Data Science & Archives

Astronomy is firmly in its "big data" era, with new discoveries and analyses being built on bigger datasets and more complex tools. The Cool Star Lab helps to create the necessary data archives while developing new machine learning search and analysis tools.

Current Projects

SpeX Archive & pySpextool

Supported by a NASA ADAP grant, our undergraduates are reducing over 20 years of IRFT/SpeX spectral data for public release, using a new python-based version of the Spextool package, pySpextool, led by Prof. Michael Cushing.

Machine Learning Algorithms

Our team is developing new machine learning tools to find and characterize brown dwarfs in large imaging and spectroscopic surveys.

SPLAT

SPLAT (Spex Prism Library Analysis Toolkit) is a long-standing project to curate low-resolution spectral data and vaious spectral analysis tools in a python catalog. SPLAT is now available through pip installation!

Advanced Model Fitting

Our team is developing new ways of conducting forward model fits to spectral and photometric data through our ucdmcmc package and machine learning approaches, as well as retrieval methods.

Recent Findings

Machine Learning Models Find Brown Dwarf Binaries

(December 2025) Former Cool Star Lab undergraduate researcher Juan Diego Draxl Giannoni reported the findings of a machine learning project to identify unresolved brown dwarf binaries from combined-light spectra. Using synthetic binaries constructed from the SpeX Prism Library, the team created a hierarchical random forest model that can both distinguish single and binary systems and classify binary components. The models achieved an overall precision of 85% and classification errors of ≤1 subtype, far better than earlier index-based approaches. This study highlights the utility of machine-learning methods for uncovering rare binaries among the large samples of spectra anticipated from SPHEREx and Euclid (read the preprint by Draxl Giannoni et al.).

Distant Brown Dwarfs Discovered as JWST RUBIES

(November 2025) Recent UCSD graduate Sara Morrissey has reported the findings of her Honors thesis, the discovery of 7 distant L and T dwarfs in deep JWST spectroscopy by the RUBIES survey. Sara used the 1-5 µm spectra to classify her discoveries and determine their temperatures, metallicities, and distances, the last reaching out to 3,000 pc from the Sun. Two of the sources show evidence of being metal-poor brown dwarfs. Congratulations on your first peer-reviewed article Sara! (see the preprint by Morrissey et al.)

Machine Learning to Classify Brown Dwarfs

(September 2025) Cool Star Lab undergraduate researcher Tianxing Zhou has led a new study investigating machine learning classification methods for low-temperature dwarfs. Drawing on a set of low-resolution near-infrared spectra from the SPLAT archive, Tianxing explored multiple ML models, and found that a k-nearest neighbors algorithm was able to classify 96% of sources to with one subtype and assign gravity and metallicity classifications with 90% accuracy. Tianxing's work advances tools for studying large samples of spectra now emerging from space telescopes such as Euclid and JWST (see the ApJ article by Zhou et al.)

Putting Machine Learning to the Test

(May 2025) Cool Star Lab PI Adam Burgasser partnered with U. Bern graduate student Anna Lueber to investigate machine learning approaches to modeling brown dwarf spectra. They compared the traditional "best fit" approach using the Monte Carlo Markov Chain (MCMC) algorithm to a machine learning algorithm know as Random Forest Retrievals that uses decision trees to determine best fit parameters. The study found that the MCMC approach yields betters fits and parameter constraints, while the RFR approach yields similar parameters and is considerably faster. The study proposes combining these methods for efficient exploration of multiple sets of models (read the preprint by Lueber & Burgasser.).

A Comprehensive Study of Metal-poor T Subdwarfs

(November 2024) CSL Director Adam Burgasser and members of the Backyard Worlds: Planet 9 team have conducted a comprehensive study of metal-poor T dwarfs, including sources discovered by citizen scientists from multi-epoch WISE data. Selecting sources based on reduced proper motion, the team identified dozens of metal-poor objects, including three "extreme" cases. They also identified three metal-rich sources with thick disk kinematics, likely ejected from the inner Milky Way. 3D kinematics enabled by Keck/NIRES observations reveal that two sources may be part of the Thamnos population, and one source part of the Helmi stream. They study also made the first metallicity classification system for T (sub)dwarfs, and defined a metallicity index for near-infrared spectra. This work helps ongoing studies that are searching for thick-disk and halo brown dwarfs in deep JWST and Euclid fields (read the preprint by Burgasser et al.)

New atmosphere models for metal-poor brown dwarfs

(May 2024) Cool Star Lab undergraduate researcher and UC LEADS Scholar Efrain Alvardo led the release of a new set of atmosphere models for metal-poor brown dwarfs. The Spectral ANalogs of Dwarfs (SAND) models were developed with form CSL graduate student Roman Gerasimov, and fills an important gap in current brown dwarf modeling suites (read the Research Note by Alvarado et al.)

Using Machine Learning to classify ultracool dwarfs

(April 2024) Cool Star Lab undergraduate researcher Tianxing "Sky" Zhou has a led a study investigating machine learning methods for spectral classification of ultracool dwarf spectra. Sky was able to demonstrate 95%+ reliability for k-nearest neighbors and random forest methods over a broad range of spectra types (read the Research Note by Zhou et al.)

JWST Uncovers Distant Brown Dwarfs on the Outskirts of the Milky Way

(Sep 2023) Adam Burgasser and Roman Gerasimov led an article analyzing JWST/NIRSpec data of three distant T dwarfs identified in the UNCOVER survey of the Abell 2744 lensing field. The NIRSpec prism data allowed full analysis of the 1-5 µm spectra, revealing all three to be T dwarfs at kiloparsec distances, two with evidence of subsolar metallicities. The coldest of the three, previously identified photometrically as GLASS-BD-1, shows evidence of phosphine in its infrared spectra, a potential new indicator of subsolar metallicity in cool brown dwarf spectra (read the preprint by Burgasser et al.).

Pop III Study highlighted in Nature Astronomy

(Jan 2023): The study of Population III star detectability led by former CSL undergraduate Mikaela Larkin was recently singled out as a Research Highlight by Nature Astronomy (read the highlight and the article).

Using Machine Learning to find binaries

(Jan 2023): The Cool Star Lab Machine Learning group published a Research Note demonstrating spectral binary identification with a random forest classifier. The paper was led by undergraduate researcher Malina Desai (read the paper by Desai et al. at RNAAS)

Page updated

Google Sites

Report abuse