A displacement-based error measure applied in a regional ensemble forecasting system

Abstract

Errors in regional forecasts often take the form of phase errors, where a forecasted weather system is displaced in space or time. For such errors, a direct measure of the displacement is likely to be more valuable than traditional measures. A novel forecast quality measure is proposed that is based on a comparison of observed and forecast satellite imagery from the Meteosat-7 geostationary satellite. The measure combines the magnitude of a displacement vector calculated with a pyramid matching algorithm and the local squared difference of observed and morphed forecast brightness temperature fields. Following the description of the method and its application for a simplified case, the measure is applied to regional ensemble forecasts for an episode of prefrontal summertime convection in Bavaria. It is shown that this new method provides a plausible measure of forecast error, which is consistent with a subjective ranking of ensemble members for a sample forecast. The measure is then applied to hourly images over a 36-h forecast period and compared with the bias and equitable threat score. The two conventional measures fail to provide any systematic distinction between different ensemble members, while the new measure identifies ensemble members of differing skill levels with a strong degree of temporal consistency. Using the displacement-based error measure, individual ensemble members are found to compare better with observations than either a short-term deterministic forecast or the ensemble mean throughout the convective period.