Benchmarking of meteorological indices for sky cloudiness classification

Abstract

Sky classification is a complex problem, due in part to such abstract conceptual definitions as clear, intermediate, and overcast, as well as other intermediate ranges. The CIE (Commission Internationale de L’Éclairage) Standard classification offers a solution to this problem, although its application requires data on the luminance distribution of the whole sky that are less commonly available. A benchmarking and classification system of ten meteorological indices is introduced in this study to classify the sky types from overcast to clear. The indices can be calculated from measurements of global, diffuse, and direct irradiance that are widely available from meteorological ground stations. The classification system uses confusion matrices, a machine-learning tool that generates a visual display of the results of supervised-learning algorithms. The CIE Standard skies classification, applied to half hourly sky-scanner measurements in Burgos (Spain), over the period June 2016 - May 2017, is used in this study as a baseline reference for a comparative review of the results from the meteorological indices and their results. They are classified by four performance ratings: Accuracy, Jaccard, Cohen, and Matthews, which feature both classification similarity and the randomness of any agreement. All meteorological indices yielded a high average degree of accuracy - close to 80% - in a detailed review of their classification. Neverthless, the results suggested that Perez’s Clearness Index based on global, diffuse and direct radiation measurements offered the most precise classification of the skies, followed closely by the Klucher Clearness Index and the Perraudeau Nebulosity Index.