Oral

IAMAS

JMP02 - Machine Learning in atmospheric, ocean and earth-system prediction: forecasting, simulation and scientific analysis

Importance of Atlantic sea surface temperature to Arctic sea ice variability revealed by deep learning

Arctic sea ice has changed dramatically over the observational period, influenced by both anthropogenic forcing and internal climate variability. While numerous studies have explored connections between internal variability of sea surface temperature (SST) patterns and Arctic sea ice changes, the relative importance and mechanisms of these teleconnections remain unclear. Here, we employ deep neural network (DNN) model to reconstruct Arctic sea ice extent (SIE) based on observed SST in three major ocean basins (Indian, Pacific and Atlantic Oceans). Our results reveal that Atlantic SST demonstrates the strongest and most stable connection with Arctic SIE, particularly with a 10-day lead time, compared to Pacific (60-day lead) and Indian Ocean (30-day lead). The linear regression model fails to capture the robust connection between Atlantic SST and SIE. Through explainable methods analysis, we identify the Caribbean Sea and Gulf Stream as key regions where Atlantic SST variability has most significant impact on Arctic SIE. Additionally, the analysis reveals non-negligible nonlinear relationships between Atlantic SST and Arctic SIE on interannual timescale, potentially mediated through SST-driven atmospheric water vapor content and circulation pattern anomalies. This paper highlights the uniqueness of DNN in understanding SST-SIE teleconnections, indicating the importance of incorporating Atlantic SST to improve Arctic sea ice predictability.