[MITgcm-support] 2deg global experiment: advection schemes

[Samar Khatiwala]

Hi Martin

You may want to reread the threads I started on advection schemes  
(some time
in Sept 2004 and again in April 2005 I think). JMC had some useful  
recommendations
to make, notably the use of staggerTimeStep when using a forward-in- 
time method
(like the DSTXX schemes). The reason has to do with stability due to  
internal waves.
In a more idealized setting, I too got dramatically different  
circulations (more so than
yours) with different advection schemes.

To quote a well known expert on the MITgcm:

"Samar,
beats me, but the different advection schemes can be dramatically  
different in their result (have a look at the documentation http:// 
mitgcm.org/sealion/online_documents/node80.html ).
Martin"

Samar

On Nov 7, 2005, at 6:47 AM, Martin Losch wrote:

> Hi there,
>
> I would like to share my experience with this (and maybe get some  
> general advice):
>
> I am running a 2deg global ocean model (much like the 2deg ECCO  
> configuration, that is, from -80 to 80 deg, 23 layers, GM, KPP) and  
> I make the interesting observation, that the solution is very  
> sensitive to the advection scheme (and the parameter  
> staggerTimeStep) that I use.
> I have reason to believe that the default 2nd-order scheme is  
> inadequate for my purposes (I restore T and S along the southern  
> boundary in order to model the effects of the Antartic shelf ice  
> caverns and thus get sharp gradients there, that lead to weird T  
> and S properties near the southern boundary), so I tried 3rd-order  
> upwind, DST3 with an without flux limiter, and 2nd-order with flux  
> limiter.
> First of all, when staggerTimeStep=.false. (default), then these  
> advection schemes (together with a set of GM, diffusivitity,  
> viscosity parameters that produced a reasonable circulation with  
> the default 2nd-order scheme: 133Sv ACC, 20Sv MOC, etc) give me a  
> dramatically intensified circulation: ACC order 400-500Sv, etc.  
> watermasses change drastically (not the patterns, just the  
> horizontal and vertical gradients, which explains the stronger  
> circulation).
> Second, staggerTimeStep=.true. brings down the values to a  
> realistic 126-130Sv (and realistic MOC, etc), except for DST3  
> without flux limiter, where the ACC is still O(250Sv).
>
> 2 years ago, Jean-Michel recommended to use the staggerTimeStep  
> flag with higher order schemes, so I am not so surprised that this  
> option salvages most of the runs, but still I am surprised that  
> these advection schemes have such a fundamental impact on the  
> solution.
> I had expected that the additional diffusion introduced by these  
> schemes (in particular the flux limited schemes) would actually  
> reduce the circulation (as they mostly do with  
> staggerTimeStep=.true.), but not increase it.
>
> Was I wrong to keep the GM-diffusivity and viscosity parameters  
> while switching the advection shemes?
>
> Martin
>
> Martin Losch // Alfred-Wegener-Institut für Polar- und Meeresforschung
> Postfach 120161, 27515 Bremerhaven, Germany
> mailto:mlosch at awi-bremerhaven.de // Tel./Fax: ++49(471)4831-1872/1797
> http://www.awi-bremerhaven.de/People/show?mlosch
>
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support
>