Applying the kriging method to predicting irradiance variability at a potential PV power plant

One-second irradiance data from forty-five sensors spaced over a one-mile square section of land were analyzed to characterize the short-term (1-s to 1-min) variability of the solar resource in Northern Arizona. The geostatistical interpolation model known as kriging was applied to our data set to better understand the method's strengths and weaknesses in accurately predicting the variations in the irradiance over this relatively small section of land. Of particular interest was to investigate the ability of the kriging method to show the variation in solar irradiance over the section of land as compared to that measured by the sensors. When using data from all the sensors as input to the prediction method, kriging performed very well compared to the sensors. However, because it is unlikely to have a large number of sensors to characterize the variability at a prospective solar site, it was also of interest to investigate how many sensors are required as input to the kriging technique in order to generate a reliable prediction. Solar data from four characteristic periods (related to the four seasons) were analyzed, and different sensor configurations, consisting of subsets of the actual sensor array, were employed using the method to demonstrate the number of sensors required to correctly characterize the short-term irradiance variability at the site. Using four measurement stations as input to the kriging method was shown to reasonably represent the variability in the 1-s to 1-min timescales.