用非参数估计技术预报风的研究  

Using Nonparametric Estimation Technique to Predict Wind

作　　者：邵明轩 刘还珠 窦以文[2] 

机构地区：[1]国家气象中心,北京100081 [2]北京市气象局,北京100089

出　　处：《应用气象学报》2006年第B08期125-129,共5页Journal of Applied Meteorological Science

基　　金：中国气象局新技术2005年推广项目(CMATG2005 M07)资助。

摘　　要：该文介绍了使用非参数估计技术制作风的预报试验。为了有一个较完整的历史样本集,试验中使用了1981—2003年共23年的逐日资料。在采用的K近邻非参数估计技术中,由于风的特殊性,其不是标量而是矢量,试验中根据天气学原理,设计和试用了一种根据过程相似性从历史样本集中搜索出近邻子集和新的从近邻子集中挑选最佳样本的方法。试验结果显示,所试用的预报方法不仅使得风的预报Ts评分有一定的提高,同时使得3 d降水量的预报Ts评分也有一定的提高,这表明该方法具有一定的实际参考使用价值。Nowadays most interpretation and application of Numerical Weather Prediction （NWP） products are processed by means of parametric methods. Through more than thirty year＇s development and use, the parametric method is proved to be useful for forecasters. But the method also shows its own weakness, which is that it needs some assumptions. If the assumption does not match with the real situation, the prediction results will not be stable. In order to overcome the problem, in the experimentation wind is predicted by kernel neighbor nonparametric estimation technique. The kernel neighbor nonparametric estimation technique is a kind of similar example reasoning methods. In the test, the historic daily data from 1981 to 2003 are used, which includes American National Center for Environmental Prediction （NCEP） reanalysis daily data and every six hours observational data of 693 stations countrywide. National Meteorological Center （NMC） T213 data are used as forecast field data. Using kernel neighbor nonparametric estimation technique, how to search for the kernel neighbor sample subclass and how to make a ＂best＂ prediction from kernel neighbor sample subclass are key techniques. Because of the particularity of wind which is a vector rather than a scalar, to forecast wind element, there is not a best way to make a ＂best＂ prediction from kernel neighbor sample subclass up to the present. In the experimentation, based on the weather process similarity and principle of synoptic meteorology, a new method, searching for kernel neighbor samples and finding out a ＂best＂ similar sample from kernel neighbor samples, is devised and tested. In the prediction test the whole forecast field is divided into five districts： northeast area, northwest area, north China area, southwest area and south China area. Experimentation results indicate that in most areas kernel neighbor nonparametric estimation prediction using new method is better than the most similar prediction not only for wind forecast but for rainfal

关 键 词：非参数估计技术 K近邻预报 样本一致性 风预报 

分 类 号：P457.5[天文地球—大气科学及气象学]