Improving Photometric Redshifts by Training Complementary Features of Galaxy Templates Using Machine Learning

This study aims to improve the photometric redshifts (photo-zs) of galaxies by combining the use of template-based and machine learning algorithms, notably the Bayesian Photometric Redshift (BPz) and Artificial Neural Networks for Redshifts 2 (ANNz2), so we can advantage-leverage the complementary aspects of both techniques and achieve improved photo-z predictions. In this work, we introduce a technique where the outputs of the template-based photo-z (the best-fit template type and photo-z) are added as inputs to ANNz2, and we see that there is an improvement in [𝜎𝜎RMS, 𝜎𝜎68] giving values as low as [0.0474,0.0471], [0.0368,0.0253] and [0.0213,0.0168] in the SDSS Stripe-82, CMASS, and LOWZ samples, respectively. This study is considered an extension of our previous work to improve photo-z values, which enhances its use in fainter and deeper sky surveys, opening broader horizons to develop these methods and finding improved methods for measuring galaxy photo-zs.