Verification of the SUNY direct normal irradiance model with ground measurements

摘要

The accurate assessment of the Direct Normal Irradiance (DNI) component of the solar resource is notoriously difficult due to the higher cost of DNI instrumentation and the much higher requirements on proper maintenance of such instruments. Nonetheless, DNI is the only component relevant to all concentrated power technologies. The evaluation of location-specific DNI variability is of current interest to the renewable energy sector due to the steep ramp rates associated with cloud effects. The present work shows the ability of the SUNY satellite-to-irradiance model in capturing the magnitude and variability of DNI when compared with high quality ground measurements. Long-term ground DNI data is compared against concurrent SUNY DNI values after an extensive data quality control. Redundant measurements at one location enables quantification of the error in the semi-automatic irradiance data quality control. This study demonstrates the general accuracy of SUNY data and quantifies the error in assessing the magnitude and variability of DNI for several different microclimates in California. A location dependent mean bias error of - 6.39% to 14.21% with a correlation coefficient (ρ) ranging from ρ = 0.90 to ρ = 0.95 is registered for long-term averages of the SUNY DNI data. Our findings indicate that the SUNY data is a valuable tool to access the DNI variability on a 30-min temporal resolution although an overestimation of small fluctuations at some locations exists. A method to correct for this overestimation of low magnitude variability is described. Overall, the SUNY data is especially important with regards to a lack of high quality ground measured DNI data and the costs associated with maintaining networks of solar instruments. However, a comparison of variability events and Variability index (VI) on 1-min resolution to 30-min resolution data reveals that the 30-min averages obscure large shares of frequency and amplitude of DNI variability. Values of Vl_(30min) are more than six times smaller than Vl_(min) can be found within the same data sets.