Improved simulation of extratropical North Atlantic atmosphere-ocean variability in HighResMIP models

The simulated North Atlantic atmosphere­–ocean variability is assessed in a subset of models from HighResMIP that have either low-resolution (LR) or high-resolution (HR) in their atmosphere and ocean model components. In general, the LR models overestimate the low-frequency variability of subpolar sea-surface temperature (SST) anomalies and underestimate their correlation with the NAO compared to ERA5 reanalysis. These biases are substantially reduced in the HR models, and it is shown that the improvements are related to a reduction of intrinsic (non-NAO-driven) variability of the subpolar ocean circulation.To understand the mechanisms behind the overestimated intrinsic subpolar ocean variability in the LR models, a link is demonstrated between the biases in subpolar ocean variability and known biases in the mean state of the Labrador-Irminger seas. Supporting previous studies, the Labrador-Irminger seas are found to be too cold and too fresh in the LR models compared to observations from EN4 and the HR models. This causes upper-ocean density and hence convection anomalies in this region to be more salinity-controlled in the LR models versus more temperature-controlled in the HR models. It is hypothesized that this may cause the excessive subpolar ocean variability in the LR models by 1) promoting a positive feedback between subpolar upper-ocean salinity, convection and Atlantic Meridional Overturning Circulation (AMOC) anomalies, and 2) weakening the negative feedback between subpolar upper-ocean temperature, convection and AMOC anomalies that is apparent in the HR models. The results overall suggest that mean ocean biases play an important role in the simulation of the variability of the extratropical ocean.