Optimizing Photometric Redshift Estimation for Large Astronomical Surveys Using Boosted Decision Trees.

Abstract

Upcoming large-scale sky surveys will obtain photometric data for over 10^8 galaxies. The unprecedented size of such data sets make full spectroscopic followup impossible. Therefore, placing precision constraints on cosmological parameters—such as dark energy—will require accurate redshift estimates based on imaging data alone. In this thesis, we describe a method for estimating photometric redshifts (photo-zs) using boosted decision trees (BDTs), which we call ArborZ. We validate ArborZ and test its performance using simulated galaxy catalogs. After showing that ArborZ is robust with respect to variations between the training and evaluation sets, we apply it to data from two major astrophysical surveys: the Sloan Digital Sky Survey (SDSS) and the Dark Energy Survey (DES). We then develop a method for applying ArborZ to estimate the redshifts of galaxy clusters. We test this in simulated data and then apply it to real data from an XCS-DES cluster catalog.