基于数值预报模式降水产品的集成应用方法试验

Integration Experiment Based on Numerical Prediction Precipitation Products

  • 摘要: 利用2008年5~7月四川省152个站点的逐日降水实况和T213、德国数值预报模式、日本数值预报模式、西南区域本地化的MM5集合降水预报模式和各成员的逐日降水预报产品进行Ts评分,然后基于西南区域集合预报8个成员Ts评分权重系数进行集成预报试验和集成方法对比试验,结果表明:基于Ts评分为权重系数的集成预报效果得到明显改进,且该方法优于逐步回归和支持向量(SVM)回归方法;最后根据上述5~7月T213、日本、德国和集合预报8个成员共11个数值模式降水预报的Ts评分按权重系数集成,预报四川2008年8~9月降水并进行效果评价,结果显示:以多个模式Ts评分为权重系数集成的降水预报效果相对于单一模式有明显提高,有较好的预报指导意义。

     

    Abstract: Based on daily live precipitation from 152 stations in Sichuan Province and daily precipitation forecasts of T213, Germany, Japan、south-west region localization MM5 ensemble forecast model and members from May to July in 2008, TS scores were conducted.Experiments of integration and integration techniques were made based on TS scores weighing coefficient of 8 members of ensemble forecast.The results show that:integration forcast has improved, and integration is superior to successive regression and SVM.At last, integration of 11models were made by using the method mentioned above.Then forecasting effect was carried out from Sichuan rainfall from August to September in 2008, the result show that:the integration rainfall forecasting has improved significantly compared with single model.

     

/

返回文章
返回