[MITgcm-support] OBC problem: spurious boundary jets with C-D coupling

[chris hill]

Hi Mark,

  As Martin notes the code for CD with OBCS has not been written.
  What resolution are you working at? Mostly CD is used for situations 
where the deformation radius is not resolved i.e. resolutions > 25km. 
For a limited area run it may not be needed. For resolutions where eddy 
scale is resolved or at least things are eddy permitting then sub-grid 
dissipation is more cleanly (although its still a dirty business) 
handled by either fixed laplacian and biharmonic viscosities or by the 
flow dependent smag/leith form (described in the online docs).

CHris
Martin Losch wrote:
> Hi Mark,
> 
> the CD scheme does not work with OBCS. I am sorry that you had to find 
> this out the hard way. I don't even think that this is documented 
> anywhere, except for some email in the support-archives. Prescribing 
> values for the D-grid velocities along open boundaries is simply not 
> implemented although it should not be so difficult: in obcs_apply_uv.F 
> one could easily set the uVelD and vVelD to appropriate values, that may 
> already be enough. One would have to sort out how to average/extrapolate 
> consistently.
> 
> For some reason, the CD-scheme is not seem to be very popular any more 
> (although it does effectively reduce grid scale noise and I use it 
> myself). One of the reasons may be that the couple constant tauCD is a 
> tunable parameter whose value is not pre-determined. If  
> tauCD=deltaT(mom), then the CD-scheme does not have any effect (then rCD 
> = 1-deltaTmom/tauCD = 0 and uVelD=uVel, see pkg/cd_code/cd_code_scheme.F 
> and model/src/ini_parms.F). tauCD should be something larger than 
> deltaT(mom), the large tauCD, the stronger the coupling between C and D 
> grid velocities and the smoother and less energetic are the results.
> 
> What can you do to reduce the grid scale noise: viscosity, in particular 
> bi-harmonic viscosity, and use StaggerTimeStep=.true. (that was 
> suggested to me once, but I haven't really seen it to have too much 
> effect). Maybe Jean-Michel has a better suggestion?
> 
> For the linear free surface the boundary values for eta are always zero. 
> If you want to change that, you'll have to use the nonlinear free 
> surface (turn on by enabling the flag in CPP_OPTIONS.h and setting some 
> parameters in data, see e.g., global_ocean.cs32x15), but that means also 
> prescribing the free surface along the boundaries.
> 
> Martin
> PS. Maybe I'll have a look at CD-scheme+OBCS in the future, as it bugs 
> me, too. Unless there is a principle problem that I have overlooked, it 
> shouldn't be too difficult. (but feel free to try it out yourself and 
> contribute your code  (o:)
> 
> On Sep 27, 2006, at 6:28 AM, Mark Hadfield wrote:
> 
>> I am attempting to apply MITgcm to limited-area ocean simulations 
>> around New Zealand. (I am also collaborating with Jill Schwarz in her 
>> Ross Sea work.) My first attempt, loosely based on the lab_sea case, 
>> was spectacularly unsuccessful. Within a few hours of starting from 
>> rest, narrow jets developed along the open boundaries. Within a day 
>> the velocities had developed to several m/s, at which point the model 
>> crashed with large Eta values where the jets impinged on topography.
>>
>> To cut a long story short, I have reproduced this problem in a minimal 
>> test case (below) and established that it occurs when the CD scheme is 
>> enabled (useCDscheme=.true.) AND the CD scheme coupling time scale, 
>> tauCD, is set to a value other than its default (which is equal to the 
>> momentum time step, deltaTMom). My first NZ region simulation is now 
>> chugging along with the CD scheme disabled and giving sensible 
>> results. (Unfortunately, it is developing some grid-scale noise in the 
>> velocity field; I am controlling that with viscosity.)
>>
>> By the way, does leaving tauCD at its default actually have the effect 
>> of disabling the CD coupling? The results I have seen suggest this is 
>> the case, but mitgcm.org won't let me look at the documentation today 
>> to confirm.
>>
>> For now, it seems that the CD scheme and OBCs are incompatible and the 
>> simplest workaround is to disable the CD scheme. I don't know how much 
>> of a limitation that will be for me. As others have noticed on this 
>> list, the model leaves the outermost Eta values at zero. I presume the 
>> CD scheme is accessing those zero values, whereas the ordinary C-grid 
>> scheme is not. Might this be solved simply with a zero-gradient 
>> boundary condition on Eta?
>>
>> In case anyone's interested, the files for the test case are in
>> ftp://ftp.niwa.co.nz/incoming/hadfieldm/dmf/work/channel/mitgcm/run01/
>> The domain is a 600 km x 600 km wide and 2000 m deep, with open 
>> boundaries on all 4 sides. Coriolis parameter f is set to -1.0E-4 
>> (this is the southern hemisphere) and beta is zero. Model grid 
>> dimensions are 40 x 40 x 1. At t = 0, the interior is at rest and a 
>> zonal jet is imposed at the western and eastern boundaries. The jet is 
>> in geostrophic balance with a tanh-shaped step in sea surface height, 
>> eta:
>> eta = Z tanh(y/D)
>> where y is N-S position relative to the centre of the channel, D is 
>> the transition zone half-width (60 km) and Z is 0.2 m. So the jet 
>> velocity is
>> ubar = -(g Z)/(f D cosh(y/D)^2)
>> where g is gravitational acceleration and f is Coriolis parameter. 
>> Peak velocity is about 0.32 m/s. (Note that eta is given here for 
>> completeness but eta data are not supplied to the model.)
>>
>> To simplify things, we solve for the velocity field only, without 
>> advection or diffusion:
>> saltStepping = .FALSE.
>> tempStepping = .FALSE.
>> momStepping = .TRUE.
>> momAdvection = .FALSE.
>> momViscosity = .FALSE.
>> In the first run...
>> ftp://ftp.niwa.co.nz/incoming/hadfieldm/dmf/work/channel/mitgcm/run01/runa/ 
>>
>> the CD scheme is enabled and tauCD is left at its default. The 
>> velocity field adjusts in less than one day to a steady state: the jet 
>> spreads more or less radially from its source at the western boundary, 
>> occupies the whole width of the channel in the centre, and converges 
>> (is confluent?) to its sink on the eastern boundary. The expected 
>> symmetry is maintained/ The difference in Eta across the jet is what 
>> you would expect from geostrophy, except that Eta at the exterior 
>> points stays zero.
>>
>> In the second run...
>> ftp://ftp.niwa.co.nz/incoming/hadfieldm/dmf/work/channel/mitgcm/run01/runb/ 
>>
>> the tauCD is set to twice deltaTmom. Here westward jets develop next 
>> to the northern and southern boundaries. These jets continue to 
>> develop over the duration of the simulation (10 d) and by the end the 
>> velocity field is dominated by a pair of gyres, with westward 
>> velocities of ~ 2 m/s at the boundaries and an eastward return flow of 
>> ~ 1 m/s in the centre.
>>
>>
>> -- Mark Hadfield "Kei puwaha te tai nei, Hoea tahi tatou" 
>> m.hadfield at niwa.co.nz National Institute for Water and Atmospheric 
>> Research (NIWA)
>> _______________________________________________
>> 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
>