Improved photometric redshifts via enhanced estimates of system response, galaxy templates and magnitude priors

Wide, deep photometric surveys require robust photometric redshift estimates (photo-zs) for studies of large-scale structure. These estimates depend critically on accurate photometry. We describe the improvements to the photometric calibration and the photo-z estimates in the Deep Lens Survey (DLS) from correcting three of the inputs to the photo-z calculation: the system response as a function of wavelength, the spectral energy distribution templates and template prior probabilities as a function of magnitude. We model the system response with a physical model of the MOSAIC camera's CCD, which corrects a 0.1 mag discrepancy in the colours of type M2 and later stars relative to the Sloan Digital Sky Survey z photometry. We provide our estimated z response function for the use of other surveys that used MOSAIC before its recent detector upgrade. The improved throughput curve, template set and Bayesian prior lead to general improvement in the predicted photometric redshifts, including a 20 per cent reduction in photo-z scatter at low redshift and a reduction of the bias by a factor of more than 2 at high redshift. This paper serves as both a photo-z data release description for DLS and a guide for testing the quality of photometry and resulting photo-zs generally.