By Ruth Doherty and Linda O. Mearns, National Center for Atmospheric Research

We have conducted a qualitative assessment of the accuracy of climate simulations by comparing observations over North America with results from two state-of the-art coupled atmosphere and ocean general circulation models. Although differences between models and observations can arise for many reasons relating to both model and data limitations, documenting and understanding differences from observations is important in designing impact studies.

The model simulations being evaluated are from the Canadian Climate Centre (CGCM1) and the British Hadley Centre (HADCM2). Seasonal-mean averages of key surface and atmospheric variables in their baseline simulations were examined to determine how their simulated climates compare to the observational data sets of Legates and Wilmott (henceforth L&W) and reanalyses of past weather observations by a team from NOAA's National Center for Environmental Prediction (NCEP) and NCAR. While one must be cautious about drawing firm conclusions due to possible limitations in available data sets, comparison of historical control simulations against the L&W surface temperature climatology revealed that both model simulations have a warm bias over much of Canada and the north-central U.S. in autumn and winter and a cold bias in the West in all seasons (possibly due in part to differences in the average height of mountains in the model and the average altitude of observation stations).

The models also show a warm bias in the Great Plains in summer and autumn. Both model simulations also displayed a wet bias over the Rocky Mountains, and a wet bias in the Northeast and Canada in the spring and summer months when compared to the L&W precipitation climatology. These biases are generally greater in the CGCM1 simulations than in the HADCM2 simulations.

Surface pressure patterns were also compared between the two model simulations and the NCEP/NCAR reanalyses. Both models simulated the Aleutian and Icelandic storm-generating low-pressure systems in winter to be a bit too intense. In general, the HADCM2 simulations underestimated and the CGCMI simulations overestimated the strength of the major high-pressure systems off the Atlantic and Pacific coasts of North America. Geopotential height patterns (i.e., the heights of pressure layers that steer the winds) were more closely represented in the CGCM1 simulations; however, the tropospheric temperatures simulated by both models were colder than observations.

Simulations made by both models to project future temperature and precipitation patterns were also examined, focusing on changes from present conditions for three ten-year time slices centered on 2030, 2060, and 2090. Because biases in model simulations of the present may also be present in the simulations of future conditions (e.g., for differences caused by not fully representing mountain heights), at least some aspects of projected changes in climate may not be seriously affected by the biases in the baseline simulations. However, this may not always be the case and therefore, care must be taken in interpreting and applying any model results. In the simulations examined, the CGCM1 simulations displayed more intense and extensive warming than the HADCM2 simulations.

However, projected precipitation changes (mainly positive) in these future periods were somewhat similar in both models, especially of an increase in precipitation in the Western coastal US in winter. However, the CGCM1 model simulated more regions of decrease, particularly in the southeastern US. Because models are only starting to be able to represent regional patterns of the climate, it should not be surprising that differences exist between these models; that they exist points to the need for considering the results of several models as impact studies are being done.

The results of these analyses are being made available to scientists involved in the National Assessment to assist in their interpretation of the results of these model simulations of climate variability and change.

For more information, contact:

Linda O. Mearns, Environmental and Societal Impacts Group, NCAR P.O. Box 3000; Boulder, CO 80307; phone: (303) 497-8124; e-mail: lindam@ucar.edu A detailed report of these analyses is available.