Thwaites Glacier:
the boundary
is a zone
A new Thwaites model tests tidal seawater intrusion against decades of observations. Its higher sea-level estimate depends on stated assumptions.
The important change is beneath the ice
A new study of Antarctica's Thwaites Glacier changes how a computer model represents the place where ice leaves the seabed and begins to float. Instead of treating that transition as a sharp line, the researchers allow relatively warm seawater to reach beneath grounded ice across a zone several kilometers wide. That change helps the model reconstruct observed retreat, thinning and acceleration, and it produces a larger contribution to global mean sea level in some simulations through 2100.
The paper, published October 12 in Nature Communications, was highlighted by NASA's Jet Propulsion Laboratory on October 20. Mattia Poinelli and colleagues used the Ice-sheet and Sea-level System Model, or ISSM, with researchers affiliated with JPL, the University of California, Irvine, Dartmouth and other institutions. Reading the paper alongside the announcement reveals why the physical change matters and why the resulting numbers should remain attached to their assumptions.
A grounding line can move across a zone
Glacial ice rests on land or the seabed until it becomes buoyant enough to float. The boundary between grounded and floating ice is commonly called the grounding line. In a model, representing it as a crisp dividing line is convenient: one side is attached to the bed, the other is floating. Satellite observations show that the real transition at Thwaites can be less tidy, especially as tides raise and lower the ocean beneath the ice.
The study describes seawater intrusions extending kilometers beneath ice near that transition. Water can periodically lift ice and deliver heat into a grounding zone rather than stopping at a fixed boundary. The researchers represent that zone explicitly, allowing ocean-driven melting inland of the model's calculated flotation line. The effect is not simply more melt spread uniformly across Antarctica. It is additional melting in a geometrically important region where the glacier is interacting with the seabed.
The past provides a demanding test
The researchers first reconstructed Thwaites from the mid-1990s into the 2020s. Their historical comparison used three kinds of evidence: how fast the ice moved, how its thickness changed, and where the grounding line migrated. These measurements are related, but they are not interchangeable. A model that matches the position of a boundary while getting ice discharge or thinning wrong has not reproduced the whole observed behavior.
The observations come from several satellite systems and methods. Radar interferometry supports estimates of ice motion and grounding-line position; elevation measurements from missions including ICESat, ICESat-2 and CryoSat help track thickness change. The paper also uses high-resolution ICEYE observations of the grounding region. Together, these records constrain a history that cannot be inferred from a single dramatic photograph of an ice cliff.
The historical work is a hindcast: researchers run a model through a period for which observations already exist, then compare the modeled changes with those measurements. It is a stronger test than selecting a visually plausible future animation. Still, matching the past does not make the future certain. The value is that it helps distinguish physical representations that are consistent with observed behavior from those that are less consistent under the tested setup.
What the fourfold comparison actually means
JPL's announcement describes the new model as matching historical satellite data four times more closely. The paper gives a more specific statistical formulation: simulations without a grounding zone were four times less likely to reproduce the historical trend than simulations with a zone a few kilometers wide. That is evidence about the tested models' fit to the historical record, not a guarantee that a future forecast is four times as accurate.
The team explored 105 simulations combining different melt rates and grounding-zone lengths. A Bayesian calibration then evaluated combinations against the observations, using a larger set of sampled parameter combinations and interpolated model outputs. The purpose was to identify combinations that could reproduce retreat, thinning and acceleration together, rather than choose one setting solely because it matched a single measurement.
The favored region was not one magic number. The paper describes central grounding-zone lengths of about 2 to 5 kilometers and melt rates of roughly 90 to 120 meters per year within the parameterization. Those melt rates describe ocean-driven basal melting in the specified part of the model. They are not a claim that the entire glacier's surface becomes 120 meters lower every year. Location, process and measurement unit all matter.
A ridge can delay retreat without stopping it forever
The 2038 date is conditional, not a countdown
The earliest unpinning date highlighted in the paper is 2038, appearing in simulations with grounding-zone lengths of 5 to 6 kilometers. That is one conditional result within an ensemble. It does not mean the authors observed an event scheduled for that year, and it does not describe the disappearance of all Thwaites ice. The modeled event is retreat beyond a particular stabilizing ridge.
Other parameter choices produced different timing. Simulations with two- to three-kilometer zones had a roughly 45 percent probability of crossing the ridge by the end of the century in the study's framework; the paper reports average crossing years of 2088 for a two-kilometer zone and 2065 for a three-kilometer zone. Wider zones consistent with recent satellite-inferred extents produced a higher modeled probability, around 60 percent, by 2100.
These probabilities belong to the experiment and its parameter distributions. They should not be presented as a complete probability of real-world glacier collapse across all possible climate futures and physical processes. Their practical meaning is comparative: the way the transition zone is represented changes whether and when the model releases the glacier from a stabilizing bed feature.
Three estimates show the sensitivity
For a selected melt rate of 110 meters per year, the paper presents a useful set of comparisons. Simulations omitting a grounding zone produce a sea-level contribution of 8.3 plus or minus 3.0 millimeters by 2100. A grounding-zone length centered on three kilometers produces 19.4 plus or minus 5.2 millimeters. A zone centered on five kilometers produces 30.1 plus or minus 5.7 millimeters. These are the paper's reported model values and uncertainty terms, not measured future water levels.
The comparison with earlier work also has a specific denominator. The paper cites an ISMIP6 multi-model ensemble mean of 15.2 millimeters, with a much wider reported spread of plus or minus 21.8 millimeters, under the same constant-forcing scenario. The roughly 30-millimeter result is about twice that mean. It is not twice all previously published estimates, and it is not outside every earlier uncertainty range.
A single calibrated modeling framework and a multi-model ensemble organize uncertainty differently. The new study's narrower numerical range does not erase all the structural uncertainty represented by other models. It shows how a particular physical representation, constrained by historical observations, shifts the result within the broader scientific problem. Comparing the absolute values and the scenario is more informative than repeating the word double without its reference point.
| Tested representation | Contribution by 2100 |
|---|---|
| No grounding zone | 8.3 ± 3.0 mm |
| Zone centered on 3 km | 19.4 ± 5.2 mm |
| Zone centered on 5 km | 30.1 ± 5.7 mm |
Thirty millimeters is not the glacier's entire ice inventory
Thirty millimeters is three centimeters, or about 1.2 inches. That conversion can help American readers picture the scale, but it does not turn a global mean contribution into an identical increase at every harbor or beach. The paper calculates sea-level equivalent from the volume of ice lost above flotation. Local relative sea level also depends on processes beyond the glacier model, so this result is not an address-level flood estimate.
The paper separately describes Thwaites as containing ice equivalent to about 65 centimeters of sea-level rise. That is an inventory-scale context figure, not its projection for 2100 in this experiment. Confusing the amount of ice associated with a glacier with the amount lost over a specified period produces a much more alarming claim than the researchers made.
Constant forcing isolates one part of the problem
The forward simulations use a baseline derived from average climate conditions over 1996–2023. The researchers hold that forcing constant while extending the model to 2100. This experimental choice helps isolate the importance of grounding-zone melting instead of combining it with a wide range of future warming assumptions. The authors describe the setup as a lower-bound forcing scenario, not as a prediction that the climate will stop changing.
That detail makes the result notable: changing the physical treatment of the grounding zone can raise the estimated contribution even without adding further climate warming to the baseline. But it does not establish a universal minimum for every possible real-world outcome. A simplified experiment remains subject to its geometry, parameters and omitted processes.
The paper also includes a sensitivity test for warmer water, linking additional thermal forcing to increased basal melting within its chosen parameterization. The authors found the tested projections more sensitive to grounding-zone extent than to moderate thermal-forcing changes. That is a statement about the range and model they tested. It should not be rewritten as a general claim that ocean warming is unimportant to Antarctic ice loss.
The seabed is measured imperfectly too
An ice-sheet model needs a description of the bed beneath the ice, but that terrain is difficult to observe. The study uses BedMachine data and discusses uncertainties in areas reconstructed from gravity information. To make the initial floating and grounded ice geometry consistent with observed 1996 grounding-line positions, the researchers applied a localized seabed elevation adjustment of up to 100 meters in specified regions under floating ice, within the stated uncertainty of that reconstruction.
This is not a trivial detail to hide behind a smooth animation. The position of the bed affects where ice can remain grounded, how water can reach it and how retreat interacts with ridges. The authors explicitly identify bathymetric uncertainty as a limit on reconstruction. Their result depends not only on adding a new process but also on creating an initial geometry compatible with the observations they are trying to reproduce.
The model uses an unstructured mesh with 89,824 elements. Resolution is about 500 meters near the grounding region and coarser in slow-moving areas. That uneven detail is a computational choice: concentrate finer spatial representation where the process under study operates. It does not mean every part of the glacier has been directly surveyed at that resolution, or that a numerical grid removes uncertainty in the underlying measurements.
What this experiment leaves out
Those limits do not invalidate the test. They define what it tests. A focused experiment can show that a previously simplified process matters while leaving other processes for further work. The strongest conclusion is not that one model has solved Thwaites. It is that actively melting grounding zones deserve systematic representation in future regional and continent-scale ice-sheet modeling, alongside improved observations.
Photographs, satellite maps and model figures play different roles
The archival photographs in this article show the glacier and the aircraft-based work associated with Antarctic observation. They are not photographs of the 2026 simulation or proof that a specific forecast is already happening. The dated Landsat images provide a different kind of evidence: observations of the ice and its surroundings at identified times. A research figure then combines observations or modeled quantities according to its caption.
That separation matters because an ice-front image can be visually dramatic while saying little about a hidden grounding zone. Radar and elevation records can detect changes not obvious in an ordinary photograph. A model links those measurements with physical assumptions to explore how the system might evolve. Each tool contributes something the others cannot supply on its own.
For coastal readers, the useful result is better evidence
For U.S. coastal communities, a study of a distant glacier contributes to a larger evidence base used to understand future sea-level risk. It is not a replacement for regional projections, local tide-gauge information, elevation data or planning standards. A global mean contribution from one glacier cannot determine the timing or depth of flooding on a particular street.
What the study offers is a more explicit link between observed behavior and one process that changes the glacier's simulated future. The historical record helps evaluate parameter choices; the ridge geometry explains why timing can change; the forward experiment shows the consequences under a stated baseline. Those are useful improvements even when the resulting answer is a range rather than a single certain date.
Comments
Post a Comment