🤖 AI Summary
Current brown dwarf surveys suffer from severe incompleteness in detecting late-type T/Y dwarfs (spectral types T8 and later), primarily due to the scarcity of empirically confirmed objects. Method: We present the first machine learning classification framework trained exclusively on synthetic photometric data generated from state-of-the-art atmospheric models (ATMO 2020 and Sonora Bobcat), enabling construction of a high-fidelity training set two orders of magnitude larger than existing observational samples. Spectral types are automatically assigned via polynomial color–spectral type relations, and classification is performed using an ensemble learner. Contribution/Results: The framework achieves >99% classification accuracy on both synthetic and real-world datasets, with spectral type predictions accurate to 0.35±0.37 subclasses. It successfully identified an uncatalogued T8.2 candidate in the Pisces field and the UKIDSS Ultra-Deep Survey (UDS) region, substantially enhancing the completeness and efficiency of ultracool dwarf surveys.
📝 Abstract
The T and Y spectral classes represent the coolest and lowest-mass population of brown dwarfs, yet their census remains incomplete due to limited statistics. Existing detection frameworks are often constrained to identifying M, L, and early T dwarfs, owing to the sparse observational sample of ultracool dwarfs (UCDs) at later types. This paper presents a novel machine learning framework capable of detecting and classifying late-T and Y dwarfs, trained entirely on synthetic photometry from atmospheric models. Utilizing grids from the ATMO 2020 and Sonora Bobcat models, I produce a training dataset over two orders of magnitude larger than any empirical set of >T6 UCDs. Polynomial color relations fitted to the model photometry are used to assign spectral types to these synthetic models, which in turn train an ensemble of classifiers to identify and classify the spectral type of late UCDs. The model is highly performant when validating on both synthetic and empirical datasets, verifying catalogs of known UCDs with object classification metrics >99% and an average spectral type precision within 0.35 +/- 0.37 subtypes. Application of the model to a 1.5 degree region around Pisces and the UKIDSS UDS field results in the discovery of one previously uncatalogued T8.2 candidate, demonstrating the ability of this model-trained approach in discovering faint, late-type UCDs from photometric catalogs.