Who are the stakeholders for PSF-Matched Coadds

Currently the choice to make PSF-matched coadds is a config in makeCoaddTempExp allowing users to make either psf-matched or non-psf-matched coadd temporary exposures (CoaddTempExps). In DRP we will want both, and storing both will require making psf-matched and non-psf-matched CoaddTempExps different butler datatypes.

We will want both types of coadds for different uses: non-psf-matched coadds for deep detection for example and psf-matched coadds for galaxy colors, optical artifact masking/rejection, etc… Help me survey the “etc”: As I plan to incorporate PSF-matched coadds into the DRP process, I’d like your input if you’re expecting your downstream algorithm to make use of psf-matched CoaddTempExps or coadds.

An RFC will follow after we develop a plan for generating and storing PSF-matched CoaddTempExps as a separate datatype.

In the short term the AP pipelines will certainly use PSF matched coadds. In the long term, we will want something more like a Kaiser coadd, but I don’t know exactly when or where that will get done.

We want to use PSF-matched coadds to perform some deblender studies.

@swinbank should confirm this, but I believe we’re not currently planning to do Kaiser coadds except as a backup option in the case that direct and PSF-matched coadds can’t do something we need. We are still planning to do likelihood coadds for deep detection, however.

Also note that Kaiser coadds would almost certainly have discontinuities where the set of input exposures changes; I had assumed this ruled them out for diffim templates.

@swinbank confirms this.

I had glossed over this detail. Thanks for pointing it out.

I must admit that I had remembered that we were planning on going away from Kaiser coadds. However, I’ve recently started thinking about them again since it allows us to use all the data. We will really be short on data for coadds early in the survey and it may help out with that.

Anyway, I have been assuming that if AP needs this then it will be up to me to plan it.

Yes, that’s likely the case for deblenders that use color information. It may not be necessary (if the deblender does its own convolution), but it would certainly be easier.