Over geologic time,
the net displacement across a fault accumulates through the action of countless
individual slip events. Measured over many displacement cycles, the average
interval between events must equal the average event displacement divided by
the remote slip rate. First principles, however, provide little guidance as
to the properties of the recurrence, which might range from a totally random
distribution of events in both space and time to identical earthquakes repeating
at fixed intervals. If recurrence is essentially random, then long-term seismic
hazard is described by the Poisson rate of activity, as discussed above. Greater
regularities and systematics in recurrence imply that useful time-dependent
forecasts of future activity can be derived from knowledge of the past behavior
of the fault system.
Results for San Andreas earthquakes have played a central role in establishing the existence of broad regularities in the recurrence process. At Parkfield, the San Andreas fault has ruptured six times since 1857 in M~~6 events with highly repeatable characteristics every 22 +/- 6 years. The latest three events, in 1922, 1934, and 1966, for which instrumental records exist, are virtually identical ( Figure 6.15; Bakun and McEvilly, 1984). Amplitude data from Milne seismographs uncovered in the preparation of Table 6.1 show that the 1901 and 1922 events produced the same surface-wave amplitudes on common stations, strengthening earlier speculations that all the 20th-century events are similar. Intensity data for the 1881 event (Toppozada and others, 1981) and for foreshocks to the great 1857 earthquake (Sieh, 1978b) place them along the Parkfield segment as well. These regularities in the size, location, and timing of all known events at Parkfield led Bakun and Lindh (1985) to propose a specific recurrence model for Parkfield earthquakes. On the basis of this model, the next in the series of characteristic events is anticipated before 1993, and its forecast represents the first formally endorsed earthquake prediction in the United States.
Geodetic analysis of the strain released in the 1966 earthquake and its subsequent reaccumulation led Segall and Harris (1987; see Harris and Segall, 1987) to identify the zone where strain accumulates and is released, the "Parkfield asperity," as the center of the 1966 rupture zone. This zone of strain accumulation appears to be effectively locked during the interseismic period and corresponds to the center of the 1966 aftershock zone (Eaton and others, 1970) between about 4- and 10-km depth. The significantly fewer events in this part of the aftershock zone than in its periphery suggests that Parkfield earthquakes occur when this locked zone suddenly releases. Aftershocks appear to result from transfer of stress to the perimeter of the asperity.
This same pattern of concentrated coseismic slip occupying a quiet region within the overall aftershock distribution characterizes several recent, well-studied events (Mendoza and Hartzell, 1988), three of which, the Coyote Lake earthquake (Aug. 6, 1979), the Imperial Valley earthquake (Oct. 15, 1979), and the Morgan Hill earthquake (Apr. 24, 1984), all have probable antecedents within the historical record. Reasenberg and Ellsworth (1982) identified the June 20, 1897, earthquake as a predecessor to the 1979 event and noted that the 82-year interval between events equaled the 1.2 m of coseismic slip determined by Liu and Helmberger (1983) divided by the long-term slip rate of 1.5 cm/yr for the Calaveras fault. Similarly, the 73-year interval between the July 11, 1911, event and the 1984 Morgan Hill earthquake (Bakun and others, 1984) well predicts the 0.8 to 1.0 m of maximum coseismic slip determined by Hartzell and Heaton (1986). The 1979 Imperial Valley earthquake is more complex because it reruptured only the northern 30 km of the May 19, 1940, fault break. Again, both the time interval between events and the fault-slip rate compare favorably with the fault slip at depth, as determined from seismograms (Hartzell and Heaton, 1983; Archuleta, 1984). Earlier ruptures of this or other segments of the Imperial fault may well be in the historical record, possibly including the April 19, 1906, event, which occurred the afternoon of the great 1906 earthquake in northern California.
Similar observations of recurrent faulting in events with characteristic magnitudes and locations from around the world (Nishenko and Buland, 1987) suggest a simple, first-order model for seismic potential. In this model, the future behavior of a specific segment of a fault can be forecast from knowledge of the size of past earthquakes, the timing and amount of slip in the latest event, and the long-term Yate of fault movement (Lindh, 1983; Sykes and Nishenko, 1984). Accordingly, the probability of an event on a recently ruptured fault segment is low until the elastic strain rebuilds, which may be estimated from the geologic slip rate. As the strain rebuilds, the probability of another earthquake increases. Empirically, the time intervals between successive ruptures of a specific fault segment define a bell-shaped distribution that may be used to estimate the odds of the next event within some future time interval, given that it has not yet occurred.
Probabilities for large earthquakes along the major branches of the San Andreas fault derived from this methodology differ markedly from Poisson estimates (Working Group on California Earthquake Probabilities, 1988). For example, the chance of a repetition of the great 1906 earthquake within the next 30 years (1988-2018) is less than 0.1. In contrast, the chance of an M=7-1/2-8 earthquake on the southern section of the San Andreas fault is 0.6. When the Working Group's report was written, the southernmost part of the 1906 fault break was assigned the highest chance of failure of any segment of the north half of the San Andreas fault. Now that it has ruptured in the October 18, 1989, Loma Prieta earthquake, the probability of another rupture will be small for several decades. A clearer understanding of past seismicity can only help to improve and refine estimates of future seismicity.