Early detection of abnormal weather conditions using a probabilistic extreme forecast index

A new method to extract information related to unusual forecast distributions from probabilistic (e.g. ensemble) prediction systems is presented. It consists of an extreme forecast index (EFI) that ranks the departure between the forecast and the model climate between -1 (forecast given 100% probability that record-breaking low values will be reached) and +1 (record-breaking high values). First the new index is derived, it is related to other measures of statistical significance, and a few properties are given. It is then demonstrated how the accumulation of ensemble forecasts every day allows the quick build-up of a model pseudo-climate that is representative enough to detect large departures from normal conditions while being representative of the latest developments in terms of resolution or physical parametrizations. The EFI is then subjected to several severe weather events that have happened in Europe over the last few years, and it is shown to be potentially useful in alerting forecasters to the risk of severe weather up to three or four days in advance. Finally, objective verification over five months in 2001-02 is presented. Although the results confirm that the model pseudo-climate is good enough to set up thresholds for severe weather evenly throughout Europe, the false-alarm rates are much larger than usually expected by forecasters or users. It is argued, however, that operating characteristics in the early medium range for severe weather have largely been ignored in the past. Although the signal is weak in the 3 to five-day range. it is undoubtedly associated with a forecast skill that might be used for setting up pre-alerts to be used either internally by meteorological services, or externally by advanced users aware of its error characteristics.