Prediction of interannual North Atlantic sea surface temperature and its remote influence over land
The quality of a multimodel of six coupled climate forecast systems initialized with observations—relative to the accompanying uninitialized system—to re-forecast the future annual-mean North Atlantic sea surface temperature (SST) departures is described. The study concludes that, measured by the anomaly correlation (AC) skill, the evolution of the leading two empirical orthogonal function modes of North Atlantic SSTs are skillfully forecast throughout the 9-year forecast range. This skill results in part from the predictions of the trend. The skill of the detrended modes, i.e., in absence of SST variability generated by the trend, is reduced, but still statistically distinguishable from zero throughout the 9-year forecast for the first mode and exclusively in the first two forecast years for the second mode. The initialization effect on the AC skill in the initialized system is statistically distinguishable from the one without initialization for the detrended first mode during the first three forecast years and the first forecast year only when the trend in North Atlantic SSTs is included. All six initialized systems of the multimodel are capable to skillfully forecast the shift of the full first mode of North Atlantic SST anomalies in the mid 1990s at all leads with HadCM 3 and EC-Earth 2.3 outperforming other systems. All systems share an intrinsic bias in simulating annual-mean SST variability in the North Atlantic. The study finds that the area-average AC skill (i.e., of a forecast containing regional information) of the North Atlantic influence on anomalous European temperature in the initialized multimodel is positive and statistically distinguishable from zero throughout the 9-year forecast for the full field case. On the other hand, a continent-wide forecast (i.e., without any regional information) of future European precipitation departures associated with North Atlantic SST variability is skillful throughout the 9-year forecast for the full field case—once the model remote influence is corrected by the observed. For the detrended case, the forecasts of the influence of the North Atlantic SSTs on both interannual temperature and precipitation departures in Europe show residual skill—which means that the multimodel is able to predict part of them beyond the trend. However, the forecasts of the North Atlantic influence on precipitation departures in the Sahel turn out not to be skillful.