Error in processing VAD winds.

[jderber-NOAA]

It was noted that a large number of observations were being eliminated by the setupw routine because there was no observation type. This observation type check was clearly added because of previous problems with this data.

It was found that there was an issue in the read_prepbufr routine because code for vad winds was added in the wrong location. This resulted in many lines containing only zeros to the output file.

The code for processing the vad was moved before the determination of the output line to fix the issue. It was necessary to move a few more lines to make everything work properly. Also one of the checks (there were multiple checks) for a zero observation type was removed from setupw. A check remains, but this condition should not occur unless another error is introduced.

This change results in the removal of many unrealistic observations from the output uv file from prepbufr (when vad winds are used). Because of the existing observation type check analysis results are unchanged.

NOTE: An additional change was made to allow the multiscale ensemble regional test I was using to work (both for the control and modified code). In constants.f90 and gsi_rfv3io_mod.f90 a change was made to allow longer file names to be used for the input ensemble files.

These changes address fixes #615

• Bug fix (non-breaking change which fixes an issue)

These changes have been tested using the regression tests and a single regional case that is using vad winds. All tests are passed with the only differences being in the output number of wind observation after distribution (the processors containing point 0,0 contained a lot of extra observations from the bug) and the output from the removal of the data by the observation type check.

Note because of the removal of many unrealistic observations in the processing, there appears to be a noticeable reduction in run time when vad winds are used.

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• New and existing tests pass with my changes
• Any dependent changes have been merged and published

DUE DATE for this PR is 10/12/2023. If this PR is not merged into develop by this date, the PR will be closed and returned to the developer.

[JacobCarley-NOAA]

I'm good with the changes pending one very minor comment.

[RussTreadon-NOAA]

WCOSS2 ctests
Install develop at 9e5aa09 and jderber-NOAA:vadbug at 335c4b2 on Cactus. Run standard suite of 9 ctests with following results

russ.treadon@clogin08:/lfs/h2/emc/da/noscrub/russ.treadon/git/gsi/pr617/build> ctest -j 9
Test project /lfs/h2/emc/da/noscrub/russ.treadon/git/gsi/pr617/build
    Start 1: global_3dvar
    Start 2: global_4dvar
    Start 3: global_4denvar
    Start 4: hwrf_nmm_d2
    Start 5: hwrf_nmm_d3
    Start 6: rtma
    Start 7: rrfs_3denvar_glbens
    Start 8: netcdf_fv3_regional
    Start 9: global_enkf
1/9 Test #8: netcdf_fv3_regional ..............***Failed  484.51 sec
2/9 Test #9: global_enkf ......................   Passed  488.58 sec
3/9 Test #5: hwrf_nmm_d3 ......................   Passed  492.74 sec
4/9 Test #7: rrfs_3denvar_glbens ..............   Passed  605.94 sec
5/9 Test #4: hwrf_nmm_d2 ......................   Passed  607.86 sec
6/9 Test #6: rtma .............................   Passed  1089.33 sec
7/9 Test #3: global_4denvar ...................   Passed  1324.26 sec
8/9 Test #1: global_3dvar .....................   Passed  1442.52 sec
9/9 Test #2: global_4dvar .....................   Passed  1563.17 sec

The netcdf_fv3_regional failure is due to a timing scalability test

The case has Failed the scalability test.
The slope for the update (1.509785 seconds per node) is less than that for the control (1.542529 seconds per node).

The wall times for the updat (jderber-NOAA:vadbug) and contrl (develop) do not show any anomalous behavior

netcdf_fv3_regional_hiproc_contrl/stdout:The total amount of wall time                        = 63.881463
netcdf_fv3_regional_hiproc_updat/stdout:The total amount of wall time                        = 63.013349
netcdf_fv3_regional_loproc_contrl/stdout:The total amount of wall time                        = 65.423992
netcdf_fv3_regional_loproc_updat/stdout:The total amount of wall time                        = 64.221177

This is not a fatal fail.

Check the nkeep value in fort.202. nkeep is the number of wind observations passed out of read_prepbufr.f90 for processing by setup routines. Find that the updat code consistently reduces the number of observation kept by read_prepbufr

	contrl	updat
global_3dvar	3168472	3147394
global_4denvar	3172060	3150982
netcdf_fv3_regional	323180	321662
rrfs_3denvar_glbens	159146	157394

While the number of kept observations decreases in updat, the number of assimilated observations is the same as contrl. This is consistent with the explanation of the bug and it's fix in this PR.

[JacobCarley-NOAA]

Looks good.

[jderber-NOAA]

Hera ctests

Regression tests were run on Hera earlier. The results appear to be essentially identical to those Russ ran on WCOSS. Even the same test failed for timing reasons. Also, I checked the nkeep values on Hera and they are again identical as would be expected.

[RussTreadon-NOAA]

Two peer reviewers have looked over changes and approved. WCOSS2 (Cactus) and Hera ctests pass with expected results.

Approve.