dbekaert / RAiDER / ace00848-29c8-45be-b172-8e1f4168e8f9

54%

dev: 53%

Pull #826

circleci

restore the half-level average in calcgeoh's full-level pressure

ECMWF hybrid coordinates define pressure on half levels; the value belonging
to a full level is the mean of the two that bracket it. calcgeoh returned
ph_lev alone -- the half level ABOVE -- while pairing it with
geopotential[idx_level], which is a full-level height. Every level was handed
a pressure belonging roughly half a model layer higher (~145 m through the mid
troposphere), so every level came out too light and the integrated hydrostatic
delay with it.

The averaging was present and correct between fc7d0f59 (2021-06-30) and
459763124 (2021-12-04), where it was commented out inside a 144-line move from
weatherModel.py to utilFcns.py; 44ebc41b replayed that onto dev. v0.1.0 is the
only release that ever shipped it correct.

On a moist US Standard Atmosphere sampled onto the 137-level HRES grid the
defect costs -42.24 mm of ZTD. Generalising the choice to
p = (1-a)*ph_above + a*ph_below, the error is linear at +84.5 mm per unit a and
zero at the ECMWF definition a = 0.5; 0.5.3 sits at a = 0.

Adds two test modules:

  test_calcgeoh_pressure.py     asserts the returned pressure is exactly the
                                mean of the bracketing half levels, guards
                                against a silent revert, and quantifies the
                                delay cost from the coefficients alone.

  test_downstream_opera_tropo.py  a contract test for OPERA L4 TROPO-ZENITH,
                                which drives the HRES model-level path
                                directly. It compares the computed ZHD against
                                Saastamoinen's closed form rather than a
                                recorded golden, so it cannot be regenerated
                                into defending a wrong answer. On its 8 KB
                                nine-column fixture a correct RAiDER gives
                                1.0 mm RMS; the unfixed... (continued)

1 of 1 new or added line in 1 file covered. (100.0%)

3479 of 6499 relevant lines covered (53.53%)

0.54 hits per line