Omit "Rejected: x" observation diagnostic output if BGQC is undefined.

[stef]

I'm using strong constraint I4DVAR along with RBL4DVAR+RPCG, although I know I4DVAR is deprecated, but I'm using it to learn because it's easier to understand. I have only tried to understand the I4DVAR code so far, but will start with RBL4DVAR soon.

I just noticed that I always forgot to #define BGQC. The reason I forgot is because in the output it always says:

Code: Select all

          Variable          IstrObs    IendObs      Count   Rejected

          temp                    1         19         19          1
          salt                   20         38         19          1

          Total                                        38          2
          Obs Tally                                    38          2

Note the "Rejected: 1" columns

This output suggests to me that these data have been excluded from the computation, but they are not unless BGQC is defined.

Now I looked at the first inner loop results of the adjoint before the conversion to v-space, and finally realized that the values don't make sense.

The reason it took me so long is because the RBL4DVAR results look fine, even if I don't define BGQC. I looked at ad_misfit.F (I4DVAR) and ad_htobs.F (RBL4DVAR) and both contain something like

Code: Select all

# ifdef BGQC
!
!  Reject observation that fail background quality control check.
!
        DO iobs=Mstr,Mend
          ADmodVal(iobs)=ObsScale(iobs)*ADmodVal(iobs)
        END DO
# endif

In my case, one (T,S) data point was bad, because I located it exactly on the bottom-most rho-point. But apparently it only wreaked havoc for I4DVAR, not for RBL4DVAR. When I switch on BGQC, it's fine in both.

Edit:
see also here:
viewtopic.php?t=6535