Overly smoothed surface currents and suppressed mesoscale eddies in FVCOM vs. MITgcm and CMEMS

[Mohammad Hossein Kharaghani]

Dear FVCOM Community,

I am writing to seek your guidance regarding an issue I have encountered with surface circulation and mesoscale eddy representation in FVCOM.

I am conducting regional baroclinic simulations covering the Persian Gulf and the Sea of Oman / northern Arabian Sea. To evaluate the results, I compared daily-averaged surface velocity fields for January 12, 2021 across three datasets (attached figure):

Panel (a): MITgcm

Panel (b): FVCOM

Panel (c): CMEMS (Copernicus Global Ocean Analysis) (Similar coherent eddy fields are also present in HYCOM)

Model Setup & Configuration

Both MITgcm and FVCOM are configured with nearly identical physics, resolution, and forcing:

Horizontal Grid & Resolution: MITgcm is configured on a structured grid with ~0.05° resolution. FVCOM employs an unstructured triangular mesh with comparable resolution (~0.06° in deep waters, refined to ~0.01°–0.02° [~1–2  km] along coastal boundaries).

Vertical Discretization: MITgcm uses 50 z-levels, and FVCOM uses 50 -layers. Both setups maintain a near-surface vertical resolution of approximately 2 meters.

Forcing & Tracers: Fully prognostic temperature and salinity, active tidal constituents, and identical atmospheric momentum (wind stress) and surface heat fluxes.

Turbulence Closures: Smagorinsky parameterization for horizontal mixing and the Mellor–Yamada Level 2.5 (MY2.5) scheme for vertical mixing.

The Problem

As seen in Panels (a) and (c), both MITgcm and CMEMS clearly capture prominent cyclonic and anticyclonic mesoscale eddies in the Sea of Oman.

In contrast, FVCOM (Panel b) produces an overly smooth, wind-aligned surface current field where mesoscale eddy structures are almost completely suppressed and damped in the upper layers.

We know that FVCOM is fully capable of resolving complex basin-to-global ocean dynamics, as demonstrated in seminal Global-FVCOM applications—such as the modeling for the Air France 447 search ("FVCOM model estimate of the location of Air France 447") and the 2011 Tōhoku tsunami assessment ("The March 11, 2011 Tōhoku M9.0 earthquake-induced tsunami and coastal inundation along the Japanese coast: A model assessment"). However, in this regional setup, the model seems to wipe out the surface eddy structures entirely.

Question

Given that both models share comparable surface forcing, similar horizontal and vertical resolutions (~2 m near the surface), and identical mixing closures (Smagorinsky + MY2.5), why is FVCOM exhibiting such heavy damping of surface mesoscale eddies, and what settings (e.g., horizontal momentum advection schemes, Smagorinsky diffusion parameters, or surface boundary condition formulations) could be responsible for this excessive smoothing?

Any insights or recommendations from your experience would be greatly appreciated.

Sincerely,

Mohammad Hossein Kharaghani

Ph.D. Candidate

Iran University of Science and Technology (IUST)

Tehran, Iran

[Siqi Li]

Hi Mohammad,

Before looking into the FVCOM numerical schemes or mixing parameters, I would suggest first checking the basic model setup, particularly the initial conditions, open boundary conditions, spin-up period, and wind forcing.

Mesoscale eddies are sensitive to these factors, and similar grid resolution and turbulence schemes do not necessarily mean that two models should produce the same eddy field at a particular time. It would be useful to first determine whether the eddies fail to develop in FVCOM, or whether they develop and are subsequently damped.

I would suggest comparing the circulation evolution from the beginning of the simulation, rather than only comparing the surface velocity fields on a single day. This may help identify when and why the FVCOM solution begins to differ from MITgcm/CMEMS.

Best,
Siqi

[Mohammad Hossein Kharaghani]

Dear Siqi,

Thank you very much for your helpful feedback and suggestions.

I completely agree that mesoscale eddy fields are turbulent and nonlinear, so an exact spatial and temporal match between different models is not expected. Following your advice, I examined the simulation's evolution from the very first days through several months, as well as the vertical structure. Here are the key details and observations:

Model Setup

Initial Conditions: Temperature () and Salinity () interpolated from HYCOM, initialized from rest ().

Atmospheric Forcing: Hourly ERA5 reanalysis fields (wind stress, heat fluxes).

Open Boundaries: Baroclinic  and  from HYCOM, with tidal elevations prescribed from the FES global tidal atlas.

Observed Flow Evolution & Vertical Structure

Immediate Difference from Day 1: Right from the first 1–2 days of the simulation (the initial spin-up stage), MITgcm already resolves the surface signatures of the mesoscale eddies. In contrast, FVCOM immediately produces an overly smooth, wind-aligned surface flow where eddy structures are completely absent.

Vertical Extent: This over-smoothing in FVCOM is strictly confined to the upper 7–8 -layers (down to roughly 30–40 meters) where the wind directly dominates. Below this layer (deeper than ~30–40 m), the mesoscale eddies are clearly present and well-developed in FVCOM.

Long-term Persistence: This pattern does not change with extended spin-up; even after running the simulation for over 4 months, the upper 30–40 meters remain heavily smoothed, while the eddies persist in the deeper layers.

Given that the eddies do develop below the top 30–40 meters, the issue remains persistent exclusively in the wind-influenced surface layers. I would greatly appreciate any suggestions or advice on what model settings or adjustments could help resolve this near-surface over-smoothing in FVCOM.

Best regards,

Mohammad Hossein Kharaghani

[Mohammad Hossein Kharaghani]

Apologies for the formatting glitch in my previous message where a few symbols were omitted. Please find the clarified Model Setup section below in plain text:

Model Setup:

• Initial Conditions: Temperature (T) and Salinity (S) interpolated from HYCOM, initialized from rest (u = v = 0).

• Atmospheric Forcing: Hourly ERA5 reanalysis fields (wind stress, heat, and freshwater fluxes).

• Open Boundaries: Baroclinic T and S from HYCOM, with tidal elevations prescribed from the FES global tidal atlas.

• Vertical Discretization: FVCOM uses 50 sigma-layers (with the smoothing occurring across the top 7–8 sigma-layers).

[Siqi Li]

Hi Mohammad,

Thanks for the additional information. Since the smoothing is mainly confined to the wind-dominated upper 30–40 m, I would suggest checking the wind forcing first.

Before running the model, it may be useful to plot the wind fields (and preferably the actual wind stress applied in FVCOM) to make sure the forcing is correct. You could also run a simple sensitivity test with the wind stress turned off to see whether the surface eddy structures recover.

This may help determine whether the issue is related to the wind forcing before changing other FVCOM numerical settings.

Best,
Siqi

[Mohammad Hossein Kharaghani]

Dear Siqi,

Thank you for your valuable feedback and guidance.

Following your recommendation, I performed the zero-wind sensitivity test and conducted detailed diagnostics on the wind forcing applied to FVCOM:

1. Zero-Wind Test Result:

When wind stress is turned off completely, the surface mesoscale eddy structures clearly emerge in FVCOM. This confirms that the model can form these vortices, but their surface expression disappears exclusively when wind forcing is activated.

2. Wind Forcing Verification & Mesh Interpolation Checks:

To verify that no interpolation errors or distortions occurred during preprocessing, I compared the raw atmospheric data directly with the field interpolated onto the FVCOM mesh (both attached for 2021-01-12 23:00:00):

Figure 1: Raw CMEMS wind speed and streamlines.

Figure 2: Wind speed and streamlines interpolated onto the native FVCOM triangular mesh.

A side-by-side comparison confirms that the interpolation is completely faithful: vector directions, magnitudes, and boundary transitions match the raw dataset without any numerical artifacts or discontinuities.

The wind field exhibits a realistic, smooth synoptic wind pattern (a characteristic winter Shamal event). However, the surface ocean circulation in FVCOM seems to rigidly mirror this smooth atmospheric wind direction, wiping out the surface imprint of the underlying eddies.

3. Consistency Across Forcing Products:

Identical Forcing: Both MITgcm and FVCOM are driven by the exact same wind forcing. Under identical conditions, MITgcm preserves the surface eddy structures, whereas FVCOM smooths them out entirely.

Direct Wind Stress vs. Wind Speed: I tested providing both 10 m wind velocities (U10, V10) and direct wind stress fields to FVCOM, but the surface smoothing remained identical.

High-Resolution Datasets: In addition to standard hourly ERA5 (~0.25 degrees), I also applied the high-resolution hourly CMEMS L4 wind product (~0.125 degrees; Product: WIND_GLO_PHY_L4_MY_012_006 / Dataset: cmems_obs-wind_glo_phy_my_l4_0.125deg_PT1H_202211), which directly supplies wind stress components. The smoothing behavior in FVCOM persisted.

Given that the wind forcing and its mesh interpolation are verified and identical between the models, and considering that the eddies appear in FVCOM when the wind is turned off, why does activating the wind lead to this complete over-smoothing of surface eddy structures in FVCOM, and what would you suggest checking or adjusting to resolve this?

[Siqi Li]

Hi Mohammad,

Thanks for the update. Since you are now debugging the forcing, I would suggest plotting the actual variables used to drive FVCOM as directly as possible, rather than derived quantities such as wind speed and streamlines.

For example, if FVCOM is driven by wind velocity, please plot U10 and V10 separately. If wind stress is used, plot Ustress and Vstress separately, before and after interpolation onto the FVCOM grid. This may help reveal issues in an individual component that are not obvious from wind speed or streamlines.

Also, since you mentioned that the surface and deeper-layer results are very different, I suggest selecting several representative locations and plotting the vertical profiles of temperature, salinity, and velocity. This may help identify where the difference starts to develop in the vertical direction.

Best,
Siqi

[Mohammad Hossein Kharaghani]

Dear Siqi,

Thank you for the guidance. Following your suggestions, I plotted the wind components separately and extracted vertical profiles across the domain to better understand the vertical structure.

I have attached the following figures for your review:

U10V10_Wind.jpg: Separate eastward (U10) and northward (V10) wind velocity components at 2021-01-12 23:00:00.

Stations_Map.png: Locations of 7 representative stations across the Persian Gulf, the Strait of Hormuz, and the Gulf of Oman / Arabian Sea.

Station_1_Profile.png to Station_7_Profile.png: Daily-averaged vertical profiles of zonal velocity (u), meridional velocity (v), potential temperature, and salinity on Day 12 of the simulation, comparing FVCOM (blue solid lines with circle markers) and MITgcm (red dashed lines with square markers).

Observations from the Diagnostic Plots

Wind Forcing Components:

The U10 and V10 component fields show smooth, consistent synoptic patterns without any vector sign flips, discontinuities, or orientation anomalies.

Temperature and Salinity Profiles:

Across all stations, the vertical profiles of potential temperature and salinity show reasonably good agreement between FVCOM and MITgcm. Both models capture the overall stratification, thermocline, and halocline structures well.

Velocity Profiles:

There is a pronounced difference in how velocity evolves vertically between the two models:

In the deep-water stations of the Gulf of Oman (Stations 3 through 7, depths ~950 m to ~3300 m), FVCOM currents below the upper ~50–100 m remain nearly stagnant and vertically uniform (close to zero).

In contrast, MITgcm develops significant baroclinic shear and active subsurface currents extending through the intermediate depths.

In FVCOM, almost all the momentum and motion remain concentrated in the uppermost layers, while the deeper water column stays largely at rest despite having the baroclinic density structure in place.

Sensitivity to Sigma-Layer Distributions:

To rule out vertical grid design as the cause, I tested three different non-uniform/hybrid vertical sigma distributions (all using 50 layers). These ranged from designs with heavy layer concentration in the top ~50 m to configurations extending refined vertical resolution down through the thermocline/pycnocline (~300–500 m) to better capture density stratification. Across all three vertical grid setups, the stagnant subsurface velocity pattern in FVCOM remained practically unchanged.

Persistence Beyond the Spin-up Period:

While these specific profile plots are shown for Day 12 (which falls within the initial spin-up stage), I have also inspected the results at much longer integration times—after 2, 3, and even 4 months of simulation. The vertical velocity pattern in FVCOM remains essentially the same over time, showing that this behavior is persistent and does not resolve itself simply with extended spin-up.

These results summarize our current setup and observations. I would greatly appreciate your thoughts on these profiles and any suggestions on what might be causing FVCOM to exhibit this vertical velocity structure.

Best regards,

Mohammad Hossein Kharaghani