release/gfsda.v16.30 updates

[RussTreadon-NOAA]

This issue is opened to document two updates in forked branch RussTreadon-NOAA/GSI/release/gfsda.v16.3.0:

1. NCO requested that GFS v16.3.0 use bufr/11.7.0
2. Tests of gsi.x compiled with -check all seg faulted in read_prepbufr.f90 due to VADWND quality control

bufr/11.7.0
The GSI references bufr routine isetprm which exercise dynamic allocation. When gsi.x is built using non-DA versions of bufr/11.5.0 or lower, the following run time message is generated

 ++++++++++++++++++++WARNING+++++++++++++++++++++
 BUFRLIB: ISETPRM - DYNAMIC MEMORY ALLOCATION IS NOT SUPPORTED FOR THIS BUILD OF BUFRLIB
 ++++++++++++++++++++WARNING+++++++++++++++++++++

bufr/11.6.0 and above include dynamic allocation in the standard modules. Tests thus far indicate that moving to bufr/11.7.0 does not alter analysis results. To build gsi.x with bufr/11.7.0, the default bufr version in modulefiles/gsi_common.lua is changed from 11.5.0 to 11.7.0. With the switch to bufr/11.7.0 the above warning message is not generated at run time.

VADWND quality control
The GFS v16.3.0 gsi.x was built with the -check all compiler option. NCO performs this test in their pre-implementation IT testing. The 2021121400 gdas was run on Cactus. gsi.x aborts in read_prepbufr.f90 when processing observation type kx=224 (VADWND) with newvad=.true.

The section in read_prepbufr.f90 which loads the cdata_all array for uv observations includes VADWND quality control.

                 !* thin new VAD wind and generate VAD superob                                                                           
                 if(kx==224.and.newvad)then
                         klev=k+5 !*average over 6 points                                                                                
                       !  klev=k    !* no average                                                                                        
                         if(klev>levs) cycle loop_readsb
                         diffuu=obsdat(5,k)-fcstdat(1,k)
                         diffvv=obsdat(6,k)-fcstdat(2,k)
                         if(sqrt(diffuu**2+diffvv**2)>10.0_r_kind) cycle loop_k_levs

Quality control tests if .true. cause program execution to jump to the top of loop_k_levs or loop_readsb. This is problematic.

The observation counter iout is incremented prior to these quality control tests. If VADWND quality control tiggers a cycle, the observation counter iout is not decremented. As a result, there are iout for which cdata_all is not loaded with actual data. In this case cdata_all(:,iout) retains the 0.0 to which the entire array was initialized. This, in turn, causes the -check all gsi.x to fail when it executes the following block later on in read_prepbufr.f90

        do k=1,ndata
           ikx=nint(cdata_out(10,k))
           itype=ictype(ikx)
           if( itype ==230 .or. itype ==231 .or. itype ==233) then
              prest=r10*exp(cdata_out(4,k))

cdata_out(10,k)=0.0 so ikx=0. Array element ictype(0) is out of bounds. Execution aborts.

A check of read_prepbufr.f90 shows that for all obstype except uv, cycle_loop_k_levs and cycle loop_readsb occur before iout is incremented. Only kx==224.and.newvad has cycle lines after iout is incremented.

The following short term patch is proposed

        do k=1,ndata
           ikx=nint(cdata_out(10,k))
           if (ikx>0) then
              itype=ictype(ikx)
           else
              itype=0
           endif
           if( itype ==230 .or. itype ==231 .or. itype ==233) then
              prest=r10*exp(cdata_out(4,k))

Tests of this patch have shown that it does not alter analysis results.

[RussTreadon-NOAA]

getsigensmeanp_smooth_ncep.f90 bug

Examination of parallel output found that getsigensmeanp_smooth.x is NOT generating atmfXXX ensemble spread files. This is due to a bug reported in GSI issue #419. PR #420 merged the bug fix into develop on June 23, 2022. This bug fix did not make it into release/gfsda.v16.3.0

The one character bug fix was added to a working copy of release/gfsda.v16.3.0 getsigensmeanp_smooth_ncep.f90. The code was recompiled and run. The atmfXXX ensemble spread files were created. This fix was committed to the forked release/gfsda.v16.3.0 at 314736c.

[RussTreadon-NOAA]

Close this issue with merge of PR #475 into authoritative release/gfsda.v16.3.0