Fix Oyr ocean resampling pipeline (enable resampling, yearly time/bounds/filename, calendar)

[rhaegar325]

PR Summary: Fix Oyr ocean resampling pipeline (enable resampling, yearly time/bounds/filename, calendar)

Overview

This change set fixes a chain of defects that prevented ACCESS-ESM1-5 ocean yearly
(Oyr) variables from being CMORised correctly, and tightens CF/WCRP compliance of the
resampled output. All fixes are in the producer (access_moppy); each is covered by unit
tests and verified end-to-end against real model output with the WCRP CMIP6
compliance-checker.

Problems fixed

#	Symptom	Root cause	Files
1	Ocean Oyr/Omon files have a dangling time:bounds but no time_bnds variable (compliance ATTR004/VAR004)	Ocean bounds calculation was commented out; method lived only on the atmosphere class	base.py, atmosphere.py, ocean.py (landed earlier, #455)
2	Oyr output still had 12 monthly steps instead of yearly	Driver never passed enable_resampling to the ocean CMORisers, and Ocean_CMORiser_OM2/OM3.__init__ did not accept it	driver.py, ocean.py
3	Single-year input crashed: "Need at least 2 time points to infer time bounds"	calculate_time_bounds could not infer frequency from one point	utilities.py, base.py
4	Yearly time coordinate landed on 31 Dec instead of mid-year	resample labels bins with the YE (year-end) boundary	utilities.py
5	Resampled output lost time:units/calendar (CF-invalid)	decode_cf moved units into encoding and they were never restored	utilities.py
6	Yearly filename included a month (010107-...)	Filename builder lumped yearly into the monthly YYYYMM branch	vocabulary_processors.py
7	Non-chronological (glob) file inputs crashed: "Index must be monotonic for resampling"	Out-of-order multi-file concat produced a non-monotonic time axis; the sort ran too late	utilities.py
8	[TIME001] yearly midpoint off by 1 day; time_bnds vs time calendar mismatch	The non-CF "GREGORIAN" label was read by cftime as the Julian/Gregorian standard calendar, while the file declares proleptic_gregorian	utilities.py

Changes

src/access_moppy/driver.py

• ACCESS_ESM_CMORiser.enable_resampling default False → True (no-op unless there is
  a genuine frequency mismatch).
• Forward validate_frequency / enable_resampling / resampling_method to both
  Ocean_CMORiser_OM2 and Ocean_CMORiser_OM3.

src/access_moppy/ocean.py

• Ocean_CMORiser_OM2/OM3.__init__ accept and forward the three resampling parameters to
  the base class (the link was previously broken at the subclass level).

src/access_moppy/utilities.py

• _shift_resampled_time_to_period_midpoint: centre the resampled time coordinate on the
  period midpoint (CMOR convention), reusing the existing per-period bounds helpers.
• resample_dataset_temporal: sort by time before resampling (monotonicity); normalise the
  "GREGORIAN" calendar label to "proleptic_gregorian" before decoding; restore the CF
  units/calendar encoding (numeric) after resampling.
• _normalise_calendar_name: map only the non-CF "GREGORIAN" to "proleptic_gregorian".
• calculate_time_bounds: add a freq_hint fallback for single-point axes; normalise the
  calendar before date arithmetic.

src/access_moppy/base.py

• _target_frequency_hint: derive a daily/monthly/yearly hint from the CMOR table.
• Pass that hint into calculate_time_bounds for the time axis.

src/access_moppy/vocabulary_processors.py

• Add a yearly branch to the filename builder: Oyr time range is year-only (YYYY-YYYY).

Behaviour change

• enable_resampling now defaults to True. Resampling is a no-op when the input
  frequency already matches the target (e.g. monthly→monthly), and only triggers on a real
  mismatch (e.g. monthly input for an Oyr table). Pass enable_resampling=False to opt
  out. Sea-ice is unaffected (the driver does not wire the flag into it); no existing test
  depended on the old default.

Tests

• test_utilities.py:
  • TestResampleTimeMidpoint: midpoint on ~2 July (incl. leap year), CF units preserved,
    unsorted input handled.
  • test_single_point_with_freq_hint, test_gregorian_label_treated_as_proleptic,
    test_normalise_calendar_name.
• test_vocabulary_processors.py: test_generate_filename_yearly_year_only.
• Full local suites green: test_utilities / test_ocean / test_atmosphere /
  test_driver / test_vocabulary_processors (the 7 failing test_cmip7_* cases are a
  pre-existing CMIP7 CV submodule issue, unrelated to this change).

module use /g/data/xp65/public/modules && module load conda/analysis3
PYTHONPATH=src python -m pytest tests/unit/test_utilities.py tests/unit/test_ocean.py \
  tests/unit/test_driver.py tests/unit/test_vocabulary_processors.py -q

Compliance verification

Run against a real CMORised file
(no3_Oyr_ACCESS-ESM1-5_historical_r1i1p1f1_gn_0101-0110.nc):

• ✅ time_bnds present (ATTR004/VAR004 cleared).
• ✅ [TIME001] time-squareness passes — time[0] = 36706.5, the proleptic_gregorian
  yearly midpoint the checker expects.
• ✅ CF §7.1 time_bnds/time calendar mismatch cleared (both proleptic_gregorian).
• ✅ Yearly filename is YYYY-YYYY.

[codecov]

Codecov Report

❌ Patch coverage is 98.14815% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 77.2%. Comparing base (7142b03) to head (8eab7f2).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/access_moppy/utilities.py	97.2%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##            main    #463     +/-   ##
=======================================
+ Coverage   76.9%   77.2%   +0.3%     
=======================================
  Files         31      31             
  Lines       5974    6024     +50     
  Branches    1107    1119     +12     
=======================================
+ Hits        4594    4648     +54     
+ Misses      1131    1126      -5     
- Partials     249     250      +1     

Flag	Coverage Δ
unit	77.2% <98.1%> (+0.3%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[rbeucher]

Can you link the issues that this will close?

[rhaegar325]

solved #454 #461

[rbeucher]

How much of that also applied to ESM1-6 and CMIP7. I think we should check to make sure that we don't just use code for CMIP6.

[rbeucher]

Most of this PR should apply to ACCESS-ESM1-6 and CMIP7, since the resampling/time-bounds/calendar fixes are in shared code paths used after CMIP7→CMIP6 mapping. Two things to double-check: CMIP7 yearly filename formatting (this change looks CMIP6-only), and OM2 source_id acceptance for ACCESS-ESM1-6 so ocean CMORisation isn’t blocked independently.