Chapter 4

Detecting Seismic Events

CONTENTS

PageIntroduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55The Meaning of ’’Detection” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55Locating Seismic Events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57Defining Monitoring Capability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

Seismic Magnitude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58Converting Magnitude to Explosive Yield. . . . . . . . . . . . . . . . . . . . . . . . . . 60Seismic Monitoring in Probabilistic Terms . . . . . . . . . . . . . . . . . . . . . . . . . 60

Limitations to Seismic Monitoring Capability . . . . . . . . . . . . . . . . . . . . . . . . 61Seismic Noise.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61Reduction of Signals at Source or Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . 61Seismic Instrumentation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62Seismic Magnitude Estimation Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

Seismic Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64Existing Networks and Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64Planned Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67Hypothetical Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

Seismic Monitoring Capability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68Calculating Seismic Monitoring Capability . . . . . . . . . . . . . . . . . . . . . . . . . 68Global Detection Capability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68Detection Capability Within the U.S.S.R. Using No Internal Stations . . 69Detection Capability Within the U.S.S.R. Using Internal Stations . . . . . 69Considerations in Choosing Monitoring Thresholds . . . . . . . . . . . . . . . . . . 70Use of High-Frequency Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70Arrays v. Three-Component Stations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

BoxBox Page4-A. Locating A Seismic Event. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

FiguresFigure No. Page4-1. Seismic Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 564-2. Seismic Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 624-3. Effect of Noise on Event Magnitude Computation. . . . . . . . . . . . ., . . . 634-4. Contributing Seismograph Stations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 654-5. Worldwide Standardized Network Station Distribution . . . . . . . . . . . . . 664-6. Detection Capability of a Seismic Station . . . . . . . . . . . . . . . . . . . . . . . . 684-7. Example of Calculated Detection Capability . . . . . . . . . . . . . . . . . . . . . . 71

Chapter 4

Detecting Seismic Events

The first requirement for a seismic monitoring network is to detect and locateseismic events that could have been caused by an underground nuclear

explosion.

INTRODUCTIONThe first requirement for a seismic monitor-

ing network is that it be capable of detectingseismic events. If the Earth were perfectlyquiet, this would be easy. Modern seismome-ters are highly sophisticated and can detectremarkably small motions. However, processessuch as the winds, ocean waves, and even rushhour traffic continually shake the Earth. Allof this ground movement is sensed by seismo-meters and creates a background from whichsignals must be recognized. Seismic networks,consisting of groups of instruments, are de-

signed to detect events like earthquakes anddistinguish them from normal backgroundnoise. The extent to which a seismic networkis capable of detecting events, referred to asthe network’s detection threshold, is depen-dent on many factors. Of particular importanceare the types of seismic stations used, the num-ber and distribution of the stations, and theamount of background noise at the station loca-tions. This chapter reviews these factors anddiscusses the capability of networks to detectseismic events within the Soviet Union.

THE MEANING OF “DETECTION”

In practical terms, detecting a seismic eventmeans more than observing a signal above thenoise level at one station. There must be enoughobservations (generally from more than onestation) to estimate the location of the eventthat created the detected signal. Measure-ments of the seismic waves’ amplitudes andarrival times must be combined according tostandard analytical techniques to give the loca-tion and origin time of the event that gener-ated the signal. The event could be any of sev-eral natural or man-made phenomena, such asearthquakes, nuclear or chemical explosions,meteorite impacts, volcanic eruptions, or rockbursts.

The seismic signals from these events travelfrom their source to individual seismic record-ing stations along different pathways throughanon-uniform Earth. Consequently, no two sig-nals recorded at separate stations will be iden-tical. Even if they came from the same source,the amplitudes, shape, and transit times of the

signals would vary according to the path theytook through the Earth. This fact has an im-portant impact on the results of the calcula-tions used to determine magnitude and location.The results will never have perfect precision.They will be based on averages and will haveassociated with them statements of what thepossible errors in the most probable solutionmight be. Origin times and locations of seis-mic events, the parameters that make up a “de-tection, ” are always based on some averagingof the data from individual stations. To deter-mine the origin time, and location of a seismicevent, the data from several stations must bebrought together at a single place to carry outthe required analysis. Seismic stations thatroutinely send their data to a central locationfor analysis are said to form a network.

The energy radiating from an undergroundnuclear explosion expands outward. Whilemost of the explosive energy is dissipated bycrushing and melting the surrounding rock, a

55

76-584 0 - 88 : QL 3 - 3

56

small fraction is transformed into seismicwaves. These seismic waves propagate out-ward and, in so doing, encounter various bound-aries and rock layers both at the surface anddeep within the Earth. The boundaries causea separation of the seismic waves into a vari-ety of wave types, some of which travel deepthrough the interior of the Earth (body waves)and some of which travel along the surface (sur-face waves). See chapter 3 for a full discussionof wave types and travel paths.

Seismic signals are detected when they aresufficiently above the background noise insome frequency band. Figure 4-1 shows seis-mograms with standard filters designed to en-hance the detection of distant events. In thiscase, the signal can be seen because it is largerthan the noise at high frequency. When dataare in digital form (as they are in the latest

generation of seismic instrumentation), filtersare applied in various frequency bands beforedetection processing.

Relatively sophisticated techniques can nowbe used to detect a seismic signal in the pres-ence of noise. In those cases where three per-pendicular components of ground motion arerecorded using a vertical and two horizontalcomponent seismometers, use can also be madeof the known particle motion of P waves todifferentiate a P wave signal from backgroundnoise. Another technique which can enhancethe probability of detecting a signal requiresa number of closely spaced seismic sensorsknown as an array. The data recorded by thesesensors can be summed together in a mannerwhich takes account of the expected signalpropagation time across the array. The arrayenhances signals from great distance that prop-

Figure 4-1.—Seismic Signals

Signal

Small seismic signals in the presence of seismic noise at four different stations.

SOURCE: U.S. Geological Survey.

57

agate vertically through the array and can re-ject noise that travels horizontally. Therefore,the array summation process tends to enhancethe signal and to reduce the noise.

For many years, most seismic verificationefforts in the United States concentrated onthe use of teleseismic signals. Teleseismic sig-nals are seismic waves which travel to dis-tances greater than 2,000 km and go deepthrough the interior of the Earth. Teleseismicwaves are used because all seismometers mon-itoring Soviet testing are located outside theU.S.S.R. and generally at distances which are

teleseismic to the Soviet test sites. The possi-bility of establishing U.S. seismic stationswithin the U. S. S. R., close to Soviet nucleartesting areas, was discussed seriously as partof negotiations for a Comprehensive Test BanTreaty. Because techniques for monitoring ex-plosions with regional signals were not as wellunderstood as techniques for monitoring withteleseismic data, research efforts to improveregional monitoring greatly increased duringthe late 1970s. Regional distances are definedto be less than 2,000 kilometers and wavespropagating to this distance travel almost en-tirely through the Earth’s outer crustal layers.

LOCATING SEISMIC EVENTS

The procedure for estimating the location ofa seismic event, using seismic data, involvesdetermining four numbers: the latitude and lon-gitude of the event location, the event depth, andthe event origin time. Determination of thesefour numbers requires at least four separatemeasurements from the observed seismic sig-nals. These values are usually taken as the ar-rival time of the P wave at four or more differ-ent seismic stations. In some cases, however,determination of the numbers can be accom-plished with only two stations by using thearrival times of two separate seismic wavesat each station and by using a measure of thedirection of the arriving signal at each station.A relatively poor estimate of location can alsobe obtained using data from only a single array.

The event location process is an iterative onein which one compares calculated arrival times(based on empirical travel-time curves) withthe observed arrival times. The differences be-tween calculated and observed arrival timesare minimized to the extent possible for each

station in the process of determining the loca-tion of the seismic event.

All seismically determined event locationshave some error associated with them. The er-ror results from differences between the ex-pected travel time and the actual travel timeof the waves being measured and from impre-cision in the actual measurements. Generally,these errors are smallest for the P wave (thefirst signal to arrive), and this is why the ca-pability to detect P waves is emphasized. Loca-tion errors are computed as part of the loca-tion estimation process, and they are usuallyrepresented as an ellipse within which theevent is expected to be. Generally, this com-putation is made in such a way that there isa 95 percent probability that the seismic eventoccurred within the area of the error ellipse.Thus, location capability estimates are usuallygiven by specifying the size of this error ellipse.Similarly, there is always an uncertainty asso-ciated with the measured depth of the eventbeneath the Earth’s surface.

58

Box 4-A.—Locating A Seismic EventEstimating the location of a seismic event can be compared to deducing the distance to a light-

ning bolt by timing the interval between the arrival of the flash and the arrival of the sound ofthe thunder. As an example, consider the use of the P wave and S wave, where the flash is the first-arriving P wave and the thunder is the slower traveling S wave. The time interval between thearrival of the P wave and the arrival of the S wave (which travels at about half the speed of theP wave), increases with distance. By measuring the time between these two arrivals and knowingthe different speeds the two waves travel, the distance from the event to the seismometer couldbe determined (figure A). Knowing the distance from several stations allows the location to be pin-pointed (figure B).

2000 4000 6000 6000 10,000

Distance from earthquake (km)

The time required for P, S, and surface waves to travel a given dis-tance can be represented by curves on a graph of travel time againstdistance over the surface. To locate an earthquake the time intervalobserved at a given station is matched against the travel-time curvesfor P and S waves until the distance is found at which the separationbetween the curves agrees with the observed P-S time difference.Knowing the distance from the three stations, A, B, and C, one canlocate the epicenter as in figure B.

Knowing the distance, say X A , of an earthquake from a givenstation, as by the method Figure A, one can say only that theearthquake lies on a circle of radius A, centered on the station. If,however, one also knows the distances from two additionalstations B and C, the three circles centered on the three stations,with radii XA, XB, Xc, intersect uniquely at the point Q, thelocation of the seismic event.

SOURCE: This example is taken from Frank Press and Raymond Seiver, Earth (San Francisco, CA: W. H. Freeman and Company, San Fran-cisco, CA, 1974).

DEFINING MONITORING CAPABILITY

Seismic Magnitude of these tasks is generally described as a func-

A seismic monitoring system must be abletion of a measure-called the seismic magnitude

to detect the occurrence of an explosion, toof an event.

locate the explosion, to identify the explosion, Seismic magnitude was first developed asand to estimate the yield of the explosion. The a means for describing the strength of an earth-capability of a seismic network to perform each quake by measuring the motion recorded on

59

a seismometer. To make sure that the meas-urement was uniform, a standard method wasneeded. The original calculation procedure wasdeveloped in the 1930s by Charles F. Richter,who defined local magnitude, ML, as the log-arithm (to the base 10) of the maximum ampli-tude (in micrometers) of seismic waves observedon a Wood-Andersen torsion seismograph ata distance of 100 kilometers (60 miles) fromthe earthquake.

Subsequently, the definition of seismic mag-nitude has been extended, so that the meas-urement can be made using different types ofseismic waves and at any distance. For bodywaves, the equation for seismic magnitude mb

is:

mb = log (A/T) + B(d,h),

where A is the maximurn vertical displacementof the ground during the first few seconds ofthe P wave, and T is the period of the P wave.The B term is a correction term used to com-pensate for variations in the distance (d) be-tween the seismic event and the recording sta-tion and the depth (h) of the seismic event. Fora seismic event at the surface of the Earth,h=0. The B correction term has been deter-mined as a function of d and h by observingseismic signals from a large number of earth-quakes.

For surface waves, seismic magnitude Ms

can be calculated by a similar equation:

M s = log (A/T) + b log d + C,

where b and c are numbers determined fromexperience. A number of formulas, involvingslightly different values of b and c for MS,have been proposed.

The terms in both the body wave and sur-face wave magnitude equations that are usedto compensate for distance reflect an impor-tant physical phenomenon associated withseismic wave propagate. This phenomenon, re-ferred to as attenuation, can be simply statedas follows: the greater the distance any par-ticular seismic wave travels, the smaller thewave amplitude generally becomes. Attenua-

tion of wave amplitude occurs for a numberof reasons including:

1. the spreading of the wave front over agreater area, thereby reducing the energyat any one point on the wavefront;

2. the dissipation of energy through natu-ral absorption processes; and

3. energy redirection through diffraction,refraction, reflection and scattering of thewave at various boundaries and layerswithin the Earth.

As a consequence, a correction term is neededto obtain the same magnitude measurementfor a given seismic event from data taken atany seismic station. The correction term in-creases the amplitude measurement to com-pensate exactly for the amplitude decreasecaused by the different attenuation factors.

Therefore, if the amplitude of the seismicwave is to be used to estimate the size of theseismic event (whether it is an explosion or anaturally occurring earthquake), a good under-standing of how amplitude decreases with dis-tance is needed for both body and surfacewaves. The distance-dependent numbers in thebody and surface wave equations representaverage corrections which have been developedfrom many observations. In general, these cor-rections will not be exact for any one particu-lar path from a particular seismic event to aparticular sensor. It is important, therefore,either to calibrate the site to receiver path orto compute the magnitude of the event usingseismic signals recorded at a number of well-calibrated stations. If multiple stations areused, the event magnitude is calculated byaveraging the individual station magnitudesfor an event. From this procedure, an averagebody wave magnitude (rob) and an averagesurface wave magnitude (Ms are determinedfor the event.

Obviously, the distance the wave has trav-eled must be known to determine the attenua-tion in amplitude and correct for it. Thereforethe seismic event must be located before itsmagnitude can be determined.

60

Converting Magnitudeto Explosive Yield

The detection and identification capabilitiesof seismic networks are described most con-veniently in terms of seismic magnitudes, typi-cally mb. This measure is used because mb isdirectly related to seismic signal strength.When interpreting capabilities in terms of ex-plosive yield, however, an additional step isrequired to translate mb to kilotons. The samemagnitude value can correspond to yields thatrange over a factor of about 10. Variations inthe magnitude-yield relationship are caused byvariations in the structure of the Earth in thevicinity of the test site (low signal attenuationversus high signal attenuation areas), the ma-terial in which the explosion is emplaced (hard,water-saturated rock versus dry, porous ma-terials), and the way in which the explosion isemplaced (tamped versus detonated in a largecavity designed to muffle the signal).

For example, if an explosion is “well-coupled,”that is, if the energy is well transmitted fromthe explosion to the surrounding rock, an mb

of 4.0 corresponds to an explosion of about 1kt. This relationship is true only for explosionsin hard rock and may vary considerably de-pending on how well the seismic waves aretransmitted through the area’s geology. Inareas that are geologically old and stable, seis-mic waves are transmitted more efficiently. Anmb of 4.0 produced by a well-coupled explo-sion in an area of good transmission might cor-respond to an explosion much smaller than akt. In areas that are geologically young andactive, seismic waves are not transmitted asefficiently and an mb of 4.0 may correspondto a well-coupled explosion larger than 1 kt.

Even greater changes in the relationship be-tween mb and yield can occur if the explosionis intentionally “de-coupled” from the sur-rounding rock in a deliberate attempt to muf-fle the seismic signal. As we will see in chap-ter 6, decoupling can be accomplished at lowyields under some situations by detonating theblast in a large underground cavity. Throughsuch evasion methods, the same 1 kt explosion

that produced a magnitude mb of 4.0 when“well-coupled” might be muffled down to aseismic signal of around mb 2.() at low fre-quencies. Lesser reductions can be accom-plished by detonating the explosion in dry po-rous material.

Because the yield that corresponds to a spe-cific mb depends so much on the scenario thatis being discussed, seismologists generally useseismic magnitude to describe monitoring ca-pabilities. In translating seismic magnitudesto yields, the reader must consider the contextin which the comparison is made. In particular,it should be considered whether the explosionis being recorded in an area of good transmis-sion and whether the explosion is well-coupled.Unless specifically stated, this report trans-lates seismic magnitudes to yields correspond-ing to “tamped” conditions, that is, a well-coupled explosion in hard rock. Situationswhere decoupling is feasible, and the effectsof such decoupling, are discussed in chapter 6.

Seismic Monitoring in ProbabilisticTerms

Whether seismic measurements are made byhand or by computer, some error is involved.Even greater additional errors arise from theimperfect estimates of how well seismic wavestravel through different parts of the Earth andhow well seismic energy is coupled to the Earthduring the explosion. All of these errors resultin some uncertainty in the final determined pa-rameters. This is true whether these parame-ters are event magnitude, location, identifica-tion characteristics, or yield. In all cases,however, it is possible to estimate a confidencefactor in probability terms for the determinedparameter. It is important to realize, therefore,that while the numbers are not presented with100 percent certainty, estimates of the uncer-tainty are known. In general, this uncertaintyis greatest for the small events and decreasesfor the larger events. A discussion of the un-certainty and what it means in terms of na-tional security concerns is presented in chap-ter 2.

61

LIMITATIONS TO SEISMIC

The strength of a seismic signal diminisheswith distance. In general, the closer the seis-mic station is to the source, the stronger thesignal will be. Hence, a principal element ofmonitoring strategy is to get close to areas ofconcern. It follows that the more high qualitystations distributed throughout a given area,the greater the capability will be to detect smallevents.

Seismic Noise

As noted previously, if the Earth were per-fectly still, detecting even the smallest seis-mic event would be easy. However, the Earth’ssurface is in constant motion. This motion isthe result of many different energy sources.Major storms over ocean areas and the result-ant wave action on continental shores causesignificant noise in the 2- to 8-second band.Wind noise and noise from atmospheric pres-sure fronts are particularly prominent onhorizontal-component seismic recordings. Thesemore or less continuous motions of the Earthare referred to as seismic noise or microseisms(figure 4-2). For purposes of siting a seismicstation, it is highly desirable to find an areathat has a low background level of seismicnoise. Generally, the lower the backgroundseismic noise at any station, the smaller willbe the seismic signal which can be detected atthat station.

Cultural activities can also generate seismicnoise that appears in the frequency range usedto monitor nuclear explosions. Generally, thisman-made noise has frequencies higher than1 Hz. Heavy machinery, motors, pumping sta-tions, and mills can all generate observableseismic noise. However, careful siting of seis-mic stations can minimize the problem of mostman-made seismic noise. From a monitoringpoint of view, it is important that noise sur-veys be made prior to the final selection of sitesfor seismic stations. If such sites are negoti-ated within other countries, provisions shouldbe made for relocating the sites should seis-mic noise conditions change.

MONITORING CAPABILITY

Seismic signals and noise are concentratedin various frequency bands. Only the noisewithin the frequency bands in which seismicsignals are observed is a problem. Even strongnoise can be eliminated by filtering as long asthe noise is outside the detection bands of in-terest.

The possibility also exists that a seismic sig-nal from one event can be masked by the seis-mic signal from another event. While this doesindeed happen, it is only a problem for moni-toring at yields around 1 kt or less without in-ternal stations. For events of interest in theU.S.S.R. above a magnitude of about mb 4.0,there are a sufficient number of stations de-tecting the event so that masking of the sig-nal at a couple of stations generally poses noserious problem. For events much below mb

4.0, a number of stations at regional distances(distances less than about 2,000 km) wouldhave to be used to avoid the masking problem.Such stations would be available if the UnitedStates obtains access to data from seismic sta-tions placed within the U.S.S.R.negotiated agreement betweenStates and the Soviet Union.

Reduction of SignalsSource or Sensor

as part of athe United

at

Poor coupling to the geologic media, eitherat the explosion source or at the seismic sen-sor, will act to reduce the amplitude of the seis-mic signal received. If the explosion is in densehard rock or in water saturated rock, the sourcecoupling will be good. If the explosion is in al-luvium, dry porous rock, or within a cavity,the coupling will not be as good. Decouplingan explosion by detonation within a cavity isan important evasion scenario which will bediscussed in chapter 6.

At the seismic sensor, signal reduction canoccur if the sensor is not placed on hard rock.In particular, if there is a layer of soil uponwhich the sensor sits, the signal-to-noise ratioat the sensor can be far less than if the sensor

62

Figure 4-2. —Seismic Noise

-

Background seismic noise at three different stations.

SOURCE: U.S. Geological Survey.

were placed on or within hard rock. Sensorsplaced in boreholes in hard rock provide su-perior coupling to the Earth and also providea more stable environment for the instrumentpackages, with a concomitant reduction innoise.

Seismic Instrumentation

Until recently, the instrumentation that wasavailable for detecting, digitizing, and record-

ing seismic signals did not have the capabilityto record all the signal frequencies of interestwith sufficient range. Further, the mechani-cal and electronic components comprising seis-mic recording systems generate internal noise,which is recorded along with true ground noiseand seismic event signals, and this internalnoise was the limiting factor in recording seis-mic signals in the high frequency range. Spe-cifically, the internal noise of the older designsof high frequency seismic detectors was higher

63

than ground noise at frequencies above about5 Hz. Thus, while ground noise is now knownto decrease with increasing frequency, the sys-tem noise remained constant or increased withincreasing frequency in the older systems.Therefore, trade-offs were made, and the en-tire frequency range of the signal was gener-ally not recorded. Specifically, in the high fre-quency range data was generally not recordedabove 10 Hz and even then was highly con-taminated in the 5-10 Hz range by internal sys-tem noise. Consequently, small seismic events,particularly small explosions, with expectedmaximum signal energy in the high frequencyrange above 5 Hz were not detected becausetheir high frequency signals were below the in-ternal system noise levels. Most existing seis-mic stations are of this type and so are limitedfor nuclear test monitoring.

Today, broadband systems capable of re-cording the entire frequency range with a largedynamic range and with low internal noise areavailable. However, the best high performancesystems are not widely distributed. To estab-lish confident detection-identification capabil-ities using high frequency seismic signals atlow event magnitudes, it will be necessary toexpand the number of high performance sta-tions and to place them in diverse geologic envi-

ronments in order to simulate the requirementsof in-country monitoring.

Seismic Magnitude EstimationProblems

As discussed earlier, the estimate of anevent’s seismic magnitude is made by combin-ing the estimates obtained from many singlestations in an averaging procedure to reducerandom errors. This procedure works well foran event which is neither too small nor toolarge.

For a small event, however, the averagingprocedure can result in a network magnitudevalue which is biased high. This follows fromthe fact that for a small event, the signals willbe small. At those stations where signals fallbelow the noise, the small signal amplitudeswill not be seen. Consequently, only higher am-plitude values from other stations are avail-able for use in computing the network aver-age. With the low values missing, this networkaverage is biased high unless a statistical cor-rection is made.

Figure 4-3 illustrates this effect. All six sta-tions (A through F) record the same magni-tude 4 event. Stations A, B, and C record the

Figure 4-3.—Effect of Noise on Event Magnitude Computation

4.8

4.24.5

3.2

STA-A STA-B STA-C

STA-D

STA-E

I

seismicnoiselevel

Average event magnitude = 4.0 (6 stations) STA-F

Computed average magnitude = 4.5 (3 stations)

SOURCE: Modified from Air Force Technical Applications Center.

.

64

event below average. Stations D, E, and F rec-ord the event above average. Normally, the sta-tions would all average out to a magnitude 4.0.For a small event, however, the stations thatrecord low (A, B, &C) do not record the signalbecause it is below the noise level. Only thestations that record higher (D, E, & F) showthe event. Without the values from the low sta-tions, the calculation of the average is madeusing only the stations that record high. Theresulting calculation biases the average to thevalues of the higher stations, giving a falseaverage magnitude value of (in this example)4.5 for a 4.0 seismic event.

In the past, computed magnitudes of smallevents were systematically biased high in thismanner. As a result, for most of the last 25years, the U.S. capability to detect seismicevents within the U.S.S.R. has, in fact, beensignificantly better than the estimates of thiscapability. Not until the late 1970s was it dem-

onstrated that the small events being detectedwere 0.2 to 0.4 magnitude units smaller thanpreviously thought. In terms of yield, thismeans that the networks were actually capa-ble of detecting events down to half as largeas previously thought possible. Within the lastfew years, analysis procedures have been em-ployed to correct for most of this bias usinga procedure called maximum likelihood esti-mation.

For large events, a similar bias problem usedto exist occasionally, but for a different rea-son. Old seismometers could not record verylarge signals without clipping the signal. Largeramplitude signals were either not available orwere under estimated. The resulting bias oflarge events, however, did not affect estimatesof detection capability and only became a prob-lem in the determination of the size of verylarge events.

SEISMIC NETWORKS

Existing Networks and Arrays

Although many thousands of seismic sta-tions exist around the world, the actual num-ber of stations which routinely report data tonational and international data centers is a fewthousand. For example, in figure 4-4 the 3,500stations are shown that routinely report datato the National Earthquake Information Cen-ter (NEIC), a center in Colorado operated bythe United States Geological Survey. Some ofthese stations report much more often thando others. The instrumentation at these sta-tions is diverse and the quality of the data var-ied. While these stations are very useful forseismic signal detection, they are less usefulfor purposes of magnitude estimation and forresearch requiring stations evenly distributedaround the world.

For purposes of treaty monitoring opera-tions and research, a well distributed networkwith a common set of instrumentation at allstations is most useful. To obtain such a stand-ard network, the United States funded the de-

velopment and deployment of the WorldwideStandardized Seismograph Network (WWSSN)in the early 1960s (figure 4-5). The WWSSNis maintained by the United States Geologi-cal Survey (USGS). The quality and perform-ance of the WWSSN is generally very good,but the recording system is limited in dynamicrange and resolution because of the use of whatis now obsolete analog equipment and also be-cause of high internal noise in the amplifyingequipment. (For example, the data are cur-rently recorded only on photographic paperrecords.)

Beginning in the early 1970s, digital record-ing seismic stations were developed by theUnited States and other countries. The datafrom these stations can be easily processed bydigital computers to enhance the signal-to-noise for signal detection and to analyze thedata for seismic source determination and forresearch purposes. These stations are includedin such networks as:

. The Regional Seismic Test Network(RSTN). These are high quality stations

●

●

● ✎●

●●

✎

.

/’-7’-.

o●

●

●

●

z

● I

●

●

● 9●

●

. ●

. “

66

● ’

67

●

●

●

designed by the Department of Energyand now operated by the USGS. They wereintended to be prototypes of in-countrystations. There are five RSTN stations dis-tributed over North America at inter-station distances that represent monitor-ing in the Soviet Union with 10 internalstations.The NORESS seismic array in Norway.This seismic array is funded by the De-fense Advanced Research Projects Agency(DARPA) and the Department of Energy(DOE). The prototype array is located insouthern Norway, an area thought to begeologically similar to the western partof the Soviet Union.The recently installed China Digital Seis-mic Network (CDSN), which is a coopera-tive program between the People’s Repub-lic of China and the USGS.The Atomic Energy Detection System(AEDS) seismic network. This network isoperated by the Air Force Technical Ap-plications Center (AFTAC). The purposeof the AEDS network is to monitor treatycompliance of the Soviet testing program.Consequently, the stations are located soas to provide coverage primarily of theU.S.S.R. The capabilities of the AEDSnetwork are described in a classified annexto this report. The USGS and the AEDSstations are the main sources of routineinformation on Soviet testing.

In addition to these networks, there is alsoa jointly operated NRDC-Soviet Academy testsite monitoring network in the United Statesand the Soviet Union. The network consistsof three stations in each country around theKazakh and Nevada test sites at distances ofabout zOO kilometers from the boundaries ofeach test site. These stations are supplyinghigh-quality seismic signal data, in the highand intermediate frequency range from 0.1 Hzto about 80 Hz. The stations are designed tobe modern prototypes of the in-country seis-mic stations required for monitoring a lowthreshold test ban treaty and they are notlimited by system noise in the high frequencyrange. Plans call for the addition of five more

such stations distributed across the SovietUnion and for several more to be similarly dis-tributed across the United States.

Planned Networks

There are a number of planned new networksthat will provide increased capability to detect,locate, and characterize seismic events aroundthe world. These networks are being developedby the United States and other countries.

Hypothetical Networks

Existing unclassified networks external tothe U.S.S.R. have an excellent capability formonitoring events with seismic magnitudesgreater than 4.0 within the U.S.S.R. However,for explosions less than a few kt, the possibil-ity exists that the seismic signals from suchexplosions could be reduced through an eva-sion method. To demonstrate a capability todefeat credible evasion scenarios that could beapplied to explosions with yields less than afew kt, seismic stations within the SovietUnion would be necessary.

Obviously, there are a number of require-ments for such internal stations and their data.Among these are the following: the data mustbe provided in an uninterrupted manner; thedata must be of high quality; the seismic noiseat the stations should be low; the operatingparameters of the stations and the character-istics of the data should be completely knownat all times; the data should be available tothe United States within a reasonable timeframe; the stations should be located for ef-fective monitoring of the U. S. S. R.; and any in-terruption or tampering with the operation ofthe station should be detectable by the UnitedStates. Obviously, these requirements canmost easily be achieved by deploying U.S. de-signed and built seismic stations within theU.S.S.R. at sites chosen by the United States.

The number of stations required within theU.S.S.R. is a function of a number of factorsincluding: the threshold level down to whichmonitoring is desired, the seismic noise at thestations, the signal-to-noise enhancement ca-

68

pability of the stations, the signal propagation posed distributions of internal stations capa-characteristics within the U. S. S. R., and the ble of detecting seismic events down to vari-possibilities for various evasion scenarios ous thresholds. The number of internal stationsthought to be effective within different areas proposed for the various distributions rangesof the U.S.S.R. Many seismologists have pro- between 10 and 50.

SEISMIC MONITORING CAPABILITY

Calculating Seismic MonitoringCapability

Calculating the detection capability of ex-isting seismic stations is straightforward. Forhypothetical stations, however, the detectioncapability must be estimated by adopting anumber of assumptions. Because there existsa range of possible assumptions which can beargued to have validity, there are also a rangeof possible capabilities for a network of hypo-thetical internal stations.

For existing stations, the average detectioncapability is easily determined by observingthe number of seismic events detected as afunction of log amplitude (figure 4-6). The sta-tion can be expected to detect all events withina given region down to some magnitude level.A cumulative plot of the detections, such asillustrated in figure 4-6, will show that a straightline can fit these values down to this magni-tude threshold level. This threshold marks the

Figure 4-6.—Detection Capability of a Seismic Station

1,000

o 3

Log amplitude

SOURCE: Modified from Air Force Technical Applications Center

point where the station fails to detect allevents. The 90-percent detection threshold (orany other threshold) can be determined fromthis plot. If the event magnitude rather thanthe observed amplitude is used, it is importantthat all magnitudes used in such plots be cor-rected for low-magnitude bias as previouslydiscussed. By examining all stations of a net-work in this manner, the station detection pa-rameters can be determined and used to com-pute overall network performance for the givenregion.

For hypothetical internal stations, no detec-tion statistics are generally available for com-puting the cumulative detection curve as afunction of magnitude. Therefore, the detec-tion capability must be estimated by assum-ing the following factors: the seismic noise atthe station, the propagation and attenuationcharacteristics of the region through which thesignals will travel, the efficiency of the seis-mic source, and the signal-to-noise ratio re-quired for the signal to be detected. All theabove factors must be evaluated in the fre-quency range assumed to be best for detect-ing a signal. The result of all these factors,when considered together, is to provide an esti-mate of the probability that a signal from asource of a given size and at a given locationwill be detected at a certain station. Individ-ual station detection probabilities are thencombined to determine the overall probabilitythat four or more stations will detect the event.Translating this capability to situations whereevasion might take place requires additionalconsiderations (see ch. 6).

Global Detection Capability

There are many seismic stations that existaround the world from which data can be ob-

69

tained. While no attempt has ever been madeto determine the global detection capabilityof all these stations, a rough estimate couldbe made by reviewing the various reportingbulletins and lists. However, the current globaldetection capability does not really matter be-cause it will soon change as various plannednetworks become installed. Consequently, anaccurate assessment of global capability is bestaddressed by discussing planned networks.

For regions external to the U. S. S. R., and par-ticularly for regions of the Southern Hemi-sphere, the greatest detection and location ca-pability will reside not with the AEDS, butwith a number of existing and planned seis-mic networks which are unclassified. This isa logical consequence of the AEDS being tar-geted primarily at events within the U.S.S.R.In particular, national networks such as thoseof Australia, China, the United States, Italy,and Canada will provide significant global ca-pability.

Given all the national and global data sources,a cautious estimate of the global detection ca-pability by the year 1991 (assuming 90 per-cent confidence of four or more stations detect-ing an event using only open unclassifiedstations) is mb 4.2. For many regions, such asthe Northern Hemisphere, the detection capa-bility will be, of course, much better. There-fore, by 1991, any explosion with a magnitudecorresponding to 1-2 kt well-coupled that is det-onated anywhere on Earth will have a highprobability of being detected and located bynetworks external to the U.S.S.R. Opportuni-ties to evade the seismic network outside theSoviet Union are limited. Because evasionscenarios require large amounts of clandestinework, they are most feasible within the bordersof a closed country such as the Soviet Union.Consequently, monitoring networks are de-signed to target principally the Soviet Union.

Detection Capability Within theU.S.S.R. Using No Internal Stations

Given that a large range of possible networksexists, a few type examples are useful to con-vey a sense of what can be accomplished. For

example, the capability of a hypothetical net-work consisting of a dozen or so seismic ar-rays that are all outside the borders of theSoviet Union can be calculated. A cautious esti-mate is that if such a network were operatedas a high-quality system, it would have 90 per-cent probability of detecting at four or morestations all seismic events within the SovietUnion with a magnitude at least as low as 3.5.This corresponds to an explosion having a yieldbelow 1 kt unless the explosion is decoupled.

The hypothetical detection threshold of mb

3.5 is considered cautious because it is knownthat a greater detection capability might ex-ist at least for parts of the U.S.S.R. For exam-ple, the single large NORSAR array in Nor-way has the potential to achieve detectionthresholds equivalent to an event of mb 2.5 orlower (corresponding to a well-coupled explo-sion between 0.1 and 0.01 kt) overlarge regionsof the Soviet Union.1

Also, fewer stations (fewer than the fourneeded above) may detect much smaller events,and this can provide useful information. How-ever, detection by one or a few stations maynot be adequate to, with high confidence, locateor identify events. Also, reductions of the de-tection threshold must be accompanied by acomparable capability to locate the events (forfocusing other intelligence resources) and toseparate nuclear explosions from earthquakesand legitimate industrial explosions.

Detection Capability Within theU.S.S.R. Using Internal Stations

Seismic stations located internalto a coun-try for the purpose of monitoring have a num-ber of important advantages: improved detec-tion capability, improved location capability,and improved identification capability.

The potential instantaneous detection thresholds for the largeNORSAR array (42 seismometers spread over an area of about3,000 km’), as described in “Teleseismic Detection at High Fre-quencies Using NORSAR Data” by F. Ringdal in IVORSARSemiannual ?’ecfmical Summary, Apr. I-Sept. 30, 1984, are:

West of Ural Mountains-mb 2.0-2.5 (possibly better)Caspian Area–rob 2.0-2.5Semipalatinsk-mb 2.5-3.0Siberia–mb 2.5-3.5

70

Although much debate is associated with thepredicted detection capabilities of internal sta-tions, improved detection capability alone isprobably not of the greatest significance at thistime because the current detection capabilityis already very good. The improvement thatinternal stations will provide to identificationcapability (differentiating explosions from nat-ural events) is by far the most important rea-son for requiring internal stations and shouldbe considered the basic requirement for inter-nal stations. The problem of detecting andidentifying seismic events in the face of vari-ous evasion scenarios will be discussed in thenext two chapters.

Based on cautious assumptions for a net-work of 30 internal arrays or about 50 three-component internal stations, it appears likelythat a detection threshold of mb 2.5 (90 per-cent probability of detection at four or morestations) could be reached. This correspondsto a well-coupled explosion of 0.1-0.01 kt, ora fully decoupled nuclear explosion with a yieldof about 1 kt. Based on more optimistic as-sumptions about the conditions to be encoun-tered and prospective improvements in dataprocessing capability, this same network couldhave a detection capability as low as mb 2.0.Detection capability contours for one such pro-posed 30-array internal network are shown infigure 4-7.

Considerations in ChoosingMonitoring Thresholds

Depending on the number of internal sta-tions, detection capabilities could either in-crease or decrease. The point, however, is thatvery low detection thresholds, down to mag-nitude 2.0, can be achieved. In fact, almost anydesired signal detection level can theoreticallybe obtained by deploying a sufficient numberof internal stations; although there maybe dis-agreements over the number and types of sta-tions needed to achieve a given threshold. Thedisagreements could be resolved as part of alearning process if the internal network is builtup in stages.

Another consideration is that all detectionestimates used in this report are based on a90-percent probability of four or more stationsdetecting an event. While this maybe a pru-dent estimation procedure from the monitor-ing point of view, an evader who did not wishto be caught might adopt a considerably morecautious point of view. (See chapter 2 for a dis-cussion of the relationship between uncertaintyand cheating opportunities.) Such concernsmight be increased by the realization that formany seismic events, there will beat least onestation that will receive the signal from theevent with a large signal-to-noise ratio. The sig-nal will be so large with respect to the noiseat this station that the validity of the signalwill be obvious and will cause a search for otherassociated signals from neighboring seismicstations. The possible occurrence of such a sit-uation would be of concern to a country con-templating a clandestine test.

Throughout all of this discussion it must bekept in mind that an improvement in detec-tion capability does not necessarily corresponddirectly to an improvement in our monitoringcapability. Although a reduced detection thresh-old must be accompanied by a reduced iden-tification threshold; occurrences such as indus-trial explosions might ultimately limit theidentification threshold. For example, the esti-mate of detection capability for internal sta-tions which is given above, mb 2.0 to 2.5, cor-responds to a decoupled explosion of about 1kt. A decoupled 1 kt explosion produces thesame mb signal as a 1/70 kt (15) ton chemicalexplosion. At such small magnitude levels,there are hundreds of chemical explosions det-onated in any given month in the U.S.S.R. andin the United States.

Use of High-Frequency Data

It has long been known to seismologists thatseismic signals of moderate to high frequen-cies can indeed be detected at large distancesfrom the source under favorable geological con-ditions. Such is the case in the eastern UnitedStates, and it is generally thought that thisis also true of most of the Soviet Union. In con-

71

72

trast, tectonic regions such as the westernUnited States and the southern fringe of theU.S.S.R. are generally characterized by strongerattenuation of high frequencies, which are there-fore lost beyond relatively short distances.

The design and development of a nuclearmonitoring strategy based on high-frequencyseismic signals calls for:

● A determination of Earth structure withinand around the region to be monitored,and an evaluation of its signal transmis-sion characteristics at high frequencies.

● A methodology for identifying and select-ing sites with low ground noise, and equip-ping such sites with high performance sen-sors and recording systems, so as to achievethe largest possible signal-to-noise ratio.

● A reliable understanding of the high-fre-quency radiation of natural (earthquakes)and man-made (explosions) seismic sources.Although empirical evidence based on di-rect observations is sufficient in principle,a predictive capability, based on theory,is required to assess properly new and un-tested monitoring conditions.

Such requirements parallel in every respectthe usual constraints placed on standard mon-itoring systems. However, direct experimen-tation pertinent to high-frequency monitoringhas been rather limited so far, and the relevantdata available today are neither abundant nordiverse. Consequently, an assessment of whetherthese requirements can be met relies of neces-sity on some degree of extrapolation from ourpresent experience, based on theoretical argu-ments and models. This situation leaves roomfor debate and even controversy.2

Recently, it has been argued that the capa-bility to detect and identify low-yield nuclearexplosions could be greatly improved by using

‘High quality seismic data is now becoming available fromthe NRDC-Soviet Academy of Sciences stations in the SovietUnion and the United States, with more widely distributed sta-tions to be added in 1988 in both countries. This data may helpreduce the necessity for extrapolation and decrease the uncer-tainties that foster the debate.

high-frequency (30 -40 Hz) seismic data.3Themajor points of the argument are:

●

●

●

that natural seismic ground noise levelsare very low at high frequencies, and thatlarge seasonal fluctuations are not antic-ipated;that careful station selection could makeit possible to emplace seismic sensors inparticularly quiet sites; andthat present seismic recording technologyallows high-fidelity recording; by suppres-sion of system-generated noise to levelsbelow ground noise even at high frequen-cies and at quiet sites.

Advocates of high-frequency monitoring ex-plain the efficiency of high-frequency wavepropagation observed in the North Americanshield in terms of a simple model for attenua-tion of seismic waves in stable continental re-gions and argue that the model applies as wellto stable continental Eurasia. Finally, they ar-

retical models of earthquakes and undergroundexplosions provide adequate predictive esti-mates of the relative production of high-fre-quency energy by various seismic sources andjustify their choice by comparison with limitedobservations.

Based on these arguments and a systematicmodeling procedure, some seismologists rea-son that the most favorable signal-to-noise ra-tio for detection of low yield (i.e. small magni-tude) events in stable continental areas will befound at moderate to high frequencies (about30 Hz). They further infer that a well-designednetwork of 25 internal and 15 external high-quality stations using high frequencies wouldyield multi-station, high signal-to-noise detec-tion of fully decoupled 1-kt explosions from anyof the potential decoupling sites within theU. S. S. R., and result in a monitoring capabil-ity at the l-kt level.

On the other hand, the inference that suchsignificant benefits would necessarily accrue

‘For example, J. F. Evemden, C. B. Archarnbeau, and E. Cran-swick, “An Evaluation of seismic Decoupling and UndergroundNuclear Test Monitoring Using High-Frequency Seismic Data, ”Reviews of Geophysics, vol. 24, No. 2, 1986, pp. 143-215.

73

by relying on high-frequency recordings hasbeen strongly questioned in the seismologicalcommunity. Indeed many scientists feel thatthe case is currently unproven. Major pointsof disagreement include:

● The concern that the theoretical seismicsource models used so far in the analysisdescribed above are too simple. Studiesaimed at constructing more realistic modelsindicate that the high-frequency wavesgenerated by seismic sources are stronglyaffected by complexities of source be-havior that the simple models do not takeinto account. On the other hand, advo-cates of high frequency seismic monitor-ing believe that the models they have usedhave successfully predicted a number ofcharacteristics of seismic sources thatwere subsequently verified and that noneof the many well-documented observa-tions of seismic wave characteristics fromlarge events are in conflict with their theo-retical model predictions. Thus they ar-gue that the model predictions, for some-what smaller events at somewhat higherfrequencies than are ordinarily studied,are reasonable extrapolations.

● The concern that it may be difficult toidentify candidate station sites where thehigh-frequency noise is sufficiently low topermit actual realization of the desiredbenefits. Experience to date is limited, andone does not really know whether a givensite is suitable until it has been occupiedand studied for at least a year. On theother hand, advocates of high frequencymonitoring feel that suitable low-noisesites are not at all rare and can be rathereasily found in most, if not all, geologicenvironments within the continents. Theyargue that stations selected so far havehad adequately low high-frequency noisecharacteristics and the selection processwas neither difficult nor time consuming.They conjecture that doubts are based onmisidentification of high-frequency inter-nal seismic recording system noise asground noise, and that once high-perfor-

mance systems with low system noise be-come more wide-spread, this concern willdisappear.

● The concern that observations which canbe employed to test directly the validityof the proposed use of high frequencies areas yet quite scant, and their interpreta-tion is not free of ambiguities. For exam-ple, the characterization of source spec-tra and the propagation and attenuationof high-frequency waves remain issueswhich are not resolved unequivocally byobservations, and yet are critical to theformulation of a high-frequency monitor-ing strategy. Similarly, available dataoften exhibit an optimal signal-to-noise ra-tio at frequencies near 10 Hz, in apparentdisagreement with the arguments enun-ciated earlier. In response to these con-cerns, proponents argue that the NRDC-Soviet observations of signal-to-noise ra-tios greater than 1 and out to frequenciesabove 20 Hz provides evidence for thepotential of high frequency monitoring.While proponents of high frequency mon-itoring agree that the observations of thelargest signal to noise ratios for signalsfrom seismic events often occur near 10Hz, this is not in disagreement with thepredictions or technical arguments ad-vanced for high-frequency monitoring.They argue that these observations are ob-tained from seismic receivers with systemnoise that is greater than ground noise atfrequencies above 10 Hz, and that manyof the observations are from industrial ex-plosions that are ripple-fired and so areexpected to have lower high-frequencycontent than a small nuclear test.

● The concern stated earlier that the moti-vation for the proposed high-frequencymonitoring approach uses contested theo-retical arguments and models to extrapo-late from our present experience and thusattempt to guide further steps towards asignificantly improved capability. In thepresent case, models are used to extrapo-late both toward high frequencies andtoward small yields, and just how far one

74

may extrapolate safely remains a matterof debate. The proponents of high fre-quency monitoring agree that the data re-lating to high frequency monitoring islimited with respect to the geologic re-gions to which it pertains. Furthermore,it is clear that experience in the system-atic detection and identification of verysmall seismic events using high-frequencydata is absent and that, as a consequence,it has been necessary to extrapolate fromexperience with larger seismic eventswhere lower frequency data is used. Propo-nents believe, however, that what limiteddata is available does support the mostcritically important predictions; these be-ing the apparent availability of low-noisesites and the efficiency of high-frequencywave propagation to large distances.

These controversial aspects notwithstand-ing, there is general agreement among seismol-ogists that good signal-to-noise ratios persistto higher frequencies than those used routinelytoday for nuclear monitoring. In particular,data in the 10-20 Hz band show clear signalswhich are undoubtedly not used optimally.Given the fact that recording of even higherfrequencies is demonstrably feasible in somesituations, and given the potential advantagesfor low-yield monitoring, the augmentation ofour experience with such data, the concomi-tant continued development of appropriateanalysis techniques to deal with them, and thevalidation of the models used in their interpre-tation are goals to be pursued aggressively.Not until a sufficient body of well-documented

observations of this nature has been collectedcan we expect to achieve a broad consensusabout the performance of high-frequency mon-itoring systems.

Arrays v. Three-Component Stations

Both small-aperture, vertical-component ar-rays such as NORESS, and three-component,single-site stations such as the RSTN stationhave been considered for use as internal sta-tions. In choosing which to use, it should berealized that many combinations of arrays andsingle stations will provide the same capabil-ity. For example, for any array network, thereis a single station network with comparablecapability; but the network of single stationsprobably requires about twice as many sites.Although a single array has advantages overa single three-component station (see chapter3), for monitoring purposes it is preferable tohave a large number of station sites with three-component stations rather than to have a smallnumber of sites with arrays. This is true be-cause it permits better accommodation to de-tails of regional geology, and better protectionagainst noise sources temporarily reducing ca-pability of the network as a whole. However,if the number of sites is limited by negotiatedagreement, but the instrumentation can in-clude either arrays or three-component sta-tions, then arrays are preferable. This is trueboth because of the inherent redundancy of ar-rays and their somewhat better signal-to-noiseenhancement capability over single three-com-ponent stations.