Flavors of Coadds as discussed at Perfect Pixels, Accurate - - PowerPoint PPT Presentation

Flavors of Coadds

as discussed at Perfect Pixels, Accurate Astrophysics, Correct Cosmology as remembered/reconstructed by Jim Bosch

Some Definitions

Direct Coadd:

• just combine per-epoch images with no modification to their PSFs
• combine per-epoch PSFs the same way to compute effective coadd PSF (requires

"combine" to be strictly linear - no clipping or medians). PSF-Matched Coadd:

• convolve per-epoch images to a target PSF
• combination can use outlier rejection
• have to drop input images with PSFs [significantly] larger than target PSF.

Some Definitions

Kaiser Coadd:

• formally optimal coadd: a sufficient statistic for the set of single-epoch images,

assuming they only contain static sources.

• an old idea that still isn't widely known, and has been rediscovered a few times
• For more information see:

○

Zackay & Ofek (2017) ○ Many more slides from me. ○ A sparse-matrix version (and generalization) from David Kirkby.

Comparing Flavors of Coadds

The Bottom Line, for LSST

Direct coadds would probably be fine (Sheldon & Armstrong, in prep):

• difference in S/N from multi-epoch fitting is very small for realistic seeing distributions
• noise correlations can be propagated via Monte Carlo (possibly even just one sample)

Kaiser coadds are even better - transforms slight concerns into non-concerns. The big open question is edges; do we:

• Also make per-object coadds (from scratch) for measurement?
• Make multiple coadds in each predefined cells and stitch them together?

How do small-cell coadds interact with blends?