To comply with an increasing demand for sustainable energy sources, a solar heating unit is being developed at the Technical University of Denmark. To make optimal use — environmentally and economically —, this heating unit is equipped with an intelligent control system using forecasts of the heat consumption of the house and the amount of available solar energy. In order to make the most of this solar heating unit, accurate forecasts of the available solar radiation are esstential. However, because of its sensitivity to local meteorological conditions, the solar radiation received at the surface of the Earth can be highly fluctuating and challenging to forecast accurately.

To comply with the accuracy requirements to forecasts of both global, direct, and diffuse radiation, the uncertainty of these forecasts is of interest. Forecast uncertainties can become accessible by running an ensemble of forecasts, and to this end, these three meteorological quantities have since August 2011 been output parameters from the high-resolution ensemble prediction system at the Danish Meteorological Institue.

The appropriateness of complementing forecast values with uncertainty estimates derived from the ensemble forecasts has been assessed by investigating the degree to which the ensemble members and the truth — here materialised by the verifying observation — are statistically indistinguishable. A degree of under-dispersion of the ensemble members is evident concerning global radiation, and the ensemble forecasts will therefore tend to express too little uncertainty in the forecast values. Under-dispersiveness is a wellknown problem in ensemble prediction. Uncertainties on obsrvations may cause some of the under-dispersiveness of the ensemble forecasts of global radiation.