Masayuki Tanaka writes:

Starting with a background, we have used all bright stars within
each CCD to calibrate the photometric zero-points. But, red stars
tend to have a large intrinsic color scatter due to metallicity
variations. For the photometric calibration, we would like to use
stars with a small intrinsic scatter only. I have looked into
the color range we should use to select stars for the photometric
calibration.

I used SRCMATCHFULL*fits from the S17A processing run. Somewhat
arbitrarily, I picked the D-COSMOS field, where the Galactic
extinction is small and the area is relatively large. The exposure
time is slightly longer than the Wide exposure, but we are talking
about very bright stars and that is probably not a concern.
I used the visits 24318-24332; this is 50%:50% combination of
30:270 sec exposures. This is the i-band data, but I realized
I should probably look at some y-band data as well to see if
the following results do not have strong filter dependence.

The first plot [gri.png] shows r-i plotted against g-r. PS1 stars are in
black. The original PS1 catalog seems to have a lot of contamination
from galaxies, but the good star/galaxy from HSC makes the matched
catalog mostly clean, although there are some outliers from
the stellar sequence and they may still be galaxies. Most of
the stars here are i<~21. The red squares are synthetic PS1 colors
from the Pickles stellar library just for reference.

As you see, the width of the stellar sequence increases around
g-r~1.2 and r-i>~0.5. This area is dominated by M-stars.
I would propose to use stars with g-r>0 and r-i<0.5 shown as
the solid line. For reference, a K6V star has g-r~1.2 and r-i~0.5.
So, I am cutting stars later than mid-K. I am not sure if we need
to cut on the blue edge, but that probably does not matter too much
because we have few stars earlier than A0V. About 40% of
the selected stars satisfy this color cut, and the number of stars
in each CCD is shown in the 2nd plot [hist.png]. The open histogram shows
all stars in each CCD and the filled histogram is for stars within
the color box.

On average, we have ~20 stars and I think that is sufficient.
COSMOS is a low stellar density field and so this is probably
a lower limit, but general-use observers may take a much shorter
exposure under poor conditions and run into a trouble. So, we may
want to have a fallback mechanism, e.g., first apply the color cut,
but if that results in too few stars (e.g., <5 stars), do not apply
the cut and do the calibration with all available stars.

John Parejko may be interested in using this information and/or code in jointcal.

John Parejko, thanks for agreeing to review this. I hope that there are some elements in here that will be useful to you for your work on jointcal.

Jenkins passed.

I've revised everything according to the existing comments, but as far as I can tell meas_algorithms hasn't yet been reviewed.

Yes, I'm still finishing the meas_algorithms review.

Thanks John. I think you're done reviewing now and have only forgotten to switch the status here.

I've tweaked meas_algorithms as well and am running everything through Jenkins. I would like to merge in the next day or so, so please shout soon if there are any objections to the current state.

Jenkins is green, merging soon.

Hey, I just responded to a couple of your points, otherwise I think you're good.

Thanks John.

Merged to master.