[MITgcm-devel] changes in OBCs pkg

[Jean-Michel Campin]

Hi Martin,

Sorry, I did not take the time to write all the things down:

1) Yes, when you switched from exp4/code/obcs_calc.F to
useOBCSprescribe=.TRUE., the results did not change in
the interior.
However, if I compare MOM dynstat_wvel_max in output.txt,
now I have (e.g., iter=10):
(PID.TID 0000.0001) %MON dynstat_wvel_max             =   2.2495491745397E-01
and before (same iteration):
(PID.TID 0000.0001) %MON dynstat_wvel_max             =   2.1847855594095E-02

2) It took me some time to figure out why we get those
large vertical velocity just in 2 grid points (in fact
2 columns of grid points), in (40,1) & (40,42):
It's because of OBCS_APPLY_UV call after CORRECTION_STEP
(all of this in momentum_correction_step.F) that set u
at N & S OBs, from 1-Olx:sNx+Olx:
with old obcs_calc.F, OBSu is set to .25 over the full
range but with useOBCSprescribe, only from 1:sNx.
The consequence is that uVel(41,1) (tile.1) is reset to
zero whereas uVel(40,1) (tile.1) is 0.25, resulting in
a large divergence; and since wVel is computed just
after (in INTEGR_CONTINUITY) (and also etaN if using exactConserv),
we get large wVel (and large etaN) at those 2 points.

Note that, strictly speaking, it's not related to an initialisation
Pb, because uVel(40,1) & uVel(41,1) are reset to OBSu several
times per iteration.

3) what I proposes:
 
> I do acknowledge that there are always problems with
> vertical velocities near the boundaires. What do you think?

Well, this is the reason why I would like to reset etaN & wVel
to zero in the OB regions, so that
a) when you plot wVel, you don't see those very weird values
 in the OB regions
b) same thing, but worst with etaN & exactConserv, because it
 accumulates (increase by 150.m /iteration).
c) since it is quiet easy to get something wrong with OBCs, it's
 better not to leave weird wVel & etaN values in the OB regions,
 that would let someone think that something is wrong.

The point is that this problem exists (and was there with
exp4/code/obcs_calc.F) but it just get worst with
useOBCSprescribe, as I explain above.

Hope it's more clear now.

Cheers,

Jean-Michel