A Bayesian classifier for photometric redshifts: identification of high-redshift clusters

Photometric redshift classifiers provide a means of estimating galaxy redshifts from observations using a small number of broad-band filters. However, the accuracy with which redshifts can be determined is sensitive to the star formation history of the galaxy, for example the effects of age, metallicity and ongoing star formation. We present a photometric classifier that explicitly takes into account the degeneracies implied by these variations, based on the flexible stellar population synthesis code of Kodama & Arimoto. The situation is encouraging, because many of the variations in stellar populations introduce colour changes that are degenerate. We use a Bayesian inversion scheme to estimate the likely range of redshifts compatible with the observed colours. When applied to existing multiband photometry for Abell 370, most of the cluster members are correctly recovered with little field contamination. The inverter is focused on the recovery of a wide variety of galaxy populations in distant (z similar to 1) clusters from broadband colours covering the 4000-Angstrom break. It is found that this can be achieved with impressive accuracy (\Delta z\ < 0.1), allowing detailed investigation into the evolution of cluster galaxies with little selection bias.