Even with aggressive emissions cuts, the ice sheet is very likely to lose mass by 2100, while very high emissions could add up to 25.4 centimetres to sea levels.
The researchers used a machine-learning system to examine how different physical assumptions affect projections of Antarctic Ice Sheet change. They also calibrated the projections against satellite observations to reduce bias in the model results.
The analysis found at least a 0.92 probability that the ice sheet is committed to mass loss this century, even with aggressive emissions reductions. Higher emissions were associated with greater mass loss by 2100, with at least a 0.89 probability; under very high emissions, the projected sea-level contribution had a median of 15.7 centimetres and a 95th-percentile value of 25.4 centimetres.
Antarctica’s projected ice loss
The researchers estimate that it is very likely—defined in the study as a probability of at least 0.92—that the Antarctic Ice Sheet will lose mass during the twenty-first century, even under aggressive emissions-reduction scenarios. They also find that higher emissions are very likely to cause greater Antarctic mass loss by 2100, with a probability of at least 0.89.
Under very high-emissions scenarios, cascading processes in the ice sheet could produce up to 25.4 centimetres of sea-level rise by 2100 at the 95th percentile. The median estimate in that scenario is 15.7 centimetres, and the projections remain consistent with satellite observations.
Why the projections matter
Antarctica is the largest source of uncertainty in projections of future sea-level rise. The findings indicate that some Antarctic ice loss may already be committed, but that emissions levels still affect how much additional loss occurs by 2100.
That difference matters for coastal communities because greater Antarctic mass loss would directly increase near-term sea-level risks. The researchers say managing those risks will require rapid emissions reductions as well as better constraints on choices used in climate and ice-sheet models.
Evidence and remaining uncertainty
This is a model-based study that uses a machine-learning emulation framework to trace how individual physical assumptions affect projections. The researchers used Bayesian calibration—a statistical method for updating estimates with observations—against satellite data to reduce projection bias.
The results still depend on emissions scenarios, climate-model choices, ice-sheet dynamics and sea-level physics. The wide gap between the median and 95th-percentile estimate under very high emissions reflects remaining uncertainty. The study specifically identifies sliding laws, ice-shelf melt parameters and climate-model selection as areas needing improved constraints.
// Source
Nature Geoscience · 2026 · DOI: 10.1038/s41561-026-02102-1
Authors: Yucheng Lin, Xuebin Zhang, Nicholas R. Golledge, Robert E. Kopp, John Alexander Church, Yi Jin, Chen Zhao, Chris R. Stokes
Institutions: UNSW Sydney, City University of Hong Kong, Rutgers, The State University of New Jersey, Durham University, University of Tasmania, Victoria University of Wellington, CSIRO Environment, Institute for Marine and Antarctic Studies, Australian Centre for Excellence in Antarctic Science, Climate Change Research Centre
