I propose to define the wavelength window to compute S/N in the r or m-arm because we reqire FLUXSTDs to have these arms for fluxCal. The exact wavelength window I suggest is 840-880nm. This is chosen as the region (1) without telluric absorption, (2) with average-ish sky lines, and (3) with CaII triplet lines, which are very important for stellar typing.

The spectra in the figure are pfsMerged.

As noted in PIPE2D-1518, this S/N cut is useful for identifying fibers that did not converge. Another reason to implement this feature. We want to implement the S/N cut not just in fitFluxCal.py but in fitPfsFluxReference.py as well so that we can exclude stars with no S/N entirely from fluxCal.

What is the definition of "S/N"? Is it sum(flux) / sqrt(sum(variance))?

Sorry for being unclear. This is the median S/N per pixel in the 840-880nm wavelength window. So, if I write the per-pixel signal-to-noise as SN_i = flux_i / sqrt(variance_i), then the S/N I meant here is the median of SN_i within the wavelength window. I am still a bit worried about the sky residual, and I would prefer to adopt a robust S/N measurement.

Could you review this pull request, too?

Merged. Thanks for reviewing PR.