Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
APPENDIX A EARTHQUAKE-PREDICTION RESEARCH IN THE UNITED STATES M. Nafi Toksoz In the United States, efforts in earthquake prediction have evolved in two directions: l. Studies of seismicity and recurrence times, statistical predic- tion, and risk evaluation based on statistical information. 2. Deterministic prediction based on changes in measurable physical parameters during the interval prior to the earthquake. There have also been experiments during the past decade in the direct control of earthquakes through injection of fluid into the fault zone. In U.S. attempts at earthquake prediction, the earliest emphasis was on statistical methods. At present, however, the deterministic tech- niques hold great potential. This appendix briefly reviews the work and results in the U.S. in earthquake prediction and in experiments toward earthquake control. SEISMICITY AND STATISTICAL PREDICTION It is generally accepted that seismicity patterns of the past hold the key to those of the present and the future. An area that has experi- enced earthquakes in the past will most likely have similar ones in the future, and the frequency and magnitudes of these future earthquakes can be generally estimated from the frequency-magnitude relationships of those of the past. Thus, early attempts to predict earthquakes were based on the seismic history of the area being studied. Using such historic and statistical data, an earthquake risk map (Figure l) was prepared for the United States(l). This map has been used extensively in construction codes and in planning. The major shortcoming of the statistical approach has been that it provides neither exact locations nor reliable recurrence intervals for the larger earthquakes(2). Statistical techniques have been used in many studies to search for periodicities or other trends conducive to more definitive prediction of earthquakes. Most of these studies suf- fered from the unavailability of data covering sufficiently long time periods. The historic data, based on observer reports rather than 37
38
39 instrumental measurements, were homogeneous neither in time nor in space. Development of seismograph networks has improved data quality, but this data base covers a time span of only a few decades, and the statistical analysis of these data has shown no well-defined period- icities. Once the clustering due to aftershocks is removed, the time distribution of earthquakes can best be fitted by a Poisson model(3, 4, 5), in which events occur randomly in time but the mean number of events per unit time is constant, and equal to the variance. The study of time-space patterns of earthquakes, which is still in a preliminary stage, may hold some promise. Simply defined, the idea is based on the concept that strain accumulates over a wide region along a plate boundary (or seismicity belt); each earthquake releases the strain energy over an area defined by the extent of faulting or of the after- shock zone. Seismic gaps along a fault that is active elsewhereâi.e., areas in which no earthquakes have occurred for long periods of timeâ may be the most likely sites for future earthquakes. This idea has been explored for earthquakes in the Aleutians(6, 7) and in the North Anatolian fault zone(8). Although more definite and quantitative studies are still required, the available results are encouraging. This type of study, combined with geodetic measurements and theoretical strain-field calculations, may be able to pinpoint the highly strained areas for very close monitoring. Other studies employing pattern- recognition concepts(9), being conducted now, may provide a multi- parameter approach to the probability of an earthquake in a given area. In summary: At present, the greatest potential of the statistical approach to earthquake prediction is to identify risky areas requiring careful monitoring, rather than to determine the exact time and location of future earthquakes. DETERMINISTIC METHODS OF EARTHQUAKE PREDICTION In recent years, significant advances have been made in identifying physical changes that preceded earthquakes. (See Reference 2 for a detailed discussion.) Most notable among these have been the changes in seismic compressional and shear-wave velocities and of their ratio (Vp/Vs), and changes in electrical conductivity, water pressure, ground tilt direction, and surface elevations in and around the earthquake source. With the aid of laboratory measurements, these precursory phenomena have been related to the physical process of dilatancy(l0, ll, l2, l3, l4, l5) put forth by U.S. and USSR investigators. The models are based on laboratory fracture studies(l6). Prior to failure, the stressed rock undergoes a volume increase or becomes "dilatant." Dilatancy is produced by the formation of cracks within the rock and increased porosity. In the U.S. "dilatancy-diffusion" model, at the initial stage of dilatancy, originally saturated rock becomes undersaturated. Then water flows into the source region and resaturates the rock. This process takes place slowly because of the low permeability of most crustal rocks. Saturation and subsequently increasing pore- fluid pressure then reduces the rock strength, and failure (i.e., the
40 earthquake) occurs. With the release of stress as a result of the earthquake, the rock returns to the "non-dilatant" state. In the "dilatancy-instability" model, rapid development of cracks during the first stage of dilatancy is followed by gradually accelerating defor- mation in the fault zone before the earthquake with a concomitant de- crease in shear stress. Outside the weakened shear zone, the cracks close again as the stress decreases and the velocity ratio increases prior to failure. The conditions that preceded earthquakes, and their effects on observable physical properties, are shown schematically in Figure 2. As discussed in the following section, the definitive field observations in the United States have been of changes of the seismic velocities or the velocity ratios and directions of the ground tilt. Additional precursory phenomena have been observed in the USSR, Japan, and China. These are discussed in greater detail in Appendix B. Precursory Velocity Changes Field tests of the feasibility of earthquake prediction have been car- ried out in the United States primarily using precursory changes in seismic velocity. Most of these studies were made after the earthquakes had occurred. In one case, however, a small (M = 2.6) earthquake was actually predicted several days before it occurred (the Blue Mountain Lake earthquake, discussed below). So far, earthquake prediction based on seismic velocity changes has been applied to earthquakes associated with thrust, normal, and strike-slip faulting. Both local and tele- seismic travel times have been used in these studies. For thrust-type events in New York State and California the results are encouraging. For others, the preliminary results are not yet definitive. The thrust-type earthquakes for which precursory travel-time or velocity changes were analyzed were those in the Blue Mountain Lake area of New York State(l2), the San Fernando earthquake of February 9, l97l (ll), and the Pt. Mugu earthquake of February 2l, l973(l7). The perti- nant data for these earthquakes are shown in Figures 3a, 3b, and 3c. These results are convincing. They imply compressional velocity changes in a fairly large area (at least a few times the fault dimension) around the earthquake source. The compressional velocity or Vp/Vs velocity ratio first decreases to a minimum, and then rapidly recovers to the normal value. The earthquake follows this recovery. The duration of the anomalous period is related to the earthquake source dimensions and the magnitude(ll, l8). Thus, the method predicts not only the time but also the magnitude of an impending earthquake. The actual prediction of the Blue Mountain Lake, N.Y., magnitude M = 2.6 earthquake was made on the basis of this type of data(l3). These data are shown in Figure 4. Based on the rapid drop of Vp/Vs ratios in the figure, and on the seismicity pattern, both the magnitude and the time of this earthquake were predicted correctly. The graph for the third event at Blue Mountain Lake, which includes additional data obtained or reduced after the earthquake, shows a characteristic behav- ior pattern also seen in the graphs for the other thrust events illus- trated (Figures 3b and 3c).
41 PRECURSOR STAGES STAGE IV EARTHQUAKE SUDDEN DROP IN STRESS FOLLOWED IV AFTERSHOCK! ⢠U1LDUPOF ELASTIC STRAIN UNSTABLE DE FORMATION IN FAULT ZONE AND PARTIAL RELAXATION OF STRESS IN SURROUNDING REGION SUDDEN DROP IN STRESS FOLLOWED IV AFTERSHOCKS OILATANCV DEVELOPMENT OF CRACKS BUILDUP OF ELASTIC STRAIN RADON EMISSION ELECTRICAL RESISTIVITY NUMBER OF SEISMIC EVENTS Fig. 2. Schematic diagrams of expected changes in some physical parameters as a function of time before, during, and immediately after an earthquake according to two models developed in the U.S. and USSR. The solid lines represent the "dilatancy-instability" model developed in the USSR. The dashed line is the "dilatancy- fluid flow" model of the U.S. The five stages are listed and the description of events at each stage is given at the top. Radon emission may be a function of both water flow and rate of creation of new surface area by the growth of cracks. The expected behav- ior of electrical resistivity in the ground has not yet been measured with sufficient accuracy to resolve between the models (l4, l5).
42 T 1 J T T T 7" I " 9 InII 12 H «_ 4^ fctil IL 44-TI _1 ^»v ^-*T ^f*^*^^ y llJI 11 125 llJ ll.4 l>l =11 1=1 lii LJ July Fig. 3a. Velocity ratio Vp/Vs as a function of time for two events in the Blue Mountain Lake, N.Y., earthquake swarm of l97l. The mag- nitudes of individual events are given along the time axis next to arrows designating the events. Bars on data represent estimated errors in data.(l2). IO 06 O.4 63 65 â¢7 69 61 63 69 67 Tim* (y*tri) 69 Fig. 3b. Variation of (a) seismic velocity ratio (Vp/Vs - l), and (b) seismic compressional (Vp)and shear (Vs) velocities as a func- tion of time before the San Fernando earthquake (magnitude =6.6) of February 9, l97l. The velocity measurements are between Pasadena and Riverside, California stations. Each point corresponds to an earth- quake whose magnitude is given along the time axis in each figure. The maximum estimated error due to time readings is shown by bars.(ll).
43 1971 1972 M M J SN,JMMJSN,J 6 *® . 800 60C iii 0 10 20 o. o O 20 10 20 600 400 200 0 Days prior to Pt Mugu earthquake ML-60. 2-21-73 Fig. 3c. Apparent velocities before the Point Mugu earthquake as determined from the recordings at seismic stations of the Santa Barbara channel array. Upper curve shows data from earthquakes at San Fernando Valley, middle curve from Los Angeles Basin.(l7) Whether this method is also applicable to non-thrust earthquakes is not yet clearly resolved. Analysis of travel-time data from earthquakes and quarry blasts in California did not show precursory velocity anoma- lies over large areas preceding some strike-slip-type earthquakes in the Bear Valley section of the San Andreas Fault(l9, 20) or for the Borrego Mountain Earthquake in l958(2l). It was not clear from these studies whether there were no precursory velocity changes or whether the dila- tancy was confined to a very small region immediately at the source area, such that its effects could not be determined within the accuracy of the data. More-recent results based on P-wave travel-time residuals immediately at the source region of the February 24, l972, earthquake near Bear Valley(22) suggested a velocity decrease prior to the earth- quake. Unlike the thrust-type events, the apparent anomalous zone for this strike-slip earthquake may have been small and probably confined to the volume defined by the aftershocks. A similar observation of undetected velocity changes prior to the earthquake was documented on the basis of explosion data for the June l, l975, Galway Lake earthquake in Southern California(24). Thus, for strike-slip earthquakes along the San Andreas, the available data seem to indicate that precursory veloc- ity changes, if they do indeed occur, probably are confined to the imme- diate focal region, and do not extend to a large area around the source. This may also be true for some strike-slip earthquakes in Japan(23). Changes of S-wave velocities have not been studied as extensively as those of P-waves. Shear-wave anomalies preceding an earthquake have been observed, but the magnitudes of these anomalies are smaller than those for P-waves.
44 1.8 1.7 1.6 1.5 1.4 1.9 1.8 1.7 1.6 1.5 1.4 * 4 "1 4 4 i J T ' - 4 > 4 4* 44 4 _ 4 ^ 4* \ { _ 1 1 - . b * t EARTHQUAKE M = 2.6 29 30 31 i 2 345 JULY AUGUST Fig. 4. Travel-time ratios (ts/tp) for the August 3, l973, Blue Mountain Lake earthquake predicted by Aggarwal et al.(l3). The upper figure (a) shows data that were available at the time of the predic- tion. The lower figure (b) shows all the data, with quarry blast and other earthquake travel times analyzed after the earthquake. The extension of these studies to areas remote from the monitoring instruments, and to larger earthquakes, is being accomplished using travel-time residuals(25). Although these interpretations are model- dependent, they may broaden the data base since large amounts of such data are available for intermediate and large earthquakes of the past. Laboratory studies of the effect of stress on rock properties have been important in understanding the field observations and putting to- gether the dilatancy model(9, l3, l5). In the laboratory, rocks exhi- bit anomalous changes of physical properties near fracture stress. The
45 magnitudes of the changes, however, are not always consistent with changes observed in the field prior to an earthquake(26). The rocks and conditions in the earth's crust near the fault zones are very com- plex and may be affected by factors not incorporated in the laboratory experiments. The laboratory studies still provide very valuable data necessary for developing a physical model and better understanding of earthquake processes. Changes in Direction of Ground Tilt The most consistent precursory phenomenon for San Andreas earthquakes has been change in the tilt direction(27). The measurements made with shallow-borehole tiltmeter arrays along the northern San Andreas indi- cate that there are definite and significant changes in the tilt direc- tion prior to a local earthquake or earthquake cluster (Figure 5). During inactive periods, the records indicate systematic (secular) tilt- ing of as much as one microradian per month in some fixed direction. Tilt directions change a few weeks to months before local earthquakes with magnitudes M = 3 to 5 as seen on instruments located within l0 AUGUST 7 EARTHQUAKE (MAGNITUDE 3.3) 7 MILES WEST OF SITE JULY 9 EARTHQUAKE (MAGNITUDE 3) 4 KILOMETERS SOUTHWEST OF SITE JANUARY 10 EARTHQUAKE (MAGNITUDE 4.3) 15 KILOMETERS NORTHWEST OF SITE 1 2 34 5 6 7 8 9 10 11 12 13 14 15 16 17 - WEST TILT (MICRORADIANS) EAST » Fig. 5. Tilt-direction changes before local earthquakes. Cumulative weekly mean tilt vectors (circles) from June 27, l973 to January l7, l974 observed at the Nutting site 7 km southwest of Hollister. Local earthquakes are shown as stars (H). North and east are the positive ordinate and abscissa, respectively.(l5, 23)
46 fault lengths of the source. This effect has already been seen for more than 20 events. The mechanism responsible for the tilt-direction change is not yet clear, but it does not appear to be simple dilatant volume expansion(28). In Summary: Precursory velocity and tilt-direction changes seem to have occurred for the earthquakes studied. However, some major questions still remain: Do velocity changes occur as a function of time without being followed by earthquakes? Are some earthquakes not preceded by velocity anomalies? Do tilt-direction changes occur prior to all local eathquakes both on the San Andreas and in other regions? Does the re- lationship between magnitude and precursor-time hold? These questions and some others need to be answered before a true assessment of the de- terministic methods can be made. Other Observations A number of other kinds of measurements, in addition to the seismic- velocity changes described above, are being conducted to identify pre- cursory changes in crustal properties prior to earthquakes. These measurements are being guided by observations in other parts of the world, by laboratory results, and by theoretical studies. They include geodetic measurements, and measurements of strain, creep (both horizon- tal and vertical motions of the crust), electrical conductivity, mag- netic anomalies, and groundwater pressure, among others. Most of these measurements were begun relatively recently and do not span sufficiently long time periods for clear-cut evaluation of their potential usefulness in earthquake prediction. Some of the preliminary work on magnetic anomalies(29, 30) and electrical conductivity(3l) is encouraging. Groundwater-pressure fluctuations show some correlation with creep rates(32). Geodetic(33, 34) and creep measurements(35, 36) have produced the most data to date. Analysis of these data, and of records of earth- quake distribution along the San Andreas(37), have provided information about crustal movements and strain fields along some segments of the fault. These kinds of information, as discussed in the early paragraphs of this report, can be very helpful in identifying areas for extensive monitoring. The most recent example has been the identification of the Palmdale uplift in California(38). Although the causes or the implica- tions of this uplift are not understood, this is clearly an area of anomalous behavior that should be watched carefully. A much larger body of measurements of changes in strain, tilt, and elevation is necessary to identify regions for extensive studies. Both the conventional mea- surements and new techniques utilizing reference points in space (qua- sars, the Moon, artificial satellites) will provide data on crustal movements.
47 EARTHQUAKE CONTROL EXPERIMENTS Earthquake control experiments in the United States have produced sig- nificant results bearing on our understanding of the occurrence of earthquakes and increasing the possibility of earthquake-risk reduction. Since the discovery of the relationship between the l962-65 earthquake sequences near Denver, Colorado, and the disposal of waste fluids by injection into a deep well at the Rocky Mountain Arsenal, earthquake control has become an important research topic. A well-planned field experiment was initiated in l967 by the U.S. Geological Survey at the Rangely Oil Field in western Colorado. Since a large number of wells were available, water could readily be injected into or pumped out of the source area and the pore pressure monitored. Meanwhile, an array of seismometers monitored the resulting changes in seismic activity. The results show an excellent correlation between fluid injection and earthquake activity, as illustrated in Figure 6(39). When the fluid pore pressure reached a threshold level (3,700 psi in this case), earth- quake activity increased. When pressure dropped as a result of water withdrawal, the seismic activity decreased. The generation of earthquakes in the field under controlled condi- tions can also be used to study and test the precursory changes in physical properties of a rock mass before earthquakes occur. MO IM .00 o : 3 Z SO EARTHQUAKE FREQUENCY AT RANOEIY OH HUD. COLORADO I All Ouol... ⢠Uki. I Inl...... .1 bMIM â¢I *lp*riK«m«l â¢*III 5000 3000 5 i 2000 1000 -FLUID WltHDItWAl Fig. 6. Earthquake frequency at the Rangely oil field, Colorado and its relation to reservoir pressure(39).
48 REFERENCES l. Algermissen, S. T. (l973). Earthquake History of the United States. (ed. by J. L. Coffman and C. A. von Hake), NOAA Publication 4l-l. 2. Rikitake, T. (l976). Earthquake Prediction. Elsevier Sci. Pub. Co., p. 357. 3. Knopoff, L. (l964). The statistics of earthquakes in southern California. Bull. Seismol. Soc. Am., v. 54, pp. l87l-l873. 4. Vere-Jones, D. (l970). Stochastic models for earthquake occurrence. J. Roy. Stat. Soc., no. l. 5. Shlien, A., and M. Nafi ToksOz (l970). A clustering model for earthquake occurrences. Bull. Seismol. Soc. Am., v. 60, pp. l765- l787. 6. Kelleher, J. A. (l972). Rupture zones of large South American earthquakes and some predictions. J. Geophys. Res., v. 77, pp. 2087- 2l03. 7. Wesson, R. L., and W. L. Ellsworth (l973). Seismicity preceding moderate earthquakes in California. EOS, Trans. Am. Geophys. Union, v. 54, p. 37l. 8. Allen, C. R. (l975). Geological criteria for evaluating seismicity. Bull. Geol. Soc. Am., v. 86, pp. l04l-l057. 9. Press, F., and P. Briggs (l974). Pattern recognition applied to earthquake epicenters in California and Nevada. EOS, Trans. Am. Geophys. Union, v. 56, p. ll50. l0. Nur, A. (l972). Dilatancy, pore fluids and premonitory variations of ts/tp travel times. Bull. Seismol. Soc. Am., v. 62, pp. l2l7- l222. ll. Whitcomb, J. H., J. D. Garmany, and D. L. Anderson (l973). Earth- quake prediction: variation of seismic velocities before the San Fernando earthquake. Science, v. l80, pp. 632-635. l2. Aggarwal, Y. P., L. R. Sykes, J. Armbruster, and M. L. Sbar (l973). Premonitory changes in seismic velocities and prediction of earth- quakes. Science, v. l80, pp. 632-635. l3. Aggarwal, Y. P., D. W. Simpson, and L. R. Sykes (l975). Temporal and spatial analysis of premonitory velocity anomalies for the August 3, l973, Blue Mountain Lake earthquake. J. Geophys. Res., v. 80, pp. 7l8-732. l4. Scholz, C. H., L. R. Sykes, and Y. P. Aggarwal (l973). The physical basis for earthquake prediction. Science, v. l8l, pp. 803-807. l5. Press, F. (l975). Earthquake prediction. Scientific American, v. 222, pp. l4-23. l6. Brace, W. F., B. W. Paulding, Jr., and C. Scholz (l966). Dilatancy in the fracture of crystalline rocks. J. Geophys. Res., v. 7l, pp. 3939-3953. l7. Stewart, G. S. (l973). Prediction of the Pt. Mugu earthquake by two methods. Proc. Conf. on Tectonic Prob. of the San Andreas Fault System, v. 473. l8. Anderson, D. L., and J. H. Whitcomb (l973). The dilatancy-diffusion model of earthquake prediction. Proc. Conf. on Tectonic Prob. of the San Andreas Fault System, v. 4l7.
49 l9. Bakun, W. H., R. M. Stewart, and D. Tocher (l973). Variation in Vp/Vs in Bear Valley in l972. Proc. Conf. on Tectonic Prob. of the San Andreas Fault System, v. 453. 20. McEvilly, T. V., and L. R. Johnson (l973). Earthquakes of strike- slip type in central California: evidence on the question of dilatancy. Science, v. l82, p. 58l. 2l. Allen, C. R., and D. V. Helmberger (l973). Search for temporal changes in seismic velocities using large explosions in southern California. Proc. Conf. on Tectonic Prob. of the San Andreas Fault System, v. 436. 22. Robinson, R., R. L. Wesson, and W. L. Ellsworth (l974). Variation of P-wave velocity before the Bear Valley, California, earthquake of February 24, l972. Science, v. l84, pp. l28l-l283. 23. Brown, R. (l973). Precursory changes in Vp/Vs before strike-slip events. Proc. Conf. on Tectonic Prob. of the San Andreas Fault System, v. 463. 24. Kanamori, H., and G. Fuis (l976). Variation of P-wave velocity before and after the Galway Lake earthquake (ML = 5.2), California, of June.l, l975. Bull. Seismol. Soc. Am., in press. 25. Wyss, M., and D. J. Holcomb (l973). Earthquake predictions based on station residuals. Nature, v. 245, pp. l39-l40. 26. Hadley, K. (l975). Dilatance: further studies in crystalline rock. Ph.D. Thesis, M. I. T., Cambridge, Mass. 27. Johnston, M. J. S., and C. E. Mortensen (l974). Tilt precursors before earthquakes on the San Andreas Fault, California. Science, v. l86, pp. l03l-l034. 28. Stuart, W. D., and M. J. S. Johnston (l975). Anomalous tilt before three recent earthquakes. EOS, Trans. Am. Geophys. Union, v. 56, p. 400. 29. Johnston, M. J. S., B. E. Smith, J. R. Johnston, and F. J. Williams (l973) . A search for tectonomagnetic effects in California and western Nevada. Proc. Conf. on Tectonic Prob. of the San Andreas Fault System, v. 225. 30. Johnston, M. J. S., C. E. Mortensen, B. E. Smith, and W. D. Stuart (l975) . Summary and implications of simultaneous observation of tilt and local magnetic field changes prior to a magnitude 5.2 earthquake near Hollister, California. EOS, Trans. Am. Geophys. Union, v. 56, p. 400. 3l. Mazzella, A., and F. Morrison (l974). Electrical resistivity variation associated with earthquakes on the San Andreas Fault. Science, v. l85, pp. 855-857. 32. Johnson, A. G., and R. L. Kovach (l973). Water level fluctuations on the San Andreas Fault south of Hollister, California. Proc. Conf. on Tectonic Prob. of the San Andreas Fault System, v. 489. 33. Savage, J. C., W. H. Prescott, and W. T. Kinoshita (l973). Geodi- meter measurements along the San Andreas Fault. Proc. Conf. on Tectonic Prob. of the San Andreas Fault System, v. 44. 34. Greensfelder, R. W., and J. H. Bennett (l973). Characteristics of strain variation along the San Andreas Fault from geodimeter mea- surements. Proc. Conf. on Tectonic Prob. of the San Andreas Fault System, v. 54.
50 35. Burford, R. O., S. S. Allen, R. J. Lamson, and D. D. Goodreau (l973). Accelerated fault creep along the central San Andreas Fault after moderate earthquakes during l97l-l973. Proc. Conf. on Tectonic Prob. of the San Andreas Fault System, v. 268. 36. Nason, R. D. (l973). Fault creep and earthquakes on the San Andreas Fault. Proc. Conf. on Tectonic Prob. of the San Andreas Fault System, v. 275. 37. Lee, W. H. K., K. L. Meagher, R. E. Bennett, and E. E. Matamoros (l972). Catalog of earthquakes along the San Andreas Fault system in central California for the year l97l. U.S.G.S., Open-File Report, 67 p. 38. Castle, R. 0., J. P. Church, and M. R. Elliott (l976). Aseismic uplift in southern California. Science, v. l92, pp. 25l-253. 39. Raleigh, C. B., J. H. Healy, and J. D. Bredehoeft (l976). An ex- periment in earthquake control at Rangely, Colorado. Science, v. l9l, pp. l230-l237.
