[machina_ex_deus]

Look up IPCC's CMIP6 project.

I think you need an expert team of software engineers, physicists and climate experts that will either build good climate models or decipher the mess that is the current models variability.

Honestly, looking at the code of some models I don't think there's much hope in understanding them to the point where you know where the deviations are coming from, it's probably easier putting together new models that are subject to much more strict testing, unit testing, evaluations against real world data than salvaging the current mess. And just throw out the models which fail the testing.