Extreme Sea Ice Loss on Subseasonal Timescales in S2S Forecast Models

Abstract:

Sea ice variability on subseasonal timescales is explored in observations and S2S forecast models at both pan-Arctic and regional scales. Particular attention is paid to the characteristics and predictability of very rapid sea ice loss events (VRILEs, which are extremes in the 5-day change in sea ice extent of a region). Rapid sea ice loss on subseasonal timescales exhibits regional and seasonal variability throughout the Arctic. In the Arctic Ocean, most VRILEs occur in the summer, but in the Atlantic region (primarily the Barents and East Greenland Seas), they occur throughout the year. S2S forecast models are able to capture the seasonality of VRILEs, though the models exhibit some biases in their mean states and variability. We explore if the predictability of sea ice is different during VRILE days versus typical days. While the forecast skill of sea ice extent is highly variable by region and by model, generally VRILE days are less predictable than typical days, especially for longer forecast lead times. Among the more skillful models, sea ice extent in many regions can be skillfully predicted for about 2 to 3 days longer on typical days than on VRILE days. When predictions of sea ice are restricted to summertime months only, the prediction skill of sea ice extent on VRILE days is even closer to that of typical days, especially at lead times of up to one week. In large swaths of the Arctic Ocean, much of the variability in the sea ice cover in the month of September can be explained by the cumulative number of VRILEs from June to August. Therefore, a path to improving September prediction is to focus attention on improving the prediction of VRILEs throughout the summer. We compare the atmospheric conditions associated with VRILEs in predictions and observations, and how they compare with forecast lead time.