Photometric redshifts for the SDSS Data Release 12

We present the methodology and data behind the photometric redshift database of the Sloan Digital Sky Survey Data Release 12 (SDSS DR12). We adopt a hybrid technique, empirically estimating the redshift via local regression on a spectroscopic training set, then fitting a spectrum template to obtain K-corrections and absolute magnitudes. The SDSS spectroscopic catalog was augmented with data from other, publicly available spectroscopic surveys to mitigate target selection effects. The training set is comprised of $1,976,978$ galaxies, and extends up to redshift $z\approx 0.8$, with a useful coverage of up to $z\approx 0.6$. We provide photometric redshifts and realistic error estimates for the $208,474,076$ galaxies of the SDSS primary photometric catalog. We achieve an average bias of $\overline{\Delta z_{\mathrm{norm}}} = 5.84 \times 10^{-5}$, a standard deviation of $\sigma \left(\Delta z_{\mathrm{norm}}\right)=0.0205$, and a $3\sigma$ outlier rate of $P_o=4.11\%$ when cross-validating on our training set. The published redshift error estimates and photometric error classes enable the selection of galaxies with high quality photometric redshifts. We also provide a supplementary error map that allows additional, sophisticated filtering of the data.