I think that the best way to go about this would be to home in on the "random" natural variation and really analyze the noise. Random noise should have very di

I think that the best way to go about this would be to home in on the "random" natural variation and really analyze the noise. Random noise should have very different spectral characteristics than real temperature data, which may have ~decadal periodicity (from the Pacific Decadal Oscillation) or similar non homogeneous frequency content. But with highly down sampled data like this a lot could be lost. That being said this is basically a Turing test with snippets of conversation selected by an antagonistic mediator. As Gelman has shown, the expected ROI is very low and any climate scientist worth her salt is better off using that effort in the grant proposal contest instead; more money and better odds.