It's all well and good to come up with a model for what the climate should do based on basic physics. The main trouble with climate prediction is that we don't have enough data points to check whether the models are really correct.
I do physical modelling for a living, but I know not to trust a fitted-but-untested model any further than I can comfortably spit a rat.
When you can't--like with climate studies--you rely on intuition instead. That's a social science.