Intrinsic Limits on Determining the Skill of North Atlantic Ocean Decadal Predictions

The northern North Atlantic is important globally through its impact on the Atlantic Meridional Overturning Circulation (AMOC), Atlantic Multidecadal Variability (AMV) and through widespread atmospheric teleconnections. The region has been shown to be potentially predictable a decade ahead with the skill of decadal predictions assessed against reanalyses of the ocean state. Here, we show that prediction skill in this region is strongly dependent on the choice of reanalysis ("truth") used for validation. Indeed, multiannual skill in key metrics such as the AMOC can be considered independent of the choice of prediction model. Instead, this skill is related to the similarity between the nature of interannual density variability in the underlying climate model and the chosen reanalysis. As such, we need to understand the past northern North Atlantic variability more fully if we can hope to have confidence in decadal climate predictions in this region. The climate models used in these decadal predictions are also used in century-scale climate projections, which raises questions about the sensitivity of these projections to the models' innate North Atlantic density variability.