主分量逐步回归在气温预测中的应用  

Application of Stepwise Regression Analysis of Principal Components in Temperature Prediction

作　　者：曾琮[1] 陈创买[2] 

机构地区：[1]广州中心气象台,广东广州510080 [2]中山大学大气科学系,广东广州510275

出　　处：《中山大学学报（自然科学版）》2006年第4期107-110,共4页Acta Scientiarum Naturalium Universitatis Sunyatseni

基　　金：广东自然科学基金资助项目(5001121)

摘　　要：以广东省冬季气温场为预报场,前一年的北太平洋海温场和北半球500 hPa高度场为因子场,分别对它们作标准化处理,然后进行主分量分析,得到主分量矩阵。通过相关分析和逐步回归,求得预报场的主分量与因子场的主分量之间的关系,对预报场的标准化主分量进行反算,得到原始场的拟合和预报。结果显示,广东省冬季气温场前4个主分量(对总方差的累积贡献达到97.5%)的预报方程都通过显著性检验,其方程复相关系数基本在0.9以上。对广东省48个代表站2001-2003年冬季气温进行预测检验,大部分预测结果的残余标准差比同期的样本标准差低,同时回归方程的预报误差略低于实际业务预报误差,因而回归方程对实际天气预报业务工作有一定的参考意义。Using the average temperature data in winter in perature field in the North Pacific and 500hPa geopotential Guangdong as the predictand field, the sea surface temheights in the Northern Hemisphere of the previous year as predictors, which are normalized, a principal component matrix is obtained by principal component analysis. The correlation between the principal components of predictand fields and the predictors is acquired by correlation and stepwise regression analyses. The fitting and forecast of primitive fields are produced by the retrieval of fields of normalized principal components. The results show that the prediction equations consisting of the first 4 principal components （97.5% of accumulated contribution to the total variance） of winter temperatures field in Guangdong pass the significance test with the multiple correlation coefficients over 0. 9. Based on the forecast verification of the winter average temperature data at 48 stations in Guangdong from 2001 to 2003, it shows that most of the residual variances of the forecasting results are lower than that of the sample, and the forecast error by regression equations is lower than that of operational weather forecasting. Therefore, the results from the study have practical implication in operational weather forecasting.

关 键 词：主分量 逐步回归 温度 预测 

分 类 号：P458.1[天文地球—大气科学及气象学]