[MITgcm-devel] OBCS_SPONGE vs RBCS

[Jean-Michel Campin]

Hi,

I started to look at OBCS_SPONGE code (see list of comments
below), and would like to propose to drop OBCS_SPONGE
and to add RBCS capability for U & V.
The main reason is that point 1 & 2 (and 6 ?) are inherent
to our obcs-sponge code, and RBCS would solve both in a clean
and reliable way.
What is your opinion on this subject ?

Cheers,
Jean-Michel

---------------------------------
There are several reasons that make the OBCS_SPONGE code
difficult to use and to maintain:

1) relaxation towards OB value is tricky when the bathymetry
is not flat in all the sponge layer region + OB section.
In the current Sponge layer, on a wet point, we relax
towards the OB values which might be in land.

2) when 2 OB (e.g., North and West) are close enough to
each other, the 2 sponge layers have an overlap.
And it's become tricky (for a numerical point of view)
to handle correctly 2 relaxation terms towards
different values.

3) there are few run-time parameters that are missing:
- a set of "useSpongeNorth/South/East/West" (can help with point 2).
- a set of "useSpongeT,S,U,V" (This is a problem when
  someone wants to use RBCS for T,S and Sponge for U,V).
And it would be a good thing to have the option to specify
different relaxation time scales for U,V and T,S.

4) OBCS-sponge is not tested (partly because of point 3) .

5) The code and documentation need improvements:
- the U and V in UrelaxObcsBound/Inner and VrelaxObcsBound/Inner
  are not related to U/V velocity (as described in OBCS.h) but
  related to East/West and North/South OB.
- the thickness of the sponge layer, more precisely, the number of
  points where relaxation is applied is not "spongeThickness"
  but "spongeThickness - 1".
- the 2 extreme values for relaxation time scale, XrelaxObcsInner
  and XrelaxObcsBound are never reached; Instead, the relaxation
  time scale is always an interpolated value between those 2
  extremes.
- in obcs_sponge.F, lot of single precision computations (results
  will change with compiler "-r8" option); not efficient (all relaxation
  time-scale calculations could be done once); not very robust
  (testing for zero value of Temp & Salt OB values).

6) I don't know if the OBCS_SPONGE code is right when using
  CS-grid with OBC that run over on several faces (with switch
  between N/S and E/W). I did not check, but the fact that we
  need to EXCH (gU,gV) is not a good sign.

That's all for now.
---------------------------------