Abstract: To model summer sea ice in Antarctica, three predictors are selected: (1) sea surface temperature prediction in the Southern Ocean (40°S–80°S) from the Climate Forecast System version 2 (CFSv2), (2) observed sea surface temperature around the Maritime Continent (100°E–130°E, 10°N–15°S) in early austral spring; (3) observed sea level pressure in the southern extratropics (20°S–90°S) in early austral spring. In this study, the singular value decomposition method is used to extract relevant information, then a regression model is applied to predict the sea ice field. When applied retrospectively to simulations for 1983–2018, this model yields significantly improved results compared with raw CFSv2 predictions: temporal correlation between predicted sea ice concentrations at single grid points and observations is significant, with an average value of 0.76 over the Southern Ocean. From cross-validation results, prediction results are significantly better than those of CFSv2, with predicted sea ice extent significantly improved relatively to CFSv2 and to persistence prediction. These results indicate that the prediction performance of the joint statistical-dynamical model described in this study is high. Therefore, this model is important to predict Antarctic sea ice, for example when planning scientific survey activities in Antarctica.

Key words:  Antarctic summer sea ice, dynamic-statistical downscaling model, Climate Forecast System, persistence prediction