Abstract
Prediction of direct solar radiation is key in sectors such as solar power and agriculture; for instance, it can enable more efficient production of energy from concentrating solar power plants. An assessment of the quality of the direct solar radiation forecast by two versions of the European Centre for Medium-Range Weather Forecasts (ECMWF) global numerical weather prediction model up to 5 days ahead is carried out here. The performance of the model is measured against observations from four solar monitoring stations over Australia, characterized by diverse geographic and climatic features, for the year 2006. As a reference, the performance of global radiation forecast is carried out as well. In terms of direct solar radiation, while the skill of the two model versions is very similar, and with relative mean absolute errors (rMAEs) ranging from 18% to 45% and correlations between 0.85 and 0.25 at around midday, their performance is substantially enhanced via a simple postprocessing bias-correction procedure. There is a marked dependency on cloudy conditions, with rMAEs 2–4 times as large for very cloudy-to-overcast conditions relative to clear-sky conditions. There is also a distinct dependency on the background climatic clear-sky conditions of the location considered. Tests made on a simulated operational setup targeting three quantiles show that direct radiation forecasts achieve potentially high scores. Overall, these analyses provide an indication of the potential practical use of direct irradiance forecast for applications such as solar power operations.
