Alice Desmons

In this project, I aimed to address the issues associated with the detection of galaxies with tidal features. As the new generation of optical imaging surveys produce larger and larger datasets of galaxy images, the detection of galaxies with tidal features using purely human classification is becoming increasingly unfeasible. To address this issue, I developed a self-supervised machine learning model to automate the detection of galaxies exhibiting tidal features. This model was trained on data from the Hyper Suprime-Cam Subaru Strategic Program (HSC-SSP) survey. The model was found to outperform other state-of-the-art supervised machine learning models available at the time, and its self-supervised architecture greatly reduced the amount of labelled data needed to train the model.

In the second project, I used this machine learning model alongside visual classification to assemble a large sample of galaxies with tidal features from HSC-SSP. I then investigated the prevalence of tidal features in galaxies, and how this prevalence changed with galaxy properties such as stellar mass, the mass of the dark matter halo in which they resided, and the redshift of these galaxies. My final sample consisted of ~34,000 galaxies, 1646 of which had visible tidal features. I found that the prevalence of tidal features increased with stellar mass, and decreased with redshift, although this relationship with redshift could be attributed to the fact that tidal features are more difficult to detect at higher redshifts. I also found that galaxies in groups had higher rates of tidal features than more isolated field galaxies, or galaxies in denser cluster environments.

In the third project, I used this same sample of galaxies to investigate the colours of tidal features in galaxies. Since the colours of galaxies reflect the age and metallicity of their stellar populations, and are correlated with their stellar masses, I wanted to know if the colours of tidal features could be used to draw conclusions about their progenitor galaxies. To do this, I designed a pipeline to model the light from galaxies in my sample and attempted to isolate the light from the tidal features. I found that tidal features were generally bluer than their host galaxies, and that red sequence galaxies with tidal features had redder outskirts than their non-tidal feature hosting counterparts, consistent with the outskirts of galaxies being composed of light accreted through mini and minor gas-poor mergers as predicted from theory and simulations. I also found evidence consistent with different types of tidal features having different formation mechanisms. More importantly, I demonstrated that an automated pipeline designed to compare large statistical samples of galaxies could be used to detect colour differences between tidal features and their host galaxies.