2021-08-12

Conventional: Working on ioda-converters for JEDI_GDAS workflow and also testing the new observations files generated by EMC (e.g., correct conventional multiple variables files, missing fields for ozone, etc.). Also working on aircraft QC and validate some obs error functions. Code into development (UKMO): pressure to height conversion and sondes horizontal drift computation, slant path finding utility, and surface observation processing

RO: wrapping up on the UKMO operator. So we have fully functional RO operator based on NOAA, UKMO, NRL (ROPP 1D), ECMWF (ROPP 2D). 

Radiance: Started working with JEDI core for providing geovals within FOV (for surface temperature averaging within FOV). ATMS development is almost done for UKMO. Started working on AMSR-2 and also RTTOV interface (scattering). Also working on thinning. 

Aerosols: New AERONET AOD is being added to UFO and other relevant code (e.g., ioda converters, fv3-jedi). This will provide an evaluation of the aerosol model using independent AERONET AOD obs.

Fabio reported that they are making good progress on the FSOI (OBS2.4) functionality in the JEDI GDAS work flow. They are currently adding the forecast sensitivities to the work flow.

Rachel reported  on the diagnostics work

• Working with Travis to provide intermediate bin files which will be used for generating common diagnostics/stats. These bin files will eventually be built automatically during the work flow.
• Modularizing the plotting code
• Discussing an alternative to jupyter notebooks. The idea is to use a dashboard instead with widgets that can be activated to create different kinds of plots

• Making great progress on the ioda file and naming conventions
  • First iteration of the documentation will be released soon
  • Still working on two issues
    • Optimization of string types (file IO and DateTime conversion using strings is slow)
    • Introduction of enum type to help with the handling of categorical data
      • Satellite ID, Cloud type
      • Allow for the encoding (numbers) and their meaning to be stored in files
• Good progress with ioda converters
  • Cory has an initial version of a ioda v2 library for python converters
    • In a PR: https://github.com/JCSDA-internal/ioda-converters/pull/604
    • Library that allows converters to directly write out ioda v2 files (ie, skip the upgrader step)
  • Ron is making progress on the generic BUFR converter
    • E.g., updates to handle delayed sequences to the NCEPLIBS-bufr repo.
• The Met Office has an initial ODB reader merged into develop
  
  • David D and Wojciech are working on this
  • The reader can handle 10-15 obs types (both conventional and radiance)
  • Ready for downstream testing
• Next Thursday the OBS3 group will lead a special topics discussion on QC and filtering

CRTM:  Roundtable tomorrow, OMS-GEMS 1/2 under development, ATMS-NG delivered.

RTTOV and Geovals:  should geovals care about what RT model is used? Ultimate goal is to be able to trivially switch between CRTM and RTTOV -- what needs to change to allow this to occur?

It appears that additional work is needed to interface RTTOV with GEOVALS completely.

Maryam is working on the mpas-ewok application interface.  Currently she is working on Orion because the ewok/r2d2 environment is more mature there, but she will move to Cheyenne when it is working.  She ran into a problem with file access (colons in file names) but she worked with the MPAS development team to solve this and is now back on track, making progress.

Substantial progress on the Cheyenne environment: We have been testing new environment modules on Cheyenne and we will transition to these new modules soon, modifying the documentation accordingly.  They are essentially the same as the current modules but the main difference is that they are now hosted from a role account called jedipara instead of Mark M's workspace.  This will allow multiple maintainers (Rick, Mark M, and Maryam) to modify the modules, increasing the responsiveness for users; if one of the maintainers is unavailable, another can step in and address problems as they arise.

As part of this effort, Rick has also set up environment modules on Casper.  The main purpose of this is to use Casper to launch the graphical user interface for ewok.  Eric has also made progress on setting up R2D2 on Cheyenne so the Joint Testbed Application environment on Cheyenne/Casper has made substantial progress recently.  Rick can say more about this in his JEDI 4 update.

Steve H was on vacation last week but now that he has returned he will focus mainly on the Fortran API for ioda-engines.  Another high priority item for ioda is more fully integrating the C++ API for ioda-engines, ObsGroup, into the ObsSpace class.  If Wojceich and Chris are available to help with this we would value their insights.

Yongang has joined the JEDI 1 team and has started work on optimizing the unstructured interpolation in oops.  This is work that Jacques Middlecoff had been working on but Jacques was tragically involved in a serious car accident a few weeks ago and is still recovering.

Mark M is making slow but steady progress in getting JEDI to run on PGI/NVIDIA compilers, debugging one test failure at a time.  Currently he is working on ufo and has 97% of the tests passing.  11 test failures remained.  The most recent problem identified was a compiler bug that was modifying the contents of a C++ structure that was being passed by value in one of the buddy check methods.  Passing the structure by reference as an argument fixes the problem but we're not sure that's the best solution - Mark is still experimenting.  After ufo tests pass, there are still a few test failures in oops (a few l95 tests) and fv3-jedi (lam/cmaq tests) to deal with.  But the good news is that these are fairly specific failures that should not effect many applications.  So, in short, fv3-bundle should be ready to run the B matrix training benchmark with pgi/nvidia compilers on Summit, provided that the pgi/nvidica compilers on Summit behave the same way as the pgi/NVIDIA compilers on AWS (no guarantee: they are a different version: pgi 20.1 vs NVHPC 21.5, spanning the NVIDIA acquisition).

Progress is also being made on setting up CI for the UKMO UM-jedi interface.  Working with Steve V, Francois V, Yash, and Steve S, Mark M has been granted access to some of the supporting code repositories at UKMO and hopes to generate a prototype development docker container (with shumlib) by the end of the day today.

Jedi 2.2 - Generalized locations (EMC liaison, Emily)

• New hire Francois Hebert joined the team. Will work on field of view for cubed sphere.

Jedi 2.6 - Ensemble DA validation (Cory, Sergey)

• Ran comparison experiment using GSI GETKF vs JEDI GETKF. Surface observations.

• Wei tried setting up experiments for multiple data sets. Still problems for scatterometer winds and AMSU-A.

Jedi 2.9 - MPAS general updates (JJ, BJ)

• MPAS-EWOK h(x) runs. 

• Can run the h(x) with multi_files.

Jedi 2.11 - UM general updates (Kristin, Steve, Steve)

• Working on the CI for UM-JEDI.

• Mark M has built a Docker container. Need to add LFRIC inputs and SHUMLIB dependencies.

Jedi 2.12 - Neptune general updates (Sarah, Nancy)

• Fast h(x) in place. 4D ready by Tuesday Wednesday.

• All eckit changes are in.

• Ioda V2 

• Variable transforms for humidity in place. Model interface and processing.

• Ben Ruston working on obs processing part.

Jedi 2.14 - Cubed sphere grid into Atlas (Dan, Marek, Olly, Steve V)

• Cubed-sphere partitioner is working.

• Close to having mesh gen working on multiple PEs.

• Equal area mesh generator is working for the cubed sphere.

Jedi 2.15 - VADER (Steve V, Dan)

• Refactoring of OOPS to remove the LinearVariableChangeBase and VariableChangeBase classes. Variable changes will be handled by traits instead.

• ewok installed on the following hosts, and successfully running experiments:
• orion
• cheyenne / casper
• discover
• user localhosts
• Installation on hera is problematic, and currently stalled, as without internet access, creating python virtual environments is non-trivial, and hasn't yet been successful.
• With working installations on HPC hosts, ewok installations in containers and AWS instances is next
• Adding user documentation for workflow structure and use
• Working on coupled DA workflows by adding the capability to execute with multiple models in one experiment run