Machine Learning based photometric redshifts for the KiDS ESO DR2 galaxies

TL;DR

This paper presents a machine learning approach to estimate photometric redshifts for over 1.1 million galaxies in the KiDS survey, achieving high accuracy and low bias, aiding cosmology and galaxy evolution studies.

Contribution

The study introduces a new catalog of photometric redshifts for KiDS galaxies using the MLPQNA model trained on combined spectroscopic datasets, with improved accuracy over previous methods.

Findings

Photometric redshifts for 1.1 million galaxies with ~0.03 uncertainty

Bias in redshift estimates is approximately 0.001

Catastrophic outlier fraction is around 0.4%

Abstract

We estimated photometric redshifts (zphot) for more than 1.1 million galaxies of the ESO Public Kilo-Degree Survey (KiDS) Data Release 2. KiDS is an optical wide-field imaging survey carried out with the VLT Survey Telescope (VST) and the OmegaCAM camera, which aims at tackling open questions in cosmology and galaxy evolution, such as the origin of dark energy and the channel of galaxy mass growth. We present a catalogue of photometric redshifts obtained using the Multi Layer Perceptron with Quasi Newton Algorithm (MLPQNA) model, provided within the framework of the DAta Mining and Exploration Web Application REsource (DAMEWARE). These photometric redshifts are based on a spectroscopic knowledge base which was obtained by merging spectroscopic datasets from GAMA (Galaxy And Mass Assembly) data release 2 and SDSS-III data release 9. The overall 1 sigma uncertainty on Delta z = (zspec - zphot) / (1+ zspec) is ~ 0.03, with a very small average bias of ~ 0.001, a NMAD of ~ 0.02 and a fraction of catastrophic outliers (| Delta z | > 0.15) of ~0.4%.