Since we just had a 5.9 earthquake centered about 100km away, I thought I would link to a recent Freakonomics Radio piece about why we can’t predict earthquakes. My favorite part of the broadcast was this exchange:
ELLSWORTH: I think there are many people who have the belief that they can predict earthquakes, and the real question is, “Can you show that you made a prediction before the fact?” and that that prediction can be tested in a scientific manner. We can certainly say that there is going to be a magnitude five earthquake in the world somewhere today. That’s an easy prediction. We can probably say that there’s going to be a magnitude six earthquake in California in the next 100 years. That’s a gimme. But in terms of making a prediction that’s really testable, that’s a much harder thing to do. One of the very interesting things that was done years ago, this was back in the days of telephone answering machines. We had a project that had a group of volunteers. People who had animals that were very sensitive to the coming of earthquakes. And the instructions were that when their animals acted up, they were to call the number and leave a message. There were never any messages left before earthquakes. But lots of messages left after earthquakes. Oh, they had meant to call, but they didn’t.
DUBNER: So in other words, “Moments before the earthquake, I noticed that my cat was pacing back and forth.”
ELLSWORTH: That’s the kind of thing. Exactly, yeah.
Notwithstanding, there has been some progress. Back in 2005 Mason professor Guido Cervone was apparently having some success predicting some large coastal earthquakes using a data-mining approach. If I understand correctly, it was essentially detecting a robust property of foreshocks. He writes to say he hasn’t worked on this in awhile, so I infer the benefit wasn’t overwhelming.
Sometimes the best you can do is probability: it’s really likely to happen within the decade, but when in that period is anyone’s guess.