What should NASA do next, they have made 4 consecutive predictions for solar cycle 24 start up, and in all cases blown the forecast big time. Kick the can, or admit their models have totally failed them, and the science contained therein. Want to bet against the can kicking?
It is with great trepidation that I then say the only way to get through is to say computer models are not science. Knowing full well that engineers who validate their models, use real data, and doing things like rolling back time and predict the past, will continue to do a reasonable job at attempting to predict future behavior. Or in fact behavior which cannot be easily simulated.
As an engineer for most of my adult life, I understand, BUT the public has clearly demonstrated they do not get it. Simplify. The last election proved it is very easy to make idiot voters out of ignorant people. It is high time we realize that for most of the American public, ignorant applies.
“It is with great trepidation that I then say the only way to get through is to say computer models are not science.”
In a sense you are correct. The AGW syllogism, if their models actually had predictive ability, would be:
1. Human CO2 causes non-linear warming.
2. Our computer models correctly predict the future using a non-linear CO2 model (not established)
Therefore Human CO2 causes linear warming.
But all number two proves is that the model correctly predicts. Not that CO2 is causative. To establish causation, any statistics 101 student can tell you that you have to manipulate the explanatory variable and measure the result of the manipulation. Then, if you manipulate the variable value and given that manipulation, the model correctly predicts the observed result, you can make valid statistical statements about causation (normally this occurs in the context of regression, not more complex models but the principle is the same).
So even if their models correctly predict using observed CO2 (not manipulated CO2), they have not performed any useful science regarding causation, only correlation via the model.
So I have three big problems with the AGW approach.
1. In modeling from observed data, you don’t really pretend that the inputs are causative. Only that they predict. If you are careful, you can put very reliable bounds around the accuracy of the predictions. But the AGW guys pretend they are establishing causation out of a process that by its nature CANNOT provide valid evidence of causation.
2. The real issue in modeling is whether model x is more predictive than model y. There are good information theoretic techniques that allow you to compare two models, one more complex and one less complex. The more complex the model, the less likely it is to be the correct model.
The AGW guys never compare their predictions to, say, using a ruler and drawing a trend line from the recent past (that technique did pretty well as a predictor—as well or better than the AGW models until 1998 and both have done equally poorly since then).
Nor do they compare their models’ predictions to, say, solar models, which are very simple—one input plus a smoothing and lagging parameter.
The AGW models are quite complex and require tuning a lot of parameters. So if the solar model (or the ruler model) produces results as good as the AGW models, you would always use the solar model in preference to the AGW model and that result is grounded solidly in information theory.
3. The nature of positive feedback is exponential growth. It looks to me like the AGW guys are forecasting short term in the linear portion of the exponential growth curve. If their models correctly predicted the short term, that is no evidence at all that the model should be regarded as useful in the exponential portion of the predictions (long term). Yet it is the long term predictions that get all the news.