[MITgcm-support] Regional high-res model configuration

Wed Apr 22 04:08:13 EDT 2020

Hi 

this is the post about tuning viscosities, that I couldn’t find earlier:
http://mailman.mitgcm.org/pipermail/mitgcm-support/2006-June/004086.html

M.

> On 22. Apr 2020, at 09:38, Martin Losch <Martin.Losch at awi.de> wrote:
> 
> Hi again,
> 
>> On 21. Apr 2020, at 19:15, Stanislav Martyanov <martyanov.sd at gmail.com> wrote:
>> 
>> Hi Martin!
>> 
>> Thanks for your reply!
>> 
>> Right to the business:
>> 
>>>> I would use viscAhGrid or viscAhGridMin for this (something small)
>> To be honest, I hardly understand the nature of viscAhGrid parameter... While viscAh can be seen as some kind of  constant physical background viscosity not resolved by the main horizontal turbulence parameterization used (Smagorinsky, Leith, etc),  viscAhGrid seems like some numerical trick. From its formula in MOM_CALC_VISC it can be concluded that it has something to do with grid size (just like Smagorinsky).
> Please have a look at the corresponding section the online documentation <https://mitgcm.readthedocs.io/en/latest/algorithm/algorithm.html#nonlinear-viscosities-for-large-eddy-simulation>, and also at the header of mom_calc_visc.F, where it should become clear how all of this is handled.
>> 
>>>> this is only 2D Smagorinsky. I would use 2D Leith instead.
>> The only reason I chose Smagorinsky parameterization was its dependence on grid resolution and dynamical regime (velocity gradients are involved). But your alternative (viscAhGrid + viscC2leith), perhaps, does the same in a combined manner.
> As far as I know (and I am by no means a specialist on turbulence), Smagorinsky parameterises 3D turbulence (which is kind of funny to then do it only in 2D by default in the code), and there’s even code to do that. The Leith scheme parameterises 2D turbulence and as long as the aspect ratio of your grid is so large (dx/dz) that makes more sense to me.
>> 
>>>> setting viscAhGridMax (something a little smaller than 1)
>>>> may be necessary
>> In MOM_CALC_VISC  the recommended value for viscAhGridMax is 1. What value would you use?
> that’s the maximum allowed for numberical “cfl-like” stability. It’s usually OK.
>> 
>>>> depending on your advection scheme, horizontal diffusivities
>>>> can be zero (I would use 77, 33 or 7 for advection)
>> I use the advection scheme 33 (3-order DST with Sweby flux limiter) for T, S and seaice.
>> 
>> Does the next set of parameters seem better than the previous one?
>> viscAhGrid = 0.5, viscAhgridmax = 1.0, useFullLeith = .TRUE., viscC2leith = 1.0, viscC2LeithD = 1.0, viscAr = 1.0e-5.
>> Diffusivities: diffKhT = 1.0, diffKhS= 1.0, diffKrT = 1.0e-5, diffKrS = 1.0e-5.  
>> Here I additionally set viscC2LeithD as it is recommented in the manual (a damping specifically targeting purely divergent instabilities near the gridscale).
> As these these are all additive, the gridMax =1 is important. viscAhGrid=0.5 is too large for a small background. See header of mom_calc_visc.F to estimate how large this will be in dimensional units (0.25*L**2*viscAhGrid/deltaT = 0.25 * 1e6 * 0.5 / deltaT in your case). Horizontal diffusivities for T and S are not necessary with your advection scheme.
> In the end you’ll have to experiment with all of that. See this post:
>> 
>>>> Don’t use the CD scheme, if necessary use small biharmonic viscosity
>>>> instead. Implicit vertical advection is probably not required either
>> OK, the CD_scheme is omitted then, thank you! But can Implicit vertical advection significally alter the results? I just hoped to use it in order to somewhat increase the time step.
> You’ll have to test that. There’s some discussion on this list. I cannot find the exact post I was looking for, though it was a good set of instructions by Baylor how to tune the viscosity parameters.
>> 
>>>> Why do you want to use the MY82 package?
>> The only reason is that I previously used it and familiar with it. Moreover, the MY is a well-known mixing scheme, and both its good and bad sides are also documented.
> Make sure that it is really is what you think it is. This is not the 2.5-level closure MY scheme, but something simpler.
> 
>>>> I’d prefer GGL90 or KPP (although at your resolution, KPP might be noisy (see this thread <http://mailman.mitgcm.org/pipermail/mitgcm-support/2020-April/012427.html>).
>> Previously our team used the KPP for another configuration, but for now I decided to omit it. I read the thread you mentioned, and it was the last straw to find another scheme :-)
>> 
>> As for GGL90, I did not previously used it at all. From the original paper by Gaspar et al, 1990, I know that it has only one prognostic equation (for TKE) and that the turbulent length scale is diagnosed rather than prognosed. From your experience, can you  recommend it over the MY82?
> GGL90 is definitely more sophisticated taht MY82. You’ll have to try it out.
>> 
>>>> but it should (like GGL90) remove the instabilities,
>>>> because it’s also Richardson number based
>> Yes, I know, and that is why I was surprized by the comment in the code "convective adjustment might be needed even with ggl90". Do you use GGL90 with or without convAdj or  ivdc_kappa?
> I never paid attention to that, but I guess it should work without cAdj or ivdc_kappa.
>> 
>> I also know that you are the author of the SEAICE package for MITgcm, so could you tell, is it OK for now to use the newly added feature SEAICE_ALLOW_MOM_ADVECTION and corresponding run-time parameters SEAICEmomAdvection=TRUE, SEAICEuseMetricTerms = TRUE, and SEAICEuseFluxForm=TRUE if my configuration uses curvilinear grid (and that's why I must use vectorInvariant for momentum)? The vectorInvariant does not use the metric terms for momentum, should I use them for SEAICE on my grid? And as I understood, there is no alternative for  SEAICEuseFluxForm?
> I have never tested the option SEAICE_ALLOW_MOM_ADVECTION properly, but just implemented it upon request. Please go ahead and try it.
> As far as I understand, the momentum advection is only correct on curvilinear grids with the vector invariant option, but that only applies to the momentum advection. The entire stress formulation uses the metric terms by default (you should not have to set it explicitly), but recently I found that I may have missed a bunch of terms because of insufficient tensor analysis background at my end. With grids that are quasi-rectangular, this is probably not much of an issue. 
> SEAICEuseFluxForm only applies to the advection scheme 2 for tracers (HEFF/AREA), which I wouldn’t use anyway (use 77 or 33).
> 
> Martin
> 
>> 
>> Stanislav
>> 
>> 
>> 
>> 
>> 
>> вт, 21 апр. 2020 г. в 14:31, Martin Losch <Martin.Losch at awi.de>:
>> Hi Stanislav,
>> 
>> generally your parameters look good, so comments below
>> 
>>> On 21. Apr 2020, at 11:23, Stanislav Martyanov <martyanov.sd at gmail.com> wrote:
>>> 
>>> Hello, dear colleagues!
>>> 
>>> I am configuring the MITgcm for the Kara Sea region. With the curvilinear grid, the model's resolution is: dx=500-1200 m, dz=2 m in upper layers, and up to 20 m in the deepest. The minimum model depth is set equal to 5 m. I use the pure r vertical coordinate (not r*) for now. Because of the curvilinear grid, I use vectorinvariant package. Other packages used are: seaice, my82, exf, cal, obcs.
>>> 
>>> Having carefully read (once again) the newest version of the online-manual and browsed the code, I have configured almost all necessary model parameters, but still some questions remain. They are qiute specialized, so I could not find the needed information in puplished papers. But for those experienced with MITgcm, they might be easy to answer. I will appreciate any advise!
>>> 
>>> 1) The newly added feature with partial cells (#undef EXCLUDE_PCELL_MIX_CODE in CPP_OPTIONS). The parameter pCellMix_select = 0 by default, which means that the enhanced mixing is OFF in bottom and surface layers. The parameters intended for the enhanced mixing in the inner layers are also OFF by default: interViscAr_pCell and interDiffKr_pCell = FALSE. Are this option and the corresponding run-time parameters really important for such configuration (Kara Sea)? Any experience with them?
>> I have no experience with this, but I think you can safely keep this turned off.
>>> 
>>> 2) I use the following viscocity and diffusivity values, intending to use the Smagorinsky viscocity for lateral turbulent exchange:
>>> viscAh = 1.0,
>> I would use viscAhGrid or viscAhGridMin for this (something small)
>> this is only 2D Smagorinsky. I would use 2D Leith instead.
>>> viscC2smag = 3.0,
>> setting viscAhGridMax (something a little smaller than 1) may be necessary
>>> viscAr = 1.0e-5,
>> depending on your advection scheme, horizontal diffusivities can be zero (I would use 77, 33 or 7 for advection)
>>> diffKhT = 1.0,
>>> diffKhS= 1.0,
>>> diffKrT = 1.0e-5,
>>> diffKrS = 1.0e-5,
>>> But for diffusivity there is no Smagorinsky analogue. What approach can be the most suitable in this case?
>>> 
>>> 3) In the paper "ECCO version 4 - an integrated framework... " by Forget et al., 2015, the authors decided to switch OFF the C-D sheme and to switch ON the Implicit vertical advection option. But they investigated the global scales, not high-res. From your experience, can their advise be applyed to the high-res ocean simulations?
>> Don’t use the CD scheme, if necessary use small biharmonic viscosity instead. Implicit vertical advection is probably not required either
>>> 
>>> 4) The MY82 package does not allow using the convective adjustment (cAdjFreq) and ivdc_kappa. Does it mean that the MY in MITgcm works fine with vertical instability cases? Actually, it should, as follows from J. Mellor's papers, but this question arose when I was investigating the GGL90 code, where I found the line "convective adjustment might be needed even with ggl90"…
>> Why do you want to use the MY82 package? I’d prefer GGL90 or KPP (although at your resolution, KPP might be noisy (see this thread <http://mailman.mitgcm.org/pipermail/mitgcm-support/2020-April/012427.html>).
>> You can just comment out the stop-statement in the code and use MY82 with other convective adjustment, but it should (like GGL90) remove the instabilities, because it’s also Richardson number based. MY82 is not really used that much (at all?).
>> 
>> Martin
>>> 
>>> Once again, I will appreciate any advise!
>>> 
>>> PS: Sorry for large wall-of-text here, but I tried to be concrete. :-)
>>> _______________________________________________
>>> MITgcm-support mailing list
>>> MITgcm-support at mitgcm.org
>>> http://mailman.mitgcm.org/mailman/listinfo/mitgcm-support
>> 
>> _______________________________________________
>> MITgcm-support mailing list
>> MITgcm-support at mitgcm.org
>> http://mailman.mitgcm.org/mailman/listinfo/mitgcm-support
>> _______________________________________________
>> MITgcm-support mailing list
>> MITgcm-support at mitgcm.org
>> http://mailman.mitgcm.org/mailman/listinfo/mitgcm-support
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mailman.mitgcm.org/mailman/listinfo/mitgcm-support