Sources of uncertainty in annual global horizontal irradiance data

The largest uncertainties were associated to operational errors such as shading (bias = −8.0%) or soiling (bias = −9.4%), which occurred frequently in low-quality monitoring networks but are rarely detected because they pass conventional QC tests. Uncertainty in estimations greatly changed from reanalysis to satellite-based products, ranging from the gross accuracy of ERA-Interim (+6.1-6.7+18.8%) to the high quality and spatial homogeneity of SARAH-1 (+1.4-5.3+5.6%). Finally, photodiodes from the Spanish agricultural network SIAR showed an uncertainty of -5.4+6.9%, which is far greater than that of secondary standards (±1.5%) and similar to SARAH-1. This is probably caused by the presence of undetectable operational errors and the use of uncorrected photodiodes. Photodiode measurements from low-quality monitoring networks such as SIAR should be used with caution, because the chances of adding extra uncertainties due to poor maintenance or inadequate calibration considerably increase.