**********************************EXCERPT***************************************ferdberple says:
For example, mean temperature fluctuations increase up to about 5 K at 10 days (the lifetime of planetary structures), then decrease to about 0.2 K at 30 years, and then increase again to about 5 K at glacial-interglacial scales.
==========
In other words, climate does not have a constant standard deviation. This is very important, because it means that the law of large numbers does not apply to climate. You cannot rule out natural variability increasing as time scales increase past 30 years, contrary to the law of large numbers.
Which means that 99.99% of the statistical methods applied to study climate will return incorrect results, because most statistical methods assume a constant deviation.
***********************************************EXCERPT********************************
Shaun Lovejoys comment on Rogers site is rather good:
In reference to Rogers disagreement with
deterministic models (GCMs) reproduce only weather and macroweather statistics (they do this quite well)
Shaun seems to be saying that GCMs can produce the statistics of macroweather quite well, not that they can actually predict or postdict the actual climate at all.
So there we have it use GCMs for creating a dummy climate for a fantasy game thats fine but dont apply the results to any real-world problems or youll be screwed.