The Atlantic Multidecadal Oscillation Inferred from the Forced Climate Response in Coupled General Circulation Models

Abstract Instrumental sea surface temperature records in the North Atlantic Ocean are characterized by large multidecadal variability known as the Atlantic multidecadal oscillation (AMO). The lack of strong oscillatory forcing of the climate system at multidecadal time scales and the results of long unforced climate simulations have led to the widespread, although not ubiquitous, view that the AMO is an internal mode of climate variability. Here, a more objective examination of this hypothesis is performed using simulations with natural and anthropogenic forcings from the Coupled Model Intercomparison Project phase 3 (CMIP3) database. Ensemble means derived from these data allow an estimate of the response of models to forcings, as averaging leads to cancellation of the internal variability between ensemble members. In general, the means of individual model ensembles appear to be inconsistent with observed temperatures, although small ensemble sizes result in uncertainty in this conclusion. Combining the ensembles from different models creates a multimodel ensemble of sufficient size to allow for a good estimate of the forced response. This shows that the variability in observed North Atlantic temperatures possess a clearly distinct signature to the climate response expected from forcings. The reliability of this finding is confirmed by sampling those models with low decadal internal variability and by comparing simulated and observed trends. In contrast to the inconsistency with the ensemble mean, the observations are consistent with the spread of responses in the ensemble members, suggesting they can be accounted for by the combined effects of forcings and internal variability. In the most recent period, the results suggest that the North Atlantic is warming faster than expected, and that the AMO entered a positive phase in the 1990s. The differences found between observed and ensemble mean temperatures could arise through errors in the observational data, errors in the models’ response to forcings or in the forcings themselves, or as a result of genuine internal variability. Each of these possibilities is discussed, and it is concluded that internal variability within the natural climate system is the most likely origin of the differences. Finally, the estimate of internal variability obtained using the model-derived ensemble mean is proposed as a new way of defining the AMO, which has important advantages over previous definitions.