The other day I went to another seminar on Uncertainty Quantification. This time, it was on the "UQ Pipeline," the software suite that runs the multi-physics simulation codes hundreds of times, mapping out the solution space and determining to which input parameters the codes are most sensitive, and how sensitive they are. This information then points the way to research that knocks down the uncertainty in those most sensitive parameters, so we can then have more confidence in the codes' predictions.
Except this time, the topic was not climate modeling. It was nuclear weapons modeling, which is what the UQ methodology was originally designed for. In other words, one of the most important aspects of global climate modeling is a spin-off of the nuclear weapons design program. In fact, the climate modeling program at my lab was itself started as a spin-off from the weapons program, when a number of weapons designers, code physicists and computer scientists left the older program to start up the newer one about 20 years ago.
But climate modeling and nuclear weapons modeling differ in many respects, of which one is crucial to remember: humans have done hundreds of nuclear weapons tests, but we have only one real test of our climate models, and it is still ongoing.