For decades, earthquake experts dreamed of being able to divine the time and place of the worlds next disastrous shock. But by the early 1990s the behavior of quake-prone faults had proved so complex that they were forced to conclude that the planets largest tremors are isolated, random and utterly unpredictable. Most seismologists now assume that once a major earthquake and its expected aftershocks do their damage, the fault will remain quiet until stresses in Earth's crust have time to rebuild, typically over hundreds or thousands of years. A recent discovery--that earthquakes interact in ways never before imagined--is beginning to overturn that assumption.
This insight corroborates the idea that a major shock relieves stress--and thus the likelihood of a second major tremor--in some areas. But it also suggests that the probability of a succeeding earthquake elsewhere along the fault or on a nearby fault can actually jump by as much as a factor of three. To the people who must stand ready to provide emergency services or to those who set prices for insurance premiums, these refined predictions can be critical in determining which of their constituents are most vulnerable.
At the heart of this hypothesis--known as stress triggering--is the realization that faults are unexpectedly responsive to subtle stresses they acquire as neighboring faults shift and shake. Drawing on records of past tremors and novel calculations of fault behavior, my colleagues and I have learned that the stress relieved during an earthquake does not simply dissipate; instead it moves down the fault and concentrates in sites nearby. This jump in stress promotes subsequent tremors. Indeed, studies of about two dozen faults since 1992 have convinced many of us that earthquakes can be triggered even when the stress swells by as little as one eighth the pressure required to inflate a car tire.
Such subtle cause-and-effect relations among large shocks were not thought to exist--and never played into seismic forecasting--until recently. As a result, many scientists have been understandably skeptical about embracing this basis for a new approach to forecasting. Nevertheless, the stress-triggering hypothesis has continued to gain credibility through its ability to explain the location and frequency of earthquakes that followed several destructive shocks in California, Japan and Turkey. The hope of furnishing better warnings for such disasters is the primary motivation behind our ongoing quest to interpret these unexpected conversations between earthquakes.
CONTRADICTING the nearly universal theory that major earthquakes strike at random was challenging from the start--especially considering that hundreds of scientists searched in vain for more than three decades to find predictable patterns in global earthquake activity, or seismicity. Some investigators looked for changing rates of small tremors or used sensitive instruments to measure Earths crust as it tilts, stretches and migrates across distances invisible to the naked eye. Others tracked underground movements of gases, fluids and electromagnetic energy or monitored tiny cracks in the rocks to see whether they open or close before large shocks. No matter what the researchers examined, they found little consistency from one major earthquake to another.
Despite such disparities, historical records confirm that about one third of the world's recorded tremors--so-called aftershocks--cluster in space and time. All true aftershocks were thought to hit somewhere along the segment of the fault that slipped during the main shock. Their timing also follows a routine pattern, according to observations first made in 1894 by Japanese seismologist Fusakichi Omori and since developed into a basic principle known as Omori's law. Aftershocks are most abundant immediately after a main shock. Ten days later the rate of aftershocks drops to 10 percent of the initial rate, 100 days later it falls to 1 percent, and so on. This predictable jump and decay in seismicity means that an initial tremor modifies Earth's crust in ways that raise the prospect of succeeding ones, contradicting the view that earthquakes occur randomly in time. But because aftershocks are typically smaller than the most damaging quakes scientists would like to be able to predict, they were long overlooked as a key to unlocking the secrets of seismicity.[break]
Once aftershocks are cast aside, the remaining tremors indeed appear--at least at first glance--to be random. But why ignore the most predictable earthquakes to prove that the rest are without order? My colleagues and I decided to hunt instead for what makes aftershocks so regular. We began our search in one of the worlds most seismically active regions--the San Andreas Fault system that runs through California. From local records of earthquakes and aftershocks, we knew that on the day following a magnitude 7.3 event, the chance of another large shock striking within 100 kilometers is nearly 67 percent--20,000 times the likelihood on any other day. Something about the first shock dramatically increases the odds of subsequent ones, but what?
That big leap in probability explains why no one was initially surprised in June 1992 when a magnitude 6.5 earthquake struck near the southern California town of Big Bear only three hours after a magnitude 7.3 shock occurred 40 kilometers away, near Landers. (Fortunately, both events took place in the sparsely populated desert and left Los Angeles unscathed.) But the puzzling contradiction to prevailing wisdom was that the Big Bear shock struck far from the fault that had slipped during Landers's shaking. Big Bear fit the profile of an aftershock in its timing but not in its location. We suspected that its mysterious placement might hold the clue we were looking for.
By mapping the locations of Landers, Big Bear and hundreds of other California earthquakes, my colleagues and I began to notice a remarkable pattern in the distribution not only of true aftershocks but also of other, smaller earthquakes that follow a main shock by days, weeks or even years. Like the enigmatic Big Bear event, a vast majority of these subsequent tremors tended to cluster in areas far from the fault that slipped during the earthquake and thus far from where aftershocks are supposed to occur [see box on page 88]. If we could determine what controlled this pattern, we reasoned, the same characteristics might also apply to the main quakes themselves. And if that turned out to be true, we might be well on our way to developing a new strategy for forecasting earthquakes.
Triggers and Shadows
WE BEGAN BY LOOKING at changes within Earth's crust after major earthquakes, which release some of the stress that accumulates slowly as the planet's shifting tectonic plates grind past one another. Along the San Andreas Fault, for instance, the plate carrying North America is moving south relative to the one that underlies the Pacific Ocean. As the two sides move in opposite directions, shear stress is exerted parallel to the plane of the fault; as the rocks on opposite sides of the fault press against one another, they exert a second stress, perpendicular to the fault plane. When the shear stress exceeds the frictional resistance on the fault or when the stress pressing the two sides of the fault together is eased, the rocks on either side will slip past one another suddenly, releasing tremendous energy in the form of an earthquake. Both components of stress, which when added together are called Coulomb stress, diminish along the segment of the fault that slips. But because that stress cannot simply disappear, we knew it must be redistributed to other points along the same fault or to other faults nearby. We also suspected that this increase in Coulomb stress could be sufficient to trigger earthquakes at those new locations.
Geophysicists had been calculating Coulomb stresses for years, but scientists had never used them to explain seismicity. Their reasoning was simple: they assumed that the changes were too meager to make a difference. Indeed, the amount of stress transferred is generally quite small--less than 3.0 bars, or at most 10 percent of the total change in stress that faults typically experience during an earthquake. I had my doubts about whether this could ever be enough to trigger a fault to fail. But when Geoffrey King of the Paris Geophysical Institute, Jian Lin of the Woods Hole Oceanographic Institution in Massachusetts and I calculated the areas in southern California where stress had increased after major earthquakes, we were amazed to see that the increases--small though they were--matched clearly with sites where the succeeding tremors had clustered. The implications of this correlation were unmistakable: regions where the stress rises will harbor the majority of subsequent earthquakes, both large and small. We also began to see something equally astonishing: small reductions in stress could inhibit future tremors. On our maps, earthquake activity plummeted in these so-called stress shadows.[break]
Coulomb stress analysis nicely explained the locations of certain earthquakes in the past, but a more important test would be to see whether we could use this new technique to forecast the sites of future earthquakes reliably. Eight years ago I joined geophysicist James H. Dieterich of the U.S. Geological Survey and geologist Aykut A. Barka, then at Istanbul Technical University, to assess Turkey's North Anatolian Fault, among the worlds most heavily populated fault zones. Based on our calculations of where Coulomb stress had risen as a result of past earthquakes, we estimated that there was a 12 percent chance that a magnitude 7 shock or larger would strike the segment of the fault near the city of Izmit sometime between 1997 and 2027. That may seem like fairly low odds, but in comparison, all but one other segment of the 1,000-kilometer-long fault had calculated odds of only 1 to 2 percent.
We did not have to wait long for confirmation. In August 1999 a magnitude 7.4 quake devastated Izmit, killing 25,000 people and destroying more than $6.5 billion worth of property. But this earthquake was merely the most recent in a falling-domino-style sequence of 12 major shocks that had struck the North Anatolian Fault since 1939. In a particularly brutal five-year period, fully 700 kilometers of the fault slipped in a deadly westward progression of four shocks. We suspected that stress transferred beyond the end of each rupture triggered the successive earthquake, including Izmits.
In November 1999 the 13th domino fell. Some of the Coulomb stress that had shifted away from the fault segment near Izmit triggered a magnitude 7.1 earthquake near the town of Düzce, about 100 kilometers to the east. Fortunately, Barka had calculated the stress increase resulting from the Izmit shock and had published it in the journal Science two months earlier. Barka's announcement had emboldened engineers to close school buildings in Düzce that were lightly damaged by the first shock despite pleas by school officials who said that students had nowhere else to gather for classes. Some of these buildings were flattened by the November shock.
If subsequent calculations by Parsons of the USGS, Shinji Toda of Japan's Active Fault Research Center, Barka, Dieterich and me are correct, that may not be the last of the Izmit quakes aftermath. The stress transferred during that shock has also raised the probability of strong shaking in nearby Istanbul over the next few years from 2 to 4 percent per year. Over the next 30 years we estimate the odds for a major quake to be 62 percent; if we assumed large shocks occur randomly, the odds would be just 20 percent [see box on opposite page].
The stress-triggering hypothesis offers some comfort alongside such gloom and doom. When certain regions are put on high alert for earthquakes, the danger inevitably drops in others. In Turkey the regions of reduced concern happen to be sparsely populated relative to Istanbul. But occasionally the opposite is true. One of the most dramatic examples is the relative lack of seismicity that the San Francisco Bay Area, now home to six million people, has experienced since the great magnitude 7.9 earthquake of 1906. A 1998 analysis by my USGS colleagues Ruth A. Harris and Robert W. Simpson demonstrated that the stress shadows of the 1906 shock fell across several parallel strands of the San Andreas Fault in the Bay Area, whereas the stress increases occurred well to the north and south. This could explain why the rate of damaging shocks in the Bay Area dropped by an order of magnitude compared with the 75 years preceding 1906. Seismicity in the Bay Area is calculated to slowly emerge from this shadow as stress rebuilds on the faults; the collapsed highways and other damage wrought by the 1989 Loma Prieta shock--as well as a flurry of small shocks since 2000 along the Hayward, Rodgers Creek and Calaveras faults--may be a harbinger of this reawakening.[break]
Bolstering the Hypothesis
EXAMINATIONS OF THE EARTHQUAKES in Turkey and in southern California fortified our assertion that even tiny stress changes can have momentous effects, both calming and catastrophic. But despite the growing number of examples we had to support this idea, one key point was difficult to explain: roughly one quarter of the earthquakes we examined occurred in areas where stress had decreased. It was easy for our more skeptical colleagues to argue that no seismicity should occur in these shadow zones, because the main shock would have relieved at least some stress and thus pushed those segments of the fault further from failure. We now have an answer. Seismicity never shuts off completely in the shadow zones, nor does it turn on completely in the trigger zones. Instead the rate of seismicity--the number of earthquakes per unit of time--merely drops in the shadows or climbs in the trigger zones relative to the preceding rate in that area.
We owe this persuasive extension of stress triggering to a theory proposed by Dieterich in 1994. Known as rate/state friction, it jettisons the comfortable concept of friction as a property that can vary only between two values--high friction when the material is stationary and lower friction when it is sliding. Rather faults can become stickier or more slippery as the rate of movement along the fault changes and as the history of motion, or the state, evolves. These conclusions grew out of lab experiments in which Dieterich's team sawed a miniature fault into a Volkswagen-size slab of granite and triggered tiny earthquakes.
When earthquake behavior is calculated with friction as a variable rather than a fixed value, it becomes clear that Omoris law is a fundamental property not just of so-called aftershocks but of all earthquakes. The law's prediction that the rate of shocks will first jump and then diminish with time explains why a region does not forever retain the higher rate of seismicity that results from an increase in stress. But that is only half the story. Dieterich's theory reveals a characteristic of the seismicity that Omori's law misses entirely. In areas where a main shock relieves stress, the rate of seismicity immediately plunges but will slowly return to preshock values in a predictable manner. These points may seem subtle, but rate/state friction allowed us for the first time to make predictions of how jumps or declines in seismicity would change over time. When calculating Coulomb stresses alone, we could define the general location of new earthquakes but not their timing.
Our emerging ideas about both the place and the time of stress-triggered earthquakes were further confirmed by a global study published in 2005. Parsons considered the more than 100 earthquakes of magnitude 7 or greater that have occurred worldwide in the past 25 years and then examined all subsequent shocks of at least magnitude 5 within 250 kilometers of each magnitude 7 event. Among the more than 2,000 shocks in this inventory, 61 percent occurred at sites where a preceding shock increased the stress, even by a small amount. Few of these triggered shocks were close enough to the main earthquake to be considered an aftershock, and in all instances the rate of these triggered tremors decreased in the time period predicted by rate/state friction and Omori's law.
Now that we are regularly incorporating the concept of rate/state friction into our earthquake analyses, we have begun to uncover more sophisticated examples of earthquake interaction than Coulomb stress analyses alone could have illuminated. Until recently, we had explained only relatively simple situations, such as those in California and Turkey, in which a large earthquake spurs seismicity in some areas and makes it sluggish in others. We knew that a more compelling case for the stress-triggering hypothesis would be an example in which successive, similar-size shocks are seen to turn the frequency of earthquakes up and down in the same spot, like a dimmer switch on an electric light.[break]
Toda and I discovered a spectacular example of this phenomenon, which we call toggling seismicity. In 2002 we began analyzing an unusual pair of magnitude 6.5 earthquakes that struck Kagoshima, Japan, in 1997. Immediately following the first earthquake, which occurred in March, a sudden burst of seismicity cropped up in a 25-square-kilometer region just beyond the west end of the failed segment of the fault. When we calculated where the initial earthquake transferred stress, we found that it fell within the same zone as the heightened seismicity. We also found that the rate immediately began decaying just as rate/state friction predicted. But when the second shock struck three kilometers to the south only seven weeks later, the region of heightened seismicity experienced a sudden, additional drop of more than 85 percent. In this case, the trigger zone of the first earthquake had fallen into the shadow zone of the second one. In other words, the first quake turned seismicity up, and the second one turned it back down.
A New Generation of Forecasts
EAVESDROPPING ON the conversations between earthquakes has revealed, if nothing else, that seismicity is highly interactive. And although phenomena other than stress transfer may influence these interactions, my colleagues and I believe that enough evidence exists to warrant an overhaul of traditional probabilistic earthquake forecasts. By refining the likelihood of dangerous tremors to reflect subtle jumps and declines in stress, these new assessments will help governments, the insurance industry and the public at large to better evaluate their earthquake risk. Traditional strategies already make some degree of prioritizing possible, driving the strengthening of buildings and other precautions in certain cities or regions at the expense of others. But our analyses have shown that taking stress triggering into account will raise different faults to the top of the high-alert list than using traditional methods alone will. By the same token, a fault deemed dangerous by traditional practice may actually be a much lower risk.
Regardless of the factors considered, chance plays a tremendous role in whether a large earthquake occurs, just as it does in whether a particular weather pattern produces a rainstorm. The meteorologists advantage over earthquake scientists is that they have acquired millions more of the key measurements that help improve their predictions.
That is why my team is building an inventory of forecasts for large earthquakes near the shock-prone cities of Istanbul, Landers, San Francisco and Kobe. We are also making assessments for southern California [see illustration on page 83] and Tokyo, where a major earthquake could wreak a great loss of lives and trillion-dollar devastation.
The Indonesian earthquake of December 26, 2004, reminds us that a great earthquake's death toll can extend well beyond the reach of strong shaking when it triggers a large tsunami. This disastrous earthquake also left a seismic legacy: large Coulomb stress increases along Indonesias Sumatra Fault and the parallel Sunda trench. Illustrating how the stress-triggering hypothesis can help in earthquake forecasting, my colleagues John McCloskey, Suleyman S. Nalbant and Sandy Steacy of the University of Ulster in Northern Ireland warned in the March 17, 2005, issue of Nature that these stresses could trigger another great earthquake on either of these faults. Just 11 days later a magnitude 8.7 earthquake occurred on the Sunda trench.
In the end, the degree to which any probabilistic forecast will protect people and property is still uncertain. But scientists have plenty of reasons to keep pursuing this dream: several hundred million people live and work along the worlds most active fault zones. With that much at stake, stress triggering--or any other phenomenon that has the potential to raise the odds of a damaging earthquake--should not be ignored.[break]
ROSS S. STEIN is a geophysicist with the U.S. Geological Surveys Earthquake Hazards Team in Menlo Park, Calif. He joined the survey in 1981 after earning a Ph.D. from Stanford University in 1980. Steins research, which has been devoted to improving scientists ability to assess earthquake hazards, has been funded by U.S. agencies such as the Office of Foreign Disaster Assistance and by private companies, including the European insurance company Swiss Re. For the work outlined in this article, Stein received the Eugene M. Shoemaker Distinguished Achievement Award of the USGS in 2000. He also presented the results in his Frontiers of Geophysics Lecture at the annual meeting of the American Geophysical Union in 2001. Stein has appeared in several TV documentaries, including Great Quakes: Turkey (Learning Channel, 2001) and the 2004 IMAX film Forces of Nature.