Antarctic Ice Loss Stays Locked In Even at 1.5 Degrees

Antarctica is very likely to lose more ice than it gains this century, even if warming is held to 1.5 degrees Celsius, a new ice-sheet study finds. The paper, published 30 September 2026 in Nature Geoscience, puts that outcome at a probability of at least 0.92 under aggressive emissions cuts.

The second finding is the one older ensembles could not lock down. Higher emissions still drive greater Antarctic mass loss by 2100, at a probability of at least 0.89, so the remaining fight is how high, not whether the continent adds water to the ocean.

The Sign of This Century Is No Longer a Coin-Flip

Lead author Yucheng Lin, an assistant professor at the City University of Hong Kong, and co-author Robert Kopp, a Distinguished Professor of Earth and Planetary Sciences at Rutgers University, wrote the public summary with colleagues including Xuebin Zhang, Nicholas R. Golledge, John A. Church, Yi Jin, Chen Zhao and Chris R. Stokes. Their study of committed Antarctic ice sheet mass loss trains a machine-learning copy of ice-sheet models, then keeps only the futures that match gravity-satellite records since 2002.

That filter is the move that ends the old argument. Unfiltered model sets still contained runs in which extra snowfall outweighed ocean-driven melt and the ice sheet grew. After the gravity check, the surviving runs point one way: net loss through 2100.

THE STUDY’S PROBABILITIES

• Net loss, aggressive cuts: Probability of at least 0.92 that the Antarctic Ice Sheet loses mass through the twenty-first century even under strong emissions reductions.
• Emissions still count: Probability of at least 0.89 that higher greenhouse-gas paths produce more Antarctic mass loss by 2100.
• Very high emissions, centre: Median Antarctic contribution of 15.7 centimetres of sea-level rise by 2100, in runs still consistent with satellites.
• Very high emissions, tail: 95th percentile of 25.4 centimetres by 2100 from cascading ice-shelf and glacier mechanisms.

A 2021 land-ice projection that fed the last IPCC assessment had found no clear emissions response for Antarctica, because extra snow competed with extra melt. Lin’s team says that competition does not survive contact with the satellite record. The sign of this century is committed. The height is not.

Gravity Satellites Threw Out the Gain Futures

Each full ice-sheet simulation can occupy a supercomputer for days, so no group can test every mix of climate path, ocean model and ice physics. The team’s shortcut was to learn those relationships from an archive of Antarctic ice simulations built by the Ice Sheet Model Intercomparison Project, then replay millions of combinations in a fraction of a second.

They limited the copy to 2100. The ISMIP6-2300 archive is too small, they write, to capture the strong nonlinearities that appear from the twenty-second century onward. Cross-checks against the original runs held up well enough to support probabilistic statements, with a Bayesian step that down-weights simulations which miss the observed mass-loss trend.

In raw simulations, ice-model choices dominate twenty-first-century uncertainty, averaging 64 percent of the variance. Sliding laws, which set how easily ice moves over bedrock, account for 21 percent of the variance from 2015 to 2100. Ice-shelf melt formulas, which turn a warming ocean into melt at the underside of floating ice, account for 20 percent. A Budd-type sliding law yields 5.0 centimetres, plus or minus 4.2, more sea-level rise in 2100 than a linear Weertman law. An observation-calibrated linear melt formula yields 7.8 centimetres, plus or minus 5.2, more than the ISMIP6 standard scheme.

ISMIP6 itself, under a very high-emissions path, had spanned from more than 7 centimetres of sea-level fall to more than 40 centimetres of rise. Lin and colleagues argue that a one-model, one-vote average can favour popular modelling choices whether or not those choices match what the ice has actually done.

The Snowy Pause Already Happened

Warmer air holds more moisture, so a hotter Antarctic atmosphere should drop more snow, especially over East Antarctica. That is the physical hope that let some projections show net gain this century. The hope has already had a field test.

NASA’s GRACE mission, which began in 2002, and GRACE Follow-On, flying since 2018, weigh the ice sheet by sensing tiny changes in Earth’s gravity. The agency’s mass maps show Antarctica shed about 135 gigatons of ice a year from 2002 to 2025, adding about 0.4 millimetres a year to global sea level. East Antarctica gained some mass from snow. West Antarctica, above all Pine Island and Thwaites, lost more.

A longer satellite reconciliation from the Ice Sheet Mass Balance Inter-comparison Exercise, published 16 September 2026, found Antarctic loss slowed to 59 gigatons a year, plus or minus 73, from 2020 to 2023. That error bar is larger than the central figure, so the four-year rate cannot be told apart from zero. Record snowfall in East Antarctica, and a near-balance on the Peninsula, did the work. Separate gravity analyses have described a multi-year mass gain after 2020 and tied it to atmospheric rivers that are not expected to hold for a century.

Those snowy years sit inside the same gravity record Lin’s filter uses. They did not reverse the 2002-to-2025 mean, and they did not keep net-gain futures in the set that matches satellites. Extra snow is already in the ledger. It is not a rescue.

Why 1.5 Degrees Still Leaves a Net Loss

The 1.5 degree Celsius target, 2.7 degrees Fahrenheit, is the most ambitious goal in the 2015 Paris climate agreement. Lin and Kopp treat it as the aggressive-cut case. Even then, the observation-filtered runs still lose ice this century.

Historical warming is the reason. The ice sheet is already out of balance with the ocean around it. Ice shelves, the floating extensions that hold back grounded glaciers, thin when that ocean warms. When they thin or break, the glaciers behind them speed up. Snowfall can pile up inland and still lose the contest if the outlets run faster than the snow arrives.

Emissions still change the total. The probability of at least 0.89 that higher paths mean more loss by 2100 is the study’s reply to anyone who reads “committed” as “nothing left to do.” Kopp put the same point in plainer words: coastal communities have to prepare for rising seas, and the amount they must manage still depends on choices made now.

What we still control is how much, and whether we set off the chain reaction that takes us to the high end.

Yucheng Lin, assistant professor, City University of Hong Kong

Cutting emissions now, Kopp said, can limit the risks facing later generations. The paper’s own language is that rapid cuts, plus better constraints on climate-model choice, sliding laws and ice-shelf melt formulas, are what manage the tail.

A 25-Centimeter Tail Under Very High Emissions

Under very high emissions, the filtered ensemble’s centre is 15.7 centimetres from Antarctica by 2100. The 95th percentile, still consistent with satellites, reaches 25.4 centimetres, which is 10 inches. Lin and Kopp describe that tail as a cascade: a fast-warming ocean thins the shelves, weakened shelves break, and glacier ice flows into the ocean faster, each step making the next easier.

ANTARCTIC SEA LEVEL BY 2100

The unfiltered spread is why earlier summaries treated Antarctica as a coin-flip. The filtered median is why “locked in” is not the same statement as “10 inches.” The 25.4 centimetre figure is the high end the satellites cannot yet rule out if the cascade runs. Lin and Kopp wrote that 25 centimetres would be enough to permanently flood the homes of more than 10 million people.

Antarctica supplied about 10 percent of global mean sea-level rise from 2006 to 2018, a modest share that does not match its leverage. The ice sheet is Earth’s largest frozen reservoir. Explainers of how ice sheets store frozen water sit beside the paper’s own figure: 57.9 metres of global sea level if the whole sheet went, about 190 feet.

Coastal Cities Must Plan for a One-Way Rise

Lin and Kopp note that about 1 billion people live in coastal areas, and about 100 million live within 1 metre, 3.3 feet, of sea level. Those are the households already exposed to flooding before Antarctica’s twenty-first-century add-on. Greenland, mountain glaciers and the expansion of warming seawater are already lifting the baseline. Antarctica now joins them as a one-way term this century, not a maybe.

Kopp said communities are deciding now on buildings and infrastructure that will last for decades, while scientists still argue about how quickly Antarctica will change. A clearer picture of the risks, he said, can help them prepare, and better projections can help them decide how to protect homes, roads and other essential works. The study’s offer is a faster way to test assumption mixes that are too costly to run one by one, which is the basis those decisions have been missing.

THE HIGH-EMISSIONS CASCADE

• Ocean first: A fast-warming ocean thins the floating ice shelves from below.
• Shelves next: Weakened shelves fragment or collapse, taking away the brake on the glaciers behind them.
• Glaciers last: Grounded ice flows into the ocean faster, and each step makes the next one easier.

Lin said the shelves act somewhat like brakes, and warmer water is how those brakes fail. The team cannot afford to study every detail of an ice-sheet model, he said, so it has to know which details matter. For coastal planning, the detail that matters is that the maybe on the sign is gone, while the high-end cascade is still a function of how hot the ocean and air around Antarctica get.

Cracks and Hidden Rivers Remain Off the Books

The authors say the study is far from the final word. Processes missing from today’s simulations cannot be learned by the emulator, including how ice cracks and breaks, and how rivers and lakes beneath the ice sheet change over time. Any of those could change how quickly Antarctic ice reaches the sea, which is the clock towns actually need when they set seawall heights and retreat lines.

THE THREE CONTROLS ON THE TOTAL

• Ice-shelf melt: How floating shelves respond to a warming ocean, including the melt formula that turns offshore heat into basal melt.
• Basal sliding: How ice moves over bedrock, with Budd-type laws producing more sea-level rise in 2100 than linear Weertman laws in the team’s tests.
• Regional warming: How fast the ocean and air around Antarctica warm, which is why climate-model choice ranks with the ice physics.

Of more than 20 physical assumptions, those three do the most to set future sea-level rise, Lin and Kopp wrote. Sharpening them would tighten the projections more than almost any other investment in the modelling chain. Until then, the observation-filtered message is blunt enough for a planning document: the continent loses ice this century even at 1.5 degrees, extra snow does not reverse the sign, and every ton of greenhouse gas still changes the height, including the 25.4 centimetre tail.

The missing cracks and subglacial rivers could still move the timetable. They do not put net gain back on the table for 2100 in the runs that match the gravity record we already have.

Frequently Asked Questions

How Do Satellites Weigh the Antarctic Ice Sheet?

The GRACE and GRACE Follow-On missions fly as twin satellites and infer mass change from tiny shifts in the distance between them as Earth’s gravity field varies. They weigh grounded ice, not floating ice shelves, which NASA’s mass maps colour grey because the shelves already displace seawater. That is why a snow gain on East Antarctica can show up in the gravity record while a thinning shelf does not, and why the 2002-to-2025 loss NASA reports is a grounded-ice number.

Why Do Some Ice-Sheet Models Still Show Ice Gain?

Warmer air can increase snowfall over East Antarctica enough, in some model setups, to outweigh ocean-driven losses at the outlets, especially when ice-shelf melt is assumed to be weakly sensitive to ocean heat. The ISMIP6 very high-emissions set included those net-gain members, which is how the unfiltered span reached more than 7 centimetres of sea-level fall. Lin’s calibration drops runs that fail to match the observed mass-loss trend, so the gain members do not survive the cut.

What Share of Recent Sea-Level Rise Came From Antarctica?

The new paper states that Antarctic Ice Sheet mass loss accounted for about 10 percent of global mean sea-level rise over 2006 to 2018. That modest share is why Antarctica still dominates the uncertainty rather than the present total: the reservoir behind it is 57.9 metres of sea-level equivalent, so small changes in outlet physics swing the century’s projection more than the recent percentage suggests.

Why Does the Study Stop Its Forecast at 2100?

The emulator was trained on the ISMIP6-2300 archive but restricted to 2100 because that ensemble is too small to capture the strong nonlinearities that emerge from the twenty-second century onward. Longer committed loss, including possible West Antarctic collapse on multi-century paths, sits outside this paper’s probability statements and would need a larger long-range set before the same gravity-filter method could be applied with the same confidence.