[cfranken]

Version 0.5.0 → 0.6.0 (breaking in 0.x), CITATION.cff, and the CHANGELOG 0.6.0 section: intro, summary of PRs #64–#87, binary regeneration requirements, and thanks to pera-noaa (NOAA GML) and Kay Suselj.

After merging, @JuliaRegistrator register will be posted on the merge commit with these release notes (owner-approved); TagBot then creates the tag and GitHub release.

AtmosTransport.jl v0.6.0

Breaking release: corrected transport numerics, observation placement and
production-matched adjoints, with expanded retrieval-column workflows.

Breaking changes

• Numerical results change: Float32 column-mass ledgers retain residuals
  (Column mass ledger: persistent per-column carry (TM5, dkg/VDIFF, CMFMC) — OF-110, OF-111 #65); Kz diffusion uses flux-form mass updates (OF-113, Kz vertical diffusion: flux-form mass update with column ledger, Float32 drift fixed (OF-113) #81).
• Surface-flux series must cover the run; static fluxes default to
  conservative regridding on every grid (Surface fluxes: time-coverage check, conservative LL/RG default, extensive regrid (#59) #70).
• ERA5 spectral pole fluxes/humidity registration (ERA5 spectral pole rows, lat-lon humidity registration, merge pass (#58, #63) #72), TM5 half levels
  (TM5 convection: ERA5 → TM5 half-level indexing; retire lat-lon physics-file route (#54, #55) #69), CFL subcycling, positivity and Lin–Rood ORD=7 seams (CFL subcycle composition; limiter; Lin-Rood ORD=7 seams (#61, #62) #68) are corrected.
• Regenerate affected ERA5 spectral lat-lon binaries and their CS regrids,
  ERA5 N320 TM5-convection binaries predating tm5_halflevel_v2, and removed
  lat-lon TM5-route binaries/regrids. Regenerate GEOS-FP C720 → C720/C180
  binaries built with the latlon_025 physics fallback (OF-119, OF-119: GEOS-FP lat-lon physics fallback — vertical order and latitude interpolation #87).
• Observation/departure files require schema v2; v1 is rejected (4D-Var: observations placed in the column, schema v2, time windows (#20, part 3) #77).
  Placement defaults to total pressure/virtual-temperature heights; dry-basis
  runs need humidity or explicit pressure_basis = "dry" (Observations: ordered vertical placement (height above ground / pressure) and total-pressure basis (#20, part 2) #76, Observation heights from total pressure and virtual temperature (OF-117) #82).
• Adjoint APIs/structs change; replay uses the forward step plan and rejects
  unsupported operators (Adjoint identical to the forward by construction: production StepPlan replayed by the adjoint (D31, OF-114, OF-115) #83). The lat-lon ERA5 physics-file TM5 route and
  its API/configs are removed (TM5 convection: ERA5 → TM5 half-level indexing; retire lat-lon physics-file route (#54, #55) #69); uncovered experimental C30 configs and
  launchers (Surface fluxes: time-coverage check, conservative LL/RG default, extensive regrid (#59) #70) and the separate adjoint replay composition (Adjoint identical to the forward by construction: production StepPlan replayed by the adjoint (D31, OF-114, OF-115) #83) are retired.

Highlights

• Placed/time-window observations, AK retrieval columns and OCO-2/OCO-3 MIP
  conversion, co-sampling and cell averages (Observations: ordered vertical placement (height above ground / pressure) and total-pressure basis (#20, part 2) #76–4D-Var: averaging-kernel retrieval columns, TM5-4DVar operator (#20, part 4) #78, OCO-2/OCO-3 MIP columns: projection onto OCO's level basis, converter, co-sampling, cell averages, sampler AK output #80).
• Single-sweep 4D-Var gradients (4D-Var: one adjoint sweep per gradient (#20, part 1) #75) and corrected host tapes/replays (Adjoint: replays and tapes follow the production step; separate tape read caches (#57, #60) #67).
• GEOS-FP instantaneous CTM endpoints (GEOS-FP native C720: window endpoints from the instantaneous CTM files #84) and physics fallback fix (OF-119: GEOS-FP lat-lon physics fallback — vertical order and latitude interpolation #87);
  GEOS-IT EFLUX enables non-local VDIFF (Preprocessing: pass GEOS-IT EFLUX through to pbl_eflux #73).
• AMD ROCm backend via AMDGPU.jl; not yet run on AMD hardware (AMD GPU (ROCm) backend via AMDGPU.jl extension #86).
• Cubed-sphere driven runs reported 2.6× faster at one tracer (Cubed-sphere driven runs: 2.6× faster at one tracer, bit-identical (met staging, fused launches, fewer syncs) #79).
• CI explicitly tests Julia 1.10, 1.12 and 1.13 (CI: test Julia 1.10, 1.12 and 1.13 #85).

Thanks to pera-noaa (NOAA GML) for issues #54–#63 and #20, and Kay Suselj for #73.
Details and migration: CHANGELOG v0.6.0.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Qfkez3tXkdGpzp8H8U6PPz

[JuliaRegistrator]

Comments on pull requests will not trigger Registrator, as it is disabled. Please try commenting on a commit or issue.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.