Yannick and Dick opened the meeting by announcing the topic of the day, which is an update on the work that the OBS team has made with recent code sprints and an overview on how they are testing H(x), the QC, and the bias correction for UFO, both independently and as a whole.
Dick referred people to a recent JCSDA news post that highlights this work and then we heard presentations from Francois, Wei, and Greg highlighting the recent UFO code sprints and the status of obs operator development and validation for radiance, satellite winds, and conventional observations. For further information consult the following slides.
Overall, these presentations reported excellent progress toward one of the key deliverables in the current JCSDA AOP, which is to replicate the current functionality of the GSI Observer system.
After Wei's presentation, Dick emphasized the role of the NRT web application in monitoring multiple observing systems in one place.
Ming asked Francois which version of GSI they were using to generate ncdiag files for ioda-converters. Francois responded that they are using a specific branch and offered to email the details.
Ling asked how clouds were handled in the examples that Wei showed. Wei responded that the results he showed focused on clear-sky radiance. GSI's current implementation does not implement an additional cloud detecting capability so, since the purpose of this is to first replicate GSI, this was also left out here. This and other obs functions in ufo will include cloud detection algorithms in the future.
Greg surveyed a number of new ObsFunctions, HofX tests, and QC steps that have been implemented recently for conventional obs and SatWinds (slide 3). Some of them are specific to certain applications but others, such as TropopauseEstimate@ObsFunction and WindDIrAngleDiff@ObsFunction are more generic and may be useful for other purposes. Many of these changes are currently in a feature branch of ufo created for the code sprint but they will be merged into develop soon.
Comparisons between GSI and JEDI for SatWinds (Greg's slide 5) exhibited only small differences. Greg (via Anna) suggested that these may be limited to measurements close to the surface where GSI has some capabilities that JEDI has not yet implemented.
Greg concluded his presentation with work that needs to be done, including a status chart that he hopes will be all green (complete) within the next few months (slide 8). In particular, he mentioned that though many of the bufr-to-ioda converters in ioda-converters are complete, there is still some work to do for conventional obs.
After the presentations finished, Yannick asked if there was a plan to do this type of comparison with other NWP systems, such as those at the UKMO and NRL. Dick responded no; doing this level of comparison for every single system would be unsustainable. But, a goal is to make it easier to do this type of work, including software tools and documents that define a protocol. The role of the OBS team would be to coordinate and support such comparisons. Greg added that he has already been actively conversing with colleagues at the UKMO and invited them to contribute to the zenhub epics if they saw anything that should be added in order to work with their system.
Ming asked how close the team was to performing many of these same comparisons with FV3-LAM. Greg responded that this still needs to be determined and expressed plans to work with Ming on this endeavor.
Ming also asked how expensive these tests were in terms of computer resources. Dick and Greg responded that the computer resources were very small compared to those required to run a cycling DA system. They focused here on H(x), which is an essential component of DA and particularly DA validation, but not usually one of the more computationally demanding components.
David S commented that one aspect of this where the UKMO struggles the most and puts in substantial effort is in dealing with auxiliary files that include supplemental information to be used for bias correction and QC. Dick agreed that this is a challenge and expressed a need in JEDI for a consistent strategy that works across systems. Greg mentioned the effort he is leading to develop ioda naming conventions as an example of work following the spirit of that intention.
Before the meeting adjourned, Yannick announced that there will be no JEDI meeting next week. Instead, we will have the JCSDA Quarterly review meeting.
Our next JEDI meeting with therefore be February 4 when we will have a focused topic discussion on the new JEDI workflow manager, EWOK. EWOK is not ready yet for general use but it's a good time to provide an update and overview.