The VIMOS Public Extragalactic Redshift Survey (VIPERS). Unsupervised classification with photometric redshifts: a method to accurately classify large galaxy samples without spectroscopic information

Techniques to classify galaxies and estimate their properties based on the photometric observations solely will be necessary for future large cosmology missions, such as Euclid. However, the classification accuracy is always lower in photometric surveys and can be systematically biased with respect to classifications based upon spectroscopic data. Our aim is to test how precisely the detailed classification scheme introduced by Siudek et al. accepted for galaxies at z~0.7 could be reproduced if only photometric data were available. We applied the FEM unsupervised clustering algorithm to 54,293 VIPERS galaxies. To construct the parameter space, we used reliable photometric redshifts and 12 rest frame magnitudes obtained by the SED-fitting with the photo-z scatter~0.03, and the outlier rate~2%. Subsequently, we compared the results to the analogous classification based on spectroscopic redshifts and corresponding absolute magnitudes. The FEM algorithm divided the VIPERS data into four main groups: (1) red, (2) green, (3) blue, and (4) outliers. Each group was further divided into 3, 3, 4, and 2 subclasses, respectively. The accuracy of reproducing galaxy classes using spectroscopic data is high 92%, 84%, 96% for red, green, and blue classes, respectively. Our method was able to efficiently separate outliers (stars and broad-line AGNs), but failed to distinguish a class of star-forming galaxies with redder r-K colours and higher SFR than remaining star-forming classes. The presented verification of the photometric classification demonstrates that large photometric samples can be used to distinguish different galaxy classes at z>0.5 with an accuracy provided so far only by spectroscopic data except for particular galaxy classes. Confirming the synergy between classifications based on spectroscopic and photometric data demonstrates the usefulness of our approach in future large surveys.