Astron. Astrophys. 332, 459-478 (1998)

8. Conclusions and prospects

In this paper we show that we can classify the galaxies of the ESO-Sculptor survey (ESS) from their flux calibrated spectra using the Principal Component Analysis (PCA) technique. The PCA allows to define a continuous spectral sequence highly correlated with the Hubble morphological type. This sequence can be written as a linear combination of a reduced set of parameters and vectors (3) which account for 98% of the total flux of each spectrum. The parameters are also sensitive to the strength of emission lines. Our main results can be summarized as follows:

1. Using Kennicutt spectra for galaxies of known Hubble types, we establish the strong correlation between the spectral galaxy type and the underlying old (red) and young (blue) stellar population within the galaxy. These populations can be quantitatively separated in the PCA approach using a sequence which arises mainly from the changes in the shape of the continuum and the relative strength of the absorption features.
2. By application to the ESS data, we show that the PCA is a flexible and powerful tool to classify galaxies using the spectral information. Galaxies can be classified using one continuous parameter (). We also find that the presence and strength of the emission lines are correlated with the spectral type (late galaxies tend to have strong emission lines), and can be quantified by a second continuous parameter (). The continuous nature of the classifying parameters and provides a powerful tool for an objective study of the systematic and non-systematic properties (i.e., peculiar objects) of spectral data. Moreover, it allows us to use one (or two) fundamental parameter(s) to construct an analytical relation between the classification parameter(s) and other quantitative properties of the galaxies (for example K-corrections, local galaxy density, etc...).
3. We illustrate using the ESS data how the PCA acts as a powerful filter of noisy spectra, inherent to deep redshift surveys. Reconstruction of the ESS spectra with 3 principal components increases the S/N from the range 8-20 to the range 35-80.
4. The spectral sequence given by the PCA is not uniformly populated: the early types are more concentrated in the classification plane than the late types. This non-uniform distribution of the different spectral types is closely related to the fact that systematic differences between two consecutive morphological types are larger among spiral galaxies than among elliptical or lenticular galaxies. This leads to construct a variable binning when comparing the observed spectral sequence with morphological classifications from other surveys.
5. (5) When making the analogy between spectral type and morphology via the Kennicutt spectra, we find that the ESS sample contains 26 7% of E/S0, 71 9% of Sa/Sb/Sc and 3 7% of extreme late spirals (Sm/Im). The type fractions for the ESS show no significant changes in the redshift interval , and are comparable to those found in other galaxy surveys at intermediate redshift. For the galaxies in the ESS analyzed here, the dominant type is Sb, followed by Sc, Sa, and early types. We do not detect any strong evolution in the ESS data as a function of redshift, up to the depth of the ESS spectroscopic catalogue (). Other surveys have detected only a marginal evolution at , like the CFRS (Lilly et al. 1996) and the Autofib survey (Heyl et al. 1997), and significant evolution appears to occur at . In the ESS sample, we note a significant excess of early types at . The nature of this excess will be further investigated in a future paper using the complete redshift sample.

Application of the PCA method to the ESS shows that it may be applied to any set of flux-calibrated spectra, and that it is a promising technique for on-going and future massive galaxy surveys. The major interest of the PCA technique is that the spectral trends followed by the sample used are independent from any set of templates (the classification space is continuously populated). The PCA technique therefore offers clear advantages over other discrete methods like the (see Zaritsky et al. 1995) or the cross-correlation method (see Heyl et al. 1997), which are fully dependent on the set of templates used: in these approaches, it is difficult to discriminate differences in the results from differences in the input templates. Also, such classification procedures are sensitive to fluctuations due to the noise of each target spectrum, an undesirable phenomenon. In contrast, the PCA offers an unsupervised classification system (Naim et al. 1995), in which one does not make any assumption on the general trends followed by the sample. Moreover, the PCA classification shows that the spectral sequence is essentially determined by the variations in the shape of the continuum. Any spectral classification method dependent only on the strength of the absorption lines (Zaritsky et al. 1995, Heyl et al. 1997) is therefore very sensitive to instrumental effects and/or physical phenomena not necessarily correlated with spectral type, and must be interpreted with caution.

The spectral classification for the ESS sample will be used to derive precise K-corrections, which are fundamental for deriving absolute magnitudes. Those in turn will allow to calculate the luminosity functions as a function of spectral type (in a subsequent paper). With the specific galaxy luminosity function, we can investigate in detail the morphology-density relation in the field (see Marzke et al. 1994) and more generally the variations in galaxy properties with local environment and location within the large-scale structure. These various analyses will be reported in subsequent papers.

© European Southern Observatory (ESO) 1998

Online publication: March 23, 1998
helpdesk.link@springer.de