Atmospheric refraction at MKO compresses the PFI field by a bit more 0.2 arcseconds at an altitude of 60 degrees, so edge cobras are offset by c. 0.1 arcseconds which is enough to lose some S/N.  At higher airmass the problem gets rapidly worse.

We deal with this by designing the plate for a particular time, but this won't always be correct;  errors of 10 degrees near zenith aren't likely to be important, but as we observe harder targets (especially ones that are setting) the effect becomes larger.

Fortunately there's an easy fix.  We already rewrite the pfsDesign file to form the pfsConfig file for a given observation.  If FPS updates the mm from (ra, dec) we can make the small corrections due to refraction at the same time.  The corrections, while significant on the scale of a fibre, are tiny on the scale of the patrol region so there is no need to redo the whole plate design.

yuki.moritani points out that the code at https://github.com/Subaru-PFS/pfs_utils/blob/tickets/INSTRM-1582/python/pfs/utils/coordinates/CoordTransp.py#L31 does exactly what is needed (note that this is the INSTRM-1582 branch, and that the API changed a little with this branch so we should indeed use this version).

We should really think about how long we expect to integrate on the field, and design it for an airmass in the middle of that time; but it's probably not worth worrying about that now. Refraction also depends on the pressure, but the changes introduced by this are almost certainly too small to be worth worrying about.