Shallow Coadds for SG Separation¶
Contact: Alex Drlica-Wagner <email@example.com>
Eli assembled a set of SVA1_SN_Field_Shallow_Coadds to enable testing at the generic SVA1 depth. We would like to examine the use of these coadds (specifically in the COSMOS and SNC fields) as training sets for the SG challenge.
Eli warned that the COADD tiles may not cover the full COSMOS or SNC fields. This shouldn't be much of a problem because we are really only interested in small regions of these fields that have been observed by other instruments. We see that we have nearly all of the COSMOS field (the missing coadd contributes only ~1600 objects to SVA1). Additionally, CDFS is completely covered, but the CANDELS UDS is not.
We now follow the procedure developed here for comparing variable distributions. We select all objects within the COSMOS field (149.5<RA<150.7 && 1.7<DEC<2.7) from Eli's combined_masked_t1000_cat.fit (~93,000 objects). We compare the distribution of these objects to a random sample of the SVA1 Gold full catalog. The discrepancy in depth no longer exists (by design), and in general the distributions appear to agree well. The COSMOS spread_model + 3*error distribution has a strange tale in the 'i' and 'z' bands around mag_auto = 24 that doesn't exist in SVA1. The better than average conditions under which COSMOS was observed do not seem to have a large impact on spread_model; however, the discrepancy is clear in mu_max (which should still not be used). The class_star distribution is quite complicated, but shows reasonable agreement between the two samples.
spread_model + 3*error¶
mag_psf - mag_model¶
mu_max - mag_auto¶