Apologies - I misunderstood the intent of what you wrote. For testing would it not be possible to use modelling instead?

E.g. in the way that you might fire drill recovery from a nuclear reactor instability without first melting part of the core.

Or how you might test a martian lunar rover before sending it off to Mars.

Even if modelling the entire system wasn’t possible with sufficient fidelity, why couldn’t you replicate the system in a geographic environment with the requisite conditions?