2022-01-06

JEDI 1

- CRTM Intel Debug workaround https://github.com/JCSDA-internal/crtm/pull/279
- Fv3-bundle nightly runs with uploads to cdash https://cdash.jcsda.org/index.php?project=fv3-bundle 
    - Small issues around code trying to upload test results multiple times when there are failures, Ryan might have a solution for this problem
- New container versions with Python packages required for ioda converters. To do: new containers with updated ECMWF packages and Steve’s bug fixes
- Test jedi-stack PR with new ECMWF codes, runs on Mac, there were issues on Cheyenne that might have to do with how the modules are created (manual changes)
    - Rick will walk Steve and Dom through the process so that more than one person can do that
- Ideas for the future: Good way to automate the creation of the software containers? Commit hook for jedi-stack repository to trigger regenerating containers when stack gets updated?
- Need discussion about cloud environments for JTD, Obs Processing and R2D2. Kubernetes cluster? I prefer something not locked to AWS for partners to also use.
- Todo: ufo-data needs CI.
- Next big ioda project is on the efficiency of the I/O, Fortran API put on the backburner for now

Updates from the JEDI2/3 meetings on 20 December and 3 January:

- Generic interpolation optimization. Yannick: work in progress, currently see 20-50 times speed up the apply part of getvalues depending on the case (test cases are fv3-jedi C786 with most operational observations).
- Testing fv3-jedi with UFS. Maryam:  Mark Potts working on updating the container. Building UFS in the bundle is being discussed.
- Localized TLM. Christian: started working on some simple cases.
- Background error covariances in fv3-jedi: Benjamin: plans to update the scripts to the latest development, and use the new fitting function (expected in a few weeks) for the new B estimation.
- Ensemble with fv3-jedi: Sergey: finished phase 1 testing: radiosonde assimilation with GSI and JEDI GETKF, some differences.
- Regional fv3-jedi DA: Ting: focusing on pure ensemble 3dvar using ioda 1 observatoins and old version of jedi. Focus is mainly on the problem of difference with h(x) for wind, the issue for the conversion between d wind and a wind for fv3 lam.
- Neptune DA: Nancy: currently running 30 days tests cycling 3dvar with neptune (completes 2 days). Focus on adding new obs, modifying yamls, checking sensitivity, integrating neptune based diagnostics into the cylc suite.
- A discussion on comparing notes on the cycling results for different systems (neptune, fv3, mpas).
- Cubed sphere grid in atlas: Marek: dual mesh works is now working. MetOffice is creating another model interface lfric-lite, mainly for H(x) capability, with focus on interpolation.

Other updates:
- Anna and Francois: several PR improving higher-level yaml validation in cost functions, applications, background errors and localizations.
- Clementine: PR in oops to allow the split of MPI communicator between the “coupled” model components for generic coupled applications.
- Anna: ongoing work on extending coupled classes to allow for an arbitrary number of model components (instead of hardcoded two).

Good progress on the workflow. 3D and 4D H(x) is working at full resolution, and they are targeting next week to have a cycling system working.

There are a number of YAML and other changes necessary for using the new workflow. Yannick is planning on giving a presentation at the end of January to inform everyone about these details.

Emphasis is on processing obs from different data provider feeds, and pushing these observations through the UFO processes. Several operational centers are on the hook to demo systems in the spring, and this work is supporting those efforts.

Having success processing ATMS, Sonde, and GNSSRO examples.

Need to add some new variable changes.

Need to address an outstanding issue in ufo (https://github.com/JCSDA-internal/ufo/issues/1778) pertaining to making YAML specification more flexible for handling cases where you want to assimilate variables different from those involved in UFO filters and checks.

Happy New Year everyone! OBS3 had its first meeting of Q3 yesterday. Here's the summary:

        • A new program, ioda-validate, will soon be introduced. This program will take a look at an obs file and will tell us if the file follows the JEDI Data Conventions. I spent a lot of time before the holidays translating our convention tables and document into YAML. Now I am linking it into our ufo-data and ioda-converters repositories so that it will automatically run whenever we make a pull request. This will greatly speed up converter development. There will be a presentation on how to use this tool at the next OBS3 meeting on January 19.
        • We are planning a test data cleanup for later this month. Our CI system is timing out because the tests are taking too long, and some tests cannot run in parallel because they write the same output files. We want to remove duplicate tests, take non-unit-tests and move them to tier 2 or nightly, fix test dependency ordering, and refactor the ufo_test_data macro to make it more flexible. This crosses over into OBS1, and Hui wants to be involved, so the bulk of the work will occur when she gets back.
        • EMC and NRL have asked us to look into how IODA and UFO can better handle superobbing. When computing Hilbert curves or performing density reductions or performing track consistency checks we need to have all locations loaded onto one CPU before distributing data to other nodes. We are meeting next week Friday to discuss.
        • Chris Thomas has draft PRs in oops and ufo that allow us to better order filters in YAML. He implements a What You See Is What You Get approach where the order in YAML matches the order of execution. This should make it easier to understand our filtering logic.
        • Thanks to work by the JEDI1 team, we have new CI containers that support all of our new python components. This will let us merge some of eccodes-Python PRs in ioda-converters. There will be another round of CI containers that deploy recent versions of eckit and odc. This will help with ODC-engine development.
        • Greg, Ben, and Youlong have been working on different aspects of BUFR file ingest. This includes work on AMDAR, ADPSFC, SYNOP, ship, buoy, GNSS and ATMS data. Additionally, outside of BUFR, Youlong developed a new converter for 9km SMAP soil moisture data.
        • Jianjun reports that JEDI-GEOS UFO can now reproduce GSI HofX for GMI. He adjusted the CRTM ObsOperator to screen satellite and solar azimuth angles, and added additional tests for rain and snow clouds.
        • We heard from Neill Bowler and David Simonin recently about the Met Office's need for new variables and dimensions in our files, particularly whenever vertical levels are used. David also has some comments in today's chat regarding mapping IODA variable names with GeoVaLs names. We can discuss today and at tomorrow's JEDI Conventions meeting. This ties in with some of the geovals vertical ordering changes, and we can discuss issues with implementing names like "air_pressure_levels_minus_one" and "geopotential_height_levels_minus_one".

JEDI notes

V2.4.1 release slightly delayed  because of impending fixes to openMP and build scripts

CRTM v3.0 now fully merged into it’s own branch, where it will remain for a bit as we finish up 2.41.   Thanks to Mark Liu

Patrick adding openMP over channels to the CRTM Forward model component, previously was just in K_Matrix.

Starting work on generation of the EPS-SG MWI

Per Dom’s comment, I’ve reviewed/approvied the workaround for the ADA_Module, but we’d really like to fully understand why this happens.

Patrick has been working on the AOD operator interface in CRTM

We’ve enabled a capability to build static and shared library for CRTM simultaneously

Cheng is working on an offline 2-stream model for quick radiance sanity checks.

Cheng streamlined the new GEOS-GOCART5 aerosol table, significantly reducing the size of the file.   The  PR is still open, but we’ll focus on closing those out shortly.

Isaac Moradi (GMAO) has been actively engaged in creating a new method for cloud coefficient generation, with a particular focus on adding new non-spherical ice particle scattering properties in support of more accurate cloudy radiance simulations.  

Isaac is also working on the active sensor module based on prior work, with the goal of enabling direct assimilation of reflectivities from Radar and Lidar.