摘要
对广西88个站冬季(12月、1月和2月)各月平均气温距平场作自然正交展开(EOF分解),选取累积方差贡献超过90%的前3个主成分作为预报量.从前期平均大气环流场和海温场中查找预报因子,对这些初选因子用偏最小二乘回归方法(PLS)进行信息筛选和成分提取,用提取的新综合变量(又称成分)作预报因子,分别建立各月平均气温前3个主成分的回归预报方程.经独立样本预报试验证明,偏最小二乘回归方法具备良好的因子信息提取能力,其预报建模方法对冬季月平均气温预报具有较好的预测效果.
This paper makes the natural orthogonal decomposition (the EOF decomposition) for the various monthly means temperature departure field of Guangxi 88 stations in winter (December, January and February), and selects the first 3 principal components of the accumulation variance contribution surpassing 90% as the predictands, meanwhile Searches the predictors from the atmospheric circulation field and the sea temperature field, and conducts the information screening and the component withdrawing with the Partial Least-Squares Regression (PLS) for these primary selecting factors, and takes the new synthesis variable as the predictor (called components) and establishes the regression forecast equation of the first 3 principal components of various monthly means temperature respectly. The independent sample forecast experiment proves that the Partial Least-Squares Regression have the good ability for factor information extraction and its forecast modelling method has the good forecast effect to the monthly mean temperature forecast in winter.
出处
《数学的实践与认识》
CSCD
北大核心
2006年第8期266-273,共8页
Mathematics in Practice and Theory
基金
国家科技部社会公益性研究专项(2004DIB3J122)
广西科学研究与技术开发项目(桂攻关0592005-2A)
关键词
偏最小二乘回归
EOF
月平均气温
短期气候预测
Partial Least-Squares Regression
EOF
monthly mean temperature
short-term climate forecast