Automatic classification of galaxy spectra in large redshift surveys.

In my thesis work I make use of a Principal Component Analysis to classify galaxy spectra in large redshift surveys. In particular, I apply this classification to the first public data release spectra of galaxies in the range 0.4<z<1.0, collected by the VIMOS Public Extragalactic Redshift Survey (VIPERS), brought to rest-frame. I apply an iterative algorithm to simultaneously repair parts of spectra affected by noise and/or sky residuals, and reconstruct gaps due to rest-frame transformation, to obtain a set of orthogonal spectral templates, or eigenspectra, that span the diversity of galaxy types. By taking the three most significant components, I can describe the whole sample without contamination from noise. I find out that my templates effectively condense the spectral information into two coefficients, that can be related to the age and star formation rate of the galaxies. I also check the consistency of the average evolution of those coefficients in time, with the expected evolution of some model galaxies. I then examine the spectrophotometric types in this two-parameter space and identify early, intermediate, late and starburst galaxies. I exploit the PCA results, in particular the comparison between observed and reconstructed spectra, to find out, automatically, peculiar galaxy types, as AGNs, whose PCA reconstruction is usually poor. I finally apply a Linear Discriminant Analysis to the Principal Components obtained, to get a quantitative separation between active and passive objects, based on the measurements of the 4000Å break intensity and the [OII] equivalent width. I also use the LDA to help again recovering spectra of the AGN type, that may have escaped another classification. In parallel to this, I develop an automatic masking routine based on an observed-frame PCA, whose eigenspectra statistically represent the sky signal.