| description abstract | The abundant spatial and temporal availability of solar energy has been fueling many researches and have been the reason for the proliferation of solar energy applications in the past decades. Many of these applications involve heavy investments and thus require highly accurate and reliable long-term average solar data for efficient deployment of solar energy technologies. Since ground stations are costly, site-specific, scarce and cannot provide long-term solar data, satellite-derived data is the next best alternative. However, satellite models are often unable to capture the complex local climatological variations of a given site. As such, short-term high precision solar ground measurements are used to train the satellite model so as to improve the accuracy of long-term solar estimates. There exist several site adaptation techniques to perform this task. However, to the knowledge of the researchers, no comparative study has been conducted to establish which site adaptation technique is the most effective. In this study, a robust methodology has been proposed to compare the effectiveness of four site adaptation techniques for monthly and yearly data sets using novel key performance indicators. Ground measurements from 12 stations in the tropical islands of Mauritius, Rodrigues, and Agalega were used to adapt satellite data obtained from HelioClim-3 database using different techniques. Three new nonlinear site adaptation techniques have been proposed: adjustment technique (Technique 2), compensation technique (Technique 3), and relationship technique (Technique 4). The first part of the study showed that 67–100% of the data sets were best approximated with sixth-order polynomials for the three nonlinear techniques. The second part revealed that Technique 1 (linear method) and Technique 2 were most appropriate for maximum and average data sets, respectively. The results were such that Technique 2 and Technique 1 provided best approximations for77.9–83.3% and 40.7–58.3% of average and maximum data sets, respectively. In the third part of the study, only Technique 2 provided remarkable improvements for all statistical metrics with respect to the original monthly data sets (113–118 data sets). The analysis reported 57.6–89.9%, 49.8–68.0%, 67.4–87.3%, 53.8–63.1%, 45.0–64.0%, 7.7–9.6% and 2.7–4.7% mean improvements for mean bias error (MBE), mean absolute bias error (MABE), mean percentage error (MPE), mean absolute percentage error (MAPE), root-mean-square error (RMSE), Nash–Sutcliffe (NSE), and coefficient of determination (COD), respectively, for Technique 2. Similar results were observed for yearly average data sets while the appreciation was shared among all four techniques for yearly maximum data sets, with Technique 1 having a slight advantage. | |
