Presentation description

Ultra-faint dwarfs (UFDs) are extremely faint galaxies, with stellar masses below 10,000 times the mass of the Sun. In contrast, the Milky Way Galaxy's stellar mass weighs around 50 billion solar masses. Ultra-faint dwarfs' low luminosity and sparse stellar population make identification inherently difficult. Yet despite their size, UFDs offer valuable insight into galaxy formation, dark matter evolution, and the early stages of the Universe.

Typically, observations with space telescopes are required to confirm the discoveries of UFDs. Due to limited access to space-based observation, initial UFD candidates must first be identified using ground-based imaging. In ground-based data, UFDs are often barely noticeable and require careful visual and algorithmic inspection of their color-magnitude diagrams (CMDs), which can help reveal underlying structure in the stellar properties of UFDs. However, the limited number of confirmed UFDs restricts the amount of training data available to validate and refine existing search methods.

We have developed a pipeline to generate artificial UFD candidates under realistic ground-based observing conditions, enabling both replication of the detection process and expansion of the available training data. We utilize a source injection code (ArtPop; Greco & Danieli 2022) to populate real sky images from the DESI Legacy Imaging Survey (Dey et al. 2019) with artificial stellar sources. We are able to generate realistic images of artificial galaxies that closely resemble known UFDs.

With access to the injected images, we recover the observed magnitudes of the stellar sources using Source Extraction and Photometry (SEP; Barbary et al. 2016; Bertin & Arnouts 1996), allowing us to construct CMDs that match those used in search algorithms. We further train a Gaussian Mixture Model on the resulting magnitude distributions to efficiently generate observed properties of hundreds of candidates without running SEP repeatedly. This framework allows us to test and refine search algorithms, supporting the discovery of the faintest galaxies.