Magnitudes in the Image Coadd Using New PSF Code

April 12, 2005

RHL developed new code to calculate the PSF based on the first run of the code on the coadd, and then using the fpObjc information to calculate a new PSF.

The residuals are much, much better. One can compare the new residuals of the image level coadd minus the catalog level coadd residuals to those from the single run minus the catalog level coadd residuals for

The PSF residials have an rms of 0.004, much better then the old version of photo, and better even then the single run, which had 0.003 mags, and the flatness of bright-dim is better.

We should run much bigger data sets.

Magnitudes in the Image Coadd

March 04, 2005

  Subject: mags from image coadd and catalog coadd
  From: James Annis
  Submitted: Fri, 04 Mar 2005 09:03:07 -0600 (CST)


We have run photo and target on one column of the coadd and compared these against catalog level coadds. The residuals show structure at the 2% level for psf magnitudes, lower than that for model magnitudes, and substantially lower for both psf and model colors.

The behavior of the psf magnitudes is interesting/worrisome, indicating investigations to be done before one wants to perform precision stellar/quasar work.

The model magnitudes and colors are stable and lack interesting behavior, so galaxy work should be able to proceed.

The state of affairs

We have two Southern coadds, a depth optmized coadd and a weak lensing optimized coadd.

We have run 1 column of the depth optimized coadd through photo, of which ~90% of the fields successfully ran. This version of photo bypasses ssc/psp and makes coadded psField files (coadded PSFs) from single run psField files.

We have calibrated these using standard target technology. At the end, we now have 680 or so tsObj files for camcol 3 of run 100001, the north strip of the depth optimized coadd.

Check of the data quality

We matched up the tsObj with the catalog level coadds of Scranton and Johnston, sdss-gneral#3030, demanding matches in both catalogs within 1" and constructed the residuals.

The answers are:

We didn't work cmodel magnitudes as the catalog level coadd didn't have psf_frac.

The plots

The plots take an explanation. s16-20-r-psf.gif shows for the first 300 fields psf<2>_im - psf<2>_cat. The green dots are the medians, the extrema of the bars contain 95% of the residuals, and the white space inside the bars contains 63% of the residuals. These are for magnitudes 16-20. The red line shows the medians of the 16-19 range minus the medians of the 19-22 range, looking for magnitude dependent systematics.

PSF mags

Ok, s16-20-r-psf.gif shows
  1. an offset of -0.057 mags, probably calibration
  2. rms residuals of 0.019, pretty good... 2% photometry
  3. bright-dim of 0.01, consistet with 0, good
  4. how squirelly psf magnitudes are
  5. bigger spreads to neg, i.e. image coadds tend to produce brighter objects
  6. a worrisome trend towards the neg after field 200, which shows up in the bright-dim as a pos trend both of which are consistent with image coadd objects having more light, increasingly so at dim magnitudes
  7. even more worrisome, near field 100, the 1-sigma scatter to pos is much greater than the 1-sigma scatter to neg
The plots of the other filters, say s16-20-u-psf.gif or s16-20-z-psf.gif show that the trend in 6) goes away in the blue and gets worse in the red. This suggests a sky subtraction issue.

Petrosian mags

s16-20-r-pet.gif shows
  1. an offset of -0.025 mags, why is this different from psf?
  2. bright-dim of 0.08
  3. how noisy petrosian magnitudes are
  4. the -very- large scatter towards the neg, so image coadd objects tend to be brighter
The other filters are similar.

Model mags

s16-20-r-mod.gif shows
  1. an offset of -0.022 mags, consistent with Petrosian
  2. bright-dim of 0.02+- 0.01, very nice
  3. how stable model magnitudes are
  4. also large scatter to the neg, im coadds tend to the brighter
  5. very little structure with commenting on.
The last comment holds for regions where the psf mags are doing something funny. The other filters are slightly less well behaved, with z showing after field 250 an offset of the medians from the center of the white space. Still, model magnitudes look very good in this comparison.


These plots are built in the same way as the magnitude plots, except we are looking at colors and over a slightly smaller magnitude range, 16-19.

PSF Colors

s16-19-gr-psf.gif shows
  1. an offset of 0.004 mags+-0.020, consistent with 0
  2. bright-dim of 0
  3. well behaved quartile distributions
This is nice behavior, except for 0.05 mag level of wandering of the residuals about 0. The other colors are similarly well behaved.

Model Colors

  1. an offset of -0.002 mags+-0.006, very nice
  2. bright-dim of 0.014+- 0.014, very nice
  3. a noisy quartile distribution
This is very nice behavior, and the other colors are similar. u-g is noisy, and i-z is perhaps the worst behaved.


To the extent that we understand how the catalog level coadds were made, these plots show that the modified verion of photo we used to run on the coadds works reasonably well. Galaxy work that uses cmodel magnitudes and model colors should be in good shape. There is little indication of systematics with magnitudes or colors. I do not understand the behavior of the psf magnitudes: they are showing with precision -something- and workers who want to use them in detail will want to make plots of residuals versus sky brightness, psf size, and number of image in the coadd.

For the basic characterization of the galaxy content, the data look to be in good shape. We could always use investigations of the star galaxy separation and the deblending.


There are a large number of these quartile residuals plots at this directory

There are averages of the residuals over 100 fields as a function of column in this directory. In these plots, the plot for the g band model-mags for cols 400-500 is typical, g psf-mags for cols 30-100 is perhaps the worst. The means for fields 30-100 are always bad.

There are a pair of quartile residuals plots of run 3384 versus the catalog level coadd in this directory. Both the psf magnitudes and the model magnitudes show much reduced scatter in the median residuals.

Jim Annis

DSS *** The information posted to SDSS mailing lists is proprietary ***
DSS  This is message 314 in the sdss-southern archive, URL
DSS  The index is at
DSS  To join/leave the list, go to
DSS  To post a message, mail it to