A model has been developed to estimate solar radiation using input data from automated sources such as the U.S. National Weather Service’s Automated Surface Observation Systems (ASOS). Previously developed solar radiation models require input data from human meteorological observers and do not account for biases introduced when using data from automated systems. The inability of ASOS to detect cloud cover above 12,600 feet is the most influential difference between the manual and automated observing methods when using these data to estimate solar radiation. Output from this model includes global, direct and diffuse solar radiation energy-integrated over each hour of interest.

During model development, nine different solar radiation observation networks across the U.S. were utilized. Together, these networks provide good spatial coverage and representation of different regional climates. Since very few stations within these solar radiation networks also provide coincident meteorological surface observations, it was necessary to pair each station with nearby ASOS locations. In most cases, a distance of at most 30 km was required between these solar radiation-ASOS station pairs before its inclusion in model development. An additional version of this model was developed to allow for the use of supplementary hourly reports of cloud information using data from Geostationary Operational Environmental Satellites (GOES). These cloud estimates are used to replace upper-level cloud information unavailable from ASOS.

Evaluation of each model version incorporates tests on independent data along with the use of cross-validation procedures. Errors obtained are comparable to other contemporary solar radiation models. Regional-, seasonal- and condition-specific biases in estimated solar radiation are also investigated and addressed, as are any systematic biases occurring between the use of manual and ASOS data as input for the model.