Constraining temperature variability projections using SMILEs that best represent observed variability

Abstract

Projecting how temperature variability is likely to change in the future is important for understanding future extreme events. This comes from the fact that such extremes can change due to both changes in the mean climate and its variability. The recent IPCC report found large regions of low model agreement in the change of temperature variability in both December, January, February (DJF) and June, July, August (JJA) when considering 7 Single Model Initial-Condition Large Ensembles (SMILEs). In this study we use the framework described by Suarez-Gutierrez et al, (2021) to constrain future projections of temperature variability by selecting the SMILEs that best represent observed variability. We consider 11 SMILEs with CMIP5 and CMIP6 forcing over 9 ocean regions and 24 land regions. We find that CESM2-LE, GFDL-SPEAR-MED and MPI-GE-CMIP6 perform best in DJF and CESM2-LE, CESM1-LE and GFDL-SPEAR-MED perform best in JJA. We find that the Southern Ocean is poorly represented in all models and that few models can represent Central America, the Amazon, North-East Brazil, and eastern and southern Asia, with western and eastern Africa poorly represented in JJA and northern Australia poorly represented in DJF. Overall models perform well over the Northern Hemisphere land masses. When we constrain temperature variability estimates, variability is generally lower, particularly over South America, Africa, and Australia – the same regions where the constraint improves projections, and where the projected change is smaller in the constraint. Where hot extremes increase so do cold extremes showing projected changes in all regions as a fattening or thinning of the distribution rather than a change in skewness.