Sensitivity of Arctic Sea Ice Extent to Temperature in CMIP6 models

The state of Arctic sea ice is strongly related to global temperature. While most global climate models in the CMIP framework have captured this general relationship, the magnitude of sea ice decline observed over the satellite era has generally been faster than forecast. The degree to which this is a result of under-prediction of temperature change or the sensitivity of the sea ice to these changes has not been thoroughly investigated.

In this study, we use linear regression to evaluate the sensitivity of Arctic sea ice to global and Arctic temperature changes across CMIP6 models. First, we examine the sensitivity under different forcings using CMIP6 pre-industrial, historical and SSP370 simulations and compare these values to observations. As in previous studies, we find a linear relationship is generally appropriate for modelling the relationship between sea ice and global temperature.

We find that both sensitivity and correlation have a strong seasonal dependence across nearly all model runs and observations. Arctic temperatures are most strongly correlated with sea ice in late summer and early fall in both models and observations. Global temperatures are less correlated with sea ice, and the relationship shows less seasonal dependence. Sensitivity to Arctic temperatures is again stronger in observations than models particularly in summer months and on annual timescales.

Across CMIP6 models, sea ice sensitivity is higher in more strongly forced scenarios (i.e. satellite era and SSP370) compared to weakly forced scenarios (pre-industrial control runs and early period of historical simulations). This indicates either a mean state dependence of the sensitivity, or that the sea ice sensitivity to temperature is dependent on the forcing scenario. Further investigation reveals that the intermodel spread in the mean state sea ice area can explain some of the variability in sensitivity, particularly in summer.

To investigate the different roles of interannual variability and forced changes, we examine the sensitivity of filtered and detrended data as well as coherence. We find that the sensitivity is much greater for the low frequency component of the relationship between sea ice and temperature than the high frequency component. The relative magnitudes of the forced sea ice response and interannual variability therefore explain some of the sensitivity differences across CMIP6 models and scenarios. Overall, these results suggest a consistent pattern in the response of Arctic sea ice to temperature across models, which may be useful for diagnosing inaccuracies in climate models and improving future sea ice projections.