A technique for extracting potentially predictable patterns from climate data

Abstract

We propose a computational technique which makes it possible to extract long-range potentially predictable patterns of interannual variability of meteorological seasonal mean fields. These patterns arise from slowly varying external forcing, such as sea surface temperatures, and slowly varying internal dynamics. The method provides a means of decomposing the covariance matrix of a seasonal mean field into covariance matrices for the potentially predictable and the chaotic, or weather-noise, components, separately. We illustrate the technique using Australian surface maximum temperatures during December-January-February (DJF) for the period 1958--1991. The dominant patterns, arising from the potentially predictable covariance matrix, are shown to be more closely related to slowly varying external forcing and slowly varying internal dynamics than those from a conventional analysis. The importance of tropical sea surface temperatures in forcing the potentially predictable patterns is also discussed.