Relationships between Monthly Precipitation and SST Variations in the Tropical Pacific Region

Abstract

The relationships between the patterns of monthly sea surface temperature (SST) variations and those of precipitation in the tropical Pacific Ocean region are examined. The rainfall data utilized are derived from satellite observations of outgoing longwave radiation. A composite empirical orthogonal function (EOF) analysis of the SST and precipitation indicates a dominant mode of variation linking SST variations in the eastern equatorial region with those of MNWI about 30° westward. One-point correlation analyses show that this general relationship is present for all points in the eastern and central ocean, but that no significant SST-rainfall correlations are evident for SST points wen of about the dateline. The one-point correlation analyses also suggest that during time periods outside of El Niñ's, the rainfall response to SST changes is largely confined to the areas of climatological precipitation maxima, suggesting only a minor alteration in the large-scale circulation. On the other hand, during El Niñ there is the strong suggestion that circulation changes give rise to complete shifts in rainfall zones. The possible influence of the strong spatial autocorrelations of SST on these results are also explored using the one-point correlations. It is concluded that the observed SST-rainfall teleconnections cannot be wholly explained by the large-scale nature of the SST variations. The inferences derived from these analyses are utilized in formulating linear regression (LR) models to specify tropical precipitation anomalies based upon a knowledge of concurrent SST perturbations. Preliminary analyses suggest that while relatively large hindcast skills are evident in various LR models, the apparent skills decrease substantially when the models are applied to new data. While these results do not prove that such models can never be very useful, they do project difficulties in greatly increasing the skill, especially as long as the available data periods are relatively short