Machine learning regressors for solar radiation estimation from satellite data

In this paper we evaluate the performance of several Machine Learning regression techniques in a problem of global solar radiation estimation from geostationary satellite data. Different types of neural networks, Support Vector Regression and Gaussian Processes have been selected as regression techniques to be evaluated, due to their good performance in similar problems in the past. The study area is located in the surroundings of the radiometric station of Toledo, Spain. In order to train the regression techniques considered, one complete year of hourly global solar radiation data is used as the target of the experiments, and different input variables are considered: a cloud index, a clear-sky solar radiation model and several reflectivity values from Meteosat visible images. To assess the results obtained by the Machine Learning algorithms, we have selected as a reference three different physical-based methods, a model based on the Heliosat-2 method (Heliosat-2), the Copernicus Atmosphere Monitoring Service (CAMS) and the SolarGIS model (Soevaluate the performance of Machine Learning regressors when the physical models are included as input variables, in a class of post-processing of these physical approaches. The results obtained show the capacity of Machine Learning regressors to obtain reliable global solar radiation estimation by using satellite measurements.