Cluster Analysis of Multimodel Ensemble Data from SAMEX

Short-range ensemble forecasts from the Storm and Mesoscale Ensemble Experiment (SAMEX) are examined to explore the importance of model diversity in short-range ensemble forecasting systems. Two basic techniques from multivariate data analysis are used: cluster analysis and principal component analysis. This 25-member ensemble is constructed of 36-h forecasts from four different numerical weather prediction models, including the Eta Model, the Regional Spectral Model (RSM), the Advanced Regional Prediction System (ARPS), and the Pennsylvania State University–National Center for Atmospheric Research fifth-generation Mesoscale Model (MM5). The Eta Model and RSM forecasts are initialized using the breeding of growing modes approach, the ARPS model forecasts are initialized using a scaled lagged average forecasting approach, and the MM5 forecasts are initialized using a random coherent structures approach. The MM5 forecasts also include different model physical parameterization schemes, allowing us to examine the role of intramodel physics differences in the ensemble forecasting process. Cluster analyses of the 3-h accumulated precipitation, mean sea level pressure, convective available potential energy, 500-hPa geopotential height, and 250-hPa wind speed forecasts started at 0000 UTC 29 May 1998 indicate that the forecasts cluster largely by model, with few intermodel clusters found. This clustering occurs within the first few hours of the forecast and persists throughout the entire forecast period, even though the perturbed initial conditions from some of the models are very similar. This result further highlights the important role played by model physics in determining the resulting forecasts and the need for model diversity in short-range ensemble forecasting systems.