Recent Seasonal Forecast Skill in a Sea-Ice Prediction System. Have We Entered a Period of Inherently Low Predictability?

Abstract:

We explore the skill in seasonal forecasts of September Arctic sea-ice extent in dynamical models that are members of the Sea Ice Outlook. We find that the multi-model ensemble only offers skill for the latest summer submission in August, and that throughout the forecasting period skill is lower than that found in hindcasts of sea-ice extent performed during earlier periods of the modern satellite record. The model-mean ensemble offers slightly higher skill, but does not beat a damped persistence forecast. We also find that the models are equally unsuccessful at predicting each other, indicating a large divergence in model physics and/or initial conditions. We explore whether the recent SIO years have been inherently more unpredictable than past decades with the help of an idealized GCM experiment. We find that in an unforced control run there are several periods of high/low persistence in the Arctic sea-ice system, but these do not translate into high/low predictability. It is thus likely that changing predictability arises from changing boundary conditions.