from California Geology, March 1988, Vol. 41, No. 3.


Condensed from A Proposed Initiative for Capitalization on the Parkfield, California Earthquake Prediction, 1986, National Academy of Sciences Press, Washington, DC, 28 pages.

Significant progress has been made during the last decade toward recognizing the increased potential for great earthquakes on seismic zones along the boundaries of the earth's crustal plates. The key to this has been the concept of the "seismic gap" as the site of a future significant earthquake. A seismic gap is a portion of a seismic zone, most easily recognized along one of the boundaries between the earth's crustal plates, within which at least one strong earthquake is known to have occurred in the past, but where such an earthquake has not taken place for a long period of time. How long a time is required for a particular site to be called a seismic gap is still largely empirical. It is generally influenced by the rates of relative plate motions across that section of the plate boundary, as well as by the documented intervals between past earthquakes. The seismic gap concept is crucial to the Parkfield Prediction Experiment.

Since 1965, when the gap hypothesis was introduced, 17 major earthquakes have occurred within previously-identified gaps. Although the locations and approximate magnitudes of these events were foretold, these successes were not true earthquake predictions in the sense that the times of occurrence were not stated reliably in advance. The existence of the gap does not in itself signify that an earthquake is imminent, but only that the potential for one is higher than in adjacent areas. The most recent large gap-filling earthquake was the magnitude 8.2 event on 19 September 1985 near Michoacan, Mexico. The site was clearly identified in the scientific literature as a seismic gap five years before the event.

Determining the average rate of recurrence of strong earthquakes on a fault is a key element in long-term hazard assessment. Recorded history is short compared to the time scale of geological processes, especially in a country as young as the United States. The recent development of techniques for extending the record of earthquakes on a fault into the more distant past by using geological evidence, has been one of the major achievements of the research under the U. S. National Earthquake Hazards Reduction Program. These paleoseismological methods have yielded, for example, the extension of the history of past major earthquakes on the San Andreas fault in southern California back about 2,000 years --- during which time 12 major earthquakes occurred on this section of the fault. Although the scatter in the time intervals between major earthquakes makes it impossible to predict the time of the next one accurately on this basis, such recurrence data, supported by known rates of plate movements, make it possible to estimate the probability of recurrence at any time after a given event.

From this kind of analysis, a probability of 50 percent has been established for the occurrence of a great earthquake in the next 30 years on the San Andreas fault near Los Angeles. As will be described in this series, similar analysis has given a probability of 67 percent for a significant earthquake on the Parkfield segment of the fault by the spring of 1993. However, a physically reasonable model that incorporates the somewhat "off-schedule" Parkfield earthquake of 1934 leads to a much higher probability for Parkfield within the next few years, and it is this calculation on which the current Parkfield prediction experiment is based.

Long-term predictions --- on the time scale of decades --- are important for planning hazard mitigation efforts, as well as for planning future scientific experiments. The accurate prediction of the time of occurrence on the scale of weeks to a few days or hours requires the identification of definite anomalous precursory phenomena.

The basic model of earthquakes that emerges from data on seismic gaps and plate tectonics theory is remarkably simple. Earthquakes are produced by sudden slippage on surfaces of failure ("faults") that relax elastic strains that have accumulated over long periods of time due to the relative movements of the earth's crustal plates. The direction and amount of slip are governed by the total strain accumulated across the fault between successive rupture events. Thus each earthquake terminates one cycle of strain accumulation and initiates the next.

To predict earthquakes on a short-term basis --- on the order of weeks or months --- requires understanding the physical processes that determine the time during the strain-accumulation cycle at which failure occurs. Laboratory analogs of earthquakes and theoretical considerations allow several possibilities that must be confirmed from field evidence. A key ingredient in the Parkfield experiment is a very dense instrumentation around the site of the anticipated rupture. We critically need measurements of the behavior of the fault during the interval from the late stages of the strain-accumulation cycle through strain release in the actual earthquake.

Direct detection of these precursory processes requires that measurements be made at very short range (a few kilometers) from the earthquake fault, because the amplitude of the deformation signals decays inversely with the cube of the distance from the fault.

Seismic waves generated by either micro-earthquakes or artificial sources are also an extremely important and powerful tool for probing the state of the fault zone.

The Parkfield Prediction

Seismologists have predicted that a moderate-size earthquake will occur within the next few years at Parkfield, California. The prediction has been painstakingly reviewed and subsequently endorsed in 1984-1985 as scientifically valid by two highly qualified panels. Nowhere else in the world is a prediction in effect with a degree of confidence as high as that for Parkfield. Here, on a specific 25-kilometer segment of the San Andreas fault about halfway between Los Angeles and San Francisco, studies during the past decade indicate a 95 percent probability that an earthquake of about magnitude 6 will occur before 1993. It is possible, although unlikely, that a large earthquake with magnitude up to 7 might occur, in which case more than 100,000 people could be affected.

A significant scientific effort is currently underway, supported by both federal and California state funding, to monitor the Parkfield area with instruments, in the hope that it will be possible to predict the earthquake on a still shorter time scale perhaps hours or days before the event. There are many scientific reasons for optimism because the prediction is based on remarkable similarities among five earthquakes that have occurred at this same location during the past 100 years.