2020MNRAS.497.2688C

The spectroscopic features of white dwarfs are formed in the thin upper layer of their stellar photosphere. These features carry information about the white dwarf's surface temperature, surface gravity, and chemical composition (hereafter 'labels'). Existing methods to determine these labels rely on complex ab-initio theoretical models, which are not always publicly available. Here, we present two techniques to determine atmospheric labels from white dwarf spectra: a generative fitting pipeline that interpolates theoretical spectra with artificial neural networks and a random forest regression model using parameters derived from absorption line features. We test and compare our methods using a large catalogue of white dwarfs from the Sloan Digital Sky Survey (SDSS), achieving the same accuracy and negligible bias as compared to previous studies. We package our techniques into an open-source PYTHON module 'WDTOOLS' that provides a computationally inexpensive way to determine stellar labels from white dwarf spectra observed from any facility. We will actively develop and update our tool as more theoretical models become publicly available. We discuss applications of our tool in its present form to identify interesting outlier white dwarf systems including those with magnetic fields, helium-rich atmospheres, and double-degenerate binaries.