If you don't know the system and the range of inputs you don't know what to model. If you can't measure the outputs, you can't validate your statistical model.

If you don't know the system and the range of inputs you don't know what to model. If you can't measure the outputs, you can't validate your statistical model. The predictions for the current solar cycle are a good example - the predictions were drastically off, the actual values outside the predicted error ranges from the statistical models. It was expected to be a more active solar cycle, based on our understanding of the system, its inputs, and the state from the previous cycle. Because we didn't (don't) understand the system, our models were wrong. If your model is not right, leaving out inputs, having the wrong range of variables, etc., it doesn't matter how good your math and statistics are, your model works...until it doesn't. Then you get to do science and generate a new hypothesis, model that, and try again! What you can't do is say in 2011, "Models show California will be drier because of global warming" then in 2017, "Models show California will be wetter because of global warming" and claim the models (both?!) have predictive power. "This will not do. You never will be able to make both of them good for any thing. Take your choice, but you must be satisfied with only one. There is but such a quantity of merit between them; just enough to make one good sort of [model]" --Elizabeth Bennett (And no, there's not nearly enough "drier" area in the new model to be compatible with the older predictions.)