Machine learning models use cloud type and cloud amount to predict rapid changes in surface solar irradiance, including short-term “ramp” events that affect grid stability. Tested at 15 sites around the world, it showed strong generalizability, matching or exceeding the original model’s predictive performance in most locations, but performance became less consistent in extreme climates.
A US research team has developed a machine learning model that uses cloud type and cloud amount as input to predict variations in surface solar radiation. The model was originally developed and trained at a single facility in Oklahoma, but researchers are now testing its performance at an additional 15 facilities around the world to assess how well it generalizes beyond the original training location.
“In 2021, Riihimaki developed a machine learning model that predicts changes in surface solar radiation from cloud type and amount based on five years of cloud radar, lidar, and surface radiation observations at the Atmospheric Radiation Measurement Program (ARM) Southern Great Plains (SGP) site in Oklahoma,” the group said. “This study complements that work by evaluating the performance and applicability of the model in an additional 15 different climates.”
For the 2021 study, the group used data recorded from 2014 to 2018 at a site in Oklahoma to train a random forest model. The model used cloud type and cloud amount as inputs to predict the mean effective transmittance (ET), the standard deviation of ET, and specifically the standard deviation of the minute-to-minute change in ET. This last metric captures rapid solar “ramp” events that are important for grid operations: sudden increases or decreases in solar irradiance caused by moving clouds.
The 2021 results showed that cloud type and cloud amount alone can explain 42% of the rapid fluctuations in sunlight caused by cloud movement. This led the authors to hypothesize that the same relationship would hold true in other climates. So they expanded their analysis to 15 additional sites, including other ARM sites in Alaska, Australia, Papua New Guinea, the Azores, Argentina, and multiple U.S. states that house the National Oceanic and Atmospheric Administration (NOAA) Surface Radiation Budget Network (SURFRAD) observatory.
However, as the prediction range expanded, modifications to the original model became necessary. In the 2021 study, cloud cover was obtained from Total Sky Imager (TSI), while in more recent studies cloud cover was obtained using RADFLUX, which estimates cloud cover from surface radiation measurements. The researchers also tested a second cloud type method used at the NOAA SURFRAD observatory, based on radiation data and ceilometer cloud base heights instead of cloud radar or lidar. This allowed us to assess whether the model remained robust beyond its original equipment configuration and could be applied more broadly.
“In terms of the coefficient of determination (r2), half of the sites (53%) have an r2 equal to or better than the original study. Of the remaining sites with low r2, almost half are within 0.1 of the original r2. This indicates that almost three-quarters (73%) of the sites have predictability as good as or better than the original study,” the academics explained. “In terms of mean squared error (MSE), the MSE for all sites is small and all within 0.0015 of the original study (0.0035). The results here support the hypothesis that this relationship is generally true for sites with other cloud climates different from the central United States.”
However, the results also showed that solar variability is less predictable for some locations and cloud types. These were primarily sites in more extreme environments than the Oklahoma reference site, including mountainous, arid, tropical, and high-latitude regions. The Alaska site showed the lowest r2 values for almost all cloud types.
The new model is described in “Predicting solar variability through cloud type and cloud cover.” solar energy. Researchers from the University of Colorado Boulder, NOAA Earth Monitoring Laboratory, and NOAA Earth System Research Institute contributed to the study.
From PV Magazine Global
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 us at editors@pv-magazine.com.
Popular content
