Yannick opened the meeting by announcing that this will be the last JEDI meeting for several weeks. The meetings on Dec 27 and Jan 3 are cancelled due to the holidays. And the meeting on Jan 10 is cancelled because many of our team members will be attending the AMS meeting. Our next JEDI Weekly meeting will be held on Jan 17. Happy Holidays!
Then we opened the meeting with EMC. Stylianos introduced a new member of the JEDI team, Shastri Paturi. Shastri works on the unification of the operational marine (ocean, ice, and waves) observations at EMC to be compatible with IODA.
Steve H mentioned the JCSDA github repository ioda-converters, which houses similar conversion scripts. Steve, Stylianos, and Shastri agreed to work together to leverage the existing content of the repository and to add any new converter functionality that EMC develops.
Stylianos also announced efforts to put together a team at EMC to contribute to different aspects of JEDI as needed, e.g. the coupled model, the QC, basic workflow, etc, starting as soon as possible.
Then the floor went to GMAO where Guillaume reported that he was still sorting our some issues with the high-res (1/4 degree) ocean model. They were able to assimilate SST but he noted some instability with the QC filters. Yannick responded that he is still working on improving the QC functionality. Guillaume also mentioned that they attempted a run with 4 million obs, but this run did not complete. When the obs data set was trimmed down to 1.5 million obs, the run did complete but was slow. This result prompted some discussion about performance optimization which will be addressed in the future.
Dan is working on computing a high-resolution (56 km horizontal spacing) 3D-FGAT increment for FV3 on 600 compute cores. Currently Bump is not converging. He also pointed out an IO bottleneck; Bump is writing out on the order of 200,000 files, and this is just for the obs operator. Yannick agreed that this is an important target for optimization in the future. Dan plans to try a GFS run at 25 km resolution next.
Chris S asked Dan if he tried running with a fixed localization and mentioned that BJ has an example of where he has done this. Dan responded that he has not tried this but will check with BJ.
Then the floor turned to Boulder.
Mark and Steve announced that they generated new Singularity and Charliecloud containers that include the ODB Python API and the NCEP BUFR Python API.
Mark also announced that this and all future updates of the containers will be announced on the main JEDI GitHub Team so be sure to subscribe to notifications from this team if you want to receive an email when the containers are updated.
To sign up for email notifications on the JEDI GitHub Team, you can follow these steps:
- Log into GitHub and go to the drop-down menu on the upper right (with the image) and select Settings
- Select Notifications from the left menu and make sure email is checked under Watching
- Go back to your GitHub home page by clicking on the cat in the upper left
- In the drop-down menu on the top left, select JCSDA to switch to the JCSDA organization
- Select view organizations in the upper right
- Select Teams on the upper right
- Scroll down and select JEDI. You should see the latest container update announcement there.
- In the menu on the left, select Watch from the drop-down menu. This will tell GitHub to send you an email when new items are posted.
Please contact Mark if you have any questions.
Mark also mentioned that he updated some of the documentation on containers and config files and these are now being reviewed as pull requests. They should be incorporated into the public JEDI Documentation soon.
Xin has been testing JEDI on Cheyenne. He made a new JEDI module available for all JEDI users (see the JEDI Documentation for instructions on how to use it). He also found problems when compiling with the Intel version 18 compiler suite. He fixed a few and is working on more. Yannick mentioned that one of the issues Xin is working on (regarding missing value representation in floating point data sets) could be related to the QC instability issue that Guillaume reported.
Yannick also announced that this week we have merged the new C++ ObsSpace (ioda) implementation that Xin has been working on.
Clementine has continued her work on ensemble DA with a block Lanczos minimizer. It''s running and converging but there are still some issues to sort out.
Steve V has continued his work with Steve H on developing scripts to convert Met Office ODB data into a netcdf format that can be read by ioda. His focus will now shift from radiosonde to aircraft.
Hailing announced that we have merged a new QC template into ioda develop. They will now use this template to develop more sophisticated Obs Filters. She is also working on cleaning up the ROPP code and expects to issue a pull request soon.
Ming had good news to report - the problem he brought up last week has now been fixed. This had to do with how Bump handles sparse data sets that may give rise to an absence of data in some processor bins used by Bump. He worked with Benjamin M to fix this and it is now working.
BJ has been working on assimilating specific humidity observations by converting data from mixing ratios. He asked if anyone else has had experience assimilating moisture observations into JEDI. Yannick responded that he is the first, as far as we are aware.
Junmei reported that she was having problems compiling the MPAS bundle on Cheyenne with GNU compilers. They reported a compiler bug which was then fixed.
Anna has continued to generate April 15, 2018 00Z observation and GeoVaLs files for new data types, working toward a full set of GSI observations. This work includes a fix for the "hack" we have been using to deal with sensible temperature in the obs and virtual temperature in the GeoVaLs.
Then the meeting adjourned with warm Holiday wishes to all!