Extracting likely scenarios from high resolution ensemble forecasts in real-time

Ensemble forecasts are calculated to give insight into the range of possible future outcomes and potential risks, but it is challenging for operational forecasters to deal with large ensemble data sets and to distil pertinent information from them, especially during high-impact events where forecasts and warnings must be issued and updated quickly with a high degree of accuracy and consistency.  Therefore, it is important to streamline this process by reducing the amount of data an operational forecaster must digest while still maintaining the necessary accuracy.  To do this, a novel clustering technique has been developed for use on ensemble forecasts to extract likely scenarios in real-time.  This technique uses k-medoids clustering and the spatial separation between frontal regions in ensemble members to group similar members together.  Frontal regions are often associated with heavy rain and strong winds, common high-impact events in the UK.  A single representative member is then extracted from each cluster to present to the forecaster as a potential weather scenario.  The method is illustrated with the UK Met Office operation ensemble forecasting system, MOGREPS-G.