search

NOAA-EMC / GSI

Gridpoint Statistical Interpolation
GNU Lesser General Public License v3.0
66 stars 151 forks source link

Recover the capability of handling model fields from operation gfs.v16.3 #785

Closed emilyhcliu closed 3 months ago

emilyhcliu commented 3 months ago

Description Make sure GSI can read the background fields from the operational gfs.v16.3 Please see details in issue #782 Resolves #782

I fixed the indentations. GitHub messes up (confuses) the code indentation change.
So, to clarify, my code changes are:

So, the real code change is just two lines (see the following link). The rest of the changes are related to indentation fixes.
https://github.com/NOAA-EMC/GSI/commit/3404ddfef03fd52c1fa04411d6613e15b7ef77da?diff=split&w=1

Type of change

How Has This Been Tested?

The branch was tested on ORION using two sets of ICs on 2024021900:

The updated code can handle background fields from both gfs.v16.3 and v17. The updated code does not change the results of regression tests.

Checklist

emilyhcliu commented 3 months ago

Working on regression tests. Will post results here soon.

emilyhcliu commented 3 months ago

Results of Regression Test on WCOSS-2:

    Start 1: global_4denvar

1/6 Test #1: global_4denvar ...................   Passed  1685.96 sec
    Start 2: rtma
2/6 Test #2: rtma .............................   Passed  970.86 sec
    Start 3: rrfs_3denvar_rdasens
3/6 Test #3: rrfs_3denvar_rdasens .............   Passed  729.70 sec
    Start 4: hafs_4denvar_glbens
4/6 Test #4: hafs_4denvar_glbens ..............   Passed  1334.85 sec
    Start 5: hafs_3denvar_hybens
5/6 Test #5: hafs_3denvar_hybens ..............   Passed  1335.36 sec
    Start 6: global_enkf
6/6 Test #6: global_enkf ......................   Passed  853.77 sec
emilyhcliu commented 3 months ago

@RussTreadon-NOAA Please add @ADCollard and @xincjin-NOAA to review this PR. Thank you.

emilyhcliu commented 3 months ago

@RussTreadon-NOAA Please add @ADCollard and @xincjin-NOAA to review this PR. Thank you.

@RussTreadon-NOAA Never mind. I just found out that I can assign reviewers myself. I will do that.

emilyhcliu commented 3 months ago

Changes look good to me. BTW, imp_physics=8 is for Thompson scheme, what value is it for the current GFDL microphysics?

The current imp_physics is 11 for GFDL Thanks for reviewing this PR.

emilyhcliu commented 3 months ago

Looks good. Confirm ctest results on Cactus. Also run stand-alone GSI script using operational backgrounds and imp_physics=11. gsi.x ran to completion. Repeat test with imp_physics=8. gsi.x hangs with

rank of data array != variable ndims (or ndims-1)

Something to consider: It is preferable for gsi.x to

  • print an informative error message explaining why the code is hung (e.g., _background inconsistent with impphysics)
  • abort execution

instead of hanging when imp_physics does not match what is in the background atmfXXX.nc file.

As things currently stand, it is not clear from the printout

nid001009.cactus.wcoss2.ncep.noaa.gov 14:  rank of data array != variable ndims (or ndims-1)
nid001009.cactus.wcoss2.ncep.noaa.gov 14: 99
nid001098.cactus.wcoss2.ncep.noaa.gov 44:  rank of data array != variable ndims (or ndims-1)
nid001109.cactus.wcoss2.ncep.noaa.gov 74:  rank of data array != variable ndims (or ndims-1)

why the code is hung.

Not sure if NCO's IT testing will catch this or not.

This is a good suggestion. I will improve the error handling. Thanks for looking into this.

RussTreadon-NOAA commented 3 months ago

@emilyhcliu , it may be easier said than done when it comes to adding code to cleanly abort when imp_physics does not align with the background atmfXXX.nc files. The rank of data array != variable ndims (or ndims-1) error message comes from ncio. The read_vardata calls for nccice and nconrd occur inside subcommunicator blocks. Not all tasks will see the error message. Thus, mpi_abort for mpi_comm_world will hang.

If you see an easy way to obtain the desire behavior, fine ... add it. If not, let's move on.

emilyhcliu commented 3 months ago

@emilyhcliu , it may be easier said than done when it comes to adding code to cleanly abort when imp_physics does not align with the background atmfXXX.nc files. The rank of data array != variable ndims (or ndims-1) error message comes from ncio. The read_vardata calls for nccice and nconrd occur inside subcommunicator blocks. Not all tasks will see the error message. Thus, mpi_abort for mpi_comm_world will hang.

If you see an easy way to obtain the desire behavior, fine ... add it. If not, let's move on.

I see. @RussTreadon-NOAA Thanks for the heads up. I will try and let you know.