I would have thought so, too. But I know for certain that with enough accurate measurements around a network, you can determine the location of an unknown with pretty good accuracy. I've seen it done.
There were geostatistical methods that made certain assumptions about routing, network propagation speeds, etc. and with a large enough targeted sample using places with known geolocations, you could narrow down to a somewhere around 50 miles (this was many years ago). I imagine techniques have improved since then (as have counter measures to add noise to the data).
If you know the general area you can probably use public satellite imagery to narrow it down a bit. I suspect there some other clever information you could use to find the locations if you really wanted to.
Even with all the information I'm not sure how feasible it is for any small group to orchestrate a real disruption given how distributed data AWS centers are. Lots of failovers exist and probably a lot people outside AWS know little about.
