The same Antarctic forcing can leave an ice shelf in two different worlds

One coastal forcing can leave an Antarctic ice shelf with two different oceans beneath it: a cold circulation that limits melting, or a warm circulation that drives more. In the model studied here, both states can persist under the same rate of salt added to the ocean as sea ice forms. That is the striking implication of hysteresis: once the system shifts into a high-melt state, simply restoring the earlier forcing may not be enough to bring it back.

The distinction matters because an ice shelf is not just a slab of floating ice. It is a brace for the grounded ice behind it. Ocean-driven thinning can weaken that brace and allow more land-based ice to flow into the sea, contributing to sea-level rise. The study does not forecast when a particular Antarctic shelf will cross a threshold. Instead, it isolates a mechanism that could make some transitions abrupt and difficult to reverse.

The Science

An ice shelf floats where a glacier or ice sheet reaches the ocean. Its underside melts when heat from seawater reaches the ice, and that basal melt can vary enormously from one Antarctic region to another. Shelves associated with cold Dense Shelf Water can experience low melt, while incursions of warmer Circumpolar Deep Water—water originating in the Antarctic Circumpolar Current—can promote much higher melt (Saddier et al., 2026).

The paper focuses on one process that helps determine which water reaches the cavity: sea-ice production in coastal polynyas. A polynya is an area of open water amid sea ice, often kept open by winds that push newly formed ice away from the coast. As water freezes, salt is mostly left behind. This brine rejection makes the remaining surface water denser. If the density increase is strong enough, the surface water sinks and stirs the water column in deep convection.

The central question is whether changing this coastal salt forcing produces a smooth, reversible response—or whether the cavity can jump between distinct circulation regimes. Realistic Antarctic models represent complex coastlines, ice geometry and changing weather, but that realism makes it costly to step through forcing levels carefully and determine whether a transition reverses. Saddier and colleagues use a deliberately simplified three-dimensional ocean model to make that controlled comparison.

They ran NEMO, a widely used ocean-modeling system, in a canonical cavity inspired by the ISOMIP+ model framework. The domain extends 250 kilometres in one horizontal direction and 50 kilometres in the other, with a flat seabed 600 metres deep and an idealized ice shelf 100 kilometres long. The ice base slopes in a simple, prescribed way. The ocean grid is 2 kilometres across and 20 metres vertically; the ice-ocean boundary calculates melt interactively, rather than imposing a fixed melt rate.

At the open-ocean end, the model restores seawater toward prescribed temperature and salinity profiles. Near the ice front, it applies a surface salt flux to represent brine from sea-ice formation in a polynya. The researchers vary this salt forcing quasi-statically—raising or lowering it in steps and allowing the model to settle—then compare the steady circulation and melt response. They also repeat experiments with different offshore temperatures and salinities, and with alternative lateral-boundary and horizontal-viscosity settings.

The standard temperature profile used to illustrate the two regimes runs from
−1.9°C at the surface to 11°C near the seabed. Its salinity structure makes the offshore water column stably stratified: denser water lies below lighter water. In one initial experiment, no polynya salt flux produces a warm regime; a much larger flux, 0.075 in the model’s prescribed units, produces a cold regime. For context, the authors note that reported extreme annual sea-ice production rates correspond to salt forcing as high as about 0.03 in Cape Darnley, among the most active production regions. The 0.075 case is therefore an intentionally strong contrast, not a direct estimate of typical conditions.

The first comparison makes the physical difference visible. In the warm case, meltwater helps sustain an overturning circulation that draws relatively warm deep water toward the grounding line. In the cold case, dense water formed under the polynya sinks and helps send cold water into the cavity; convection is prominent beneath the polynya. The cold regime also includes a circulation feature near the ice front that can locally enhance melting even while melt across much of the cavity remains low.

What They Found

The headline result is hysteresis. During a gradual change in polynya forcing, the warm-to-cold transition occurs at a different point from the cold-to-warm transition. Between those transition points, the model can maintain either state, depending on the system’s history. This is not simply a delay while the ocean catches up: the researchers identify two distinct steady responses under the same external forcing.

That distinction changes how to think about reversibility. If a cold cavity shifts into a warm, high-melt regime, a later return to the original salt forcing need not automatically restore the cold circulation. The ocean can remain on the warm branch of the response. Hysteresis is familiar in other climate systems, where feedbacks make the route into a state different from the route out. Here it emerges from the interaction between brine-driven convection, cavity circulation and meltwater.

The thermocline—the layer where temperature changes most sharply with depth—provides a useful marker of these changes. The researchers track how its depth near the polynya evolves during the forcing experiments. Its position differs between regimes, making it a candidate diagnostic for the circulation state, even though the study’s idealized model does not establish a direct observational early-warning threshold for a real shelf.

The mechanism behind bistability is a gyre near the ice front. A gyre is a rotating circulation pattern. In the cold regime, brine-driven density differences help create an anticyclonic gyre that straddles the front. Under the right conditions, that circulation recirculates meltwater within the front area instead of carrying it away. The retained meltwater affects local density and circulation, helping reinforce the state that produced it. The authors’ central finding is that bistability appears when this front gyre traps and recycles meltwater.

The feedback does not mean that every gyre or every polynya causes an abrupt transition. Rather, it identifies a route by which a local change in sea-ice production can become self-maintaining. The initial forcing changes density and circulation; the circulation alters where meltwater goes; and the changed meltwater distribution helps shape the circulation in turn. That feedback can give the system two stable outcomes instead of one.

The width of the bistable range shifts with offshore conditions. Warmer far-field ocean temperature makes that range larger, according to the study. Stronger background stratification makes it smaller. Stratification is the layering of the ocean by density; strong layering resists vertical mixing, while weaker stratification makes it easier for surface cooling and brine rejection to affect deeper water. These sensitivities mean that the same polynya forcing cannot be interpreted in isolation from the offshore water mass it acts upon.

The model also shows that melt rate does not respond identically in the two regimes. The authors propose separate scaling laws: a quadratic fit for warm-mode melt as a function of thermal driving, and a linear fit for cold-mode melt. Thermal driving is the difference between the ocean temperature and the local freezing temperature at the ice base—the heat available to melt ice. The paper’s supplied results describe these functional forms, but do not provide the fitted coefficients or transition-threshold values in the material available here; it would be misleading to infer numerical melt rates from the plotted curves.

Offshore temperatures tested in the experiments

Label	Value
TEMP−1.9	-1.9 °C
TEMP−1	-1 °C
TEMP0	0 °C
TEMP0.5	0.5 °C
TEMP1	1 °C
TEMP1.5	1.5 °C
TEMP2	2 °C

Values are imposed surface temperatures in the experiment table; they do not represent observed Antarctic trends.

That caution is important when interpreting the experimental parameter range. The temperature bars show imposed far-field surface temperatures used in sensitivity runs, not observed warming or melt responses. The authors span conditions from
−1.9°C to 2°C, allowing them to test how the model’s response changes as the surrounding ocean properties change. The point is not that Antarctica’s offshore water will follow this exact sequence, but that background conditions can shift the threshold behavior.

Why This Changes Things

Most projections ask how much an ice shelf will melt under a future scenario. This paper asks a different question: if the ocean has already moved into a high-melt state, can it return when the original forcing returns? That is a consequential distinction for risk. A response that tracks forcing reversibly can ease as conditions ease; a hysteretic response may retain some of its change, with implications for how quickly ice-shelf thinning could recover.

The study also helps explain why regional modeling papers can reach different conclusions about reversibility. Horizontal viscosity is a numerical representation of how unresolved small-scale motion spreads momentum. In the experiments, increasing horizontal viscosity limits bistability. In effect, changes to the model’s representation of local circulation can weaken or remove the feedback that permits two states. The authors suggest this may help explain why earlier simulations have reported differing behaviors.

Horizontal viscosity settings tested

Label	Value
NUH1.3	1.3 ×10⁷ m⁴ s⁻¹
Standard	6.4 ×10⁷ m⁴ s⁻¹
NUH32	32 ×10⁷ m⁴ s⁻¹

These are tested model settings, not measured ocean viscosities.

This is not a reason to dismiss either realistic simulations or idealized ones. The simplified domain strips away many features that matter in Antarctica: irregular seabeds, complex ice drafts, tides, changing winds, seasonal sea-ice cover and spatially variable ocean properties. It also uses parameterizations for vertical convection and small-scale mixing. Those choices make the mechanism easier to isolate, but they limit how directly one can map its thresholds onto a named ice shelf.

The model’s most useful contribution is therefore mechanistic. It shows that a polynya is not merely a source of cold dense water. Its effect depends on whether the resulting circulation brings water into the cavity, how a gyre organizes flow around the front, and whether meltwater is exported or recirculated. This makes local ocean dynamics and offshore stratification part of the explanation for why neighboring shelves—or different models of the same shelf—might respond differently.

The wider stakes are substantial but indirect. Ice shelves buttress grounded Antarctic ice, and their thinning can make it easier for that ice to flow seaward. The paper does not translate its idealized melt rates into sea-level projections, nor does it demonstrate that a real shelf is currently approaching a tipping point. It instead identifies a feedback that future regional studies can test against observations and more realistic simulations.

What’s Next

The next step is to ask whether the front-gyre feedback survives in realistic, high-resolution configurations. A useful test would resolve the coastline, ice draft and bathymetry of a particular shelf while systematically changing polynya forcing in both directions. Such experiments could establish whether the two branches of the response occur under plausible seasonal and long-term conditions, and whether they persist when winds, tides and sea-ice production vary over time.

Observations could help by tracking more than melt alone. The study points to thermocline depth as an indicator of regime changes. Measurements of temperature and salinity across the ice front, combined with current observations and estimates of polynya sea-ice production, could reveal whether a front gyre is retaining meltwater or allowing it to escape. No single indicator would prove that a transition is underway, but together they could constrain the circulation processes models must get right.

There are also clear model caveats. The geometry is idealized; the forcing is prescribed; and the ocean model’s viscosity and convection treatments influence the behavior being studied. The paper’s scaling laws and transition logic are therefore tools for framing hypotheses, not ready-made thresholds for Antarctic monitoring. The authors’ own sensitivity results make that point especially strongly: a parameter intended to represent unresolved dynamics can alter whether the transition is reversible.

That is a valuable warning for climate projections. A model that captures the average melt rate but misses the circulation feedback could also miss whether a change persists after forcing shifts. By identifying a specific candidate mechanism—the gyre that recirculates meltwater at the front—Saddier and colleagues give future work something concrete to test. The ultimate question is not only how warm the water becomes, but whether the ocean circulation can find its way back.