Challenges faced by the Large Ensemble Single Forcing Model Intercomparison Project models in representing the response of the North Atlantic atmospheric and oceanic circulation to external forcings

Reliable projections of surface climate over Europe depend on the trustworthiness of simulations of the North Atlantic atmospheric circulation from climate models. Recent examples of discrepancies between models and observations have raised the possibility that models cannot capture the long-term observed trends in the North Atlantic circulation. Here, we examine the ability of models contributed to the Large Ensemble Single Forcing Model Intercomparison Project (LESFMIP) to simulate historical changes in the atmospheric and oceanic circulation in the North Atlantic sector. From 1951 to 2014, the wintertime North Atlantic jet has strengthened, and the NAO has trended towards its positive phase, like the Northern Annular Mode in the stratosphere. All-forcing historical simulations show only a very weak trend that is missing from nearly all individual models. Nonetheless, the models skilfully predict the observed multi-decadal variability of the NAO, and the sum of the individual forcing simulations simulates trends closer to that observed. Specifically, the hist-GHG ensemble captures the observed signal better than historical; however, there is substantial non-additivity between the sum of each of the individual forcings and historical. Some of this nonadditivity appears to be linked to sea surface temperature warming near the ice-edge in the Barents-Kara Sea and near Greenland.  The divergence between models and observations is much less pronounced over land or when the period since 1979 is considered. In this period, the models vs. observations discrepancy is mainly in boreal summer and appears to be due to aerosols. Our results suggest that projections of atmospheric circulation in the Euro-Atlantic sector may be unreliable because they underestimate the response to human emissions or the magnitude of multidecadal-to-centennial time scale internal variability; we cannot rule out observational uncertainty as a potential cause.