Photo-z performance for precision cosmology

Current and future weak-lensing surveys will rely on photometrically estimated redshifts of very large numbers of galaxies. In this paper, we address several different aspects of the demanding photo-z performance that will be required for future experiments, such as the proposed ESA Euclid mission. It is first shown that the proposed all-sky near-infrared photometry from Euclid, in combination with anticipated ground-based photometry (e.g. PanStarrs-2 or DES) should yield the required precision in individual photo-z of sigma(z)(z) <= 0.05(1 + z) at I-AB <= 24.5. Simple a priori rejection schemes based on the photometry alone can be tuned to recognize objects with wildly discrepant photo-z and to reduce the outlier fraction to <= 0.25 per cent with only modest loss of otherwise usable objects. Turning to the more challenging problem of determining the mean redshift < z > of a set of galaxies to a precision of vertical bar Delta < z >vertical bar <= 0.002(1 + z) we argue that, for many different reasons, this may be best accomplished by relying on the photo-z themselves rather than on the direct measurement of < z > from spectroscopic redshifts of a representative subset of the galaxies, as has usually been envisaged. We present in Appendix A an analysis of the substantial difficulties in the latter approach that arise from the presence of large-scale structure in spectroscopic survey fields. A simple adaptive scheme based on the statistical properties of the photo-z likelihood functions is shown to meet this stringent systematic requirement, although further tests on real data will be required to verify this. We also examine the effect of an imprecise correction for Galactic extinction on the photo-z and the precision with which the Galactic extinction can be determined from the photometric data itself, for galaxies with or without spectroscopic redshifts. We also explore the effects of contamination by fainter overlapping objects in photo-z determination. The overall conclusion of this paper is that the acquisition of photometrically estimated redshifts with the precision required for Euclid, or other similar experiments, will be challenging but possible.