Scenario-based Forecast Verification of U.S. East Coast Winter Storms in Multi-Model Ensembles

Cool-season extratropical cyclones along the U.S. East Coast often have significant impacts on the safety, health, environment and economy of this most densely populated region. It is vital to forecast these high-impact winter storm events as accurately as possible by numerical weather prediction (NWP), including in the medium-range (3-7 days). Ensemble forecasts are appealing to operational forecasters when forecasting such events because they can provide an envelope of likely solutions to serve user communities. However, information from ensemble forecasts is often underutilized in the operational process mainly due to the lack of simple and quantitative tools to extract useful information from large multi-model ensemble datasets and the corresponding ensemble verification to assess model errors and biases in forecasting significant events. This work applies a fuzzy clustering method to multi-model ensemble forecasts of East Coast winter storms, and proposes a scenario-based verification method for ensemble forecasts of these storms in the cool seasons (November to March) from 2007/08 to 2014/15.

The multi-model ensemble includes the 50-member European Centre for Medium-Range Weather Forecast (ECMWF), the 20-member National Centers for Environmental Prediction (NCEP), the 20-member Canadian Meteorological Centre (CMC) ensembles retrieved from The Observing system Research and Predictability Experiment (THORPEX) Interactive Grand Global Ensemble (TIGGE) archive. The “January 2015 blizzard” will be investigated to illustrate the application of the fuzzy clustering tool to scenario separation and forecast verification. Five clusters are determined based on the leading Empirical Orthogonal Function (EOF) patterns of 3- or 6-day ensemble forecast uncertainties. An “analysis group” is defined to represent the scenario of real development best. This case study will demonstrate that the EOF/fuzzy clustering tool is able to efficiently and objectively separate different scenarios within the multi-model ensemble. We then examine 115 winter storms using TIGGE ensemble data to statistically assess the performance and biases of the different modeling systems. Our results suggest that the ECMWF members on average have a higher chance to be included in the “analysis group” for medium range; however the model tends to show an on-shore bias for both short- and medium-range forecasts in the cyclone cases. On the other hand, the CMC members have the lowest chance among the three to be included in the “analysis group” and show a more severe under-prediction of cyclone intensity in medium-range forecasts than ECMWF model does. Within the domains we selected to verify, all three models have poor error-spread relation in forecasting winter storms based on the metrics we used. The NCEP ensemble is the most under-dispersed and the ECMWF one is the least under-dispersed in the medium range. Future applications of the clustering tool for both operational and research purposes will also be discussed. This work sheds light on improving the understanding of winter storm predictability and NWP bias, and provides a new perspective to communicate forecast uncertainty in predicting high-impact winter storms.