Evaluating solar radiation forecast uncertainty

The presence of clouds, and their characteristics, has a strong impact on the radiative balance of the Earth and on the amount of solar radiation reaching the Earth's surface. Many applications require accurate forecasts of surface radiation on weather timescales, for example, solar energy and UV radiation forecasts. Here we investigate how operational forecasts of low and mid-level clouds affect the accuracy of solar radiation forecasts. Four years of cloud and solar radiation observations from one site – Helsinki, Finland, are analysed. Cloud observations are obtained from a ceilometer and therefore, we first develop algorithms to reliably detect cloud base, precipitation and fog. These new algorithms are widely applicable for both operational use and research, such as in-cloud icing detection for the wind energy industry and for aviation. The cloud and radiation observations are compared to forecasts from the Integrated Forecast System (IFS) run operationally and developed by the European Centre for Medium-Range Weather Forecasts (ECMWF). We develop methods to evaluate the skill of the cloud and radiation forecasts. These methods can potentially be extended to hundreds of sites globally. Over Helsinki, the measured Global Horizontal Irradiance (GHI) is strongly influenced by its northerly location and the annual variation in cloudiness. Solar radiation forecast error is therefore larger in summer than in winter, but the relative error in the solar radiation forecast is more or less constant throughout the year. The mean overall bias in the GHI forecast is positive (8 W m−2). The observed and forecast distributions in cloud cover, at the spatial scales we are considering, are strongly skewed towards clear-sky and overcast situations. Cloud cover forecasts show more skill in winter when the cloud cover is predominantly overcast; in summer there are more clear-sky and broken cloud situations. A negative bias was found in forecast GHI for correctly forecast clear-sky cases and a positive bias in correctly forecast overcast cases. Temporal averaging improved the cloud cover forecast and hence decreased the solar radiation forecast error, but made little impact on the overall bias. The positive bias seen in overcast situations occurs when the model cloud has low values of liquid water path (LWP). We attribute this bias to the model having LWP values that are too low or that the model optical properties for clouds with low LWP are incorrect.