An improved algorithm for estimating incident daily solar radiation from measurements of temperature, humidity, and precipitation

We present a reformulation of the Bristow-Campbell model for daily solar radiation, developed using daily observations of radiation, temperature, humidity, and precipitation, from 40 stations in contrasting climates. By expanding the original model to include a spatially and temporally variable estimate of clear-sky transmittance, and applying a small number of other minor modifications, the new model produces better results than the original over a wider range of climates. Our method does not require reparameterization on a site-by-site basis, a distinct advantage over the original approach. We do require observations of dewpoint temperature, which the original model does not, but we suggest a method that could eliminate this dependency. Mean absolute error (MAE) for predictions of clear-sky transmittance was improved by 28% compared to the original model formulation. Aerosols and snowcover probably contribute to variation in clear-sky transmittance that remains unexplained by our method. MAE and bias for prediction of daily incident radiation were about 2.4 MJ m^-2 day^-1 and +0.5 MJ m^-2 day^-1, respectively. As a percent of the average observed values of incident radiation, MAE and bias are about 15% and +4%, respectively. The lowest errors and smallest biases (percent basis) occurred during the summer. The highest prediction biases were associated with stations having a strong seasonal concentration of precipitation, with underpredictions at summer-precipitation stations, and overpredictions at winter-precipitation stations. Further study is required to characterize the behavior of this method for tropical climates.