CFHTLenS and RCSLenS: testing photometric redshift distributions using angular cross-correlations with spectroscopic galaxy surveys

Abstract

We determine the accuracy of galaxy redshift distributions as estimated from photometric redshift probability distributions $p(z)$. Our method utilises measurements of the angular cross-correlation between photometric galaxies and an overlapping sample of galaxies with spectroscopic redshifts. We describe the redshift leakage from a galaxy photometric redshift bin $j$ into a spectroscopic redshift bin $i$ using the sum of the $p(z)$ for the galaxies residing in bin $j$. We can then predict the angular cross-correlation between photometric and spectroscopic galaxies due to intrinsic galaxy clustering when $i \neq j$ as a function of the measured angular cross-correlation when $i=j$. We also account for enhanced clustering arising from lensing magnification using a halo model. The comparison of this prediction with the measured signal provides a consistency check on the validity of using the summed $p(z)$ to determine galaxy redshift distributions in cosmological analyses, as advocated by the Canada-France-Hawaii Telescope Lensing Survey (CFHTLenS). We present an analysis of the photometric redshifts measured by CFHTLenS, which overlaps the Baryon Oscillation Spectroscopic Survey (BOSS). We also analyse the Red-sequence Cluster Lensing Survey (RCSLenS), which overlaps both BOSS and the WiggleZ Dark Energy Survey. We find that the summed $p(z)$ from both surveys are generally biased with respect to the true underlying distributions. If unaccounted for, this bias would lead to errors in cosmological parameter estimation from CFHTLenS by less than $\sim 4\%$. For photometric redshift bins which spatially overlap in 3-D with our spectroscopic sample, we determine redshift bias corrections which can be used in future cosmological analyses that rely on accurate galaxy redshift distributions.