Details
-
Type:
Story
-
Status: Done
-
Resolution: Done
-
Fix Version/s: None
-
Component/s: None
-
Labels:None
-
Story Points:16
-
Epic Link:
-
Sprint:DRP S20-3 (Feb)
-
Team:Data Release Production
-
Urgent?:No
Description
When SFM is run, we produce and persist a "calibrated" exposure (calexp) upon which we perform measurements to compute the src catalogs. Subsequently, an "external" calibration can be run (i.e. jointcal/fgcmcal) to compute a more refined set of photometric and astrometric calibrations. Finally, a global sky correction can be performed and applied. The latter two steps are not reflected in the persisted visit-level src catalogs, yet these are the catalogs on which many "visit-level" metrics are computed and tracked. Indeed, the external photo/astrom calibrations can be applied at the catalog level, so their effects can be measured and tracked. However, there is currently no check on the photometry post global sky-correction and just pre-warping into what actually gets included in a coadd. It would be good to know if/what differences there are at these different stages, particularly in assessing the chaining of the measurement and application of these calibrations. In particular, the global sky correction is measured independently of any external photometric calibration and applied subsequently. Does this have any (negative) effects on the effective calibration of the image going onto the warp & coaddition stage?
Current plan is to make use of the new functionality added on DM-23352 which allows for forced CCD photometry to apply any combination of the above mentioned calibrations (as config options). Forced catalogs will be created for the different permutations and combinations and the the resulting source catalogs will be compared.
Note to self: If any "issues" are uncovered, it may be worth a subsequent ticket to look into adding the ability to perform the global sky subtraction (pipe_drivers' skyCorrection.py) on images that have had the external photometric calibration applied.
Attachments
Activity
While I doubt I will be merging any code in the short term, I do hope to have a few more illustrative plots made within the next few days. Regardless, I will provide a summary here by the end of the week.
Would you mind having a look at this (the attached pdf)? This is the outcome of my analysis-sprint efforts and you are likely the person to care the most. This is not a typical "review" as no code is getting merged as part of this (the pipe_analysis scripts will get added on DM-23550 and I'm not sure my adaptations to meas_base will make it into the stack...discussion is ongoing, but you can have a look at them on the ticket branch listed here.)
In particular, I'm interested in how you feel about the results, particularly if the differences are small enough to ignore or if you think we should really be running things in more consistent manner (my feeling is that, if a given run is going to measure and apply the skyCorrection, then running fgcmcal on calexp's to which the skyCorrection has been applied is a step in the right direction?) Also let me know if you would like to see anything in addition to what I've shown (it might not get prioritized, but we can create new tickets and see if we can get them into a future sprint).
This is very interesting, thanks for getting this going. Here is a non-prioritized list of comments/suggestions, some easier than others.
- There are a couple of possibly intertwined effects that could happen by running the calibrations on non-properly-sky-subtracted calexps (as is currently done). The first (primary) one is that the faint stars are going to be biased more than the bright stars, because this is an additive effect. You can see this in the curvature of the mga comparisons on slide 4. The second (secondary) is that the bias in the faint end can then bias the calibration at the bright end. I know I saw this at the few mmag level for fgcmcal; from these tests it may be that jointcal is not as affected, due to the small scatter in the psf/psf comparison. At the same time, for the 12pix aperture used for calibration there is a scatter of ~1% at S/N > 100 due to the sky issues. It's almost problematic then.
- From slide 2, it looks like visit 1228 (used for the sample plots) is much better behaved than visit 29350. So I think we need to look at a lot more sample visits (in particular 29350) before we can say the differences are small enough to ignore. I have no doubt that for many visits this is not a problem, probably correlated with sky brightness.
- Another thing to look at which would be interesting (but definitely more work!) is the effect of running jointcal/fgcmcal on the sky corrected images (which should be our goal). That is, one could expect that slide 3 would look different if the photocalib is computed after the sky correction, as well as applied in addition to the sky correction. As above, I would expect this to vary depending on the visit.
Thanks for the comments Eli. I agree with all of them (and sorry this only used jointcal...the fgcmcal results from the PDR2 run did not exist yet, so I had to go with the latest RC2 run). As per your second point, I did look at a few more visits, but didn't include plots in the slide deck so as to keep from overwhelming. The numbers bounce around a little, but none were striking enough to include there...but, since you asked, I've attached plots for an example HSC-G and HSC-Z visit here.
I particularly agree with your 3rd point...if others agree, I'm happy to do it/be involved.
Is there something that can be merged this week? Like a PDF of the plots in your presentation or should I open and close a new ticket with your 2 weeks worth of analysis-sprint story points?