The simplest way is to use the difference between PSFMag and KronMag. This has the advantage that it is available for all objects in the survey.
This example is from a stack sample around Abell 1656 matched to SDSS and coloured by SDSS classification (red - stars; black - galaxies). A cut in iPSFMag-iKronMag of 0.05 (shown) does a reasonable job of separating stars and galaxies. |
---|
You can calculate this using data from the MeanObject, StackObjectThin or ForcedMeanObject tables. You might improve the separation by combining results from more than one band or using a more sophisticated cut. Note that we do not expect PSF-Kron to be exactly 0.0 for stars, as Kron magnitudes by definition require a correction to take them to total magnitudes.
Faintward of i ~21 this simple cut becomes unreliable (usually over-predicts the number of stars). Also be aware that once stars become saturated (brighter than i ~14 in this example) they shift to the galaxy side of the cut. A full discussion of this method applied to Pan-STARRS data can be found in Farrow et al. (2014).
StackObjectAttributes (or StackObjectView) has a quantity psfLikelihood for each filter. This is around 0 for galaxies but can be either +1 or - 1 for stars. The best way to see what is happening is to plot a quantity like log10(abs(psfLikelihood))
Same sample as above, but now using log10(abs(ipsfLikelihood)) as the separator. |
---|
Either momentsR1 or KronRad can be used as a separator (KronRad is 2.5*momentsR1 so it doesn't matter which you use.
Same sample as above, now using log10(imomentR1) as the separator. |
---|
You can also try momentRH, which is the NOT the half light radius, but the expectation value of r^0.5:
Same sample as above, but noe with log10(imomentRH) as the separator. |
---|