息矩阵识别这种函数关系的方法,并以面雨量预报为例建立了集成方案。利用欧洲中期天气预报中心(ECMWF)、日本气象厅(JMA)两种数值模式降水预报产品分别计算输出了淮河流域面雨量预报产品。利用地面观测雨量资料,采用泰森多边形法计算流域面雨量实况。将2011年6-8月ECMWF、JMA面雨量预报产品和面雨量实况资料作为学习样本,构建了模糊关系矩阵。利用2012年6-8月ECMWF、JMA两种数值模式面雨量预报产品进行了集成预报试验。采用降水分级检验评分(TS)方法,对比分析了ECMWF、JMA以及集成预报(CFCST)的预报性能。结果表明:基于信息矩阵方法的面雨量集成预报性能总体上高于单个预报成员。

Consensus forecast of rainfall forecasting of NPWs is an effective way to overcome the shortcoming of single forecast mode and improve the accuracy of rainfall forecast. The essence of any forecast consensus method is to establish relationship between inputs and outputs,and the relationship can be mathematically represented by functions. In this paper,a novel approach,called information matrix,was introduced to recognize the function relationship. To display the advantage of the model,a consensus forecast method of areal rainfall was proposed based on the information matrix method. First,the numerical areal rainfall forecast results of Huaihe River Basin were calculated and exported from the quantitative precipitation forecast of ECMWF and JMA,respectively. Then,based on the rainfall data of conventional weather station,the actual situation of areal rainfall was also calculated by Thiessen polygon method. Next,fuzzy relation matrix was constructed by learning from the samples of all the attempt of data and two results of areal rainfall,and was exported from ECMWF and JMA for the period from June to August 2011. Finally,an attempt of consensus forecast was conducted by utilizing the numerical precipitation forecast of ECMWF and JMA for the period from June to August 2012. The performances of the methods ofECMWF,JMA and consensus forecast( CFCST) were analyzed and compared each other by rainfall gradation test score( TS). The results indicate that,the performance of areal rainfall consensus forecast based on information matrix method is overall better than that of individual method in general.