the gravity field, magnetic field, electrical resistivity, water flow, groundwater chemistry, atmospheric chemistry; and many other parameters that might be sensitive to stress, cracks in rock, or changes in the frictional properties of rocks. The literature is extensive (135); only a few examples are discussed here.

A logical successor to the seismic-gap model is the hypothesis that earthquake occurrence is accelerated or decelerated by stress increments from previous earthquakes. One version of this hypothesis is the stress shadow model—that the occurrence of large earthquakes reduces the stress in certain neighborhoods about their rupture zones, thus decreasing the likelihood of both large and small earthquakes there until the stress recovers (136). The stress model differs from the seismic-gap model in that it applies not just to a fault segment, but to the region surrounding it. Furthermore, because stress is a tensor, it may encourage some faults and discourage others. In some regions near a ruptured fault segment, the stress is actually increased, offering an explanation for seismic clustering. At present, the model offers a good retrospective explanation for many earthquake sequences, but it has not been implemented as a testable prediction hypotheses because the stress pattern depends on details of the previous rupture, fault geometry, stress-strain properties of the crust, possible fluid flow in response to earthquake stress increments, and other properties that are very difficult to measure in sufficient detail.

Seismicity patterns are the basis of many prediction attempts, in part because reliable seismicity data are widely available. Mogi described a sequence of events that many feel can be used to identify stages in a repeatable seismic cycle involving large earthquakes (137). In this model a large earthquake may be followed by aftershocks of decreasing frequency, a lengthy period of quiescence, an increase of seismicity about the future rupture zone, a second intermediate-term quiescence, a period for foreshock activity, a third short-term quiescence, and finally the “big one.” Any of the stages may be missing. This behavior formed the basis of an apparently successful prediction of the M 7.7 Oaxaca, Mexico, earthquake of 1978 (138). Unfortunately, there are no agreed-on definitions of the various phases that can be applied uniformly, nor has there been a comprehensive test of how Mogi’s model works in general (139).

Computerized pattern recognition has been applied in several experiments to recognize the signs of readiness for large earthquakes. V. Keilis-Borok and Russian colleagues have developed an algorithm known as “M8” that scans a global catalog for changes in the earthquake rate, the ratio of large to small earthquakes, the vigor and duration of aftershock sequences, and other diagnostics within predefined circles in seismically active areas (140). They report significant success in predicting which circles are more likely to have large earthquakes (141). Since 1999, they



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement