|Title||Verification of extreme weather events: Discrete predictands|
|Publication Type||Technical memorandum|
|Secondary Title||ECMWF Technical Memorandum|
|Authors||Magnusson, L, Richardson, D, Haiden, T|
Forecasting severe weather is one of four main goals in the ECMWF strategy 2011-2020. In this report we evaluate the forecast performance for extreme events of temperature, wind speed and precipitation, all verified against SYNOP observations. We compare the high-resolution forecast (HRES), the ensemble control forecast (CTRL)and the ensemble forecast (ENS). We focus on the frequency bias, the SEDI score and potential economical value. Comparing the evolution of the SEDI score for three different percentiles since 2002, we found that SEDI for the 98th percentile hasimproved more over the past 10 years than the 50th and 80th percentile, with the clearest result for 7-day temperature forecasts. This indicates that forecasting extremes have benefited even more from improvements in the forecast system (data assimilation and model) than the forecasting of more average weather. We acknowledge that for many places and parameters the 98th percentile cannot be classified as severe. However, sample size for higher thresholds is not sufficient to give robust results. For really extreme/severe events (with return periods of several years), we believe that one has to look into each specific case.