CNN photometric redshifts in the SDSS at $r\leq 20$

TL;DR

This paper presents a CNN-based method to estimate photometric redshifts for approximately 14 million SDSS galaxies, achieving high accuracy and reliability, and providing publicly accessible data and tools.

Contribution

The study introduces a CNN trained on galaxy images to produce accurate photometric redshifts with well-behaved PDFs, outperforming existing methods and addressing biases in training data.

Findings

Median bias of 0.0015 in redshift estimates

Catastrophic failure rate of 4% at r<19.8

Redshift distributions align with tomographic analyses

Abstract

We release photometric redshifts, reaching $\sim$0.7, for $\sim$14M galaxies at $r\leq 20$ in the 11,500 deg$^2$ of the SDSS north and south galactic caps. These estimates were inferred from a convolution neural network (CNN) trained on $ugriz$ stamp images of galaxies labelled with a spectroscopic redshift from the SDSS, GAMA and BOSS surveys. Representative training sets of $\sim$370k galaxies were constructed from the much larger combined spectroscopic data to limit biases, particularly those arising from the over-representation of Luminous Red Galaxies. The CNN outputs a redshift classification that offers all the benefits of a well-behaved PDF, with a width efficiently signaling unreliable estimates due to poor photometry or stellar sources. The dispersion, mean bias and rate of catastrophic failures of the median point estimate are of order $\sigma_{\rm MAD}=0.014$, <$\Delta z_{\rm norm}$>$=0.0015$, $\eta(|\Delta z_{\rm norm}|>0.05)=4\%$ on a representative test sample at $r<19.8$, out-performing currently published estimates. The distributions in narrow intervals of magnitudes of the redshifts inferred for the photometric sample are in good agreement with the results of tomographic analyses. The inferred redshifts also match the photometric redshifts of the redMaPPer galaxy clusters for the probable cluster members. The CNN input and output are available at: https://deepdip.iap.fr/treyer+2023.