Given the large number of unexpectedly bad natural disasters the past couple decades I suspect many risk estimates were low balled in the past. '100 year' floods seem to happen every 20 years and '500 year' ones are commonplace. They better not be designing midwestern nuke plants for only T3 tornadoes or we'll find how well a T5 can scatter radiation. Siting plants near large natural water supplies can be very beneficial- where'd they be without easy access to sea water now? But you need a generous safety margin of elevation. Run computer models of worst theoretically conceivable river floods, tsunamis, etc. Seattle area better model lava flows or lahars from Mt. Rainer.
Consider the probability that a natural disaster strong enough to break one thing could well break several things. I'd also like to see robots capable to investigating high radiation areas and debris strewn areas. That need has been obvious since Chernobyl. It's shocking that Japan apparently doesn't have such available. Our military has made much progress in robots in Iraq and Afghanistan. Some of that tech and experience needs to be spun off and pre-positioned for such use. It should be technically possible to remotely pilot helicopters for these water drops as well as those pump trucks. There may still be some radiation hazard involved in fueling and servicing them, but performance could be better without having to protect a pilot. A robot that could deliver and position a fire hose would also be great.
The robots would definitely help with the water cannons right now!
Does sound like Japan, unlike the U.S., has not done any updated modifications for an event like this. Read my post above, the U.S. has done modifications to all our nuclear plants.
But the thing that sticks out like a sore thumb in this latest accident is the common mode failure. The plants withstood the seismic event. They shutdown upon detection of that, the diesels started when offsite power went out, everything was running until the tsunami came ashore. Anyone who has ever designed a system knows that you always try to avoid common mode failure where you can, but here things obviously were inadequate. I think that is where the focus will be in the "lessons learned" to come from all of this. And, historically, the industry has been pretty diligent in applying those lessons. I think we'll see that again here.