[MITgcm-support] High vertical velocity/noise next to land

[Martin Losch]

Hi David,

I have a few suggestions to your "data" file, some are unrelated to your problems, e.g:
instead of
>  deltaTmom = 2400.,
>  deltaTtracer= 2400.,
>  deltaTClock = 2400.,
>  deltaTfreesurf= 2400.,
you can cuse deltaT = 2400., which sets all other parameters, in case you do not want asynchronous time stepping (makes life a little easier).

AFAIK the CD-scheme is working but not really popular even with its author, so I'd try to avoid this in high resolution experiments.

For high resolution experiments I found it crucial that the horizontal grid has nearly square grid cells because there is only one viscosity parameter for both i and j direction. If dx and dy are very different you'll have a hard time finding the good values for the viscosity parameters. So I'd go for a grid with dy scaled with cos(lat) for a lat-lon-grid (you need to replace dyspacing by a delYFile='yourfile with dy'). Along with this I'd use the non-dimensional viscosity parameters. Instead of viscAh and viscA4 use viscAhgrid and viscA4grid. These are scaled by deltaT and dx*dy so that a value of one for these parameters give the cfl-criterion for viscosity. See pkg/mom_common/mom_calc.F for details and recommendations for values, there are also other parameters that you can use to limit the viscosity. In general you need bi-harmonic for numerical stability.

Further you can try variable viscosity parameterization: Smagorinski or Leith, here you should probably use the SmagD and LeithD variants, too. I found the documentation on this very helpful: 
<http://mitgcm.org/public/r2_manual/latest/online_documents/node85.html>, and the header of the aforementioned mom_calc_visc.F

I am not sure if setting the ivdc_kappa parameter (turns on convection by increased vertical diffusivity) is a good choice for high resolution runs.

hFacMin=.05, might be too small, try 0.1 or even 0.3
For advection, I'd use either 33 (DST3FL) or the undocumented 7 (7th order monotonicity preserving scheme, requires an overlap of >=4), there is also the Prather advection (80), which is very stable (but has restrictions for open boundaries, as far as I know)

Staggered time stepping is (in my experimence absolutely) required for the multi-dimensional advection schemes (e.g., 30, 33, 77, 7)

Hope this helps,

Martin

n Jan 25, 2010, at 12:36 PM, David Munday wrote:

> Dear MITgcm supporters,
> 
> I'm using MITgcm in a narrow sector model (20 degrees longitude by 120 degrees latitude with the southern-most 20 degrees as a re-entrant "circumpolar" channel) and want to use as similar a configuration at as wide a range of resolutions as possible. Unfortunately, I'm getting quite noisy velocity fields at half-degree and finer grid spacing, which most visibly manifests as very strong (~1E-4 m/s) grid-scale oscillations in vertical velocity (the configuration is hydrostatic).
> 
> At my coarsest resolution of 2x2 degrees I get quite strong upwelling (~1E-6 m/s) next to my western boundary, in the first grid cell, and comparable downwelling in the first grid cell next to the eastern and northern boundaries. I don't think this is the Veronis effect, since I'm using GM to do my temperature/salinity dissipation, and my western boundary currents are about 2-3 grid cells wide. At half-degree and finer, these coherent bands break down into grid scale oscillations and the horizontal velocities start to visibly show the noise that's causing the strong vertical velocities. My first thought was that there might be a problem with my boundary conditions, and so I tried freeslip conditions on the walls to no avail.
> 
> I've searched through the support archive to try and get some help and found a number of suggestions that had been made previously. From those discussions I got a few suggestions to try, such as switching to the vector invariant form of the momentum equations, using the Jamart wet point options, highOrderViscosity advection, and making sure that I've got staggerTimeStep set appropriately. I've also fooled around with the different tracer advection schemes. None of that really helped.
> 
> I've had the most success with using a Shaprio filter, although I only know what is said about them in the documentation and I spent last friday fairly blindly fiddling to see if I could get better results. The Shapiro filter causes the vertical velocity to form bands of alternating upwelling/downwelling for 2-3 grid cells next to the boundaries, although there are grid scale oscillations still pleasant. The CD scheme does work to prevent the worst of the noise occurring, but makes the flow very viscous, so much so that I can reduce my viscosity by more than an order of magnitude without causing the solution to explode! Just increasing viscosity, by a factor of ten doesn't help, If I wack up the GM co-efficient to suppress all eddy activity the noise remains in coherent bands...Except now I don't have eddies :(
> 
> Has anyone got some suggestions for what I'm doing wrong? I've attached the data file that I'm using at half degree without any of the things I've tried over the last three weeks included. For anyone that's familiar with it, I tried a carbon copy of PARM01 from the quarter degree set-up in the repository and that didn't improve the noise.
> 
> Thanks for reading my long-winded cry for help, fingers crossed,
> 
> Dave
> 
> <data>
> 
> 
> *****************************************************************************
> 
> Dr. David R. Munday
> University of Oxford
> Atmospheric, Oceanic, and Planetary Physics
> Clarendon Laboratory
> Parks Road
> Oxford
> OX1 3PU
> U.K.                 01865 272 093
> 
> *****************************************************************************
> 
> "But I, being poor, have only my dreams;
> I have spread my dreams under your feet;
> Tread softly because you tread on my dreams."
>     - W.B. Yeats, He Wishes for the Cloths of Heaven -
> 
> *****************************************************************************
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support