Fixed Astrometry in SVA1 Finalcut

Following up on Fixing Astrometry in SVA1 Finalcut, I (Eli) have run the quick-and-easy astrometry on all of SVA1 Finalcut with dec > -62.

The method (recap)

I used the CCD-by-CCD solution method in Fixing Astrometry in SVA1 Finalcut. The i-band coadds were used to generate a reference star catalog, and then solve the astrometry solution for each individual ccd. It's fast, and seems to work well. There are limitations in that it does not use all of the information (in terms of instrument stability); it only works where we have i-band coadds (which covers all of SVA1 gold of course); and doesn't take into account tree rings and other high order distortions.

The astrometric reference catalog I used is available here, in scamp-compatible FITS_LDAC format. (The data was taken from the SVA1 database, with a simple MODEST_CLASS type selection):


The scamp config file is attached as sva1_rerun_scamp.conf. Please be aware that you'll have to put in the full path name to the reference catalog.


Attached are a bunch of plots; here are some as a sample.

MAG_PSF offset: Testing GCM

This shows the mean mag_psf offset of all stars in the CCD relative to the mean magnitude for each star. Large outliers here show that there was some issue with the GCM solution, in that the stars are significantly offset. Red dashed line is a Gaussian fit to the distribution. I recommend using images that are significant (>4sigma) outliers with caution.

RA offset

After solving the astrometry, this just confirms that the positions of the stars in the ccd are indeed consistent with other measurements of the stars. This is not identically zero because the test was performed with a different star selection than the solution. In some cases there are gross outliers (not shown) where the astrometry solution failed. These are mostly bad images that should/have been rejected already (bad sky, ghosts, etc.)


This is the RMS of the offsets. For most images, we are around ~12 mas rms in RA (and dec as well) in riz. g is a bit worse. The large outliers are images with some problems. Note that the total rms is ~17 mas. According to Gary, the best we can possible hope to achieve before taking into account tree rings and other higher order distortions is ~15 mas, so I'm pretty pleased with the general performance.


Overall, there were 200713 ccds that I ran through...

  • 180 had no scamp solution (at boundary, no i band coverage)
  • 80 scamp crashed (similar to above; unsure on why crash)
  • 6226 had too few (<50) good star matches (at boundary)
  • 3740 have a bad MAG_PSF offset (>4sigma)
  • 2784 have a bad RA offset (>4sigma, bad solution)
  • 4639 have a bad RA rms
  • 3653 have a bad Dec offset
  • 3892 have a bad Dec RMS

In all 174567 (87%) pass all my recommended flag cuts for inclusion into (a) multiepoch shape fitting and/or (b) new coadds (if it comes to that...)

The new headers have been passed to Erin Sheldon who is rebuilding SVA1 meds files.