Future North American Temperature Extremes as Simulated by a Hierarchy of Global and Regional Climate Models

In this study we use Generalized Extreme Value theory to examine future changes in extreme maximum and minimum temperature over North America in different classes of climate models. Projections for the period 2041-70 are compared to those for the recent past (1979-2000) using simulations from ensembles of: (i) fully coupled atmosphere-ocean GCMs (AOGCMs); and (ii) regional climate models (RCMs) driven by prescribed sea surface temperatures and meteorological data from the AOGCMs. For summertime extreme maximum temperatures, the RCMs and AOGCMs show similar statistically significant increases in the mean (~3 K) and range (~2-5 K) over a large region of the United States and southern Canada. However, extreme maxima in the RCM multi-model mean in the present climate are biased high relative to observations and AOGCMs. By contrast, extreme wintertime minimum temperatures increase less in the RCMs (~2 K) than the AOGCMs (~3 K), and with a narrower range; but extreme minima in the RCM multi-model mean are much closer to observed. The changes in extremes are attributed almost entirely to shifts in the mean (or location), rather than changes in the shape or scale, of the extreme value distribution. This suggests a linear response in extremes to climate change, at least within the range of warming expected by 2070.