Evaluation of selected models for estimating solar radiation on horizontal surfaces

Within Task IX of the International Energy's Solar Heating and Cooling Programme, 12 models which simulate solar irradiance on horizontal surfaces were evaluated with data from seven countries. This paper presents the results of this study. Simple, widely applicable models were considered which use standard meteorological observations. Two models use cloud layer amount and type, three use total cloud amount, four use fractional sunshine, while the remaining four models partition measured global irradiation into direct beam and diffuse components. Data from 15 stations were used; four Australian, four European, three Canadian, and four American. In most cases three years of data were used for each station. Cloudiness regimes from relatively cloud free (Australia and Albuquerque) to relatively cloudy (European stations) were represented. Model performance was assessed from the mean bias error (MBE), the mean absolute error (MAB) and the root mean square error (RMSE). Quantitative ranking of models from these measures showed that cloud layer models provided the best global irradiation estimates (MBE < |0.2|; RMSE < 2.1 MJ/m2/day) and the partitioning models the best direct beam and diffuse irradiation estimates (MBE < |0.6|; RMSE < 1.6 MJ/m2day). However, RMSE results for 30-day means showed that layer models performed as well as the partitioning models for direct beam and diffuse irradiation. The performance of the better models showed little variation with either season or cloud cover: thus, they seem to be generally applicable. The performance of the better global irradiation models is probably limited more by inadequate meteorological measurements and observations than by defects in the models themselves.