• A machine learning model is further evaluated for solar variability predictability. • Predictability is generalizable to locations globally with different climates. • Predictability is similar even for different observations (networks, instruments). Riihimaki et al. (2021) (R21) 1 developed a machine learning model that predicts surface solar irradiance variability from cloud type and cloud cover from five years of cloud radar, lidars, and surface radiation observations at the Atmospheric Radiation Measurement Program (ARM) Southern Great Plains (SGP) site in Oklahoma. This study complements that study by evaluating R21 model’s performance and applicability in different climates at 15 additional sites. The additional sites include other ARM sites located globally and National Oceanic and Atmospheric Administration (NOAA) Surface Radiation Network (SURFRAD) sites located across the continental United States. The observed relationship of the standard deviation of the minute-to-minute change in effective transmissivity ( σ ( Δ E T ) ) varying with cloud type and cloud cover was found to be site agnostic and agreed with R21. The predictability results were found to confirm the R21 results (r 2 = 0.42) that cloud type and cloud cover provide σ ( Δ E T ) predictability: three quarters (73%) of the sites have the same predictability or better than R21, noting r 2 values of 0.37–0.54. Furthermore, all sites have small mean squared errors and all are within 0.0015 of R21 (0.0035). Locations with less predictability correspond to more extreme environments such as mountainous and high-latitude. The observed relationship and predictability of σ ( Δ E T ) were also both found to not be overfit to specific cloud type and cloud cover data products. This study represents another step forward in predicting day-ahead solar variability for energy market purposes from numerical weather prediction model output with confidence that R21 is broadly applicable to different climates.
Balmes et al. (Thu,) studied this question.