[MITgcm-support] On the vertical mixing parameterization

[Stefano Querin]

Dear MITgcm-ers,

...and what about writing a sort of a "compendium" on mixing schemes/parameterizations/values for simulations on different spatial and temporal scales?
For example, when trying to simulate deep water thermohaline trends, I found extremely useful the second-order moment (SOM) advection scheme [Prather, 1986] to reduce the spurious diapycnal mixing (especially in the deep layer) due to the vertical z-level discretization. That scheme is cited only marginally in the documentation (it has been described more thoroughly in the very nice paper by Hill et al. [2012]).
This is just an example, but there are many situations in which choosing the correct approach could narrow the (very wide) range of options that a user can test for each particular case study; e.g., Leith and/or Smagorinsky scheme? KPP or other vertical schemes? Harmonic and/or biharmonic coefficients? What should I prefer for coastal/open sea problems? ...and so on...
Of course, every numerical experiment is strongly site and resolution dependent, but a very synthetic collection of (up-to-date) "rules of thumb" could be helpful for both beginners and experts. Then, as suggested by Andrea, a systematic comparison with field observations is essential.
I have some experience on problems with resolutions ranging from a few kilometers to a few tens of meters, on mid-latitude marginal seas (Mediterranean and sub-basins), and I will be happy to share my expertise on these case studies.
I am thinking about a shared Google document or spreadsheet in which we could (very briefly) summarize, review and compare (and criticize!) our experiences (area/basin, resolution, time-stepping, numerical scheme, coefficients), taking into account (of course!) the official MITgcm documentation, recommendations and references. I collected also a quite large amount of information about mixing schemes/coefficients from the previous posts of this mailing-list.

If somebody thinks it could be a good (and useful) idea, please reply to this e-mail or contact me directly (to avoid spam...).
Any other idea, proposal, criticism is most welcome!

Cheers,

Stefano


On 17 Jan 2017, at 10:01:37, Martin Losch wrote:

> Hi Chengyan Liu,
> 
> KPP computes the turbulent mixing coefficients for the entire water column from different processes. Generally, in the ocean interior the vertical diffusion/mixing is caused by processes like internal wave breaking. The background diffusion is energetically relevant, because it mixes dense deep water upwards against buoyancy gradients and hence forms an important (albeit poorly constrained and hence somewhat “ad-hoc") energy source for the global overturning. For this the “canonical value” is something between 1e-4 and 1e-5m^2/s. You provide this value via diffKhT/S and then KPP uses it whereever the mixed layer mixing is below this value.
> 
> With sufficient numerical diffusion (e.g. due to a diffusive advection scheme), you don’t need any background diffusion to make the model stable, but you will not explicitly include the mixing energy. Having said that, the numerical diffusion is often larger than the canonical 1e-5m^2/s.
> 
> Martin
> 
> 
>> On 17 Jan 2017, at 08:52, Andrea Cimatoribus <andrea.cimatoribus at epfl.ch> wrote:
>> 
>> Hi,
>> from my experience with MITgcm and other models I can confirm your tests: background diffusivities are very important. In fact, in my experience, the results are much more sensitive to changes in background diffusivities than to changes in the turbulence model parameters (or, in fact, to using different turbulence models).
>> How much is the right amount of background diffusivity then depends on many details of your simulations. In general, I would say that higher resolution simulations need less background diffusivities (I get for instance very good results with molecular background values at ~200m resolution), but a final answer can only come from systematic comparison with observations. As far as I know, there can be no a priori answer.
>> 
>> Ciao,
>> 
>> Andrea Cimatoribus
>> postdoctoral researcher
>> EPFL ENAC IIE ECOL
>> https://people.epfl.ch/andrea.cimatoribus
>> 
>> On 16/01/17 22:39, whale wrote:
>>> Dear friends
>>> 
>>> 
>>> 
>>> I am confused on the treatment of the vertical mixing parameterization
>>> in MITgcm.
>>> 
>>> In my experiment, I have employed the KPP (Nonlocal K-Profile
>>> Parameterization for Vertical Mixing). As the manual explained, the KPP
>>> could control the vertical mixing by supplying vertical diffusivity and
>>> viscosity.
>>> 
>>> 
>>> 
>>> The question is if I still need to set the other parameters of vertical
>>> mixing and viscosity in /date file/, such as diffKrT and diffKrS, when I
>>> have already employed the KPP.
>>> 
>>> 
>>> 
>>> In my mind, these two parameters seem likely to be linked to the
>>> background vertical mixing, and I do not know if my understanding is
>>> right or not. Actually, I have already run two experiments with and
>>> without the diffKrT and diffKrS, respectively. The results of my two
>>> experiments are different from each other, and confirm that the diffKrT
>>> and diffKrS still have a great effect when the KPP is employed.
>>> 
>>> 
>>> 
>>> Thank you for your consideration.
>>> 
>>> Cheers
>>> 
>>> Chengyan Liu
>>> 
>>> 
>>> 
>>> _______________________________________________
>>> MITgcm-support mailing list
>>> MITgcm-support at mitgcm.org
>>> http://mitgcm.org/mailman/listinfo/mitgcm-support
>>> 
>> 
>> _______________________________________________
>> MITgcm-support mailing list
>> MITgcm-support at mitgcm.org
>> http://mitgcm.org/mailman/listinfo/mitgcm-support
> 
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support