Star Clusters: Constraining Gas Clearing Timescales With HST H? Imaging and Classifying Cluster Morphology With Machine Learning

This dissertation focuses on two topics related to the properties of star clusters as resolved by HST imaging taken by the Legacy ExtraGalactic UV Survey (LEGUS) and Physics at High Angular Resolution in Nearby GalaxieS (PHANGS) surveys: First, the analysis of star cluster ages in tandem with the detailed morphology of any associated HII regions can provide insight into the processes that clear a cluster’s natal gas, as well as the accuracy of cluster ages and dust extinction derived from Spectral Energy Distribution (SED) fitting. We classify 3757 star clusters in 16 nearby galaxies according to their H? morphology (concentrated, partially exposed, no emission), using HST imaging from LEGUS. We find: 1) The mean SED ages of clusters with concentrated (1-2 Myr) and partially exposed HII region morphologies (2-3 Myr) indicate a relatively early onset of gas clearing and a short (1-2 Myr) clearing timescale. 2) The extinction of clusters can be overestimated due to the presence of red supergiants, which is a result of stochastic sampling of the IMF in low mass clusters. 3) The age-reddening degeneracy impacts the results of the SED fitting - out of 169 clusters with M* > 5000 solar masses, 46 have SED ages which appear significantly underestimated or overestimated based on their environment, and the presence or absence of H?. 4) Lastly, for galaxies at 3-10 Mpc, we find that uncertainties in morphological classification due to distance-dependent resolution effects do not affect overall conclusions on gas clearing timescales when using HST H? images, whereas ground-based images do not provide sufficient resolution for the analysis. Secondly, the time required to visually inspect star cluster candidates identified in HST imaging of nearby galaxies has limited the availability of star cluster catalogs. To greatly expand upon these samples, deep transfer learning has recently been proven capable of creating models to accurately classify star cluster morphologies at production-scale for nearby spiral galaxies (D < 20 Mpc). In order to optimize the reliability of such models, we use HST UV-optical imaging of over 20,000 cluster candidates from the Physics at High Angular Resolution in Nearby GalaxieS (PHANGS) survey to create and evaluate two new sets of models: i) distance-independent, which uses the candidates from 18 galaxies regardless of galactic distance, and ii) distance-dependent, which splits the sample into three distance bins (9–12 Mpc, 14–18 Mpc, 18–23 Mpc). From our experiments, we find 1) the overall accuracy of our models are comparable to previous automated star cluster classification studies (∼60–80%), 2) for PHANGS-HST star cluster classification, our models show an improvement of a factor of two in the accuracy of classifying asymmetric and multi-peaked clusters (Class 2 and 3 objects) with respect to the previous method (Wei et al. 2020), 3) while we observe, at best, a weak negative correlation between model accuracy and galactic distance, we find that training separate models for the three distance bins does not significantly improve model accuracy. Lastly, 4) we identify dependencies between model accuracy and the properties of clusters such as brightness, color, and SED age