A gradient boosting and broadband approach to finding Lyman-α emitting galaxies beyond narrowband surveys

¹ Instituto de Astrofísica e Ciências do Espaço, Universidade do Porto, CAUP, Rua das Estrelas, PT4150-762 Porto, Portugal
² Departamento de Física e Astronomia, Faculdade de Ciências, Universidade do Porto, Rua do Campo Alegre 687, PT4169-007 Porto, Portugal
³ DTx–Digital Transformation CoLab, Building 1, Azurém Campus, University of Minho, PT4800-058 Guimarães, Portugal
⁴ Celfocus, Avenida Dom João II, 34, Parque das Nações, 1998-031 Lisbon, Portugal
⁵ Instituto de Astrofísica e Ciências do Espaço, Universidade de Lisboa, OAL, Tapada da Ajuda, 1349-018 Lisbon, Portugal
⁶ Departamento de Física, Faculdade de Ciências, Universidade de Lisboa, Edifício C8, Campo Grande, 1749-016 Lisbon, Portugal

^⋆ Corresponding author.

Received: 15 April 2025
Accepted: 28 July 2025

Abstract

Context. The identification of Lyman-α emitting galaxies (LAEs) has traditionally relied on dedicated surveys using custom narrowband filters, which constrain observations to specific narrow redshift intervals, or on blind spectroscopy, which although unbiased, typically requires extensive telescope time. This makes it challenging to assemble large statistically robust galaxy samples. With the advent of wide-area astronomical surveys producing datasets that are significantly larger than traditional surveys, the need for new techniques arises.

Aims. We test whether gradient-boosting algorithms, trained on broadband photometric data from traditional LAE surveys, can efficiently and accurately identify LAE candidates from typical star-forming galaxies at similar redshifts and brightness levels.

Methods. Using galaxy samples at z ∈ [2, 6] derived from the COSMOS2020 and SC4K catalogs, we trained gradient-boosting machine-learning algorithms (LGBM, XGBoost, and CatBoost) using optical and near-infrared broadband photometry. To ensure balanced performance, the models were trained on carefully selected datasets with similar redshift and i-band magnitude distributions. Additionally, the models were tested for robustness by perturbing the photometric data using the associated observational uncertainties.

Results. Our classification models achieved F1-scores of ∼87% and successfully identified about 7000 objects with an unanimous agreement across all models. This more than doubles the number of LAEs identified in the COSMOS field compared with the SC4K dataset. We managed to spectroscopically confirm 60 of these LAE candidates using the publicly available catalogs in the COSMOS field.

Conclusions. These results highlight the potential of machine learning in efficiently identifying LAEs candidates. This lays the foundations for applications to larger photometric surveys, such as Euclid and LSST. By complementing traditional approaches and providing robust preselection capabilities, our models facilitate the analysis of these objects. This is crucial to increase our knowledge of the overall LAE population.