Probabilistic Forecasts of Solar Insolation

Meteorological forecasts of incident solar radiation are valuable for solar resource owners and others. Most previously described forecast methods provide a single predicted value. However, a well-calibrated forecast probability distribution is more useful in that it could be used to make optimum decisions under any decision rule. We demonstrate methods of constructing and evaluating probabilistic forecasts of cloudiness (as observed from geostationary satellites) for a given location by combining climatology with weather prediction model output. We use metrics from information theory for forecast skill assessment. We consider two locations, New York City (representing a moist temperate climate) and Nevada (representing a sunny desert climate). We find that probabilistic forecasts conditioned on weather prediction model output improve on climatology at both locations. At Nevada the key limitation for accurate forecasts appears to be the weather model’s representation of cloudiness, while at New York weather predictability is also important.