Postprocessing multimodel ensemble data for improved short-range forecasting

A short-range multimodel ensemble forecasting system was developed and run during the summers of 2002-2004 in support of a National Oceanic and Atmospheric Administration program to improve near surface forecasts over the New England region. It was quickly realized that some form of postprocessing was needed to yield the desired improvements in the forecast variables. A simple bias correction approach was developed in which National Weather Service (NWS) surface station data are used over the preceding 12 days to bias correct today's forecast for each model variable, station location, and forecast time. This bias correction is applied separately for each ensemble member. Results indicate that the ensemble mean forecasts for 2-m temperature and dewpoint temperature are more accurate than those produced by any of the Model Output Statistics (MOS) packages available operationally. In addition, the probabilistic forecasts are quite reliable and skillful and provide added value to decision makers. A simple extension to this technique is developed to provide forecasts for any location within the ensemble forecast domain. Finally, a new method for postprocessing the ensemble precipitation data also is developed and provides reliable and skillful probability forecasts over 3-h to 24-h periods. Results from both the bias-corrected ensemble technique and the precipitation adjustment technique will be shown and discussed.