Detecting Metal Absorbers in Quasar Spectra: A Machine Learning Approach

Large scale surveys such as SDSS, DESI and upcoming 4MOST surveys provide an unprecedented amount of quasar spectra to explore. One of these surveys is our upcoming 4MOST community survey 4HI-Q (preliminary title). The survey will obtain ~ 1 million R=20 000 quasar spectra to search for HI and metal absorption line systems in quasar sightlines at low to high redshifts. Given the vast number of spectra, it is unfeasible to explore them individually to infer systems of interest. Therefore, in preparation for our survey, we are aiming to create a pipeline which efficiently fits the quasar continuum and subsequently identifies systems including metal absorption lines. Machine learning has become an important tool in astrophysics to approach such problems (e.g. Parks+2018, Wang+2022). Using TNG50 from the Illustris Project (e.g. Nelson+2019), we create realistic metal absorption lines and inject them into quasar mock spectra including Lyman-alpha forests. We use this dataset to train an autoencoder for quasar continuum fitting and a convolutional neural network (CNN) for the detection of metal absorption lines within quasar sightlines. This approach of deriving metal absorption lines from simulations, will also aid in exploring the physical parameters of the systems responsible for the absorption lines in the future.