This article compared the performance of three forecasts combining schemes (mean, median and trimmed mean) on solar radiation data. Solar radiation data covering January 2003 to December 2016 were collected from the National Cereals Research Institute, Baddegi, Bida, Niger State, Nigeria. A total of five forecasting models (Multiplicative and additive versions of decomposition model, multiplicative and additive versions of Holt-Winter’s model, and SARIMA model) were fitted. A comparison of the three forecasting schemes on out-of-sampling forecast performance on the basis of mean square error showed that the mean scheme performed best with mean square error of 0.85227, followed by trimmed mean with mean square error of 0.88711 and then the median with mean squared error of 1.0849564. The article concluded supremacy of the mean as forecasts combining scheme over the median and trimmed mean for the solar radiation data as far as the applied forecasting methods are concerned.
Keywords: solar radiation; forecasting; decomposition; holt-winter’s; SARIMA
