Solar energy applications require readily available, site-oriented, and long-term solar data. However, the frequent unavailability of diffuse irradiation, in contrast to its need, has led to the evolution of various regression models to predict it from the more commonly available data. Estimating the diffuse component from global radiation is one such technique. The present work focuses on improvement in the accuracy of the models for predicting horizontal diffuse irradiation using hourly solar radiation database from nine sites across the globe. The influence of sunshine fraction, cloud cover, and air mass on estimation of diffuse radiation is investigated. Inclusion of these along with hourly clearness index, leads to the development of a series of models for each site. Estimated values of hourly diffuse radiation are compared with measured values in terms of error statistics and indicators like, R2, mean bias deviation, root mean square deviation, skewness, and kurtosis. A new method called “the accuracy score system” is devised to assess the effect on accuracy with subsequent addition of each parameter and increase in complexity of equation. After an extensive evaluation procedure, extricate but adequate models are recommended as optimum for each of the nine sites. These models were found to be site dependent but the model types were fairly consistent for neighboring stations or locations with similar climates. Also, this study reveals a significant improvement from the conventional k-kt regression models to the presently proposed models.
Diffuse solar radiation and canopy photosynthesis in a changing environment
