GitHub Issue NOAA-EMC/GSI#538 Add options to tune weights of mixed ensemble for FV3-LAM EnVar#541
Conversation
…semble for FV3-LAM EnVar
|
Is it possible to ask @TingLei-NOAA and @lbi2018 to review this PR? I think they are well-qualified because they are major contributers of #451. |
| write(6,*)'n_ens_gfs and n_ens_fv3sar won"t be used if not regional_ensemble_option==5' | ||
| endif | ||
|
|
||
| weight_ens_gfs=one |
There was a problem hiding this comment.
Here, the namelist parameters like l_both_fv3sar_gfs_ens had been readin as well as weight_ens_gfs and weight_ens_fv3sar, right? then those assignments of the later two parameters (weight_ens??) will override the readin values for them. So, they will always be one.
Do I misunderstand something here?
There was a problem hiding this comment.
In the case of l_both_fv3sar_gfs_ens=.false., only either global or regional ensemble is used, so weight_ens_gfs and weight_ens_fv3sar should be one (only either is used). To force that, I added this part. (If ensemble B should be tuned in this case, please don't set weight_ens_gfs or weight_ens_fv3sar but set beta_e0.) In the case of l_both_fv3sar_gfs_ens=.true., weight_ens_gfs and weight_ens_fv3sar are not overwritten and the parameters in the namelist are used.
There was a problem hiding this comment.
@shoyokota Thanks for your explanation . That does make sense.
This question is resolved
There was a problem hiding this comment.
@shoyokota Could you please run a regression test and document the test result here.
There was a problem hiding this comment.
@ShunLiu-NOAA I tried regression tests on Orion, but global 3dvar/4denvar/4dvar tests were failed because accesses to some files belonging to group "rstprod" are not permitted. I requested access to the NOAA RDHPCS Program, but it has not been accepted yet after two weeks. Even after it is accepted, I can run tests only on Orion. Could I ask someone to run the regression tests on my behalf?
There was a problem hiding this comment.
@TingLei-NOAA Could you please run a regression test on HERA or WCOSS2?
There was a problem hiding this comment.
@ShunLiu-NOAA Sure, I will give an update here when I have finished the regression tests.
|
@ShunLiu-NOAA On hera, two failed tests: 2:global_4dvar;7:rrfs_3denvar_glbens. Seems GSI/fix/anavinfo.rrfs is missing for 7 while GSI/fix/global_anavinfo_qlqi.l127.txt is missing for 2 in this PR. It seems I still use an old fix dir. But "git submodule update" gave the error with message |
|
@ShunLiu-NOAA , seems I should first run "git submodule sync" . After that, "fix " dir was updated . The regressiong tests of those two failed ones is ongoing. |
|
I don't have any regression test on Hera recently. @hu5970 Could you work with Ting on this regression test? |
|
@ShunLiu-NOAA An update on my regression tests. Thanks to Russ, the current GSI regression tests procedure works fine for me on hera. But a few re-run of the ctest always gave me a couple of failed tests (different for different runs) for their wall clock time exceeding the time limits (even I changed all of them to 30:00). I will try more required time while I am wondering what the culprit is . |
|
@TingLei-NOAA , the most important aspect of the ctests is reproducibility independent of task and thread count. A few of the checks in the current suite of ctests need to be revised. While the test indicates Failed it is not a fatal fail. For example, slow I/O on Orion can lead to an anomalously large wall time. This is a system fileset issue, not a gsi or enkf code issue. |
|
@RussTreadon-NOAA Thank you for your clarification/reminding. Yes. the job killed for exceeding time would cause the regression test become idle. Since, such tests with this issue were different for different runs of the whole ctest (they succeeded in different runs), I would conclude that the current PR passed the current regression tests . @ShunLiu-NOAA . |
…semble for FV3-LAM EnVar (#541) This PR adds options (weight_ens_gfs and weight_ens_fv3sar) to tune weights of mixed ensemble for FV3-LAM EnVar (NOAA-EMC/GSI#538). Regression tests for global 3dvar/4denvar/4dvar are not completed yet, but for other tests, issues are not found except for "failed the scalability test" and "exceeded maximum allowable hardware memory limit" on Orion. Fixes #538 Co-authored-by: Sho Yokota <syokota@Orion-login-1.HPC.MsState.Edu>
…le for FV3-LAM EnVar (NOAA-EMC#541) This PR adds options (weight_ens_gfs and weight_ens_fv3sar) to tune weights of mixed ensemble for FV3-LAM EnVar (NOAA-EMC#538). Regression tests for global 3dvar/4denvar/4dvar are not completed yet, but for other tests, issues are not found except for "failed the scalability test" and "exceeded maximum allowable hardware memory limit" on Orion. Fixes NOAA-EMC#538 Co-authored-by: Sho Yokota <syokota@Orion-login-1.HPC.MsState.Edu>
5 participants
This PR adds options (weight_ens_gfs and weight_ens_fv3sar) to tune weights of mixed ensemble for FV3-LAM EnVar (#538). Regression tests for global 3dvar/4denvar/4dvar are not completed yet, but for other tests, issues are not found except for "failed the scalability test" and "exceeded maximum allowable hardware memory limit" on Orion.
Fixes #538