Chaotic tremors within a computer.

Experiments with synthetic earthquakes created in a computer suggest that the San Andreas fault behaves much less predictably than previously thought -- a finding that carries both good and bad implications for residents living nearby.

Steven Ward and Saskia Goes at the University of California, Santa Cruz, made their discovery with a computer model that mimics earthquake recurrence over thousands of years, something researchers cannot determine from examining the relatively short historic quake record in California. The model works by gradually increasing stress on the San Andreas, which is divided into several segments that have different propensities for producing quakes. When the crustal stress reaches a critical value for a given fault segment, that part of the San Andreas slips, generating an earthquake that transfers stress onto adjoining segments.

By watching the model simulate 100,000 years, Ward and Goes see that each segment behaves irregularly. After one large tremor on a patch of fault, the next one in that same spot could come as soon as 50 years later or wait as long as 450 years. Ward believes the model presents an accurate picture of quake behavior because it does well at simulating the record of smaller, more numerous earthquakes within this century.

The finding raises questions about seismic hazard studies, notably a 1988 report by a panel of scientists that estimated the probability of earthquakes along each segment of the San Andreas for the next 30 years. To do this, the panel relied heavily on calculations of the average interval between the last few earthquakes on each segment. But the computer simulations by Ward's group show that intervals can vary substantially -- so records from the last few earthquakes cannot provide a true estimate of the probability that another will strike in the next few decades. Indeed, the Santa Cruz researchers calculate that the 1988 panel overestimated the long-term risks along each segment of the fault. That might seem like good news. But the work also indicates that scientists face a much more difficult task estimating seismic hazard.