Notes taken by Ben Johnson
INFRA
- Liam investigated using GCC's OpenMP offload in conjunction with the CUDA runtime's memory management. Worked on setting-up Azure VM for GPU offload development, including setting up spack-stack and jedi-bundle.
- Francois finished work to make the generic interpolator more robust on structured grids, unlocking generic interpolations for saber blocks (spectral B, GSI, etc)
ALGO
Upcoming merges: next Thursday the merge for using vader for xb/fg variable changes in saber: https://github.com/orgs/JCSDA-internal/discussions/143
Updates from the team:
- Clementine isorking on the 3dvar geos skylab experiment, working with GMAO on geos-jedi exchange. Working on the fv3-jedi geos interface to make it as compatible with the current one as possible.
- Nate set up a test using spectral filters for localization with mpas, and is experimenting with different resolutions.
- Steve V is working on improvements to linear variable changes in vader that would allow to use vader for linear variable changes in saber.
- Steve H finalized specifying multiple files for the obs space; this is ready for thorough testing.
- Christian is working with Tom Hill on HTLM with pseudo model; outer loops in skylab. Starting to look into adding more observations at each outer loop (for continuous DA).
MetOffice
Observation processing
- A coupled (atmosphere and ocean) trial has successfully completed with JOPA being used for the first time in atmosphere and ocean.
- IASI, CrIS and AIRs are now part of our JADA setup. This is our first use of correlated observation errors, and this is working in a comparable way to our current system.
- Work has started on the reflectivity forward operator. The idea is to provide a generic shell with a simple operator to start with. Ultimately, we want a full operator accounting for all the physics involved.
General points:
-Our covariance work is currently concentrated on diagnostics and calibration.
SOCA
nothing exciting, but if you really need something:
- OASIM has been added to the skylab marine hofx test
- working on ensemble TLM 4dvar with regional soca
- Travis continues to delete SOCA code, hopefully he doesn't go too far.
Hernan:
- The ROMS-JEDI has successfully passed the LinearModel unit test, a significant milestone in our project. The symmetry test between the tangent linear and adjoint model operators yields a dot product difference of an order of 10E-13 between the initial and final state vectors, regardless of the initialization values. It means that the discrete symmetry between TL/AD operators is within the acceptable roundoff value, ensuring the convergence of the 4D-Var minimization.
- We are working and using VADER compliance in preparation for SABER requirement in the next few days.
- The currently fully functional ROMS-JEDI interface provides a solid foundation for us to continue our data assimilation work with JEDI confidently.
- We started playing with 4D-Var applications and having some technical difficulties reported on https://github.com/JCSDA-internal/oops/issues/2697
NCAR:
- Jake Liu: found recent code MPAS-JEDI memory use has a big jump, traced to PR on June 21, made an OOPS issue. Traced to 2 possible lines of code.
- BJ made an issue, Kate Huxtable at MO looking at alternative solution, being worked on.
NASA:
- Tested cycling varbc with the full observing system
- Evaluated results by comparing GSI and JEDI over an extended period, exceeding a week
- Examined preliminary results produced by cycling ocean + sea-ice DA using SWELL
OBS:
IODA converters
- Collaboration with NASA and Isaac Moradi, have an active draft PR for CloudSat Precipitation Radar. Isaac is now running Skylab on Discover, and we plan to test the converter and adapt it for the GPM DPR.
- NRL provided some ionosonde data, a decoder has been introduced into IODA converters and is in draft state. There will be subsequent work in IODA conventions to implement this data.
UFO development:
- A yaml option was introduced to the ROPP interface, this then requests the compressibility to the atmosphere, deviating from the ideal gas assumption.
- Dr. Nghi Do from UCSD (Jennifer Hasse's team) was in Boulder for an in-kind visit, this saw the implementation of the airborne radio occultation observations from the Atmospheric River reconnaissance field program (ar_recon) into Skylab. These have been tested with JEDI using both MPAS and FV-3 model interfaces.
- NOAA EMC, and Dr. Emily Liu helped to verify new features in UFO, specifically the correct passing of OOPS flags into UFO operators, particularly with the CRTM interface was validated. Secondly, Dr. Liu tested the new action to reduce observation space based on an OBS pre-filter, this testing uncovered an issue that prompted an OOPS PR that was merged and this action is now behaving across pre-filters as expected.