Abstract |
Gaia is a European Space Agency (ESA) cornerstone mission which was launched in 2013.
While its main objective is to take a census of the stellar content of our Galaxy, it will also
observe a large number of many other objects, among them over a million of unresolved galaxies.
The objective of the Thesis is to extend, towards larger redshifts, the module Unresolved Galaxy
Classifier (UGC) that is being developed for the Gaia ground-based pipeline by the Greek team,
member of the Data Processing and Analysis Consortium (DPAC). The UGC is being developed to
use low-resolution spectra of galaxies observed by Gaia's BP/RP spectrophotometer to classify
the galaxies and to estimate specific astrophysical parameters. Our purpose is to develop an
automatic procedure, based on supervised machine learning algorithm, through the Support
Vector Machines (SVM), to predict the redshift of galaxies from BP/RP spectra up to z = 0:6.
As a very first step we found which sources in the first Gaia's archive are galaxies by cross-
matching the Gaia DR1 with the SDSS DR13 spectra table, which is composed of sources that
have been observed spectroscopically. This led us to an estimate that more than 1.5 million of
galaxies are observed by Gaia. In order to be able to create the redshift estimator, we selected
a properly restricted subset of the matched galaxies (Gaia DR1 with SDSS DR13 BOSS) and
downloaded their SDSS's photometric and spectroscopic parameters (among them the redshifts).
The corresponding SDSS BOSS spectra have been modelled for the BP/RP spectrophotometer
forming with the parameters an empirical library" of Gaia's galaxies spectra. In a series of
experiments with the SVM, trained and tested using the spectra and redshifts from the library,
we defined an optimal SVM model to be used in the UGC. In this optimization the error in
redshift prediction for galaxies with Gaia's magnitude G=17 within the range z=0.0-0.6 is 0.056,
while in the middle part of the range it is as small as 0.029. Ways to improve this performance
are discussed.
|