JEDI-based ensemble recentering and analysis calculation #3312
Conversation
|
Interesting. I updated to d9e0275. No shellcheck warnings. I am using a copy of Is this version of shellcheck too old? I see that g-w has a Now I see shellcheck warnings ... but these warnings are not the ones we see from github actions. Any ideas @WalterKolczynski-NOAA, @aerorahul , @KateFriedman-NOAA as to (a) what's going on and (b) how to get the github shellcheck action to pass? |
DavidNew-NOAA
requested review from
RussTreadon-NOAA,
WalterKolczynski-NOAA,
aerorahul,
junwang-noaa,
KateFriedman-NOAA,
DavidHuber-NOAA,
CoryMartin-NOAA and
CatherineThomas-NOAA
as code owners
|
@aerorahul @WalterKolczynski-NOAA This PR is ready for review. It has a pending companion GDAS PR, but I'd like some feedback on this PR before I merge the GDAS PR and break things. |
|
WCOSS2 g-w CI Install All jobs ran up to the 20240224 00Z cycle. Jobs enkfgdas_fcst_mem001 and enkfgdas_fcst_mem002 failed in this cycle with the error message
The fv3gfs modules the @WalterKolczynski-NOAA , @aerorahul , @DavidHuber-NOAA : any suggestions for how best to resolve this issue? |
|
WCOSS2 Test As a test make the following modification to It is necessary to load udunits and gsl before nco. With these changes the enkfgdas_fcst_mem jobs run to completion. This is not a solution. It is only a test to demonstrate that defining the path to ncbo allows the enkfgdas forecast jobs to run to completion. The C96C48_ufs_hybatmDA completed all jobs with only the following jobs remaining A rocotocheck of enkfgdas_earc_vrfy shows that this job can not be queued because enkfgdas_echgres of the given cycle is not SUCCEEDED. The changes in this PR remove echgres from JEDI based DA. Job echgres is no longer in the C96C48_ufs_hybatmDA xml.
We need add logic to exclude the echgres dependency when running JEDI atmospheric DA. |
|
WCOSS2 Test 2 Updated and regenerated C96C48_ufs_hybatmDA_pr3312.xml. While this allowed enkfgdas_earc_vrfy to be submitted, the job failed due to a required file not being found Directory Templates in |
|
WCOSS Test 3 Locally modify
With these changes in place the enkfgdas_earc_vrfy jobs ran to completion. With these and other local modifications to the Cactus working copy The modified copy of The change to I did not make any changes in My guess is that these changes are related to the gefs, gfs, and sfs builds of the UFS weather model. |
There was a problem hiding this comment.
g-w CI for C96C48_hybatmDA (GSI-based atmospheric DA) and C96C48_ufs_hybatmDA (JEDI-based atmospheric DA) run on Cactus. All jobs in C96C48_hybatmDA successfully run to completion. The enkfgdas forecast and archive jobs failed. Modifications to g-w files allowed these jobs to successfully run to completion.
|
Thanks @RussTreadon-NOAA for the thorough testing. The archive job issue seems like an easy fix. I'm not so sure about the |
|
@DavidNew-NOAA : Two questions for others
|
The `gfs_arch_tars` job currently does not depend on `gempak` jobs, even though it archives data produced by them. This PR will introduce that dependency. Additionally, there are several missing dependencies for cleanup when the arch_tar job is not executed. Nearly all of the job's dependencies need to be replicated for cleanup in case arch_tar doesn't run. This PR will address this problem as well. Resolves NOAA-EMC#3294
Adds sfs as a valid option for NET. To start, the GEFS system is generally just copied wholesale for SFS. This includes the extract_vars job. Other than base and resources, config files link to the GEFS versions, just as GEFS config files point to the GFS versions except where they have needed to be changed. The temporary SFS_POST option has been removed. The existing SFS test is copied and slightly modified for a PR-level CI test. Resolves NOAA-EMC#2271
DavidNew-NOAA
requested a review
from TerrenceMcGuinness-NOAA
as a code owner
Description
COORDINATED MERGE
This PR implements ensemble recentering and analysis calculation in the Global Workflow, using JEDI-based applications to replace certain GSI utilities when JEDI is turned on in the workflow. If using GSI, then the workflow will remain unchanged. This PR also (finally) implements native-grid DA increments into the worflow.
The
gdas_analcalcandenkfgdas_ecenjobs will be replaced bygdas_analcalc_fv3jediandenkfgdas_ecen_fv3jedijobs respectively. Theenkfgdas_echgresjob is eliminated, since changing of resolution of the deterministic backgrounds is done internally in the JEDI-based recentering application.The design for this PR is based on discussions between the DA team and GW team a few months ago. Explanation of the flow of data through the workflow:
The
gdas_analcalc_fv3jedijob dependencies do not change. The native-grid backgrounds andincrements are staged, and then the GDASApp JEDIfv3jedi_add_incrementsapplication is run to add them and interpolate to the Gaussian grid. The Gaussian-grid backgrounds are also staged, and then a simple Python function inserts to these analysis variables into the histories, which become the Gaussian analyses. This is done this way to guarantee that the resulting Gaussian analyses are in the exact format required by UPP.The
enkfgdas_ecen_fv3jedino longer depends on the analysis calc job, since the ensemble-resolution variational analysis is computed/interpolated internally in the JEDI-based recentering application. All other job dependencies remain the same. We no longer need to compute the ensemble mean analysis in this job, since it can be outputted the the JEDI local ensemble DA application in theenkfgdas_atmensanlsoljob and just staged for recentering. The variational increment and deterministic backgrounds are also staged to compute the ensemble-resolution variational analysis. The output of this job is no longer the recentered ensemble increments, but rather the "correction increment", which when added to ensemble increments becomes the recentered increments. The prefix for the "correction increment" iscatminc.The
enkfgdas_fcstjob now stages both the ensemble increments and the correction increment. They are added together withncboinforecast_postdet.shto generate the recentered increment.All forecast increments, both deterministic and ensemble, are now on the native cubed-sphere grid
Resolves #3248
Type of change
Change characteristics
How has this been tested?
Checklist