Thanks for the feedback. What I obviously forgot is that a computer model is just as prone to error or misinterpretation as any other attempt at physical modeling.
I should have qualified my comment on models. Some are good approximations and many are not, garbage in garbage out so to speak.
However, models are extremely useful in predicting events. I use models every day. Often there are no real solutions to these models (just gives an approximation). However, they still are "good enough" to get the job done so to speak.
Also, you may have heard it argued all theories are really models based on empirical data/evidence, direct and indirect observation, experimentation, etc.
Modeling the atmosphere is hard. It is am extremely chaotic environment. However, they are getting better at it.
Indeed mistakes have been made. I remember the "nuclear winter" models and arguments. Unfortunately, the people who created those models failed to take into account the exchange between the oceans and the atmosphere, which when done so, showed that their results did not conform to what would have happened in the “real world” should such an event have really happened.
To return to my hadronic shower modeling: one model we had developed was looking quite promising. The energy deposition was dead-on; the spectra were indistinguishable from real data. There was just this one little teeny problem that the jet distributions were somewhat off. Well, who's going to notice that? Just a detail, right? The model wasn't supposed to be perfect, after all.
A little digging showed that in each shower, the ratio of hadronic to electromagnetic energy was way off from experimental reality. We realized that if the model couldn't get that right, then it couldn't be right. The good-looking distributions were spurious. We had to start over.
The climatological models get so many things "right" (i.e. in line with people's expectations) that people accept their results as correct, and believe that if it gets a few things wrong, a few minor tweaks will clear all that up, and the major results won't change. But it's not that simple or easy.
You see, in the case of our hadronic shower model, it took me a long time to admit that there was a problem. (When I said we "realized" the model couldn't be right, what I meant was we finally admitted it.) My strong desire was to ignore my instincts, to gloss it over, to handwave it away. We ultimately threw the model out, but it was painful to flush away that much work. And we had an enormous advantage over the climatologists: we weren't trying to use the model to predict anything, but were just trying to reproduce something well-known. We had more (and firmer) reality checks, and a much simpler process to model, but the correct path was still difficult to find, both technically and emotionally.
The warning signs are there that the models are wrong. The model-builders have a strong emotional incentive to ignore and downplay the signs. The audience doesn't know that the signs are potentially important.