Photometric Redshifts and Signal-to-Noise

Abstract

We investigate the impact of photometric signal-to-noise (S/N) on the precision of photometric redshifts in multi-band imaging surveys, using both simulations and real data. We simulate the optical 4-band (BVRz) Deep Lens Survey (DLS, Wittman etal 2002), and use the publicly available Bayesian Photometric Redshift code BPZ by Benitez (2000). The simulations include a realistic range of magnitudes and colors and vary from infinite S/N to S/N=5. The real data are from DLS photometry and two spectroscopic surveys, and explore a range of S/N by adding noise to initially very high S/N photometry. Precision degrades steadily as S/N drops, both because of direct S/N effects and because lower S/N is linked to fainter galaxies with a weaker magnitude prior. If a simple S/N cut were used, S/N>17 in R (corresponding, in the DLS, to lower S/N in other bands) would be required to keep the scatter in Deltaz = (zspec-zphot)/(1+zspec) to less than 0.1. However, cutting on ODDS (a measure of the peakiness of the probability density function provided by BPZ) greater than 0.4 provides roughly double the number of usable galaxies with the same scatter. Ellipticals form the tightest zspec-zphot relation, and cutting on type=elliptical provides better precision than the ODDS>0.9 cut, but this eliminates the vast majority of galaxies in a deep survey. In addition to being more efficient than a type cut, ODDS also has the advantages of working with all types of galaxies (although ellipticals are overrepresented) and of being a continuous parameter for which the severity of the cut can be adjusted as desired.