color outliers can identify SGA galaxies contaminated by stars

[moustakas]

Reported by @arjundey in legacysurvey/legacypipe#626 but moving to the SGA repo:

I have been examining the color outliers with the following criteria (note: in all cases the color is constructed from the 23.5 mag/sq.arcsec. SB aperture and I only looked at SGA galaxies with SGA_ID NE -1 and IN_DESI EQ 'T' and IN_DESI_GRZ EQ 'T'):

[A] (W1-W2) < -2 OR (W1-W2) > 1

• mostly OK; some objects made up of multiple sources; some too faint to be visible with WISE (e.g., Virgo dwarfs). May be a good way of finding large LSBs.

The following have -2 < (W1-W2) < 1

[B] (g-r) < -2 OR (g-r) > 5 OR (r-z) < -5 OR (r-z) > 5

• great way of finding objects where a galaxy and one or more bright stars lie within the SGA ellipse. The ones that are exceptions are where data are missing in one of the bands or where the galaxy lies along a bleed trail

I think it would be useful to identify all the SGA sources where a Gaia DR2 star lies within the SGA elliptical mask and set those aside for re-fitting in a better way (or at least tag them appropriately). Do we already do this, in MASKBITS perhaps? From what i can see, we only check stars with G < 16, but perhaps we need to include the fainter ones for the purpose of determining reliability of the SGA photometry? (most of the galaxies satisfying [B] do not have MASKBIT 11 set).