New methodologies and improved models in the estimation of solar irradiation stars

Abstract

The wide development of solar energy, technical agriculture and climate monitoring happening in these years requires a better knowledge of incoming solar irradiation. Although solar irradiation can be measured with pyranometers with high accuracy if correctly maintained, there is a lack of these sensors in most of the countries and regions. Besides, the high spatial and temporal variability of solar irradiation make measurements from relatively nearby stations not reliable for certain applications. As a result, solar irradiation must be frequently modelled and estimated. Many different approaches have been proposed in the last decades for generating solar irradiation out of other commonly measured meteorological variables such as temperatures, rainfall and sunshine duration. More recently, other techniques using sensors onboard satellites are able to provide solar irradiation with a higher spatial coverage. Furthermore, novel machine learning techniques can generate accurate estimations of solar irradiation. However, despite the massive development of all these techniques, still there are some drawbacks and issues in the estimation of solar irradiation directly affecting accuracy. In this context, this thesis focuses on two main blocks of studies: the temporal and the spatial methods for the estimation of solar irradiation. Beginning by traditional models, models were benchmarked based on the errors and robustness and their capacity of spatial generalization was also evaluated. From this point, more complex techniques such as support vector regression with a special optimization procedure were proposed and results were compared to parametric models. To end the block of temporal models, a wide range of satellite-based models were studied to evaluate the sources of uncertainty and error in the estimation of global and also beam irradiation under different scenarios. Regarding the spatial methods, satellite-based estimations were compared to on-ground measurements and then combined to generate more accurate maps of solar irradiation, not only for global horizontal irradiation but also for the effective irradiation on three different tilted angles. Furthermore, a very precise downscaling method for satellite-based estimations was proposed taking into account the topography and geostatistics using some on-ground records. The results of these proposed methods show a useful insight on the improvement of the estimation of solar irradiation. Some of the methods proposed in this thesis were provided as free R programming code, available as supplementary material in the articles. This code was generated with the aim of being useful for future replications and applications of the proposed methods in different regions and was one the most relevant final products of this thesis. To conclude, the contributions presented in this thesis prove the great field of improvement in the temporal and spatial estimation of the main energy input in our planet, the solar irradiation.