Skillful seasonal predictions of Antarctic sea ice in a dynamical forecast system

Compared to the Arctic, seasonal predictions of Antarctic sea ice have received relatively little attention. In this work, we utilize a coupled dynamical prediction system to assess both the current seasonal prediction skill and the limits of predictability for regional Antarctic sea-ice extent (SIE). We perform two complementary suites of ensemble prediction experiments using a global atmosphere-ocean-sea ice-land model developed at the Geophysical Fluid Dynamics Laboratory. First, we consider a set of retrospective initialized seasonal predictions spanning 1981-2018. These retrospective forecasts are found to skillfully predict detrended winter SIE in the Amundsen and Bellingshausen, Indian, and West Pacific sectors at lead times up to 11 months in advance and in the Weddell Sea at leads times up to 6 months. Prediction skill in these regions generally exceeds that of a persistence forecast. Second, we consider a suite of “perfect model” predictability experiments performed with the same coupled model, in which ensembles are initialized from nearly identical initial states. These experiments quantify the potential prediction skill achievable in this system, and show long-lead (> 1 year) predictability in the Weddell, Indian, and Ross sectors. We examine the sources of prediction skill in both experiment suites, finding that advected upper ocean heat content plays a critical role for winter SIE predictions. Overall, these results suggest a promising potential for providing operational regional Antarctic sea-ice predictions on seasonal timescales.