Non-ideal gas operator

[Huw Lewis]

Update forward operators to account for non-ideal gas/compressibility in the refractivity formula.

Sean Healy will lead on this work and has prototype code. A branch will be created (from v4.1 trunk) for this work.

[Ian Culverwell]

Test3_bangle.png

[Ian Culverwell]

Test3_bangle_frac_diff.png

[Ian Culverwell]

IT-FM-06_ifort9_linux.png

[Ian Culverwell]

Test2_refrac_good.gif

[Ian Culverwell]

Test2_bangle_good.gif

[Ian Culverwell]

Test2_bangle_all.gif

[Ian Culverwell]

This work has been carried out by Sean Healy in development branch sbh_ROPP5.

Compressibility/non-ideal gas effects, as described in GSR-09, are effected by means of a -comp switch in ropp_fm and ropp_1dvar tools.

Builds and runs under ifort9 on linux.

Scientific tests:

Tests of ropp_fm using data in IT-FM-05.nc (2D, 500 profiles) shows a pretty uniform increase in bending angle , consistent with the increase in k1 and the presence of Zd<1. The mean fractional increase in BA resulting from this would be expected to be about 0.115%. In practice it's this sort of magnitude, but varies with height, becoming negative above ~10km: Image(Test3_bangle_frac_diff.gif

Sean thinks latter effect is likely due to effects of compressibility in hydrostatic eqn. In GSR09 he suggests this could lead to dalpha/alpha ~-0.1% at 100hPa, which tallies with what we find.

Chris Burrows has run this through a modified version of IT-FM-02 (1000 randomly selected ECMWF model profiles) in the test folder. As for the 2D results above he found mean percentage changes in refractivity and bending angle of ~0.1% near the ground and -0.05% at height (bottom row):

Tests of ropp_1dvar using data in IT-1DVAR-04 (5 profiles) shows very small differences in {T, q, p} when using refractivities ({~0.02K, ~0.001g/kg, ~0.01hPa} resp)

but a bit larger when using bending angles ({~0.2K, ~0.2g/kg, ~0.1hPa} resp):

This is probably OK: the one profile that gave the large dq has a big change from the background in both cntl and test:

Omitting this profile we'd probably say mean dq ~0.02g/kg.

Sean has said these are the sorts of differences he'd expect to see. Ready for formal testing.

[Ian Culverwell]

Formal testing completed. Changes cause no problems. The "make test" tests (as carried out by buildpack, for example), have been extended to test the -comp options, and corresponding example figures have been added to the distribution.

[Ian Culverwell]

NB: typo in earlier text. Mean over all 500 profiles of fractinal difference in bangle resulting from comp changes is:

[Ian Culverwell]

Mark as fixed by agreement with Sean Healy.