Sheep Herd Accuracy?

The next UN Climate Conference will take place in Copenhagen in December under Swedish chairmanship of EU. UN Climate chief Yvo de Boer hopes the conference will in particular reach agreements to limit the growth of emissions in developing countries, required to be necessary by predictions of catastrophical global warming.

The most advanced computational methods for long-time predictions of the global climate, the AOGCM coupled Atmospheric Ocean General Circulation Models, use computational grids of size 

• surface: 2.5 x 3.75 degrees, height: 19 levels HadAM3
  
• surface: 1.4 x 1.4 degrees, height: 26-40 levels CCSM3

to simulate turbulent convective radiative reactive phenomena. This means a surface grid of 30 x 30  to 60 x 60 for a quarter hemisphere including complex geometry of land and ocean.

From my experience from computing turbulent flow like the flow around a car or wing, the resolution of HadAM3 is definitely not impressive, and the grid of CCSM3 also seems too coarse. For such grids turbulence modeling is required, which however is an open problem.

What about the accuracy of these models? According to IPCC TAR underlying the Kyoto Protocol

• The model mean exhibits good agreement with observations.
• The individual models often exhibit worse agreement with observations.
  

The logic here is to take the mean of many incorrect models, to get a correct model! As a scientific principle this is wonderful, because you can generate so much knowledge out of ignorance: 

• The mean value of many incorrect values, is the true value!
  

But of course you say: What about the sheep herd effect? Just because all sheep rush in certain direction, does it mean it is the right direction? Not even in politics, and definitely not in science.