Abstract

Contributed Talk - Splinter Learning

Wednesday, 23 September 2020, 11:55   (virtual room G)

Predicting resolved galaxy properties from photometric images using convolutional neural networks

Tobias Buck, Steffen Wolf
Leibniz Institut Potsdam, Universität Heidelberg

Multi-band images of galaxies reveal a huge amount of information about the morphology and structure of galaxies. However, inferring properties of the underlying stellar populations such as age, metallicity or kinematics from those images is notoriously difficult. Traditionally such information is best extracted from expensive spectroscopic observations. Here we test the information content of photometric multi-band images of galaxies and establish a connection between photometric galaxy images and their underlying physical stellar and gaseous properties via a machine learning approach. We utilize SDSS ugriz mock images from the public data release of the Illustris simulation as a proof of concept as uncertainties and systematics in the data are exactly known. We show that a thorough connection between 2d maps of stellar mass, metallicity, age and gas mass, gas metallicity with the galaxy morphology in the wavelength bands u,g,r,i,z can be learned by neural network. Thereby we recover the true stellar properties with only little scatter (<10%, compared to 20-30% statistical uncertainty from traditional mass-to-light-ratio based methods) on a pixel by pixel basis.