[MITgcm-support] seaice model

[Martin Losch]

Hey,
I think I found the problem! Pointing at timestepping really helped, 
Dimitris!

The pkg/seaice model uses consistently DELTAT for the timestep, except 
when surfaceTendencyTice is computed, where it uses deltaTtracer (which 
is correct, as far as I can see).
BUT ... I would think, since we are dealing with the thermodynamic 
properties, ALL DELTAT should be replaced by deltaTtracer, except for 
the advective processes, where it should be deltaTmom (which sets 
deltat) (I don't know about the dynamic solver, I guess you can control 
the timestep by setting npseudo, but I don't understand yet, how that 
relates to model time). I did this in growth.F, that is, replaced all 
DELTAT with deltaTtracer and now the model is stable and gives 
reasonable results!
I can only guess that the reason why this has not been a problem, is 
that asynchronous timestepping was not used so far? Or at least not 
with such a big timestep as in global_ocean.cs32x15 or 
global_ocean.90x40x15.

I cannot oversee, where exactly the deltaT should be replaced, but I am 
almost positive, that it should be done everywhere in growth.F.

What do you think?

Martin

On Thursday, April 29, 2004, at 12:12 PM, Martin Losch wrote:

> Dimitris,
>
> BTW, I think I have been CC-ing all of my emails to Jinlun, haven't I?
>
> On Wednesday, April 28, 2004, at 03:49 PM, Dimitris Menemenlis wrote:
>
> The CS-grid is fine, but I cannot think of any reason, why the 
> seaice-pkg shouldn't work on the global_ocean.90x40x15 grid. I mean, I 
> don't expect any realism (without an Artic basin nor Ross nor Wedell 
> Sea), but it should be stable; I would like to understand these 
> extreme growth rates (which lead to the high salinities), and why they 
> occur. Maybe I should start sending plot. These extreme ice 
> thicknesses occur only in singular grid cells (admittedly 4x4 degrees 
> wide).
>
>>> But the problems I have there are the same, plus, with the cubed 
>>> sphere
>>> grid I no longer know what I am doing (o: But essentially, the same
>>
>> There's a bunch of matlab and other diagnostics for looking at stuff.
>> I think it's worth the effort getting used to this grid ... eventually
>> you will get hooked and become an addict like the rest of us, except
>> for Alistair who leaves us to swallow his dust and is always moving to
>> newer and greener pastures.
> As you saw from my last email to support, I already demonstrated my 
> greenhornishness.
>
>>> thing happens as with lat-lon grid. Ice thicknesses become very 
>>> large,
>>> huge freshwater flux out of the ocean, strong horizontal divergences 
>>> in
>>> the ocean, large vertical velocities and violation of the vertical
>>> cfl-criterion. If I reduce the timestep even further, the model runs
>>> longer but eventually it blows up, too (all after less than a year of
>>> integration).
>>
>> I was looking back at some preliminary comparisons that I did
>> of pkg/thsice and of pkg/seaice, with and without dynamics
>> in the 32x32x6x15 config.  All three integrations ran fine
>> for five years and there does not seem to be any anomalous
>> behavior.  But I am integrating all three with 1 hour time
>> steps.   So maybe it's just a time step issue?
>
> This may very well be that case: Certainly, with a 
> tracerTimeStep=3600s, you won't run into the CFL problems, that I 
> have. But I only have these problems with useSeaIce=.TRUE., with out 
> seaice I can use a long tracer timestep of 86400s (172800s in the 
> global_ocean.90x40x15 experiment) so which part of the seaice-pkg 
> needs the small timestep?
> I am now refering to the global_ocean.cs32x15 configuration:
> Sea ice dynamics:
> - with SEAICEuseDynamics=.false. I have the same problems as before,
> - with npseudo = 24 which for tracerTimeStep=86400 effectively means 
> SEAICE_DT=3600s, I have the same problems
> - I also turned off the advection (not turned of the the flag 
> SEAICEuseDYNAMCICS, which BTW uses the same timestep as the 
> thermodynamics. In the context of asynchronous time stepping this 
> should maybe be the momentum timestep deltaTmom? I don't know. Anyway, 
> turning off advection does no help either (doesn't surprise me because 
> U/VICE should be zero if SEAICEuseDYNAMICS=.false.)
>
> This leaves me to believe it has to do with the thermodynamic part. I 
> also ran the model with the thsice-pkg, and that does not appear to 
> need the short time step.
>
> Martin
>
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://dev.mitgcm.org/mailman/listinfo/mitgcm-support