Interpreting CMIP6 projections of 21st century Antarctic sea ice loss

Compared to CMIP5, the CMIP6 ensemble shows modest improvements in some aspects of Antarctic sea ice simulation and in particular a reduction of inter-model spread in historical sea ice area (SIA). However, large biases remain. Results from CMIP5 have previously suggested that ensemble regression and model selection techniques may provide solutions to the challenge of making projections of future Antarctic SIA in the presence of large historical biases. Here we revisit and extend this analysis with the new ensemble, focusing on strong forcing scenarios.

In summer (February) the historical climatology of SIA is a strong linear constraint on projections of SIA, in both generations. This is because the strong forcing leads to the loss of the majority of summer SIA in each model, so that the models that start with greater SIA exhibit greater reductions. This relationship is stronger in CMIP6 than CMIP5, which derives from the fact that CMIP5 contains many more models that have very large positive biases in historical SIA and do not lose the majority of ice.

In winter (September), inter-model spread in SIA climatology explains just under half the variance in projections of SIA change for both ensembles. The mean historical winter climatology is similar between generations, as is the regression slope of SIA change against SIA climatology. However, there is a greater reduction of SIA in CMIP6 than CMIP5. We find this to be statistically related to greater global mean warming in CMIP6 than CMIP5, and therefore to the well-known larger climate sensitivity in the CMIP6 ensemble.

These findings imply that, depending on season, a different balance of local (SIA climatology) and global (GMST change) drivers can be used to explain inter-model and inter-generation spread in projections of SIA loss. They also connect our ability to project Antarctic SIA loss to our understanding of the fidelity of higher CMIP6 climate sensitivity. Questions remain about whether models are correct in their simulation of Antarctic SIA sensitivity to global surface temperature.