Diag: Clear Air Turbulence

[bblay-mo]

This ticket is a directive for Claude:

Migrate the UM diagnostic "Clear Air Turbulence" to LFRic.
The UM code can be found in atmosphere/PWS_diagnostics/pws_cat_mod.F90.

Edit: also max cat and max cat pressure.

[bblay-mo]

Claude's plan summary:

What the UM does.
The Clear Air Turbulence family is three STASH items in one driver block. Item 16 is Dutton's empirical CAT predictor on pressure levels. Per level it takes vertical wind shear from a six-point cubic-spline derivative of height, u and v with respect to pressure, horizontal shear from centred differences of the 300, 250 or 200 hPa winds, combines them as 10.5 + 1.5*(ShearH + ShearV), and maps that onto a 0 to 7.5 index through three fixed knots. Items 17 and 18 are the per-column maximum of item 16 and the pressure at which it occurs.

Proposed LFRic port.
A bundle of three fields under the aviation__ prefix, one new algorithm module, three new kernels and their unit tests, called from the diagnostics driver next to the freezing-level diagnostic:

• A column kernel porting the UM spline derivative line for line.
• A cross-stencil kernel for the horizontal derivatives and the Dutton knot interpolation, using the same per-cell cubed-sphere geometry as the issue 781 vorticity plan.
• A small in-place maximum kernel for items 17 and 18.

Item 16 is Group I with the Group G activation modifier. Items 17 and 18 are Group C. Prior-art check found nothing on this branch, upstream/stable or upstream/main. Complexity is rated Complex.

Decisions I need from you (full list of eleven in the plan):

1. Scope. The issue names only item 16, but items 17 and 18 come from the same block for one extra kernel. I planned all three. Confirm or restrict.
2. Pressure levels. Default is the standard pressure_levels axis with values computed only at the slots that round to 300, 250 or 200 hPa and missing data elsewhere. Alternatives are all axis levels, or a bespoke axis that would need a rose-stem change.
3. Maximum tie-breaking. The UM keeps the first level in its STASH list. LFRic would scan in axis order, which gives 300 hPa first. This matters for item 18 wherever the index saturates or is zero at more than one level.
4. Field ids. aviation__clear_air_turb and siblings, or plev__clear_air_turb for the pressure-level one.
5. Edge cells. Missing data at cells with a missing stencil branch, which is what the UM does at its LAM edges and polar rows.

Two things worth knowing that shaped the plan. The UM silently reuses stale winds when a requested level is not one of the three valid ones, so those slots are garbage there. I did not port that and mark them missing instead. Also, the Group I skill reference says pressure-level kernels need PSyKAl-lite wrappers, but the real constraint is the real-array level argument. Passing the level as a scalar, one invoke per level as the UM loop does, avoids the wrapper. That is recorded as a skill gap.