Empirical forecasts of tropical Atlantic sea surface temperature anomalies

Abstract

The interannual variability of the tropical Atlantic is characterized by warmings and coolings similar to the Pacific ones (El Nifio), and by an interhemispheric signal of decadal variability. The magnitudes of the Gulf of Guinea warmings are less and, therefore, they do not significantly influence the earth's climate, as the El NiiidSouthern Oscillation (ENSO) does. In the past, they have been studied because of their connections with the recurrent droughts in the Sahel region. Recently, a number of modelling studies have tried to establish their dependence on ENSO. The real existence of an interhemispheric decadal signal, and its predictability, is also a widely discussed topic. Forecast studies have recently appeared for both the north tropical Atlantic and the Gulf of Guinea regions, and are now operationally available from the National Oceanic and Atmospheric Administration.In the present work we try first to understand the tropical Atlantic variability in terms of forcings external to the basin. These are identified from 48 years of monthly anomalies of sea surface temperature (SST) data obtained by combining the Comprehensive Ocean and Atmosphere Dataset (COADS) and the Integrated Global Ocean Services System (IGOSS) dataset, and then using the Bayesian theory of estimation. Besides the ENSOrelated scales, our analysis retains a decadal time scale in the data variability.Next, a model is built to forecast the most important features of the equatorial Atlantic interannual variability. These features are monitored through two indices, the Gulf of Guinea and the north tropical Atlantic indices ('the predictands'). Predictor fields are identified from our preliminary analysis, as those time series significantly correlated with the predictands. These correspond to grid points in the tropical Pacific (mainly the Nifio3 region, (5OS-5' N. 150°W-900W)) and tropical Indian oceans. Forecasts were issued for 28 years, at three-monthly intervals. For the north tropical Atlantic index, we have a good forecast skill at leads greater than four months with predictors obtained from east equatorial Pacific time series. For the Gulf of Guinea index, a good forecast skill can be obtained only when we include time series of the equatorial Indian ocean, as well as the east equatorial Pacific, among the predictors. Any of the forecasts presented here show useful forecast skill that, at least, beats persistence at leads greater than four months