Photometric redshifts for the SDSS Data Release 12

Abstract

We present the methodology and data behind the photometric redshift database\nof the Sloan Digital Sky Survey Data Release 12 (SDSS DR12). We adopt a hybrid\ntechnique, empirically estimating the redshift via local regression on a\nspectroscopic training set, then fitting a spectrum template to obtain\nK-corrections and absolute magnitudes. The SDSS spectroscopic catalog was\naugmented with data from other, publicly available spectroscopic surveys to\nmitigate target selection effects. The training set is comprised of $1,976,978$\ngalaxies, and extends up to redshift $z\\approx 0.8$, with a useful coverage of\nup to $z\\approx 0.6$. We provide photometric redshifts and realistic error\nestimates for the $208,474,076$ galaxies of the SDSS primary photometric\ncatalog. We achieve an average bias of $\\overline{\\Delta z_{\\mathrm{norm}}} =\n5.84 \\times 10^{-5}$, a standard deviation of $\\sigma \\left(\\Delta\nz_{\\mathrm{norm}}\\right)=0.0205$, and a $3\\sigma$ outlier rate of $P_o=4.11\\%$\nwhen cross-validating on our training set. The published redshift error\nestimates and photometric error classes enable the selection of galaxies with\nhigh quality photometric redshifts. We also provide a supplementary error map\nthat allows additional, sophisticated filtering of the data.\n