JEDI Algorithm Team (Christian Sampson)
PR adds OpenMP to HTLM
updateIncTLandupdateIncAD.Will test speedups on larger test cases.
Starting deeper work on continuous DA with hybrid + ensemble B.
Git-LFS test files separated for SABER/VADER (PRs merged).
Doruk Ardag (Marine / SWELL / R2D2 / Workflow)
Shutdown
...
: Only contractors &
...
co-ops working.
Ricardo (pre-furlough):
Multi-outer-loop tests
...
: UFO filters may not apply in later loops unless a
...
switch is enabled (
...
name TBD)
...
→ consider default True.
Possible
...
outer-loop state
...
propagation issue tied to SWELL workflow
...
config.
Marine:
...
Local coupled tests; R2D2v3 now
...
tested in SWELL.
Onboarding:
...
Jerome
...
/Maryam
...
ramping on SWELL; leveraging
...
R2D2
...
; interest in EVA (EMC-relevant
...
).
...
Workflows / Infrastructure (Eric L. / Ashley / Evan / Dom)
cylc in Skylab:
...
Toy models testing progressing (Ashley).
R2D2:
...
Client/server
...
debt
...
cleanup (Eric).
spack-stack: Feature branch with
...
Spack 1.0 integration
...
→ basis for
...
2.0 (target early Nov) (Evan).
CI:
...
Issue causing excessive timeouts in 5–10% of builds
...
; triage ongoing (Evan).
UKMO (Michael Cooke
...
)
ECTRANS: Plan for multiple split MPI communicators
...
to run ensemble waveband B models
...
in parallel (
...
remove sequential
...
hybrid-B
...
overhead).
4D-Var:
...
Focusing on
...
initial setup
...
speed (
...
profiling/mitigation in progress).
Regional JOPA: Observation
...
configs nearly ready for final trials.
OBS (IODA/UFO/USO, François V.)
IODA
...
PR-1701 (MRMS): Code fixed; smaller files.
UFO: Surface operator ↔ GFS operator
...
interface concerns under review.
Unified Super Operator:
...
Owners/next steps with Ricardo / François V.
...
CRTM (Ben Johnson)
Sea-ice model investigation ongoing.
- IFX testing and evaluation of ongoing segfaults, flang compilation testing ongoing
- working with NOAA STAR on AMSR3 coefficients (ODPS+ODAS), Cloud Coefficients (extending to submm)
- working with Isaac Moradi at GMAO on Cloud Coefficients (extending to submm)
- Cheng Dang returning on PolSIR funding.
NOAA/EMC
...
NWS exempt from furloughs.
JEDI R&D / Real-time: Two real-time “upgradeS” GFSv17 full-res parallels; GCAFS (aerosols) repo: https://github.com/NOAA-EMC/GCAFS.
Retrospectives: Start end of month; science freeze for retro; 2 hurricane seasons; ~6 months to generate. Large res + ocean increases compute.
Obs format for JEDI ingest: Finished.
Regional: Background error model work; PR in SABER to turn on all testing for GSI-BEC.
HAFS-JEDI: Running real-time for multiple storms.
Optimization: With GDIT; NetCDF chunking for HAFS and RRFS.
NCAR/MMM (Jake Liu)
spack-stack 1.9.2 working well; switching to it.
François Hébert — Interfaces & Optimization
(Update w/ Dan Holdaway)
Furlough status: EMC excepted; business-as-usual.
GFSv17: Real-time full-res parallel with JEDI for snow/ocean/sea-ice DA.
GCAFSv1: Real-time full-res parallel with JEDI aerosols.
Regional: Finishing static B using GSI-BEC.
Atmosphere: Low-res real-time JEDI-atmos parallel running alongside degraded GSI.
GNSS-RO: Continuing operator debugging in JEDI.
Throughput/I/O: Ongoing NetCDF chunking optimization work for HAFS/RRFS with GDIT.
(From prior) Retrospectives for GFS planned to start end of month; science freeze during retro; 2 hurricane seasons (~6 months to generate).
NCAR/MMM (Jake Liu) — Modified
MPAS-bundle (develop) currently building with spack-stack-1.8.0 on Derecho.
Tests show spack-stack-1.9.2 gives comparable memory & speed to 1.8.2 for Intel & GNU; Intel build uses slightly more memory.
Plan: Update MPAS-bundle develop to spack-stack-1.9.2 after additional tests.
Interfaces & Optimization (François Hébert)
Static spectral B: Multithreading complete
...
; next: waveband ensemble B, variable changes, and HTLM items for next deliverable.
OSDF-based reader:
...
Load step nearing completion this sprint.
CO₂ units
...
: Standardized to mol/mol (model-interface → GeoVaLs → UFO
...
).
obs2ioda: yaml-cpp
...
→ eckit.
...
SABER/BUMP /
...
Fortran allocatables & methodology: Memory not released in some paths; need disciplined allocate if not allocated / deallocate pattern. Will discuss with Benjamin M.
Tutorial (Hernan Arango)
ROMS-JEDI tutorial completed; thanks to contributors.
Mac (M4) builds: Spack-Stack built with gfortran; several Mac adjustments identified and reported; aware that Mac support is no longer official.
BUMP runtime: Found cases of DEALLOCATE on non-allocated local → runtime errors / leaks; appears to be a compiler bug.
Coding guidance: Use guards:
IF (.NOT. ALLOCATED(var)) ALLOCATE(...)IF (ALLOCATED(var)) DEALLOCATE(var)
Apply broadly in BUMP and related Fortran to harden runs (HPC/cloud)
...
.