Antarctic fast ice in Terra Nova Bay is crucial for regional climate, ecosystems, and scientific logistics but has recently shown unprecedented decline, impacting operations at Jang Bogo Station. Despite its significance, there remains limited predictive understanding of Antarctic fast ice dynamics, especially regarding thickness and stability, with most existing studies constrained to observational monitoring of extent and seasonal variability. This study develops a predictive model for fast ice area using XGBoost machine learning, integrating atmospheric reanalysis data (skin temperature and wind), satellite‐derived polynya area, and fast ice extent. Incorporating time‐lagged variables (up to 5 months) significantly improved model (lagged model) accuracy (validation R = 0.57) over a model with no‐lag variables (R = 0.38). SHapley additive explanation (SHAP) analysis revealed that Manuela automatic weather station skin temperature with a 1‐month lag and zonal wind with a 2‐month lag were key predictors in the lagged model, highlighting the importance of antecedent atmospheric conditions. These findings demonstrate the utility of machine learning for forecasting fast ice, offering vital insights for adapting to changing Antarctic coastal environments and supporting logistical planning.
Kim et al. (Wed,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: