Several methods have been tested in the effort to learn how to predict earthquakes. Among the more serious methods which have been examined are seismicity changes, changes in seismic wave speed, electrical changes, and groundwater changes.
``Seismicity changes'' is really a fancy way of saying ``foreshocks''. A foreshock is an earthquake which precedes a larger earthquake and is ``near'' the epicenter of the larger earthquake. The problem for earthquake prediction is that not all earthquakes have foreshocks, and we don't yet understand what to look at to figure out if a given small earthquake will become a foreshock. This is an active area of research.
In the 1970s in the Soviet Union, several seismologists noted changes in the speeds of seismic waves in regions where later there were earthquakes. They noted that first the speeds dropped by a small amount up to a year ahead, and then a few weeks to months prior to the earthquake, the speeds began to increase again. Unfortunately, while this topic caused a lot of excitement in seismology 20 years ago, it has not lived up to the hype.
Changes in the Earth's electrical conductivity and electric and magnetic fields have sometimes been noted in areas near the future epicenter of a large earthquake. These changes could be brought on by changes in deep fluids along a fault plane or by stresses in the rocks along a fault. For example, just prior to the 1989 Loma Prieta earthquake, a group of scientists from Stanford University detected a massive increase in the noise level on one of their electric field recorders located in the Santa Cruz Mountains only a few kilometers from the San Andreas Fault. This got lots of people excited, and several experiments are in progress, but for now, the predictive value of these changes has not yet been demonstrated.
Changes in groundwater are also sometimes noted before earthquakes. These changes can include flow rates (sometimes wells dry up or dry wells become flowing again), taste or smell changes, and changes in chemistry, particularly in the concentration of the radioactive gas radon. All these have been noted in some earthquakes, but again, they are not universal features.
Are you beginning to see a pattern here? The real problem with earthquake prediction is that nobody has yet found a single factor or set of factors that always behaves in the same fashion before every earthquake of a given size. Until such a set of factors is found, we won't be able to make successful short-term earthquake predictions. In fact, some seismologists are now coming to believe (after 20-30 years and millions of dollars) that earthquakes may be fundamentally unpredictable -- that earthquake prediction is, in fact, completely impossible.
OK, so by now you are probably wondering if there has ever been a successful earthquake prediction anywhere in the world. The answer, surprisingly enough, is yes -- one. In February, 1975, Chinese seismologists predicted that a large earthquake would hit the Haicheng area within 72 hours. They made this prediction based on a long-term observation of increased seismicity in the region, coupled with short-term changes in groundwater, and widespread observations by farmers, fishermen, and the like that animals were behaving very strangely (the key thing here is that it was widespread). Despite the fact that it was February in northern China and bloody cold, the populace was evacuated to the city squares in the area. 65 hours later, a magnitude 7.3 earthquake hit the area and destroyed thousands of buildings. Due to the massive evacuation, fewer than 30 people lost their lives -- the toll could have been in the hundreds of thousands (including surrounding towns) otherwise.
The problem is that the very next year, in July 1976, the very same Chinese seismologists neglected to predict the devastating Tangshan earthquake. The city of Tangshan was nearly completely leveled. The official death toll stands at about 225,000. However, American seismologists who visited the area shortly after the earthquake have gone on record stating that more than 650,000 people were actually killed -- because the earthquake had not been predicted.
So, yes, there has been one successful prediction. No, there have been no more since then.
Just out of fairness, I should say that the American seismological community has had no better luck with earthquake prediction. Some of you may have heard of the Parkfield prediction. Parkfield is a tiny town (population 33, I think) in central California which seems to have magnitude 6 earthquakes at regular intervals, about every 22 years. Since the last one was in 1966, a prediction was made in 1984 that there was a 90% chance of another magnitude 6 earthquake in Parkfield by 1993.
It hasn't happened. And many millions of dollars worth of geophysical equipment -- everything from seismometers to tiltmeters to strainmeters to radon detectors to laser distance measuring systems to electric and magnetic field records -- sits in the hills there waiting for the next quake. When it comes, we should get loads of potentially very valuable data on what happens near the epicenter of a large earthquake just before that earthquake happens. But it must be said that the prediction itself has failed -- the earthquake did not come in the specified window of time.
You may detect a note of frustration on my part with earthquake prediction. While I believe that it would be useful to be able to make short-term earthquake predictions, I personally doubt we will ever get there. My frustration comes from having dealt with many amateur predictors who simply don't understand that they need to prove they can predict earthquakes, rather than just say they can. OK, you ask, but how would you go about proving that you can predict earthquakes?
That's a good question. Here are the five crucial features that any good earthquake prediction must have: