Multi-annual predictability of the Atlantic Meridional Overturning Circulation

Decadal climate predictions have the main feature of being initialized, hence lying midway between initialized seasonal forecasts and forced multi-decadal projections. The North Atlantic is among the few places where decadal variations are considered potentially predictable with an added value of the initialization due to the Atlantic Meridional Overturning Circulation (AMOC), which exhibits slow multi-annual fluctuations. A correct representation of this process is fundamental to skillfully predict climate variability in the Northern Hemisphere at these timescales. In this thesis, AMOC predictability is investigated in the CMCC-CM2-SR5 (CMCC Coupled Model v2 in standard resolution) decadal system. The ability of the model to forecast the AMOC is evaluated in both a deterministic and probabilistic way, comparing a set of hindcasts initialized between 1960 and 2018 with observations, ocean reconstructions, and a non-initialized historical simulation. Special attention is devoted to the analysis of AMOC biases. Indeed, it is documented that predictions suffer from initial shocks and tend to drift towards the model's equilibrium state. We find that the potential predictability of the system is high up to a ten-year forecast range, but this is not reflected in the AMOC transport forecast skill, which undergoes a sudden reduction after the first year. An interesting finding is that the drift of the model is start-date dependent: we leverage on this feature to propose a new post-processing approach for the drift adjustment, different from the usual one in which drifts are treated as stationary. The experimented approach significantly increases the forecast skill. Furthermore, we identify a reduction of convection in the Labrador Sea, a feature that previous studies linked with the model drift of the AMOC. Further research with an increased ensemble size of both initialized and historical simulations and with a multi-model set is envisaged.