Relating the uncertainty of short-term wind speed predictions to meteorological situations with methods from synoptic climatology

In order to provide an uncertainty estimate for short-term wind power predictions the accuracy of the underlying wind speed prediction is assessed quantitatively for different meteorological situations. With methods from synoptic climatology an automatic classification scheme is implemented using measurements of wind speed, wind direction and pressure at mean sea level to characterize the local weather conditions at a site. The classification procedure involves principal component analysis to efficiently reduce the data to the most relevant patterns. Cluster analysis is used to group days with similar meteorological conditions into common classes. A comparison of these clusters with weather maps shows that typical weather patterns are successfully captured by the classification scheme. The mean forecast error of the wind speed prediction of the German Weather Service is calculated for each of the clusters. It is found that different meteorological situations have indeed significant differences in the prediction error where the highest rmse can be by a factor 1.3 to 1.6 larger than the smallest rmse. Typically, high uncertainties in the forecast have to be expected in situations where low pressure systems quickly pass north of the site while stationary high pressure situations have smaller forecast errors.