Details
-
Type:
Story
-
Status: Done
-
Resolution: Done
-
Fix Version/s: None
-
Component/s: None
-
Labels:None
-
Story Points:1
-
Epic Link:
-
Team:Data Release Production
-
Urgent?:No
Description
Erik Dennihy noticed that the local_PhotoCalib columns don't match the calexp.photoCalib. Colin Slater noticed that photoCal runs after measurement.
Symptom example:
butler = dafButler.Butler('/repo/main', collections=['HSC/runs/RC2/w_2022_04/DM-33402'])
|
|
|
calexp = butler.get('calexp', visit=30504, detector=54)
|
src = butler.get('src', visit=30504, detector=54)
|
print(calexp.getPhotoCalib().getLocalCalibration(calexp.getBBox().getCenter()))
|
print(np.median(src['base_LocalPhotoCalib']))
|
# >>>0.18701489832531293
|
# >>>0.04394189128521331
|
Easy solution (without thinking about it too hard) would be to add a second round of measurement after photocal.
For DP0.2. we can run the RecalibrateSourceTable task I'm working on right now under DM-33959, with calexp.photoCalib as input, as an afterburner.
Attachments
Issue Links
Activity
IsrTask.roughZeroPoint() sets a photoCalib in the postISR exposure using config.fluxMag0T1. Do we really need a "rough" (and in fact likely entirely wrong!) calibration going into characterize/measurement? If that didn't exist, the local photoCalib plugin would have failed where it was being run, because exposure.photoCalib would be None.
fluxMag0T1 is only set in obs_subaru, so that postISR calibration is nonsense for every other instrument. I'd like to get rid of it entirely if we can.
I'll bet we have the same problem with EvaluateLocalWcsPlugin, and any of those other summary plugins: if they run before CalibrateTask, they'll have junk in them.
Wait, what? The local calibration is not even using the single-frame photocal fit? OMG.
Note that a second round of (full) measurement after calibration would be very slow and overkill. Just using the photometric calibration and wcs and doing the computation in calibrate would take no time and be 4 lines of code.
Thanks Jim Bosch,
4 PRs: The actual (minimal) change in pipe_tasks, 2 regressions tests, and a bug in ci_builder revealed by adding a second test to ci_imsim.
- https://github.com/lsst/ci_hsc_gen3/pull/68
- https://github.com/lsst/ci_imsim/pull/17
- https://github.com/lsst/pipe_tasks/pull/653
- https://github.com/lsst-dm/ci_builder/pull/9
extra Jenkins run for everything here: https://ci.lsst.codes/job/stack-os-matrix/36179/display/redirect
I'll note that this is more serious than just the local photoCalib columns being wrong: all calibrated nJy fluxes in our sourceTable and consolidated visit table files are wrong!