A team of ocean modelers at Los Alamos National Laboratory and the Pacific Northwest National Laboratory has taken one of the most ambitious attempts yet to make a global ocean model think like a coastal one. In a study published in Geoscientific Model Development, Dylan Schlichting and colleagues pushed the Model for Prediction Across Scales–Ocean (MPAS–O), the ocean component of the Department of Energy’s Energy Exascale Earth System Model, to resolutions rarely attempted in a global framework. Over the Gulf of Mexico, the model’s unstructured mesh narrows to roughly 1.4 kilometers, while the same simulation stretches out to 100-kilometer cells in the Pacific and Indian Oceans. The target of this computational zoom was the Mississippi–Atchafalaya river plume, one of the most dynamic and socially important freshwater systems in North America, which feeds the seasonal hypoxic dead zone on the Texas–Louisiana shelf.
The scientific challenge is deceptively simple to state. Global climate models run at horizontal resolutions of tens of kilometers, far too coarse to resolve submesoscale processes, the swirling fronts and filaments a few hundred meters to a few kilometers wide that dominate the exchange of heat, salt, carbon, and nutrients between the coast and the open ocean. These features are characterized by Rossby and Richardson numbers of order one, meaning that rotation and stratification no longer dominate in the familiar geostrophic balance. Regional models have long captured such dynamics, but they operate on limited domains and require artificial boundary conditions. Variable-resolution unstructured meshes promise the best of both worlds: kilometer-scale fidelity where it matters, coarse resolution everywhere else, all within a single global simulation.
Schlichting’s team tested that promise against a demanding benchmark. Their study site, the Texas–Louisiana shelf, undergoes a dramatic and well-documented seasonal transformation. Each summer, weak upcoast winds near-resonant with the local inertial period let freshwater from the Mississippi and Atchafalaya rivers pool on the shelf, forming a surface-advected plume whose strong lateral salinity gradients spin up a field of submesoscale eddies 20 to 50 kilometers across, with a deformation radius of roughly 5 kilometers. In the non-summer months, downcoast winds squeeze the plume against the coast and storms suppress coherent eddies, leaving a mess of unorganized fronts and filaments. Any model claiming coastal credibility must reproduce this annual cycle.
As a reference, the researchers used a validated configuration of the Regional Ocean Modeling System (ROMS) at a mean resolution of 1.7 kilometers, a model with a long track record of reproducing the shelf’s eddy kinetic energy and salinity structure. They ran two MPAS–O experiments. The first, MPAS–O: JRA, stayed close to the settings used in fully coupled climate simulations, drawing river discharge from the Japanese Reanalysis dataset. The second, MPAS–O: USGS, introduced several innovations: pseudo point sources that insert streamflow from nine gauged rivers into specific grid cells, river temperatures estimated from air temperature rather than copied from the nearest ocean cell, and passive dye tracers that tag freshwater so its fate can be tracked across the entire Gulf.
The river improvements were necessary because the default global treatment spreads runoff over hundreds of kilometers of coastline, a scheme originally designed for the stability of the Amazon outflow. That approach pushed Mississippi water far beyond the shelf and severely underestimated Atchafalaya discharge, which is roughly 30 percent of the Mississippi’s flow via the Old River Control structure but missing entirely from the reanalysis product. The point-source alternative brought the simulated freshwater distribution dramatically closer to the ROMS reference and to observations, an important technical contribution in its own right for the growing community of kilometer-scale global modelers.
In the open Gulf of Mexico, the news was encouraging. The model generated vigorous submesoscale turbulence, visible as swirling fronts and filaments with Rossby numbers approaching one, and the statistics confirmed it: probability density functions of surface vorticity skewed toward cyclonic values and divergence skewed toward convergence, with broader tails in winter than summer. That seasonal modulation matches the established physics of mixed layer instabilities, which feed on the deeper winter mixed layer’s reservoir of available potential energy. By these open-ocean metrics, MPAS–O behaves like other state-of-the-art unstructured models such as FESOM and ICON, demonstrating that regional refinement can deliver submesoscale-permitting dynamics inside a global simulation.
On the shelf itself, however, the story turned. During winter, when the real eddy field is naturally suppressed, MPAS–O agreed qualitatively with ROMS, producing plausible small-scale fronts, filaments, and vortices. But in summer, the defining field of 20-to-50-kilometer baroclinic eddies simply failed to materialize. Cross-sections revealed the likely culprit: a persistent brackish lens in the top several meters of the water column, absent in ROMS, associated with markedly weaker vertical mixing. Within the boundary layer, vertical eddy viscosity in the MPAS–O configuration could fall more than an order of magnitude below the ROMS values, and the CVMix implementation of the K-Profile Parameterization used in MPAS–O lacks an explicit bottom boundary layer scheme that ROMS possesses.
The freshwater budget told the same story from a different angle. Tagging river water with passive dyes showed that MPAS–O: USGS held 42 percent more Mississippi freshwater and 58 percent more Atchafalaya freshwater on the shelf than ROMS on average, occasionally more than double in summer, with spurious late-summer accumulation peaks. Because eddies are the dominant mechanism for exporting freshwater across the shelf break, suppressed eddies mean trapped freshwater. The authors emphasize that the suppression likely arises from interacting parameterization and structural design choices, including the CVMix mixing scheme’s need for coastal tuning and the absence of horizontal momentum in river inflows, rather than from fundamental numerical deficiencies. Notably, even halving the vertical layer thickness to half a meter and minimizing lateral mixing did not restore the eddy field.
There are genuine silver linings. The global mesh lets researchers follow Mississippi water hundreds of kilometers into the open Gulf, past the Loop Current, onto the Florida shelf, and even into the Gulf Stream, something a limited-domain model cannot do without boundary artifacts. MPAS–O also captured the plume’s qualitative seasonal migration, its winter confinement along the coast and summer spreading over the shelf. The authors recommend that developers of other unstructured global models adopt an idealized model hierarchy of increasing complexity to systematically isolate such failures, and note that the Department of Energy is now evaluating two-way nesting of ROMS within MPAS–O as an alternative path to seamless coastal–global simulation.
The broader lesson is sobering for anyone hoping that brute-force resolution will solve coastal climate modeling. Grid spacing of 1.4 kilometers, achieved here on roughly 5,000 computing cores over multi-year runs, proved necessary but not sufficient: the mixing physics, river forcing infrastructure, and vertical coordinate design inherited from coarse global climate configurations all leave fingerprints on kilometer-scale coastal dynamics. The authors suggest MPAS–O may perform better in coastal systems where rivers matter less, such as the Mid-Atlantic Bight, and that validation against real interannual observations is the essential next step. Until then, the summer eddies of the Texas–Louisiana shelf remain a benchmark that global models have not yet met, a reminder that the boundary between the coastal ocean and the global ocean is as much a modeling frontier as a physical one.
Subject of Research: High-resolution realistic simulations of coastal submesoscale baroclinic instabilities and river plume dynamics with the MPAS–O global ocean model
Article Title: Towards high-fidelity simulations of coastal submesoscale baroclinic instabilities with MPAS–O (vE3SM3.0.0) – Part 2: Realistic experiments
Article References: Schlichting, D. R., Smith, K. M., Petersen, M. R., Hetland, R. D., Hinson, K. E., & Engwirda, D. (2026). Towards high-fidelity simulations of coastal submesoscale baroclinic instabilities with MPAS–O (vE3SM3.0.0) – Part 2: Realistic experiments. Geoscientific Model Development, 19(19), 9627-9650. https://doi.org/10.5194/gmd-19-9627-2026
Image Credits: AI Generated
Keywords: MPAS-O, submesoscale processes, Mississippi-Atchafalaya plume, Texas-Louisiana shelf, baroclinic instability, ocean modeling, river plumes, variable-resolution mesh, ROMS, Gulf of Mexico, K-Profile Parameterization, E3SM
News Source: Violet Maxwell. (October 12, 2026). Kilometer-Scale Global Ocean Model Captures River Plume Seasons but Misses Coastal Eddies. Scienmag.
