Improved photometric redshifts with colour-constrained galaxy templates for future wide-area surveys

TL;DR

This paper introduces a new galaxy template library constrained by deep photometry, significantly improving photometric redshift accuracy for future wide-area surveys like LSST, Euclid, and Roman.

Contribution

We develop a color-cube-based template library trained on deep photometry to enhance photometric redshift estimates with limited band data.

Findings

Achieved σ_NMAD of 0.026 and 6% outlier fraction with limited bands.

Improved photo-z precision by about 30% at redshifts 2-3.

Reduced outlier fraction by approximately 13% compared to existing methods.

Abstract

Cosmology and galaxy evolution studies with LSST, \Euclid, and {\it Roman}, will require accurate redshifts for the detected galaxies. In this study, we present improved photometric redshift estimates for galaxies using a template library that populates three-color space and is constrained by \HST/CANDELS photometry. For the training sample, we use a sample of galaxies having photometric redshifts which allows us to train on a large, unbiased galaxy sample having deep, unconfused photometry at optical-to-mid infrared wavelengths. Galaxies in the training sample are assigned to cubes in three-dimensional color space, $V-H$, $I-J$, and $z-H$. We then derive the best-fit spectral energy distributions of the training sample at the fixed CANDELS median photometric redshifts to construct the new template library for each individual color cube (i.e. color-cube-based template library). We derive photometric redshifts (photo-z) of our target galaxies using our new color-cube-based template library and with photometry in only a limited set of bands, as expected for the aforementioned surveys. As a result, our method yields $\sigma_{NMAD}$ of 0.026 and an outlier fraction of 6\% using only photometry in the LSST and \Euclid/{\it Roman} bands. This is an improvement of $\sim$10\% on $\sigma_{NMAD}$ and a reduction in outlier fraction of $\sim$13\% compared to other techniques. In particular, we improve the photo-z precision by about 30\% at $2 < z < 3$. We also assess photo-z improvements by including $K$ or mid-infrared bands to the $ugrizYJH$ photometry. Our color-cube-based template library is a powerful tool to constrain photometric redshifts for future large surveys.