Generative AI models capture realistic sea-ice evolution from days to decades

The introduction of GenSIM marks a significant advancement in climate modeling, particularly in understanding Arctic sea-ice dynamics. This generative AI model, trained on two decades of sea-ice-ocean simulation data, predicts the evolution of sea ice at 12-hour increments and extends its forecasts up to 30 years. By accurately reproducing the dynamical properties of sea ice, including its leads and ridges, GenSIM not only captures long-term trends in sea-ice volume but also learns hidden interactions between multi-year ice and ocean conditions. This capability is crucial, as sea ice plays a vital role in stabilizing the Earth system. The model's ability to extrapolate from short-term forecasts to decadal simulations while maintaining physical consistency could revolutionize how scientists approach climate predictions and inform policy decisions regarding climate change mitigation and adaptation strategies.