What factors shape the radio luminosity of star-forming galaxies? A new calibration from LoTSS-DR2

TL;DR

This study investigates how galaxy properties influence the radio luminosity of star-forming galaxies, providing a new calibration of the RC--SFR relation using LOFAR data and decision-tree analysis.

Contribution

It introduces a novel calibration of the RC--SFR relation based on a large sample and decision-tree regression, emphasizing the roles of SFR and stellar mass.

Findings

SFR is the most important predictor of radio luminosity.

Stellar mass also significantly influences radio luminosity.

The new calibration is consistent across different SFR and stellar mass estimation methods.

Abstract

Radio observations offer a dust-unobscured view of galaxy star formation via the radio continuum-star formation rate (RC--SFR) relation. Emerging evidence of a stellar mass dependence in the RC--SFR relation raises the broader question of how other galaxy properties may influence this relation. In this work, we study the dependence of the global RC--SFR relation on galaxy properties in local ($z\,\leq$\,0.3) star-forming galaxies (SFGs) using the second data release of the LOFAR Two-Metre Sky Survey (LoTSS-DR2). Employing a non-parametric decision-tree regression algorithm, we identify the most important galaxy properties for estimating the radio luminosity using a sample of 18,828 emission-line-classified SFGs based on spectroscopic data from the SDSS-DR8. Along with the spectroscopically obtained SFRs and stellar mass values, we also use SFRs and stellar masses derived using photometric SED-fitting from the \textit{GALEX}--SDSS--\textit{WISE} Legacy Catalogue (GSWLC) for the same sample. We find that a galaxy's SFR is most important for predicting the radio luminosity, followed by the stellar mass, at $>5\sigma$ significance. Complementing the LoTSS catalogue 150\,MHz flux densities with aperture photometry for the rest of the emission-line classified sample (35,099 galaxies in total), we obtain a new calibration of the RC--SFR relation, which does not change significantly whether we use spectroscopic or photometrically derived SFRs and stellar masses, despite the fact that the methods probe star formation on different characteristic timescales. Our results highlight the utility of decision-tree algorithms for handling censored radio-selected galaxy samples, which will be useful for future spectroscopic surveys of radio sources.