JEDI Infrastructure: Evan Parker

We had two presenters at UIFCW 24

  • Eric gave a talk on R2D2 in the in System Architecture session
  • Ashley presented a poster on JEDI Infrastructure projects

R2D2:

  • Eric tracked and fixed a two bugs that had affected the record update functions in some cases.

Spack-stack

  • FMS 2023.04 did not play well with MacOS, where we saw most fv3jedi crests failing in jedi-bundle. This has been repaired and a workaround is in place until we can get the proper fixes in place. Contact SteveH is you are needing this workaround (any MacOS).
  • Appears that we are converging  on Atlas 0.38.1 for the upcoming spack-stack-1.8.0 release (Aug/Sep). If you have comments on this decision, see spack-stack issue 1155.
  • We've added more updates to the spack-stack AWS PCluster site config allowing use of the 3.9.3 release from June of this year


Algo Team: Steve H

SteveV

  • Enhancement to VADER linear variable changes is near completion.
  • All model interfaces that have Vader linear variable changes implemented will need to adapt since the Vader methods have changed their signatures, so people should keep an eye out for an announcement on that coming shortly.

Christian

  • Beginning to really frame the changes needed in ioda and oops to add new observations between outer loops.
  • HTLM with dual model, one for the non linear control and one for the ensemble as well as the general pseudo model is ready for a PR in EWOK/SKYLAB

Clémentine:

  • A simple 3dvar case with GEOS is in review in skylab. I am still working on adding ensemble B, GSI static B and observations / yamls similar to the ones used at GMAO
  • Working on running the converter for restart files so we can add GEOS backgrounds to the ingest suite in our workflow
  • Continuing conversation on how to form an analysis correctly (adding increment and propagating the information to all dependent variables). This is a conversation that is led by the models interface team

Nate

  • Added a feature to the generic StdDev SABER block that allows you to apply a multiplicative factor to the block (useful for tuning Jo/Jb ratio)
  • Made a small modification to SpectralToGauss block to add/preserve atlas field metadata which was needed for compatibility with oops interpolator
  • Been working on a single obs MPAS experiment with spectral filter as a static B. Made good progress thanks to PR in MPAS-jedi that converts MPAS field names to CCPP names.

SteveH

  • Work is underway to create an ObsSpace::append function that will support continuous DA
  • Work is underway to enable instantiation of the new ObsDataFrame memory container inside an ObsSpace. This will enable us to test with real world data, and do comparisons with the current ObsGroup container

Models Interface: Francois Herbert

- Liam is continuing investigations of gcc's suitablility for openmp offload with cuda memory management (primarily working to understand and mitigate issues arising from the interplay between compiler optimizations and openmp offload). Tested building jedi-bundle with non-spack-stack atlas.

- Francois is cleaning interfaces for the generic global interpolator as work towards applying the global interpolator for resolution changes in fv3jedi + mpasjedi; look for PRs improving the resolution-change interpolations in these model interfaces in the weeks ahead.

Observations: Hui

  • Space weather data decoder: Ionosphere decoder PRs are ready for ionosonde and TEC. Also added ctest and tried with multiple files. 
  • PyIRI-JEDI is being updated to be built with OOPS in develop. Additional utilities were also added for horizontal interpolation. To be used in space weather
  • Space weather workshop (JEDI kick-off meeting) will be on Aug 28-29, Boulder, CO. Hybrid option is provided. It is a collaboration meeting with ~30 ppl already including in-kinds and external collaborators. 
  • Adding QC flags to TL/AD, together with MetOffice in-kinds. It will adopt the QC flags from all filters and therefore reduce the computing time and memory use. 
  • Radar operator PR from MetOffice: The OBS team is working to coordinate this work with existing capabilities to ensure the code is generic. 
  • Continue to verify OSW observations using JEDI-skylab and focus on the impact studies using different scores.
  • Also remind people to label their PR with "OBS". The OBS team is clearing the Zenhub board by reviewing issues and PRs. 

Question from Chris Snyder: is the idea with radar to have a a single operator that works for everyone? Hui: yes, but we are seeing if we need two b/c there is overlap but differences. Greg: yes, the idea, very adaptable to different schemes. 

CRTM: Cheng Dang

  • Both Ben and Cheng presented at UIFCW this week, on the recent developments of CRTM.

Developing:

  • Two PRs are ready in preparation for the CRTM v3.1.1 release. Cleaning up a lot of issues from the past. 
  • CRTM team is working with the PolSIR project team for coefficient generation and CRTM general support.

Aerosol:

  • CRTM team is working with STAR in-kind contributors on simulations for aircraft-level and ground-based sensors. New features in following quarter. 


Question from Jerome Barre: going to use JEDI with James/GEMS?. Cheng: not right now, depends on MPAS dev. They need to update their side.

composition: Jerome Barre

  1. geos-cf trace gas cycling with eda and high res runs. Using geos fp system and IAU to perturbs the meteorology on forecasts.
  2. working on ingest of new observation: aeromma field campaign for scientific evaluation (from last August) of the tras gas suite
  3. progress on L1b aerosol assimilation. working on Filtering cloud vs aerosol pixels for reflectance DA with CRTM. 

Soca: Travis Sluka

  • Hamideh: continuing work on coupled OASIM hofx
  • Kriti: continuing work on ensemble TLM for SOCA 4DVAR
  • Travis: continuing my march toward deleting/refactoring lots of SOCA code.
    • I'm about to issue several PRs that switch the skylab marine experiments to using the new generic diffusion in saber, and deleting the model specific diffusion code from SOCA.
    • I'm in the process of refactoring all of our linear variable changes into proper saber blocks. These will live in the soca repo, for now, but some useful blocks (such as the parametric ocean stddev) can be moved into saber at a later date.

Travis asked if people noticed Orion or Hercules being really slow or just SOCA. Nate commented that he did not notice. 

EMC: ...

GMAO: Ron Gelaro

Not much to say. Him and Ricardo have been on vacation.

Tracking down stuff with the atmospheric observation system. Seeing differences in increments in ozone in 10mBar gsi. Looking at jacobians. Other thing they are working on is seeing ocean surface increments. More info next week as people return. 

From atmospheric side: Doruk said nothing right now. 

MettOffice: Phil Underwood

Summer time, not big update. 

We are playing with SABER to test different ways to reduce the interpolation errors in our B matrix. 

Steve H said: Travis check out this comment in spack-stack issues: https://github.com/JCSDA/spack-stack/issues/1166#issuecomment-2226591030

Navy: Sarah King

Starting to look at high altitude in the model. Will start looking at the observations for that soon. 

ROMS: Hernan Arango

Lots of progress with 4dvar. The issue is: since ROMS has a staggered grid for velocity, he is dealing with adjoint - initialize the TL and then need to multiply with transpose but BUMP doesn’t allow for that transformation so need to put at center of cell and put back into the increment. Bump out of commission so he used the identity matrix, but it sucks. So he went to the diffusion operator, he likes how it spreads into the diffusion but he suspects a bug b/c not doing the adjoint. He sees a straight line when putting it at the center. So he doesn’t know where the diffusion is doing the changes. He thought Travis did this in atlas. Travis said, yes atlas is doing the exchange. But it could be a bug. 

MMM: Chris Snyder

Main thing is working with MPAS code. Parts they need come from JCSDA internal fork. The plan is to pull directly from the main MPAS repo. they are close. They have PRs back in MPAS for their branch. They are testing. Hoping to switch over soon in a few weeks (not months). He started a discussion if people want to keep track: https://github.com/orgs/JCSDA-internal/discussions/145. There are changes in how the file IO is working.

  • No labels