Yannick announced that, due to the U.S. fourth of July holiday, there will be no meeting next week. Next meeting will be on July 11th.
Yannick requested that everyone send in a few highlights (one-liners) to included in his JEDI project quarterly report. This is for the period from April through June for 2019. The highlights are due to Yannick by July 5th.
Yannick reported that ECMWF said that ODC (new API for ODB files) should be available to us within a few days.
Then we collected updates from everyone.
Boulder
Steve H reported that he has created a branch in the IODA repository called "feature/nc-file-checker". It contains a new ctest that checks all of the netcdf files checked into the IODA repository (currently 59 files) for compliance with the IODA netcdf conventions. These conventions are: use netcdf fill values for missing marks, use single precision, don't use invalid numeric values (nan and inf) and make the QC marks integers. Currently 11 of the 59 files fail these checks, and ZenHub issues have been submitted to get these fixed. When a file gets repaired, please commit it to the feature/nc-file-checker branch in the IODA repository.
Steve H announced that his next step is to replace the netcdf C API calls in the IODA netcdf reader/writer with netcdf C++ API calls. This requires that everyone use the JCSDA fork of ecbuild which contains Rahul's fix for the FindNetCDF4.cmake script. All of the containers and AWS have been updated to the JCSDA fork of ecbuild so they are ready to go. If anyone has their own build system outside of the containers, they will need to update to the JCSDA fork of ecbuild before this update is merged in.
Steve V reported that the MPAS jedi interface repository is going to be renamed from "mpas" to "mpas-jedi". This is being done to reduce confusion with the repository that holds the MPAS code, and to use naming that is similar to other JEDI model interfaces.
Clementine is developing parallel execution (MPI level) of ensemble forecast and H(x).
Maryam reported that an update for Travis-CI in the UFO repository has been merged which allows the use of dependent branches in other repositories. Before this change, when you had, for example, a branch in UFO and another in IODA that need to be tested together, Travis-CI in UFO would only checkout the UFO branch and use develop for IODA and the tests would fail. After this change both the UFO and IODA branches will be checked out and tested together. For this to work, the branches from different repositories that you want tested together must be named identically.
A discussion ensued where a concern was brought up that someone may accidentally use the name of a branch in another repository and cause that branch to be checked out when you don't want that. It was decided that we should go ahead with the methodology of using identically named branches, and to address any issues if and when they surface. To help mitigate an accidental match, specific (rather than generic) branch names should be used. Another important practice is to delete branches after they have been merged.
Chris H has forecasting working in the shallow water (SW) toy model. He is currently trying to figure out how to get analytic initialization and observation locations working together. The current methods in JEDI are not well suited for this. Chris has met with Mark M and Yannick and they have figured out a way forward.
Chris H reported that Intel 17.0.1 compiler is working on Theia, but Mark M has run into issues with Intel 17.0.1 on Cheyenne and AWS. Chris tried Intel 17.0.1 on Cheyenne and noted that the issues only happen in optimized mode. He is currently debugging these issues, and suspects that they are caused by a defect in the compiler.
Yannick noted that, despite the Intel compiler issues, it would be useful for others to have the SW model (with forecasting) merged into develop as a guide for MPI aspects of the model interface. Chris agreed and will merge his code in now.
EMC/GMAO
Ben will be upgrading the sea surface emissivity model in CRTM after he receives an update from the emissivity model developer today. Ben also added a ctest for CRTM that exercises every sensor built into CRTM. He eventually wants to do a similar test (that exercises all the sensors) at the FV3 level. Ben asked if Travis-CI is ready for CRTM. Maryam responded that it is not, and she could start working on that today.
Hamideh has been conducting comparisons between GSI and JEDI FV3, and the assimilation of GMI obs looks good. She is also repairing the GMI and SMAP obs files so they are in compliance with the IODA netcdf conventions.
Guillaume announced that he is developing a list of experiments to run on Discover (since they have a large allocation) and that EDA is on the list. He asked if anything EDA related for the ocean was ready. Yannick responded that EDA is technically ready and should run okay, but he added a warning that some tuning may be necessary to be able to process some obs correctly. Guillaume will work with Clementine to get the EDA experiments on Discover running properly.
Jong has a develop branch set up for CICE6 and is making progress with the interface.
Cory is working to resolve some issues with generating/assimilating surface winds obs from GSI. He should have a PR submitted soon.
Dan has been helping Victor with machine learning techniques initially for emulating AOD obs simulation.
NRL
Sarah reported runtime crashes when using the Intel compiler. She is using version 17.1.043 of the Intel compiler, and the issue is related to variables being passed to subroutines. Creating a temporary variable and passing that to the subroutine makes the crash go away. Some discussion ensued (version 17.0.5.239 works on Cheyenne, try compiling in debug mode, messages about temporary arrays are warnings and the issue may lie elsewhere) and it was decided to have Sarah push her code to github so others could help out (debug, review, etc.).
UKMO
Marek has a prototype working for doing time interpolation in EnVar. He wants to clean up the prototype before pushing the code to github. Yannick mentioned that he placed some code examples in github that Marek should find useful. Marek will take a look at Yannick's examples as he fixes up his prototype.