Hybrid statistical–dynamical seasonal prediction of summer extreme temperatures in Europe

Abstract

Although summer extreme temperatures over Europe are potentially predictable on seasonal time‐scales, state‐of‐the‐art dynamical seasonal prediction systems (SPSs) exhibit low skills in predicting such events in central and northern Europe. This limitation arises from the underestimation of predictable components of climate variability in the model ensemble. However, recent studies suggest that the skills in predicting extratropical climate can be largely improved through statistical postprocessing techniques, which increase the signal‐to‐noise ratio in the model ensemble. In this study, we evaluate the potential for improving the seasonal prediction skills of European summer extreme temperatures in a multimodel ensemble (MME) of SPSs by applying a teleconnection‐based subsampling technique in the hindcast period 1993–2016. This technique is applied to the North Atlantic Oscillation (NAO) and East Atlantic (EA) modes, which are key drivers of summer extreme temperatures in Europe. Results show that the subsampling substantially improves the MME prediction skills of both the summer NAO and EA. Specifically, correlations between the observed and subsampled MME NAO indices improve from to 0.77, and for the EA they improve from −0.11 to 0.84. Similarly, the root‐mean‐square error of the subsampled MME NAO (EA) index improves from 1.06 (1.02) to 0.65 (0.56). Moreover, retaining those ensemble members that accurately represent the NAO teleconnections enhances the MME prediction skills for the summer European climate, including the occurrence of summer extreme temperatures. This improvement is particularly pronounced in central and northern Europe; that is, the regions where current SPSs show the lowest skills in predicting European heat extremes. In contrast, selecting ensemble members that accurately represent the EA teleconnections does not improve the predictions of summer extreme temperatures. This is likely associated with the model deficiencies in realistically representing the spatial pattern of the summer EA and, thus, the physical processes driving summer extreme temperatures.