Star / Galaxy / CR Separation in the IPP

-- Eddie Schlafly & Eugene Magnier

Star / Galaxy Separation Summary

Star / Galaxy separation by the IPP has multiple driving goals. One goal is to allow scientists to select objects from the database (DVO or PSPS) which are "likely" to be either stars or galaxies, depending on the science driver. Another is to act as an initial seed to the Morphology Server and Photometric Classification Server with guesses for more detailed analysis of possible stars or galaxies. Another is to guide the IPP as it choses which objects should have which measurements performed.

The IPP performs several different extended source (ie, galaxy) analyses. Not all objects necessarily have the full set of measurements applied. The different types of extended source measurements are listed below, in order of the CPU requirements:

  • PSF and Kron magnitudes : all sources in all images (single exposures or stacks) have PSF and Kron magnitudes measured.
  • moments (2nd, 3rd, 4th) : all sources in all images (single exposures or stacks) have moments measured.
  • Circular radial apertures : all sources in the yearly stacks have the PSF-matched circular radial aperture photometry measured. Currently, we expect this to be performed for 2 target PSF sizes (1.2 & 1.6 arcsec?).
  • Petrosian and Elliptical apertures : these are measured only for the yearly stacks, and for some subset to be defined, limited potentially by S/N and by galactic latitude constraints.
  • non-linear fits (Exponential, deVaucouleur, Sersic) : these are measured only for the yearly stacks, and for some subset to be defined, limited potentially by S/N and by galactic latitude constraints (need not match the Petrosian & Elliptical apertures).

We have used GPC1 observations of Stripe 82 to define the PS1 Star / Galaxy / CR separation. The advantage of this region is that SDSS data exists for both single pass and deep stack analysis.

We looked at several parameters measured by the IPP psphot tool which may be sensitive to the difference between stars and galaxies. We used the SDSS star/galaxy separation from both the single pass and the deep stack images as 'truth'. Our main conclusion is that a simple comparison between the PSF and Kron magnitudes for objects does a good job of reproducing the SDSS star/galaxy separation for signal-to-noise > 10.0 For signal-to-noise between ~6.5 and 10.0, the two populations start to blurred together significantly. For S/N < 6.5, we are unable to distinguish stars and galaxies.

We found that the moments did not improve the star/galaxy separation. Instead, because of the variable PSF, including moment information tended to increase the noise in the measurement.

Star / Galaxy Separation Examples

This first figure shows an example of the GPC1 / SDSS comparison. The green dots are the SDSS-identified stars; the blue dots are the SDSS-identified galaxies; the red dots are not identified in SDSS. The plot on the left shows the Kron magnitude - PSF magnitude vs PSF magnitude (all PS1 measurements). The plot in the middle shows the SDSS model magnitude - SDSS PSF magnitude from the deep stack (this is essentially the SDSS star/galaxy separation analysis). The plot on the right shows the (Kron magnitude - PSF magnitude - median offset) / sigma (PSF error, Kron magnitude error, 0.01 mags all added in quadrature). We generated plots like the above for 15 exposure, and they all exhibit more-or-less the same behavior.

The right hand plot is the PSF-significance: it represents the significance (in terms of a Gaussian sigma) of the PSF nature of the object. The larger the value in this plot, the less likely the object is taken from a population of PSF-like objects. We consider this to be a measurement of the a-priori star probability.

The second figure shows a series of histograms of the right hand plot above, summed over all exposures. Each histogram is a different PSF magnitude range. This sequence shows the reliability of the PSF-significance as a measurement of the a-posteriori galaxy probability as we proceed to fainter magnitudes. In these histograms, the black line is all objects; the green line is the SDSS-identified stars; the blue line is the SDSS-identified galaxies; the red line is the objects not identified in SDSS.

The third figure shows an equivalent series of histograms, but this time in terms of the PSF signal-to-noise.

Cosmic Ray detection

We also are able to examine sources not identified by SDSS, many of which are cosmic rays. We find that the second moment minor axis (M_minor_) is the most useful indicator of the cosmic rays. As an example, the figure below shows M_minor_ for one of the Stripe 82 exposures above.