"Computational Fluid Dynamics (CFD) works great if it has been fully tweeked against flight data or wind tunnel data for a specific set of Reynolds and Mach numbers. It has been my experience that CFD usually doesn't compare well for conditions that are significantly different from what it was originally coded for (you should never use CFD without a validation case). Well executed experimental data always trumps CFD data. Good flight data always trumps experimental data. The G-rated version of what guys in the flight test business call CFD: "Completely Fudged Data". Unfortunately flight data is typically a hundred times more expensive than experimental data and experimental data is a hundred times more expensive than CFD. The wind tunnels were shutdown mainly because CFD was vastly cheaper even though CFD was incorrect most of the time. This leads to the old good news - bad news joke. The bad news: All we have to eat is manure. The good news: We have LOTS of it."

This is precisely the point I'd tried unsuccessfully to make when discussion climate "models". Sure there are many deterministic, computable things we model in eng & science but there are many others that modeling is a faux art.