As part of our processing, we have compared our dataset to the existing NREL solar dataset for the continental United States as well as several observational datasets from the Surface Radiation Network (SurfRad), the Baseline Surface Radiation Networks, the Indian Meteorological Department, and the Bureau of Meteorology in Australia. In all cases our data had a low RMSE (Root Mean Squared Error) and compared favorably with our competitors. The smaller RMSE implies that our dataset is more useful as a time-series of values. In other words, our estimates of irradiance are closer to actual observed values than our competitors’ estimates on an hour-by-hour basis.

Find white papers detailing our validation by region here.

Note that in some areas with highly reflective terrain, such as salt flats and some areas with permanent snow, the satellite algorithms have difficulty distinguishing clouds from the terrain. As a result, the cloudiness estimates in these areas are higher than they should be, thus the amount of global horizontal and direct normal irradiance is underestimated and the diffuse radiation in these areas is overestimated. Known areas affected by this problem include the Bonneville Salt Flats on the Nevada-Utah border and the White Sands Missile Range in New Mexico.