Quantifying Photometric Redshift Errors in the Absence of Spectroscopic Redshifts

Abstract

Much of the science that is made possible by multiwavelength redshift surveys requires the use of photometric redshifts. But as these surveys become more ambitious, and as we seek to perform increasingly accurate measurements, it becomes crucial to take proper account of the photometric redshift uncertainties. Ideally the uncertainties can be directly measured using a comparison to spectroscopic redshifts, but this may yield misleading results since spectroscopic samples are frequently small and not representative of the parent photometric samples. We present a simple and powerful empirical method to constrain photometric redshift uncertainties in the absence of spectroscopic redshifts. Close pairs of galaxies on the sky have a significant probability of being physically associated, and therefore of lying at nearly the same redshift. The difference in photometric redshifts in close pairs is therefore a measure of the redshift uncertainty. Some observed close pairs will arise from chance projections along the line of sight, but it is straightforward to perform a statistical correction for this effect. We demonstrate the technique using both simulated data and actual observations, and discuss how its usefulness can be limited by the presence of systematic photometric redshift errors. Finally, we use this technique to show how photometric redshift accuracy can depend on galaxy type.