Extreme weather events and their probabilistic prediction by the NCEP ensemble forecast system

In this presentation first general issues related to extreme precipitation events, and weather events in general, will be discussed. It will be argued that extreme events are generated by the same physical processes as other events, and forecasting them requires the same tools as those used for general weather prediction purposes. The extreme events, by definition, lie on the edge of the distribution of all weather events in the phase space. Specific combinations of different factors contribute to the occurence of extreme events. Apparently,strong nonlinearities play an important role in limiting the scope of physically plausible events, thus determining also the extremes of the distribution of weather events. It will be argued that the same nonlinear effects will also prevent even perfect models from making forecasts that are not biased toward the climatological mean. Looking at forecasts made for the most extreme observed cases, any errors in initial conditions are likely to result in forecasts with less extreme conditions. It will be pointed out that despite the bias problem and other possible difficulties, and contrary to expectations, most verification measures indicate that extreme events are better predicted than near normal events of the same spatial and time scales. The second part of the presentation will provide examples of probabilistic quantitative precipitation forecasts (PQPF), generated by the NCEP ensemble, for extreme events. Verification statistics for different measures of probabilistic forecast performance will also be presented, along with a procedure to calibrate the ensemble-based PQPF forecasts.