Contrasting the internal and external components of Atlantic Multidecadal Variability in CMIP6 historical simulations

Over the observed period, North Atlantic Sea surface temperatures have gone through cycles of anomalous warming and cooling relative the global mean. This variability has become known as the Atlantic Multidecadal Variability (AMV), and it has been associated with important regional climate impacts. However, in recent years there has been considerable controversy over the origins of AMV. In particular, there is debate over whether AMV is a natural phenomenon (e.g., an expression of internal variability or natural external forcings), or whether it was caused by human activity through the impact of anthropogenic aerosol forcing.

Here, an analysis of CMIP6 multi-model historical simulations is presented which isolates the internal and externally forced AMV. The analysis shows that, although there is substantial externally forced AMV in the CMIP6 historical simulations, the forced variability is part of a wider hemispheric signal and is not specific to the North Atlantic like in observations. Therefore, the magnitude of the externally forced variability is highly dependent on the definition of the AMV index used. Ocean circulation changes consistently lead the internal AMV across models, but there is no-clear relationship for the external AMV. AMV is also associated with broader changes than just sea surface temperatures, but this multivariate fingerprint of AMV is significantly different between the internal and external components. For example, internal AMV is associated with salinity anomalies and increased turbulent heat loss across the subpolar North Atlantic that agree broadly with observations. However, in contrast, the externally forced AMV is associated with freshening and reduced heat loss across the subpolar North Atlantic and especially in models with the strongest aerosol forcing. Overall, the analysis suggests that internal variability remains a likely hypothesis to explain AMV, but questions remain on whether models adequately simulate the forced response.