A new statistical methodology for classification of sky luminance distributions based on scan data

In the present work, the sky luminance distribution in Garston, south England, is studied using the multivariate statistical methods of Factor Analysis and Cluster Analysis. The database consists of sky luminance scan data for the year 1992. The whole methodology is applied for 5 solar altitude intervals (6°–18°, 18°–30°,…, 54°–66°). At first, Factor Analysis defines the main categories of sky conditions (overcast, partly cloudy, clear) and then Cluster Analysis gives a more detailed grouping of the available scans. For the final classification of the sky luminance distributions, the Cluster Analysis results along with the functions of “relative gradation” and “relative scattering indicatrix”, which have been introduced by Kittler et al., are taken into account. For an assessment of the methodology proposed, the results have been compared to the ones found by using the Standard Sky Luminance Distribution method (Kittler et al.). For solar altitudes up to 42° the differences in the frequency of occurrences, for the three main categories, is less than 10%. The statistical approach used in this paper is recommended as an additional means to “standard” methods for classification of sky luminance conditions, when scan measurements are available.