Abstract. The shrinking Arctic sea ice cover observed during the last decades is probably the clearest manifestation of ongoing climate change. While climate models in general reproduce the sea ice retreat in the Arctic during the 20th century and simulate further sea ice area loss during the 21st century in response to anthropogenic forcing, the models suffer from large biases and the model results exhibit considerable spread. The last generation of climate models from World Climate Research Programme Coupled Model Intercomparison Project Phase 5 (CMIP5), when compared to the previous CMIP3 model ensemble and considering the whole Arctic, were found to be more consistent with the observed changes in sea ice extent during the recent decades. Some CMIP5 models project strongly accelerated (non-linear) sea ice loss during the first half of the 21st century. Here, complementary to previous studies, we compare results from CMIP3 and CMIP5 with respect to regional Arctic sea ice change. We focus on September and March sea ice. Sea ice area (SIA) variability, sea ice concentration (SIC) variability, and characteristics of the SIA seasonal cycle and interannual variability have been analysed for the whole Arctic, termed Entire Arctic, Central Arctic and Barents Sea. Further, the sensitivity of SIA changes to changes in Northern Hemisphere (NH) averaged temperature is investigated and several important dynamical links between SIA and natural climate variability involving the Atlantic Meridional Overturning Circulation (AMOC), North Atlantic Oscillation (NAO) and sea level pressure gradient (SLPG) in the western Barents Sea opening serving as an index of oceanic inflow to the Barents Sea are studied. The CMIP3 and CMIP5 models not only simulate a coherent decline of the Arctic SIA but also depict consistent changes in the SIA seasonal cycle and in the aforementioned dynamical links. The spatial patterns of SIC variability improve in the CMIP5 ensemble, particularly in summer. Both CMIP ensembles depict a significant link between the SIA and NH temperature changes. Our analysis suggests that, on average, the sensitivity of SIA to external forcing is enhanced in the CMIP5 models. The Arctic SIA variability response to anthropogenic forcing is different in CMIP3 and CMIP5. While the CMIP3 models simulate increased variability in March and September, the CMIP5 ensemble shows the opposite tendency. A noticeable improvement in the simulation of summer SIA by the CMIP5 models is often accompanied by worse results for winter SIA characteristics. The relation between SIA and mean AMOC changes is opposite in September and March, with March SIA changes being positively correlated with AMOC slowing. Finally, both CMIP ensembles demonstrate an ability to capture, at least qualitatively, important dynamical links of SIA to decadal variability of the AMOC, NAO and SLPG. SIA in the Barents Sea is strongly overestimated by the majority of the CMIP3 and CMIP5 models, and projected SIA changes are characterized by a large spread giving rise to high uncertainty.
Seasonal Prediction from Arctic Sea Surface Temperatures: Opportunities and Pitfalls
