Abstract
Shifts in the frequency of atmospheric circulation patterns throughout the year can characterize seasonality and provide a means to evaluate the ability of atmosphere‐ocean general circulation models (AOGCMs) to reproduce key dynamical structures influencing regional climate. However, the identified characteristics of these patterns can depend on the clustering method used, the physical properties of the model generating the circulation fields, and even their spatial scale. Using two statistical clustering methods, a mixture model and a hierarchical method, we show that these factors can result in distinctly different atmospheric patterns characterizing seasonal circulation over eastern North America based on NCEP and ERA‐40 geopotential height and sea level pressure (SLP) fields. Consistency is improved through constraining the number of clusters at each level to 4 or 5, and using the mixture model method. In general, this method tends to produce more consistent results across the various datasets and is more sensitive to day‐to‐day variations in pattern frequencies than the traditional hierarchical method. Applying the mixture method to PCM and CCSM3 simulations reveals some clear differences relative to reanalysis‐based patterns. In particular, the AOGCMs do not reproduce the observed duration and/or location of summer, with seasonal shifts (PCM) and summer extensions (CCSM) that are strongest at lower levels. However, taking into account the uncertainty introduced by the different factors, these AOGCMs do successfully capture many of the observed large‐scale drivers of seasonality. These results lend support to circulation‐based downscaling, but also highlight some systematic model biases, and hence the ongoing potential for improvements in model parameterization and dynamics. Copyright © 2006 Royal Meteorological Society