27 September 1968, Volume 161, Number 3848

The Denver Earthquakes

Disposal of waste fluids by injection into a deep wellhas triggered earthquakes near Denver, Colorado.

J. H. Healy, W. W. Rubey, D. T. Griggs. C. B. Raleigh

Scientists and public officials areseriously considering the question ofwhether removal of fluid from a deeplyburied reservoir will reduce the likeli-hood of a destructive earthquake nearDenver, Colorado. The question at is-sue in the discussions is not so ab-surd as it might seem. It is widely,though not unanimously, held that in-jection of chemical-waste fluid into thereservoir in the Denver basin triggeredthe earthquakes. We attempt here topresent the statistical evidence for cor-relating the two events—fluid injectionand earthquakes—and to develop a hy-pothesis relating the two as cause andeffect.

In 1961 a deep disposal well wascompleted for the U.S. Army at theRocky Mountain Arsenal, northeast ofDenver, Colorado. The well was drilledthrough 3638 meters of nearly flat-lyingsedimentary rocks in the Denver basininto Precambrian crystalline rocks; itsdepth at completion was 3671 meters.The disposal of waste fluids from chem-

, 'cal-manufacturing operations at the ar-senal had been a difficult problem, forwhich the deep well appeared to be an

j "deal solution. Injection of fluids on a' routine basis was begun on 8 March

1962 and continued through 30 Septem-. ber 1963 at an average rate of about 21

million liters per month. No fluid was

Dn. Healy and Raleigh are affiliated with">e U.S. Geological Surrey, Mcnlo Park, Cali-

, «>rnia; Drs. Rubey and Griggs. with the Institutei *f Geophysics and Planetary Physics, University

<* California, Los Angeles.

;., W SEPTEMBER 196*

injected from October 1963 to August1964. Then for a time fluid was putinto the well under gravity flow at anaverage rate of about 7.5 million litersper month. Gravity flow was continueduntil 6 April 1965, when injection un-der pressure was resumed, at an averagerate of 17 million liters per month. On20 February 1966 injection of fluid wasstopped because of a suggested connec-tion between the well and earthquakesin the Denver area.

Two seismograph stations were oper-ating in the Denver area in 1962, oneby the Colorado School of Mines, atBergen Park, about 34 kilometers westof Denver, the other at Regis Collegein Denver. Both stations began to re-cord earthquakes from the region north-east of Denver, starting on 24 April1962. As the sequence continued, theU.S. Geological Survey established sev-eral additional stations, and Yung-LiangWang, a graduate student at the Colo-rado School of Mines, undertook astudy of all the available seismic re-cordings. He located many of the earth-quakes (/) within a region about 75kilometers long, 40'kilometers wide,and 45 kilometers deep (Fig. 1, left).It was pointed out later that most ofthe earthquakes located by Wang werewithin 8 kilometers of the disposal well.

Father Joseph V. Downey, directorof the Regis College Seismological Ob-servatory-, was among the first to suggestthe possibility of a relationship betweenthe disposal well and the earthquakes.

In November 1965 David Evans (2), aconsulting geologist in Denver, showeda correlation between the volumes offluid injected into the well and the num-ber of earthquakes detected at BergenPark, and publicly suggested that a di-rect relation did exist (Fig. 1, right).

The proximity of the earthquakes tothe Denver metropolitan area createdconsiderable public interest and con-cern. A number of the larger earth-quakes, of Richter magnitude between3 and 4, were felt over a wide area,and minor damage was reported nearthe epicenters. The sudden appearanceof seismic activity close to a major cityposed serious questions, and the possi-bility that the earthquakes were causedby operations at the Rocky MountainArsenal had to be evaluated as quicklyas possible.

The Preliminary Program

The U.S. Army Corps of Engineerswas called upon for technical supportand advice, and the U.S. GeologicalSurvey, in cooperation with the Corpsof Engineers, began a program of in-vestigation to evaluate the Evans theory.The Colorado School of Mines playeda major role in those investigations,with support from the State of Colo-rado, the Corps of Engineers, and theEnvironmental Science Services Admin-istration.

A search of the available instrumen-tal and historical records was one ofthe first investigations undertaken. Anyearthquakes that had occurred beforethe start of fluid injection would lessenthe correlation between water injectionand earthquake occurrence. The seismo-graph station at Bergen Park beganoperation only a few months beforethe start of water injection, and noearthquakes were recorded from theDenver area during that period. Thestation at Regis College had been inoperation since 1909, but it is locatedin an area of high background noiseand, for most of its history, was oper-ated at low magnification. Thus, earth-quakes of small magnitude could have

1301

Table 1. Magnitude and frequency of occurrence of Denver earthquakes.

Magnitude ''

196:196319641<*6<19661967

TotalAverage!

7;89

26168

61

62

478

83.2

2934

864

IS:9

is:30.5

•* <; "• 9

49

62*

-,156:

9.4

:3

6•v

4

17

2.6

11

4

14

111.4

11

1 1

4 30.4

189284

72550186306

1584

• To the nearest ( l . l -mapnitudc unityearly activity 1962 66.

'• Total includes all ear thquakes reported. 1 Average

escaped detection hy the local stationsand the regional networks.

Between 1954 and 1959 a short-period seismograph station was operatedby Warren Longley at the University ofColorado. Boulder. A search of the rec-ords from that station (3) revealed 13events that might have been earth-quakes in the Denver area, but all hadoccurred during normal working hoursand were probably the result of con-struction blasting or disposal of explo-sives at the arsenal.

Hadsell (4). in a search of news-paper accounts and other historic rec-ords, uncovered reports of a number ofearthquakes in Colorado in historictimes. None had epicenters near thezone of current activity, with the pos-sible exception of an earthquake in1882 that was felt in the Denver-Boul-der area as well as at other, widelyseparated points in Colorado. It appearsthat there is no evidence of seismic ac-tivity before 1962 similar to the earth-quakes that have occurred since 1962.

A very dense network of seismicstations was established by the U.S.Geological Survey in the vicinity of the

arsenal well, to obtain accurate loca-tions of earthquake hypocenters. Slight-ly modified explosion seismology equip-ment was used, in eight small arrays.Each array consisted of six verticalseismometers at half-kilometer inter-vals, arranged in an L-shaped pattern,and two horizontal seismometers lo-cated at one of the vertical-seismometerpositions. This special network was op-erated during January and February of1966 for about 6 hours each day.'. Dur-ing this 2-month period, between oneand five earthquakes large enough to belocated occurred each day during thehours of recording. A survey of theseismic velocities of the rocks pene-trated by the well and a nearby seismicrefraction profile were used to deter-mine a velocity-depth function (5).The dense net of seismic stations andthe unusually good control on seismicvelocities made it possible to locatesmall earthquakes with a high degreeof accuracy. Well-recorded earthquakeswere located with a precision of 0.3kilometer in the horizontal plane and0.5 kilometer in the vertical plane. Notall the earthquakes were so well record-

ed, but all of the calculated locationsare thought to be within 1 kilometer ofthe true locations.

Sixty-two earthquakes for which ac-curate locations could be obtained oc-curred in a roughly ellipsoidal zoneabout 10 kilometers long and 3 kilo-meters wide, which includes the disposalwell (Fig. 2. top). The long axis of thezone trends N 60°W. The depths of theearthquakes ranged from 4.5 to 5.5kilometers.

The first motion of compressional-wave arrivals reveals a pattern that isconsistent with right-lateral strike-slipmotion along fault surfaces that areparallel to the trend of the zone ofepicenters. In many cases a nodal planecrossed a single array, showing a changein the direction of first motion withinthe half-kilometer interval between seis-mometers (Fig. 3).

First motions on the surface over thehypocenter of a vertical strike-slip faul thave a quadrantal pattern. Most of thesurface first-motion patterns in the studydiscussed here were quadrantal in na-ture (Fig. 4).

Following completion of the prelimi-nary program, the Colorado School ofMines installed additional short-periodseismograph stations and the U.S. Gco-logical Survey installed a 17-elementarray over the zone of hypocentors.These networks are continuing to recordearthquakes.

Change in the Pattern of Activity

Injection of fluids in the disposal wellwas terminated in February 1966. bi»earthquakes have continued at a ratevarying from 4 to 71 per month. &

£2

3f40'lIDS' 10' 104*20'

Longitude

Fig. 1. Observations on which David Evans in 1965 based his theoryRocky Mountain Arsenal. Denver. Colorado. (Left) Epicenters (solidfrom the Bergen Park and Regis College stations and from temporary U.S. GeologicalEvans, between the number of earthquakes and the volume of fluid injected.

1302

EARTHQUAKE FREQUENCY__

L

ry of the relation between fluid injection and earthquake* •' ^id circles) of earthquakes as calculated by Wang i'sir|f ^iry U.S. Geological Survey stations. (Right ) Correl:il">ns-

ScrENCE. 161

'£*''•"

at the Bergen Park Observa-tory. During that period, hundreds ofvery small earthquakes were detectedon the U.S. Geological Survey networkcentered on the hypocentral zone. On10 April 1967, the largest earthquakeof the series to that date shook theDenver area. Its magnitude was esti-mated at 5.0 by Maurice Major (6),on the basis of the Bergen Park Obser-ntory records. The U.S. GeologicalSurvey network located the hypocenterWhin the zone of previous activity. Infact, all the earthquakes located fellwithin or close to the bounds definedby the 2-month preliminary study (Fig.2, top). A number of aftershocks oftte 10 April earthquake define a linearpattern which trends about N 60°W (7)(Fig. 2, bottom), as the earlier earth-quake locations did. From the statisticsfor the frequency and magnitude of theentire sequence, the large earthquakeof 10 April might reasonably have beenexpected. But there was more to come.On 9 August there was a second largeearthquake of magnitude between 5'.4tnd 5V2. and on 26 November therewas a third, of magnitude 5.1 (Fig. 2,bottom). The three large earthquakes,*ith their foreshocks and aftershocks.introduce a dramatic change in the pat-tern of activity.

Some seismologists have argued thatfte magnitude-frequency plot, a histo-pam of the common logarithm of theManber of earthquakes relative to theirMgnitude, can be used to establish thete»ri of seismicity in an area and to|*edjct the rates of recurrence of largettftthquakes over a period of years. AWnskJerable number of data support&b view, but there are some startlingexceptions (8).

With the exception of the Los An-Basin, the magnitude-frequencyams for all the southern Califor-

il" regions studied can be fitted withijitraight line having a slope between3B.8 and —1.02. The average for south-**« California is —0.86,

The magnitude-frequency relation-*kip for earthquakes in Denver was de-fcrtnined first by Yung-Liang Wang (/).*fo obtained a value of —0.78 for the

. A recent determination of —0.82made by Major and Wideman (9).

|i» the basis of data obtained through*'"~ 1967.

:y the data from the Bergen Parkatory were used in determining

. . ^itudes for the Denver earthquakes."*fcause it is the only station that hasS*n in operation at high magnificationftroughout the period of activity.

•^SEPTEMBER 1968•Sff."

Recently, Ruth Simon at the Colo-rado School of Mines has reexaminedall the Denver earthquake seismogramsand has prepared a new list of earth-quakes for the entire period of earth-quake activity ( 6 ) . These data wereused in preparing the magnitude-fre-quency plots given here. The data onmagnitude and frequency, by year, aregiven in Table 1.

The yearly plots (Fig. 5A) reveal aremarkably consistent pattern from1962 through 1966. with the possibleexception of 1964, for which the slopeis somewhat lower. The number ofearthquakes recorded in 1964 is toosmall to provide an adequate measureof the seismic activity.

With the exception of the data for1967, the available data can be repre-

Carthquake Epicenter

DENVER Aftershocks of10 April 1967 earthquake

Aftershocks of9 August 1967 earthquake

Recording location

| 0 I km

DENVER

Fig. 2. (Top) Epicenters of earthquakes located in January and February of 1966 bymeans of a dense network of temporary seismic stations. The locations are accurateto within 1 kilometer. (Bottom) Location of ( i ) the earthquake of 10 April 1967 andits aftershocks: (ii) aftershocks of the event of 9 August 1967; and (iii) the epicenterof the 26 November 1967 earthquake.

1303

VI

V2

V3

V4

V5

V6i•">• , ,-•• .V vw.—. V.V.-.V, v. ^

"— 1 second ->

Fig. 3. Seismogram (27 December 1965) showing compressional (upward motion) anddilatational (downward motion) arrivals. The nodal plane apparently intersects thearray near seismometer VI.

sented by the relationship log N = a -rbM. where M is the Richter magnitude.N is the number of earthquakes peryear with magnitude between M — 0.25and A/ ~ 0.25. a is a constant whichindicates the level of seismic act ivi tyduring the period studied, and h is aconstant which determines the slope(Table 2).

The average for the data for theyears 1962 through 1966 suggests thatone earthquake with a magnitude ofabout 4.5 could be expected within a5-year period, the total period of earth-quake activity. The rate of recurrencefor earthquakes of magnitude between6.0 and 6.4 is more than 100 years. Theoccurrence of a single earthquake inApril 1967 having a magnitude of 5.0was not in serious conflict with expecta-tion. However, the two succeeding

earthquakes of magnitude greater than5 in 1967 clearly showed that the earth-quakes were not following this statisti-cal pattern, established from 1962 to1966. which predicts about three suchearthquakes in 60 years (Fig. 5B) .

When the frequency-magnitude rela-tionship for 1967 is examined moreclosely, it becomes clear that a signif-icant change has occurred. The datafor earthquakes smaller than magnitude4.5 are best fitted by a straight linehaving a slope of —0.61. which is sub-stantially smaller than the magnitude-frequency slope for other years (Fig.5B). However, because of the occur-rence of three earthquakes with magni-tude between 5.0 and 5.5. with noearthquakes in the magnitude range4.5 to 4.9. it is not possible to fit allthe points for 1967 with a straight line.

*

TREND OF EPICENTRALZONE DISPOSAL WELL '

NODAL PLANES OFRIGHT LATERAL FAULTS

J.

Fig. 4. Seismic radiation patterns of first arrivals from epicentral locations. (Blackquadrants) Compressional first motions; (crosses) azimuthal orientations of the near-vertical nodal planes, one of which is the fault plane for the event. The compass rosediagram at lower left gives azimuthal orientations of right-lateral strike-slip faults.

1304

The significance of this change in theearthquake pattern is more clearK re-vealed if we consider the seismic enery\radiated as a funct ion of magni tude.Richter (10. p. 366) gives the follow-ing relationship between energy, E. andmagnitude. M: log E = 11.4 — 1.5 A /From this relationship, a graph ofcumulative energy release for each 6-month period was prepared (Fig. 6).

Because the energy of earthquake?increases by a factor of about 31 foreach 1-unit increase in magnitude, thethree earthquakes with magnitudegreater than 5 that occurred in 1967account for almost all of the energyreleased in the earthquake series. Itseems cle;;r that something new is hap-pening, with unknown portent for thefuture.

Correlation of Earthquakes

with Fluid Injection

We have, to this point, presented theinformation more or less in the orderin which it became available, so thatthe reader can reconstruct the historicalsequence of events. In the rest of thearticle we depart from this chronolog-ical presentation and turn to the formu-lation of a physical model that, wesuggest, explains the observations. Butfirst, an approximate statistical com-parison can be made. What is the prob-ability that an earthquake sequencesuch as that at Denver could occur h>pure chance close to the time and placeof the injection of fluids into the base-ment rock? The occurrence of swarmsof small earthquakes in Colorado isprobably a more common event thanmost people recognize. During 'hc

course of our work, several groups piearthquakes were noted that might havebeen similar to the Denver earthquakesAn earthquake sequence is occurringnear Rangely, Colorado: another i* oc

curring some distance to the north «<Denver, perhaps in Wyoming: and 0'ers may be occurring in the San Ju J

Mountains of southern Colorado. M1^of this activity was detected by onbfew stations and has not been sllld*f

in detail. It is probable that mostthese areas are somewhat less a^ ̂than the area near Denver. Hadse^(4) list of Colorado earthquakes shei ^70 earthquakes between Decem

for

1870 and January 1967. The Iisithe years prior to 1962 is based ̂reports of felt earthquakes. The 1>S ^1962 and subsequent years incluo"

SCIENCE. VOl. '

earthquakes in Colorado, exclusive ofthe Denver area, of magnitude greaterthan 3.0 and all earthquakes in theDenver sequence of magnitude greaterthan 3.1. Of the 70 earthquakes. 28were in the Denver sequence. We cansay conservatively that not more thanten earthquake swarms similar to theDenver swarm are occurring in Colo-rado at any one time. Colorado has anarea of 270,000 square kilometers. Theaccurately located earthquakes in theDenver sequence are all contained in anarea of less than 65 square kilometers.The probability of finding a swarm in arandomly selected 65-square-kilometerarea is therefore 1 in 4150.

What is the probability that an earth-quake swarm would start within 7weeks of the beginning of a fluid-injec-tion program? The only known possiblysimilar seismic activity near the disposalwell is an earthquake that occurred in1882. The location of this earthquakeis given in Hadsell's list as Vail Pass,but it was felt in the Denver area, soit may possibly have occurred near the

Table 2. Values of constants a and b (seetext ) computed by least squares.

Year

196:19631964196?19661967

a

3.063.332.433.653.192.77

h

-0.80- .85- .64— .8'- .90- .60

present location of the disposal well.If one assumes that the earthquake

of 1882 occurred near Denver ratherthan at Vail Pass and that the recur-rence rate for this activity was 80 years,the probability that an earthquakeswarm would start within 7 weeks ofthe beginning of fluid injection in thedisposal well is 1 in 600. The jointprobability that the earthquakes wouldbe so closely associated in both timeand space, with the disposal well is1/600 X 1/4150 = 1/2.500.000. Theoccurrence of a natural earthquakeswarm so closely related to the disposalwell would thus be an extremely un-likely coincidence.

Since shallow earthquakes are causedby movements along faults, seismic ac-tivity at the level of the Denver earth-quake sequence will, if repeated at fre-quent intervals through geologic history,probably result in large tectonic dis-placements. These displacements shouldproduce recognizable deformation ofthe sedimentary rocks overlying theearthquake zone. A seismic-reflectionsurvey of the epicentral region of theDenver earthquakes revealed no exten-sive folding or faulting in the sedimen-tary rocks overlying the earthquakes( I I ) - Several small faults, of displace-ment less than 30 meters, were discov-ered, and topographic relief over thearea of the survey was less than 60meters. This evidence suggests that therehas been very little seismic activity atthis site since Precambrian time.

In a search for clues to the earth-quake mechanism, the original pressurecharts were reexamined and, whereverpossible, the average daily wellheadpressure and the volume of fluids in-jected were redetermined. The daily

100

.10

1.7 2.2 2.7 3.2 3.7 4.2 4.7 52 5.7Magnitude

*^fc 5. (A) Frequency-magnitude relationships for the Denver**rthqnakc series. The lines are least-squares fits to histogramsfor each year and for the total period. The histograms werePrepared in 0.5-magnitude units. (B) Comparison of the f re-gency-magnitude relationship (seismicity) for 1967 with thetv«age for the preceding 5 years.

*7 SEPTEMBER 1968

0.1

l 1 1 r-

+ Averoge seismicityfor 1962-1966

• Seismicity for 1967

1.7 2.2 2.7 3.2 3.7 4.2 4.7 5.2 5.7 6.2Magnitude

1305

data (Fig. 7A) were used to determinean average pressure for each month.and these averages were plotted againstthe number of earthquakes per month(Fig. 7B). Figure 7 shows that thereis a strong correlation between fluidpressure and the level of seismic ac-tivity for the years 1962-66. but theincrease in seismic activity in 1967 ap-pears to rule out any simple direct rela-tionship between these parameters. Itcould be argued that the increase inseismic activity during a period of de-creasing fluid pressure is not consistentwith the theory that fluid pressure trig-gered the earthquakes, but we will showthat this activity pattern when a zoneof increased pressure is spreading out-ward from the well is not surprising.

The cross correlation between theearthquake series and the fluid-pressureseries was calculated (Fig. 8A). to de-termine whether there was a lag be-tween peak fluid pressure and peakseismic activity. The cross correlationwas calculated as:

where X = data from the earthquakeseries. )' = data from the pressureseries, Rj.v(p) = cross covariance orcross correlation, n = number of dis-crete data points, and p = lag: for — p.x and y are reversed in the above equa-tions. In actual computation, the pro-cedure of Lee and Cox (12) was adoptedto treat data gaps in the pressureseries.

The cross correlation is not symmet-rical and indicates that the peak ofseismic activity follows the peak pres-sures by approximately 10 days. Thelack of symmetry in the cross correla-tion is a consequence of continued seis-mic activity for a number of monthsafter the injection of fluid.

For comparison, a correlation co-efficient was computed by means of theRank difference method (13). with theearthquake and pressure data specifiedin 10-day intervals (Fig. 8B).

Mechanism of the Earthquakes

An understanding of the mechanismof these earthquakes is of great impor-tance, not only for controlling thecurrent sequence at Denver but alsofor predicting where other such se-quences might occur. It now seems

1306

I962i 1963 I 1964 ! 1965 ; 1966 I 1967 I

Fig. 6. Cumulat ive seismic-energy release inDenver earthquakes, by 6-month periods.

clear that, contrary to some early con-jecture. the energy released by theearthquakes was stored in the basementrock as nonhydrostatic elastic strain.The evidence pointing to a tectonicorigin for the elastic strain energy re-leased is twofold.

1 ) The frequency-magnitude rela-tionship for the early earthquakes issimilar to that for tectonically activeareas such as southern California. \

2) The seismic-radiation patterns'are consistent wi th right-lateral strike-slip motion on vertical fault planesaligned with the trend of the seismiczone. The prolongation of the epicen-tral zone in a west-northwestern direc-tion parallel to one of the two possiblefault-plane sets (Fig. 4) strongly sug-gests that a zone of vertical fracturesexisted along the trend prior to theinjection of fluid. Examination of corefrom the basement rock has shownthe presence of vertical fractures priorto such injection (14). These observa-tions suggest that the earthquakes areproduced by a regional stress field oftectonic origin.

From this evidence we consider ithighly probable that the release of thestored tectonic strain was triggered bythe injection of fluid into the basementrock. This hypothesis has a rationalbasis established theoretically by Hub-bert and Rubey (75) and tested exper-imentally by several workers. In arock containing fluid in its pore spacesat a pressure p, the total stress S isresolved into two components (16), aneutral stress p and an effective stress a,

The neutral stress p is the pressure ofthe ambient fluid and has no shearcomponents. Consequently, faulting orother nonhydrostatic deformation of therock must be in response only to theeffective stress a. In the fracture of

rocks at high pressure, the shear stress- on the fau l t plane at failure is c i \enby the empirical relation

T = T.. — 5,, tan a 1 1 >.

where 5,, is the normal stress acrossthe fault plane, tan <i> is the coefficientof friction, and -„ is the coheshestrength. If a pore fluid is present atpressure p. Eq. 1 becomes

T — r, -i- (S,, — p) tan 0

<r,. = S,. — p

T = T. -j- ff. tan 0 '-'

The effect of increasing pore pressureis to reduce the frictional resistance tofracture by decreasing the effective nor-mal stress, tr,,, across the fracture plane

The parameters in Eq. 2 can be eval-uated for the Denver earthquakes onthe basis of the following assumption^and relations between T. a,,, and theprincipal stresses.

P: — ffs . *r r= Sin 2a ( 3 )

cos 2<. 14)

where a is the angle between <r, and thenormal to the fracture plane.

According to conventional mode!-of the dynamics of faulting ( / " > •strike-slip faulting occurs where theleast and greatest principal stresses archorizontal. The magnitude of the leaMprincipal stress. S3, can be est imatedfrom data on pressure relative to thevolume injected.

The rate of injection of fluid, dnrmcperiods of rapid injection, was found tfvary approximately linearly with I1""1

pressure down to a bottom-hole pri>-sure of about 362 bars. Below th i spressure the injection rate was negligi-ble, suggesting a sharp discontinuity inthe reservoir characteristics with pi"cv

sure. Over the 5 days preceding shut-down of the injection program, on 20

February 1966,'the average bottom-holepressure was 368 bars ( I S ) : the ; i \ t f 'age injection rate was 114 lit^ Pcr

minute. Within 1 day after shutdown,the fluid level in the well dropped 3|a rate indicating that injection was con-tinuing at 0.076 liter per minute, at anaverage bottom-hole pressure of - -bars. A decrease of 6 bars in "u^pressure resulted in reduction of 'injection rate by a factor of I50^i_'rvan early injection test on 22 121962, raising the injection ratementally from 379 to 1414 liters

fction I[be

tprat]

proxiiater |

[ thethl

sses Iinteeqtjconateperl

71 miS. TH; must I

stimd[.injec|

as• as

hij8

'atf f dpearnues|innbej

feitaticbef

f the

voirons.

pori

timl1962|

pro|next

30to

ITS

ie fatafore

SCIENCE vo. 1*1EMBl

minute required a linear increase ofpressure of about 7 bars per 379-Iiter-per-minute increment. A linear extrap-olation of the bottom-hole pressurefrom these data gives a pressure of362 bars (19) at an injection rate ofzero.

Such a large discontinuity in injec-tion rate with pressure has a closeparallel in numerous cases where oil-bearing reservoirs have been hydraul-ically fractured to increase their per-meability. From the theory of hydraulicfracturing (20), the fluid pressure, p...required to hold fractures in the for-mation open to rapid injection is foundto be equal to the magnitude of theleast principal stress. S3. In the reservoirrock at the Denver well, p,. would beapproximately 362 bars, some 93 barsgreater than the initial fluid pressurein the reservoir.

If the greatest and least principal| stresses are assumed to be horizontal., the intermediate stress, 52, is verticali and equal to the lithostatic pressure.

The conventional pressure gradient forsaturated sedimentary rocks is 0.226bar per meter; on this basis. 5., at3671 meters below the surface is 830bars. The maximum principal stress.$1, must therefore be at least 830 bars.

Estimates of the pore pressure beforethe injection of fluid (/?„) vary widely,from as much as 328 bars (21) to aslittle as 269 bars (22). The first figureis too high; the fluid level in the wellas of 8 April 1968 corresponds to 311bars at depth of 3671 meters, and therate of falloff is still 0.027 bar per day.'t appears that, if the rate of falloffcontinues to vary with the depth offluid in the well, as it did from 23September 1966 through 8 April 1968."•* static head may be about 900"neters below the well mouth, in whichcase the original reservoir pressure was269 bars. This value is taken as thereservoir pressure in the following cal-culations.

The pore pressure at the time of thefirst recorded earthquakes can be esti-'"ated from the pumping records of thearsenal (23). The volume injected daily*' the time of the events recorded in^Pril 1962 was about I.I million liters,°<tt no pressure records were kept. In

next month excess pressure of30 bars at the wellhead was

to achieve the same volume,^h* pore pressure in the reservoir at'•he time faulting was initiated (p,) isffcrefore taken to be 389 bars.

Given St = 830 bars, 5, = 362 bars,27 SEPTEMBER 196S

and p,, = 269 bars, the effective shearand normal stresses on a potential faultplane, from Eqs. 3 and 4. are - — 203bars, cr» = 210 bars. The angle a istaken to be 60 degrees, the value foundtypically for experimental shear frac-tures. In the Mohr-Coulomb criterionfor failure (Eq. 1) the resistance tofracture is provided by the frictionalterm a,, tan <j> (<(, is about 30 degreesfor many rocks) plus the cohesivestrength, -„. In the reservoir rocks priorto injection, ra would have to be at leastas great as 82 bars to prevent fracture.

During fluid injection, when the porepressure was raised to 389 bars, rwould be 203 bars and <rn would be 90bars. The occurrence of faulting uponreduction of the frictional term a, tantt> by 69 bars indicates a value for -„ of151 bars or less.

The cohesive strength for sound,crystalline rocks deformed in the lab-oratory is about 500 bars. For thefractured rocks present in the reservoir,a cohesive strength of 150 bars seemsa reasonable assumption. Additionalfracturing during the earthquake se-

__ 2290--<C~ 1990:

.= 1690-

£ "D-i ic-

•I 25-

f 20-

~ 15-

" 10-ow 5-

-I o-

800 1000

Number of days1600

1968

Fig. 7. (A) Daily pressure plotted relative to number of earthquakes. The dashedportions in the pressure curve are interpolated where only data on the volume offluid injected were available. (B) Comparison of number of earthquakes and pressure,on a monthly basis.

1307

2.0

I.S

•f;l[ l.oE

5•Zi

1 "

0

2'

: 1 1 1

1

'. ' • '-'/^/ • " '

, J

/

1'

^

rIA/V

,/, , ;K> 150 IOC 50

» '. . l .1 ' '' Hj l V- ..'l

.•• ' L . i

, ,0 5C

Lag (in days)

Fig. 8. (A) Cross correlation of earthquakespeak at about -f 10 days indicates a lag infollowing fluid injection. Data gaps were hamand Cox ( 1 2 ) . (B) Rank difference correlatiquakes in 10-day intervals and

— i 1 :A

.

J

l ' ! ' \ 1/1\ ':• n i!< '

t i'l ' 1 ^ f fl

' " ' , , .

0.5

_

„.!.<!t> °c C

v o|| k . - /'i _ -~ °

k ^ j

v,o

11' • ^l " ~ O >lJ

i ' L \ i °^ 'u i

1 1 •100 ISO 20O 250

and fluid pressure. The -2.earthquake occurrences

led as described by Leeon (13) between earth-

,'

Q,'*

"" -o--"""f'

,'

'i

,'

i i0 20 10

EcrThquoke leod( i n doys)

' ,'"X ' ' B/'

\ \

V

-

S131fWPOIlocf tic

i'mo

§aesW^ *

I ]/&cenIS; i

H?Ia

Igtoti

t 1 (D 10 20 30

E o r t h q u o k e log,' in doys i

HVald

H| 2

pressure averaged over 10-day intervals. H^ H

quence would be expected to furtherdecrease the cohesive strength and theresistance of the reservoir rock to frac-ture and frictional sliding. Upon lower-ing of the pore pressure, the earth-quakes would be expected to continueuntil the frictional resistance plus thereduced cohesive-strength term exceed-ed the shear stress. A reduction of thepore pressure to less than the init ialpressure p,, might be required beforeseismic faulting was finally inhibited.If TO = 50 bars, p0 would have to beless than 214 bars.

We have assumed in the foregoingdiscussion, which is consistent withHubbert and Willis's theory and experi-ments, that the hydraulic fractures inthe well open normal to S3. The alterna-tive—that the fault planes alreadypresent in the basement become opento rapid flow at pressure of 362 bars—is worth considering. Migration of theepicenters along the west-northwest-trending zone parallel to the instrumen-tally determined fault planes suggestssuch a possibility. In this case, the totalnormal stress, 5n, across the fault sur-face must be exceeded at 362 bars. Thehorizontal least principal stress, 53, isthen given by

5,=2S* - 5, (cos 2a+ 1)

1 — cos 2a(5)

Taking 5, = 830 bars and a = 60degrees, as before, we have Sx = 206bars. In this case the effective stressdifference, CT, — o3, would be 624 bars,rather than 468 bars where S3 wastaken to be equal to the critical pres-sure pr. At the pressure (pr — 362 bars)for hydraulic fracturing, the least prin-cipal effective stress, aa, would betensile and equal to — 156 bars. Rocks

1308

generally have a tensile strength ofabout 50 bars and therefore could notsupport such a large tensile stress. Wefavor, therefore, the model discussedabove, in which hydraulic fractures areopened perpendicular to 5S at potepressure of 362 bars. The pore fluidthen permeates away from these frac-tures along preexisting faults parallel tothe trend of the epicentral zone.

This analysis shows that, given rea-sonable values for the frictional andcohesive strength of fractured rock,the basement rock at the Denver wellwas stressed to near its breakingstrength before injection of fluid wasinitiated. This conclusion is in accordwith the seismological evidence of con-sistency of trend and sense of slip onthe fault planes, features indicating aresponse to stresses of tectonic origin.If the unknown principal stress, 5,. isgreater than 52. rather than equal toSo as assumed, this conclusion isstrengthened. Increasing the pore pres-sure in the reservoir rocks to 389 barsby means of fluid injection reducedthe frictional resistance to fracture byabout 69 bars and initiated the earth-quake sequence.

The distribution of the Denver earth-quakes in space and time has two fea-tures which at first glance seem anoma-lous but in fact are consistent with theHubbert-Rubey theory.

1) Although there is a net migrationof epicenters away from the well con-sistent with the advance of a pressurefront during the period of fluid injec-tion (24), earthquakes appear to havebeen occurring over the whole of theexisting epicentral zone at the end ofthe pumping period (Fig. 2, top).

2) Since the cessation of fluid injec-

tion in February 1966. earthquake ac-t iv i t y has continued despite reduction ofpore pressure in the reservoir at thewell to 311 bars in April 1968. Seismicactivity near the well has ceased, butearthquakes of larger magnitude thanthose that occurred during the in jec t ionperiod have occurred near the north-westerly terminus of the epicentral zone(Fig. 2, bottom).

We assume that the rocks in the f a u l tzone contain a large number of crack*that vary in length and have a normaldistribution of orientation about thetrend of the fault zone. Furthermore.the cracks parallel to the fault zone areassumed to have orientations such t h a tT/CT,, has a maximum value — that "*•assumed to have the most favorableorientation for propagation. In lhe

cracks where r/a, exceeds the coeffi-cient of sliding friction, irregularitiesalong the fracture surface and 'hf

crack terminations may be consideredlock points restricting movement '"effect the strength of these lock pom"constitutes the cohesive-strength term.r0, in the Mohr-Coulomb criterion forfailure (Eq. 1) . At the t ime of »"earthquake, when one of thc<.c locpoints fails, the break may pror-'F-'''until it is inhibited by a stronger lo*point along the fracture surface. ^crack would also tend to arrest a* ^passed from a region of higher mioregion of lower pore pressure.

This conceptual model of the *ture process is illustrated inFractures already present in the ^

thament which have orientationsfavorable for propagation icrac

)fl,and B) and orientations that ar ̂favorable (crack C) are sno*"lengths shown correspond to

avol

vaditarjc h iyori

e

SCIENCE, vou :

" tance along each crack between lockpoints. The greater the distance betweenJock points the greater the concentra-tion of stress at these points and themore likely a given fracture is to prop-agate.

In the sequence shown (Fig. 9),certain features may be noted.

1) Propagation of crack A takesplace with seismic velocity upon in-jection of fluid at the well and continuesuntil the mean pore pressure over theincreased length of fault is below thevalue required for continued propaga-tion, or until the crack encounters astronger lock point.

2) Crack B is shorter and, thoughnearer to the well, requires an addition-al increase in fluid pressure to propa-gate.

3) With continued injection of fluid,cracks A and B propagate to greaterlengths, and faulting also takes placeon crack C, which is shorter and lessfavorably oriented with respect to <TJ.

4) As shown in the idealized profilesof reservoir pressure, cessation of fluidinjection results in a rapid reduction ofpressure near the well but in a continuedadvance of the pressure front at greaterdistance from the well. The cracks,such as A. which have propagated outbeyond the pressure front during fluidinjection become reactivated as thefront advances, even though the injec-tion has ceased. The shorter and unfa-•Vorably oriented cracks within the zoneBear the well, in which fluid pressure isdecreasing, become inactive.-;, A relation between the length of a',»alt rupture and the magnitude of the^associated earthquake has been estab-lished from data on faults which break0* surface (25). If this relation holdsfor the Denver earthquakes, our con-(*ptual model, as outlined above, would}B*dict a predominance of earthquakesj*? greater magnitude following cessa-_^on of fluid injection, when only thelinger cracks are capable of propa-l«ion. Moreover, faulting near the^cD would diminish, in comparison

that during the injection period,seismic activity would be moredistributed over the entire epi-zone and greater numbers of

'fwhquakes of smaller magnitude would'

•JJost investigators working on theearthquakes have concludedinjection of fluid triggered the

SEPTEMBER i96«

seismic activity. However, some investi-gators believe that the earthquakesmight have occurred even if the wellhad not been drilled, and there is awide range of opinion on the advisa-bility of attempting to terminate theearthquake sequence by removing fluidfrom the well ( 2 6 ) .

We consider the possibility of acoincidental occurrence of earthquakeswith the onset of fluid injection atDenver to be remote. Furthermore,there is now evidence that fluid injec-tion and swarms of earthquakes occurtogether at other locations (27). Themechanism by which fluid injectiontriggered the earthquakes is the reduc-tion of frictional resistance to faulting,a reduction which occurs with increasein pore pressure. Other mechanisms,including reduction in the strength ofthe rocks, due to chemical alteration,have been investigated, but so far noneof these appear to explain the seismicityadequately. The implication of thepore-pressure mechanism—that the

rocks were stressed to near their break-ing strength before the injection offluid—is in accord with the availabledata. Faulting is predominantly right-lateral strike-slip on faults which areparallel to the epicentral zone, an ob-servation which also suggests release oftectonic stress rather than disturbancesproduced by fluid injection alone.

Prior to 1967. the frequency of oc-currence of earthquakes of differentmagnitudes in the Denver area wassuch that the likelihood of a reallydestructive earthquake could reasonablybe considered remote. In view of the1967 earthquakes, however, there isno longer any assurance that a de-structive earthquake will not occur.Hence, the question of what might bedone to lessen the earthquake hazardnow confronts those earth scientistswho have been studying the problem.

In our view, it might be possible toreduce the size and number of the earth-quakes by removing substantial quanti-ties of fluid from the reservoir. If the

N

WELL

PC

t

Fig. 9. (Right) Idealized profile of pore pressure in the reservoir during successivestages (a-d) of fluid injection. (X) Distance from well; (p.) initial reservoir pressure;(PC) hydraulic fracturing pressure. Between stages c and d, injection ceases. (Left)Idealized view of fault zone during successive stages (a-d) of fluid injection. Arrowson fractures present before injection indicate propagation and release of seismic energyduring advance of the pressure front. The assumed orientations of horizontal principalstresses. 5,, S». are shown.

1309

Hubbert-Rubey pore-pressure effect isthe cause of the earthquakes (and, toour knowledge, it is the only mechanismyet demonstrated to explain them ade-quately), reducing the pore pressureshould effectively strengthen the reser-voir rocks and thus reduce the sizeand number of the earthquakes.

Despite the theoretical attractivenessof this approach, the engineering diffi-culties and costs involved in removingsufficient quantities of fluid from thereservoir to reduce significantly the oc-currence of earthquakes may proveprohibitively expensive. Early with-drawal tests conducted before fluid In-jection was begun indicated a reservoirwith very low transmissibility. If thislow transmissibility is characteristic ofthe reservoir today, the removal offluid would probably be impractical.One technique used successfully to in-crease the production of oil from reser-voirs having low transmissibility is thepropping open of fractures with coarsesand. Steps are now being taken toevaluate the properties of the reservoirand to design preliminary tests to deter-mine the feasibility of removing thefluids (28).

Given sufficient time, the pore pres-sure in the focal zone will dissipatenaturally, and it is hoped that, as thezone of high pressure spreads outwardfrom the well, the maximum pressuresin the reservoir will fall below the levelrequired to trigger seismic activity. Un-fortunately we do not have precisemeasurements of the original staticpressure in the reservoir and, therefore,we cannot tell how much excess pres-sure remains in the reservoir rock.Because the level at which the fluidstands in the disposal well is still fall-ing, we know that a substantial over-pressure must still exist and that there-

fore the zone of high pressure is stillmoving out from the well. There is noreason to believe that the excess pres-sure will dissipate naturally before theepicentral zone is extended by furtherfaulting and attendant large earth-quakes. It appears that, unless the re-moval of fluid can be shown, eithertheoretically or by pumping tests, toconstitute a real hazard, every' reason-able effort should be made to removefluids from the reservoir in order toreduce the reservoir pressures andminimize the earthquake hazards. Tobest achieve this goal and to test thehypothesis presented here, we favorremoval of fluid by means of a secondwell drilled into the currently activefocal zone.

References and Notes

1. Y. Wang, thesis. Colorado School of Mines(19651.

2. D. Evans, Mountain Geologist 3, No. 1, 23(1966).

3. H. L. Krivoy and M. P. Lane, in "Geo-physical and Geological Investigations Re.lating to Earthquakes in the Denver Area,Colorado." U.S. Geol. Sun'. Open File Rep.(1966), pt. 2.

4. F. Hadsell, "History of Earthquake Activityin Colorado," Colorado School of MintsRep. (1%7).

5. J. H. Healy, W. H. Jackson, J. R. VanSchaack. in "Geophysical and GeologicalInvestigations Relating to Earthquakes in theDenver Area, Colorado," U.S. Geol. Sun.Open File Rep. (1966), pt. 5.

6. M. Major and R. Simon. Quart. ColoradoSchool of Mines 63, 9 (1968).

7. The location of the earthquake of 10 Apriland of its aftershocks was determined byD. B. Hoover of the U.S. Geological Survey.

8. The nature of the magnitude-frequency plotsfor southern California, together with a gooddiscussion of the pros and cons of usingthese curves for predicting large-magnitudeearthquakes, is given by C. Allen, P. St.Amand, C. Richter, and J. Nordquist [Bull.Seismol. Sac. Amer. 55, 753 (1965)].

9. M. Major and C. Wideman. "Seismic Studyof Derby Earthquakes," Colorado School ofMinn Rep. (1967).

10. C. Richter, Elementary Seismology (Freeman,New York, 1958).

11. D. B. Hoover, unpublished manuscript.12. W. H. K. Lee and C. S. Cox, J. Geophys.

Res. 71. 2101 (1966).13. Handbook of Chemistry and Physics (Chem-

ical Rubber Company, Cleveland, Ohio. td.47, 1967), p. A-245.

14. D. M. Sheridan. C. T. Wruckc. R. E. Wilcox.in "Geophysical and Geological Invesiica-tions Relating to Earthquakes in the DenverArea. Colorado." U.S. Geol. Sury. Open FileRep. (1966). pt. 4.

15. M. K. Hubbert and W. W. Rubey. Bui:.Geol. Sac. Amer. 78. 115 (1959).

16. K. Terzaghi. Eng. Xc*'s-Record 116. 501(1936).

17. E. M. Anderson. The Dynamics of Faulnne(Oliver and Bovd, Edinburgh, ed. 2. 19M)p. 15.

18. The pressure gradient in the injected fluidis taken to be 0.0988 bar per meter (DB. Hoover, personal communication. 196«i.

19. The pressure gradient in the fresh wa te rused in the initial tests is 0.0979 bar permeter.

20. M. K. Hubbert and D. G. Willis. Tram.AIME 210, 153 (1957).

21. F. C. Ball and G. R. Downs, in "InkaionWell-Earthquake Relationship, Rocky Moun-tain Arsenal. Denver. Colorado." U.S. Arm^Corps of Engineers, Omaha District, Her-(1966).

22. H. K. Van PooUen. report prepared for theU.S. Army Corps of Engineers (January 1968).

23. F. C. Ball, H. H. Sells, G. R. Downs, reportprepared for the U.S. Army Corps of En-gineers (May 1966).

24. M. Major, personal communication. 196".25. D. Tocher, Bull. Seismol. Soc. Amer. 4*.

147 (1958).26. It would be extremely difficult to summ.ime

all of the opinions that have been expressedregarding the Denver earthquakes. Mam irt-vestigators who have worked on the proMemhave not published their contributions, amimost of the people working on it have mod-ified their views as additional evidence hasbecome available. The current views of work-ers at the Colorado School of Mines are sum-marized by J. C. Hollister and R. J. Weirnerin Quart. Colorado School of Mines 63. Nc1

1, 2 (1968).27. One example that has been studied is a-

Rangely oil field in western Colorado: see JH. Healy, C. B. Raleigh. J. M. CoaUrypaper presented before the 64th annual m-'C'-ing of the Seismological Society of America.1968.

28. Pumping tests were initiated at the arsera.well on 1 September 1968.

29. Publication of this article (publication V1

679 of the Institute of Geophysics and Plane-tary Physics) has been approved by the *•rector of the U.S. Geological Survey. Tt*work described has been supported in P*11

by NSF grant G_A. 277. We thank M. MaV^and R, Simon of the Colorado School o-Mines for providing us with a lisi of lt*times of occurrence and the magnitudes <\Denver earthquakes recorded at the Cc^H. Green Observatory i Bergen Park. Co**rado; W. H. K. Lee for making his spe^»'analysis program available to us; and M ^Hubbert for discussions on hydraulic frj-.ta''ing.

|micrifdeatl

dencJ[been Ifkbsoij

dudeUuced

nicletwcj

fphotc

hwjJerWes: IOgletl

1310 SCIENCE, VOL. 1(1