fix for zero pressure for total ozone causing Inf in enkf_obsmod.f90

[jswhit]

Russ discovered this problem and here is a quote from his email

"Jim Jung compiled NOAA-EMC/GSI in debug mode. enkf.x seg faulted. We traced the problem back to the log pressure line shown below from src/enkf/enkf_obsmod.f90

! calculate locations of obs that passed initial screening in cartesian coords.
allocate(obloc(3,nobstot))
allocate(oblnp(nobstot)) ! log(p) at ob locations.
allocate(corrlengthsq(nobstot),lnsigl(nobstot),obtimel(nobstot))
lnsigl=1.e10
do nob=1,nobstot
oblnp(nob) = -log(obpress(nob)) ! distance measured in log(p) units

Total column ozone observations (e.g., omi, ompstc8) have a Reference_Pressure of 0.0 written to the netcdf diagnostic file. -log(0.0) is Infinity.

Looking at src/gsi/setupoz.f90, the 0.0 Reference_Pressure written to the ozone diagnostic file originates from the ozinfo file. This file has a pressure column. 0.000 is specified for total column ozone observations. It's not clear what Reference_Pressure actually means for a total column variable. This may be why we use 0.000 in the ozinfo file."

and my response

"If obs space localization is used ( modelspace_vloc=F) then the reference pressure is used as the ob location in the vertical for localization. If modelspace_vloc=T, I don't think it is used at all. The vertical location of a vertical integral is not a well defined quantity, so it's not clear what it should be set to. I agree it should be reset to a positive value if it's zero though, to avoid generating an infinite value."

[RussTreadon-NOAA]

PR #626 merged. Closing this issue.