Pz Cats: Photometric redshift catalogs based on DES Y3 BAO sample

TL;DR

This paper evaluates four photometric redshift estimators on DES Y3 BAO data, exploring PDF-based galaxy selection to improve redshift accuracy and bias, and publishes resulting catalogs for cosmological analysis.

Contribution

It introduces a novel PDF-based galaxy selection method to enhance redshift estimation accuracy and compares four estimators on DES Y3 data, providing publicly available catalogs.

Findings

ANNz2 is the most reliable estimator across criteria.

PDF selection improves colour representation and reduces bias.

Despite smooth PDFs, catastrophic errors still occur.

Abstract

Over the years, photometric redshift estimation (photo-z) has advanced through various methods. This study evaluates four distinct photo-z estimators-ANNz2, BPZ, ENF, and DNF-using the Dark Energy Survey Y3 BAO Sample. Unlike most studies, we explore selecting optimal galaxies based on their redshift Probability Distribution Function (PDF) by either reducing noise or identifying those approximating a Gaussian distribution. We cross-matched 25,760 galaxies drawn from four spectroscopic surveys with the photo-z sample to comprehend redshift bias and its 68th percentile $\sigma_{68}$. The lowest $\sigma$ for all estimators was found in the range $0.79<z_p<0.85$. Among the estimators, DMF exhibited the greatest bias, while ENF, ANNz2, and BPZ showed decreased precision outside 0.7 to 0.9 redshift range. To select galaxies with minimal bias, ANNz2 emerged as the most reliable algorithm across all criteria. PDFs selection significantly improves colour representation over the spectroscopic sample, underscoring the role of magnitude space in selection. While ANNz2 achieved superior precision, ENF poorly selected Gaussian PDFs, leaving few galaxies for LSS evaluation. Despite smooth PDFs, catastrophic redshift errors were present. Though DNF had the poorest precision, it offered enough galaxies for cosmological use. Subsampling galaxies with secondary peaks less than 30\% of the main peak height, termed Small Peaks, showed ANNz2 excelled. The catalogs produced have been published as Pz Cats within the ZENODO repository.