The morphological indicators of gas mass fraction for low-redshift galaxies
Abstract
The growth of galaxies is regulated by the amount of cold gas available in the ISM. Neutral atomic hydrogen (HI) gas comprises the bulk of z ~ 0 galaxies' ISM masses, but remains difficult to detect due to the faintness of its 21-cm emission line. To circumvent this observational challenge, astronomers have devised optical-wavelength heuristics in order to estimate gas mass fraction (MHI/Mstar) of a galaxy. Unfortunately, these methods often trade off interpretability in favor of accuracy, or vice versa; in other words, simple color-based estimators typically have larger scatter, whereas complex models perform better but can have millions of parameters. We train a deep convolutional neural network (CNN) to accurately predict galaxies' gas mass fractions using only gri SDSS imaging crossmatched with ALFALFA and xGASS. We visualize trained CNNs using the Grad-CAM algorithm, which highlights morphological features that are associated with high or low gas mass fraction for an input galaxy image. Bright HII regions and wispy blue ISM features indicate gas richness, while redder galaxy bulges and smooth stellar distributions convey gas poorness. Interpretable and accurate deep learning results will multiply the scientific returns of future large-area optical and HI surveys.
- Publication:
-
American Astronomical Society Meeting Abstracts #235
- Pub Date:
- January 2020
- Bibcode:
- 2020AAS...23520814W