[MITgcm-support] exactConserv & OBCs ?

[Jean-Michel Campin]

Hi Alistair,

I have noticed also (in exp4 set-up) that when I turn on exactConserv,
because the same velocity is imposed on 2 consecutive rows at the OB,
the gradient of bathymetry in the OB region results in a different
barotropic transport between those 2 rows of points, and the eta field
starts to change, but only in the "dead" region of the domain (= excluding
the part that is outside the OB).
This has no impact on the solution, and the 1.e-11 difference looks
OK for me.

To use the non-linear free-surface with OBCs, you need to specify the
velocity and the free-surface position at the OBs
(OBN,S,E,Weta in OBCS.h, obcs_apply_eta.F ...),
and the free-surface elevation is then only used to update the model
geometry (hFacC,W,S) but is still not used to compute pressure
gradient at the OB. I think it make sense, because this way, you
control the input-output flow through the OB.
It happens also that those unphysical values of etaN in the "dead" 
region of the domain were causing a lot of warnings, so that I decided
to apply the OBN,S,E,Weta also to the eta field in the "dead" region 
of the domain, without any impact on the solution (at least I hope so).

Now, with NLFS, I think that if you don't read-in or compute those 
OBN,S,E,Weta with NLFS, they are simply set to zero (in obcs_calc.F).
So, even in this case, you totally control the input-output flow at the
OB, base on the assumption that etaH==0 at the OB.

I remember that I did some test, but cannot guarantee that there 
is not a small, well hidden problem somewhere.

The part that is still missing is the Orlanski with NLFS, 
but I don't see an immediate need until we get a well-tested
barotropic component of the Orlanski scheme.

Jean-Michel