Title: Multi-decadal Changes in the Tropics Drive Variability and Predictability of Antarctic Sea Ice
Abstract
Advancements of dynamic and statistical models have progressively improved weather forecasts and climate projections of polar sea ice concentration (SIC). However, a gap remains at the subseasonal scale (1-8 weeks) due to limited understanding of ice-related physical mechanisms. We recently developed a deep learning model, the Sea Ice Prediction Network (SIPNet), that can predict SIC without accounting for complex physical processes. Compared with mainstream dynamical models from ECMWF, NCEP, and GFDL, as well as the LDEO linear Markov Model, SIPNet outperforms them all, effectively filling a gap in subseasonal Antarctic SIC prediction capability. SIPNet has the highest skill in autumn but the lowest in spring. In addition, the Weddell Sea exhibits higher sea-ice predictability, whereas predictability is low in the West Pacific. SIPNet also reveals that ENSO events exert cross-timescale influences on sea ice's subseasonal predictability, through generating large sea ice anomalies. ENSO becomes a key source of sea ice predictability at lead times of 3 weeks and longer, with El Niño enhancing linear predictability more than La Niña in the Amundsen and Bellingshausen Seas, the Ross Sea, and the Indian Ocean. La Niña mainly enhances the nonlinear predictability of sea ice, particularly in the Ross Sea ...
(full abstract available at https://docs.google.com/document/d/1-K10HNghzPuN5vKRwIY8XI8L6dfVIPnC/edit )
