General Announcement

  • Meeting was held during a shutdown period with reduced attendance
  • Quarterly review has been postponed to November

INFRA (Evan)

  • Public repository csta-jedi-bundle has been updated with Jedi model data and SonarQube integration for cybersecurity scanning
  • R2D2 server changes were completed as part of a two-month migration
    • FetchByIndex feature has been eliminated which may break existing client code
    • Users experiencing issues need to update their fetch commands
  • Git LFS quota issues reported at the end of the month
    • Strict budget constraints exist to prevent excessive capacity bills
    • Quota cycle resets the next day
    • Team is working to offload Git LFS traffic to reduce costs

Model Interface (Francois H)

  • UFO: added missing ctest for CosineOfLatitudeTimesOrbitNode predictor, added channels to Difference Check filter, removed the unused atmsfcinterp operator, work underway to clean redundant constant definitions
  • IODA: running initial JEDI experiments with IODA's OSDF backend, and identifying and fixing some uncovered issues
  • optimization: merged last PRs adding OpenMP multithreading to var changes identified by UKMO, now working with UKMO to set up benchmarking experiments
  • maintenance: identifying and fixing bugs for running JEDI with atlas 0.44, in anticipation of bumping JEDI's required atlas version

ALGO (Christian)

  • Lots of small cleaning/bugfixes in SABER/QUENCH
  • Anna Shylaeva added the capability of running ProcessPerts application (for doing spectral filtering) using the explicit diffusion block as a filter
  • We are preparing to do a coordinated merge to simplify how 2D variables are handled in BUMP. Waiting on a few last PR approval, and keep eye out for Discussion
  • PR is open and under review for adding HofX application and a ctest using ionosonde observations in PyIRI-JEDI
  • Added Open MP to Skylab to help enable optimization testing with larger experiments.
  • Doing some benchmarking with the HTLM optimizations
  • Adding some workflow changes to enable multi resolution with 4dvar.
  • Continuous DA is continuing...

OBS (Francois V)

IODA converter:

  • New converter for TEC data from Madrigal Database (space weather) #1658
  • New converter for TENET json data (space weather) #1717
  • Update the AHI decoder in obs2ioda #1714

UFO: 

  • The JCSDA OBS team continues to review and update the issues in UFO and ioda-converters.
  • There has been some technical debt reduction with the maintenance of IODA conventions (PR3797 (Co2 units) and PR3845 (Corrected the mismatch variable names in RadialVelocityUFO)
  • Zeroing out negative cloud-related quantities (mass content & effective radius) before passing to CRTM #3847.
    • Yannick added: negative value should be taken care of in all operators, not only CRTM.
  • Further we have removed unused operators PR3852 (Remove atmsfcinterp operator) and added a missing ctest for a bias predictor that wasn’t being tested PR3844 (Ctest for Predictor CosineOfLatitudeTimesOrbitNode).  
  • We are developing a backlog of items to be prioritized and working internally and with partners to address them.

CRTM (CD)

  • Evaluating AMSR3 coefficients delivered by NOAA-STAR.
  • Collaborating with the PolSIR PI (Vanderbilt University) to set up CRTM for the nature run.
  • Enabling future UFO simulations and data assimilation (DA) for PolSIR data, including setting up unit tests for the PolSIR instrument, updating the CRTM fix folder to include new coefficients, and preparing data ingestion once the nature run is ready.
  • Continue working with ShihWei on the UFO lidar operator PR.
  • General support for community users.

EMC (Dan)

  • Team found scaling issues with the RFS system
    • IO for FV3 fields identified as bottleneck
    • System doesn't scale well with MPI processes; optimal setup is about 4 MPI processes per node
    • Interpolation between model backgrounds and observations becomes slow as MPI processes increase
  • Exploring possibility of moving IO for FV3 into Atlas structure
  • Running 3D VAR on North America domain at 3km resolution takes about 550 seconds (vs. 750 seconds for GSI)
  • Experimenting with scale-dependent localization using Bump with offline scale separation

GMAO (Dorukhan, Maryam)

  • We are in the process of adapting to R2D2, so that's been the exciting part.
  • Maryam is working on Swell workflow to add GEOS-CF, successfully ran hofx experiment with R2D2 version 3.
  • Looking into using GOMSaver filter to save GeoVals for tempo observation

UKMO (Mike)

  • Trying to improve or allow for the increased number of processors by using split communicators, particularly to improve the performance of the covariances.
  • Issues discovered with multi-resolution 4DVAR - linear obs operator trajectory gets lower resolution when it should be higher resolution

NAVY (Dom)

  • New OneAPI compiler version 2025.3.0 fixes previous bugs but has issues with OpenMP threading

MMM (Chris)

  • Progress on MPAS configuration with higher top. Some plausible results from 3DVar.
  • Junmei working on running GMI H of X

  • LETKF. Double or single precision. The results we have say it matters, but not very much. So it's not a really a very definitive result on whether to make the change or not.

ROME (Hernan)

  • 4DVar testing on 3km resolution on US East Coast. 20 hours for 1 outerloop and 16 inner loop. Probably more results in the next week.
  • No labels