Notes from jointcal photometry discussion

Notes a discussion about jointcal photometry

These are notes I took during a conversation between @KSK, @RHL, @jbosch, and @parejkoj about how to move forward on how to output jointcal’s fitted results, what we need to do to get it to match meas_mosaic’s output, and what we’d like to do further in the future.

Any errors in the below are mine: please correct me if I misheard something.

Saving catalogs

Jim: there’s a “little known” way to save the calibrated catalogs, but it may have bit-rotted.

Less destructive:

  • save something like meas_mosaic’s output as a math.boundedField

More destructive:

  • redo calib to force people to specify their corrections.
  • add a boundedField thing to calib.
  • have a calib that is just a magnitude zeropoint.

Spatial vs. SED dimensions

What to do with the spatial-varying component?

  • save the jacobian, or not?
  • have to include the wcs with the calib?
  • spatial structure in the filters: may be color-dependent.
  • Are all of these things multiplicative factors that are commutative?
  • If they don’t all commute, then we have to think harder about the order and may have to make the interface less general.

What about the SED dimension?

  • Do we just have a 3d polynomial (spatial+SED)?
  • Polynomial is probably not the most appropriate choice.

Reference catalogs

  • All our refcats are afw.tables, so how do we represent the colors.
  • Our refcats have an entry for each object in each flux.
  • We need to generate obs_*-appropriate fluxes and colors (which each obs package should be able to do).
  • We should add placeholders for colors to jointcal, so we’re ready in the future.

Next steps

  • Try reducing the catalog to just one star, to see the spatial structure: even just fitting a single zeropoint should be an improvement over single-frame procesing.
  • Save the “single zero-point per chip” that jointcal can currently fit as a new Calib (probably as part of the Exposure that is persisted in the “wcs” dataRef).
  • Figure out how the Jacobian behaves before figuring out how to put it into Calib.
  • Currently assuming uniform illumination of the flat-field-screen, if that’s not true the Jacobian can matter a lot.
  • match meas_mosaic: spatially-varying polynomial over each chip, and the Jacobian. Put all that information into the calib.
  • There may be discrete jumps at the edges of the chips because of coatings on the chips: cannot assume a smooth function across the whole field.
  • Eventually fit all the parts separately like we plan to do with the WCS.
  • We should do some experiments to decide how we want to represent the spectral and spatial dimensions.
  • How do we cover the “known spatial structure in the filters”?
  • We need star SED info to go into jointcal.
  • Can we extract the known spatial structure in the HSC filters from the photometry?
  • Can we look at time-varying water vapor as a factor of position?
  • Can we say that there is one function per CCD?
  • Look at the change in y-band between exposures (and between CCDs) as a function of i-z.
  • Represent transmission as a function of up to three separable things: spatial, time, wavelength, with each function only dependent on one of those.

I think the color (wavelength) dependent terms will have to be spatially varying as well. Depends exactly what you’re talking about doing and using them for here, but yes… the color of an object creates a measurable difference in magnitude (so must be calibrated) and this difference varies across the fov by a measureable amount. Estimates of the amount available in document about lvl 1 photo metric calibration.

(My apologies if I’ve completely misconstrued this discussion, you’ve probably already accounted for this).

And I suggest that they be made to vary over the detector rather than over the full field-of-view. I’ve seen some CCDs with very strong variation in color terms. And especially if we get a Frankencamera for LSST we will want the color terms to vary discontinuously over the focal plane.