Importance of photometry calibration at CCD level for coadded images

Data Management
#1

This is probably a very basic question but I am wondering how the zero point determined while reducing individual CCD images are propagated to the coadded images ? In other word; how important is it to determine the zero point for individual CCD if one performs a calibration like stellar locus regression on the coadded images ?

#2

For HSC, we have been using meas_mosaic to set a position-dependent magnitude zero-point (kinda like “ubercal”). @parejkoj is in the process of putting the same capability into jointcal.

Getting the zero-points aligned on the individual CCDs (so that sources have the same calibrated flux level) is important, even if you’re planning to calibrate the coadds independently: if they aren’t aligned you’re going to introduce additional noise in the coadd.

#3

Ok, that makes sense. But are the coadd actually recalibrated in the current pipeline ? If yes where is it done ? measureCoaddSources ?

#4

No, we currently don’t have a facility for recalibrating the coadd. I think QA on the coadd measurements indicates that this is not necessary.

1 Like
#5

But you certainly shouldn’t let that stop you! If you put a stellar locus regression Task together, it’ll be a very useful QA tool.

#6

We are working on 2 fronts for the DESC cluster analysis on CFHT data:

  • Improving photometric calibration at CCD level by using SNLS catalogs as in Betoule et al. 2013
  • implementing stellar locus regression at the coadd level

But I think that we should also improve getZeroPoint() in pipe_tasks/…/photocal.py as it does not take into account the errors on magnitudes from the reference catalog (unless this has been changed recently, but I don’t think so)

#7

I’d suggest we skip that, and spend our efforts improving the photometric fit in jointcal (which I plan to start very soon). That would be our go-to source for calibration in the future, no?

1 Like
#8

I agree that it is better to put the efforts on improving the photometric fit in jointcal now, but it’s a little bit embarrassing to keep a bugged CCD photocal in our pipeline. This is also bad for future multifit algorithms which will be running on individual exposures, not coadded ones.

#9

I’m embarrassed by the current behavior of PhotoCalTask as well, but I also don’t think it’s doing much harm. As for multifit, it will definitely use JointCal outputs for both photometry and astrometry, so the initial CCD-level versions of those don’t matter there either. Where they will matter much more is in Alert Production.