SV-A1 Color Uniformity From Clusters

Using the average color of red galaxies in clusters we can track relative changes in the color calibration across the survey. The basic idea is that given, for example, the average g-r color of a cluster red sequence we should be able to accurately predict the r-i color. Or to look at it another way, if we know the g-i color the r band should be easily predictable. Importantly, this test relies purely on photometric data

If we pixelize the sky we can compute the relative color offsets and look at the histogram of residuals as well as structure across the sky. For comparison, in SDSS DR8, using the redMaPPer catalog we can see that the scatter in <g-r> - <r-i> is ~0.005 mag across the full 10000 deg^2 survey.

Uniformity in Raw SVA1 Catalog

In the raw SVA1 catalog, I measure the red sequence using MAG_AUTO\[2\] as the total i-band magnitude (for luminosity) and MAG_DETMODEL\[0:3\] for computing griz colors.

As can be seen, there is quite a lot of scatter in both sets of colors. In addition, there are large offsets between discontiguous regions (not surprising) as well as significant gradients within SPT-E (more worrying).

However, even with these issues this is a significant improvement relative to Y1C1, which due to bugs in the pipeline had a large number of truly discrepant regions.

It is also apparent at first glance that the residual patterns are similar to those in Bob Armstrong's SLR residuals. This is a good thing! Again, in Y1C1/2 due to the image registration bugs the stars (via MAG_PSF) and galaxies (via MAG_DETMODEL) were doing completely different things on a tile-by-tile basis.

Uniformity in the SLR SVA1 Catalog

I adjusted all galaxy colors using Bob Armstrong's SLR calibration in the rearmstr.slr_cal table. Of the 208 deg^2 covered in the above cluster run, 201 deg^2 have SLR values in the table. I then recalibrated the red sequence and ran redMaPPer using the new "SLR SVA1 Catalog". Note that I applied each SLR correction on a tile-by-tile basis. Some improvements may be able to be achieved by doing bilinear interpolation of the SLR table, or (perhaps) by running on smaller areas than coadd tiles.

The uniformity now looks like this:

This is much improved! Note the improvement in the scale: the scatter in <g-r>-<r-i> has been reduced from 3.5% to 2.0%, and the scatter in <r-i>-<i-z> has been reduced from 2.2% to 1.0%. This is still not at the level that we need for cosmology, but it is a significant improvement.

Uniformity in the Huan Lin Galaxy Locus SVA1 Catalog

Next, I adjusted all galaxy magnitudes using Huan Lin's "Galaxy Locus" calibration from here . I applied the DETMODEL magnitude corrections (bzpt_detmodel) to the reddened (uncorrected) galaxy magnitudes to get reddening and galaxy locus corrected magnitudes. I recalibrated the red sequence and ran redMaPPer using the new "Huan SVA1 Catalog". Note that comparisons of Bob's SLR map to Huan's Galaxy Locus map show significant discrepancies (no link) at a level that should be detectable by this red sequence test.

The uniformity now looks like this:

This shows slight improvement over the raw calibration, but is significantly worse than the SLR performance. Furthermore, with the SLR there are no tiles that are surprising (>3 sigma) outliers, while there are several regions that are significant outliers with the Galaxy Locus.


  • The raw calibration is much improved relative to Y1C1/2, showing no obvious bugs shifting things at the 10% level.
  • The stars and galaxies are shifting in the same way. Hooray for reasonable behavior!
  • The SLR improved things, but we still require better photometric uniformity.
  • The GCM is not achieving the relative uniformity within SPTE that is required (or even the level of SLR; In the end I expect the GCM method to outperform SLR for relative colors).