The skill of empirical and combined/calibrated coupled multi-model South American seasonal predictions during ENSO

Abstract

Abstract. This study addresses seasonal predictability of South American rainfall during ENSO. The skill of empirical and coupled multi-model predictions is assessed and compared. The empirical model uses the previous season August-September-October Pacific and Atlantic sea surface temperatures as predictors for December-January-February rainfall. Coupled multi-model 1-month lead December-January-February rainfall predictions were obtained from the Development of a European Multi-model Ensemble system for seasonal to inTERannual prediction (DEMETER) project. Integrated (i.e. combined and calibrated) forecasts that incorporate information provided by both the empirical and the coupled multi-model are produced using a Bayesian procedure. This procedure is referred to as forecast assimilation. The skill of the integrated forecasts is compared to the skill of empirical and coupled multi-model predictions. This comparison reveals that when seasonally forecasting December-January-February South American rainfall at 1-month lead-time the current generation of coupled models have a level of deterministic skill comparable to those obtained using simplified empirical approaches. However, Bayesian combined/calibrated forecasts provide better estimates of forecast uncertainty than the coupled multi-model. This indicates that forecast assimilation improves the quality of probabilistic predictions. The tropics and the area of South Brazil, Paraguay, Uruguay and Northern Argentina are found to be the two most predictable regions of South America. ENSO years are more predictable than neutral years, the latter having nearly null skill.