JEDI2 and JEDI3 update (Anna Shlyaeva):
- Benjamin Menetrier's PR on anisotropic functions in BUMP (https://github.com/JCSDA-internal/saber/pull/164) will be merged on Friday, Feb 18.
- Benjamin M developed a testbed for testing SABER blocks: https://github.com/JCSDA-internal/saber/pull/171
- Benjamin M is working on implementation of non-Gaspari-Cohn functions in NICAS (as mix of weighted GC functions).
- Steve Vahl is working on porting spectral B from um-jedi (developed by Marek Wlasak) to saber.
- Cory Martin and Dan Holdaway made changes to generalize GEOS and UFS IO in fv3-jedi.
- At the last JEDI2/3 meeting we had a discussion on saving full increment from JEDI variational application to use in IAU.
- Fan Han added Philips vertical interpolation for computing mid-layer pressure in fv3-jedi.
- Mark Potts reworked the UFS linkage to JEDI and pulled it out of fv3-bundle into ufs-jedi-bundle.
- Ting Lei working to upgrade fv3-jedi regional cycling system from ioda1 to ioda2 files.
- Chris Snyder reported that mpas team working on getting ready for interpolation refactoring.
- Nancy Baker: Sarah King updated neptune bundle up to January. Obvious how much it is improved.
JEDI4 update (Yannick Tremolet):
r2d2 and ewok in beta-testing and updating phase.
JEDI1 update (Dom Heinzeller):
- Options for turning OpenMP on and off on a per-repository basis
- Spack-stack updates: planning for a first version for beta testers end of February
- Problems on hera resolved for Intel 2020.2 - RDHPC sysadmins want us to move to 2021.3.0 because of security issues with 2020.2 - jedi-stack available for testing with those versions
- Recent request to build JEDI and SOCA on gaea (Dom will help)
- Soca-ioda udunits problems resolved
- Ongoing issues with ukmo software stack (eckit/fckit/atlas version incompatible)
- Discover: jedi-stack with 2021.4.0, most oops and ufo tests pass, but fv3-jedi segfaults during initialization
- Negotiations with AWS and UCAR management to use AWS Proserve to help us setting up deeper CI/CD pipelines until we hire a new cloud systems architect (replacement for Rick)
- Ioda writer development to improve performance and to reduce the number of output files
- This is work in progress and the first step is being tracked in IODA #20
- The first step is to create a single pool of MPI tasks that collect from the world communicator group and write out one file per MPI task in the pool.
- Expect a PR toward end of next week
- Next steps are: Modify writer pool to write a single output file using HDF5 parallel IO, add capability to instantiate multiple writer pools which enables the writing of ioda files in parallel
- jopa container updated with MKL for um-bundle (docker, singularity)
- Bug in SABER release related to getting test files from dash - fixed and created new release tags in jcsda public repository
- Updating cdash server - now on AWS t2xlarge instance, issues with https access to container
- New feature to download CRTM data from S3 instead of cdash (currently being tested)
- Discussion at JEDI1 meeting yesterday:
- Need better support for profiling in containers, or better in general: profiling tool that works everywhere.
- Do our containers have Intel vTune?
- Update C++ standard now that many partners are moving to newer Intel compilers?
- need to find out what standards the Intel compilers support https://en.cppreference.com/w/cpp/compiler_support
- Anna was asking about C++-17 support
- C++-20 has better support for datetime code that would allow removing a lot of code from oops
- Asked our in kinds to follow up with their teams
OBS (Ben Ruston):
* Metadata: talking with various groups and people on observations which are not instantaneous such as RADAR, wind profiler and 8-day products for instance for vegetation. At this point the UFO designers are using these data and aware of their characteristics but should come up with some general standards. Investigate what is done for CF and what we can use from that, but keeping in mind somewhat the focus of environmental DA.
* diagnostic monitoring: have been discussing with folks useful things for diagnostic monitoring and what is routinely used for both for quick looks and longer tests. Are inviting Brett C. and David S. from UKMO and potentially Andy Smith (BOM). We would like to continue to expand the tools, and potentially create a new home JEDI-plots.
* simObs: Benjamin Johnson has been leading the whitepaper and ideas for a simObs capability, and we've been in discussion with Nikki Prive about considerations for OSSE capabilities, particularly how to create realistic observation errors for clouds and AMVs, etc.
* MergeIODA: tracking down some issues, usually working well on output from the UFO; however, with some conventional data types have run into issues merging after the converters where the individual files work but the merged files do not. This is likely due to derived variable in the UFO, but tracking this down and getting it sorted out turning out to be a good exercise in identifying again what is in the "raw" data files and what is derived.
OBS1 (Hui Shao):
UFO repo: Many changes were made to update the UFO with the top-down geoval order, which came in from the UKMet code sprint about three weeks ago. Since then, there are a few PR merged to update the existing capabilities, including DrawObsErrorFromFile obsFunctions, UKMet RO observation uncertainty, assignError filter bug fixes.
Discussed about the request to remove "AtmSfcInterp" filter. This is incomplete filter following GSI implementation for surface observations. There was some intention to remove this filter due to its incompleteness and its use of the "reorderzdir" which has been flagged for removal and will require all operators to check and assume top-down vertical coordinates. We had discussion with EMC and figured we need to keep this filter for O-B diagnostics at this point since no other option is available yet for GDAS applications. We plan to remove this once an alternative is tested.
Building a cycling workflow for 3DEnVar. The caveat is how to use the GFS ensemble in our experiments. The GFS doesn't save the ensemble restart files at each analysis/cycling hour. Instead, it saves the ensemble restart files from the last cycle and IAU increments for the analysis hour. Therefore, a 6-hr forecast with IAU on is required before starting a full cycling system. JEDI-FV3 doesn't have IAU at this point. On the other hand, GFS does save ensemble forecasts at each cycling hour on Gaussian grids. Now we are experimenting to convert those Graussion files to restart files so that we can use them in JEDI cycling.
Updating yaml files for conventional obs, which will be used for our skylab experiments.
Converters work are continued for various data types from WMO-bufr.
OBS2 (Francois Vandenberghe):
- We are discussing with AWS about getting help from their Professional Service Team to develop further automatic testing and CI/CD
- We have set-up a prototype web server to display diagnostics from JEDI HofX applications.
OBS3 (Ryan Honeyager):
We want to hold a IODA-converters code sprint on the afternoons of March 8 - 11. The purpose of this sprint is to begin applying the new JEDI Data Conventions to our data. So, we will want to change variable names and units, and we want to switch over to a better representation of observation dates and times in our files. This will be the first in a series of sprints to deploy our conventions, and for now we want to restrict our work to just the converters. There are downstream dependencies in UFO, R2D2, and in the model interfaces, and resolving these will require additional preliminary work before we can host a sprint for those repositories. So, in ioda-converters we may need a temporary feature branch or software switch to go between names.
Yesterday I asked the OBS3 group for volunteers for the sprint, and today I am asking if any of the applications groups or the broader partners are interested in attending. If yes, please email me. Even if you do not have much to contribute, this is a good chance to learn how JEDI gets its observation data, and I can add you to the mailing list and shared drive.
Beyond that, OBS3 met yesterday and we had roundtable updates. Everyone is making steady progress.
Greg has a PR in ufo for ingesting SYNOP, buoy, and ship data direct from an LDM server, and is working on a Python AMDAR aircraft converter.
Jo is working on decoupling filter variables and simulated variables, and is experiencing the expected difficulties of dealing with ObsVectors that now can have differing lists of variables (processed variables or assimilated variables).
Youlong has a new 9km SMAP soil moisture converter PR. He is also validating bufr2ioda snow depth data for Europe. He reports problems with missing station ids.
Mike is working on 1DVAR for IASI and ATOVS. He had questions on how to re-calculate H(x) after retrieving surface emissivity and skin temperatures. The group tentatively suggested a UFO filter, but were unsure about interactions with assimilation.
CRTM (Patrick Stegmann):
- Patrick has fixed a number of bugs in the CRTM OpenMP-over-channels
parallelization that occur for edge cases, like computing only a single channel for an
instrument.
- The rest of his time went into IASI-NG validation calculations for including heavy
molecules in the absorber set.
- Benjamin Johnson has created a preliminary
- Cheng has been working on adding RTTOV aerosol optical properties to the CRTM
and preliminary work on adding netCDF I/O to the CRTM surface structure.
COMPO (Jerome Barre):
- first set of UFO updates done on the averaging kernel operator. Now all types of satellites instruments for trace gases can be assimilated in the system. Converters have to be coded though.
- working on the GEOS-CF (GEOS with reactive chemistry) interface now.
LAND (Andy Fox):
- land project is ready for SABER update.
SOCA (Travis Sluka):
- soca-science uses R2D2 and updates to R2D2 can break soca-science if they are not coordinated.
Questions and discussions:
- A discussion on using IAU with fv3-jedi, file formats in fv3-jedi and conversion between different formats (Hui Shao, Dan Holdaway, Cory Martin, Hailing Zhang).
- A discussion on updating C++ standard to C++17 (Dom Heinzeller, Ryan Honeyager, David Simonin, Yannick Tremolet, Dan Holdaway).