Yannick opened the meeting announcing that there are no special topics, and that we would go around with updates. Yannick also announced that we will meet next week (Dec 20), then take off the following three weeks (Dec 27, Jan 3 and Jan 10) due to the holidays and AMS.
Ming reported that he got an hofx example working with WRF after updating to the correct bump configuration. His example is using only temperature and he will add in more variables next. Ming indicated that he is doing a regional simulation with WRF and that his obs file has only a handful of observations that lie within the model domain. He will work with Steve H to create a new obs file with more obs located in his model domain.
JJ has completed the move of the Locations class from IODA to UFO. This change has been merged into the develop branches. He is also working on having mpas-bundle pull the MPAS code directly from its github repository, but he is running into issues with slow response to requests for minor (but necessary) updates to MPAS. For now, he will work around this by creating a fork of the MPAS github repo.
Steve V is working on a UK Met Office ODB file converter to the IODA netcdf file format. He is working with Steve H on this, and we have a preliminary working conversion for a radiosonde case.
Steve H is working on a GSI ncdiag file converter to the IODA netcdf file format. Anna has successfully created an AMSU-A sample ncdiag file from GSI and the current ncdiag converter works on this sample. Steve is also reorganizing the ioda-converters repository so that all python converter scripts use a common netcdf writer (which creates the IODA netcdf file). The Met Office ODB converter is using the common netcdf writer, and the GSI ncdiag converter still needs to be migrated over to the common netcdf writer. Marek asked about what format is being written by the common IODA netcdf writer, and Steve H responded that it is a new format that Steve V and he designed. The switch to the new format for the existing IODA netcdf files will have to be coordinated with an update to the IDOA reader, and this is estimated to take place in a couple weeks.
Editor's Note: After the close of this meeting, the transition plans for converting all of the existing IODA netcdf files to the new format were discussed. We decided to give the common IODA netcdf writer (in the ioda-converters repo) the capability to write both the existing IODA netcdf format and the new format. We will initially have the converters write the existing format so that we can do the coordinated transition later on in a separate and independent effort.
Clementine reported that her ensemble work is in progress.
Xin announced that he has a pull request, which is under review, containing the C++ ObsSpace implementation. He has merged in the numerous other updates that have occurred over the past weeks (including the Locations move from IODA to UFO).
Francois reported that Hailing has submitted a pull request containing fixes for the GNSSRO QC checks.
Mark announced that he has successfully built a Vagrant configuration file that is set up for JEDI. This includes 6 virtual CPUs and both Singularity and Charliecloud containers. All a user will need to do is fire up Vagrant and run the container of their choice. Mark will update the documentation for this after the meeting, and include a link to the documentation on the jcsda.org website.
Marek has successfully run the LFRic aqua-planet simulation using 6 process element and 24 process elements. 3DVar is running to completion, but the results appear to have errors of which are being investigated. Marek has created a radiosonde pseudo-obs file for his testing. Yannick commented that pseudo-obs should be able to be generated by an OOPS facility called "makeobs". Makeobs may be useful for eliminating the work going into creating the pseudo-obs file. Marek will investigate makeobs. Once 3DVar is working, he will move onto 4D.
Chris has a JEDI module (the user environment manger) for Theia working. He found a fix for an issue where ecbuild would find MPI for one bundle, but subsequently fail to find MPI for another bundle. The key is that CMake version 3.9 works, but CMake version 3.10 does not work. Apparently, CMake 3.10 has changed the way that the CMake find_mpi macro works and that change is incompatible with ecbuild's find_mpi macro. Chris will file an issue with ECMWF, plus Yannick will email the ecbuild developers to prompt them to look at this issue. Chris has switched over to integrating a shallow water toy model into JEDI. This is work in progress.
Chris reported that the cmake --trace facility is a useful debugging tool for cmake issues. (This is accessed through ecbuild using the '--' option: ecbuild -- --trace). Chris also mentioned that he wasn't able to run all the Charliecloud tests on Theia due to lack of root permission. Discussion of this ensued and it was determined that the root permission was only needed to build the image, and users do not need root permission to run the image. Thus, the image can be built on a system other than an HPC system, and then run on the HPC system. One can bypass the tests that require root privileges by setting the following environment variable before running the tests:
export CH_TEST_PERMDIRS=skip
Dan has forecasting working with FV3-GEOS. He is working on getting the State interface fully functional now.
Guillaume is debugging a high resolution ocean (SOCA) run. He also announced that he has a branch on UFO that contains QC checks that ensure obs are located inside the (regional) model domain.
Sarah reported that NEPTUNE integration into JEDI is progressing nicely. They are targeting to get DA run results before the holiday.