zELDA II: Reconstruction of galactic Lyman-alpha spectra attenuated by the intergalactic medium using neural networks

¹ Observatori Astronòmic de la Universitat de València, Ed. Instituts d’Investigació, Parc Científic. C/ Catedrático José Beltrán, n2, 46980 Paterna, Valencia, Spain
² Departament d’Astronomia i Astrofísica, Universitat de València, 46100 Burjassot, Spain
³ Universität Heidelberg, Institut für Theoretische Astrophysik, ZAH, Albert-Ueberle-Str. 2, 69120 Heidelberg, Germany
⁴ Max Planck Institute for Astrophysics, Karl-Schwarzschild-Str. 1, 85748 Garching, Germany
⁵ Department of Astronomy, Physics Building, Tsinghua University, 100084 Beijing, China
⁶ Unidad Asociada “Grupo de Astrofísica Extragaláctica y Cosmología”, IFCA-CSIC/Universitat de València, València, Spain
⁷ Instituto de Física de Cantabria (CSIC-UC), Avda. Los Castros s/n, 39005 Santander, Spain
⁸ Institute of Science and Technology Austria (ISTA), Am Campus 1, 3400 Klosterneuburg, Austria

^★ Corresponding author: gurung.lopez@gmail.com

Received: 20 December 2024
Accepted: 12 March 2025

Abstract

Context. The observed Lyman-alpha (Lyα) line profile is a convolution of the complex Lyα radiative transfer taking place in the interstellar, circumgalactic, and intergalactic media (ISM, CGM, and IGM, respectively). Discerning the different components of the Lyα line is crucial in order to use it as a probe of galaxy formation or the evolution of the IGM.

Aims. We aim to present the second version of zELDA (redshift Estimator for Line profiles of Distant Lyman-Alpha emitters), an open-source Python module focused on modelling and fitting observed Lyα line profiles. This new version of zELDA focuses on disentangling the galactic from the IGM effects.

Methods. We built realistic Lyα line profiles that include the ISM and IGM contributions by combining the Monte Carlo radiative-transfer simulations for the so-called shell model (ISM) and IGM transmission curves generated from TNG100. We used these mock line profiles to train different artificial neural networks. These use the observed spectrum as input and the outflow parameters of the best fitting ‘shell model’ as output along with the redshift and Lyα emission IGM escape fraction of the source.

Results. We measured the accuracy of zELDA on mock Lyα line profiles. We find that zELDA is capable of reconstructing the ISM emerging Lyα line profile with high levels of accuracy (Kolmogórov-Smirnov<0.1) for 95% of the cases for HST/COS-like observations and 80% for MUSE-WIDE-like observations. zELDA is able to measure the IGM transmission with typical uncertainties below 10% for HST/COS and MUSE-WIDE data.

Conclusions. This work represents a step forward in the high-precision reconstruction of IGM-attenuated Lyα line profiles. zELDA allows the disentanglement of the galactic and IGM contribution shaping the Lyα line shape and thus allows us to use Lyα as a tool to study galaxy and ISM evolution.