Earthquakes occur when rock fractures beneath the surface of Earth. There are many types of earthquakes, each with characteristic patterns of vibrations, or tremors. The study of these vibrations is called seismology, after the Greek word for "motion." Some are associated with magma rising towards the surface of a volcano. Others are associated with rockbursts in deep mines. The vast majority of earthquakes, however, are modelled by seismologists as two faces of rock slipping past each other along an approximately flat internal surface, called a fault. Faults can be seen and touched at the surface, if fault slip extends this far, or if erosion exposes older faults. If slip on a fault breaches the surface, it can be observed at fault scarps, along which one can measure the relative motion of the two rock faces. H. R. Reid published a comprehensive survey of slip along the San Andreas Fault after the great 1906 San Francisco earthquake. He proposed that earthquakes are an elastic dislocation in rock, like the snapping of a ruptured rubber band. Reid's concept continues to influence our thinking about earthquakes.

When most rocks are stretched or squeezed by tiny amounts (one part in 100,000 or less), they behave elastically, and can vibrate like a spring. The vibrations dont last forever, because friction within the rock eventually transforms the vibrational energy into heat. The tremors that emanate from an earthquake are called seismic waves, elastic vibrations that propagate in response to travelling stresses (a type of force) within rock. The dislocation, or slip, along a fault excites seismic waves much as a stone dropped into water excites water waves. In solid rock, unlike fluid water, two types of elastic waves are possible: P waves and S waves. P waves are compressional, involving the expansion and contraction of material in response to travelling stresses. S waves involve a sideways shearing of material as the stress travels. Figure 1 demonstrates, by using a spring as a model for rock vibrations, that P waves cause material to oscillate back and forth in the direction the wave travels. S waves cause material to oscillate sideways relative to the wave's propagation direction. P waves can propagate in fluid and gaseous material, where we recognise them as sound waves. S waves propagate only in solid material.

P waves travel faster than S waves, and so are recorded first by instruments called seismometers, which electronically amplify and record small vibrations. The record of ground motion is called a seismogram. In the first seismometers, seismograms were drawn by a pen on a sheet of paper on a slowly rotating drum. Drum seismometers can still be found in the display cases of universities and museums. Modern research seismometers are digital electronic devices, which translate ground motion into bits and bytes in memory chips, disk drives or magnetic tape. Seismometers that record both vertical and horizontal vibrations are especially useful for distinguishing P and S waves, since the vibration directions of the two waves differ. For earthquakes in the crust and shallow mantle, much of the vibrational energy emanated by fault motion is trapped near the surface in the form of a surface wave, similar to a pulse travelling along a carpet. Surface waves come in two types as well: Love waves are composed of shear motion only, and Rayleigh waves are a combination of P and S wave motion.

Unlike ripples on a pond caused by a dropped stone, seismic waves from an earthquake do not have equal amplitude in all directions. By comparing P and S waves on seismograms from a network of seismometers, seismologists can reconstruct the orientation of the fault, and the direction of slip (Figure 2). These reconstructions indicate that earthquake faults have a wide range of orientation and slip preferences. There are three major types of earthquakes, thrust (or reverse), normal, and strike-slip. Strike-slip earthquakes are the easiest to visualize, as these occur when two vertical faces of rock slide laterally past each other. Both thrust and normal earthquakes involve vertical motion on a fault, which is usually tilted at an angle from the vertical. On such a dipping fault, seismologists identify the lower rock face as the foot wall, and the upper face as the hanging wall. In thrust earthquakes the foot wall dives under the hanging wall, which rises up. This type of earthquake is common where plates or crustal blocks collide. In normal earthquakes the foot wall is uplifted as the hanging wall sinks. This type of earthquake is common where plates or crustal blocks are pulled apart. Both thrust and normal earthquakes can, after many repeated earthquakes, give rise to mountain ranges. The Himalayas rise above a network of thrust faults along the borders of India, Nepal and China. The Wasatch Mountains rise above Salt Lake City along a north-south normal fault system that splits the state of Utah in two.

Seismologists have several methods to measure the size of earthquakes. The seismic moment Mo is a rough measure of the physical work done by slip on a fault. It is determined by multiplying the fault area A by the average slip on the fault D and the rock's shear stiffness {mu}, so that Mo={mu}AD. Although seismologists cannot directly measure any of the three factors {mu}, A, or D, the product Mo can be estimated from the amplitude of seismic waves recorded by seismometers. Rock fracture in detectable earthquakes can occur on scales ranging from a few meters to 1000 km, with slips from millimeters to tens of meters, so seismic moment can span many orders of magnitude. Therefore the size of an earthquake is more conveniently expressed as a magnitude on a logarithmic scale. The Richter magnitude M can be estimated from the amplitudes of either P, S or surface waves, and is the measure of an earthquake most widely quoted by the news media. As devised by Charles Richter at the Califirnia Institute of Technology in the 1930s, each step on the magnitude scale correponds to a 101.5 increase in seismic moment and energy release, approximately a factor of 32. Two steps correspond to a factor of 1000 increase. One meter of slip on a circular fault of radius 10 km gives rise (approximately) to a magnitude M=7.0 earthquake. Assuming that rock stiffness stays constant, a magnitude M=9.0 earthquake would scale up to 10 meters of slip on a circular fault of 100 km radius. Only two earthquakes exceeding magnitude 9.0 have occurred in the 20th century, in 1960 in Chile, and in 1964 in Alaska. Earthquakes with magnitudes between 7 and 8 are much more frequent, occurring worldwide on average more than once per month.

The modified Mercalli magnitude scale is used to express the damage caused by an earthquake, rather than the size of the earthquake itself. This magnitude varies from place to place, according to proximity to the fault, local geology and the clustering of man- made structures. The Mercalli scale ranges from I ("Tremors felt by instruments, but not bystanders") through VI ("Felt by all, many of whom run outdoors. Some damage to plaster and brick chimneys.") to XII ("Near-total destruction. Waves actually seen by bystanders along ground surface. Objects tossed into the air."). Mercalli magnitudes, though often somewhat subjective, are a useful for policymakers who wish to relate earthquake type and size to actual societal hazards. In many earthquake-prone regions, the most damaging earthquakes have occurred before 1900, before the advent of drum seismometers. To estimate Richter magnitudes for these historical earthquakes, seismologists combine measurements of fault scarps and other ground disturbance with Mercalli magnitudes based on old damage reports.

In order to explain where earthquakes most commonly occur, one must consider two primary factors. First, external forces must be straining the rock, causing it to move. Second, the rock must respond to these applied forces by suffering brittle fracture, not steady plastic flow. On a large-scale, both of these factors can be related to the theory of plate tectonics. Earthquakes occur predominantly at the boundaries of a global patchwork of surface plates, where neighboring plates move past each other. On a finer scale, especially relevant to detailed hazard mitigation efforts, local conditions become important. For instance, earthquakes are most likely to occur along faults, planar surfaces that have already suffered previous earthquake slippage. Within the faults themselves, earthquakes often start at favored locations, called asperities, which may be characterized by roughness in the fault surface, a patch of high-strength rock, or other special properties. As scientists narrow their attention from the perspective of tectonic plates down to the perspective of how the rock within an asperity fractures, our understanding of the earthquake process becomes less and less certain. We can forecast with some confidence the broad regions where earthquakes are likely to occur within the next few decades or centuries. However, our current scientific understanding is inadequate to provide the precise location and timing of the next earthquake to occur, for instance, near San Francisco. At the present time, earthquake risk can best be addressed by society on the broad scale, even as scientists continue to study the details of how, why and when earthquakes occur.

In plate-tectonic theory, the surface of Earth is divided into roughly 14 major plates, ranging in size from the Arabian peninsula to the nearly all the Pacific Ocean. The plates are rigid, so that motion at the surface occurs primarily along plate boundaries. The plates move across the surface at velocities typically between 10 and 100 mm/yr. The forces applied to rocks along the plate boundaries build up slowly, and large earthquakes along different portions of a boundary appear to occur independently. In the United States, the Pacific plate borders the North American plate within California, trending north- northwest approximately from the southern Mojave Desert near the Salton Sea to Cape Mendocino near the city of Eureka. Geologists traditionally locate this strike-slip plate boundary along the San Andreas Fault, along which the Pacific plate slides northwards relative to North America. From Cape Mendocino to the Canadian border the North American plate borders the Juan de Fuca plate, the last remnant of a larger oceanic plate, most of which has subducted into the mantle under the western United States. All along the southern coast of Alaska, the Pacific plate borders the North American plate again, sliding northwards along Alaska's eastern "panhandle." Subduction occurs where the coastline veers west to connect with the Aleutian Island chain. One of the largest earthquakes in the 20th century, with magnitude M=9.2, ruptured the central portion of this plate boundary on Good Friday 1964. It devastated the city of Anchorage and generated tsunami water waves that flattened military bases on the coast of Kodiak Island. The plate boundary continues along the volcanic islands of the Aleutians, trending from a pure subduction to a pure strike-slip boundary as it approaches the Russian Kamchatka peninsula.

The central and eastern United States does not contain or border a major plate boundary, and so has many fewer earthquakes. Several US island territories are lie along plate boundaries, however. Puerto Rico lies along the boundary of the North American plate and the Caribbean plate. Guam lies along the boundary of the Pacific plate and the Philippine plate. Guam suffered a magnitude M=8.1 earthquake on August 8, 1993. The Hawaiian islands lie far from any plate boundaries, but are prone to earthquakes associated with volcanism associated with a mantle hot spot.

Earthquake hazards in other countries are also concentrated along plate boundaries. The islands of Japan lie along subduction zones at the eastern edge of the Eurasian plate, and have a long history of devastating shocks. The Pacific coast of Mexico, Central America and South America parallels an active subduction zone for thousands of kilometers, leading to substantial risk. From southern Italy eastward through the Middle East to Iran stretches a complex pattern of earthquake zones associated with the confluence of the African, Eurasian and Arabian plates. Finally, the collision of India, riding on the Indo-Australian plate, with the Eurasian plate gives rise to frequent earthquakes along the base of the Himalayan mountains. The effects of this continental collision are felt most severely beyond the Himalayas in China, where it causes many of the most damaging earthquakes worldwide, in terms of death toll.

Plate tectonics helps to explain where subsurface rocks are brittle. Through most of the Earth interior, the pressure and temperature of rocks, combined with exceedingly slow motion of plate tectonics, makes them more susceptible to plastic flow than to brittle fracture. The rocks are quite solid, but nevertheless stretch and shear without breaking. Only near the planet's surface, or in unusually cool internal regions, is brittle fracture more likely. The outermost 10-100 km of Earth stiffens into rigid plates as it cools and loses heat to the oceans and atmosphere. At mid-ocean ridges, where oceanic plates are created, the cooling has "just started," the plate is thin, and earthquakes occur only in the upper few kilometers. At subduction zones, the oceanic plate has cooled sufficiently to sink back into the mantle, and brittle fracture extends more deeply. In fact, seismologists can trace the downward path of many plates to depths of 600-700 km by following the earthquakes that continue to fracture its cooler, stiffer rock. The largest earthquakes worldwide occur in shallow subduction zones. Many of these are located beneath sparsely-populated areas, like the Tonga Islands in the south Pacific Ocean. However, large cities like Tokyo and Seattle are at risk from subduction-zone earthquakes of magnitude 8 or more. The deepest earthquakes are rarely large and rarely damaging. However, a magnitude 8.3 earthquake at 500 km depth beneath South America on June 9, 1994 rattled a wide area of Bolivia, and was felt as far away as Toronto, Canada.

On a regional rather than a global scale, the theory of plate tectonics by itself fails to explain many details of earthquake locations and faulting. The largest earthquakes may follow the plate boundary geometry, but smaller, damaging earthquakes with different geometry tend to surround the boundary. The city of Kobe, Japan suffered a devastating earthquake on 17 January 1995. This earthquake occurred on a strike-slip fault that bounded this port city on its inland side, not on the thrust faults associated with the subduction zone offshore. In southern California, the strike-slip San Andreas Fault has not experienced a "great" earthquake with magnitude in the M=7-8 range since Fort Tejon earthquake of 1857. The region's largest earthquake in several decades was located in the Mojave desert well north of the plate boundary: the 28 June 1992 earthquake near Landers, California (M=7.4). More importantly, "moderate" earthquakes in the M=6- 7 range have been surprisingly frequent on blind-thrust faults within the heavily- populated Los Angeles Basin. The most recent of these, on 17 January 1994 beneath Northridge, California (M=6.9), is currently the most damaging US earthquake in terms of dollar cost. Fault slip on "blind" thrust faults is not seen on fault scarps at the surface. Even so, blind-thrust faults can give rise to mountainous topography, including most of the east-west trending hills within the Los Angeles Basin. Although located on the San Andreas Fault itself, the 17 Oct 1989 Loma Prieta earthquake in northern California also had a blind-thrust character. The steep mountains between the cities of Santa Cruz and San Jose testify that this characteristic was not a fluke, as these and other lines of mountains in the San Francisco area are likely to have had a similar origin.

Details of faulting patterns can often be related to the regional response of crustal blocks to forces at the plate boundary. The thrust faults of the Los Angeles Basin have been related to a large bend in the San Andreas Fault north of the city, which appear to impede smooth sliding along the plate boundary. As for the Kobe earthquake, strike-slip faults are common within the overriding plate at many subduction zones, allowing slivers of crust to slide sideways along the trench, like a watermelon seed squeezed between your fingers. In fact, some geologists hypothesize that the San Andreas Fault began this way, transporting slivers of crust northward along the California coast behind a long-vanished ocean trench. Farther inward from a subduction plate boundary, it is often noticed that the overriding plate is stretched as though tugged outward by the trench itself. Such extension is thought responsible for the normal earthquakes in the Aegean Sea and surrounding portions of Greece and Turkey. Residual forces from past subduction may contribute to the normal earthquakes in the Basin and Range region between the Sierra Nevada Mountains of California and the Wasatch Mountains of Utah, encompassing Nevada, western Utah, and parts of southern Oregon and Idaho. Earthquakes near Yellowstone National Park in northwestern Wyoming, including faulting that raised the Grand Teton range, appear related to volcanism associated with a mantle hot spot centered beneath the park.

The nature of the continental crust may help explain why earthquake activity spreads far from plate boundaries on land, but much less so in the ocean. Continental crust is thicker (35-40 km on average) than oceanic crust (7 km), and is composed of a different assortment of rock types. Laboratory experiments using quartz, a mineral common in continental crust but less so in oceanic crust, suggest that continental crust undergoes a transition from brittle to ductile behavior in the middle of the crust, at depths near 15 km. Significantly, nearly all earthquakes in the continental crust are confined to the upper crust, suggesting that the lower crust flows plastically in response to plate-tectonic forces. The existence of a brittle-ductile transition within the continental crust would imply, for instance, that the network of small mountain ranges in the Los Angeles Basin, each associated with one or more "blind" thrust faults, does not reflect a similar crumpling of the mantle beneath. In fact, there are no moderate-to large earthquakes at sufficient depth beneath California to assure us that the mantle portion of the Pacific-North American plate boundary is truly confined along the narrow trace of the crustal San Andreas Fault. Here and perhaps elsewhere in the world, plate motion in the mantle may occur over a broader shear zone, transmitting forces upward through the ductile lower crust to cause brittle fracture in the upper crust. This conceptual model may help explain intervals where the San Andreas Fault is accompanied by parallel strike-slip faults, such as the Hayward fault on the eastern side of San Francisco Bay. Since large-scale plate motion could, in principle, be accommodated by slip on either of these parallel faults, earthquake risk becomes more diffuse and less predictable.

Plate tectonic theory does little to explain the pattern of earthquakes in the central and eastern United States. Here tremors are very infrequent, but can be devastating. Earthquake hazards are estimated largely from historical accounts of earthquakes prior to the 20th century, plus field surveys of past ground disturbances. Historical damaging earthquakes and small contemporary tremors are concentrated in the New Madrid region near the common borders of Missouri, Arkansas and Tennessee, the Appalachian mountains along the Tennesee/Virginia border, the Charleston area of South Carolina, and a coastal strip along New Jersey, New York and New England, extending into the Maritime provinces of Canada. The St Lawrence Seaway in Quebec is the site of many moderate earthquakes, including a M=7.0 earthquake in 1925. The underlying causes of these earthquakes are not well known. In many cases, earthquake slip along the Atlantic seaboard occurs on long-inactive faults, thought to be associated with the initial rifting of the Atlantic Ocean. Beneath the New Madrid region, geologists have identified an ancient rift that failed to split North America apart more than one billion years ago. Some have speculated that contemporary motion on these old faults is induced by the viscous rebound of North America after the retreat of 1-km-thick continental ice sheets roughly 10000 years ago. Although much slower than the motion at plate boundaries, the pattern would be similar: plastic flow in the mantle, drawn back into the depression where the ice sheet had formerly lain, strains the brittle upper crust, causing earthquake fracture.

Seismologists must be cautious when advising society about impending earthquakes. They distinguish between earthquake prediction and earthquake forecasting. There is currently no agreed-upon method to predict the location, size, and probable time of an impending earthquake on a particular fault. However, it is currently possible to forecast the likelihood of earthquakes of a given size in a fault region over time periods of decades to centuries. Earthquake "prediction" and "forecasting" differ mostly in their claimed precision in time, and in the advisable societal response. If earthquakes could be predicted with an accuracy of days or even months, policymakers could alert emergency-response agencies and perhaps move people out of dangerous areas while waiting for the earthquake to occur. It is difficult to keep an urban population in a high-alert status for long periods, however, so the societal response to a long-term earthquake forecast should involve more permanent measures, such as assembling maps of locations that suffer the most intense shaking during earthquakes, using such maps to formulate zoning laws and buiding codes to limit construction in dangerous areas, and educating the general public. Society can use an earthquake forecast to limit property damage and loss of life, but cannot expect to escape them altogether.

What are the prospects for more accurate predictions of earthquake occurrence? Skepticism is justified by the fact that geologists do not yet know how rock fracture initiates during an earthquake, nor what determines how far a fracture proceeds once it starts. Without a clear idea of how the system works, it is difficult to justify the accuracy of an earthquake prediction. However, earthquake research has advanced greatly in the past few decades, so this disappointing situation may improve.

Many seismologists argue that earthquakes in a fault zone exhibit self-similar behavior. A "self-similar" physical system behaves identically on all length and/or time scales, and to some extent can be described using the mathematical theory of fractals. If rock fracture during earthquakes is a self-similar process, small and large earthquakes behave identically. Rupture initiates and proceeds according to identical "rules." One expects there to be no intrinsic upper or lower limit to the size of earthquakes. Also, the frequency of earthquake occurrence would follow a power law with respect to size, in which the number of "large" earthquakes per year, divided by the total number of earthquakes per year, is a predictable ratio. To a large extent, observed sizes and frequency-of-occurrence in seismicity catalogs of past earthquakes follow the predictions of self-similarity. Though the maximum size of an earthquake is limited by the extent of brittle rock in the shallow earth, detailed studies have demonstrated that earthquakes with Richter magnitudes M<0 are quite numerous, consistent with the lack of a lower limit to earthquake size. In addition, seismologists have long recognised that the frequency of earthquake occurrence tends to follow a power law with respect to size. Define N to be the number of earthquakes per year above a certain magnitude M. The power law for earthquake occurrence takes the form

log N = a - bM

Both parameters a and b are determined empirically by a statistical fit to data in seismicity catalogs. The relative frequency of large and small earthquakes is described by the parameter b, sometimes called the b-value. The parameter a scales with the total number of earthquakes. For worldwide seismicity, a is roughly 8 and b =1 to a good approximation. Using these values, one can predict N=100 earthquakes per year worldwide with Richter magnitude M > 6.0, and N=10 earthquakes per year worldwide with Richter magnitude M > 7.0. The analogy with the theory of fractals can be visualized by thinking about how a population of earthquakes varies with the b-value. As the b-value increases, larger earthquakes become less numerous, and fault motion at plate boundaries becomes "smoother," with fewer large jumps. As the b-value decreases, larger earthquakes become relatively more numerous, and fault motion becomes "rougher." Special-effects animators exploit a similar power-law behavior to simulate smooth or jagged background topography in computer-generated images.

If earthquakes are self-similar, the physical laws that govern their behavior can be studied in the laboratory in experiments, or with simple friction theory. The simplest of these theories is the elastic dislocation model, proposed by H. R. Reid for the 1906 San Francisco earthquake. It argues that rocks along a fault can accumulate strain until a predictable failure threshhold is reached. At this point, an earthquake occurs. This type of behavior is observed in fracture and friction experiments on laboratory rock samples. Elastic dislocation theory can be used to predict the timing of large earthquakes if plate- tectonic motions are steady and constant. One expects earthquakes on an isolated fault to occur cyclically, with a predictable time interval between shocks as strains accumulate towards the fracture threshhold. All one needs to measure is the recurrence interval between large earthquakes.

Reid's model underpinned earthquake-prediction and forecasting efforts in the United States in the 1970's and 1980's. Researchers had only a few areas where a recurrence interval could be estimated with confidence from historical records. M>7 earthquakes on a single fault are typically spaced by a century or more, and historical records are sparse in places like the Western US. A seismic gap is identified as a portion of a fault that has not broken in a long time. Earthquake hazard in a seismic gap is assumed to increase as the time since the last earthquake increases. The segment of the San Andreas Fault that slipped in the 1989 Loma Prieta earthquake was identified by the US Geological Survey as a seismic gap. Other seismic gaps in California, some more prominent than the Loma Prieta segment, have not suffered earthquakes. The recurrence interval of earthquakes is unknown for most of these, which seriously limits the use of seismic gaps as a prediction tool. If only the timing of the last large earthquake in a region is known, seismic gaps are more useful for long-term forecasting of earthquake hazard.

In the late 1970's, the first careful study of disturbed soils near the San Andreas Fault north of Los Angeles, California suggested a cyclic sequence of large earthquakes with an irregular spacing of 150-250 years. This fault segment last ruptured in the 1857 Fort Tejon earthquake with estimated magnitude M=8.3. The estimated recurrence interval suggested increased hazard towards the end of the 20th century. Earthquake prediction efforts focussed on the place where the 1857 quake had initiated, near Parkfield, California. A short segment of the San Andreas Fault near Parkfield had suffered M=6 earthquakes with striking regularity: 1857, 1881, 1901, 1922, 1934, and 1966 -- an approximate recurrence interval of 22 years. The apparent agreement with simple elastic dislocation theory encouraged seismologists to focus attention on the Parkfield fault segment. The next M=6 earthquake was thought most likely to occur near 1988, and scientists monitored the fault segment intensively in a 1984-1993 prediction window. At the end of 1993, however, no earthquake larger than M=5 had occurred on the segment.

Three factors are likely to interfere with the cyclic earthquakes predicted by simple elastic dislocation theory. The first factor is fault interaction. Single faults cannot be treated independently of neighboring faults. After a fault slips during an earthquake, the forces on all neighboring faults are affected. Seismologists have documented cases where aftershocks of moderate and large earthquakes tend to cluster in areas where theory predicts that stresses have been increased by the main shock, and are scarcer in regions where stresses have been decreased. The 1994 Northridge earthquake in Los Angeles occurred where stresses had been increased by the nearby 1971 San Fernando earthquake. In simple computer models, the interactions between different faults, or different segments of the same fault, lead to unpredictable recurrence intervals between large earthquakes. Interactions with a single fault are also likely. Many faults are observed to concentrate stress release at asperities, patches of the fault whose rupture typically triggers slip in a wider area.

A second factor that complicates elastic dislocation theory is velocity-friction behavior. In the laboratory, some materials suffer velocity weakening, that is, frictional resistance to sliding decreases as the sliding velocity increases. Other materials exhibit velocity strengthening properties, in which frictional resistance increases as velocity increases. If velocity-weakening material lines the faces of a fault zone, repeated stick-slip motion occurs, as the fault offers little resistance to sliding once slip begins. If velocity strengthening material lines the faces of the fault zone, sliding will occur in the form of steady, slow creep. If mixture of rocks with these properties lines a fault zone, slow creep on velocity-strengthening patches gradually loads the velocity-weakening patches to the point of failure, in turn generating more creep. Computer models of such systems predict complex chaotic behavior, in which the largest earthquakes occur without stable recurrence intervals. On the positive side, models of this kind predict behavior that resembles observed fault motion e.g. self-similarity in earthquake sizes. However, these models suggest that seismologists would need to know many details about the rocks and their interactions within a fault zone in order to predict where and when the next large earthquake is to occur.

The third factor is fluid-fault interaction. Earthquake faults form a network of pathways for water, carbon dioxide and other volatile compounds in the brittle upper crust. There are three major sources of these volatiles: 1) rainwater, which percolates downward through surface fractures and porous rock, 2) mantle outgassing, principally as a byproduct of magma migration, eruption and emplacement, and 3) metamorphic dehydration reactions, which reverse the effects of chemical weathering reactions. The presence of fluids in fault zones complicates earthquake forecasting, because the dynamics of large-scale fluid migration and fluid-rock chemical reactions are poorly understood and cannot easily be duplicated in the laboratory. Fluids trapped between the faces of a fault will decrease its resistance to failure, much as a banana peel causes characters to slip and fall in children's cartoons. This occurs because the fluid pushes back against any external force that presses the fault faces against each other. In this manner, fluids weaken fault zones beyond the effects of the pre-existing rock fracture itself, adding to the tendency for rock strain to be concentrated along pre-existing fractures. Fluids in fault zones also lubricate sliding motion, so that frictional heating during the earthquake process is greatly reduced. This lubrication is thought responsible for the heat-flow paradox of the San Andreas Fault in California. If one extrapolates from laboratory data on rock fracture, one expects frictional heating of the rocks surrounding an active fault, leading to a significant increase in the heat flow from subsurface rocks to the atmosphere. No such heat-flow increase is observed along the San Andreas Fault, leading seismologists to conclude that its sliding friction is quite low. Another argument for fluid-filled fault zones comes from observations of aftershock enhancement. In several well-monitored regions after significant earthquakes, small aftershocks appear to be concentrated in places where the main earthquake has increased rock stresses, and less numerous where stresses have been decreased. Since the theoretical stress increases are quite small, it appears that most of the small faults that generate aftershocks were close to failure, consistent with a weakened fluid-filled condition.

If the past history of earthquakes offers only a general guide to the timing and sizes of future earthquakes, seismologists must turn to earthquake precursors, unusual phenomena that precede an earthquake. The list of potential precursors is long. Foreshocks are small earthquakes that precede the main shock, and can offer clues to where stresses on the fault are building towards the breaking point. Several other phenomena may indicate imminent rock failure: unusual strains and tilts of the ground; changes in water level within wells, which suggest that water is being squeezed out of microscopic cracks in compressed rocks, or else sucked into the cracks of stretched rocks; radon gas emissions, released by the opening of cracks in rocks; and electromagnetic signals caused by rock strain or by the motions of saline fluids within the fault zone. Changes in groundwater chemistry and geyser activity have also been documented prior to earthquakes. The use of these phenomena for earthquake prediction suffers from the small number of "controlled experiments" to determine which combination (or combinations) of signals is a reliable indicator of an imminent earthquake. The Parkfield, Calif monitoring effort in 1984-1993 was only partly an attempt to verify elastic dislocation theory. It was intended primarily as a controlled experiment to document the phenomena that immediately precede a significant earthquake. Attempts of this kind will continue, in order to collect the data necessary to determine whether near-term earthquake prediction is possible.

Most reports of precursory earth movements have been based on data from only a handful of instruments, and therefore difficult to cross-check. future progress in the study of earthquake precursors, if any is possible, will require broader data-collection efforts. Because large earthquakes are unpredictable at this time, seismologists will need to be lucky. The US and other nations in earthquake-prone regions are upgrading regional seismographic networks, which monitor small tremors in active fault zones, to take advantage of increased computational and telecommunication facilities. Japan will soon install a nationwide network Global Positioning System (GPS) sensors, numbering in the thousands, to monitor slow earth movements. Researchers in the US are considering similar networks of GPS sensors, which determine absolute location to a precision of millimeters by referencing orbiting satellites. These efforts have been made possible by the mass production of cheap, accurate sensors, and of computer networks capable of handling the immense volumes of scientific data.

Figure 1: How material moves as seismic waves travel through it. A) P waves involve compression of material, which oscillates in the direction the wave travels. B) S waves involve the shearing of material, which oscillates perpendicular to the direction the wave travels. The "P" and "S" stand for "primary" and "secondary" waves, so named because S waves follow P waves on a seismogram. You may also find it useful to remember "P" and "S" as "pressure" and "shear," the types of motion associated with the two wave types.

Figure 2: P and S waves radiated by slippage on a fault. Shear forces are greatest along the plane of the fault, parallel to the slip direction, and perpendicular to the fault. S waves are larger in these directions. Compressional forces are greatest for waves that emanate at 45 degree angles to both the fault and the slip directions. P waves are largest in these directions. The initial motion in the wave, up or down for P waves, sideways right or left for S waves, helps diagnose the slip direction.

Figure 3: (photograph from lecture) The Grand Teton mountain range in northwest Wyoming has been generated by normal earthquakes along a north-south-trending near-vertical fault. Erosional material from the mountains has accumulated on the hanging-wall side of the fault, creating vast meadows between the town of Jackson Hole and the southern entrance to Yellowstone National Park.

Figure 4: US plate boundaries & tectonic provinces

Figure 5: Western US Seismicity

Figure 6: The figure shows the location of a 5 December 1997 earthquake in Kamchatka, Russia (M=7.7) and the 82 seismic stations which contributed data the SPYDER system operated by the Incorporated Research Institutions for Seismology (IRIS). The curved lines connecting the epicenter and stations indicate paths taken by surface waves along the earth's surface. Each path is part of a great circle, the shortest path between two points on a spherical surface.

Figure 7: Cartoon of earthquake precursors.

Boxed Reading 1: Moment-Magnitude relations. In the 1920's Charles Richter of the California Institute of Technology (CalTech) devised a magnitude scale to express earthquake size to newspaper reporters. The number was computed from the logarithm of seismic motion as measured by a standard type of seismometer at CalTech, with corrections applied for the distance between earthquake and seismometer. The Richter magnitude M was (conveniently) a number between 0 and 10 for all the earthquakes encountered by Richter, though in principle it could be applied to larger and smaller quakes. Seismologists worldwide adopted M, calibrating it to different seismometers, different locales, and different seismic waves. For distant earthquakes, Richter magnitudes can be estimated either from body waves (oscillation periods near 1 second) or surface waves (oscillation periods > 20 seconds). For earthquakes with M>6.3, surface-waves give more reliable magnitudes than body waves. The seismic moment Mo was initially measured from surface waves with 200-second period, but combines the amplitudes of P-waves, S-waves and surface waves with computer models of how waves propagate through the earth. It has been a standard measure of earthquake size starting in the 1980's. Hiroo Kanamori of CalTech devised a formula to calibrate M and Mo in 1977:

log Mo = 1.5M + 9.1

where Mo has the units of newton-m = kg-m^2/sec^2.

Boxed Reading 2) Real time seismology with global networks. Modern seismometers record ground motion in digital form, that is, as numbers that represent the motion of the seismometer at evenly-spaced intevals of time. Since all data is stored in electronic format rather than on paper or film, it can be rapidly transferred over computer networks. Once assembled in a single location, seismic data can be processed to estimate the location, size and faulting style of an earthquake. Most processing is done automatically by computer software, whether for a small tremor recorded by a regional networks near Los Angeles CA or a M=7 earthquake in the Fiji Islands, picked up by a global network. In most cases a large earthquake anywhere in the world can be characterized within a few hours, and help to direct rapid-response disaster relief. Many of these systems are publicly accessible via the Internet. After any earthquake with M>5.5, the SPYDER system of the IRIS Data Management Center in Seattle, WA interrogates seismometers around the world to obtain records of seismic motion (Figure 6). You can reach the SPYDER system though the IRIS web page:

http://www.iris.edu

From the IRIS site, you can connect to many other sources of earthquake information. For instance, a list of the last 10-20 significant earthquakes worldwide can be obtained by a network query to the US Geological Survey:

finger quake@gldfs.cr.usgs.gov

Main References:

T. Lay and T. C. Wallace, Modern Global Seismology, Academic Press, 1995.

C. H. Scholz, The Mechanics of Earthquakes and Faulting, Cambridge Press, 1990.

C. H. Scholz, Earthquakes and friction laws, Nature, v391, 37-42, 1998.

B. Bolt, Earthquakes, Freeman, 1988.

E. Roeloffs, The earthquake prediction experiment at Parkfield California, Reviews of Geophysics, v32, 315-336, 1994.

footnote: Incorporated Research Institutions for Seismology, a consortium of universities with research programs in seismology.

back to global change home page