Realism of Simulated Internal Variability in Arctic Sea Ice

Arctic sea ice has decreased dramatically over the last few decades, particularly in the summer months. The observed decline is faster than most CMIP5 models, but if internal variability is considered, models and observations are not inconsistent. With only one realization of reality in observations, it is difficult to disentangle the role of internal variability from the forced response. We directly compare one metric of internal variability by resampling both observations and models. So far we have compared five CMIP5 models from the CLIVAR multi-model large ensemble archive (CanESM2, CESM1, CSIRO MK36, GFDL ESM2M, and MPI ESM1). For the pan-Arctic, the selected models were found to have similar internal variability to observations in the summer, but overestimated internal variability in spring, peaking at 50-100% overestimation in March. Spatially, we find the variability in ice edge region is consistently modeled well in March. In September, the simulated internal variability is too high in the central Arctic often by a factor of two or more, the marginal seas in general have a low bias of the order 25%, but this varies considerably between models and regions.