Low/High Redshift Classification of Emission Line Galaxies in the HETDEX Survey

TL;DR

This paper evaluates methods for distinguishing high- from low-redshift emission line galaxies in the HETDEX survey, using spectroscopic, photometric, and statistical techniques including machine learning.

Contribution

It introduces and compares three separation methods—equivalent width cut, Bayesian, and machine learning—for efficient galaxy classification in large surveys.

Findings

Support vector machines perform well in galaxy classification.

Bayesian method offers probabilistic separation.

Simple equivalent width cut is effective for quick filtering.

Abstract

We discuss different methods to separate high- from low-redshift galaxies based on a combination of spectroscopic and photometric observations. Our baseline scenario is the Hobby-Eberly Telescope Dark Energy eXperiment (HETDEX) survey, which will observe several hundred thousand Lyman Alpha Emitting (LAE) galaxies at 1.9 < z < 3.5, and for which the main source of contamination is [OII]-emitting galaxies at z < 0.5. Additional information useful for the separation comes from empirical knowledge of LAE and [OII] luminosity functions and equivalent width distributions as a function of redshift. We consider three separation techniques: a simple cut in equivalent width, a Bayesian separation method, and machine learning algorithms, including support vector machines. These methods can be easily applied to other surveys and used on simulated data in the framework of survey planning.