Applying Fuzzy Clustering Analysis to Assess Uncertainty and Model Performance in Forecasting Cool Season High-impact Weather over the U.S. East Coast

Cool-season extratropical cyclones near the U.S. East Coast often have significant impacts on safety, health, environment and economy to this most populated region. For example, the “January 2015 nor'easter” or “2015 blizzard” caused thousands of flights cancellations, travel bans enacted in five states, and two related deaths. Hence it is crucial to forecast these high-impact weather (HIW) events as accurately as possible, including in the medium-range (3-7 days). Ensemble forecasting systems are applied in operations to show an envelope of likely solutions for HIW systems. However, it is generally accepted that ensemble outputs are underused in NWS operations partly due to the lack of verification to assess model biases and efficient tools to communicate forecast uncertainties. In this study, we have applied a fuzzy clustering tool to diagnose the performance of different modeling systems in forecasting HIW using multi-model ensembles and to communicate uncertainties.

To illustrate the application of the fuzzy clustering tool in verification and separation of scenarios, the “2015 blizzard” is first explored using the multi-model ensemble including 90-members from ECMWF, Canadian Meteorological Center and NCEP ensemble datasets. Fuzzy clustering analysis based on the Principal Components of the two leading Empirical Orthogonal Function patterns of the 1- to 6-day ensemble forecasts are computed to group ensemble members into N (in our case 5) clusters. For after the fact verification, the analysis can be included as an additional ensemble member in the computation. We then examine 60 cool season HIW cases (2008–2015) using TIGGE ensemble data to statistically assess the performance of different modeling systems in capturing the scenario that includes the analysis. In actual operational application of the fuzzy clustering tool, the ensemble mean can be included as an additional member to objectively identify members that are closest to the mean. In summary, the clustering tool can efficiently separate different scenarios in a multi-model ensemble in targeted regional domains, provide forecasters an effective and objective method to compare forecast uncertainties among different operational models, and can be used as a tool to assess model performance.