The Generalized Additive Model for the Assessment of the Direct, Diffuse, and Global Solar Irradiances Using SEVIRI Images, With Application to the UAE

Generalized additive models (GAMs) can model the nonlinear relationship between a response variable and a set of explanatory variables through smooth functions. GAM is used to assess the direct, diffuse, and global solar components in the United Arab Emirates (UAE), a country which has a large potential for solar energy production. Six thermal channels of the spinning enhanced visible and infrared imager (SEVIRI) instrument onboard Meteosat second generation (MSG) are used as explanatory variables along with the solar zenith angle, solar time, day number, and eccentricity correction. The proposed model is fitted using reference data from three ground measurement stations for the full year of 2010 and tested on two other stations for the full year of 2009. The performance of the GAM model is compared to the performance of the ensemble of artificial neural networks (ANNs) approach. Results indicate that GAM leads to improved estimates for the testing sample when compared to the bagging ensemble. GAM has the advantage over ANN-based models that we can explicitly define the relationships between the response variable and each explanatory variable through smooth functions. Attempts are made to provide physical explanations of the relations between irradiance variables and explanatory variables. Models in which the observations are separated as cloud-free and cloudy and treated separately are evaluated along with the combined dataset. Results indicate that no improvement is obtained compared to a single model fitted with all observations. The performance of the GAM is also compared to the McClear model, a physical-based model providing estimates of irradiance in clear sky conditions.