GAME: GAlaxy Machine learning for Emission lines

TL;DR

GAME is an improved machine learning code that accurately infers interstellar medium properties from galaxy emission spectra, now including more spectral features and noise handling, validated against real data and other methods.

Contribution

The paper introduces an optimized version of GAME with expanded spectral library, noise inclusion, and uncertainty evaluation, enhancing the accuracy and efficiency of emission line analysis.

Findings

GAME shows very good agreement with empirical metallicity estimates.

Ionization parameters from GAME are higher due to library differences.

The code is fast and uses all spectral lines simultaneously.

Abstract

We present an updated, optimized version of GAME (GAlaxy Machine learning for Emission lines), a code designed to infer key interstellar medium physical properties from emission line intensities of UV/optical/far infrared galaxy spectra. The improvements concern: (a) an enlarged spectral library including Pop III stars; (b) the inclusion of spectral noise in the training procedure, and (c) an accurate evaluation of uncertainties. We extensively validate the optimized code and compare its performance against empirical methods and other available emission line codes (PYQZ and HII-CHI-MISTRY) on a sample of 62 SDSS stacked galaxy spectra and 75 observed HII regions. Very good agreement is found for metallicity. However, ionization parameters derived by GAME tend to be higher. We show that this is due to the use of too limited libraries in the other codes. The main advantages of GAME are the simultaneous use of all the measured spectral lines, and the extremely short computational times. We finally discuss the code potential and limitations.