Antarctic sea ice prediction with A convolutional long short-term memory network

Abstract

Antarctic sea ice predictions are becoming increasingly important scientifically and operationally due to climate change and increased human activities in the region. Conventional numerical models typically require extensive computational resources and exhibit limited predictive skill on the subseasonal-to-seasonal scale. In this study, a convolutional long short-term memory (ConvLSTM) deep neural network is constructed to predict the 60-day future Antarctic sea ice evolution using only satellite-derived sea ice concentration (SIC) from 1989 to 2016. The network is skillful for approximately one month in predicting the daily spatial distribution of Antarctic SIC between 2018 and 2022, with the best predictive skill found in austral autumn (MAM) and winter (JJA). ConvLSTM also performs well in real-time prediction in February and September when the Antarctic sea ice extent (SIE) reaches the seasonal maximum and minimum, with the monthly mean SIE error mostly below 0.2 million km². The results suggest substantial potential for applying machine learning techniques for skillful Antarctic sea ice prediction.