New Image Statistics for Detecting Disturbed Galaxy Morphologies at High Redshift

TL;DR

This paper introduces three new image statistics (multi-mode, intensity, deviation) to improve detection of disturbed galaxy morphologies at high redshift, demonstrating their effectiveness with machine learning and highlighting limitations of human classification.

Contribution

The paper presents three novel statistics for identifying disturbed galaxies at high redshift and shows their effectiveness using machine learning, surpassing human annotation capabilities.

Findings

MID statistics are most effective for disturbance detection.

Machine learning classifier achieves high accuracy with new statistics.

Human annotators have limited effectiveness at high redshift.

Abstract

Testing theories of hierarchical structure formation requires estimating the distribution of galaxy morphologies and its change with redshift. One aspect of this investigation involves identifying galaxies with disturbed morphologies (e.g., merging galaxies). This is often done by summarizing galaxy images using, e.g., the CAS and Gini-M20 statistics of Conselice (2003) and Lotz et al. (2004), respectively, and associating particular statistic values with disturbance. We introduce three statistics that enhance detection of disturbed morphologies at high-redshift (z ~ 2): the multi-mode (M), intensity (I), and deviation (D) statistics. We show their effectiveness by training a machine-learning classifier, random forest, using 1,639 galaxies observed in the H band by the Hubble Space Telescope WFC3, galaxies that had been previously classified by eye by the CANDELS collaboration (Grogin et al. 2011, Koekemoer et al. 2011). We find that the MID statistics (and the A statistic of Conselice 2003) are the most useful for identifying disturbed morphologies. We also explore whether human annotators are useful for identifying disturbed morphologies. We demonstrate that they show limited ability to detect disturbance at high redshift, and that increasing their number beyond approximately 10 does not provably yield better classification performance. We propose a simulation-based model-fitting algorithm that mitigates these issues by bypassing annotation.