You are on page 1of 12

Bulletin of the Seismological Society of America

Vol. 80 June 1990 No. 3

SOURCE INVERSION OF T H E 1988 UPLAND, CALIFORNIA, EARTHQUAKE: D E T E R M I N A T I O N OF A FA U L T PL A N E FOR A SMALL E V E N T BY JIM MORI AND STEPHEN HARTZELL ABSTRACT
We examined short-period P waves to investigate if waveform data could be used to determine which of two nodal planes was the actual fault plane for a small (ML 4.6) earthquake near Upland, California. We removed path and site complications by choosing a small aftershock (ML 2.7) as an empirical Green function. The main shock P waves were deconvolved by using the empirical Green function to produce simple far-field displacement pulses. We used a leastsquares method to invert these pulses for the slip distribution on a finite fault. Both nodal planes (strike 125 , dip 85 and strike 221 , dip 40 ) of the firstmotion focal mechanism were tested at various rupture velocities. The southwest trending fault plane consistently gave better fitting solutions than the southeasttrending plane, We determined a moment of 4.2 x 1022 dyne-cm. The rupture velocity, and thus the source area could not be well resolved, but if we assume a reasonable rupture velocity of 0.87 times the shear wave velocity, we obtain a source area of 0.97 km 2 and a stress drop of 38 bars. Choice of a southwesttrending fault plane is consistent with the trend of the nearby portion of the Transverse Ranges frontal fault zone and indicates left-lateral motion. This method provides a way to determine the fault plane for small earthquakes that have no surface rupture and no obvious trend in aftershock locations.

INTRODUCTION A small earthquake (ML 4.6) occurred near the town of Upland, California, on 26 June 1988. It was located (34 8.09' 117 42.58') close to the frontal fault zone of the Transverse Ranges at a depth of 7.9 km. This is an area of relatively low seismicity, and little is known about previous historical fault movements (Cramer and Harrington, 1987; Morton and Matti, 1987; Pechmann, 1987). The location and well-constrained focal mechanism, obtained from 124 network first motions (L. Jones, personal comm.), are shown in Figure 1. The routine Southern California Network locations of about 50 aftershocks cluster in a volume with dimensions of a few kilometers and, as is typical for events of this size, do not show an obvious trend that can be identified as the fault plane. We have used waveform data from the Southern California network to try to resolve which of the two nodal planes was the fault plane for the earthquake. The short-period (1 to 8 hz) seismograms for the distances used in this study (42 to 85 km) are quite complicated; consequently, it is difficult to extract information about the source. One solution to this problem is to use a smaller earthquake from the same location as an empirical Green function. Deconvolution of the seismograms of the main event, using the empirical Green functions, should eliminate all of the path, site, and instrument effects (Frankel et al., 1986; Li and Thurber, 1988; Mori and Frankel, 1990). We have used this method to obtain 5o7

508

J. MORI AND S. HARTZELL

54 -

lib

117

FIG. 1. Focal mechanism (lower hemisphere) and location of the 1988 Upland earthquake. Triangles represent stations which recorded data used in this study.

relatively simple far-field P-wave displacement pulses. We then inverted this data for slip on a fault plane to test which of the two nodal planes better fits the data.
DATA AND EMPIRICAL G R E E N FUNCTION DECONVOLUTION

Data used in this study are from three stations (LJB, RAY, SIL) of the Southern California Network and from the Pasadena (PAS) Very Broad Band instrument (Fig. 1). The three short-period network stations, which have high-and low-gain channels with a combined dynamic range of about 72 db, and the large dynamic range (140 db) Pasadena instrument, allowed on-scale recording of both the ML 4.6 main shock and the ML 2.7 aftershock that was used as the empirical Green function. The Nyquist frequency for the PAS instrument is 8 hz; consequently, all of the data used in this study were low-pass filtered, using a third-order Butterworth filter with a corner at 8 hz. In choosing an event to use as an empirical Green function, we looked for aftershocks that had waveforms that were most similar to the main shock, and so would indicate similar focal mechanisms and locations. The particular aftershock that was used as the empirical Green function was chosen because it was the smallest well-recorded aftershock that had waveforms similar to the main shock. The smallest aftershock was chosen in order to minimize the source contribution in the waveform. We found that the epicenters of the main shock and aftershock were within 250 m of each other when they were relocated using the same set of stations. This is less than half a wavelength of the highest frequency (8 hz) used in the data, so that, initially, all of the propagation effects should be the same for the two earthquakes. However, the rupture extent for the main shock was estimated to be 1 to 2 kin, so that this assumption may not be as valid for later parts of the waveform. The deconvolution of the main shock, using the empirical Green function, was

SOURCE INVERSION

OF T H E

1988

UPLAND EARTHQUAKE

509

done by dividing the Fourier Transform of the main shock by the transform of the aftershock (Mori and Frankel, 1990). A water-level criterion (set at 0.1 per cent of the maximum amplitude) was used to fill up the "holes" in the spectrum of the aftershock, in order to stabilize the deconvolution against division by small values in the transform of the empirical Green function. The main shock and aftershock waveforms and the resultant deconvolutions are shown in Figure 2. All four of the stations show relatively simple deconvolved pulse shapes, interpreted as being the far-field P waves with all of the effects of the path, site, and instrument re~.~oved. In the inversion, we used only the pulse shape data, so the areas under the displacement pulses were normalized to give the same moment at each station. Before normalizing the waveforms, we calculated the moment for the main shock using the following expression from Boatwright (1980): D Mo = 4~[p(Xo)p(x)c(x)]l/2C(Xo?/2 -F 5~ where (Xo) = 2.80 and (x) = 2.28 g/cm 3 are the densities at the source and receiver, respectively; c ( x o ) = 6.56 and c ( x ) = 3.40 km/sec are the velocities at the source and receiver, respectively; D is the epicentral distance; F is the radiation pattern which also includes a free surface correction at the receiver; and ~ is the area under the
PAS

LJB
148691 758

36808 381

SIL 95365 220

0.00

1'.00

2'.00

3'.00

FIG. 2. On t h e left are pairs of waveform data showing t h e m a i n shock (top) a n d t h e aftershock used as t h e empirical G r e e n f u n c t i o n (bottom). T h e n u m b e r s on t h e left are the m a x i m u m amplitudes in digital c o u n t s of each trace. T h e waveforms on t h e right are t h e far-field d i s p l a c e m e n t pulses resulting from deconvolution of t h e m a i n shock u s i n g t h e aftershock. T h e time scale is in seconds.

510

J. MORI AND S. HARTZELL

TABLE 1
M O M E N T ESTIMATE Station Distance (km) Azimuth F U (cm sec) Moment (dyne-cm)

PAS LJB RAY SIL Average

42.5 52.1 83.5 84.7

272 346 97 73

0.88 -1.04 0.68 0.65

6.23 x 10-4
7.58 x 10 .4

8.57 X 10-4 6.69 x 10-4

1.94 x 1022 2.46 x 1022 6.75 x 1022 5.59 x 1022 4.19 x 1022

displacement pulse. T h e results, shown in Table 1, give an average value of 4.2 x 1022 dyne-cm for the four stations. This value for the total moment, together with an assumed value of 4.0 x 1011 d y n e / c m for the rigidity, was used to scale the results of the inversion for slip distribution. INVERSION METHOD We followed Hartzell and H e a t o n (1986) in calculating the inversion of the waveforms for the distributed slip on the fault plane. In this method, the fault plane is divided into discrete subfaults, and synthetic seismograms are calculated at each station for a unit a m o u n t of slip on each subfault. A synthetic waveform at each station can t h e n be formed from a linear combination of the subfault seismograms. A linear least-squares inversion, which minimizes the error between the synthetics and the data, is used to solve for the amounts of slip on each of the subfaults. In this study, the program QUAKE7 by P. Spudich was used to calculate the synthetic seismograms for each subfault (Spudich and Frazer, 1984). T h e fault plane was divided into triangular shaped subfaults, which enabled us to make a simple and fast calculation for the source-time function as the rupture sweeps across a subfault area. For all of the fault models, a constant rupture velocity and a constant rise time of 0.06 sec were assumed. Figure 3 shows the 11 by 11 grid of triangles, 0.36 by 0.60 km in size, t h a t we used for all of the inversions. T h e grid is centered around the network location and oriented to match one of the nodal planes of the first-motion mechanism. T h e plane striking southeast at 125 and dipping 85 to the southwest was tried first (plane 1), t h e n the plane striking southwest at 221 and dipping 40 to the northwest (plane 2) was used. Because the data we are modeling are assumed to be the far-field displacement pulse shapes, all t h a t is needed for each subfault synthetic is the source-time function generated by t h a t subfault. All of the synthetic source-time functions were simple unipolar triangles, not complicated waveforms, so this inversion was quite stable. A positivity constraint (Lawson and Hanson, 1974) was placed on the solution, but no damping or minimization constraints were necessary. T h e inversion was designed to resolve the shape of the azimuthally dependent waveforms, consequently all of the displacements pulses were normalized to the average m o m e n t of 4.2 x 1022 dyne-cm, and the focal mechanism was kept fixed. T h e inversion procedure solved the following equation for the vector x t h a t contains the unknown values of slip for each of the 121 subfaults:
Ax=B

T h e A matrix (300 by 121 pts.) contains the synthetic seismograms for the 121 subfaults at each of the four stations, sampled at 50 samples/sec for 1.5 sec.

SOURCE INVERSION OF THE 1988 UPLAND EARTHQUAKE

511

STRIKE

i-"1

0'.00

1'.00
KM

2'.00

3~.00

FIG. 3. Subfault grid used for the inversion on both fault planes. The filled circle is the hypoeenter.

The B vector contains 1.5 sec of data sampled at 50 samples/sec for the four stations (300 pts). The inversion was carried out assuming various rupture velocities from 2.6 to 4.6 km/sec (corresponding to 0.70 to 1.2 times the S-wave velocity, assuming a P- to S-wave velocity ratio of 1.7) on the southeast-striking nodal plane, plane 1, and then on the southwest-striking nodal plane, plane 2. The solution for each of the cases was evaluated by using the variance (2), a2 = ((A x) N is the number of degrees of freedom, N = ndata - nsol - 1, where ndata is the fixed number of data points in B and nsol is the number of subfaults that had nonzero slips. The value of nsol varies because of the positivity constraint, which solves only for those subfaults which it determines has nonnegative slip. Comparisons of the variances show which models produce the synthetics that best match the data.
D E T E R M I N A T I O N OF F A U L T P L A N E

B)2/N

The variances for a range of rupture velocities on the two nodal planes are shown in Figure 4. In all cases, except for the slowest rupture velocity of 2.6 km/sec, plane 2 gives a better fit to the data. The variance reduction using plane 2 compared to plane 1, is 2 to 12 per cent for the slower rupture velocities and 20 to 30 per cent for the higher rupture velocities, indicating a significant improvement of the fit to the data for plane 2.

512
0.020

J. MORI

AND

S. H A R T Z E L L

0.015

tJ <9 Z

<
< 2>

0.010

0.005

Plane 2

0.000 2.5

. . . . . . . . .

' 3.5

. . . . . . . . .

' 4.5

. . . .

RUPTURE VELOCITY (KM/SEC)


FIG, 4. V a r i a n c e s of t h e s o l u t i o n s for a r a n g e of r u p t u r e v e l o c i t i e s o n t h e two n o d a l p l a n e s . T h e a s s u m e d S - w a v e v e l o c i t y i n t h e s o u r c e r e g i o n is 3.8 k m / s e c .

Figure 5 shows the slip solutions obtained for several rupture velocities on planes 1 and 2. When plane 1 is used the rupture spreads out rather symmetrically from the hypocenter. For the preferred plane 2, the rupture propagates more updip and toward the southwest. The updip rupture propagation obtained for plane 2 is consistent with the observation that many earthquakes initiate near the bottom of the source area and rupture upward toward the surface (Sibson, 1982). The slip distribution on the preferred plane 2 is what one might qualitatively expect from looking at the waveforms in Figure 2. The main feature to note is that PAS and LJB look similar in shape and RAY and SIL look similar, with the PAS/ LJB pair being slightly broader than the RAY/SIL pair. This suggests some directivity toward the east, with the rupture moving either southeast on the southeast-trending plane (plane 1) or updip on the southwest-trending plane (plane 2). It appears that the reason the inversion results favor one fault plane over the other is also due to the similarity of the P A S / L J B waveforms and the SIL/RAY waveforms. For the favored fault plane 2, the take-off angles from the fault normal are relatively similar for PAS (133 ) and LJB (116), so it is not difficult for the inversion to find solutions which produce waveforms that are similar at PAS and LJB. For fault plane 1, the take-off angles are quite different to PAS (123 ) and LJB (50), consequently the inversion does not do as well in making similar synthetic waveforms at PAS and LJB. The same reasoning applies for the similar waveforms at SIL and RAY, where the take-off angles are similar for plane 2 (61 , 53 ) but slightly greater for plane 1 (39 , 62). Thus, the fault-plane determination depends on the difference between PAS/ LJB waveforms and the RAY/SIL waveforms. Even taking into account the noise level seen in the waveforms, there appears to be a distinct difference in shape

SOURCE INVERSION OF THE 1988 UPLAND EARTHQUAKE


NW

513

SE

NW

qqF

NW

50
2O 10 0

2.8 KM/SEC

3.4 KM/SEC

4.0 KM/SEC

PLANE

NE

SW

h,~

~;w

NF

30 (

2.8 KM/SEC

3.4 KM/SEC

4.0 KM/SEC

PLANE 2

0.0

1.0 KM

2.0

3.0

FIG. 5. Solutions of slip distribution for three rupture velocities on each of the two nodal planes. The vertical pin in the middle of the grid marks the hypocenter.

between the P A S / L J B waveforms and the RAY/SIL waveforms, that can be used for the fault plane determination.
R U P T U R E VELOCITY

For both planes, the variances become smaller with higher rupture velocities. This result indicates that the best-fitting solution is for the highest rupture velocity tested, 4.6 km/sec, implying a transonic rupture velocity of 1.2 times the shearwave velocity. However, we believe this result is just a consequence of the poor resolution we have on the rupture velocity. Because all the stations are far from the source, it is difficult to measure the directivity effects needed to obtain a rupture velocity. There is a direct relation between the rupture velocity and the lateral dimensions of the source. Assuming higher rupture velocities, the source dimension increases, given the fixed pulse duration observed in the data. When the source dimension increases, more subfaults (variables) contribute to the solution, consequently the solution can be more complicated and can better fit the data. Thus, the reduction of the variance for higher rupture velocities is believed to be simply a result of solving the problem with more variables. The number of nonzero values in the immediate region of the hypocenter increased from 10 for 2.6 km/sec to 22 for 4.6 km/sec. The increasing complexity of the solution with increasing rupture velocity, can be seen in the solutions on plane 1 of Figure 5. We also tried to resolve the rupture velocity using the Akaike information criterion, AIC (Akaike, 1974), which is a statistical test used to compare models with different numbers of parameters. However, the results were similar to the variance and did not show any preferential rupture velocity. Both statistical tests

514

J. MORI AND S. HARTZELL

did show a very weak local minimum at a rupture velocity of 3.3 km/sec. For lack of any other evidence, we have simply assumed this reasonable value (which corresponds to 0.87 times the shear-wave velocity) for our preferred rupture velocity.
RESULTS

The solution of slip distribution for the rupture velocity of 3.3 km/sec on plane 2 is shown in Figure 6. An estimate of the uncertainty in the slip solution is given by the diagonal elements of the model covariance matrix multiplied by some estimate of the error in the data. We used the standard deviation in the absolute moment determination from the four stations (Table 1) as an estimate of the data error. The resulting uncertainties in the slip of each subfault are shown on Figure 6. The main area of slip near the hypocenter is significantly larger than the error estimates, indicating a reliable part of the solution, whereas the smaller amounts of slip on the downdip edge of the grid are within the uncertainty estimates and not considered a resolvable part of the slip solution. Assuming that the earthquake ruptured the area of the nine subfaults near the epicenter, the source area is 0.97 km 2. The static stress drop is given by Kanamori and Anderson (1975)
Aa = C~(D/L)

where C is a geometrical factor near 1.0, tL is the rigidity, D is the average slip, and L is a fault dimension. Using C -- 1.0, ~ = 4.0 x 1011 dyne/cm 2, an average value of slip for the nine subfaults, (D = 9.1 cm) and a fault dimension equal to the square root of the area of the nine subfaults (L = 0.99 km), gives a stress drop of 38 bars. Using the solutions for the other rupture velocities gave stress drops of 11 to 37 bars, with the slower rupture velocities having generally smaller rupture areas and thus higher stress drops.
NE SW / NE SW

..50 20 10 0

5.5 KM/SEC

UNCERTAINTY

6.o

1'.o
KM

i.o

i.o

FIG. 6. (Left) Preferred solution for the slip distribution using a rupture velocity of 3,2 km/sec on plane 2. (Right) Uncertainties in the solution estimated from the model covariance matrix and the standard deviation of the absolute m o m e n t estimate.

SOURCE INVERSION OF THE

1988 UPLAND EARTHQUAKE

515

A,
I I

I L,

i RAY

().0

(~.5

1'.0 SEC

1'.5

i.O

FIG. 7. Modelsynthetics(dottedlines)for a rupture velocityof S.2 km/secon plane 2 and deconvolved displacementwaveforms(solidlines) used in the inversion. In Figure 7, the synthetic seismograms for the preferred solution are shown, together with the far-field displacement data that were used in the inversion. The synthetic seismograms from Figure 7 were also convolved with the empirical Green functions and are compared with the original low-pass filtered data in Figure 8. In both figures, the synthetics match the data reasonably well for all the stations, even including the high-frequency features. DISCUSSION The method described above provides a way of determining fault planes for small events that have neither surface rupture nor aftershock locations which clearly define a fault plane. Using a small event as an empirical Green function has the advantage of easily correcting the waveforms for the complicated path and site effects. Extremely detailed knowledge of the velocity structure would have been necessary to construct synthetic Green functions to model an earthquake of this size in the 1 to 8 hz frequency band. Further, the simple displacement pulses which resulted from the deconvolution are easy to invert for fault slip. The inversion is much more stable than inverting complicated seismograms which include all the path and site effects. With good azimuthal station coverage, fault-plane determination should be possible for most strike-slip events with source dimensions of a few kilometers, because the directivity effects due to the finiteness of the source would be significantly different for the two nodal planes. Dip-slip events may be more difficult, because the azimuthal dependence of the directivity effects would be much harder to see for two nodal planes that have a similar strike. Dip-slip events would probably require

516 2800/',,
. . . . . . . . . . . . . I',."",/
6049 ',j'

J. MORI AND S. HARTZELL PAS


. _

RAY

48955,,,,, , . . . . . . . . . . .
5680

o
, ~,,,',t . . . .
ir i, i,

,
',4"~,

8 728 5 ........... 148691

-.

,;,,

,,

/ ;'~\~ ," ,, .._

LJB 16og8o - . . . ...... . 95563

SIL

,~" ,'~ f, ', ~ . , I \;1~ , - , ; , ,. i, ,

" ~Jl'\'i_l~

0.00

1.00

.00 SEC

.00

4.00

FIG. 8. Model synthetics from Figure 8 convo]vedwith the empirical Green functions (dotted lines) and the original low-pass filtered data (solid lines).

30 '118 O'/~,._

45'
I

30'
I

15'

L.
o

I
io

2o KM

15'

34' O'

Fro. 9. Map showing inferred left-lateral motion for the small southwest trending section connecting the Cucamonga and Sierra Madre thrust faults. Fault locations and displacements are taken from Cramer and Harrington (1987).

one or more stations close to the source with upward take-off angles, in order to resolve the effects caused by the difference in the dip angles of the nodal planes. The size of earthquakes t h a t can be studied in this m a n n e r is somewhat limited. The event has to be large enough so t h a t the finiteness of the source can be observed in the frequency b a n d of the data. If the source is too large, however, the deconvo-

SOURCE INVERSION OF THE

1988 UPLAND EARTHQUAKE

517

lution of the waveforms using a smaller earthquake will not work, because the small event would not be an appropriate Green function for the entire areal extent of the larger event. For short-period network data in the frequency band of 1 to 20 hz, magnitude 4 to 5 earthquakes are suitable. The inversion results for the 1988 Upland earthquake indicate that the southwesttrending plane was the fault plane. This implies that the slip was left-lateral on a southwest-trending plane dipping toward the northwest. This orientation is consistent with the trend of a nearby portion of the Transverse Ranges frontal fault zone, as it bends to the southwest connecting the Cucamonga fault with the Sierra Madre Fault. From focal mechanisms of a few small earthquakes in the region, Cramer and Harrington (1987) inferred that slip on this portion of the fault was left-lateral strike-slip, similar to the movement we described in this study (Fig. 9). Strike-slip faulting on this southwest-trending portion of the fault and the adjacent thrust faulting on the east-west-trending portions on either side is consistent with the general north-south convergence across the Transverse ranges (Morton and Yerkes, 1987).
CONCLUSIONS

By using a small aftershock as an empirical Green function, complicated path and site effects can be deconvolved from P waveforms recorded at regional distances. This process results in far-field source-time functions with a high-frequency resolution of up to 8 hz. This frequency range is higher than can be modeled using only synthetics with our present knowledge of velocity structures. The deconvolved waveforms were inverted for the slip distribution on a discretized fault plane, testing the two nodal planes of the focal mechanism at various rupture velocities. Because the data fit the southwest-trending nodal plane better than the southeast-trending nodal plane for almost all of the rupture velocities, we believe that this approach enabled us to conclude that the southwest trending plane was the fault plane for this earthquake.
ACKNOWLEDGMENTS We benefited from the helpful reviews of A. Frankel, T. Heaton, W. Ellsworth, A. Snoke, and D. Boore. L. Jones provided locations and focal mechanisms from Southern California Network. REFERENCES Akaike, H. (1974). A new look at the statistical model identification, I E E E Trans. Autom. Control AC-19,716-723. Boatwright, J. (1980). A spectral theory for circular seismic sources: simple estimates of source dimension, dynamic stress drops, and radiated energy, Bull. Seism. Soc. Am. 70, 1-28. Cramer, C. H. and J. M. Harrington (1987). Seismicity and tectonics of the Cucamonga fault and the eastern San Grabriel Mountains, San Bernardino County, In Recent reverse faulting in the Transverse Ranges, California, U.S. Geol. Surv. Profess. Paper 1339, 7-26. Frankel, A., J. Fletcher, F. Vernon, L. Haar, J. Berger, T. Hanks, and J. Brune (1986). Rupture characteristics and tomographic source imaging of ML - 3 earthquakes near Anza, Southern California, J. Geophys. Res. 91, 12633-12650. Hartzell, S. and T. Heaton (1986). Inversion of strong ground motion and teleseismic waveform data for the fault rupture history of the 1979 Imperial Valley, California, earthquake, Bull. Seism. Soc. Am. 76, 649-674. Kanamori, H. and D. L. Anderson (1975). Theoretical basis of some empirical relations in seismology, Bull. Seism. Soc. Am. 65, 1073-1095. Lawson, C. L. and R. J. Hanson (1974). Solving Least Squares Problems, Prentice-Hall, Inc., Englewood Cliffs, New Jersey, 340 pp.

518

J. MORI AND S. HARTZELL

Li, Y. and C. H. Thurber (1988). Source properties of two microearthquakes at Kilauea Volcano, Hawaii, Bull. Seism. Soc. Am. 78, 1123-1132. Mori, J. and A. Frankel (1990). Source parameters for earthquakes associated with the 1986 North Palm Springs, California, earthquake determined using empirical Green functions, Bull. Seism. Soc. Am. 80,278-295. Morton, D. M. and J. C. Matti (1987). The Cucamonga fault zone: geological setting and Quaternary history, In Recent reverse faulting in the Transverse Ranges, California, U.S. Geol. Surv. Profess. Paper 1339, 179-203. Morton, D. M. and R. F. Yerkes (1987). Introduction, In Recent reverse faulting in the Transverse Ranges, California, U.S. Geol. Surv. Profess. Paper 1339, 1-5. Pechmann, J. C. (1987). Tectonic implications of small earthquakes in the central Transverse ranges, In Recent reverse faulting in the Transverse Ranges, California, U.S. Geol Surv. Profess. Paper 1339, 179-203. Sibson, R. H. (1982). Fault zone models, heat flow and the depth distribution of earthquakes in the continental crust of the United States, Bull. Seism. Soc. Am. 72,151-163. Spudich, P. and L. N. Frazer (1984). Use of ray theory to calculate high-frequency radiation from earthquake sources having spatially variable rupture velocity and stress drop, Bull. Seism. Soc. Am. 74, 2061-2082.

U. S. GEOLOGICAL SURVEY 525 SOUTH WILSON AVENUE PASADENA, CALIFORNIA 91106 Manuscript received 12 September 1989

You might also like