SPECIAL ISSUE

MODELING AND EARTH STRUCTURES

Seismology has burgeoned into a modern science-force-fed by federal fund- ing to advance technology for detecting underground nuclear explosions and predicting earthquakes, and by industry to improve tools for gas and oil exploration. Computers, seismic instrument systems, telemetry, and data reduction have played key roles in this growth.

MICHAEL R. RAUGH

In conjunction with modern technology, the sophisti- FOREWARNED IS FOREARMED: A cated methods of modern seismology now enable earth HYPOTHETICAL CATASTROPHE IN CALIFORNIA scientists to peer into an opaque earth. The purpose of Many of California’s most attractive features are due to this article is to discuss some of those tools and the the same tectonic processes that we generally consider information they reveal about earth processes and destructive. A geological time-lapse film, beginning structures. The subject is broad; hence the focus is nar- around 140 million years ago, would show the Pacific rowed to several studies centered in three regions of Ocean lapping against the western slope of the ances- California, the best-studied sei.smically active state in tral High Sierras, a softly undulating range of low-lying the nation. mountains not at all like the lofty granite peaks of to- The article begins with a sobering discussion of the day. As millions of years roll by, leading edges of the potential impact of a major quake in California. Indeed, Farallon and Pacific plates would bump and jostle off- a disastrous shock with severe human and economic shore against the western edge of the North American consequences is expected within the next generation, plate, causing subduction of oceanic plates and crum- but good planning c:ould alleviate some of the hardship. pling and upward folding of the marginal terrain, grad- Next, for historical perspective, a few selected mile- ually extending the continental plate by aggregation of stones are sketched to show how rapidly seismology island arcs and new mountains arising out of the sea to has advanced in the past 25 years. In order to provide a the west. The sea would advance and retreat repeat- technical perspective, the two most extensive seismic edly, but ultimately retreat farther and farther to the instrumentation systems in California are described, il- west. We would see volcanoes exploding at various lustrating a complex system of instrumentation, teleme- places on land and offshore, and massive lava flows try, data reduction, and computers. Finally, a sampler rushing out over the ancestral Sierras. At stages, semi- of seismic studies that would have been inconceivable tropical forests would develop in central California before the present era of instrumentation and computer around the marshy edges of the Great Valley leaving systems is presented, centerin,g on three regions of Cali- coal deposits, then an ice age and glaciers would appear fornia (Figure 1): the Imperial in the Salton scouring the High Sierras rising ever higher to the ac- Trough (site of the best-documented in his- companiment of violent earthquakes. We would see the tory), the Long Valley Caldera east of the High Sierras final draining of sea water from the Great Valley, the (where a potentially explosive volcano threatens to cut carving of the Golden Gate, and the encirclement of off Owens Valley water to ), and the Big San Francisco Bay. What we would not see is the swal- Bend (a puzzling change in direction of the San An- lowing of the Farallon plate beneath the shores of dreas Fault through the Transverse Ranges of southern southern and central California, an invisible process California]. that continues today north of Cape Mendocino. But, ACM/IEEE-CS Joint Issue8 1985 ACM OCIO~-0782/65/1100-1130 750 when the present reached California,

1130 Communications of the ACM November 1985 Volume 28 Number II Special issue

FIGURE1. Mapof Californiaand SurroundingLandscape Showing SomeLocaffons and Geological FeaturesReferred to in the Text

The light background north of the Transverse Ranges approximates the region instrumented by the USGS California Seismological Network (CALNET), and the darker back- ground to the south approximates the region instrumented by the Array, operated in cooperation by the USGS and Cal- tech. Owens Lake at the southern end of Owens Valley, designated in the figure as a drainage basin, was until early in the century a large body of alkaline water more caustic than Mono Lake; it has been reduced to an extensive alkali flat by diversion of the Owens River into the Los Ange- les Aqueduct.

November 1985 Volume 28 Number 11 Communications of the ACM 1131 Special Issue

shoving the Farallon plate beneath the continent some The most famous earthquake in American history 20 million years ago, the subduction of oceanic plates was that of 1906 in San Francisco. Two other California changed to horizontal slip south of Cape Mendocino, earthquakes at least as large also occurred within his- creating the highly visible system. In torical times, but are less well known because less the final sequence, we would see the edge of California damage was associated with them. The great quake of west of the San Andreas gliding briskly northward (see 1857, sometimes referred to as the Fort Tejon earth- Figure 2). The film would conclude with dramatic quake, produced approximately 350 km of faulting all views of the major landmarks of California, a region the way from Parkfield in Monterey county southeast justly famous for its natural resources: mountains, through Fort Tejon in the Tehachapis to Cajon Pass in beaches, deserts, lakes and rivers, abundance of min- San Bernardino county. In fact, some regions of the erals, and fertile plains. California’s beauty and wealth Tejon fault slipped as much as 11 m, compared to 7 m are due to tectonic processes. Yet even today these for the earthquake of 1906. The Owens Valley earth- same relentless processes also cause the catastrophes quake of 1872, centered near Lone Pine, may have been associated with earthquakes and volcanoes. the largest of all. In this century, there have been

Generation of Generation of Ridge axis Volcanic (source) oceanic crust oceanic crust chain and lithosphere / - and lithosphere \

According to plate , the eisrth’s lithosphere dstrong dgntinent, creating an ocean trench associated with the line outer layer) is divided into about a do?en large oontiguous where Su~udtiotif3agins. As the ocean plate descends into plates, each of which is in slow motii corresponding to thehot &ihtlebf t#e earth, ii partly mefts, and tighter por- convection occurring within the mantle beneath We plates tir$rs float back uptoward the surface, causing volcanic flowing plastically at a rate measured in centimeters per year. eruptiOns and the uplift of mountains. It is a matter of debate Earthquakes occur when one plate scrapes against enother. whether convection drives the plates, or gravitational forces In special areas of the earth where plates diverge, c@led dsSvethe conveotion by pulling down upon the uplifted ocean midocean ridges, molten material (magma) rises conve&yely tic&es @k& push*) and pulling the descending slabs down from the earth’s mantle and fills the gap between the diverg- (aslabpull”). Although easy to grasp in principle, the modem ing plates, creating tractions that pull the ocean ridge apart. theoty of plate teqtonics was not easy to perceive and verffy. Fresh magma flows into the cracks, where it adheres and GaologikM processes are notoriously slow, and the global solidifies within the ridge. When an oceanic and conlinental s patterns in which they operate become apparent only when plate converge (as at the western boundary of South Amer- abseWed systematically on a, worldwide basis. (After Dewey, ica), the denser ocean floor is driven (subducted) beneath the J.P. Plaita tectonics- Sci. Am. (1972).)

FIGURE2.

1132 Communications of the ACM November 1985 Volume 28 Number 11 Special Issue smaller destructive earthquakes in Santa Barbara, Long capability may be incorporated as they become avail- Beach, Tehachapi, San Fernando, Imperial Valley, and able, without adverse affect on the functioning of the Coalinga. overall system. Based on the observed regularity of earthquakes and Initial scenarios are planned for developed regions in other structural factors of known California fault sys- California where high probabilities exist for large-to- tems, earth scientists expect an earthquake comparable great-magnitude earthquakes in the near future. Pre- to that of 1906 to strike somewhere, probably in south- liminary studies have yielded ominous results. ern California, within the next 30 years. A major earth- Steinbrugge and others (1981) found that direct prop- quake near Los Angeles or Silicon Valley, on the scale erty losses for a magnitude 7+ shock centered in highly of 1906, would have a terrible impact on the economy developed regions of Los Angeles or in the San Fran- of California and on the nation. Such an earthquake is cisco Bay region would be in the $40-60 billion (1980) inevitable; the only question is when. Although earth- range, an order of magnitude more damage than the quakes cannot be prevented, they can be anticipated worst hurricane in U.S. history (Hurricane Agnes, and planned for. 1969). But that may be just the tip of the iceberg. What The Federal Emergency Management Agency has not been included are estimates of the indirect (FEMA) is responsible for developing contingency plans losses, for example, unemployment and loss of supplies for responding to and planning the orderly recovery and services due to shutdown of factories, disruption of from an earthquake disaster as well as creating emer- lifelines such as highways, water, and electric-power gency policies in anticipation of such a disaster. Key communications and sanitation, disruption of banking, to the planning process is a modular system of compu- and loss of the tax base. According to a pilot study by ter programs being developed to provide rapid and Bob Wilson of FEMA to assessthe long-term economic comprehensive analyses of earthquake impact for any impact, a major quake in the Palmdale bulge region designated earthquake epicenter and magnitude in near Los Angeles would result in indirect losses of 7 or California. more times the direct property losses. If this indirect-to- The programs specify the region affected by the direct loss ratio were to hold for a magnitude 7+ in Los earthquake and, for each commercial sector, estimate Angeles or the San Francisco Bay area, the total loss direct and indirect losses including (1) the physical would be staggering: as much as $1 trillion or more. damage caused by ground shaking; (2) the percent of And that says nothing of the loss of human life. It is loss of function including the time required to restore obvious from such studies that plans for preassessing facilities to preearthquake usability; (3) the loss ex- and mitigating earthquake disasters should receive high pected to arise from collateral hazards such as inunda- national priority (see Figure 3). tion, ground failure, and fire; and (4) the number of injuries and fatalities. A TOUCH OF HISTORY To assessthe impact of a disaster on the U.S. econ- The seeds of the modern age of computational seismol- omy, a joint supply-side/demand-side economic impact ogy were planted on September 2, 1959. That is the model was developed by the Engineering Economics date when the Advanced Research Projects Agency Associates of Berkeley and is now being implemented (ARPA) began funding research into seismic detection by FEMA. It measures the economic flow between in- of underground nuclear explosions. An early ARPA ob- dustries and various commercial sectors within the jective was to stimulate the field of seismology across a United States, as affected by the region of impact. The broad range of topics (including improvements in seis- model is based on a classification known as the Stand- mic instrumentation and intensified studies of large ard Industrial Classification of the U.S. Input-Output earthquakes recorded worldwide, microearthquakes re- Table Update for 1977. corded by local networks, and earth crustal structure There are three major input components: (1) a based on active experiments with explosive sources) ground-shaking model, provided by the USGS, to deter- and institutions (including universities, governmental mine the geographic distribution of seismic intensity research agencies, and private research corporations). resulting from a given location and extent of faulting; When ARPA began to channel its support into studies (2) data on the distribution of man-made structures, that held the most promise of solving its limited objec- including their value, economic functions, and the tive of nuclear test detection, other agencies adopted number of occupants as a function of time of day; and promising elements of the program that supported their (3) an engineering model, developed by the Applied missions: for example, the U.S. Geological Survey for Technology Council of Palo Alto, that gives the degree earthquake-hazard reduction (including earthquake of damage to each structure type as a function of seis- prediction) and the National Science Foundation for en- mic intensity and estimates the time required to fully gineering seismology and seismological research in sup- recover the function of the damaged facilities. port of graduate training in seismology. These programs Thus every engineering structure, every economic have continued to interact strongly with each other and activity housed by every structure, and every compo- with parallel research programs in the petroleum ex- nent of regular dollar flow in California are classified ploration industry on the critical problems of instru- and accounted for by the FEMA system. Because the mentation and data reduction and analysis. All have system is modular, improvements in data and modeling received invaluable ongoing stimulation from develop-

November 1985 Volume 28 Number 71 Communications of the ACM 1133 Special Issue

Station C Earthquake locations are determined S by using basic facts about elastic wave propagation. An earthquake Statton B / \ radiates energy in the form of \ P-waves, which are longitudinal vi- Curves show how travel tlrnes of brations like sound, S-waves, which Secsmogram for P-and S-waves are transverse vibrations that occur station A only in rigid material, and surface increase with waves, which are confined, like distance a; t E ocean waves, to the surface of the ; earth and do not travel deep within. The P-waves travel some 73 percent faster than the S-waves and arrive at surface kxations first. Because the earth’s elastic properties vary with depth (and to a lesser degree with Travel lime latitude and longitude), P- and of P-wave from S-waves are refracted (bent) as they earthquake to move out from the canter of the station C quake toward distant points. For most practical purposes, geometric ray-tracing techniques may be used to plot the course of a P-wave X4 istance through the earth from the earth- -X B quake source to a point on the sur- . * xc face. An integration along the ray path, taking into account the vekxity of the wave at each point, determines the travel time. The earthquake, only wave an&al times are needed. The wave- time and location of an earthquake may be estimated either form itself is of no concem, If more information is required, by observing P-wave arrivals timed at several (@tfeast four) such as the direction of slip and magnitude of the earth- seismographic stations, and finding the point in the earth quake, then more of the waveform must be considered, whose ray-path travel times to each seismograph are most namely, the polarity of the wave and the maximum ampli- consistent with the observed data, or by observing separa- tude. (From Press, F., and S!ever, FL Earth. W.H. Freeman tions between P-wave and S-wave arrivals at three seismo- and Co., San Francisco, Calif., 1982.) graphic stations (see figure). Note that, for location of an

FIGURE3. Locating Earthquakes

ments in microelectronics and c:omputers that have ephone lines or microwave. The instruments them- permitted simpler, more effective solutions to old data- selves are highly varied. There are vertical component collection and analysis problems and have opened pos- analog seismometers, x-y-z component digitizing accel- sibilities for comprehensive automatic data-collection erometers, and mechanical devices that aim a wiggling and analysis systems that were hardly dreamed of a light beam at an advancing film. Some historical exam- decade ago. ples of telemetered networks are mentioned in the his- The introduction of computers in the early 1960s for tory sidebar. locating earthquakes was an important step forward, An exciting new application of computers to seismol- but there were still serious dela;ys in the collection and ogy involves microcomputers to control various compo- reading of seismic data for analysis. Seismograms were nents of data-acquisition systems. Electrical signals gen- recorded locally at the seismometers, and phase read- erated by a seismometer as the ground shakes must be ings or seismograms were sent by mail to a central calibrated, timed, amplified, filtered, and converted to a laboratory for further analysis. Weeks or months were digital signal for processing and analysis on a digital consumed in this process. computer. Microcomputers provide a powerful means Computers without a convenient source of data are of controlling such hardware components. Recently de- like Pegasus without wings. Automated instrumenta- veloped data-acquisition systems such as that devel- tion and telemetry systems are necessary adjuncts of oped at the U.S. Geological Survey by R. D. Borcherdt modern seismology. Instrument arrays come in all sizes and his colleagues are designed to isolate system func- and shapes, from a few cassette recorders that can be tions on a uniform set of hardware modules via a gen- deployed easily at the scene of a disaster to track after- eral computer bus. This design approach has permitted shocks, to regional networks of permanently installed the development of a General Earthquake Observation seismometers that transmit signals continuously via tel- System (GEOS) that is easily adapted to a wide variety

1134 Communications of the ACM November 1985 Volume 28 Number 11 Special Issue of seismic experiments including seismic refraction acquisition systems that within the last two years have studies, near-source strong motion studies, and tele- yielded data sets of unprecedented quality on experi- seismic studies for the purpose of nuclear detection. ments ranging from strong motion studies with ground Isolation of system functions on separate modules per- motions as high as 1 g in acceleration to earth strain mits straightforward adaptation of the system to im- studies with motions as small as 5 x lo-” strain units. provements in technology. This application of micro- As a result of advances of this type, the Committee on computers has resulted in construction of data- Opportunities for Research in the Geological Sciences

RANDOMHISTORICAL REFERENCE POINTS 1928: Seismological Method of Determining Earthquake marily for the detection of short-period body waves from Source Mechanism teleseismic sources. The process was carried out digitally. Perry Byerly, at UC Berkeley, formulated the seismological 1967: Empirical Demonstration of Planar-Fault Surface method of estimating fault planes from distant recordings, In the fall of 1965, the USGS began a detailed study of the enabling scientists to determine fault motions and stress San Andreas Fault in central California. The first seismic patterns at distant quake locations. The method provides a station was set up at Gold Hill, near Parkfield. In June 1966, mathematical foundation for modern calculations of earth- a magnitude 5+ earthquake near Parkfield produced a zone quake fault mechanisms by computers. of ground rupture that split right past the USGS seismic 1949: Remarkable Early Effort in Radio Telemetry station at Gold Hill. Following the quake, eight new low- P. S. Gane led a long-term effort to study rock bursts in power portable stations were targeted on the southern part deep lode mines of Witwatersrand, South Africa. Data from of the 1966 zone of surface rupture in an effort to map the local analog instruments, telemetered over distances of sev- fault rupture surface underground by the small aftershocks eral kilometers to a central analysis station, were used to that might occur along it. The experiment, directed by seis- Iocate spontaneous explosions of distressed rocks in deep mologist Jerry Eaton, was the first in which locations were lodes. computed accurately enough to confirm that earthquakes result from slippage of the earth along a planar fault. 1958: Telemetered Networks Continuous telemetry of analog seismic data from the field 1970-: Large-Scale Dense Regional Networks to a central recording site was first introduced as a means Large regiona telemetry networks were developed under of improving network performance and accelerating tradi- the Earthquakes Hazard Reduction Program and its close tional analysis of the records. By 1958 a four-station teleme- antecedents during the 1970s. The number of telemetered tered network was operated by the USGS around the sum- stations in the USGS network (Calnet) mit of Kilauea Volcano in Hawaii; and substantial parts of and the cooperative USGS-CIT southern California array sparse regional networks operated by UC Berkeley in rose to 189 and 94 respectively by 1975, and to 303 and 211 northern California and Caltech in southern California respectively by 1982. Data from these large dense networks were on telemetry by 1961 and 1967 respectively. overwhelmed traditional methods of analysis and forced the development of more effective, computer-based meth- 1980: Computer Program to Locate Earthquakes ods for recording and analyzing the data. Bruce Bolt, now director of UC Berkeley’s Seismographic The first large-scale effort to develop automatic process- Stations, in 1961 reported the use of a computer for carry- ing for a dense regional network was carried out on the ing out the calculations required for locating distant earth- northern California telemetered network by Sam Stewart in quakes, using procedures similar to ones described by the early 1970s. Stewart’s system employed a CDC-1700 to Geiger in 1910. At about the same time, J. M. Nordquist at analyze 110 stations in real time. Detection and timing of Caltech developed a least-squares program for applications events, collation of data from stations across the network to to local earthquakes: Caltech locations from 1961 to the distinguish earthquakes from noise, and calculation of the present have used this program. earthquake hypocenters were carried out on this one ma- 1963: Seismically Delineated Tectonic Plate Boundaries chine. Though successful as a prototype, this system was Lynn Sykes at Lamont and Robert Engdahl at National not sufficiently flexible and economicaf to expand to cover Oceanic and Atmospheric Administration were quick to the’much larger seismic network that then existed in apply computers to accurately locate thousands of oceanic northern California. earthquakes around the globe. Their accurate delineation 1982: First Regional Digitally Telemetered of faults beneath the seas drew early attention to the im- Microearthquake Array portance of midocean ridges and continental trenches as The Anza array monitors the San Jacinto fault region in tectonic plate boundaries. southern California. It comprises 10 three-component seis- 1965: Large-Scale Computer-Based Seismic Array mometers that continuously transmit digital signals via mi- The first large-scale application of computer technique in crowave to a seismological laboratory at UCSD’s Scripps the recording and analysis of large seismic networks was Institution in La Jolla. Seismic signals are recorded with a deveIoped in the nuclear test detection research program broader frequency bandwidth and dynamic range many for the LASA (Large Aperture Seismic Array) network in times that possible in an analog system. The goal at Anza is Montana. LASA was a roughly circular 300-km-wide array to use the entire waveform to unravel the source properties of 300 to 500 seismometers arranged in 21 subclusters of 16 of each earthquake and synthesize them into a composite to 25 seismometers each. The LASA network was used pri- picture of the evolving forces that drive the fault.

November 7985 Volume 28 Number II Communicafions of the ACM 1135 Special Issue

WHAT IS SE Seismic waves bend, reflect, scatter, slow down, speed up? and fade in response to the media through which they propagate. Although such behavior seriously distorts earth the earth and sourcehypotheses, A mm&son af ihe sim- imagery and complicates analysis, it is this very sensitivity ulatedresponse and tIie empi&aIIy mqpqvredground re- of seismic waves to inhomogeneities in the earth that sponseis made to supportor reject the hypotheses. makes them carriers of valuable hints of the materials and Nonlinearinverse methods assume @at the earth struc- structures through which they have traveled. ture and seismicsource under sttidy are reasonablywell Wave probes are produced randomly by earthquakes and known,so that correctionsta the a p&i modelmay bq volcanoes, or purposefully by explosionsor shakingma- determinedfrom the databy lineart&&r& theory.Al- chines(as in gasand oil exploration).To detectthem, net- thoughinverse methods are elegant in theory and &ici& worksof seismicinstruments serve as sensors to measure computationally,they often reqQiremore d pr&ri know& and recordground motion at selectpoints of the earth’s edgeof the earth than do the forwardmethcads, surface.The datamay be eithercollected on magnetictape It is interestingto note that in many s&&s; only the then physicallytransported to a centralprocessing location, wave travel timesare of inter&e The wa+&qm &elf is or telemetereddirectly via phonelines or microwave,for discarded.When more inf@rm&ion Is required,r&&v&y reductionand computeranalysis. moreof the wavemay be used.In same Bdvancedstudies The primary toolsof analysisare forwrd mefhods and (asin Archuleia’sstudy reviewedbeIaw), in which fine- inverse methods: Forwardmethods &mpute a simulatedre- graindetails of earth struetumoridt rupture ares&&, sponseof the earth to a givenwave source,based on hy. muchof the waveformmay be sllmulaturd, pothesesabout elastic properties of the interveningearth It isclear that computersare necessary:tsicarry but ti andrupture propertiesof the seismicsource. To a first massivecalculations required by e&rnoI~gy.But, equally approximation,the lawsgoverning P- and S-wavepropaga- important,seismic imaging is an gnderdetemlned probkrn, tion are very similarto the lawsof geometricoptics, with which meansthat similarresutts could be aichievsld ,)y the P- and S-wavevelocities taking the placeof the index morethan oneset of hypothesesf&h@ earth and rupture. of refractionin optics.Hence, usually, the methodsof geo- Hencegreat amounts of accuratti~~~~‘~~~~~~~ ta con? metric opticsmay be used.In moreexacting studies, the strainthe possibilitieswithIn ra~s~~~~~~c&da Tti ?a elasticwave equation may be usedto modelpropagation why automatedinstrument&m syst&& h&e b&&e sb’’ moreaccurately, usually requiring a greatdeal of computa- importantin modernseism&gy.

(1983)has recommendedthe development of a national Menlo Park, where signalsare immediately digitized, instrumentation resource for the earth-science commu- earthquake P-waves are picked automatically, and nity, comprising as many as 1000 instruments for earthquake locations and magnitudes are computed studying the continental lithosphere, and expansion and displayed within minutes. Calnet covers the terrain and improvement of the worldwide network. The mi- north of the Transverse Rangesand is concentrated in crocomputer offers the possibility of developing hard- regions of high seismicity. During typical earthquake ware modulesthat can be configured in a variety of swarm activity, such as occurred at Mammoth Lakes ways to serve theseas well as most seismicdata- (seebelow) or at Coalinga, a’8many as 1200 to 1400 acquisition needs. events may be plotted in a single hour. The‘\primary purposeof Calnet is to monitor poten- TWO SEISMIC NETWORKS: CALNET AND tially hazardous situations. Accordingly, emphasisis THE SOUTHERN CALIFORNIA ARRAY placed on efficient real-time detection and graphics dis- Before moving on to a sampleof seismicstudies cen- play of seismicactivity at any designatedregion tered in the three California regions,we will briefly throughout the net. RTP savesonly an abstract of each describe the two largest regional seismicnetworks in earthquake event for archival storage;waveforms are California: the USGS Northern California Seismological discardedand cannot be used for subsequentanalysis. Network, known as Calnet, and the Southern California Built by Rex Allen and Jim Ellis (beginning in 1977), Array. Most of the instruments in these two networks the core of RTP is a homegrown parallel processing are inexpensive single-component(vertical) seismome- system of 33 16-bit TI-9900 basedmicroprocessor sub- ters that transmit analog signalscontinuously via phone systems.With one exception, each of the subsystems lines to their respective data-reduction centers: the processesthe data from eight instruments. Significant USGS in Menlo Park or the Caltech SeismologicalLabo- alterations in the signal from any one instrument trig- ratory in Pasadena.Although neither network exhibits ger deposit of critical data from that instrument into a advanced instrumentation, each is extraordinarily use- ring buffer shared by all of the subsystemswhere it is ful becauseof the large amounts of regional data ob- allowed to remain for three minutes. The exceptional tained and the automated data-reduction techniques subsystemacts as an associator. When a fresh record has employed. been placed in the ring buffer, the associatorsearches Calnet currently usesapproximately 400 instru- through the ring buffer for other records that correlate ments, 264 of which feed signalscontinuously by phone in time and spacewith the fresh one. If there is a lines to the RTP (real-time picker) computer system in sufficient number of simultaneous events at neighbor-

1136 Communications of the ACM November 1985 Volume 28 Number 11 Special issue

ing seismometers, RTP signals that an earthquake has to the Naval Weapons Test Range at China Lake in occurred and passes the relevant reports to a minicom- southern California. puter (PDP-II/~@ for computing the location and plot- The Southern California Array, operated in partner- ting the earthquake in the context of a regional image ship by the USGS and Caltech, includes approximately (see Figure 4). 250 instruments, distributed fairly evenly throughout In the days before RTP, when locating earthquakes southern California from the southern Sierras to the required manual intervention, days, weeks, even Mexican border. Analog data are transmitted continu- months could be lost before locations of earthquake ously into the CUSP data-reduction and storage system swarms were plotted and interpreted. Now scientists at Caltech. Unlike RTP, CUSP was designed for effi- need only wait for an automatic, digitized-audio phone cient storage of ground-motion data for subsequent call from RTP to be advised within minutes that signifi- analysis rather than for real-time display: The success cant activity is taking place. A smaller version of RTP of CUSP can be seen in the fact that the USGS, with an has been installed at Caltech’s Seismological Laboratory operational staff of four, is now recording and catalog- in Pasadena to monitor swarms that are occurring in ing 25,000 earthquakes per year, most of a magnitude the vicinity of the dormant Coso volcanic field adjacent less than 2.0 in southern California. Compared to pre-

37.

37,

37’.

37 .6

-119 -118.9 -118.8 -118.7

RTP uses graphics terminals to display Calnet seismic activ- 24 hours, blue for older than 24 hours). Magnitude is indi- ity. At any time any terminal may be used to display seismic- cated by the proportionate size of the circle. RTP is coupled ity on any subnet. A map of the region is displayed with to an automatic phone dialing system to advise scientists grean traces delineating known faults. Events are depicted when an earthquake of magnitude 4.5 or greater is detected. by circles on a background map of the region, appropriately (Data provided by Rob Cockerham and Rick Lester, U.S. cdored to indicate how recently they occurred (red for within Geowical Survey, 1985.)

FIGURE4. RTP Display of Long Valley

November 1985 Volume 28 Number 11 Communications of the ACM 1137 Special Issue

CUSP productivity, this amounts to a tenfold increase are executed through well-designed graphics tools in the number of events and a fourfold increase in the (see Figure 5). number of stations processed, while significantly in- The CUSP design is predicated on a novel combina- creasing accuracy and reducing overall staffing require- tion of a process sequencer coupled with a relational ments. In the design of CUSP, considerable thought database in such a manner that the body of earthquake was given to automating just those aspects of data re- data is “self-evolving.” Events enter the system in an duction that are best done by computers. Interactions arbitrary initial state and are carried through required

PIDB

UUUBO

CBK 17

Jut 20

VA0 21

BRG 31

PI0DSE3 HQT 10

An operator sitting at a computerterminal observes a display PIODSE3, which describe the onset of seismic energy. The like that in the fqure and, if necessary, may correct automatl- two time codes, WWVB and TIME, represent an international caky generated picks of seismic wave arrival times at select standard and a locally generated observatory time, respac- seismic stations indicated by three-letter codes on the left- tively. The seismograms have been automatically shied into hand side. The topmost window of the figure, called the alignment by a theoreticat calculation of P-wave arrivals precision timing window, provides an expanded view of the based on the epkxmtrai distances displayed in kilometers indicated portion of a seismogram, enabling the operator to following the corresponding station names on the left; misa- verify or improve the accuracy of the pick. The indivfcfual lignment would indicate an erroneous pick. picks are labeled by short character SttingS, such as PfOUor

FIGURE5. Interactive Display of CUSP in Event-Timing Mode

1138 Communications of the ACM November 1985 Volume 28 Number 11 Special Issue processing using a sequence of state transitions result- 6). For the first time, strong ground shaking in the near ing eventually in their inclusion in a research database. source region of a large-magnitude (6.5) earthquake was This approach is particularly appropriate for the reduc- well documented. The first recordings from a dense tion of seismic network data since it provides the flexi- differential array were used for studying fault-rupture bility required to deal with various analysis require- dynamics (Spudich and associates), and an instru- ments that vary spatially and with earthquake magni- mented building provided the first well-documented tude. The CUSP paradigm stands in sharp contrast to observation of failure of a modern engineering struc- earlier methods (sometimes referred to as the cattle ture. In addition, the Geological Survey maintained 18 prod approach) in which all discrete steps in the analy- accelerographs that surrounded the most-affected re- sis must be individually initiated through some human gion north of the epicenter. Each instrument recorded intervention. three orthogonal components of motion: one vertical Carl Johnson, while a graduate student at Caltech in and two horizontal. Thus complete timed records of 1975, began developing CUSP in order to study a large ground motion were available. Moreover, the region ensemble of small-magnitude earthquakes in the Impe- previously had been the subject of an extensive seismic rial Valley. Now working for the USGS in Pasadena, refraction survey (see seismic-refraction-study sidebar), Johnson has been improving the system and adapting it which provided a detailed velocity structure that made to current technology. A CUSP system with a capacity accurate analysis possible. of about 500 stations and a sample rate of 100 Hz has The time was ripe for studying the temporal and spa- been set up in Menlo Park on Calnet. As in the south, tial evolution of the faulting process. The information the system will be used to prepare accurate earthquake to be determined included (1) precise delineation of the catalogs and to preserve digital archives of both re- fault surface: (2) the amount of slip at each point on the duced data and seismic traces for further study. fault: (3) the direction, rate, and duration of slippage for A new CUSP system was installed this year at the each point; and (4) the speed at which the fault rupture USGS Hawaiian Volcanic Observatory on the island of (crack tip) advanced-that is, the rate at which the slip Hawaii. This system, comprising two VAX-11/780s region spread out from the initial point of rupture. (VMS operating system) installed at the summit of Ki- Several outstanding studies, notably those of Olson lauea volcano, provides the additional capability of and Apsel, and Hartzel and Heaton, have analyzed picking and locating volcanic earthquakes as they occur these issues using inverse methods. Here we discuss and has resulted in a z&fold increase in analysis pro- another method by Ralph Archuleta of UC Santa Bar- ductivity. A unique aspect of the Hawaiian system is bara, which uses a forward method and illustrates the the provision for volcanologists to tap into the real-time use of synthetics. We present Archuleta’s work because data stream for tremor analysis and microearthquake the detailed results and straightforward analysis yield counting, without disrupting the routine real-time pro- direct insight into the inner workings of an earthquake. cessing. Although Archuleta’s method was straightforward, the We turn now to an exhibit of seismic studies con- results were not achieved without cost; the large num- ducted recently in three regions of California. The un- ber of iterations and extensive computation required derlying theme is that these studies could not have unusual tenacity. been performed without the aid of modern instrumen- First, it was assumed that slip (motion of one side of tation and computer systems and the attendant innova- the fault relative to the other) occurred on a fault sur- tions in data reduction and analysis methods. face, the exact position of which was to be determined. The surface was dissected into small rectangular cells, EARTHQUAKES AS A SOURCE OF and adjustable parameters were assigned to each cell to SUBSURFACE ILLUMINATION define the slip motion of the points within each cell. The Imperial Valley region in southeastern California The parameters included the starting time, direction, has been called the birthplace of the San Andreas fault rate, and duration of slip for each cell. Any complete system. It lies within the , a tectonic rift selection of parameters would constitute a discretized zone, in which a portion of the earth’s crust is stretch- model of the fault plane and rupture behavior. Second, ing and thinning out, creating a valley depression along it was necessary to compute the ground motion that the upper surface and an opposite upward bowing would be caused by such a rupture throughout the sur- along the lower surface. Crustal extension and apparent rounding region, particularly at the location of each intrusions of magma into areas of the lower crust ex- accelerograph; as will be discussed, the technical core plain the presence of hot springs and three small vol- of the work resides in how this was accomplished. canoes at the southern shore of the . The Third, by laborious trial and error, a tentative slip Imperial Fault lies on the boundary between the Pacific model was specified, the theoretical effects of such slip- Plate to the southwest, and the page were computed for each accelerograph location, to the northeast. Geologists have long recognized the and the computed ground motion was compared with region as earthquake country, especially after a magni- the actual ground motion observed during the earth- tude 7.2 that occurred in 1940. quake. The objective was to iteratively readjust all of The best-documented earthquake in history occurred the parameters until a good fit was obtained. After sev- on the Imperial Fault on October 15, 1979 (see Figure eral months and nearly three hundred executions of his

November 1985 Volume 28 Number 11 Communications of the ACM 1139 Special Issue

(a) Archuleta’s Map of imperial Valley Low passed 323” horizontal compo- ARCHULETA: 1979 IMPERIAL VALLEY EARTHQUAKE nent of particle velocity superposed on a map view of the Imperial Valley. 33:Ol (From J. Geophys. Res. 89, B6 (June r&IL ’ ’ ’ ’ ’ ’ ’ .-: 10,19&l), 4563.)

Eoze .

Brawley EO3W fault zone

EOU& A-

” ElIw- r(\ %? ! \

FIGURE6a. Data Used in ArchMa’s Study of the Imperial Valley Earthquake of 1979

program on a VAX-11/780 (VMS operating system), rial Valley, quantitative descriptions of ground shaking Archuleta acquired the insights needed to obtain a at each accelerograph location induced by an impulse solution. located anywhere on the Imperial Fault. By combining The method used by Archuleta to compute ground Green functions to simulate the effects of any given slip response as a function of a given slip model illustrates a model, Archuleta was able to synthesize the response common feature of modern seismology. To accurately of the earth to any one of his fault models. predict wave behavior throug:hout a region, it is first Here are the results: The Imperial Fault plane was necessary to construct a model of local inhomogeneities found to incline 10" from the vertical. Slip occurred to in the earth’s crust that can affect wave motion. The a depth of nearly 13 km and at a distance 35 km north process can be compared to calibrating a lens to com- from the epicenter lying to the south of the border with pensate for the effects of distortion. Typically the re- . Slip was almost entirely horizontal with a quired information is obtained from seismic refraction slight vertical component in sediments at the northern experiments (see sidebar). end. The largest slip rates were confined to depths be- Fuis and associates, using seismic refraction methods, tween 5 and 13 km with maximum speeds of about 1 m had carefully constructed a velocity profile of the Impe- per second, the duration of slip for a particle was less rial Valley, which Archuleta incorporated into a com- than 1.9 s, a period considerably shorter than the rup- puter program to simulate the elastic properties of the ture time (approximately 12 s) for the entire quake. Imperial Valley. The program permitted Archuleta to Rupture speed on the Imperial Fault was highly vari- compute the response of the earth throughout the Im- able, on the average moving slower than S-wave perial Valley to impulse excitations at various cell posi- velocity. tions within his fault model. This was like determining In addition to motion on the Imperial Fault, it was the regional effects of a hammer blow at selected points known from surface measurements that the Brawley on the fault. Thus it was possible to determine (Inde- Fault had slipped. The questions were when (the sur- pendently of the slip model!) Green functions for Impe- face measurements were not made until 24 hours after

1140 Communications of the ACM November 1985 Volume 28 Number 11 Special Issue

3230 530 OOWN a 10 20 30 0 10 20 30 0 10 20 30 (b) Data and Synthetics I I I 1 I I I 1 Comparison of three components of synthetic particle velocity time histo- ries with the data at stations EOl , E02, E03, and E04. All components lEol v 0.5 are plotted to the same amplitude SYN. scale. For stations where absolute SO time was available (EOl , E02, and E04), the synthetics have been -0.5 F +mLJ --I aligned accordingly; for stations where absolute time was not avail- E02 v --&w-L M able (E03), the synthetics have been shifted for the best fit. This shift is SYN. generally less than 1 .O s from the trigger time of nearby stations. Sta- tions EOl and E02 were not used to constrain the faulting model. (From J. Geophys. Res. 89, B6 (June IO, 1964) 4570.)

I t I I I 1 I I I I I I 0 10 20 30 a 10 20 30 0 10 20 30 TIME IS1 TIME IS1 TIHE IS1

FIGURE6b. Some Results of Archuleta’s Study of the Imperial Valley Earthquake of 1979

AN EXAMPLEOF A SEiSMiCREFRACTION EXPRRIMEWj' Recently Gary Fuis and associates at the USGS in Menio perturbation of velocity modelsiunnin~ on a HoneyweE) Park completed an extensive (30 man-year) seismic refrac- ~uktcs system, calculation of ra$ pa&?, siitid&c&@ ,pf s tion survey of the Imperial Valley region. Thefr preliminary arrival UWS of rays to each of the ane hpnd&d in@&- data,published in 1982,were used~byArchuleta to model merits,1t was possibleta construct a rr@el d wavu’velod- seismicvelocity variations in the imperial fault zone. To ties as,asfumtion ef depth throughout the ImperM $idt gather data, one hundred portable seismicinstruments regi&. The fnformatian QwSacq&#,di $miiinedyith.data were arrangedlinearlyat approximately one-km intervals. l&m bpra samples’ throughout ths v#&q eizak$ed I%& and t$n explosive charge was fired, sending P-wavesdown into ass&fatesto ~anstruct a s~~~t~~ra~~~.,~~,d~i,d the Imperial the earth along ray path; that curved and resurfacedat , fault kegion.Ray tracing describes~,~pp~~~~~eti~n of We points within a radius of 200 km. Any particnlar in&u- wave”‘mot&e and work5 well only ‘~~,[email protected] ment thus receiveda signal that had ,passedthrough sonic si~f~~a~fiy she&r than the size ~~~~~~~al,~~~~~~a~ portion of the earth.‘ssurface, Generally, a ray detected net&s &d in circu’mstaneesrrvhqre t&a @Wits gwdisnt far away may have traveled to greaterdepth than one doe¬ ohangeta i$ly, In rogipn~~~~~~~~~~~~a~~ meth- measuredcloser to the sour&, or both rays may have ’ ods arenot applicaii le, ~~~.~~i~~~~~~~~~~“~~~~,‘~qn~iooa gazed a structural dfscontinnity-such as a layer of harder mu$ often be developedby e~~~~;~~~~;~~~~~~~aor ~timte Soil, shale, or compactedgravel-been diffracted along the ’ ebmbnt methodsand may rdfluire ~~~~~~~~b~~&QGxter sameboundary far semedistance, and radiated back up power. Suc;hcomplications arise f$equentl~i&,gas and oil toward the instrument. In any case,each ray provides a , expieration, where the e&h ie t~p~~~~~~‘ia~,r~d:i~COW sampleof subsurfaceelastic properties.The experiment plexSstratawith sharply ;diffeTerin~‘vvi;K,:~~l~~~~~~or is was repeated40 times with various shot points and align- Erushedand convolttted as in a fat&Terre. Thesecomplica- merits of the linear array of instruments. By trial-and-error tions do not arise within the ImReri,alfault SHFC “:I,, ’

November 1985 Volume 28 Number II Communications of the ACM 1141 Special issue

the main shock) and how much. Archuleta found that he LONG VALLEY CALDERA: A POSSIBLE LINK could not fit the data without including a slip model for BETWEEN VOLCANISM AND EARTHQUAKES the Brawley Fault. The Brawley quake was quite sub- Long Valley is a popular recreational area on the east stantial (of magnitude 5.7), and judging from the timing side of the about 40 km southeast of of the rupture, it must have been triggered by rupture Yosemite Valley. Famous for its Mammoth Lakes ski on the Imperial Fault. The Brawley Fault was 10 km area, fishing in Lake Crowley, and numerous hot long and 8 km wide. Although it contributed less than springs, it has become a vacation and summer residen- four percent to the total energy release, it had a large tial spot for thousands of Californians. Long Valley also impact on local accelerations. has the distinction of standing at the headwaters of the The most dramatic discovery was that the synthetics Owens River, long submerged in political controversy could not be matched to the true data without rejecting as a major water supply for Los Angeles. Since 1978 a common assumption, namely, that the rupture must Long Valley has been shaken by a series of intense and advance more slowly than an S-wave. To obtain a good unusual earthquakes. Each of the larger quakes, rang- fit, it was necessary to allow the rupture to race at an ing in magnitude from 5.7 to 6.3, was associated with apparent velocity exceeding l.he P-wave speed. This swarms of thousands of smaller quakes during the pre- was an especially satisfying result. As early as 1972, ceding or following weeks. The unusual nature of these Robert Burridge, a mathemat:ician at the Courant Insti- quakes has been puzzling earth scientists and has tute, suggested theoretical conditions under which a prompted a wide variety of monitoring and research rupture could become transonic, that is, exceed S-wave activities in the area. speed. The idea was that, because the two sides of a Typical California earthquakes consist of one large fault are slipping in opposite (directions, the P-wave ra- quake, called the main shock, usually followed, and diation from the crack tip cart induce a substantial occasionally preceded, by many smaller events; then a shear on the fault ahead of the rupture. If such a shear subsequent long period of calm. The smaller events happened to occur at a point sonthe fault already commonly occur upon the same fault plane as the main stressed at a level near its yield point, it could trigger a shock. But in Long Valley the pattern is different. In- new earthquake there before the arrival of the original stead of calm, there seems to be a trend toward recur- rupture. Burridge’s result has a maximum rupture ve- ring episodes of large quakes. In this recent episode of locity equal to the P-wave speed. D. J. Andrews of the activity, the first large quake occurred in October Geological Survey and Steven. Day of Systems, Science 1978-a magnitude 5.7. Between May 25 and May 27 of and Software later demonstrated the feasibility of the 1980, four magnitude-6 events occurred. In January idea with a computer model of a transonic crack tip. 1983, three magnitude-5.0-5.7 events were followed by Although transonic rupture velocities had been in- thousands of smaller earthquakes. On November 23, ferred for other earthquakes, because of the paucity of 1984, a magnitude-5.7 earthquake occurred. And dis- data and theoretical basis, the results were generally persed among these larger events have been smaller discounted. The result has important implications for short-lived swarms of hundreds of events each. Fur- seismic engineering, because if rupture velocity is near thermore, the swarms associated with the larger quakes a wave speed, radiated energy can be strongly focused do not appear to occur on a plane. They are confined to in the direction of rupture. a restricted volume, as though the crust were snapping Archuleta’s work involved the direct computation of at random locations within the volume, rather than seismic waves: hence his technique falls into the cate- slipping along a well-defined fault plane. gory of forward modeling (see the sidebar on seismol- Long Valley sits in the midst of an area of intense ogy). As mentioned, other authors studied the same prehistoric volcanism. In the early 197Os,geologist Roy earthquake by linear inverse methods. An advantage of Bailey, during the course of mapping supported by the inverse methods, besides being efficient computation- Geological Survey’s geothermal research program, ally, is that they place bounds, on how well each pa- proved that 730,000 years ago Long Valley was the site rameter is determined. An advantage of forward meth- of a giant volcanic explosion. To compare it with a ods, despite the fact that they may entail vastly more cataclysm of our time, Mount Saint Helens spewed 0.25 computation than inverse methods, is that they allow cubic km of ash onto the surrounding region, but the greater flexibility in exploring models that depend non- explosion at Long Valley hurled 600 cubic km of ash, linearly on one or more parameters. Archuleta has burying 1500 square km of countryside under incandes- shown that ground response is very sensitive to the cent deposits of hot, steaming ash and spreading air- rupture propagation speed, a highly nonlinear variable. borne ash over the entire southwest and east as far as It is indicative of present-day seismology that each of Kansas, leaving a caldron-shaped depression 17 km long the Imperial Valley studies has offered only a kinematic by 32 km wide. Major eruptions continued intermit- model. Each has specified motions on a fault without tently within the caldron, the last occurring about attempting to describe the stresses that induced them. 50,000 years ago. Now somewhat cooled and apparently A dynamic model of a real earthquake-a model taking mellowed with age, the caldron’s interior has softened account of the stresses-may be the next step forward into scenic Long Valley. Because of a broad domelike in earthquake modeling. uplift in its center, Long Valley is known as a resurgent

1142 Communications of the ACM November 1985 Volume 28 Number 11 Special Issue

caldera. Many older calderas of similar kind have been nia earthquakes and volcanic eruptions with certainty. identified in North America, but two young ones of Earth scientists must forever seek to balance the possi- similar age include the Yellowstone Caldera, Wyoming, bility of errors in prediction against the consequences and the Valles Caldera, New Mexico. Five km south- of failing to advise the public of an impending catastro- west of Long Valley is Devils Postpile National Monu- phe. We certainly will not try to resolve that complex ment, where spectacular vertical stone columns were debate here-the scientific as wei1 as the political created by a cooling of lava flow, Immediately north- issues are intractably controversial. Here we merely west of Long Valley are the Inyo Domes and Mono review some of the evidence for potential volcanism, Craters, where violent volcanic explosions occurred as using it as a means of exhibiting some of the modern recently as 550 years ago. Long Valley is in fact sur- tools of seismology. rounded by evidence of active volcanism: black lava Early in the 1970s geophysicist Art Lachenbruch and mountains, red cinder cones, obsidian domes, pumice his colleagues, while participating in the Geological layers, and youthful explosion craters, ample evidence Survey study of Long Valley as a possible source of that similar eruptions could occur in the future. geothermal energy, noted that the magnitude of the One difficulty in understanding Long Valley seismic- observed heat flow in the area required an intense heat ity arises from the volcanic deposits. themselves. The source at depth. Lachenbruch speculated on the like- region is shot through with a complex pattern of lihood of an active magma chamber lying beneath the magma dikes and covered by thick accumulations of valley. By determining the thermal flux as measured in volcanic ash, lava flows, and welded ash deposits. In specially drilled bore holes, he was able to place con- addition the earth’s crust beneath the valley is inter- straints on where such a chamber might lie. As part of penetrated by a network of aquifers that conduct super- this multidisciplinary geothermal research effort, seis- heated water and steam from volcanic chambers deep mologist Dave Hill and his colleagues conducted a seis- in the crust to thermal springs at the surface. In such mic refraction study of Long Valley, similar in kind to intensely fractured and variable terrain, earthquake the one discussed for the Imperial Valley (see sidebar), waves are scattered and attenuated unpredictably. It is but smaller in scope. The objective was to model the difficult to interpret seismic waves passing through the velocity structure of the crust and, working in concert caldera and to compute earthquake locations with pre- with field geologists and geochemists, determine the cision. In a sense, volcanic regions tend to hide them- stratigraphy and structure of the region. Unexpectedly, selves behind a dark veil of geologic complexity. on one of the refraction lines Hill discovered a second- Another difficulty is due to the fact that Long Valley ary reflected phase possibly from the top of a T-km- lies at the boundary between two very different geolog- deep magma chamber. At the same time Hill’s col- ical provinces, the Sierra Nevada to the west and the leagues, Don Steeples and H. Iyer, found that P-waves Basin and Range to the east. The former is a massive arriving from distant large earthquakes were delayed granite batholith with a solid root extending to a depth by as much as a third of a second if they passed of 45-55 km. The latter is a broad, shallow-crusted, rift through certain parts of the caldera. They postulated zone. that molten rock to depths as deep as 15 km were One of the largest earthquakes in California history slowing down the P-waves. Thus, beneath the region occurred 140 km south of Long Valley centered near mapped as a resurgent dome by Bailey, and speculated Lone Pine on the eastern escarpment of the Sierra Ne- by Lachenbruch as the site of the necessary heat vada in 1872. Earthquake scarps throughout the region source, Hill, Steeples, and Iyer found seismological evi- show that large earthquakes have regularly accompa- dence strongly supporting the existence of molten nied the regional uplift that continues to hoist the magma. Sierra Nevada. Earthquakes at Long Valley can be re- Early in 1980 Alan Ryall, taking note of the 1978 lated to any number of causes: to Sierran uplift and the magnitude67 quake and subsequent swarms, pre- inevitable grinding conflict between the adjacent prov- dicted that a large quake could be expected in the re- inces, to cracking of the crust by extensional rifting, or gion within the foreseeable future. In May of 1980 four to the forceful intrusion of magma into dikes, sills, or quakes of magnitude 6 occurred within hours, followed magma chambers. Most geologists would agree that any by a magnitude 5.9 in September 1981, and three of these could be the cause of quakes in Long Valley. In magnitude-5.0-5.2 quakes in January 1983 within a sum, there is increasing evidence that the magma swarm of thousands of smaller quakes. Repeated geo- chamber beneath Long Valley’s resurgent dome has be- detic surveys of the region have shown that the resur- come reactivated, with the possibility of an eruption in gent dome has been stretching laterally and has been the near future. lifted vertically approximately 50 cm since 1979. Jim Geologists are well aware of the possible dangers and Savage and Malcolm Clark of the Geological Survey are watching developments at Long Valley carefully. It have mathematically modeled the swelling as due to was not too long ago that similar swarms of tremors expansion of a spherical chamber at a depth of lo-11 beneath Mount Saint Helens gave hints of the impend- km beneath the resurgent dome by the injection of 0.15 ing explosion. The sciences of seismology and volcanol- cubic km of magma (see Figure 7). ogy are too young to make claims of predicting Califor- In 1983 Chris Sanders, working at the Seismological

November 198.5 Volume 28 Number 11 Communications of the ACM 1143 Special Issue

Long Valley Calcfera was produced by eruption of Bishop surrounding earthquake activity indicates renewed filling of Tuff 730,000 years ago, which fills the caldera above the the chamber with magma from depth and possibly injection subsided Sierran basement block. Partial emptying of the of a dike along the caldera ring fracture below the south upper part of the chamber during these eruptions paused moat. The shallower cupola (hump) of magma on the right, collapse of the roof of the chamber and sinking of the Sienan based on Sanders’s seismic attenuation studies, appears to block about 2000 m into the magma below. Subsequent be the center of the current uplift. The heavier lines at the renewed magma pressure caused arching of the caldera upper left and lower right margins of the contemporary ffoor to form the resurgent dome about 7OO,OWyears ago. magma chamber mark the location of seismic phase transi- Since then the magma chamber has been cooling and crys- tions detected by Hill, leutgert, and colleagues. (Cross sec- tallizing from the walls inward (fed), gradually reducing the tion prepared by Roy Bailey and Dave Hill, U.S. Geological size of the partially molten chamber (yellow-oraixge>. ¢ Survey, Apr. 30, 1985. Publication by permission of the (1979-l 985) renewed rise of the resurgent dome (0.8 m) and American Geophysical Union.)

FIGURE7. Cross Section through Long Valley Caldera and Its Inferred Subjacent Magma Chamber

Laboratory of the Mackay School of Mines in Nevada, means of locating regions of apparent partial melt be- used the Nevada seismic network and seismic modeling neath Long Valley. The concept is that, because elastic programs on a PDP-11/70 to analyze S-wave energy solids are capable of transmitting shear waves but liq- received from Long Valley earthquake swarms as a uids are not, the shear-wave characteristics of criss-

1144 Communications of the ACM November 1985 Volume 28 Number 1I Special issue crossing ray paths through Long Valley can be used as with determining just a few basic facts about an earth- magma sensors: The more completely molten the re- quake, such as the location, direction of fault slip, ori- gion, the more attenuated the shear waves passing entation of the plane of faulting, and magnitude. Ob- through it. With enough rays in different directions, it served from a distance, it is almost impossible to distin- should be possible to identify regions that have no melt guish finer features. In principle, given data from a and those that are partially or completely molten. number of distant seismic stations, a trial-and-error for- Sanders analyzed 1200 ray paths from 281 small earth- ward method could be used to determine a rupture quakes that were recorded by a high-density seismic location, orientation, and average distance of slip, con- array spreading from Long Valley in eastern California sistent with the observed seismograms. But, because at to Lake Tahoe and to Walker Lake in western Nevada. large distances, the data cannot distinguish between Sanders concluded that two shallow magma bodies ap- small slips on a large surface and large slips on a small pear to lie in the central and northwest caldera, both surface, the same results would be obtained by increas- lying partly beneath the resurgent dome, as well as two ing the slip surface and proportionately decreasing the smaller, more diffuse magma areas in the southern slip distance. The features thus described, understood caldera and beneath Crowley Lake. The top of the cen- as an average, provide a phenomenological description tral magma body is as shallow as 4.5 km, and the north- of a source and are referred to as an earthquake mecha- west body as shallow as 5.5 km. At depth the two bod- nism. ies probably coalesce and extend to at least 13 km be- Until recently, a forward method was used to com- neath the surface. This is the first direct evidence of pute earthquake mechanisms. In its most efficient im- large bodies of molten rock in the caldera, since only a plementation, it could easily consume 20 minutes or liquid will affect S-waves in such a drastic manner. It is more on a VAX-11/780 to compute the mechanism of a also the first detailed subsurface map of such a body. single earthquake. In 1971 geophysicist Freeman Gil- This study provides new evidence indicating magma as bert formulated a powerful mathematical tool that has shallow as 4.5 km, compared to the i’-lo-km depth drastically reduced computation times to seconds: the originally suggested by other studies. seismic moment tensor. The essence is that an elastic In 1983 James Luetgert and Walter Mooney at the medium responds linearly to a moment tensor mecha- Geological Survey used seismic refraction profiles from nism, that is, the parameters that enter into the defini- eight earthquakes north of Mammoth Lakes in the tion of the moment tensor operate linearly upon the depth range of approximately 5-8 km to complement earth model, so that the efficient inverse methods of Hill’s explosion refraction study with additional veloc- linear algebra may be applied to derive the moment ity information at greater depth. An unusual, possibly tensor from the observed data. The moment tensor has unique, feature of the study is that the dense array of been an important innovation, not only numerically, seismic instruments used for recording data was the but conceptually as well, for it provides a concise de- same as that used in the earlier shot-charge study, ena- scription of a variety of different possible earthquakes, bling comparison of information derived from deeper not just of those that arise from lateral slip on a fault, earthquakes with that derived from shallower explo- but also of those arising from explosions. sions. By noting strange secondary arrivals on these Julian has underscored the fact that the moment ten- profiles and interpreting them as reflections from the sor may also be used to describe earthquakes caused by bottom of a magma chamber, Leutgert and Mooney the forceful intrusion of magma into the earth’s crust. were able to define a possible lower boundary for the Such magma intrusion yields no net material expan- region of partial melt, thus complementing Sander’s sion, but does incorporate simultaneous expansion and upper and lateral boundaries. Combining their results contraction along orthogonal axes. Although this fact with those of Sanders, they estimated that the magma had been noticed before, Julian was first to apply it to body in the southcentral caldera must have lateral di- the Long Valley caldera earthquake data. Other investi- mensions of about 10 km x 5 km and a height of about gators, when calculating mechanisms, constrained the 12 km, and they deduced a volume of partial melt on moment tensor to be equivalent to a fault-slip model the order of 600 cubic km. and arrived at inconsistent results, depending on which As noted earlier, because of the complexity of the set of seismograms was being analyzed. But Julian volcanic formations in Long Valley, the seismic data are sought the best-fitting unconstrained tensor and arrived especially difficult to interpret. Nevertheless, there is at an intrusion model that reconciles the data. increasing suspicion, if not consensus, among geologists, Whether alleged magmatism beneath Long Valley is geophysicists, and seismologists that an active magma indeed the cause of the recent earthquakes or is instead chamber is stirring beneath the resurgent dome. One an effect of tectonic rifting that cracks the crust and additional line of reasoning in support of this suspicion opens it to magmatic intrusion is still very much a has been provided by Bruce Julian of the Geological matter of controversy. Julian’s conclusions have been Survey. disputed, but his work demonstrates that the seismic Julian has noted inconsistencies in interpretations of moment tensor can be used effectively to study intru- seismic data from Long Valley and has proposed an sion models as a cause of earthquakes. interpretation that suggests that the recent earthquakes I want to mention one further seismic study con- may have been magmatic in origin. To understand ducted at Long Valley even though it has nothing to do Julian’s thesis, note that seismologists are often satisfied with magmatism. Seismic-reflection data have been

November 1985 Volume 28 Number 11 Communications of the ACM 1145 Special issue

used extensively by the petroleum industry to con- nary deflection of the San Andreas through the Trans- struct images of geological strata in prospecting for verse Ranges is called the Big Bend. structures that accompany deposits of gas and oil. The The Big Bend is a puzzle for geologists: What is it method exploits the fact that the wave equation is re- doing there? If the Pacific Plate west of the San An- versible in time. This means, in particular, that record- drea; is moving northerly in a direction parallel to the ings of excitation of a free boundary by incoming waves general trend of the San Andreas, how does it manage may be used in reverse as a source of waves that are to slip around the Big Bend? Some interesting proposals identical, but opposite in direction, to the incoming have been put forward by scientists at Caltech that may waves. Thus, when a shot charge is fired and waves help to explain the Big Bend and resolve other puzzles that are reflected by subsurface strata are recorded by a as well. sufficiently dense array of instruments, the data may be The major tool is tomographic seismology, which is propagated backward into the earth through a pro- related to tomographic image reconstruction used in grammed solution of the wave equation. By plotting the medical CAT scanning. In the latter, a patient lies still results at various time steps, the wave front may be while a movable X-ray beam sweeps in a plane at right observed to converge upon subterranean structures. angles to the long axis of the patient’s body. One full The seismically detected structures are the clues used revolution of the sweeping beam about the patient’s by geologists to determine whether gas and oil may be body thus provides sufficient information to reconstruct present. There are numerous technical difficulties an image of one slice (or transverse section) of the body. caused by refraction, diffraction, and multiple reflec- Imagery of adjacent sections can be adjoined in order to tions, as well as limitations in the data. Nevertheless, reconstruct an image of, say, the patient’s head or spine. reflection seismology has proved to be a tool of im- The basic idea of medical tomography is to imagine mense importance to the petroleum industry-an in- that a transverse section of heterogeneous matter is di- dustry that has purchased morle than a dozen Cray and vided into a checkerboard of discrete elements, each of a small number of CDC Cyber-205 supercomputers for uniform size. A typical X-ray beam, as it passes through a variety of applications including reflection work. a line of such material elements, will be attenuated by George McMechan of the Texas University Center for an amount that depends only on the total mass strung Lithospheric Studies is the first to use seismic-reflection along that line. The attenuation (and, indirectly, the techniques to construct an image of an earthquake total mass encountered) is measured at the point where source (see Figure 8). Working with colleagues at the the beam reemerges from the patient’s body: The Geological Survey, McMechan used reverse propagation greater the cumulative amount of matter traversed by of earthquake data from a dense array of instruments the beam, the greater the reduction of intensity of the located in Long Valley. Selecting three quakes from the beam. Obviously it is not possible to reconstruct the January 1983 swarm, McMechan reversed the data for density of individual elements along a single ray path each quake through a finite difference implementation without more information. But, if there are enough rays of the wave equation. Beginning with the waves re- (beams) in different directions or, more precisely, .if corded at the surface of the earth, he reversed the prop- there are typically more rays of varied directions agating direction such that the wave field converges to through each element of area than there are elements reconstruct an approximate image of the earthquake along a typical ray path (such that there are more inde- source. The important aspect of McMechan’s work is pendent equations than unknowns), then in principle, a the imaging of the seismic source rather than earth least-squares algebraic method can be used to estimate structure as is normally done in the petroleum indus- the density of each material element. Several computa- try. The black dot indicates the location of the source tionally efficient methods are now available to solve as computed by more traditional means. this problem, the most common being ART (Algebraic Reconstruction Technique). A related method called REFINING IMAGES OF THE INNER EARTH: SIRT (Simultaneous Iterated Reconstruction Technique) THE BIG BEND AND THE MINIPLATE is also used. Plate tectonics has provided a global understanding of Tomography may also be ap$lied to determine the processes that create continents and cause volcanoes variation of seismic wave speeds within the earth. and earthquakes. Now, using massive amounts of high- Since wave speeds correlate with temperature and den- quality data and sophisticated processing techniques, sity of the earth, the method has become an important earth scientists are beginning to fill in details of plate research tool in modern . The idea is to subdi- structure and to understand finer points of the physical vide a region of interest into congruent rectangular processes that create regional topography. bricks that are as small as possible, but large enough to Southern California is a good example. Take a look at be resolved by the data. Material within each brick is the map of the San Andreas Fault. For most of its assumed to be uniform, so that the elastic properties length, the San Andreas trends north by northwest. But are identical throughout any one brick. The time note that where it approaches the southern end of the required by a seismic wave to traverse the brick is San Joaquin Valley the San Andreas swings eastward inversely proportional to wave velocity within the past San Bernardino, where it resumes a north by brick and directly proportional to the length of the ray northwest trend into the Salton Trough. The extraordi- segment within the brick. To determine the travel time

1146 Communications of the ACM November 1985 Volume 28 Number 11 Special Issue

km km 6 10 12 14

W

Extrapolation of an observed (seismogram) wave field back- equal to zero. The round black dot in (d) is an estimated ward in time produces an image, in both time and space, of hypocenter computed by a standard method using regional the source. Four time steps in the process are displayed network travel times. (Figure from McMechan, G.A., Leut- showing the numerically reconstructed wave field at (a) t = gert, J.H., and Mooney, W.D. Imaging of earthquake sources 3.00 s, (b) t = 1.50 s, (c) t = 0.75 s, and (d) t = 0.00 s, where in Long Valley Caldera, Calii., 1983. Undated manuscript pro- the inferred time of earthquake at the source is arbitrarily set vided by Walter Mooney at USGS in Menlo Park, Calif.)

FIGURE8. Backward-Extrapolation Image of the Source of an Earthquake of January 1983 in Long Valley Caldera

of a wave from an earthquake source to a seismograph add up the traversal times of the ray through each station at the earth’s surface, it is only necessary to brick in its path. As in the case of medical tomography, compute the ray path from the source to the station and the measurement made for one ray is inadequate to

November 1985 Volume 28 Number 11 Communications of the ACM 1147 Special issue

resolve the character of each brick along the ray path, gence of the crust from opposing sides of the mountain. but if there are enough rays in enough directions The latter type of sustaining force is called dynamic through each brick, then an average wave velocity for compensation.We shall see below that, for the Trans- each brick can be determined. verse Ranges at least, there is good reason to believe A complication for tomogra:phy in seismology is that that dynamic compensation is involved. But first it is the data arise from random ea.rthquake sources and important to know more about crustal thickness. cannot be controlled to provide uniform coverage from Pn-rays were used to determine the thickness of the all directions. Until the advent of instrument networks crust and variations of P-wave velocities in the upper and semiautomatic processing of earthquake records, mantle. Hearn’s analysis of Pn-arrivals resulted in a the data were too sparse for tomographic methods. But profile of crustal thickness (Moho depth) throughout in recent years the CUSP system has gathered adequate southern California and a profile of upper mantle wave data from the Southern California Array to make to- velocities. The crust is thinnest (20 km) in the Salton mography feasible for studyin,g the crust and upper Trough near the and thickest (34 km) mantle beneath southern California. farther northwest in Ventura County. In fact, the re- At Caltech Rob Clayton and his associates have gional topography of the boundary between the lower adapted medical tomography to study the geology of crust and upper mantle, known as the Moho, is quite southern California. They have used an SIRT program complicated, reflecting the complexity of the surface on a VAX-11/780 (UNIX@ operating system). In order to topography. “see” through the earth’s crust to determine the charac- Having obtained knowledge of Moho depths through- ter of the earth’s mantle, it is necessary first to under- out southern California, as well as an average crustal stand the structure of the crust. For his thesis, Tom velocity, it became possible for scientists to trace and Hearn applied tomography to (data from thousands of time ray arrivals from distant earthquakes with greater regional earthquakes to determine the P-wave veloci- accuracy than ever before. In this respect, Hearn’s crus- ties and crustal thickness variations beneath southern tal study serves a purpose like that of a seismic refrac- California. He analyzed both Pg- and Pn-wave arrivals tion experiment carried out on a much larger scale. recorded at approximately two hundred seismographs Because an inhomogeneous crust distorts imagery of the Southern California Array. Pg-waves are P-waves through wave refraction, Hearn’s work made it possible that originate at an earthquake source within the crust to remove the distortions. Gene Humphreys, working and propagate directly to the seismic station without with Clayton, used Hearn’s results to refine a tomo- ever leaving the crust. Pn-rays are P-waves that de- graphic image of the P-wave velocity in the southern scend to the bottom of the crust from an earthquake’s California mantle, derived from teleseismic events (dis- source within the crust, run at high speed along the tant earthquakes) recorded by the Southern California upper mantle, radiating rays back upward through the Array. Humphrey’s early work was done on a Prime crust at reduced speed to the surface. computer, but subsequently a VAX-11/780 (UNIX) was Analysis of Pg-arrivals yielded a velocity map that used. As before, the region of interest (this time the reveals spectacular discontinuities across several of the underlying mantle) was dissected into rectangular major faults of southern California (for example, the blocks, each of supposedly uniform elastic properties. San Andreas, Garlock, San Jacinto, and Elsinore faults) The earthquake sources were sufficient in number and demonstrating the existence of regional blocks divided variable enough in location to obtain a well-resolved along fault boundaries. A somewhat puzzling result is image of the upper mantle using SIRT, the same tomo- that the velocity profiles beneath the Transverse graphic technique used by Hearn (Figure 9). Ranges and the Peninsular Range are slow like that of The results are fascinating. A slow region was found the surrounding terrain rather than fast like that of the beneath the Salton Trough. Assuming that wave veloc- mountains themselves, indicating that the mountainous ity is a decreasing function of temperature, the result material does not extend downward beneath the level implies that this mantle is relatively hot and less dense of the surrounding terrain. This is puzzling because than the surrounding material. Because the mantle is most mountains have a root that is deep enough to presumed to flow plastically in geological time, it is compensate for the elevated p(art, much as an iceberg reasonable to infer that differential density of material floats partly submerged to provide buoyancy for the in the mantle must induce upwelling beneath the Sal- part that rises above sea level. One hypothesis is that ton Trough and that resulting basal tractions on the the roots were shorn from beneath the mountains by a crust cause the thinning and extension known to occur detachment fault, the plane of which is parallel to the in that region. A more remarkable result, however, is earth’s surface. Presumably the region now beneath or that there is a large block of relatively fast-velocity surrounding a mountain has flexed downward to sup- material directly beneath the Transverse Ranges. As- port it. Another hypothesis, perhaps complementary to suming a lower temperature in the block (and therefore the first, is that part of the force that sustains the higher density) compared to the surrounding mantle, it mountain may be generated d:ynamically by tractions is reasonable to believe that the block is downwelling. due to mantle convection undlerneath causing conver- If that is so, then upper mantle material, drawn in from opposite sides to replace the top of the downwelling UNIX is a trademark of AT&T Bell Laboratories. block, must exert tractions upon the overlying crust,

1148 Communications of the ACM November 1985 Volume 28 Number 11 Special Issue

Results of the inversion of teleseismic P-delays. In the upper panel, a W -, E cross section (A-A’) through the Transverse left panel, a horizontal section at a depth of 100 km is shown Range anomaly is shown. In this projection the anomaly ap- superimposed on a location map of southern California. The pears as a wedgelike feature that is deep&f on the eastern locations are shown for the three cross sections (A-A‘, side. A S 3 N cross section (B-3’) through the Transverse B-B’, and C-C’) that are displayed in the other panels. The Range anomaly is shown in the lower right panel. The anom- tick marks surrounding the horizontal section show the loca- aly appears as w slablike feature that dips slightly to the tions of the block centers used in the inversion. All panels are north. Iti the upper right panet, a SW or, NE cross section displayed with no relative exaggeration. The contour interval through the Balton Trough anomaly is shown. The anomaly is 1.5 percent relative velocity deviations, with > 1.5 percent is about 2-4 percent slow and extends down to 75-l 25 km. indicated by dotted areas and < -1.5 percent by the (From Beophys. Res. Left. 17,7 (July %964),626.) hatched areas. The zero contour is dashed. In the lower left

FIGURE9. A Tomographic Image of Mantle Structure beneath Southern California causing the kind of convergence mentioned above that stood, such tractions appear to be aligned in such a way could dynamically support a rootless mountain. Al- as to maintain the curvature of the Big Bend. though the mechanisms are not yet completely under- In order to test the self-consistency of dynamic com-

November 1985 Volume 28 Number II Communications of the ACM 1149 Special Issue

pensation. Brad Hager and Gene Humphreys employed CONCLUSION a finite-element simulation of viscous flow driven by a I have tried to show, by presenting a few contemporary mantle density gradient. Their programs were run on a examples, that modern computer and instrumentation variety of computers, including a VAX-11/780 (UNIX systems have catalyzed the rapid growth of seismology operating system), Prime (Primos Operating System), over the past three decades, leading to increasingly ac- and Ridge (UNIX operating system). At present Hager is curate knowledge of the structure and physical pro- using a Celerity (UNIX operating system) for some of cessesof the Earth’s interior. In effect, we are peering his work. Assuming density differences commensurate into the opaque Earth, not with crystal clarity yet, but with the tomographically inferred velocity variations in more than just a shadowy, suggestive way. The rapid beneath the Transverse Ranges, they were able to show progress of recent years has been astonishing. Networks that the seismically inferred density anomalies would of seismic instruments, automated data-reduction sys- result in viscous flow supportive of the Transverse tems, efficient database storage of huge volumes of Ranges. Further support for dynamic compensation is data, and rapid and flexible computer analysis tech- given by the fact that the flow models predict varia- niques have each played a part; and they have all come tions in gravitational potential equal to the empirically about within one generation. Most of the key systems measured regional gravity field. and analytical techniques described in this paper either It is interesting to note in passing that Clayton, Hager, came into being or were generally known only after and their colleagues are experimenting with concurrent 1970. Without these technical innovations, the studies processing; they are converting some of their finite- reviewed here concerning the Imperial Valley earth- element convection modeling programs and finite- quake, the Long Valley Caldera, and the tomography of difference seismic modeling programs for execution on southern California would have been impossible. the Mark II hypercube, a parallel computer with 16-128 Our knowledge of the Earth’s interior will continue processing elements, each based upon an Intel 8086/ to advance rapidly, in pace with developments in 8087 chipset and 256 Kbytes of memory, under devel- graphics, parallel computation, expert systems, com- opment jointly at Caltech and NASA’s Jet Propulsion puter networks, and instrumentation systems. Hope- Laboratory in Pasadena. fully that knowledge will not only illuminate science, Additional understanding of the Big Bend is being but also permit society to plan for and mitigate the gained by kinematic studies of southern California. effects of inevitable earthquake disasters. Geodesic trilateration measurements made by the Acknowledgments. This article could not have been USGS have yielded accurate relative motions between written without the generous cooperation of earth sci- benchmark points throughout southern California. Ray entists at the US Geological Survey, Caltech Seismologi- Weldon of the USGS, working with Humphreys, has cal Laboratory, Stanford University, and the University considered that such motions must be simultaneously of California, most of whose names appear in the text. I consistent with geologically mleasured slip rates of would also like to thank Dick Baugh, Adam Dziewon- known faults in southern California and measured ex- ski, Joe Fletcher, Stewart Levin, Al Lindh, and Chris tension rates of the Basin and IRangeprovince east of Rojahn for helpful discussions and comments on an the Sierra Nevada. Their conclusions are that the Sierra early draft. Nevada and Great Valley block is moving westward SUGGESTED READINGS against the San Andreas Fault north of the Big Bend, Applied Technology Council. Earthquake damage evaluation data for driven by the Basin and Range extension. That exten- California. Rep. ATC-13, Applied Technology Council, Palo Alto, sion, combined with the basal tractions described by Calif.. 1985. Bott. M.H.P. The Interior of the Earth. Elsevier North-Holland, New York. Hager and Humphreys, is driving a counterclockwise 1982. rotation of southern California about a point that is Hoffman, A. Vision or Villainy: Origins of the Owens Valley-Los Angeles Water Conrraversy. Texas A&M Univ., College Station, Tex., 1981. approximately 650 km southwmestof the San Andreas Howard, A.D. Geologic History @Middle California. Univ. of California Fault in the Big Bend. Thus WIeldon and Humphreys Press, Berkeley, 1979. assert that southern California must occupy its own Iacopi, R. Earthquake Country. Lane, Menlo Park, Calif.. 1971. Rinehart. C.D.. and Smith. W.C. Earthquakes and Young Volcanoes along miniplate, situated between the North American Plate the Eastern Sierra Nevada. Kaufmann, Los Altos, Calif., 1982. and the Pacific Plate. In theory, the Big Bend and the Wesson: R., and Wallace, R. Predicting the next great earthquake in southern portion of the San Artdreas Fault are simply California. Sri. Am. 252, 2 (Feb. 1985) 35-43. the outer arc along which the miniplate rotates. As CR Categories and Subject Descriptors: B.3.2 [Memory Structures]: attractive as the theory may be, not all scientists agree. Design Styles--mass storage; 1.6.4[Simulation and Modeling]: Model Validation and Analysis; I.2 [Physical Sciences and Engineering]: earth Thomas Jordan of MIT and his associates, who are and atmospheric sciences studying satellite ranging measurements made by Additional Key Words and Phrases: ART (Algebraic Reconstruction NASA’s Crustal Dynamics Project, do not believe that a Techniques), CUSP system, SIRT. Tomography miniplate is needed to account for the kinematics of Author’s Present Address: Michael R. Raugh, Research Institute for Ad- vanced Computer Science, Mail Stop 230-5, NASA Ames Research Cen- southern California. Undoubtedly data from NASA’s ter. Moffett Field, CA 94035. highly accurate new Global Positioning Satellite system Permission to copy without fee all or part of this material is granted will help to clarify plate motion in southern California. provided that the copies are not made or distributed for direct commer- Although the problems posed by the Big Bend will con- cial advantage, the ACM copyright notice and the title of the publication and its date appear. and notice is given that copying is by permission of tinue to stir debate, earth scientists are unlimbering the Association for Computing Machinery. To copy otherwise, or to powerful tools that may soon provide the answers. republish, requires a fee and/or specific permission.

1150 Communications of the ACM November 1985 Volume 28 Number 1 I