Statistical models for predicting rainfall in the Caribbean

Abstract

Separate predictive models are created for the Caribbean early wet season (May–June–July) and late wet season (August–September–October). Simple correlations are used to select predictors for a Caribbean rainfall index and predictive equations are formulated using multiple linear regression. The process is repeated after long term trends are removed from the Caribbean rainfall index and the models validated using a number of statistical methods. Four variables are confirmed as predictors for the early season: Caribbean sea surface temperature anomalies, tropical North Atlantic sea level pressure anomalies, vertical shear anomalies in the equatorial Atlantic, and the size of the Atlantic portion of the Western Hemisphere Warm Pool. Only the first two are retained in the late season model. On the interannual time-scale, equatorial Pacific sea surface temperature anomalies become significant in both seasons. The NINO3 index is retained among the predictors for the early season, and zonal gradients of sea surface temperature between the equatorial Pacific and tropical Atlantic are retained for the late season. The results also indicate spatial variation in the importance of the seasonal predictors.