SLR corrections for MAG PSF and MAG AUTO¶
Keith's SLR corrections for Gold were done with MAG_PSF star colors. But there are problems with MAG_PSF! So...is this a problem if we're using MAG_AUTO galaxy colors?
I (Eli) have run BigMACS SLR, nside=128, using several catalogs/magnitudes to test the biases.
- SVA1 Gold Stars (modest_class selection, >10sigma in each griz), MAG_PSF colors
- SVA1 Gold Stars, MAG_AUTO colors
- SVA1 raw stars, MAG_AUTO colors
The first run is simply a sanity check vs. Keith's runs, which had a slightly different star selection and higher resolution (and interpolation). The second run is to look at directly the offsets between MAG_AUTO and MAG_PSF (after taking into account gross biases in the original GCM run). The third run can be used to make a new gold-auto SLR correction...
These all look good, but there's a hint of an offset at the south end in r-i; not sure about this. However, the widths of the histograms are all ~5mmag, approximately the error in the estimation of the SLR zeropoint. So this looks okay. Yay!
Here things are a bit more problematic, especially in r-i. There are significant biases in some regions, probably caused by differing systematics in the r and i bands (while g/r and i/z are well-matched). I've scanned through all the systematics maps, and the only one that has the same patterns is the sky brightness (and of course sky sigma which is basically sqrt(sky)). Is there something in the sky brightness that is affecting star (or galaxy?) measurements with MAG_AUTO or MAG_PSF? How to check?
Note in particular the sw corner in r band compared to the bias in the g-r auto offset. Also, the northern strip in r-i that seems to be lower sky brightness in r relative to i.
These don't tell us anything new...they look a lot like the MAG_PSF maps from Keith, of course. The MAG_PSF-MAG_AUTO offsets are smaller than the original offsets. Offsets are now relative to GCM, with fiducial dereddening applied.
SN X3 Field¶
The SN X3 field is particularly important, because it is the source of many of our deepest spectroscopic training sets (VIPERS, GAMA). Also, there are hints that the redMaGiC galaxies have an overabundance of outliers here (though is this just because of all the deep spectra?). In any event, this deep SN field has mostly undithered observations, with some offset pointings that fill in the chip gaps and create significant depth variations in the Gold coadd catalog. This creates particular problems for the PSF modeling, and creates problems with MAG_PSF. The question: do the corrections, based on MAG_PSF, create other problems with MAG_AUTO (which is currently used for colors in many studies)?
There are definitely regions, especially in g-r, that are very "red": significantly offset between the AUTO and PSF fits. (Don't worry about the overall zero-point...this is not 0 because of a dumb extra reddening offset that I forgot to remove).
Here is a plot of the spectra in X3 (grid is vipers; dense squares are GAMA). Red points are my comparison stars below:
And here is a comparison of the stellar locus with MAG_PSF and MAG_AUTO from Gold:
The locus is way noisier with MAG_PSF due to the depth variations (and is actually worse in some outlier regions of X3, especially the NE, but these are less populated by spectra). And I believe this noise could be negatively impacting the SLR corrections (certainly adding noise to them!)
However, I had previously been worried that the PSF SLR fits were imprinting these offset biases on the stars, but I no longer believe that to be the case; the spatial variations are on a smaller scale than the SLR pixelization, so the SLR fits do seem to be grabbing the "main locus", but this is by no means guaranteed at every point in the survey.
Background: The input to the coadds as well as ngmix is the relative flux scaling from the SVA1 run of the global calibration module (GCM). This had known problems (basically fixed for Y1A1!) This necessitated the SLR recalibration of the zeropoints for Gold in the first place. All offsets that I'm looking at below are much smaller than the large systematic color offsets corrected by applying the SLR recalibration.
This test: The GCM outputs zeropoints based on MAG_PSF for every CCD. The local calibrations are quite good, with reproducibility of ~3mmag generally. I match all the stars in SVA1 finalcut (using the GCM zeropoints) to get average MAG_PSF magnitudes for every star. I then run the same SLR code as above.
Bottom line: the coadd MAG_PSF appears to be working well overall, expect for a number of complete outlier pixels. Based on visual inspection, we also know that there are problems at finer scales that this relatively coarse test isn't picking up (except for the crazy pixels). I'm not sure why the MAG_PSF colors aren't centered at zero, and this could be a worry.
The coadd MAG_AUTO has a lot more scatter and structure, though there aren't any crazy outliers. So from this test it appears that in the above tests it is MAG_PSF that is more reliable, in spite of it being on a coadd (modulo the small structure problems!) But it also shows there may be dangers in trusting the MAG_PSF SLR recalibration for AUTO colors...