Make domain_stats use spatial_average and handle 0-360 longitudes

[brianpm-ucar]

Closes #262.

domain_stats now takes its mean with spatial_average instead of its own copy of the cos(lat) weights, as the issue asked.

While doing that I found a bug in the same function. make_polar_plot calls it with [-180, 180, 45, 90], but the regridded data has longitude 0 to 360, and the old sel(lon=slice(-180, 180)) kept only longitudes 0 to 180. The Mean, Max and Min in the polar-map header therefore described half of the cap. The old code also returned an empty selection for latitude that runs north to south.

Change

• The region is picked with masks on lon % 360 and on latitude, so the longitude convention of the data and of the domain no longer has to match. If west is greater than east the box wraps through 0/360, and a span of 360 degrees or more is the whole circle.
• The mean is spatial_average(region). Min and max are as before.
• New tests in lib/test/unit_tests/test_domain_stats.py: same answer for 0-360 and -180-180 data, descending latitude, a box across 0/360, box edges on 0, 180 and 360, NaN points left out, and the global box equal to spatial_average. Each expected value is a separate numpy calculation. Against the old function, 5 of the 6 tests fail.

Effect on real output (PS climatology from the t1 model-vs-model run, longitude 0 to 358.75; mean, max, min in hPa):

Region	Old	New
North cap (45N-90N)	985.71, 1031.65, 756.02	984.57, 1031.65, 661.54
South cap (90S-45S)	928.81, 1011.12, 580.60	952.89, 1011.12, 580.60

For data already on -180 to 180 with ascending latitude, old and new give identical values.

Checking

• pytest lib/test/unit_tests: 269 passed. black --check lib/ is clean.
• A separate read-only Claude agent reviewed the change with no blocking findings. I tightened the docstring (2-D input; the old code silently averaged over any extra dimension, the new one raises for a time dimension longer than 1) and two tests after its review.

Not checked

• A full ADF run with the polar-plot headers. The only caller, make_polar_plot, passes 2-D seasonal fields.
• The black version pinned for CI (I used the p12 environment's).

Not changed here: make_polar_plot still fills the percent-difference panel with 0 before its statistics, and still selects latitude with sel(slice) for the plot itself.

Claude-assisted: written with Claude Code.

🤖 Generated with Claude Code