What was done
The authors of this intriguing paper conducted "an evaluation of the performance over western Europe of an ensemble of General Circulation Models (GCMs) used to simulate climates at the present-day, the mid-Holocene and the Last Glacial Maximum (LGM)," comparing simulations of surface air temperature and precipitation among the different models and with observed and proxy data sets.

What was learned
In the words of the authors, "for absolute values, there was a higher inter-model correlation for temperature than there was for precipitation for all months and all time slices."  In terms of differences between the present and the mid-Holocene, however, they found that the temperature correlations "are no longer robust and show a large variation ? where any climate signal was swamped by the inter-model variability."

Hoar et al. also found that "experiments performed with models from the same institution tend to cluster," and that "statistical comparisons of the models with observed and proxy data sets demonstrate a lack of consistency in model performance between months, transects and time slices."

Most amazing of all, perhaps, the three East Anglia researchers found that for the LGM, "a more realistic simulation of the ocean, as given in the sensitivity study of Kitoh et al. (2001), widened [our italics] the difference between simulated and proxy derived winter temperatures in western Europe."  And in their very next sentence, they say that "this does not necessarily imply that the models are getting worse (although there is undoubtedly a need for further development), but rather emphasizes the need for more accurate and better spatially resolved palaeoproxy data sets for the LGM."

What it means
Taken as a whole, the several findings of Hoar et al. reveal a number of serious model deficiencies.  The one just mentioned, however, is astounding: incorporation of a more realistic simulation of the ocean actually leads to temperature simulations that are worse than they are with a less realistic ocean, demonstrating that while a little knowledge can be a dangerous thing, a little more knowledge may sometimes be even more dangerous.

Just because climate models tend to become more complex with the passage of time does not insure they are getting better; they may well be stagnating or actually on a retrograde course in terms of their ability to faithfully represent the final climatic outcome of a small perturbation of the atmosphere's composition, especially when that perturbation involves the concentration of a trace gas (CO2) that stimulates all sorts of phenomena in nearly all of earth's plants, many of the physiological processes of which have significant climatic implications.  Consequently, the discrepancy between simulated and proxy temperatures discovered by Hoar et al. may well indicate a need for greater model introspection and less questioning of the palaeoproxy data sets with which the model simulations disagree.

Reference
Kitoh, A., Murakami, S. and Koide, H.  2001.  A simulation of the last glacial maximum with a coupled atmosphere-ocean GCM.  Geophysical Research Letters 28: 2221-2224.
Reviewed 7 July 2004