2007A&A...470..761T

The Gaia astrometric survey mission will, as a consequence of its scanning law, obtain low resolution optical (330-1000nm) spectrophotometry of several million unresolved galaxies brighter than V=22. We present the first steps in a project to design and implement a classification system for these data. The goal is both to determine morphological classes and to estimate intrinsic astrophysical parameters via synthetic templates. Here we describe (1) a new library of synthetic galaxy spectra, and (2) first results of classification and parametrization experiments using simulated Gaia spectrophotometry of this library. We have created a large grid of synthetic galaxy spectra using the PEGASE.2 code, which is based on galaxy evolution models that take into account metallicity evolution, extinction correction, emission lines (with stellar spectra based on the BaSeL library). Our classification and regression models are Support Vector Machines (SVMs), which are kernel-based nonlinear estimators. We produce a basic library of about 3600 zero redshift galaxy spectra covering the main Hubble types over wavelength range 250 to 1050nm at a sampling of 1nm or less. It is computed on a regular grid of four key astrophysical parameters for each type and for intermediate random values of the same parameters. An extended library reproduces this at a series of redshifts. Initial 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. Comparing the colours of our synthetic library with Sloan Digital Sky Survey (SDSS) spectra we find good agreement over the full range of Hubble types and parameters.