2021-07-01

JEDI 1

For the past few weeks, the JEDI 1 team has been primarily occupied with the JEDI-FV3 1.1.0 / IODA 2.0.0 release (which occurred officially on June 14) and the 6th JEDI Academy (which occurred last week, June 21-25).  Yesterday the core JEDI 1 team met to discuss a few remaining issues from the release that we will revisit now that the academy is over.  Chief among these are the ioda v2 file output performance issue (https://github.com/JCSDA-internal/ioda/issues/252), better handling of the test data in saber, download of tagged test data from UCAR DASH when the tags don't match (e.g. patches), automatic (rpath) linking of the ioda python API, and (over a longer term) elimination of the `get_db` and `put_db` interfaces in favor of better use of ioda-engines.

There are 6 open PRs in ioda.  One that is useful to mention here is https://github.com/JCSDA-internal/ioda/pull/233 by Wojciech which introduces an atlas mesh as an option for specifying the obs distribution in ioda.  This will probably be merged this week so anyone interested is invited to have a look if you haven't already.

Mark also mentioned that we want to eliminate the JCSDA-internal forks of fms and GFDL_atmos_cubed_sphere.  These are public forks of public repos from NOAA/GFDL.  Currently we have forks of these repos on both our public JCSDA GitHub organization and our JCSDA-internal organization.  There is no reason for us to maintain both forks.  Since these are part of our public release, we'll keep the forks on JCSDA and remove the forks on JCSDA-internal.  As part of the release process, the release-stable branches for public were synchronized with internal.  So, for any bundles that use these repos (including fv3-bundle and soca-science), you should just need to remove the `-internal` from the GitHub URL.  When this is done, we'll remove the internal forks.

Maryam and Rick have also been working on the MPAS application.  Rick has worked with the sys admins to set up a role account on Cheyenne where he has started installing modules and ewok/r2d2.  No change is needed immediately but after we verify everything is working, we will in the future change the location of the Cheyenne environment modules to the role account so multiple members of the JEDI team can help maintain it.

6/21/21 JEDI 2 Meeting

Attendees JJ, Olly, Wojciech, Catherine Thomas, Ting Lei, Carwyn Pelley 

JJ Guerrette - JJ mentioned the ongoing performance issues with IODA V2 that have been documented in Github issues (https://github.com/JCSDA-internal/ioda/issues/252). He also mentioned a problem he’s see with with radiance files when using channel subsetting (https://github.com/JCSDA-internal/ioda/issues/312). He’s seeing all channels in the output files even when subsetting in yaml.

Marek - Said he was close to finishing a factory PR for Atlas cubed-sphere. This PR allows panels to be chosen more flexibly, generalizing what Dan originally did. He created a factory for the tiles.

Olly - Also working on Atlas. He’s been working on the Mesh-generator. Adding ghost points so things plot correctly in 2D. Also looking into interpolation of vector fields like wind. There’s complications at the edges of the tiles.

Steve Vahl - Has been finishing and cleaning up his initial phase of VADER work which provides a framework for generalizing non-linear variable changes. (It is based on a non-develop branch of OOPS.) Also has been learning Atlas and cubed-sphere code and concepts so that he can contribute to the UKMO Atlas cubed-sphere implementation.

Wojciech - Has been reading some papers on calculating adjoints using TLM methods. Still in the research phase. Hasn’t gotten far on how it could be implemented in JEDI.

Ting Lei - Has been able to successfully get a IODA v1 to IODA v2 conversion into his workflow. But mentioned that the IODA v1 files are still a necessary step in the workflow though. Also he was working on tests to compare GSI and JEDI. At first he didn’t see any localization impact but then found a problem in the yaml. Now works as expected. 

Mark Potts has been working on the UFS interface, specifically the UFS - FV3 side. It now works with the ‘develop’ branch on one side and a feature branch on the other. He’s trying to get a smaller test case for file size purpose.

Christopher Harrop - Reported that he was able to put in fixes into shallow-water for some long-standing issues. Some tests were failing on certain compilers and platforms. He changed some initialization routines and that fixed the issues for some reason. Also updated a convenience build script that is in the bundle to use the latest JEDI stack. Also updated to IODA V2.

JEDI3 (DA Algorithms) update:
- Yannick opened a pull request that runs change of variables only once in 4D H(x): oops#1255.
- Wojciech is working on a series of pull requests for assimilating derived variables: oops#1259, ioda#316, ufo#1304.

• The core team has introduced several models into EWOK and R2D2
  • The following should be working for the H(x) experiment: GDAS, GEOS, MPAS, SOCA
• We are currently porting EWOK to different HPC systems (i.e., other than Orion)

Ben gave updates for the OBS team as a whole, including OBS 1 and 3 because Hui and Ryan are away from their offices today.

OBS:

Hui is out until 12th, contact myself and Greg for OBS1

Greg is out this morning

Ryan is out until Tuesday

https://github.com/JCSDA-internal/ioda/issues/15 Variable metadata, such as units, are not retained when written to an output file, iceboxed pending ioda variable schema?   needs to be elevated.   

Focuses:   Moving forward with GDAS, SKYLAB, UM, GEOS,  IODA v2 updates, Academy, Workshop, ITSC-23 conferene.

Greg is wrapping up conventional obs, satwinds.  Keep track of the BUFR / prepBUFR to IODA conversions via this sub-epic: https://github.com/JCSDA-internal/ufo/issues/764 

Fabio: Debugging 4d-h(x) issue with ioda-v2 files. Interesting that the 3d-h(x) uses the same set of observation files without any issue.  

observing systems suffering of the issue with the ObsErrorFactorQuotient observation function (see https://github.com/JCSDA-internal/ufo/issues/1109). The observing systems using this function are listed as: aircraft, satwind, scatwind, sfc, sfcship, sondes, vadwnd

Hailing:  ioda-converters update:  https://github.com/JCSDA-internal/ioda-converters/pull/581

Still IODA1 output, need to use ioda-update.x to update to IODA2; tested GNSSRO jedi_gdas file Cory generated.

Francois gave updates for OBS2:

• Supported the JEDI Academy (both lectures and practices).
• Fixed some problems with the GNSSRO converter (both from GSI diags input files and from BUFR input files) to adjust to IODA2 format.
• Conducted a walk through of the Diagnostics Notebook to a number of NASA staff. Working on improving the efficiency and scalability of the data read in. 
• Worked with the Met-Office on building the um-bundle code on Orion. The code builds and passes most of the ctests (just missing data files). 

- Getvalue and lineargetvalues for roms (Hernan)

- regular mom6 interface maintenance (Travis)

- jedi-godas/ng-godas: 3D h(x) & 3DVAR with EWOK (Travis, Claude, Eric, Rick, Guillaume)

- ioda v2 saving is slow: ~8 min for 6M obs

- Using R2D2 for obs in soca-science (G)

- letkf/3dvar with multi-domain ufo (Hamideh)

- Regional ocean DA: diagnostic work (Kriti)

For the National Water Model:

Folks are out on PTO, so other than merging a couple of PRs that had been awaiting OOPS/UFO PRs we’ve been largely preoccupied with a “Concept of Operations” document for  NOAA Office of Water Prediction management.

For the UFS work:

We did have a team meeting this week with folks from EMC and PSL, to update on current status and coordinate short-term activities. 

• Some real progress with the snow work, and we are currently in the final stages of tuning some length scale parameters in the new obs space localization with the LETKF-based OI to get it to match existing capabilities. 

• Still need to incorporate some additional QC filters for the station snow obs, and are working on that right now, but then the capabilities for assimilating these station obs will largely be complete.

•  The team are working on tidying up some preprocessing scripts used with the IMS snow cover obs, and will need to update the converter for that.

• Over the next few weeks we’re going to be figuring out what needs changing and adding as we switch from the NOAH model to the NOAH-MP model 

CRTM:

Patrick: Coefficient generation package being released "soon" still need to work up the announcement and last minute PRs, etc., IASI-NG still working on

v2.4.1 preparation, documentation.   Need to discuss a general transition plan to using CRTM develop / release tags so we're not stuck on crtm_jedi branch.  2.4.1 gives additional openMP advantages over crtm_jedi, and includes some bugfixes. 

Cheng focusing RTTOV intercomparison via UFO, keen on gettng RTTOV-SCAT in.    Our focus will be on improving VIS/near-IR capabilities in CRTM. 

Continued interest in slant-path capabilities and/or 3-D conic sections of model space to perform more accurate computations of radiances.  

Strong interest in developing physically-based land surface emissivity model, maybe coordinated effort with LAND?