Solar Radiation Dataset
The dataset used in this study is based on pyranometric global solar irradiance ground measurements which belongs to stations of the Spanish National Radiometric Network from Agencia Espanola de Meteorologia (AEMet) sited in Madrid (40.45°N, 3.72°W, 680 m.). The period of data acquired goes from 1st of January 1979 up to 31st of December 2003 having 17376 of half daily values.
The quality of pyranometric available data and temporal acquisition of measurements are two crucial factors related to adjustment of proposed models. In this study an exhaustive quality analysis has been done on both important factors which influence high quality measurements (based on BSRN recommendations). Manufacturer of pyranometers used is the Dutch firm Kipp&Zonen and belong to CM11 series.
The mean daily solar radiation presents a seasonal variation due principally to the variation of the position of the earth to the sun and a stochastic aspect introduced by atmospheric components (aerosols, H2O, O3,...).
Half daily ground solar radiation shows a seasonal trend with fluctuations from day to day due to cloudiness, changing air mass and aerosols. The seasonal trend can be separated in several different ways: (1) “clearness sky index” based on extracting the influence of latitude, dividing between measured ground solar radiation and extraterrestrial solar radiation values, (2) subtraction of the annual harmonic (first harmonic of Fourier analysis), (3) “lost component” which is based on the subtraction of each value of the extraterrestrial radiation the corresponding value of ground solar radiation. In general, approach (1) leads to a procedure much simpler from the point of view of the user, but it is not suitable because statistical forecasting methods needs predictors with stationary behaviour and Gaussian distribution.