Background
The authors write that today "there are more than two dozen different climate models which are used to make climate simulations and future climate projections." But although it has been said that "there is strength in numbers," most rational people would still like to know how well this specific set of models does at simulating what has already occurred in the way of historical climate change, before they would be ready to believe what the models predict about earth's future climate.

What was done
The two U.S. researchers began their study of 28 pre-industrial control runs, as well as 70 20th-century forced runs, derived from 23 different climate models, by analyzing how well the models did in hind-casting "networks for the 500 hPa, surface air temperature (SAT), sea level pressure (SLP), and precipitation for each run."

What was learned
In the words of Steinhaeuser and Tsonis, they report that (1) "the models are in significant disagreement when it comes to their SLP, SAT, and precipitation community structure," that (2) "none of the models comes close to the community structure of the actual observations," that (3) "not only do the models not agree well with each other, they do not agree with reality," that (4) "the models are not capable to simulate the spatial structure of the temperature, sea level pressure, and precipitation field in a reliable and consistent way," and that (5) "no model or models emerge as superior."

What it means
In light of their several sad findings, the team of two suggests that "maybe the time has come to correct this modeling Babel and to seek a consensus climate model by developing methods which will combine ingredients from several models or a supermodel made up of a network of different models." But with all of the models they tested proving to be incapable of replicating any of the tested aspects of past reality, even this approach would not appear to have any promise of success.