[MITgcm-support] SSH drift and obcs again

[Thomas Haine]

Hi Martin, Chris,

OK, I understand that bcs revert to periodic when they're unspecified
and there's no wall in the bathy file. But then why does my case i)
work?  E/W obcs are unspecified here and there is an open periodic
boundary in the east/west direction too.

Also, when I set the corner obc speeds as you suggest:

a) Case i) before still works the same. No surprise.

b) With all 4 obcs open (the case I'm really interested in): the source
term is O(1) again and eta is drifting but it integrates for many steps
OK. Why is there a cg2d source?

If I remove the corner obc points altogether in data.obcs (i.e. obcs
from 2 -> NX-1, not 1 -> NX etc) I get similar results to b) except it
dies after 12 steps (w overflow).

More generally, I don't understand how obcs_apply_uv.F is supposed to
work. Why is the mask shifted relative to the grid in the 2nd obc
assignment, e.g.:

        vFld(I,OB_Js(I,bi,bj)+1,K,bi,bj)=OBSv(I,K,bi,bj)
     &  *_maskS(I,OB_Js(I,bi,bj)+1,K,bi,bj)

        vFld(I,OB_Js(I,bi,bj),K,bi,bj)=OBSv(I,K,bi,bj)
     &  *_maskS(I,OB_Js(I,bi,bj)+1,K,bi,bj)

Is this consistent with solve_for_pressure.F where cg2d_b and cg2d_x are
set at the obcs points themselves, not 1 step in on the S/W bdy, or at
multiple points as in obcs_apply_uv?  I can't see how.

Tom.