Assessing the Skill of Six CMIP6 Climate Models in Predicting Arctic Sea Ice to Narrow the Uncertainty in Projections of the September Ice-Free Arctic

Arctic sea ice is an important indicator and regulator of the state of the global climate. As global temperatures rise due to climate change, Arctic sea ice is undergoing unprecedented rapid melting, bringing vast physical and socio-economic implications for the global Earth system and its inhabitants. In this thesis, I aim to characterise the implications on Arctic sea ice of the SSP 5-8.5 “worst-case” high-emissions scenario, and to narrow the broad uncertainty in timing of summer sea-ice-free conditions by identifying models with higher relative predictive skill. I compare key spatial and temporal trends and EOF principal modes of variability of sea ice concentration (SIC) for both the historical and SSP 5-8.5 projections of six models, where the CanESM5 model emerges as the most likely to predict observed SIC. I find a statistically significant, almost 1:1, linear relationship between early and late predictive skill, based on a comparison of root-mean-square-error between observed and simulated SIC for early and late portions of the data. Given that historical predictive skill of CanESM5 translates to high predictive skill for future projections, we infer that summer sea-ice-free Arctic will likely follow the trajectory outlined in CanESM5 projections: September sea-ice free by 2033, with a narrower 5%-95% confidence interval of 2027-2039. These findings highlight the importance of taking concrete action on Paris Agreement 1.5ºC targets to avoid the realisation of the stark consequences of the SSP 5-8.5 trajectory.