Statistical properties of clear and dark duration lengths

The focus of this study is on the properties of consecutive intervals consisting of pairs of clear periods (the Sun is shining) and dark periods (the Sun is obscured by clouds). Namely, the distribution of the duration length of clear and dark periods is studied and illustrated with results obtained for the climatic regime of Timisoara (Romania, Eastern Europe). Usual simple moment-based characteristics such as means and variances cannot be unambiguously estimated due to censoring. Seasonal changes in the length distributions have been studied through the (nonparametric) Kaplan-Meier estimates. Both dark and clear duration distributions are different among months. The correlation between clear and dark periods within one dark-clear pair is of substantial interest. Therefore, a rigorous approach based on Spearman's rank correlation coefficient is used instead. The dependence of duration distribution upon covariates is studied using the Cox regression model, which fits the hazard ratio incurred by a unit change in the explanatory variables. It is shown that increasing the extraterrestrial irradiance on horizontal surface and the clearness index during dark/clear periods increases significantly the dark/clear period end risk. The Cox regression model enables more complicated analyses, such as the non-additive effects (i.e. interactions) of several covariates. The clearness index of a dark period (CID) has a much stronger effect than the extraterrestrial irradiance on horizontal surface of a dark period (EID) and the CID-EID interaction is still significant after the interactions when the months are included.