An analysis of short range ensemble forecasts for cold season extratropical cyclones. Part II: upper Midwest cases from the winter of 2008-09

The winter of 2008-09 produced a number of significant storms across the upper Midwest. The storms resulted in a variety of hazardous weather conditions, including heavy snow, sleet, freezing rain, and extremely cold temperatures. Forecasts from the Short Range Ensemble Forecast (SREF) system have become a key component in the forecast of weather phenomena in the first few days of the forecast, particularly in the watch and warning phases for significant events. The SREF can be used to help augment deterministic model output by providing information on forecast uncertainty and the probability of the occurrence of various conditions. Due to the overwhelming amount of information available from the SREF system, forecasters have become reliant on using mean and probability fields in order to quickly synthesize the output. An analysis of SREF mean and probability fields for several significant events during the winter of 2008-09 indicated a consistent inability to highlight the most likely location of adverse weather until events were within 24 hours of occurrence, providing very little information to forecasters beyond what was available from the deterministic guidance. A review of the output from individual members during these events provided important clues as to why the mean and probability fields performed poorly. In several cases, numerous SREF members produced forecasts which a human forecaster could quickly dismiss as being unreasonable and outside the scope of possibility based on past experience with similar events. However, such solutions were used in deriving the forecast means and probabilities, and given equal weight in the calculations. In addition, the clustering of forecasts for each model type used with the SREF and the unequal weighting of model types further compromised the mean and probability fields.

Three events will be examined to demonstrate how the SREF provided little additional value to forecasters, and could have misled them into improperly diagnosing the likelihood, significance, timing, and location of hazardous weather. Forecasts from the North American Mesoscale (NAM) and Global Forecast System (GFS) models during these events tended to latch onto the correct solution 12 hours prior to the SREF, and the corresponding solution was often significantly different from the SREF mean and within very low forecast probability areas. An evaluation of output from additional deterministic solutions for these events, including the European Center for Medium Range Weather Forecasting (ECMWF) and Global Environmental Multiscale (GEM) models, would have provided forecasters with more appropriate estimations of forecast uncertainty and probability of occurrence at longer lead times. A “poor man's ensemble” such as this highlights the likely importance of including multiple models within an ensemble framework, supporting one key characteristic of the SREF system and underscoring the potential benefit of using it within the forecast process. However, without an evaluation of output from the individual members of the system, forecasters may not have a complete picture of its derived fields and could be misled.