JEDI 1 - 

  • Has been busy with release (v1.1.0) work and JEDI Academy mostly. Orion remains a problem with slurm issues.
  • Had a meeting with AWS regarding support. (Services vs. costs.) Decided to remain at current service level and not increase. At current level all support requests must go through our specified AWS points of contact, and then go into a queue.


JEDI 2 

No report since Dan is on leave and no recent meeting due to Memorial Day holiday.


JEDI 3

Code updates:

Sergey added another option to HybridGain application (average of increments): oops#1227.

Wei Huang issued a bugfix for GETKF with QC filters: oops#1233.

Benjamin updated qg obs file format to use groups, to make the files more user-friendly: oops#1234.

Wojciech documented bias predictors available in JEDI: jedi-docs#197.


Discussions and updates at JEDI3 meeting (June 14):

- Travis plans to use variational bias correction for a simple but realistic 2D SST data assimilation system. Expectation is that for simple predictors it should work reasonably well even before the preconditioning is updated for VarBC. This would be a good testbed for VarBC tests.

- Clementine presented ideas on improvement of MPI communications in 4D variational algorithms, particularly 4DEnVar, by using sendreceive / reduce (from discussions with Andrea Piacentini). Part of the following discussion focused on non-reproducibility of MPI reduce, which is a dealbreaker for operational DA. There are several places in the code where we already use MPI reduce. The ways to go around the non-reproducibility of MPI reduce should be investigated.

- Anna will be setting up a meeting to discuss addition of observation error with cross-channel correlations some time in the next few weeks.

- Sergey and Wei plan to work on profiling LETKF with 2 or 3 types of obs (amsua, conventional obs, and iasi or gps) to figure out where optimization is needed in observation operator computations.

- Benjamin implemented a generic computation of u/v from psi/chi on a regular grid in BUMP (and its adjoint). He used a Savitzky–Golay filter to compute smooth derivatives and get continuous results. This has solved the problem with discontinuous u/v fields generated by fv3-jedi static B.


JEDI 4

  • Is supporting testbed applications that we want to get running.
  • Goal is to run HofX application with all the models using generic EWOK. Implies storing all the backgrounds in R2D2. Making good progress and hope to have it in place by the end of the month.


Ben Johnson

  • Has been working on Academy mostly.
  • JEDI GDAS. Good progress.


Hui Shao 

  • Testing practical sessions for Academy.
  • JEDI GDAS - generating needed files. Once this is done, they’ll have an hourly background.
  • Want to resume regular JEDI GDAS validations after Academy. Right now just using diagnostics.
  • Had a meeting about missing functionalities.
  • Field kickoff meetings are happening. In-kind people are reporting what they’ll be doing.


Francois Vandenberghe

  • Working on IODA and R2D2 in sagemaker framework. 
  • Working on tutorial.
  • Trying to build um-bundle on S4. Having problems with some code that doesn’t compile.


Ryan Honeyager - OBS3

  • OBS team had a documentation sprint. PR’s have been merged.
  • Team members presented at JCSDA workshop. 
  • Also merged release.
  • Team will be presenting at JEDI Academy. New emphasis on configuration instead of development.


Guillaume Vernieres

  • GDAS Ocean applications have backgrounds ready.
  • Having an issue with ObsSpace save, which is taking 8 minutes per cycle.
  • Ocean diagnostics are being worked on.


Andrew Fox

  • Sergey and Clara(?)(Sorry not sure about the name here) Working on reading orography from static file.
  • Clara(?) also looking at interpolation


CRTM Update

  • Working on CRTM v2.4.1. This will have numerical differences from existing version, so need to work on transition plan. Has greater OpenMP support. And new aerosol coefficient files. 
  • CRTM RTTOV scat(?) being worked on.
  • Principal-component based work is being discussed.


David Simonin

  • ODB backend is a priority
  • Obs processing work is going well.
  • Genesis is going well. A bit behind on conventional obs.


Jake Liu - MPAS

  • Took a lot of time to convert to IODA v2.
  • Upgrade to build system.
  • Recently tested yaml QC filter under EWOK JEDI GDAS. Had some trouble with a lot of obs rejected.
  • Background error covariance - 3D-Var from BUMP NICAS seems slower than 3D-Envar.
  • Yannick mentioned that we should be able to do experiments to compare MPAS and FV3 regarding QC filters.
  • No labels