Spread Model Maps¶
One question that has been raised is how the spread model star selection works in different areas of the sky. Is the stellar locus peaked at spread_model = 0 everywhere? In fact, we know that it isn't peaked at 0 everywhere, due to issues with PSF modeling. In particular, Emmanuel has pointed out that in dense fields of galaxy clusters the psfex star selection can get confused and choose faint galaxies to put into the template, which leads to a biased spread_model selection.
I took all the objects from the SVA1 catalog with griz observations, and applied Bob's SLR (which isn't really important). I pixelized the sky and in each pixel fit a Gaussian to a histogram of objects with |spread_model_i| < 0.01 and mag_psf_i < 21.0. This gives a mean value and sigma for the "stellar locus" in spread_model space. Sigma of course will be biased slightly high due to errors in spread_model_i (which I have not taken into account, but by keeping to brighter objects this shouldn't be too much of a problem.)
The maps are attached below. I have two sets. The "sm" maps show the mean spread_model_i value in each pixel, and the "nsig" maps show mean/sigma, the number of sigma the bias is away from 0. The "all" maps show all of SVA1, and the "spte" maps zoom in on SPTE.
The bulk of SPTE and SPTW look fine to my eye, both in terms of mean values and bias, in that using spread_model_i should give a relatively uniform selection of stars (relative to the spread_model locus) across the survey. But there are some exceptions.
First, the stellar modeling in several of the supernova fields appears to be off: the stellar locus is peaked at spread_model_i ~ -0.001, or more than 1 sigma low. I'm not sure why this should be the case for the SN fields, except perhaps that they were taken with worse seeing.
Second, the stellar modeling in the southern end of SPTE, where we have the LMC, is biased high, such that spread_model_i ~ 0.0004, or ~0.4 sigma high. Yet another reason to be cautious in this region!
SN Fields (W. Wester)¶
I looked at spread model for bright objects in the SN fields selected with flags=0 and 17<mag(psf)<20. For each of the four SN regions (all 10 fields), in each filter, I plot the spread_model distribution for objects from the SVA1 Coadd catalog. I then fit the peak to a Gaussian where I exclude tails beyond 2 sigma of the Gaussian. Results are in the attached file and show the mean and width of the Gaussians depending on the field and filter at a level that is not small compared with a typical 0.0001 spreaderr_model for such objects.
Y1C2 SPTE: With PSF Homogenization (E. Rykoff)¶
Nacho asked if I could take a look at the same thing with Y1C2 which was run with PSF homogenization. The biases are smaller by a factor of ~2, but the same pattern emerges especially in the LMC region. I don't have the Y1C2 SN fields on hand, however (were they run on Y1C2?) which is where the larger issues are.
Y1C1: With PSF Homogenization¶
And while I'm at it, I added in Y1C1. Please take with several grains of salt...we all know that the Y1C1 reduction wasn't ideal. But I do have the SN fields available. And...it looks like SVA1. The stellar locus is biased ~1 sigma to the negative side in the SN fields, though the absolute magnitude of the shift is smaller (due to a tighter locus in the PSF homogenized images).