Cyclostationary EOF Modes of Antarctic Sea Ice and Their Application in Prediction

Abstract

AbstractThis study examines the modes of sea ice concentration (SIC) in the Southern Ocean from a new perspective using the cyclostationary empirical orthogonal function (CSEOF) method, which decomposes a space‐time field into a combination of modulating variability on phase‐locked annual cycles. The first CSEOF mode of SIC (72.55% of total variance) represents the normal annual cycle modulated by a low‐frequency oscillation. The second mode of SIC (3.39%) represents the trend pattern, which reveals that the SIC increases in most regions of the Southern Ocean except for those adjacent to the Antarctic Peninsula, wherein the decreasing SIC trend is evident in austral warm seasons. This trend mode also explains that the sea ice in the Weddell Sea is increasing in austral summer and decreasing in austral winter. The Principal Component (PC) of this model is negatively corrected with the Southern Annular Mode index with a correlation coefficient of about −0.5. The third mode of SIC (2.48%) represents a wavenumber‐2 Antarctic Circumpolar Wave mode that is linked to ENSO. As the CSEOF method decomposes the space‐time SIC field into low‐frequency modulations of phase‐locked annual cycles, it can be used for SIC prediction, by taking the spatial modes as the basis functions and predicting their PCs from the autoregressive models. However, this CSEOF‐based SIC prediction model exhibits unsatisfactory forecast skills, even with optimized model parameters. Therefore, it is still necessary to study more SIC prediction models and evaluate their forecast skills in the future.