Decadal Temperature Prediction Skill in the North Atlantic Region in CMIP6

Abstract:

We analyze the representation of observed North Atlantic SST variations since 1960 in a multi-model ensemble composed of 58 models. Initialized decadal hindcast and historical simulations from the 6th phase of the Coupled Model Intercomparison Project (CMIP) show better prediction skill for North Atlantic SST (72% observed variance explained by the hindcasts), than equivalent simulations from CMIP5 (56% observed variance explained by the hindcasts). Reduced improvement of correlation skill from initialization for SST in the subpolar gyre (SPG) region in CMIP6 compared to CMIP5 can be traced back to much better agreement between CMIP6 historical simulations and observations since the 1980s – ~63% explained variance in CMIP6 versus ~36% in CMIP5 – which is partly caused by a realistic response of CMIP6 models to volcanic and solar forcing, explaining ~55% of the observed SST variance. In CMIP6, hindcast initialization continues to be valuable for capturing the amplitude of change and reducing the uncertainty of decadal SPG SST predictions.

SPG SST skill differs between individual model ensemble means in CMIP6 hindcasts. We identify two main groups of models, one with high SPG SST hindcast skill across lead time, and one with high skill at early lead time and deterioration of skill over lead time. This classification of different hindcast systems is robust across several popular deterministic skill metrics. Prediction skill for summer surface air temperature over Europe appears to be regionally connected to the individual models’ skill at predicting SPG SST, illustrating the societal value of understanding SPG SST prediction skill.