Performance of empirical models for diffuse fraction in Uruguay

Knowledge of diffuse solar radiation is required for the estimation of global irradiation on inclined surfaces or for estimating DNI for CSP applications. Since diffuse irradiance data is comparatively scarce relative to global horizontal irradiance (GHI) data, several methods are used to estimate the diffuse component of GHI. These methods have a local component and most of them have been developed using data recorded in the northern hemisphere, where long-term reliable measurements of diffuse irradiance are available. This work considers ten models for hourly diffuse irradiation and evaluates their performance, both in their original and locally adjusted versions, against data recorded at five sites from a subtropical-temperate zone in the southern part of South America (latitudes between 30°S and 35°S). The raw data has been quality-assessed by using a set of seven sequential filters which preserve the natural spread of the data while removing unphysical data points. The local adjustment and performance evaluation are done using random-sampling cross-validation techniques on an ensemble. The best estimates result from locally adjusted multiple-predictor models, some of which can estimate hourly diffuse fraction with uncertainty of 18% of the mean.