Robert Watkins reckoned I missed the point on tsunami warning - saying that this one was detected in minutes (in Hawaii) but there was no way to get the message out. Nick said in comments that it was a 'Act of God', something that humanity can do much about (my post had referred to general politico-economic ills).
Regarding getting the message out, sure, this is undoubtedly one of the hardest parts of the problem. The other hard part is ensuring appropriate response procedures take place on the ground after a warning has been received. The easier parts are acquiring the raw data, processing it (to make predictions) then notifying local hosts of the incoming danger. Like I mentioned before, the distribution to end users would be most appropriate as a publish/subscribe system. Bob Wyman (who has an existing distribution system of adequate scale, pubsub.com) gave some explanation of the problems from that point of view. He highlighted lack of good standards for expressing earthquake data in XML, and aggregator support for type-specific rendering of messages. Like other syndication systems this could be approached in a totally per-case manner, but a more generalised one is possible.
I think it's reasonable to assume that at least most of the number crunching could be done relatively easily server side. I hope so, because a simple client will make a lot of difference in being able to maximise the number of potential receivers. Related to this is the question of how much info the syndication system needs to carry. At minimum would be a warning including predicted hazard time and simple description. This should have an identifier available for tracing back where the prediction came from, a paper trail. At the other end of the scale would be the publishing of the raw data and letting client-side tools draw their own conclusions. I think the practical solution probably falls between the two, with the warning message carrying some of the technical details (either as content blobs or more usefully RDF statements), with adequate linkbacks for a sophisticated client to retrieve and work on the source data itself.
The minimal client software needn't be much. A starter might just display:
<description>Possible tsunami from earthquake; general area Zz Ocean; magnitude m; coords: x,y; </description>
But it would be important to have either push pubsub support or (probably localized) servers that could handle frequent polling.
Damn, I've just seen the time - better get on with some work. I'll continue this later, ramble on about the 'last mile' a little…[Danny]