Beneath the floating fringes of Antarctica, a hidden battle between warm ocean water and glacial ice is quietly shaping the future of global sea levels. The rate at which seawater melts the undersides of ice shelves, the vast floating extensions of glaciers that buttress the flow of grounded ice toward the sea, has become one of the largest sources of uncertainty in projections of Antarctic mass loss. Now, a team of Dutch researchers has unveiled a substantially upgraded tool designed to close that gap: version 2.0 of the one-Layer Antarctic model for Dynamical Downscaling of Ice–ocean Exchanges, known as LADDIE. In a paper published in the journal Geoscientific Model Development, Erwin Lambert of the Royal Netherlands Meteorological Institute and colleagues at Utrecht University describe a model that captures the fine details of sub-shelf melting at a computational cost low enough to run across the entire continent.
The problem LADDIE addresses is one of scale. The most realistic simulations of ice–ocean interaction come from fully coupled, three-dimensional ocean models, but these demand enormous computing resources. Studies cited by the team indicate that ocean models need a resolution of roughly two kilometres or finer to accurately represent melting near the grounding line, the critical zone where ice first begins to float, and even higher resolution to capture the narrow basal channels that channelise melting on many ice shelves. Continental-scale coupled simulations today typically run at resolutions of four to fourteen kilometres below ice shelves, and only regional configurations achieve the two-kilometre threshold. Running such models over the multiple centuries relevant to ice sheet dynamics remains unfeasible, which is why most Antarctic sea-level projections still rely on stand-alone ice sheet models with simplified melt parameterisations.
Those parameterisations, which translate offshore ocean temperatures into fields of melt rates beneath the ice, come in several flavours, including quadratic temperature dependencies, the box model PICO, and one-dimensional plume models. Yet detailed comparisons against three-dimensional ocean models and observations have shown that they struggle to reproduce the horizontal pattern of melting, and earlier work by the same group demonstrated that ice sheet dynamics are highly sensitive to precisely that spatial pattern. LADDIE was conceived as a middle way: a two-dimensional model of the upper ocean mixed layer beneath ice shelves that resolves the influence of topographic steering and Coriolis deflection on meltwater flow, without the cost of simulating the full three-dimensional cavity.
Version 2.0 represents a deep technical overhaul. The original LADDIE was written in interpreted Python on a square grid with finite difference numerics, which limited it to coarse pan-Antarctic resolutions. The new version has been translated into compiled Fortran and rebuilt on an unstructured mesh of triangles and Voronoi cells using finite volume numerics, drawing on the Utrecht Polar SYstem toolbox developed for the UFEMISM ice sheet model. Ocean velocities are solved on the triangles, while melt rates and other variables are resolved on the cells surrounding the triangle vertices, an arrangement equivalent to the Arakawa-B discretisation used in the three-dimensional ocean model FESOM. This design gives the velocity field an effective resolution roughly twice as fine as that of the melt field, sharpening the simulated melt plumes.
The time stepping scheme has also been replaced. Where version 1.0 used a modified LeapFrog scheme with a Robert–Asselin filter, the new version employs a third-order Forward–Backward Runge–Kutta scheme, a generalised form of the scheme used in NEMO 5. With carefully chosen forward–backward weights, this scheme maintains numerical stability with time steps approximately twice as large as standard Runge–Kutta approaches. Combined with parallelisation across up to 128 processor cores, these changes cut computation time by an order of magnitude compared with the previous version. In a benchmark simulation of the Amundsen Sea region at roughly one-kilometre resolution, the team found that doubling the core count reduced computing time by about a third up to 64 cores, with 16 cores arguably the optimal trade-off for that particular domain.
The evaluation against three-dimensional ocean models is striking. In the idealised ISOMIP+ Ocean1 experiment, which prescribes an 80-kilometre-wide ice shelf under warm Amundsen Sea-like forcing, LADDIE 2.0 was compared with an ensemble of twelve ocean models. Its melt pattern closely resembled the multi-model mean, with a mean difference of just 0.3 metres per year and a root mean squared difference of 5.5 metres per year, placing it among the models closest to the ensemble average. The model also reproduced the melt patterns of the four models that cluster tightly around the mean, namely MITgcm-BAS, MITgcm-JPL, MPAS-Ocean and NEMO-CNRS, suggesting it captures the essential physics that establish those patterns.
At continental scale, the team configured LADDIE at approximately two-kilometre resolution, a mesh of more than 450,000 vertices, and compared it with the RISE multi-model ensemble of nine pan-Antarctic ocean models. Bulk metrics such as melt sensitivities, which describe how strongly melting responds to ocean warming, fell close to the ensemble median and mean. A comparison against four satellite-derived estimates of sub-shelf melting showed good overall agreement in integrated melt rates per ice shelf, notably without any regional tuning. The largest discrepancies, along the Bellingshausen Sea and at the Borchgrevink ice shelf, appear to stem partly from a warm bias in the ISMIP6 ocean forcing dataset, and preliminary tests with updated forcing significantly reduce the mismatch. The model does underestimate refreezing in cold cavities such as the Filchner–Ronne and Ross ice shelves, likely because it underestimates barotropic currents and tides there.
Perhaps the most visually compelling result comes from the Pine Island ice shelf, one of the fastest-melting glaciers in West Antarctica. Run at a nominal resolution of 120 metres, with a mesh of over 430,000 vertices, LADDIE reproduced the fine-grained network of basal channels observed in high-resolution satellite studies, including the well-known Y-shaped channel and the branching of wide along-flow channels into narrower ones deflected by Coriolis forces and topographic steering. This capability matters because channelised melting is underrepresented in large-scale remote sensing products, and because satellite altimetry cannot measure melt rates close to the grounding line, where bridging stresses break the assumption of hydrostatic balance and where melting may matter most for ice sheet stability.
The consequences for sea-level projections could be substantial. In an idealised coupled simulation integrating LADDIE with the UFEMISM ice sheet model on the same adaptive mesh, the team compared the dynamic melt model against a conventional quadratic melt parameterisation under identical high-melt ocean forcing. Although initial melt rates were tuned to be nearly identical, the coupled simulation produced a threefold increase in grounding line retreat and volume above floatation loss. The reason lies in feedbacks: with the parameterisation, ice shelf thinning lifts the ice out of the warm water layer and suppresses melting, a negative feedback, whereas LADDIE’s simulated meltwater flow sustains strong melting along steep basal slopes and shear margins, a partial positive feedback that amplifies the loss of buttressing.
The authors are careful to stress that two-dimensional melt models cannot replace three-dimensional ocean models, and that LADDIE’s rapid equilibration and simplified physics carry known limitations, including the absence of seasonal intrusions of warm water into shallow cavity regions. But by making a physically detailed, open-source melt model fast enough for centennial-scale, continent-wide simulations, and by fully integrating it with an ice sheet model on a shared mesh, the team hopes to push future intercomparison efforts such as ISMIP7 toward more realistic representations of ocean-driven melting. As the ocean continues to gnaw at the underside of Antarctica’s floating ice, tools like LADDIE may prove essential for turning scattered observations into credible forecasts of how much the sea will rise.
Subject of Research: A two-dimensional ocean mixed-layer model for simulating ocean-driven melting beneath Antarctic ice shelves and its coupling to ice sheet dynamics
Article Title: The one-Layer Antarctic model for Dynamical Downscaling of Ice–ocean Exchanges (LADDIE) version 2.0
Article References: Lambert, E., Jesse, F., & Berends, C. J. (2026). The one-Layer Antarctic model for Dynamical Downscaling of Ice–ocean Exchanges (LADDIE) version 2.0. Geoscientific Model Development, 19(19), 9441-9461. https://doi.org/10.5194/gmd-19-9441-2026
Image Credits: AI Generated
Keywords: Antarctica, ice shelves, sub-shelf melting, sea-level rise, ocean modelling, LADDIE, ice sheet dynamics, grounding line, basal channels, UFEMISM, model intercomparison, cryosphere
News Source: Violet Maxwell. (October 8, 2026). Faster Antarctic model brings ocean-driven ice shelf melting into sharper focus. Scienmag.
