Towards a library of synthetic galaxy spectra and preliminary results of classification and parametrization of unresolved galaxies for Gaia - II

TL;DR

This paper presents an extended library of synthetic galaxy spectra for Gaia, demonstrating promising initial results in galaxy classification and parameter estimation using simulated data.

Contribution

It introduces a new, extensive library of synthetic galaxy spectra based on improved models, covering most SDSS galaxy types, and applies machine learning for classification and parametrization.

Findings

Synthetic library covers 28,885 spectra across galaxy types.

Good agreement with observed galaxy spectra.

Initial classification results are promising with low bias and variance.

Abstract

This paper is the second in a series, implementing a classification system for Gaia observations of unresolved galaxies. Our goals are to determine spectral classes and estimate intrinsic astrophysical parameters via synthetic templates. Here we describe (1) a new extended library of synthetic galaxy spectra, (2) its comparison with various observations, and (3) first results of classification and parametrization experiments using simulated Gaia spectrophotometry of this library. Using the PEGASE.2 code, based on galaxy evolution models that take account of metallicity evolution, extinction correction, and emission lines (with stellar spectra based on the BaSeL library), we improved our first library and extended it to cover the domain of most of the SDSS catalogue. We produce an extended library of 28885 synthetic galaxy spectra at zero redshift covering four general Hubble types of galaxies, over the wavelength range between 250 and 1050 nm at a sampling of 1 nm or less. The library is also produced for 4 random values of redshift in the range of 0-0.2. It is computed on a random grid of four key astrophysical parameters (infall timescale and 3 parameters defining the SFR) and, depending on the galaxy type, on two values of the age of the galaxy. The synthetic library was compared and found to be in good agreement with various observations. The first results from the SVM classifiers and parametrizers are promising, indicating that Hubble types can be reliably predicted and several parameters estimated with low bias and variance.