Estimation methods for global solar radiation: Case study evaluation of five different approaches in central Spain

Three different tiers of estimations were obtained being SARAH and OK the best performing methods overall. The SARAH dataset (MAE=1.10±0.13 MJ/m2, MBE=0.22±0.36 MJ/m2) generated estimates with the lowest spread, but led to a slight overestimation in low-altitude flat areas. The OK (MAE=1.10±0.25 MJ/m2, MBE=0.00±0.31 MJ/m2) outperformed SARAH in these flat areas (high density of stations), but at the expense of a higher variability. Alternatively, SARAH surpassed Ordinary Kriging (OK) when the distance to the closest station exceeded 20-30 km. The ERA-Interim reanalysis and the XGBoost were in the second tier of estimations, and the parametric model yielded the worst results overall. ERA-Interim exhibited a systematic overestimation. The locally trained Antonanzas and XGBoost struggled to model the atmospheric transmissivity, showing large positive errors in spring months and a small underestimation of clear-sky days. Finally, a summary with the strengths and weaknesses of the five methods provides a deeper understanding for the selection of the adequate estimation approach.