Yannick opened the meeting by prompting a discussion about the logistics of this weekly meeting. He suggested moving this meeting to a bi-weekly format (everyone attends), and then filling in the alternate weeks with special topics (only those interested in the topic attend). It was decided to try this out for a while and adapt the format as needed. Suggestions for the special topics meetings are:
- Definitely do these
- How to handle surface obs below lowest model level
- QC filters
- Additional suggestions
- Fold in some of the existing specialized meetings, such as models and B-matrix
- Assistance with specific needs of a particular center, such as the Met Office
Yannick announced that the beta version of the ECMWF ODC API has been delivered to the core team. The core team will begin testing the ODC API and doing some performance benchmarking. It has been discovered that ODC requires eckit version 1.1.0 (we are currently using 0.23.0) so the core team will also be working on doing an upgrade of eckit.
The we collected updates from everyone
Boulder
Dan was visiting Boulder this week and he is developing the capability to do multiple outer loops using different resolutions with FV3.
Mark O is working on the eckit upgrade. He has ufo-bundle building, but is encountering some test failures. He has discovered that fckit also needs to be updated. He is making good progress with fixing the causes of the test failures, and will be issuing PRs soon.
Clementine continues to make progress with MPI communicator schemes for her block ensemble methods.
Hailing has GNSSRO assimilation working for multiple cycles in an FV3 forecast.
Steve H has a PR under review for moving the IODA netcdf interface from the netcdf C API to the netcdf C++ API. This requires the use of the JSCDA fork of ecbuild, so this needs to be installed on all of the HPC systems. Cheyenne is working at this point using the jedi-stack modules.
Steve H has created two new repos to facilitate testing ODC: JCSDA/odc and JCSDA/odc-bundle. odc-bundle successfully builds and all ODC ctest tests pass. Please do not distribute these repos to anyone since ECMWF is not ready quite yet to release ODC. We are also using these repos to track any changes we need to make and feed those back to ECMWF.
Work has started for attaching units specifications (metadata) to variables in netcdf obs files. Steve H announced that we need to follow the netcdf COARDS conventions for the unit names, etc. The COARDS spec includes more variable metadata beside units, so a discussion ensued about how much of the COARDS spec we want to include. It was decided for now to only include the units metadata and we will add more variable metadata in the future as needed.
JJ has 2 PRs under review in UFO related to CRTM. The first contains preparation for handling hydrometeors and water vapor in CRTM. This one should not affect model work. The second enables the addition of clouds and absorbers into CRTM. It is set up to add absorbers one by one, making it easy to add in new absorbers in future PRs.
BJ has fixed an issue with variable transforms in MPAS, and is currently working on mesh transforms for MPAS.
Ryan Honeyager is a new JEDI core team member who will be in College Park working as a JEDI-EMC liaison. He will start with UFO related tasks and may move into FV3 related tasks in the future.
Chris H has getValues working in the shallow water (SW) model now. For the short term, this involved creating a subclass of GeoVaLs which holds some model specific items. We want GeoVaLs to be model agnostic, so the long term plan is to move this temporary solution out of the GeoVaLs class. Chris reported that he needs synthetic obs to test the SW model, and requested help with developing the MakeObs utility for this need. A discussion ensued about how to go about this task, and it was decided to take this discussion offline.
Chris S asked about the status of ODC and what the ODC package contained. Steve H responded that:
- The ODC package builds successfully and all ctest tests pass
- The package includes APIs for C++, C and Fortran
- The package does not include a Python API, and we have inquired with ECMWF about this omission.
College Park
Guillaume reported slow execution on Theia using SOCA with a 1/4 degree resolution in a 3DVar run. He noted that they don't see this issue on Discover. The issue is not in BUMP, but does appear to be related to an "invert-multiply" function.
Guillaume also asked about the status of some of the QC filters (in the develop branches) as they have experienced some broken functionality. Yannick acknowledged that there are some test "holes" and issues with some of the filters, and asked Guillaume to send a list of issues that they encounter with QC filters so these can be resolved.
Cory has a PR under review in ioda-converters that adds units metadata to the GSI netcdf variables. He generalized this solution so that the converters beside the GSI diag converters can utilize his work.
Hamideh is making progress on assimilating brightness temperature, humidity and salinity.
Andrew has resolved issues with radiance channel selection. He also reported, in general, issues with various builds. Any issues that come up should be reported through ZenHub so that their resolution can be tracked.
Met Office
Marek is making progress with time interpolation in 4D EnVar runs using LFRic. The Locations class extension is working properly now, but he is experiencing a program crash during the obs operator function. After a short discussion it was suggested to create an H(x) test which should facilitate debugging the crash. This test could be copied from the 4DVar H(x) test, and then just add the time interpolation.
Marek also asked about how to do adjoints transform between different momentum field representations, and reported compile failures with the SABER repo. The adjoints transform discussion will be taken offline, and it was noted that the SABER repo is not quite ready for use yet. The plan is to get SABER released next week, and any issues that come up should be commented in the upcoming PR (for doing the initial release of SABER).
NRL
Sarah asked if jedi-stack is supported as they have experienced some build issues. The answer is definitely yes and any issues that come up should be reported through ZenHub so that their resolution can be tracked.