Segmentation fault on occultations with negative refrac at high altitudes

[Stig Syndergaard]

In a few cases (two found), the refractivity, even after stat.opt., is negative at around 100 km. This cause trouble downstream in the codes, with log taken to negative values, and eventually result in a large array index (largest negative integer possible), and then the crash.

The trouble starts in ropp_pp_invert_refraction.f90, line 144, where the scale height becomes NaN:

scale = -(impact_nh(n+nh) - impact_nh(3*(n+nh)/4)) / &

(LOG(bangle_nh(n+nh)) - LOG(bangle_nh(3*(n+nh)/4)))

And ends with a crash in ropp_pp_tdry.f90, where KZ becomes -2147483648.

The problem should be solved by checking the refractivity at an early stage, and stop there. A long term solution would be a better stat.opt., less sensitive to the noise from ionospheric scintillations.