Answers are grounded in this paper and cite their sources.
This analysis demonstrates improved accuracy of a deep learning forecasting system for sea ice concentration, suggesting it outperforms traditional models.
There has been a steady increase in marine activity throughout the Arctic Ocean during the last few decades, and maritime end users are requesting skilful high-resolution sea ice forecasts to ensure operational safety. Different studies have demonstrated the effectiveness of utilizing computationally lightweight deep learning models to predict sea ice properties in the Arctic. In this study, we utilize operational atmospheric forecasts, ice charts, and sea ice concentration passive microwave observations as predictors to train a deep learning model with future ice charts as ground truth. The developed deep learning forecasting system predicts regional ice charts covering parts of the East Greenland and Barents seas at 1 km resolution for 1–3 d lead time. We validate the deep learning system performance by evaluating the position of forecasted sea ice concentration contours at different concentration thresholds. It is shown that the deep learning forecasting system achieves a lower error for several sea ice concentration contours when compared against baseline forecasts (persistence forecasts, sea ice free drift, and a linear trend) and two state-of-the-art dynamical sea ice forecasting systems (neXtSIM and Barents-2.5) for all considered lead times and seasons.
No takes yet. Share an insight, caveat, or question.
Kvanum et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: