I came across this video (possibly from a tweet by Chris Colose) that describes climate models. I thought it was quite good, but I’m not a climate modeller so maybe others would disagree. The one thing I thought it did quite well was that it was careful to refer to what climate models do (wrt the future at least) as projections. You’ll often hear people claim that climate models have failed because observations have not matched model results over the last decade or so. Some immediate problems with this claim are that climate models are not actually optimised for decadal predictions/projections and what you’re often shown are ensemble averages, which tend to smooth out short-term variability.
A more fundamental point, though, is that when climate models are used to consider the future, they’re actually making projections, not predictions. One has to make assumptions about our future emissions (greenhouse gases, aerosols, …), solar variability, and other – as yet – unknown factors. Therefore, climate models are really telling us something about what might happen in the future if we follow various different pathways. Formally, therefore, if one wants to claim that a climate model has failed, one has to also know whether or not the pathway we actually followed was the same (or similar to) what was assumed in the model. Having said that, as Tom Curtis points out in this comment there is some indication that the model trends are different to the observed trends, suggesting that climate models may be too sensitive (I haven’t had much chance to look at this myself, so would quite like to know more about the significance of this). However, simply pointing out that the observations don’t match the model ensemble for the last decade or so, is not really sufficient to indicate that climate models have failed.
The other thing the video does quite well is talk about how climate models can be used to consider past changes. This can be used to try and understand what caused past climate changes, and can – presumably – also be used to see if these models can actually explain past changes. It also allows one to test for attribution. In the case of 20th century warming, models without anthropogenic forcings cannot explain the surface warming (both globally and regionally) after 1950. Models with anthropogenic forcings can. That’s one reason why the IPCC is very confident that most of the warming since 1950 has been anthropogenic. Anyway, I’ve said enough. I recommend watching the video yourself.