Application of statistical techniques to the analysis and prediction of ENSO: Bayesian oscillation patterns as a prediction scheme

Here we study the low-frequency variability of the tropical Indian and Pacific basins with a new statistical technique, Bayesian oscillation patterns (BOP). To describe the climatic system in this region, zonal wind and sea surface temperature (SST) are the selected variables. Their variability can be explained in terms of a reduced number of frequencies and spatial patterns. These are identified for each field by a statistical procedure. With the help of the patterns and the frequencies a predictive scheme is devised and applied in two forecast experiments. In the first, zonal wind anomalies are predicted using patterns and frequencies identified in the wind field. A more sophisticated scheme, a linear model which includes non-harmonic oscillations and interactions between patterns, is used when forecasting SST seasonal anomalies in the Niño3 region. In this case, the predictors include the values of the frequencies identified in the BOP analysis of both wind and SST fields, and thecorresponding patterns.