The researchers used the Coastal and Regional Ocean Community model, known as CROCO, a hydrostatic regional model built on the Regional Ocean Modeling System. They analyzed a set of Atlantic-wide simulations called GIGATL at horizontal resolutions of 1, 3 and 6 kilometers, with 100 vertical levels in the finer cases and 50 in the coarsest one. Some runs included tidal forcing derived from the TPXO7.2 and GOT99 tidal databases, while others omitted tides entirely, allowing the team to separate the effects of resolution and tides. All configurations were forced at the surface with hourly momentum and heat fluxes from the Climate Forecast System Reanalysis and initialized with the Simple Ocean Data Assimilation reanalysis, and the team focused on the common period of July 2009.

The study region covers the part of the Brazil Basin occupied by the tracer release experiment in 1997, where abyssal waters dominated by Antarctic Bottom Water flow northward from the Southern Ocean through the Vema and Hunter channels. The basin’s seafloor is famously rough, featuring canyons that span 10 to 100 kilometers and abyssal hills with typical crest-to-crest distances of 6 to 8 kilometers, slopes that are critical or near-critical for internal tidal motions. That ruggedness makes the basin a demanding test for any model, because the small-scale topography is precisely what drives much of the deep-ocean mixing that models must capture.

To measure mixing in the simulations, the team released clouds of 10,000 strictly passive tracer particles at three depths: in the bottom-most model layer about 40 meters above the seafloor, at 3,000 meters and at 4,000 meters. Five clouds were released every two days during the first nine days of July 2009, and each particle was tracked for 20 days. The initial vertical spread of the tracers was set to 9 meters, matching the value measured during the field experiment. Diffusivity was diagnosed in two ways: from the spreading rate of particles in density space using a tracer-moment method, and from a layer-mean estimate within the bottom boundary layer, which was defined by a density threshold and typically ranged from tens of meters to more than 200 meters thick over steep slopes.

The simulations reproduced the broad observed pattern: modeled diapycnal diffusivity is enhanced toward the bottom and decreases moving up into the water column, consistent with the tracer release measurements. Near the bottom boundary layer, the 1 and 3 kilometer simulations produced diffusivities comparable in value to the observational estimates, which showed diffusivities of 2 to 4 times 10 to the minus 4 square meters per second about 500 meters above an abyssal hill, rising above 10 to the minus 3 within the boundary layer. The 6 kilometer run substantially underestimated these near-bottom values, producing diffusivities of only 6 to 8 times 10 to the minus 5 square meters per second that remained nearly constant throughout the water column, controlled mostly by numerical and parameterization choices rather than resolved dynamics.

The trouble emerged higher in the water column. Above the bottom boundary layer, the modeled diffusivities overestimated the observational values in all runs, and the bias grew with increasing resolution, with the worst overestimate in the 1 kilometer tidal simulation. The culprit, the analysis showed, was an eddy field far more energetic than reality. In the 1 and 3 kilometer runs, many mesoscale and submesoscale eddies populated the deep interior, with the 1 kilometer cases producing eddies with local Rossby numbers near one, surrounded by rings of intense opposite-signed vorticity. These eddies, close in size to the baroclinic Rossby radius of deformation of about 50 kilometers and extending to about 2,000 meters depth, formed independently of tidal forcing and were almost entirely absent in the 6 kilometer run. Tracers released in and around eddy cores mixed roughly two to four times faster than tracers released in the background flow in the 3 kilometer case, and more than three times faster in the 1 kilometer simulations.

The root cause, the authors argue, lies in how models treat the seafloor. To keep terrain-following vertical coordinates stable, the model bathymetry, interpolated from the SRTM30plus dataset, is smoothed with a Gaussian kernel so that topographic steepness never exceeds a fixed ratio. Even in the 1 kilometer run, only topographic features at an effective scale of about 2.5 kilometers are well resolved, while the real basin hosts roughness at scales of 1 kilometer and smaller. Spectral analysis of the topographic height revealed that even the finest model grid underestimates the second moment of the bathymetric spectrum, which governs internal tide energy density, by nearly an order of magnitude, and underestimates the fourth moment, which relates to shear variance and wave breaking, even more severely. In other words, the smoothed model seafloor cannot efficiently generate and scatter the high-mode internal tides that, in the real ocean, break and dissipate energy near the bottom.

That missing energy sink has consequences. In the real basin, rough topography provides a pathway through which geostrophic and tidal motions transfer energy into the internal-wave field, feeding small-scale shear and turbulent dissipation. With smoothed topography, the model generates fewer of these waves, so the eddy field loses a principal energy sink and persists at higher amplitude, contributing to elevated diffusivity that extends into the ocean interior. A sensitivity experiment reinforced this interpretation: running the model at 1 kilometer resolution but with the smoother 3 kilometer bathymetry produced far fewer coherent deep eddies and a circulation resembling the coarser run, showing that bathymetric detail, not resolution alone, controls the simulated eddy field. Energy exchange diagnostics further showed that while eddy kinetic energy in the 1 kilometer run was roughly nine times that of the 3 kilometer run, the resolved internal-wave kinetic energy increased by only about 30 percent, leaving the internal-wave pathway an insufficient energy sink.

The authors caution that other factors contribute as well. The k-epsilon vertical mixing scheme used in the runs is sensitive to background settings, and numerical mixing inherent to terrain-following coordinates can become substantial over steep slopes even when rotated tracer-advection schemes are employed. Additional topographic smoothing reduces that numerical mixing but simultaneously erases small-scale bathymetric structure that is dynamically important for flow-topography interactions, a genuine trade-off for model designers.

The broader lesson is sobering for the kilometer-scale climate models now under development and for the machine-learning ocean emulators increasingly trained on submesoscale-permitting simulations. Resolution alone is not the answer: pushing a model to submesoscale-permitting resolutions over rough topography may produce unrealistic, overestimated diffusivities in the ocean interior, counterintuitively more so than coarser configurations that cannot resolve the instabilities generating the eddies. The team calls for nonhydrostatic simulations at finer horizontal and vertical resolution, and for new observational constraints to establish how often and by how much the relative contributions of eddies and internal waves near the seafloor are misrepresented in state-of-the-art ocean models, since those contributions shape the residence time of heat and carbon across the water column.

Subject of Research: How model resolution and bathymetric smoothing affect simulated deep-ocean diapycnal mixing in the Brazil Basin

Article Title: Diapycnal mixing in submesoscale-permitting simulations of the deep Brazil Basin

Article References: Huang, Y., Bracco, A., Polzin, K., & Gula, J. (2026). Diapycnal mixing in submesoscale-permitting simulations of the deep Brazil Basin. Ocean Science, 22(5), 3415-3433. https://doi.org/10.5194/os-22-3415-2026

Image Credits: AI Generated

DOI: 10.5194/os-22-3415-2026

Keywords: diapycnal mixing, Brazil Basin, ocean modeling, CROCO, submesoscale eddies, internal tides, bathymetric smoothing, bottom boundary layer, tracer particles, BBTRE, ocean diffusivity, deep ocean