A machine learning model uses cloud type and cloud cover to predict rapid changes in surface solar radiation, including short-term events that affect the stability of the power grid. When tested in 15 locations worldwide, it showed strong generalizability, with most locations matching or exceeding the predictive performance of the original model, although extreme climates performed less consistently.
An American research team has developed a machine learning model that predicts the variability in solar radiation at the surface using cloud type and cloud cover as input. The model was originally developed and trained at one location in Oklahoma, and the researchers have now tested its performance at fifteen additional locations worldwide to evaluate how well it generalizes beyond the original training location.
“Riihimaki does developed in 2021 a machine learning model that predicts surface solar radiation variability based on cloud type and cloud cover based on five years of observations of cloud radar, lidars, and surface radiation at the Atmospheric Radiation Measurement Program (ARM) Southern Great Plains (SGP) site in Oklahoma. This study complements that study by evaluating the model’s performance and applicability in different climates at 15 additional locations.”
In the 2021 study, the group used data collected at the Oklahoma site between 2014 and 2018 to train a random forest model. The model used cloud type and cloud cover as input to predict the mean effective transmission rate (ET), the standard deviation of ET, and in particular the standard deviation of minute-to-minute changes in ET. The latter metric captures rapid solar “catastrophe” events – sudden increases or decreases in solar radiation caused by moving clouds – that are important to the operation of the electric grid.
The 2021 results showed that cloud type and cloud cover alone could explain 42% of the rapid fluctuations in sunlight caused by moving clouds. This led the authors to hypothesize that the same relationship would hold in other climates. So they expanded the analysis to 15 additional sites, including other ARM sites in Alaska, Australia, Papua New Guinea, the Azores, Argentina, Texas, Colorado, and California, as well as National Oceanic and Atmospheric Administration (NOAA) Surface Radiation Budget Network (SURFRAD) stations in Illinois, Nevada, Montana, Mississippi, Pennsylvania, South Dakota, and Colorado.
However, as the scope of the prediction was expanded, adjustments to the original model were necessary. In the 2021 study, cloud cover was derived from a Total Sky Imager (TSI), while in the more recent work it was obtained using RADFLUX, which estimates cloud cover from surface radiation measurements. The researchers also tested a second cloud-like method used at NOAA SURFRAD stations, based on radiation data and ceilometer cloud base heights instead of cloud radar and lidar. This allowed them to assess whether the model remained robust beyond the original instrument configuration and could be applied more broadly.
“In terms of coefficient of determination (r2), half of the sites (53%) have the same r2 or better than in the original study. Of the remaining sites with a smaller r2, almost half are within 0.1 of the original’s r2. This indicates that almost three-quarters (73%) of the sites have the same predictability or better than the original,” the academics explained. “In terms of the mean square error (MSE), all locations have small MSEs, and all are within 0.0015 of the original study (0.0035). The results here confirm the hypothesis that the relationship is largely applicable to locations with other cloud climatologies that differ from the central United States.”
However, the results also showed that some locations and cloud types showed lower predictability of solar variability. These were primarily locations in more extreme environments than the Oklahoma reference location, including mountainous, arid, tropical, and high latitudes. The Alaska location showed the lowest r2 values for almost all cloud types.
The new model was described in “Prediction of solar variability by cloud type and cloud cover”, published in Solar energy. Researchers from the University of Colorado Boulder, NOAA Global Monitoring Laboratory and NOAA Global Systems Laboratory contributed to the study.
This content is copyrighted and may not be reused. If you would like to collaborate with us and reuse some of our content, please contact: editors@pv-magazine.com.
