Earthquake prediction is much like taking on the stock market; you can see patterns, you have an idea what is going on, but can you make a profit? For stocks it is a purely a matter of money, for a social benefit like earthquake prediction, it is a matter of whether the prediction truly helps anybody.

Earthquakes were once thought to be random geologic events without cycle or pattern, and in a way they behave a lot like stocks, wind bursts, solar flares, and other events that have properties described by chaos theory. The method of attempting to predict the stock market by direct examination of historic price swings has its adherents, and its detractors. In fact there are fundamental laws working at the smallest to largest scale, and the frequency and size of events correspond to what is called a power law. For stocks, there is a pattern that the number of daily price-swings of 1%, is, perhaps, 8 times the number of days when the swing is 2%, which is eight times the number of days at 4%. Once in a while, there is a major 'stock-quake' with a big swing!

The number of earthquakes and their size generally follow a power law as well (see Zipf's law). Earthquakes are rated in size by a logarithmic scale called magnitude (calculated in different ways, one such being the Richter magnitude scale). An M3 earthquake has about 30 times more energy than an M2. Also, M2 earthquakes occur 10 times more often than M3's, which are 10 times more frequent than M4's, etc. Some natural phenomena follow power laws because they are fractal in nature, being self-similar over all scales. As a result of this ubiquity and intrinsic perceptual biases, people generally see 'patterns' or 'things' in any fractal distribution. Thus, the background of stars has its constellations, or one may see a duck in a fluffy cloud. One branch of mathematics that deals with pattern analysis explicitly is called Ramsey Theory.

Like stocks, the pattern of earthquakes is quite capable of being correlated with anything — once! People have 'associated' the onset of an earthquake with such things as animal behavior, the weather, motion in the level of water wells, etc. Unfortunately, unlike clouds, patterns which might be useful in predicting quakes are not as evident as that rain is more likely when it is cloudy than when it is not. The science of statistics is primarily concerned with discovering patterns and quantifying evidence of associations or correlations in data, regardless of cause. For example, a statistical link may be established between consumption of fatty food and cardiovascular disease, just as there is a statistical link between cigarette smoking and various illnesses.

To be socially useful, earthquake predictions do not have to be ultra precise in magnitude, time and place. Even predictions of a general nature can be quite useful if they are based on scientific principles. For example, the region near the town of Parkfield in California has experienced a magnitude 6 earthquake approximately every 22 years since some time in the 1800s. This led researchers to predict that a similar quake would hit the region in the mid-1980s. Because of the potential value of the scientific data that could be obtained from monitoring seismic data prior to a quake, and because the Parkfield area is relatively quiet - in comparison to most urban areas with respect to man made seismic activity, the region was heavily instrumented with all varieties of monitoring equipment.

The predicted quake failed to materialize on the expected fault, however a sizable quake did occur in nearby Coalinga, California in 1983. Perhaps the Coalinga quake released some of the stress on San Andreas Fault near Parkfield, and was in effect a substitute for the missing quake. If that is the case, then one would have expected that the next quake in the Parkfield region would be sometime in the mid 2000s. Indeed another killer quake occurred near Parkfield, this time in San Simeon, California in December 2003. Regrettably, the San Simeon quake of December 2003 produced two fatalities in the town of Paso Robles.

It was not until November of 2004 that the expected (but not predicted) Parkfield earthquake arrived, with analysis of the collected data offering no immediately obvious indications that could be used for prediction.

While it might be desirable to be able to predict a specific quake, of a particular magnitude on a given day, the more socially useful predictions in fact are the predictions that a particular geographic region might be especially likely to have a major seismic event within a particular time frame. That is because if it could be determined that a killer quake was definitely going to hit an area, even as vaguely as 'soon', then it becomes possible for regional planners to allocate resources for such projects as urban redevelopment, retrofitting, etc., in those areas where the commitment of a portion of otherwise finite public capital will have the greatest public benefit.

Predicting such things as a small earthquake in California 'any day now', might be somewhat like saying a horse will win the Kentucky Derby - in fact a group of scientists at the University of California, Los Angeles, lead by Dr. Vladimir Keilis-Borok, predicted that a quake similar to the Paso Robles quake would strike in a 12,000 square mile (31,100 km) area of Southern California within a time frame of a few months. However, this predicted earthquake never materialized.

As scientists study earthquakes they may become more precise in their estimates of seismic hazard, using such advanced tools as real-time GPS.

External link

• The USGS view on earthquake prediction http://www.geophys.washington.edu/SEIS/PNSN/INFO_GENERAL/eq_prediction.html