Seawater seeping under Thwaites Glacier doubles its 2100 sea-level forecast

Warm seawater is reaching kilometres beneath Antarctica’s Thwaites Glacier, and once a computer model accounts for it, the glacier’s projected contribution to sea-level rise this century roughly doubles. The new model reaches that figure without assuming any further warming of the ocean or the air. Unlike most of its predecessors, it also reproduces what satellites have actually watched the glacier do: retreat, thin and speed up, all at once.

That matters well beyond Antarctica. Thwaites, widely nicknamed the Doomsday Glacier, holds enough ice to lift global seas by roughly two-thirds of a metre if it all melted, and the stretch of coast where it sits already accounts for about a third of the ice Antarctica loses. Every extra millimetre raises the baseline from which high tides and storm surges reach the coast.

The study, led by glaciologist Mattia Poinelli with Eric Rignot of the University of California, Irvine, and colleagues at NASA’s Jet Propulsion Laboratory, Dartmouth, UCLA and UC San Diego, projects up to 30 millimetres of global sea-level rise from Thwaites alone by 2100, with an uncertainty of about 6 millimetres either way. The average of the international model ensemble behind recent projections, run under the same conditions, gives about 15.

“Most projections to date have struggled to match the rapidity of the mass loss because of missing physics,” said Rignot, a professor of Earth system science at UC Irvine.

The model also points to a specific weak spot. The glacier’s fastest-moving section, its main trunk, is currently held back by the Mouginot Ridge, a bump in the seafloor a few kilometres across that sits about 950 metres below sea level. “The Thwaites main trunk is being temporarily stabilized by the Mouginot Ridge,” Poinelli said. In the simulations that best match what satellites see today, there is about a 60% chance the glacier slips off that ridge before the end of the century. After that, the seafloor deepens inland, and the ice has nothing comparable to grip until the next high point, the South Ridge.

How they did it

A glacier that ends in the ocean has a grounding line, where ice resting on bedrock lifts off and floats. Most ice-sheet models draw it as a sharp boundary, and in many common formulas ocean melting fades to zero right at it.

Satellite radar tells a messier story. Images from the ICEYE radar satellite constellation show the grounding line of Thwaites shifting by kilometres with the tides. On the main trunk, the ice repeatedly lifts off the bed up to 10 kilometres inland of the Mouginot Ridge, so pressurised seawater flushes in and out beneath ice that models treat as firmly grounded. The team calls this band a grounding zone.

The researchers built that zone into the Ice-sheet and Sea-level System Model, known as ISSM, letting ocean melting reach a few kilometres beneath the grounded ice. They started the glacier in the state satellites recorded in the mid-1990s, ran it forward to 2023 and checked it against three independent records: ice speed, thinning and grounding-line position.

Without the grounding zone, the model could only keep pace with the early years by assuming melting of 150 metres of ice per year beneath the floating shelf, and it still fell short on the ice discharged, the thinning and the retreat of later years. With a zone about 3 kilometres wide, a melt rate of 120 metres per year reproduced the retreat of both the main trunk and the eastern ice shelf. Rates of 100 to 120 metres per year, in line with satellite estimates, matched the 32 metres of thinning measured across the grounded ice over 25 years.

A statistical calibration then tested a range of zone widths and melt rates. The best fits clustered at zones 2 to 5 kilometres wide and melt rates of 90 to 120 metres per year. Simulations with no grounding zone were four times less likely to reproduce the observed history than those with a zone a few kilometres wide.

How the new number compares

The yardstick is ISMIP6, the international intercomparison that pooled many ice-sheet models for recent climate assessments. Under the same steady climate, its average for Thwaites is 15.2 millimetres by 2100, but with a spread of plus or minus 21.8 millimetres, wide enough to include runs where the glacier slows down late in the century.

The new model narrows that range by tying itself to the observed past, and its answer depends on the zone’s width. With no zone, the projection drops to 8.3 millimetres, below the ensemble average and out of step with the historical record. A 1-kilometre zone gives 14.3 millimetres, close to the ensemble mean. A 3-kilometre zone, the statistically best fit, gives 19.4 millimetres. A 5-kilometre zone, the width radar sees today, gives 30.1 millimetres, twice the ensemble mean.

Simply adding a grounding zone raises the projected contribution by 51%, and each extra kilometre of width changes it by about 22% on average. Warming the ocean by an extra 0.5 to 0.7 degrees Celsius, by comparison, adds only about 15% more ice loss in the model.

For scale: since 1996, the main trunk has retreated about 25 kilometres along a deep seafloor channel, roughly 800 metres a year, while the eastern ice shelf has pulled back 10 to 20 kilometres. Thirty millimetres would be close to a twentieth of everything Thwaites holds, released by one glacier in less than a century.

What the study does not settle

The headline figure is a lower-bound scenario, not a forecast of the worst case. It holds the ocean and climate at their 1996 to 2023 average, so it leaves out further warming and a possible rise of the warm water layer, both of which would increase melting. The 60% chance of unpinning applies only to simulations with a zone 4 to 6 kilometres wide. With a 2 to 3 kilometre zone, the chance falls to about 45%, and with no zone or a narrow one, the glacier stays on the ridge until 2100.

The model also simplifies several things on purpose. It holds the width of the grounding zone fixed, although in reality it varies with tides, seafloor slope and water beneath the ice. It keeps the ice front in place, so it ignores calving, the breaking-off of icebergs, which has likely contributed to past retreat. It uses a simple formula for ocean melting instead of simulating how water circulates in the cavity under the ice shelf, and it keeps the friction under the glacier constant over time, which the authors say may underestimate thinning and speed-up.

To match the 1996 grounding line, the team also raised parts of the mapped seafloor by up to 100 metres, within the error of the maps. And the comparison sets one calibrated model against a multi-model ensemble whose spread partly reflects differences between models; other teams will need to reproduce the result.

What happens next

The authors argue that actively melting grounding zones should be built into regional and continental ice-sheet models, including the international intercomparison projects. They also call for more measurements of temperature, salinity and currents beneath the Thwaites ice shelf.

“These results predict greater risk of and exposure to coastal flooding than anticipated in the coming decades,” Rignot said.

Common questions about Thwaites Glacier

What is the grounding zone of a glacier?

It is the band where a sea-ending glacier passes from resting on bedrock to floating. At Thwaites, radar shows it is several kilometres wide and shifts with the tides, letting warm seawater melt ice that looks grounded. Most models treat it as a single line.

How much will Thwaites Glacier raise sea level by 2100?

The new model projects up to 30 millimetres, plus or minus about 6, by 2100, with the climate held at its recent average. That is about twice the average of the ISMIP6 model ensemble under the same conditions, which is 15.2 millimetres. The whole glacier holds about 65 centimetres of sea-level rise.

Is Thwaites Glacier going to collapse?

The study does not predict a collapse by 2100. It finds a 60% chance, in the simulations that best match today’s radar data, that the main trunk detaches from the Mouginot Ridge before 2100, as early as 2038 in the widest-zone runs. Every simulation adds to sea level throughout the century with no sign of slowing.

Why is Thwaites called the Doomsday Glacier?

Thwaites holds about 65 centimetres of potential sea-level rise, sits in the sector behind about a third of Antarctica’s ice loss and rests on a seafloor that deepens inland, which makes retreat hard to stop once it starts.

The study was published online on 12 October 2026 in Nature Communications and is open access. The authors name the next modelling step: linking the zone’s width to the glacier’s changing shape and the water beneath it, so models can predict it instead of prescribing it.

Reference: Poinelli et al., “Sea-level contribution from Thwaites Glacier doubles in an observation-constrained ice-sheet model,” Nature Communications, 2026. DOI: 10.1038/s41467-026-77976-1

Tags: climate, Antarctica, Eric Rignot, glaciers, NASA JPL, sea-level rise