史久恩,瞿栋根,孙山泽,崔玉玺. 1964. 统计学长期天气预报方法的若干研究(一) 逐步回归技术的应用[J]. 气象学报, 34(4):507-518, doi:10.11676/qxxb1964.049
统计学长期天气预报方法的若干研究(一) 逐步回归技术的应用
STUDIES ON LONG-RANGE FORECASTING BY STATISTICAL METHODS:PART I——APPLICATION OF STEPWISE MULTIPLE REGRESSION TECHNIQUE
投稿时间:1964-03-10  
DOI:10.11676/qxxb1964.049
中文关键词:  
英文关键词:
基金项目:
作者单位
史久恩 中央气象局气象科学研究所 
瞿栋根 中央气象局气象科学研究所 
孙山泽 北京大学数学力学系 
崔玉玺 中国科学院计算技术研究所 
摘要点击次数: 1913
全文下载次数: 2114
中文摘要:
      本文采用了逐步迴归技术,可以在一大堆预报因子中选择出迴归方程中的主要变量,组成一个最后的迴归方程--预报方程。为了具体说明逐步迴归方法应用于长期预报的计算情况,将1932-1962年华北五站(北京、天津、保定、石家庄、营口)7,8月份平均降水总量的资料作为因变量y,以表征太阳辐射影响的因子、前期环流影响的因子和其它若干气象要素等作为自变量进行分析研究。对我国若干重点地区的月、季降水的迴归分析指出,各地区与降水有关的预报因子是不完全相同的。就1963年夏季和6-10月份月的降水长期预报进行检查。各重点地区42次预报的结果说明,用这个方法所作的降水趋势预报尚好,比偶然性预报约高10%。最后,作者就迴归分析的改进作了简略的讨论,并提出今后除了用于长期预报的研究和业务工作外,对中、短期和专业天气预报以及气候学方面等的研究工作也将会是一个有用工具。
英文摘要:
      In this paper,a stepwise multiple regression procedure is used to screen from a large number of predictors that are most significantly related to a particular pcedictand.If or more predictors are not statistically significant,they may be eliminated from equation.The stepwise technique is here used to solve the problem of monthly and seasonal forecasting of precipitation over China.As an example,from the mid-summer data of 31 years (1932-1962),the summary of the calculation procedure of monthly ptecipitation over North China is illustrated.The analyses indicates that the monthly precipitation in each tract is related to different predictors,such as solar radiation indices,parameters of general circulation,and other factors for the periods prior to rainfall.The linear regression equations (for predicting the monthly precipitation over several regions of China) based on these correlations are tested on an independent sample including 42 cases.The results are compared with those generated by chance.It is found that the prediction scores are higher than above verification standard.Finally,a brief discussion is given concerning the improvement of regression analysis and suggestions are made for further statistical work on that problem.
HTML   查看全文   查看/发表评论  下载PDF阅读器
分享按钮