Since our last meeting was three weeks ago, the main agenda item for today was to touch base on what everyone has been up to since then.

NOAA - EMC

Update from SOCA (Guilamme)

  • SOCA team is making good progress integrated changes from the code sprint several weeks ago into develop - still some work to be done
  • Working with Rahul on refactoring ufo integration
  • Progress on covariance modeling - added trajectory
  • Bump has been integrated into SOCA

Update from Sylianos

  • Has decided how to move forward with integrating wave model into JEDI but will be busy in the next few weeks with preparation for an upcoming summer school

Update from Ben (CRTM)

  • Progress with Anna on recent code enhancements (I didn't catch the details  - MM)
  • debugging Jacobian issues
  • new TL/AD implementation working


NCAR

Update from MPAS (BJ)

  • implemented and tested new bump interface
  • comparing bump to nicas - close on some aspects but off on others - still investigating
  • implemented geoval for radiance operator
  • ran into some parallelization/communication issues with global operations (e.g. min/max) that needed to be addressed
  • Currently can only use a single observation operator
  • To do
    • analytic interpolation test
    • coding norms test

Question (BJ) - how do we implement multiple instruments for Radiance operator?

Yannick: There are several possible ways to proceed - we need to discuss this

Reminder from Yannick: We need to Review our ZenHub boards - make sure they are up to date

NASA GMAO

Update from FV3 (Dan)

  • successfully switched to Bump
  • 3Denvar seems to be working with GNU compilers but there are still issues with the intel compilers
  • 4DVar now working
  • Team has been working on the CRTM interface recently - initial results promising - seems to be producing a reasonable brightness Temperature
  • coding norms test successfully implemented
  • Analytic interpolation test successfully implemented so now we have a comprehensive State test
  • Next on the agenda
    • balance operators

Warning: A recent ufo merge broke 3DVar (GeoVaLs allocation problem) - Dan has seen this with FV3 but it will likely impact everybody.  Dan will look into fixing this today.

NRL

No progress to report at this time

NOAA (College Park)

Update from Rahul

  • Has been working on the separation of GSW - making build optional - work to do on instantiation 
  • Working with SOCA team on integrating results of code sprint from several weeks ago into JEDI develop branches
  • Meeting today to discuss how to proceed with integrating marine data into ioda (convert to ncdiag?)

UK Met Office

Update from LFRic (Steve)

  • Have been working on EnVar - it now runs but has not yet been verified 
  • 3DVar works with readiosonde
  • Bump has been implemented - needs to be checked

Issue raised: We need to decide on standards in ioda for the representation of latitudes and longitudes (units, range) so that models can be made to be consistent.

Update on ODB

  • converted some ioda and ufo test files to ODB2 for testing
  • also working on ODB1
  • now using a local ODB build 

Yannick: It would be desirable to include some ODB implementation in JEDI repos but we have to be careful about licensing and distribution.  We have access to ODB software through our ties to ECMWF but some of it is not on GitHub - available only through tar files.

Issue raised: Do we test coding norms for Fortran? Answer: Not yet.  We do not know of any existing tools that can be used for this and it is not a high priority for us to develop our own at this time.  If anyone knows of software that is available to test Fortran coding norms, please let us know!

In the meantime, it helps to turn on syntax checking options when compiling

JCSDA (Boulder)

Doxygen update (Mark)

We decided that it makes the most sense to build the doxygen documentation at the bundle level in order to capture the interaction between all relevant namespaces.  A prototype for this now exists in the feature/doxygen branch of the ufo-bundle repository.  We encourage you to follow this example to add doxygen documentation to your own bundles.  Steps include

  • modify CMakeLIsts.txt to include an optional BUILD_*_BUNDLE_DOC flag
  • Add a Documenation directory that contains its own CMakeLists.txt file, along with a Doxyfile.in config file.   Feel free to copy these files to your own bundle and make minor modifications (names and path names) . If you have any questions, feel free to contact Mark (miesch@ucar.edu)

Yannick commented that, though it is convenient to work with the bundles now, we are considering other options for the future for organizing and building JEDI applications.  Currently some developers (including Yannick himself) work outside the bundles, using CMake flags to compile the repositories they need.

Update on MPI implementation in OOPS (Chris)

  • the feature/bump branch is having problems finding MPI on Theia (Rahul later commented that he has successfully run the fv3-bundle on Theia so he may be able to help with this - it may be sensitive to environment)
  • Administrators for Theia did install git-lfs but they did not yet open any ports for data transfers.  Since GitHub uses Amazon web services for their LSF stores, they may be reluctant to take the security risk.  The LFS fetches can be bypassed through appropriate git configuration settings or environment variables.  This allows the code to build (i.e. build scripts run), but tests will fail since the data has not been downloaded.

Issue Raised: We may need an alternative to Git-LFS for some HPC systems that have particularly stringent security protocols

Chris (S) later asked whether this is a problem on Cheyenne.  Chris (H) answered no - because NCAR/CISL does not block outgoing internet traffic from Cheyenne.  The problem is that NOAA's security protocols are more strict than NCAR's, and also more strict than NASA.

Update from Clara:

Has been working with Steve to convert observational data files to netcdf, for interfacing with ioda.

Issue Raised: Is netcdf the best way to proceed?  The files are very large.

A few responses: currently we're duplicating header information in at least some of the data files, which can increase the file size significantly.  This was a quick implementation and needs to be improved.  More generally, it was pointed out that file sizes and conversion time can become problematic if you're not careful about how you write the files.  This is a priority for Steve when he returns from holiday and travel.

Update from XIn

Xin has been working on developing a JEDI environment container on Cheyenne using Inception and Singularity.  A problem he has been running into is that inception disables internet connections from within the container so one cannot run operations such as cloning git repositories.  Chris (S) mentioned that they are running fine on Cheyenne outside of any containers so there is no need to proceed with this just for their sake.  In any case, CISL is actively working on these issues.

More generally, Yannick commented that getting the containers to work on HPC systems may take time - this is an emerging technology. 

Xin is also looking into implementing multiple variables for each observation type

Update from Francois

  • progress on integrating GNSSRO measurements into JEDI
  • Not currently using ncdiag - instead sticking with buffers for now
  • working on Adjoint



  • No labels