It's not that bad. We have various records going back hundreds of thousands, in some cases hundreds of millions, of years. We have a fair understanding of various mechanisms. And the natural fluctuations you are talking about on such long frequencies are not at issue - we are only worried about the next few hundred years at most. The effects of continental drift we can safely ignore in our predictions. Likewise asteroid impacts, enormous volcanism, and the like - they have had a lot of impact, but there's no point in building models around them as likelihoods. Also, our lack of daily or weekly resolution in time and square-meter resolution in space is not that important either. If we have an idea of snow in winter, heat in summer, and the generally average values of them over general areas of interest, we have answered a lot of important questions. So the questions we are asking are - a bit of luck - exactly on the scale we have the best hope of finding answers for - a few dozen years, sections of continents and major ecological and agricultural zones. Furthermore, rough qualitative estimates are valuable -we don't have to nail Iowa's average nighttime evapo-transpiration deficit in July twenty years from now (important as that will be) to get a valuable warning of the possible effects of the CO2 boost on Iowa corn about when the current CRP contracts are running out. So when our models check out against the recent past, and then predict the first recorded hurricane off the coast of Brazil, and then tell us we face the possibility of stronger circumpolar Antarctic winds creating essentially permanent drought in much of Australia, it would be best to keep an eye on the trends in Australian rainfall, no? Maybe set up the preliminaries for handling the most immediate problems that would create ?