The dataset is based on actual, half-hourly, high-resolution visible satellite imagery observations via the broadband visible wavelength channel at a 2 arc minute resolution. These data have been processed to create more than 10 years of hourly of Global Horizontal Irradiance (GHI), Direct Normal Irradiance (DNI), and Diffuse Horizontal Irradiance (DIF) at a horizontal resolution of roughly 3 km. 3TIER processes the satellite images based on a combination of in-house research and algorithms published in peer-reviewed scientific literature. These algorithms contain parameters and coefficients that are based on empirical fits to observational data. The specific satellites and length of time varies slightly by region due to differences in the availability of data from region to region. See the chart below for more details.

To develop and validate our model, 3TIER used observations from the SurfRad, Baseline Surface Radiation Networks, National Solar Radiation Database, Bureau of Meteorology (Australia), National Institute of Water and Atmospheric Research (New Zealand), Indian Meteorology Department, NREL Annex II, Linke Turbidity Database from Ecole des Mines de Paris, MODIS Aerosol Optical Depth and Water Vapor Datasets, and snow data from the high resolution dataset developed by the National Snow and Ice Data Center.

The basic processing scheme follows the SUNY model used by NREL with a few key improvements made within the 3TIER algorithms. These improvements include: a higher spatial and temporal resolution, an in-house developed seasonal variability correction factor, an in-house developed empirical fitting of the data to ground station measurements, and the integration of instantaneous irradiance values to determine the hourly value. Each of these improvements results in a similar bias, but a significantly lower RMSE (Root Mean Squared Error) compared to the NREL dataset. 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 the NREL estimates on an hour-by-hour basis.