Site-adaptation of Global Horizontal Irradiance applying Machine-learning techniques in Coastal Regions of Ecuador

The dispersion of meteorological stations in Ecuador and the Andean Region makes it difficult to evaluate, among others, the solar resource for energy generation purposes. Currently, some alternatives allow obtaining data from satellites of climatological variables with high spatio-temporal resolutions. However, this information usually presents over- or underestimates in areas where they have not been validated with surface measurements. In this paper, a machine-learning based model is proposed to validate and forecast global horizontal solar radiation from remote sensing in a coastal region of Ecuador. Two statistical learning algorithms, stepwise selection, and best subset selection, were employed. In situ solar radiation measurements, solar geometry, and climatological variables were used as predictor variables of the model. As a result, an improved global horizontal radiation dataset was achieved, decreasing the scattering errors from an initial R
²
of 0.607 to an R
²
of 0.876. The proposed methodology may be applied to obtain more reliable satellite-derived data, in areas that have not been validated and have a high cloud cover index. The developed model could be used for future regional adaptations.