[MITgcm-support] Re: OBCS

Martin Losch mlosch at awi-bremerhaven.de
Wed Oct 27 09:13:36 EDT 2004


Hi again,

Alistair, I am not using NLFS nor z* (but it's good to know that it 
works with your fix), and I get a nice solution in the interior for my 
4-open-boundaries box. However w is an order of magnitude larger for 
the northern corner points, that is: (1,-1), (-1,-1), or 
(1,sNy),(sNx,sNy). Do you see something similar? (it's a problem for 
me, because the passive tracers see everythings a periodic and are 
advected by the values on the boundaries/corners as well).

Martin

On Oct 26, 2004, at 3:18 PM, Alistair Adcroft wrote:

> I've recently been sorting out the interaction between OBCs and NLFS. 
> There is a missing OBCS_APPLY_ETA:
>
> RCS file: /u/gcmpack/MITgcm/model/src/integr_continuity.F,v
> retrieving revision 1.9
> diff -r1.9 integr_continuity.F
> 168a169,171
> >        IF ( useOBCS )
> >      &    CALL OBCS_APPLY_ETA( bi, bj, etaN, myThid )
> >
>
> which I will check in shortly but otherwise everything does work. 
> There are no missing exchanges - this must have been fixed by Patrick.
>
> With the above patch, NLFS and z* work with OBCS at 1 and -1 (i.e. the 
> first points in the domain) which I prefer and there are no rediculous 
> values coming out of the pressure solver. Note that topography must 
> always be flat at OBCs (i.e. at H(I_OBW)=H(OWB+1)). There is no need 
> to have a false wall in topography when you put OBCs at 1,-1; the OBCs 
> override any periodicity that would otherwise appear. I haven't 
> checked but a false wall might be necesary if the OBCs are interior to 
> the computational domain (e.g. at 2,-2).
>
> A.
> Martin Losch wrote:
>> Hi Jake,
>> thanks for your answer.
>> I do have the open boundaries on the outermost grid-points (so only 
>> 1's and -1's in the data.obcs), so that the corners are part of both 
>> open boundaries (both zonal and meridional ones). The physical 
>> solution indeed does not look affected by these large |w|-values in 
>> the corners. I don't use the CD-scheme. I think all exchanges are now 
>> correct in the current code. At least I have run an experiment with 4 
>> tile vs 1 tile and they all gave the same results.
>> I do use the passive tracer package, though, for which there are no 
>> open boundaries implemented. I am worried about the passive tracers 
>> being affected by the strange w-values on the boundaries. But I guess 
>> my real problem is the fact, that passive tracers are treated with 
>> periodic boundaries while U,V,T,S are not. So I'll have to do 
>> something about that.
>> Martin
>> On Oct 25, 2004, at 3:30 PM, Jake Gebbie wrote:
>>>
>>> Hi Martin:  My memory is fading at this point, but here's my 
>>> assessment.
>>>
>>> If you're setting the o.b. points to be the outermost grid-points in 
>>> the
>>> domain, then the vertical velocity field (or sea surface height) at 
>>> those
>>> points don't affect the interior solution. In other words, those 
>>> points
>>> don't get used in the global pressure solver. But the large or 
>>> strange
>>> values should be only on the open boundaries and not in the interior.
>>>
>>> So, to be clear, what are the coordinates of the "corner" versus the 
>>> open
>>> boundary coordinates?
>>>
>>> One other idea: I believe the CD scheme and open boundaries are
>>> incompatible (can anyone else back me up?). Is CD turned off?
>>>
>>> Dan Lea has also found that there was a missing exchange in the OB
>>> routines. Hopefully, it has found its way into the main trunk.
>>>
>>> Cheers,
>>> --Jake
>>>
>>>
>>>
>>>
>>>
>>> On Mon, 25 Oct 2004, Martin Losch wrote:
>>>
>>>> Hi,
>>>>
>>>> I have made an obseration  with the OBCS package, and I can't quite
>>>> figure out what it is, mayby someone can help:
>>>>
>>>> I have a box with 4 open boundaries, where I prescribe U and V (and 
>>>> T
>>>> and S). From looking at obcs_apply_uv (which is called from dynamics
>>>> and momentum_correction_step, so before integr_continuity), I would
>>>> think that u(1,:) = u(1-1,:)=u_obcs(east),
>>>> u(nx,:)=u(nx+1,:)=u_obcs(west), v(:,1)=v(:,1-1)=u_obcs(south), and
>>>> v(:,ny)=v(:,ny+1)=v_obcs(north). So at the corners, dw/dz = 
>>>> du/dx+dv/dy
>>>> = 0 and w(z) = const.
>>>> But that's not what I observe. There is a vertical profile (nearly
>>>> linear in 3 of 4 corners). This w a the corners is more or less
>>>> consistent (smooth transition) with surrounding points at the 
>>>> southern
>>>> points (iy=1), but very singular at the at the northern points
>>>> (iy=sNy). In fact, w is large (factor 100 larger than interior) and
>>>> positive at (ix=1,iy=sNy) and large and negative at (ix=sNx,iy=sNy).
>>>>
>>>> This looks very much like a bug to me, but before I spend too much 
>>>> time
>>>> on hunting it down, I would like to know if anyone has made similar
>>>> observations. Maybe there's a simple explanation, why at these 
>>>> corners
>>>> w has to be so strange ...
>>>>
>>>> Martin
>>>>
>> _______________________________________________
>> MITgcm-support mailing list
>> MITgcm-support at mitgcm.org
>> http://dev.mitgcm.org/mailman/listinfo/mitgcm-support
>
>
> -- 
> Dr Alistair Adcroft            http://www.mit.edu/~adcroft
> MIT Climate Modeling Initiative        tel: (617) 253-5938
> EAPS 54-1624,  77 Massachusetts Ave,  Cambridge,  MA,  USA
>
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://dev.mitgcm.org/mailman/listinfo/mitgcm-support




More information about the MITgcm-support mailing list