Modification of spectral UV irradiance by clouds

[1] Approximately 10,000 UV irradiance spectra resulting from 2 years of continuous measurements in Germany were used as a database to analyze the effect of cloudiness on spectral UV irradiance. Values of spectral cloud modification factors (CMF) were derived by modeling a corresponding clear-sky irradiance spectrum for every UV measurement under cloudy conditions. The total set of CMF values was used to train neural networks using different sets of input data (parameter records) to describe the clouds, resulting in different, optimized, algorithms (CMF parameterizations). These different CMF parameterizations were evaluated by asking how the quality of the derived CMFs depended on the information content of different parameter records. It was shown that a visual description of cloudiness is not adequate to determine CMFs for an actual case (deviations of 50% and more), even if it was known whether or not the solar disk was obscured by clouds. Improvements for the determination of actual CMFs are possible, with deviations mostly below 15% if the parameter record comprises an actual broadband irradiance measurement. It was shown that such a CMF parameterization is able to provide a good estimation of actual CMFs, also for places with a different cloud climatology. The sensitivity of CMFs to wavelength and solar zenith angle was investigated on the basis of the derived CMF parameterizations. The relations found depend on the kind of CMF parameterization, i.e., the parameter record. In particular the separation of those cases when the solar disk is visible from those cases when the solar disk is obscured may lead to different dependencies of CMFs on solar zenith angle and wavelength.