If you really want to know stop looking at the surface. All you care about is seeing blue though, and whinging when you don't.
It all happens upstairs. Relatively minor differences in timing/interactions between shortwaves can make a big difference in sensible weather for a given locale, and there are going to be differences between operational cycles at range because each iteration gets a fresh set of initial parameters to integrate/extrapolate, and the further out in time the model does that, the more likely there will be a different outcome at particular time/place than than the previous cycle.