During the boreal cool season regional climate in the United States is strongly impacted by extreme temperature regimes (ETRs), heavy rain events and snow/ice storms. In particular, extreme temperature regimes such as cold air outbreaks (CAOs) significantly impact energy consumption and human safety (via exposure). A minimum requirement for accurately predicting the future behavior of ETRs is that the climate model used to pursue such adequately replicates observed ETR behavior during recent past decades. To this end, we study the behavior and low frequency modulation of ETRs in the high resolution coupled climate model simulations associated with CMIP5. We employ CMIP's “baseline” simulations of past climate designed for model evaluation. Our focus will be on the latter portion of the 20th Century to facilitate direct comparisons with parallel results obtained from reanalysis datasets. Our primary focus is on CAOs and warm waves (WWs) occurring over three distinct geographical regions: the Midwest (MW), Northeast Megalopolis (NE), and Deep South (SE).

We first study regional ETR behavior using 3 different metrics for each cool season: 1) Number of extreme (cold or warm) Days, 2) a cumulative “Impact Factor” and 3) Peak normalized anomaly Value. This is followed by analyses of long-term trends, decadal variability and the low frequency modulation of ETR events by model-specific modes including the North Atlantic Oscillation (NAO or AO), Pacific-North American, Pacific Decadal Oscillation and the El Nino-Southern Oscillation. Linear correlation and regression analysis is used to isolate the associations with prominent simulated modes of low frequency variability. These results are then directly compared and contrasted with parallel results derived from observational reanalysis data. Part of this assessment is a validation of the model representation of the low frequency modes, themselves. An ancillary goal is to assess the impact of stratospheric resolution upon ETR behavior. This is pursued via contrasting parallel simulations from so-called “high-top” and “low-top” model configurations.