You don’t rerun everything from scratch. Especially weather simulations can be checkpointed at places you have high certainty, and keep running forks after that point with different parameters. This is extremely common with for example trying to predict wind patterns during forest fires, you simulate multiple branches of possible developments in wind direction, humidity, temperature, etc. If the parameters you test don’t cover every scenario that is plausible you might sometimes engineer it into the simulation just to see the worst case scenario, for example.
And in medicine, especially computational biochemistry you modify damn near everything