Assessment of Arctic and Antarctic Sea Ice Predictability in CMIP5 Decadal Hindcasts

This work examines the ability of coupled global climate models (CGCMs) to predict decadal variability of Arctic and Antarctic sea ice. We analyze decadal hindcasts/predictions of 11 CGCMs from the Coupled Model Intercomparison Project Phase 5 (CMIP5). Large discrepancies between the decadal hindcast and observed sea ice extent are found. This is particularly true for the models having large systematic biases and using a full-field initialization approach, in which the predicted sea ice extent quickly drifts away from the initial constraint, causing the decadal climate prediction to deteriorate. The anomaly correlation analysis between the decadal hindcast and observed sea ice suggests that in the Arctic, for most models, the areas showing significant predictive skill become broader associated with increasing lead times, covering large parts of the Arctic Ocean at 6-8 years ahead. This area expansion is primarily because nearly all of the models are capable of predicting a decreasing Arctic sea ice cover, consistent with the observations. Sea ice extent in the north Pacific side has better predictive skill than that in the north Atlantic side (particularly at a lead-time of 3-7 years), but there is a re-emerging predictive skill in the north Atlantic side at a lead-time of 6-8 years. In contrast to the Arctic, Antarctic sea ice hindcasts do not show broad predictive skill at any time scales, and there is no obvious improvement linking the areal extent of significant predictive skill to lead-time increases. This might be because nearly all of the models predict a retreating Antarctic sea ice cover, in contrast to the observations. The multi-model ensemble mean (MMEE) shows predictive skill over large parts of the eastern Pacific sector of the Antarctic. In general, for both the Arctic and Antarctic, the predictive skill of the MMEE outperforms most models and the persistence prediction at longer time scales.