The space‐filtering of monthly surface temperature anomaly data in terms of pattern, using empirical orthogonal functions

Abstract In statistical long‐range forecasting the large volume of data makes some form of filtering imperative both to reduce the data‐handling problem and to exclude random elements which do not contribute to understanding or prediction. In this study, which is an exercise in the statistical technique of factor analysis, fields of 30‐day surface temperature anomaly for the years 1881–1960 are filtered in terms of sets of patterns specific to each calendar month. These are derived empirically, and so truly reflect the characteristics of the original fields. In the resulting representation, the volume of data may be reduced to one quarter while retaining 80 per cent of the original variance, and a study of seasonal variation shows that continentality and zonal thermal advection are at all times of year the most important factors in shaping the 30‐day anomaly fields.