Yannick is back in Boulder from travel in Europe. He announced that he is currently working on the JCSDA Quarterly Report with Tom, and then he opened the floor for updates from everyone.
Steve H reported that the C++ implementation of IODA is getting near a prototype that has a working Radiosonde obs operator. This work is divided into three pieces:
- boost MultiIndex data structure for the memory store
- C++ netcdf interface
- C++ ODB interface
Steve H is working on the netcdf interface and has the initial version completed. Steve Vahl (NCAR) is working on the ODB interface and this is work in progress. Xin is working on the MultiIndex memory store and he is very close to having a radiosonde example working. Xin reported that he is working in ufo-bundle and getting reasonable values in the Radiosonde H(x) calculation. The code is seg-faulting at the end where it seems that a destructor is trying to free memory that has already been freed. Xin will have this repaired soon. Within a few days, he expects to have a working version of the test_ufo_radiosonde test.
Steve H also mentioned that he is working on a scheme for IODA where either the current Fortran implementation or the new C++ implementation can be selected on an obs type basis. This will enable us to split up the work of porting over the different obs operators to the new C++ IODA implementation.
We made tentative plans to devote the JEDI Weekly Meeting in two weeks (Oct 25) to a presentation about the new ioda ObSpace implementation.
Francois mentioned that they have been working on using a licensed software package with JEDI. They ran into problems with licensing issues but they are now near a solution.
Dan reported that he is currently working on FSOI. He is building a linear forecast app which will enable testing FSOI and he is nearing submission of a pull request for this code. Dan also announced that his 3Dvar TL/AD assimilation of Radiance obs code is under review and should be merged in soon.
Guillaume said that he is waiting on the new C++ IODA implementation to begin porting the marine obs operators over. Steve H responded that the code to enable this work (see above) should be ready sometime next week.
Rahul mentioned that he is close to having the pseudo model reading the state of a forecast advance from disk. He is using QG output state files for initial testing. Rahul requested assistance with the QG interface and Yannick will assist offline. The question was raised on whether it is best to define the pseudo-model in a separate class and header file (PseudoModel.h) or whether it should be an option in Model.h. Further discussion on this will proceed after the meeting.
Rahul is planning on getting started back up on building FV3-GFS on AWS next week.
Hui reported that they are nearing completion of an obs QC filtering interface that goes into UFO (which is used during the execution of obs operators). Initially, the QC filtering will be working on AMSU-A and GNSSRO obs data. They are also working on a post-filter that does obs thinning. Hui asked questions about how to organize the code in regard to these filters. Some of the filters are common tasks that multiple obs operators can leverage. Yannick responded that we need to discuss this further, and one topic is how to make these filters generic.
Steve S is still working on the aquaplanet configuration for LFRic and Marek announced that they are nearing a pull request for the JEDI LFRIC interface.
BJ reported that the JEDI MPAS interface to create GeoVals for GNSSRO is almost working. Hailing is working with them to check the test results. They are getting reasonable results for refractivity, but not yet for bending angle at this point.
JJ requested details about the next JEDI Academy as he is interested in attending. Yannick responded that they have ~50 people interested in the upcoming November academy, and that he will be working with Tom on how to handle this many people. We may be offering another academy soon after the November academy, but this has not been decided yet.
More generally, we have received several queries about how to register for these academies. The answer is that there is currently no registration process. For now, the JEDI academies should be considered as invitation-only events designed to train a selected group of test users. As the project matures, we will hold future JEDI academies that will have an open registration process.
Marek asked if it would make sense to add diagonal scaling to the covariance matrix in order to facilitate the validation of increments. He also asked if there is such a variance scaling capability in BUMP. Yannick responded that the variance scaling does make sense, and that BUMP may have this capability. We will have to follow up with Benjamin about the details of how to use the BUMP implementation.
Steve S asked about the status of the BUMP MPI port to fckit. Mark responded that the code is under review. Benjamin found a couple issues that Mark will fix today, and he is waiting to get review feedback from Yannick before merging. The merge is likely to take place next week. As described on the pull request, please try the feature/fckit-mpi branch of oops to make sure it works with your model before we merge this into develop.