[MITgcm-support] ice layer modeling -> cube corners?

[Martin Losch]

Hi Elja,

topologically "llc" and the cs32 are very similar, but I was careless  
in putting in my 5-cents-worth:  In the llc configuration we rotate  
the grid so that the "cube" corners are in land. Then the sea-ice  
model never sees these corners.

I am afraid that you found a limitation of the current sea ice  
dynamics implementation: it does not work with cube-corners. Try  
setting Pstar to zero (SEAICE_strength=0) and set SEAICE_deltaTevp =  
100. (= your deltaT). This will give a free drift ice model (no  
internal ice stresses),  but that may actually work (although there  
is no guarantee) and it includes the effect of tidal velocities etc.  
(just without internal stress) ... (Also you will use the EVP model  
implementation which I don't recommend in general, but here it's hack  
to overcome any iteration that involves MPI-exchanges across cube  
corners).

Martin


On 21 Jul 2008, at 18:06, Elja Huibregtse wrote:

> Hi Dimitris, Martin,
>
> Thank you very much for your help:
>
> >> In your output file, do:
> >> grep advcfl_ STDOUT.0000
> The values are all well below zero (1E-3), only for the last 2  
> timesteps they increase rapidly and become larger that 1. DeltaT is  
> in my case 100 seconds.
>
> >> One thing that you can do to test above hypothesis is to turn off
> >> sea ice dynamics and see if it helps.
> >> SEAICEuseDYNAMICS  = .FALSE.,
> I turned this parameter off, and the odd 'corner-behavior' indeed  
> dissapeared. (Unfortunately, the model still became instable.) But  
> I am wondering what is going on, now. Does this mean that the  
> seaice thickness only depends on thermodynamics and that the seaice  
> does not respond to the prescribed tidal potential?
>
> >>it's correct, that we never cared about cube corners in the sea-ice
> >>model, but in my tests, (with the llc-configuration), I did not have
> >>any problems so far, especially not at coarse resolution.
> Martin, I am using the 'global_ocean.cs32x15' configuration. What  
> is the difference between your llc-configuration and the  
> global_ocean.cs32x15?
>
> My goal is to study the ocean dynamics, influenced by a strong  
> tidal potential. I expect that the seaice plays an important role  
> (because the forcing works on the ice, as well). So, I would like  
> to take the seaice dynamics into account. What would you recommend:  
> using another model, like the llc-configuration or ..?
>
> Thanks in advance!
>
> Elja
>
>
> -----Oorspronkelijk bericht-----
> Van: mitgcm-support-bounces at mitgcm.org namens Martin Losch
> Verzonden: ma 7/21/2008 9:45
> Aan: mitgcm-support at mitgcm.org
> Onderwerp: Re: [MITgcm-support] ice layer modeling -> cube corners?
>
> Hi Elja, Dimitris,
>
> it's correct, that we never cared about cube corners in the sea-ice
> model, but in my tests, (with the llc-configuration), I did not have
> any problems so far, especially not at coarse resolution.
>
> Martin
>
> On 18 Jul 2008, at 15:42, Dimitris Menemenlis wrote:
>
> > Elja, the correct overlap in your case is OLx=Oly=8, I think.   You
> > can test this by gradually increasing the overlap for a very short
> > problem without optimization, until the pickup files are bit-
> > identical.  But I don't think that this is what causes trouble at
> > the corners.  The change in the solution from OLx 5 to 8 will be
> > very small.
> >
> > The corners of the cube have smaller dimensions than the rest of
> > the cube so it's easier to violate CFL conditions.  What are your
> > typical velocities, UVEL, VVEL, WVEL, relative to your grid
> > dimensions, dx, dy, dz?
> >
> > In your output file, do:
> > grep advcfl_ STDOUT.0000
> >
> > If any of these values gets much larger than 1, then you may have
> > to decrease your deltaT even further.
> >
> > But deltaT is already quite small in your case if I remember
> > correctly so this may not be the problem.
> >
> > A second possibility is that sea ice model will not work at the
> > corners.  All the tests that I have run with sea ice model, the ice
> > never gets close to the corners.  In particular, the metric terms
> > are ignored in the sea ice dynamics, so close to the corners I have
> > no idea what will happen.  Also, I don't think that the ice
> > momentum equations are compatible with corners.
> >
> > Does anyone else on this list have experience with running the sea
> > ice model across the cube corners?
> >
> > One thing that you can do to test above hypothesis is to turn off
> > sea ice dynamics and see if it helps.
> >
> > SEAICEuseDYNAMICS  = .FALSE.,
> >
> > in data.seaice
> >
> > Dimitris Menemenlis
> > DMenemenlis at gmail.com
> >
> > On Jul 18, 2008, at 6:00 AM, Elja Huibregtse wrote:
> >
> >> Unfortunately, I still didn't find the solution to fix my unstable
> >> ocean+ice model. While checking the data, I found, however, that
> >> velocity-fields show deviant behavior at the corners of the cube
> >> (I use a cubed sphere configuration). I attached one horizontal
> >> velocity plot, where this behavior is clearly visible. The model
> >> becomes, as expected, unstable in one of these 'corner-regions'
> >> and blows up.
> >>
> >> Could there be something wrong with the coupling between the
> >> tiles? Or do I need a larger overlap (in my case: OLx=Oly=5)?
> >> Do you have any idea what is wrong and how to fix it?
> >
> > _______________________________________________
> > 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