Bright Galaxy Impurity


As shown in A Modest Proposal for Preliminary Star/Galaxy Separation, there is relatively high impurity in the bright end of the galaxy selection in COSMOS. What's up with that?

Note that this can be seen in context of the following plot:

Black/red: SVA1 selection (using dumb spread_model cut); green: SDSS DR8; blue: Aardvark

The black and red are SVA1 (not sure of difference...sorry). The green is SDSS DR8, and blue is Aardvark. From 19<i<21 then SVA1/DR8/Aardvark all agree. Fainter than this SVA1 and Aardvark match. Good! But at the bright end there are two issues. First, the hump at the very bright end. These are misclassified stars, which are taking care of by a CLASS_STAR cut here: A Modest Proposal for Preliminary Star/Galaxy Separation. But there's also a "knee" where we have an excess of bright "galaxies" at i<18.

Are these objects the same as the impure galaxy selection from COSMOS?

The spurious bright objects in COSMOS

What are these things? Visual inspection shows most of them to be artifacts of bright stars:

There are 2 i~18 "galaxies" on either side of a masked saturated star. These have flags==3 set (deblend) but in general these objects do not have other suspicious flags or flags_weight set. Because mangle has not been run on the cosmos fields, I can't tell what mangle is telling us here.

But there's something else: I cannot find any examples of this type in SPTE (though I don't have the truth to look for and I didn't spend too long looking). But there aren't any pairs of objects like these.

So: what's going on in COSMOS? And we still don't know the source of the extraneous objects in SPTE!

The two obvious differences between COSMOS and SPTE fields are the length of the individual exposures, and the total depth. Huan will be producing full-depth equivalent COSMOS coadds shortly which will help address this. But my suspicion is that the longer 300s COSMOS field exposures create longer bleed trails and thus make this problem worse. (However, I should add that in SPTE brighter stars that have equivalently long bleed trails don't have obvious spurious detections attached to them!)

Thoughts/suggestions welcome...


Thanks to Ryan Keisler for pointing out that a lot of these bad bright objects actually do not have MAG_PSF measured properly (MAG_PSF_I 99). Huh!

Turning this around, of the 1043 objects in COSMOS with MAG_PSF_I 99, 1043 (77%) are junk. Of the other ones, 17% are very faint and only marginally detected (<5sigma). So it looks like this is a reasonable cut to make.