Tim Lambert has another nice post Global Warming Sceptic Bingo, with a pleasing number of the refutations from RC or here... its working, folks. I've been a bit busy building recently, but I notice that Tim doesn't have a refutation for We can’t predict the weather a week in advance. How can we do it 100 years in advance?. This may well be a really stupid argument, but lots of people make it, so its worth debunking:
To get you in the right frame of mind, consider an analogy: sea level at the beach. You can't predict the height of the next wave, or even when a wave will strike ten minutes from now. Yet the tide can be easily predicted years (centuries, probably millenia) ahead.
A better analogy is with throwing a (fair) die: you can't predict whether the next throw will be 1, 2, 3, 4, 5 or 6; yet you can be fairly sure that if you throw it 1000 times you'll get very bored. Sorry: If you throw it 1000 times, the average throw will be close to 3.5. And to pursue this a bit more, if you make the die a bit unfair (by, say, making the 6 twice as likely to come up) then you still can't predict what the next throw will be. But you can be sure that the long-term average will go up to... thinks... about 4 1/3.
Weather is, we are fairly sure, intrinsically unpredictable past a certain limit that isn't precisely known, but is about 1 month. In practice, current NWP (limited by computer size, model completeness and starting obs) can be useful out to about 2 weeks at a stretch. The weather is thus chaotic. As far as can be seen, climate isn't. Obviously, some features of climate can be predicted far further out: we know that winter will be warmer than summer, for a trivial example. But what about the trends in global mean temperature over 100 years? In this situation we have an imposed forcing (increasing GHG concentrations) on top of a chaotic system (weather) which appears to average to a non-chaotic system with intrinsic predicatability (climate).
If you trust the models, you can test this: if you start an atmosphere-only model with differences in the starting state differing only at the numerical-accuracy level (which is far below the obs error the NWP starts from) they diverge within a month. But if you then take the long-term mean of the runs, they are the same. If you start coupled models off with small differences and GHG forcing, they again differ on the weather, and the year-to-year variations in the global mean; but the long term trends match.