LIMIT_VVISC

[pmaccc]

I see in this ticket from Hernan: https://www.myroms.org/projects/src/ticket/736 that the ability to limit the eddy viscosity and diffusivity was introduced about 5 years ago (e.g. if LIMIT_VVISC is defined in your .h then the code will use AKV_LIMIT from your .in, and the default in ocean.in is 10^-3).

My question is, does anyone have guidance about using these in practice? I note that in the example cases distributed with the current ROMS there are several that define GLS_MIXING, but none that define LIMIT_VDIFF or LIMIT_VVISC.

Tanks for your help,

Parker

[xupeng66]

Here is a follow-up on Parker's questions. While inspecting the vertical eddy viscosity and diffusivity computed by turbulence closures in ROMS, I found that the results were different by several orders of magnitudes between GLS and Mellow-Yamada Level 2.5 in an idealized simulation of the circulation off the Washington Coast. I have attached a plot showing the comparison along a zonal transect at 47 deg N. Note Akv and Akt are plotted on log scales, so the range of the colorbar is 10^-9 to 10^-6 m^2/s for MY25 (left) and 10^-4.5 to 10^-3 m^2/s for GLS (right). It seems that one of the two closures must be problematic. I then came across this paper (Robertson and Dong, 2019, "An evaluation of the performance of vertical mixing parameterizations for tidal mixing in the Regional Ocean Modeling System (ROMS)"). The authors found that the eddy viscosity and diffusivity computed using GLS were so large that they suspected there was a coding error. I am curious to know if this issue with GLS_mixing has been solved in a more recent release of ROMS. Thanks!

[jcwarner]

ok.
First of all let's tackle the AKV_limit issue. This option is intended to prevent the values of Akt and Akv from getting very large.
I do a lot of estuary work with lots of mixing, so I recommend setting these limiter values be at something like AKV_limit=1.0d0 and Akt_limit=1.0d0.
IF you are unsure what to set, then DO NOT use this option.
This option is an option, not required.
DO not make these values too small, or the Akv will be limited to be very small and the shear will get too large. This is probably what those authors did (Robertson and Dong, 2019), made this number too small, there was no mixing and the shear was too large. I do not remember seeing a post on that.

If there is something wrong, I will fix it. But I am not finding anything wrong.
We have a multitude of studies and publications for ~15 years that show how well the GLS scheme is doing. We are simulating the vertical stratification in the Hudson, Delaware, and US east coast hurricanes very well.

Please look at my most recent hudson paper ( many other older ones):
Warner, J. C., Geyer, W. R., Ralston, D. K., & Kalra, T. (2020). Using tracer variance decay to quantify variability of salinity mixing in the Hudson River Estuary. Journal of Geophysical Research: Oceans, 125, e2020JC016096. https://doi.org/10.1029/2020JC016096
Look at Figure 3 - comparison of salinity at several stations along 50km of the Hudson for over 100 days. I am actually very amazed how the model can just keep running and produce results that well.

Here is an older but great paper:
https://agupubs.onlinelibrary.wiley.com ... 04JC002585
Ming show comparisons of Akt in his Fig 10.

We run a forecast model daily using GLS, and our Atlantic ocean is not mixing away.
I randomly searched and found this paper:
https://gmd.copernicus.org/articles/14/ ... 7-2021.pdf
Their Figure 1 shows vertical diffusivity ~ 1e-5. So that agrees with your figure 7 for GLS.

I am not an expert but i think MY25 showing AKV at 1e-8 is rather small.
That is less than the background value we use.
Maybe i am misunderstanding those plots?

In summary, I do not see / experience any issues with GLS.

-john

[xupeng66]

Thanks for the detailed explanation, John. To answer your question about the small viscosity and diffusivity computed using MY25, there is no surface or tidal forcing in my idealized simulations. The dynamics are driven by the lateral boundary conditions constructed from daily averaged output of a global reanalysis product (GLORYS12V1). I think the lack of surface or tidal forcing may be the reason why the MY25 viscosity and diffusivity are quite small. Note that I have also set the background values (Akv_bak, Akt_bak) to zeros.

The ROMS tested by Roberson and Dong (2019) is version 3.4. I am using version 3.7. Could you please let me know the version of ROMS you used to generate the results in your 2020 paper?

[jcwarner]

i dont think GLS has changed much in several years, so it should be the same.
-j

[zhang]

It is really strange to see Robinson and Dong's claim that the ROMS GLS code has issues for simulating internal tides. I have been using GLS for simulations of internal tides at a shelf edge for years. I have no problem with the code. The modeled viscosity and diffusivity look fine to me, and the resulted internal wave field looks fine as well. I certainly haven't seen any of my simulations turning into barotropic (vertically uniform) in 1 day as Robinson and Dong described in their paper (Figure S2).

Gordon