Date Published:

July 1, 2017

Abstract:

With an eye towards the computational requirements of future large-scalesurveys such as Euclid and Large Synoptic Survey Telescope (LSST) thatwill require photometric redshifts (photo-z's) for ≳ 10⁹objects, we investigate a variety of ways that 'fuzzy archetypes' can beused to improve photometric redshifts and explore their respectivestatistical interpretations. We characterize their relative performanceusing an idealized LSST ugrizY and Euclid YJH mock catalogue of 10 000objects spanning z = 0-6 at Y = 24 mag. We find most schemes are able torobustly identify redshift probability distribution functions that aremultimodal and/or poorly constrained. Once these objects are flagged andremoved, the results are generally in good agreement with the strictaccuracy requirements necessary to meet Euclid weak lensing goals formost redshifts between 0.8 ≲ z ≲ 2. These results demonstratethe statistical robustness and flexibility that can be gained bycombining template-fitting and machine-learning methods and provideuseful insights into how astronomers can further exploit thecolour-redshift relation.

Website