Application of Additive Logistic Regression Models for Convective Hazards to NWP

The German Weather Service (DWD) has expressed interest in increasing its capability to assess the probability of severe convective weather events in the 12 – 36 hour forecast range. On this forecast timescale, extrapolation of trends in observations have little skill and further a strong dependency on guidance from NWP models exists. We developed a tool based on statistical models and NWP data that enables the DWD to forecast convective hazards such as hail exceeding 2 cm in diameter, severe wind gusts  25 m/s and significant tornadoes in the 12 – 36 hour forecast range. The occurrence of a convective hazard can be expressed by the product of the probability of a thunderstorm and the probability of a hazard given that a storm occurs. The probability of a thunderstorm is computed using several convective parameters derived from ERA-Interim global atmospheric reanalysis and lightning data from the European Cooperation for Lightning Detection. Severe weather reports from the European Severe Weather Database in Central Europe during the period 2008-2013 are additionally used to calculate the probability of a convective hazard given a storm occurs. Additive logistic regression is used to compute the probability functions. A calibration of the models with one year of ICON-EU and three years of ECMWF NWP data was carried out and evaluated with a number of recent convective storm cases. We will present the developed forecasting tool for the probability of severe convective weather events for the DWD as well as its application to NWP data.