Understanding the properties of spectral distributions of daylight and its dynamical changes at different sites with varying atmospheric conditions is an active research area. Results obtained in these studies are of interest in many fields such as color vision, meteorology, biology, photography, etc… Many efforts to measure and model daylight spectra originate in the 60s [1, 2, 3, 4, 5]. The SMARTS model [6] is one example of recent tools to compute clear sky spectral irradiance from a description of atmospheric conditions, time and solar geometries. In this paper we describe our investigation of daylight spectra based on the principal component analysis (PCA) of spectra. We will show that the structure of the space leads to the well-known Poincaré disk model of hyperbolic geometry. The natural group SU(1,1) operating on the disk is then used to model the PCA representations of measured daylight spectra sequences and spectra generated by the SMARTS model.