支持向量机(SVM)方法在降水分类预测中的应用

支持向量学习机(SVM)是基于统计学习理论的模式分类器,将SVM方法应用于降水异常的分类预测中尚属首次。主要利用1958—2003年逐月的74个环流特征量、NINO 3,NINO 4海温指数、相关区域海平面气压、500 HPA、100HPA有关指数资料等,分别建立了四川盆地5片区降水距平百分率大于50%(特多)和小于-50%(特少)的2个SVM推理模型,并进行了降水分类预测试验和2005年1-3月实际预测,结果显示出所建SVM推理模型的Ts评分较高,具有一定的预测能力,展示了SVM的优越性和推广前景,可在短期气候预测业务中参考应用。

APPLICATION OF SUPPORT VECTOR MACHINE (SVM) IN RAINFALL CATEGORICAL FORECAST

Two support vector machine(SVM) reasoning models were established for five areas with abnormal rainfall forecast of 50% excess or less than-50% in Sichuan Basin,using the monthly data of 74 circumfluent eigen values,sea surface temperature index of Nino 3 and Nino 4,index of sea level pressure of 500 HPA and 100 HPA height from 1958 to 2003.Rainfall categorical forecast was performed with these two models from January to March in 2005.The results indicated that both models performed satisfactorily,with relatively high threat scores(Ts).Empirical results demonstrated that their performance was better than that given by standard statistic analysis and forecast methods.