Statistical Weather Forecasting

Abstract

This paper illustrates the use of statistical methods such as multiple correlation analysis, discriminant analysis and principal component analysis for weather forecasting. Two examples are presented: the first is a qualitative local prediction of daily precipitation for the next day, using pressure measures of the present day and of past days, by means of discriminant analysis; the second is an analysis of the 500-mbar geopotential heights over Western Europe by means of principal component analysis, followed by a quantitative synoptic prediction of the evolution of these geopotential heights for the next week to come, by means of multiple correlation analysis. For each of these two prediction problems, good predictors are chosen among a great number of candidate ones by a special stepwise selection procedure.