Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Bulletin of the Seismological Society of America, Vol. 84, No. 3, pp.

646-659, June 1994

The Co-Seismic Slip Distribution of the Landers Earthquake


b y J. F r e y m u e l l e r , N. E. K i n g , and P. Segall

Abstract We derived a model for the co-seismic slip distribution on the faults
which ruptured during the Landers earthquake sequence of 28 June 1992. The
model is based on the inversion of surface geodetic measurements, primarily
vector displacements measured using the Global Positioning System (GPS). The
inversion procedure assumes that the slip distribution is to some extent smooth
and purely right-lateral strike slip. For a given fault geometry, a family of so-
lutions of varying smoothness can be generated.
We choose the optimal model from this family based on cross-validation,
which measures the predictive power of the data, and the trade-off of misfit and
roughness. Solutions which give roughly equal weight to misfit and smoothness
are preferred and have certain features in common: (1) there are two main patches
of slip, on the Johnson Valley fault, and on the Homestead Valley, Emerson,
and Camp Rock faults; (2) virtually all slip is in the upper 10 to 12 km; and
(3) the model reproduces the general features of the geologically measured sur-
face displacements, without prior constraints on the surface slip. In all models,
regardless of smoothing, very little slip is required on the fault that represents
the Big Bear event, and the total m o m e n t of the Landers event is 9.1019 N-m.
The nearly simultaneous rupture of multiple distinct faults suggests that much
of the crust in this region must have been close to failure prior to the earthquake.

Introduction
The Landers earthquake (Mw = 7.3) occurred at 11:58 ments, including displacements determined by Stanford
UTC on 28 June 1992, and was followed about 3 hr later and the USGS along with those of several other groups.
by the Big Bear earthquake (Mw = 6.2). The Landers We analyzed most of the available pre-earthquake and
earthquake ruptured five major faults, with a total rup- postearthquake GPS data to determine the co-seismic dis-
ture length of approximately 85 km (Fig. 1). It occurred placement field (Fig. 2). We then inverted the co-seis-
in the southern part of the Eastern California Shear Zone mic displacements and line-length changes to determine
(ECSZ), a broad zone of active shear that extends from the distribution of slip on the faults that ruptured during
the San Andreas fault north to the east edge of the Sierra the earthquake sequence.
Nevada (Dokka and Travis, 1990a; Dokka and Travis,
1990b; Savage et al., 1990). In the area of the Landers Landers-Big Bear Sequence Overview
rupture, the shear zone consists of an 80-km-wide band The Landers mainshock initiated at a depth of 3 to
of active right-lateral strike-slip faults (Dokka, 1983; 6 km on the Johnson Valley fault at 34.2 ° N, 116.4 ° W
Dokka and Travis, 1990a). The Big Bear earthquake did (Hauksson et al., 1993). The rupture propagated to the
not rupture to the surface. There remains some uncer- north, rupturing portions of the Johnson Valley, Kick-
tainty as to the structure which generated it, and it may apoo, Homestead Valley, Emerson, and Camp Rock faults
have been a double event, rupturing a pair of faults of (Sieh et al., 1993). Surface rupture was also reported on
conjugate orientation (Jones and Hough, 1993; Jones and other faults, including the Burnt Mountain and Eureka
Helmberger, 1993). Peak faults to the south of the mainshock. These faults
Within 30 hr of the earthquake, field crews from were probably ruptured by an aftershock which occurred
several institutions began to deploy Global Positioning within the first few minutes after the mainshock (Hough
System (GPS) receivers in order to measure co-seismic et al., 1993).
and postseismic displacements. Intensive field measure- Surface slip was variable, averaging 2 to 3 m over
ments continued for 3 weeks, and additional GPS sites much of the rupture (Ponti, 1993; Sieh et al., 1993).
were occupied later in the year. Hudnut et al. (1994) Two clear peaks in the slip distribution were observed,
present a summary of the measured co-seismic displace- a southern peak on the northern part of the Johnson Val-

646
The Co-Seismic Slip Distribution of the Landers Earthquake 647

117 ° 30'W 117" 15'W 117" 0 ~ W 116" 45'W t16" 30'W 116'15'W 116" 00'W 115"45'W
35" 00'N I . . . . . . . . . . . . . . ~. . . . . . . . . . . . . . ~. . . . . . . . . . . . . . , .............. j
. . . . . . . . . . . . . . ~. . . . . . . . . . . . . . I .............. 35"00'N

~SOAP //Z~HARVARD

~..IECT
34" 45'N 34" 45'N
A7007
,~o?

, OOl i
~CAJO
~EGUNDO

34" 15'N ABEAR .-" ,ILMEEKS 34" 15'N


•" ALMTECC

I 'Lu
..'" ~ONYX
•"' !
...'" OOW

,ooN l -o o ..1, / .... ooN

&c/u~

33" 45'N 33" 45'N


11T 30W 1 1 T 15'W 117" 00'W 116" 45'W 116" 30'W 116" 15'W 116" 00'W 115" 45'W

Figure 1. The GPS and Geodolite (EDM) sites used in this study. Sites ob-
served by EDM only are shown as open triangles, and sites with GPS observations
are shown as filled triangles. The EDM lines used are also shown. The mapped
surface rupture, based on Sieh et al. (1993), is shown in solid lines, and the
model faults are shown with dotted lines.

ley fault and a broader northern peak on the Emerson Measurement (EDM) both before and after the earth-
and southern Camp Rock faults. Peak surface displace- quake. An initial geodetic slip model was derived by
ments of 6 m were measured along the Emerson fault. Murray et al. (1993) based on this data set. They mod-
The slip distribution estimated from the moment-rate eled the slip distribution using 10 uniform-slip disloca-
function observed at the TERRAscope station PAS shows tions, and obtained a reasonable fit to most of the data.
a similar slip distribution, with two peaks separated by Their slip distribution did not match the observed surface
a trough with very little moment release (Kanamori et slip very well, especially in the southern part of the rup-
al., 1992). ture, but their estimate of the moment released, 8" 1019
N-m, was in agreement with other estimates, which range
Previous Geodetic Results from 8 to 11 • 1019 N-m (Kanamori et al., 1992; Sieh et
Far-field displacements were measured at the five al., 1993).
continuously operating GPS sites of the Southern Cali-
fornia Permanent GPS Geodetic Array (PGGA) (Bock et Observed Geodetic Displacements
al., 1993; Blewitt et al., 1993). These sites are at least
80 km from the nearest surface rupture. Bock et al. and The observed co-seismic displacements relative to
Blewitt et al. each report co-seismic displacements of up the far field are shown in Figure 2. These displacements
to 48 ram, based on the same raw GPS data but with are based on 38 displacement vectors measured with GPS,
independent analysis techniques. The displacements re- and 58 line-length changes measured by EDM. In this
ported by the two groups agree to within their uncer- figure, the GPS site at Vandenberg Air Force Base ( - 3 7 5
tainties. We included a subset of the data used by Bock km from the epicenter) is assumed not to have moved
et al. and Blewitt et al. in our analysis, and our dis- co-seismically. Only line lengths are measured with EDM,
placements for the PGGA sites agree with the results of so there are rotational and translational null vectors in
both groups to better than the uncertainties. the derived displacement field (e.g., Segall and Mat-
The U.S. Geological Survey (USGS) measured 58 thews, 1988). These null vectors are removed by esti-
line lengths between 30 sites by Electronic Distance mating displacements based on both the EDM and GPS
648 J. Freymueller, N. E. King, and P. Segall

117" 3 0 ' W 117" 15'W 117" 0O'W 116" 4 5 ' W 116" 3 f f W 116" 15'W 116" 0 0 ' W 115" 45'W
35"00'N i .............. i .............. , .............. ~. . . . . ~ . . . , . . . 1 ~ ............. i .............. ~. . . . . . . . . . . . . . J 35"00'N

~x

34" 45'N 0 34" 45'N

34" 30'N 34" 30'N

34" 15'N 34" 15'N

34" 00'N • 34" 00'N


k~

zk
2m

33" 45'N ............. i .............. [ .............. i .............. ~ ........... I. i . I . . . . . . . . . . . . . . i ............ It ' 33" 45'N
1 1 T 30'W 11T 15'W 11T 00W 116" 45'W 116" 30'W 116" 15'W 1160 O0'W 115" 45'W

Figure 2. Observed (heavy vectors) and model (light vectors) co-seismic dis-
placements with 95% confidence ellipses. These displacements are based on a
combination of GPS and EDM data. Sites with large error ellipses are those which
have only single-frequency GPS data prior to the earthquake. The model dis-
placements are based on the best model, with surface-slip constraints applied.
The different models of Figures 5 and 6 produce very similar model displace-
ments. On the scale shown here, the model displacements are indistinguishable.

data sets. The rotational ambiguity in the EDM displace- The GPS data used in this study comes from a va-
ments is resolved by the GPS displacements at the seven riety of sources. More details are given by Hudnut et al.
sites, which were measured by both methods. The trans- (1994). Prior to the Landers earthquake, no single in-
lational ambiguity is removed by assuming that Van- stitution had done a complete high-precision geodetic
denberg had no displacement. The slip inversions done survey of the entire area surrounding the Landers rup-
in this study used the relative GPS displacements and ture. Instead, several institutions had surveyed portions
EDM line-length changes rather than the derived dis- of the area. These data sets are overlapping and com-
placements of Figure 2, so they would not be affected plementary. With the exception of two pre-earthquake
if Vandenberg did in fact move. surveys carried out and analyzed by the USGS, all of the
raw GPS data used to derive the displacements used in
GPS Vector Displacements this study were analyzed at Stanford. Pre-earthquake data
When several sites are measured simultaneously came from five principal sources: the San Bernardino
within a region of a few hundred kilometers in size, one County Surveyor's Office GIMMS network, the Califor-
can use GPS to determine intersite vectors with subcen- nia Department of Transportation High Precision Geo-
timeter precision using standard data collection and anal- detic Network (HPGN) survey, the USGS, the Caltech
ysis techniques. The positions of the sites relative to a and County Surveyors' Inter-County surveys, and the
distant global network of GPS sites (Fig. 3) can also be permanent stations of the PGGA. The GIMMS network
determined with a precision of roughly 1 cm. Displace- is notable in that single-frequency GPS receivers were
ment vectors are measured by differencing the GPS-de- used, which seriously limits the precision of the survey
termined positions from two epochs. For this study, the and requires special care in data analysis. The precision
displacements have been computed relative to the per- of that survey is further limited by the very short (1 hr)
manent GPS site at Vandenberg Air Force Base, the clos- observation sessions used, in contrast to the 8 to 24-hr
est permanent site unaffected by the earthquake. sessions used in a typical high-precision GPS survey.
The Co-Seismic Slip Distribution of the Landers Earthquake 649

OKR

".,A"

Figure 3. Global GPS network. The global GPS sites used in this study are
shown with triangles. Solid triangles are used for the sites which were con-
strained--see the text for details.

However, this data is valuable, because many of these ospheric models were generated using dual-frequency data
sites are located in areas near the rupture zone where no from nearby permanent GPS sites (Goldstone, Pinyon,
higher-precision GPS sites were measured prior to the JPL, and Scripps). The relative ionospheric delays de-
earthquake. Postearthquake GPS data taken by Caltech, termined from these sites were used to estimate coeffi-
Jet Propulsion Laboratory (JPL), the Southern California cients for a polynomial expansion of the TEC as a func-
Earthquake Center (SCEC), Scripps Institution of Ocean- tion of position and time. We used a model that assumes
ography, Stanford, and USGS were used in our solu- all free electrons are concentrated in a thin spherical shell
tions. 350 km above the surface of the earth. The point where
the ray from the satellite to the receiver pierces the shell
Single-Frequency GPS Data. The analysis of single- can be computed from the receiver position and satellite
frequency GPS data poses several challenges. The data orbits. The model parametn'.'zes the path delay as a func-
must be treated specially, as it is biased by errors re- tion of the latitude and hour angle of the piercing point,
suiting from propagation delays as the GPS signals pass and the angle of incidence of the incoming ray (Geor-
through the ionosphere. With dual-frequency data it is giadou and Kleusberg, 1988). The hour angle is deter-
possible to use the measurements on the second fre- mined by longitude and time of day.
quency to eliminate the ionospheric path-delay error. We The GIMMS survey was carried out using two to seven
used the Bernese GPS software for this part of the anal- 60 to 120-rain sessions per day. The receivers did not
ysis because it is well suited to handling single-fre- record data between sessions. An independent iono-
quency data. To simplify the analysis of this data, we spheric model was generated for each session, using a
used the Scripps PGGA orbits rather than the less precise second-order polynomial in hour angle and a first-order
broadcast ephemeris. The error induced by the propa- polynomial in latitude. On average, these models re-
gation delay is proportional to the integrated total elec- move the systematic bias but cannot remove the local
tron content (TEC) along the propagation path. The path distortions. The integrated TEC was highly variable over
delay is spatially and temporally variable. Uncorrected the time of the single-frequency survey, which ran from
ionospheric path-delay error will systematically bias the October 1991 to May 1992. The modeled TEC varied by
estimated coordinates for all stations (the entire network up to an order of magnitude between the most active
of stations will appear to shrink), and small-scale vari- days and the quietest days. Variation within each day
ations in the integrated TEC will cause local distortions was much smaller but still substantial. Rapid variations
in the estimated coordinates as well. in the TEC are not modeled well with this parameteri-
To correct the single-frequency data, empirical ion- zation, and some residual ionospheric errors may remain
650 J. Freymueller, N. E. King, and P. Segall

on such days. The stations in the northern part of the cluding the PGGA Goldstone site (Goldstone tracking
G1MMS network (especially 7000, 7001, and DUMP) were complex near Barstow), are on the North American side
observed on days with large and variable TEC, and these of the San Andreas fault system, but the PGGA JPL (at
solutions could contain biases. Jet Propulsion Laboratory) and Pinyon (Pinyon Flat Ob-
Each day of the GIMMS network was analyzed in- servatory southwest of Palm Springs) sites are on the
dependently. Data from all sessions on each day were Pacific side. The positions of sites in the Landers area,
analyzed together to give a single estimate for the co- which were measured at different times, are related to
ordinates of the stations observed that day. All of the each another via the PGGA sites, which are assumed not
independent daily solutions were then adjusted together to deform over the entire time period. Any motion of the
to determine the best-fitting coordinates for all of the PGGA sites would introduce a bias in the positions of
sites based on all of the data. The covariance of the co- sites in the Landers area. The magnitude of the bias is
ordinates was scaled by the misfit of the adjustment so not easy to estimate, but is small compared to the co-
that the reduced chi-square statistic was unity. The rms seismic displacements.
misfit of the adjustment was reduced by roughly 50% by The Stanford GPS solutions used a global network
applying the ionospheric models. Even so, the precision of GPS stations (Fig. 3) to estimate precise satellite orbits
of the pre-earthquake positions for these stations is fairly and provide a consistent frame of reference for deter-
low, with the best being determined to about 30 mm. mining co-seismic displacements. The global network
Most positions were determined to about 50 mm. These consists of ~25 high-quality GPS receivers which track
data are inherently noisier than the more precise dual the GPS satellites 24 hr a day. These data were analyzed
frequency data, and despite the best efforts may contain simultaneously with the dual-frequency data from sta-
systematic biases owing to the ionosphere. Such a bias tions in the Landers area. Station coordinates were es-
would appear as an anomalous expansion or contraction timated for all stations, and precise satellite orbits were
of a part of the network. estimated for all satellites. The positions of several sta-
tions distributed around the globe (ALGO, CANB, FAIR,
Dual-Frequency GPS Data. We analyzed most of the HART, KOKR, ONSA, and PENT; see Fig. 3 for loca-
dual-frequency GPS data at Stanford using the GIPSY/ tions) were assigned a priori uncertainties of 5 cm; the
OASIS II software developed at the Jet Propulsion Lab- a priori site positions were based on the Goddard Space
oratory of the California Institute of Technology, Pre- Flight Center VLBI model GLB753. These positions
earthquake positions for a few sites were provided by should in theory be accurate to 2 to 3 cm so a 5-cm a
USGS, computed using the Bernese GPS software. Com- priori constraint is conservative. Independent solutions
parison of the coordinates of two common sites, relative were computed for each day of data. The remaining sites,
to the PGGA site Pinyon, showed agreement at the level including all those in the Landers area, were given a
of - 1 cm. Therefore, combining the two data sets should priori uncertainties of 1 kin. The positions of most sites
not introduce significant errors into the displacements measured with dual-frequency receivers are determined
despite the differences in the analysis. to about 10 mm or better. The exceptions are some sites
The pre-earthquake data come from three time pe- which were observed for short periods of time; these are
riods. The earliest data are from the HPGN (we used 6 determined to about 20 to 30 mm. The positions of the
days from the period April through July 1991). Data from PGGA sites are determined to 3 to 4 ram.
the Caltech-led Inter-County network (9 days in April Coordinates from all of the pre-earthquake solutions
1992) and from the 5 days of PGGA data immediately were adjusted together to form a complete, consistent set
preceding the earthquake complete the pre-earthquake of pre-earthquake coordinates. The adjustment was car-
data set. Postearthquake data comes from the 21 days ded out in two steps. In the first step, all the of the data
immediately following the earthquake. from each of the four pre-earthquake networks (GIMMS,
No corrections were made for tectonic motion. Over HPGN, Inter-County, and June 1992 PGGA) were ad-
the year between the HPGN survey and the earthquake, justed to form a complete set of coordinates for each
up to about 3 cm of accumulated relative displacement individual network. The four resulting sets of coordi-
(a rough approximation based on the San Andreas fault nates were then adjusted together to form the final set
average slip rate) could have accumulated across the en- of pre-earthquake coordinates. In the final adjustment,
tire network, including the PGGA sites. Roughly half as only California sites were adjusted together, to avoid the
much displacement could have accumulated after the ob- problem of correcting for global plate motions. The po-
servation of portions of the GIMMS network. Secular de- sition of the Landers-area stations relative to the global
formation is not an issue with the single-frequency GIMMS network was based only on the June 1992 PGGA data.
network, however, since the uncertainties in the dis- A similar process was used for the postearthquake data,
placements for those sites are considerably higher than except that only one step was required since the post-
the potential accumulated secular deformation. Virtually earthquake data set spanned only 21 days. In each ad-
all of the HPGN and other sites used in this study, in- justment, the covariance of the coordinates were scaled
The Co-Seismic Slip Distribution of the Landers Earthquake 651

by the reduced chi-square statistic of the weighted ad- mm may have accumulated between the last survey and
justment residuals. Prior to scaling, the reduced chi-square the earthquake.
statistic (for the global solutions) was typically in the The Mojave EDM network spans the Homestead
range of (2 to 2.5) 2, indicating that the formal uncer- segments of the Landers rupture zone. Most of these lines
tainties on a global scale were slightly underestimated. are west of the rupture. The USGS observed the lines of
The covariance of the single-frequency GIMMS survey the western Mojave network only twice, with EDM in
was scaled by a factor of about 82, a result of both the 1982 and with GPS in 1992. Thus the "co-seismic"
ionospheric errors which remain in that data set and the changes for the lines of the Mojave network contain 10
tendency of the Bernese software to underestimate the yr of secular deformation. Savage et al. (1990) used
covariance of the estimated coordinates. triangulation and EDM data to calculate the deviatoric
strain rate from 1934 to 1982 for the western Mojave
EDM Data network. The principal strain rates are 0.08 --- 0.02 and
-0.08 --- 0.02 microstrain per year, with the axis of
Four separate EDM networks (Fig. 1), consisting of maximum strain oriented N86 ° W --- 5°. Triangulation
58 lines between 30 sites, span the Landers rupture zone data does not allow the dilatational component of strain
(Savage et al., 1993). The networks which span the to be estimated. We assume that these results describe
southern part of the Landers rupture zone and the region the secular deformation between 1982 and 1992, and use
to the south were observed shortly before and shortly them to estimate a preseismic correction for each line of
after the earthquake. The coverage of the northern Lan- the western Mojave network. In the approximately 10 yr
ders rupture zone is not as good; most lines are north or between the two USGS surveys, the principal strain rates
west of the rupture, and the last pre-earthquake EDM predict that lines parallel to the maximum or minimum
surveys were done 3 to 10 yr prior to the earthquake. To axis should change by approximately 0.8 or - 0 . 8 parts
correct for the varying index of refraction along the beam per million. Most of the predicted secular changes are
path, temperature and humidity measurements are made less than 10 mm in absolute value. The largest is a 21.2-
from a small plane during ranging. With these measure- mm contraction for Lucerne South Base-Ord, the long-
ments, plus endpoint pressures, the precision of the Geo- est line (26.8 km) of the western Mojave network. For
dolite EDM is (a2 + b2L2)1/2, where a = 3 mm, b = each line, we subtracted the estimated secular change
2.10 -7, and L is baseline length in kilometers (Savage from the observed 1982 to 1992 change. We assume that
and Prescott, 1973). The Geodolite lines in the Landers the remaining signal is the co-seismic deformation as-
area are 6.7- to 31.7-km long, with corresponding pre- sociated with the Landers earthquake.
cision of 3.3 to 7 mm.
The 38-line Joshua EDM network spans the southern Estimation of the Co-Seismic Slip Distribution
half of the rupture zone (Landers and Johnson Valley
segments). It also spans the Pinto Mountain fault east of The estimation of the slip distribution on a fault from
the epicenter, and extends southward to the Joshua Tree geodetic displacements is inherently nonunique. Many
epicenter and the San Andreas fault. The USGS surveyed different slip distributions can produce the same dis-
the Joshua network in late April 1992, after the Joshua placements at any set of points on the surface [for a fur-
Tree event and just 2 months before the Landers earth- ther discussion, see, for example, D u e t al. (1992)]. We
quake. The postseismic survey was done in the first week will restrict ourselves to models having pure strike-slip
of August 1992, about 1 month after the earthquake. The faulting on a given set of faults which approximate the
Edom monitor array (Fig. 1) spans the San Andreas fault trace of the surface rupture. Furthermore, the sense of
about 15 km southwest of the Joshua Tree epicenter. Three slip will be restricted, so only models with right-lateral
lines southwest of the Joshua Tree epicenter. Three lines slip on the Landers faults and left-lateral slip on the Big
southwest of the San Andreas fault supplement the Joshua Bear fault will be used. To regularize the problem, we
network coverage of the 1992 earthquakes. The USGS will also require the slip distribution to be smooth by
observes this array with both EDM and GPS. The last minimizing a weighted linear combination of the data
preseismic EDM and GPS observations were done only misfit and the roughness of the slip distribution.
2 months before the earthquake. Postseismic EDM and The fault model is specified by the location, orien-
GPS surveys were done I month after the earthquake. tation, and dimensions of all faults which slipped in the
The Barstow EDM network is north of the Landers rup- earthquake. Each fault is subdivided into a regular grid
ture. The southernmost station, Lava, is about 15 km of subfaults. We model the Landers rupture using a sim-
northwest of the northern tip of the rupture zone. The ple five-plane model, with each plane subdivided into 2-
USGS Observed two lines out of Lava five times between by 2-km elements (Table 1). The five planes correspond
1979 and 1989, and resurveyed these lines with GPS af- to the five major faults that ruptured, and are comparable
ter the earthquake. The preseismic rates of change were to the parametrization used by Murray et al. (1993). This
1 to 2 mm/yr in absolute value, suggesting that about 7 model ignores the discontinuities in the surface rupture;
652 J. Freymueller, N. E. King, and P. Segall

Table 1
Model Fault Parametrization*
Extends From Extends To
Number of Horizontal
Fault Latitude Longitude Latitude Longitude Subfaults

Camp Rock 34°39.75 ' - 116°41.20' 34°34.05 , - 116%4,32' 7


Emerson 34°34.05 ' - 116°34.32 ' 34°30.22 ' - 116°31.33' 4
Homestead Valley 34°30.22 ' - 116°31.33' 34°23,50 ' - 116°28,00 ' 7
Landers 34°23.50 ' -116°28.00 ' 34°18,50 ' -116°26.17 ' 4
Johnson Valley 34018,50 ' -116°26.17 ' 34°08,17 ' -116°25.06 ' 9
*All faults extend from the surface to 18-kin depth, and have 9 vertical subfaults.

it assumes a continuous rupture across all five faults. A that the misfit is nonzero even when no smoothing is
comparison of the model geometry to the actual surface used, indicating that no model of the assumed geometry
rupture is shown in Figure 1. The geometry is a good satisfying the positivity constraints can perfectly fit the
approximation to the surface rupture in the northern part data. The misfit is nonzero without positivity constraints
of the rupture, but has shortcomings where the surface as well.
rupture stepped over from the Johnson Valley fault to One might choose the smoothest solution which does
the Homestead Valley fault (via the previously un- not appreciably increase the misfit above the level of the
mapped Kickapoo fault). unsmoothed solution. Alternatively, the smoothing weight
Additional fault elements were introduced at each could be chosen so that the maximum slip at depth re-
end of the grid and below the grid. These elements sim- mains "reasonable." However, these methods of deter-
plify the boundary conditions for the smoothing operator mining the optimum smoothing weight are very subjec-
and were constrained to have zero slip. The model grid tive. The question of the proper weight to be given to
extended to 18-km depth, significantly deeper than the smoothing is critical--with less smoothing the solutions
depth of the deepest aftershocks, in order to minimize have higher maximum slip and a smaller rupture area,
the effect of the lower boundary condition. The model while solutions with a larger weight given to smoothing
grid was extended beyond the edges of the principal sur- will have smaller slip amplitudes and a larger overall
face rupture in order to minimize the effect of the end rupture area.
boundary conditions. The model is not significantly al- One objective means to determine the optimum
tered if the slip on these elements is not constrained, and smoothing is to choose the solution with shallow slip
the preferred estimate for these elements is zero slip. which best agrees with the observed surface-slip distri-
The model relating the observed geodetic data d to bution. In this case, agreement with the surface slip did
the fault slip distribution s is not prove to be a useful method to determine the opti-
mum smoothing weight, since there were two places on
d = Gs, (1) the fault where the geodetic slip model disagreed sig-
nificantly with the observed surface slip for all values of
where the equations have been weighted so that the data /3 which were tried (this included models which weighted
have unit covariance. The model matrix G is computed roughness more than 2000 times more than misfit). One
assuming a uniform elastic half-space with a Poisson's of these two areas of disagreement with the observed
ratio of 0.25. We minimize the functional surface slip remained even when the surface-slip obser-
vations were included as prior information, and will be
F(s,,8) = IGs - dl 2 +/32LHs[ 2, (2) discussed later.
The statistical method of cross-validation provides
subject to inequality constraints s, --- 0 for all subfaults another objective means to determine the optimum weight
i. Dextral slip is considered to be positive, so for the given to smoothing (Matthews and Segall, 1993; Arna-
assumed sinistral Big Bear rupture s, is replaced by -si. dottir, 1993). Cross-validation measures the ability of a
The smoothing operator H is a finite-difference approx- model to predict independent data. A model is deter-
imation of the Laplacian operator (Harris and Segall, mined using all but one of the data points, and this model
1987). The first term of (2) is the squared L2 norm of is used to predict the value of the datum that was left
the residual vector, and will be called the misfit. The out. This process is repeated for each datum. The sum
second term will be called the roughness. The term/32 of the squares of the prediction errors (called the Cross-
is a coefficient that weights the penalty on roughness Validation Sum of Squares, or CVSS) is a measure of
relative to misfit. Thus, varying/3 produces a family of the model's predictive ability. The model with smallest
solutions of varying roughness and misfit (Fig. 4a). Note CVSS is in principle the optimal model. The CVSS for
The Co-Seismic Slip Distribution of the Landers Earthquake 653

models with different weights given to smoothing are


25OO shown in Figure 4b. The minimum CVSS is found for a
model with f12 = 0.0625. This model is quite rough, and
2250 .
features an unexpectedly high slip maximum of 19 m.
A possible bias in determining the optimum smooth-
2000
ing comes from the simplified model geometry used in
1750 this study. By ignoring some of the complexity of the
real faults, we force the model to become rougher in
1500 I ~,--More s m o o t h i n g (beta = 1.0) order to continue to satisfy the data. This bias can affect
~[ ~Preferred solution(beta= 0.5)
cross-validation as well as the more subjective measures
"~ 1250 ' = ' of determining the optimum smoothing. In particular, the
near-field sites are most sensitive to both real variations
lOOO
in slip and inadequacy in the model geometry. Prediction
750
residuals from near-field sites dominate the CVSS for all
values of smoothing. Arbitrarily downweighting the near-
500 field sites would counter this possible bias but would
also reduce our ability to determine whether there were
250 any real short-wavelength features in the slip distribu-
tion. For now, we prefer a slightly smoother model than
0
5 10 15 20 25 the model with the CVSS, but the question of the opti-
Roughness mum smoothing remains only partially answered.
(a) We could not obtain an acceptable fit to the data
from three EDM lines and one GPS vector. The misfit of
these EDM lines was of the order of 15 to 20 tr, while
the misfit of other lines was 2 to 5 tr on average. The
i ,
GPS vector which could not be fit was the site Meeks,
\
10500 - j \ located in the foothills of the San Bernardino Mountains
i \ west of the hypocenter. One of the EDM lines which
f '\
could not be fit included this site. Since the EDM and
9000" GPS data were misfit in a consistent way by about l0
'\\ cm, we concluded that there was probably local defor-
"\
mation near this station and did not use its data in the
\\\
~ ~ ..~ ~ " ~O~ ~ J ' ~ " ~ ~ r ~ ' J "~~
inversions. The other EDM lines with a large misfit, Old
L)
W o m a n - R o c k (west of the northern part of the rupture)
and Edom-Warren (crossing the San Andreas fault south
of the rupture), were also deleted. It is worth noting that
these data, including the lines to Meeks, were not sig-
nificantly misfit by the model of Murray et al. (1993),
but the three lines which were significantly misfit (24 to
47 o') by their model are fit very well by our model.
The preferred slip model is shown in Figure 5a, with
the smoothing parameter/32 = 0.25 km 2. Models with
• • , . . . . , . . . . , . . . . i . . . .

5 10 15 20 less (the minimum CVSS model) and more smoothing


Beta
are shown in Figures 5b and c, respectively. A model
(b) which includes constraints based on the observed surface
slip is shown in Figure 6. Certain features are clearly
Figure 4. (a) Trade-off curve between rough- visible in all of these models. Virtually all the slip is in
ness and misfit. Varying the smoothing parameter
produces a family of solutions of varying misfit the upper 10 to 12 km, in excellent agreement with the
and roughness. The preferred solution is the one aftershock distribution reported by Sieh et al. (1993).
which does not unduly increase the misfit to the Two major patches of slip release are clearly identified,
data. Based on this criterion, the preferred solu- a southern patch on the Johnson Valley fault and a north-
tion is for fl = 0.5. (b) The CVSS as a function ern patch on the Homestead Valley and Emerson faults.
of the smoothing parameter ft. There is no sharp
minimum, but the preferred solution has ~ = 0.25. The centroid of the southern patch is 2- to 4-km deeper
This is the solution which is best able to predict than that of the northern patch. The southern patch is
independent data, based on the "leave one out" located about 8 km north of the hypocenter. There is
approach of cross-validation. only about 2 m of slip at the hypocenter. Observed and
654 J. Freymueller, N. E. King, and P. Segall

a. Preferred model ( [52 = 0.25 )

18km

b.

L._ z~,,;~ ::: I IIII I

Right lateral slip (m)


18kin
Figure 6. Model with surface constraints. This
model uses the same smoothing parameter as the
C, preferred model of Figure 5a, but adds constraints
o
based on the surface slip. The essential features
of the models of Figure 5 are not changed by add-
ing surface constraints.

18 km

Table 2
Moment Release Summary (The Total Moment
Released was 9 × 10~9 N-m)
Fault Fraction of Total Moment

Camp Rock 7%
Emerson 23%
r-- i I III III I
Homestead Valley 38%
, Right lateral slip (m) Lenders ! 1%
Johnson Valley 21%
Figure 5. Models with no surface constraints.
All three models are shown in perspective view,
with north to the left. The three models are (a) the
preferred model, (b) a model with less smoothing,
and (c) a model with more smoothing. The es- Rock faults, so there could be trade-offs in the estimates
sential features of the preferred model are also of moment release for the individual faults.
present in the other models: virtually all slip above
10- to 12-km depth, and two distinct patches of The surface slip predicted by our model is larger and
high slip on the Johnson Valley fault and the more variable along strike than the observed surface slip,
Homestead Valley and Emerson faults. The mag- unless surface-slip constraints are added (Fig. 7). The
nitude of the slip maximum decreases as smooth- surface-slip distribution used for comparison is from field
ing is increased. mapping done by the USGS (Ponti, 1993). The surface
slip constraints are added by including the geological
surface-slip observations as data in the inversion. The
model displacement vectors for the preferred model are uncertainties in the surface-slip data were assigned based
shown in Figure 2. The rms misfit, rms [ ( o b s e r v e d - on the variability of the observed surface slip within each
c o m p u t e d ) / t r ] , of the preferred model is 2.4 for the 167 element (subfault). Typical assigned uncertainties ranged
data used to derive the model. from 50 to 200 cm. With the surface-slip constraints
The total seismic moment of the model is 9" 10 t9 added, the misfit of the GPS and EDM data increased by
N-m, in good agreement with the moment obtained by 18%, but the overall features of the model are not changed.
seismic means (Kanamori et al., 1992; Sieh et al., 1993). The geodetic and surface data are incompatible only in
In our preferred model, 61% of the moment was released one location, on a portion of the southern Homestead
by slip on the Homestead Valley and Emerson faults, Valley fault. At this location, 2.5 m of surface slip was
and 21% on the Johnson Valley fault (Table 2). Our model observed, but the model predicts 9 m of slip. The site
geometry does not distinguish between the overlapping Maumee (GPS and EDM) is located very close to this
segments of the Homestead Valley, Emerson, and Camp fault segment (Fig. 1). A subsidiary fault not included
The Co-Seismic Slip Distribution of the Landers Earthquake 655

1500.

-,-.-o--

1000- ....... .@._

":A
..-- ? il :':" :J~
:
.... •o - .

500 -

.." .: .~ , ,;, /" .:


,~ i ":.'. :
i " i ,: !
.~ " -: A' 0
i " £-q
0
-20 0 20 4O 60

Distance along fault trace (km)

Figure 7. Surface-slip comparison. The surface slip predicted by the models


with and without surface-slip constraints are compared to the observed surface
slip (Ponti, 1993). Without surface constraints, the geodetic model generally
overestimates the observed surface slip, but the major peaks in both models are
located in similar places. With surface constraints, the only significant difference
between the model and the observed surface slip is at 22 km from the epicenter,
where the model predicts close to 900 cm of slip, versus the 250 cm observed.
The GPS and EDM site Maumee is located very close to the fault at this location
(see text).

in the model which had up to 50 cm of surface offset The Big Bear Fault
(Ponti, 1993) is located close to Maumee. The high sur-
face slip in the model at this point is probably a result Regardless of the degree of smoothing, our models
of local motion at Maumee which cannot otherwise be placed very little slip on the Big Bear fault. Peak slip in
explained with our model fault geometry. Because dis- our model for Big Bear is just over 1-m left lateral, but
placement decreases as - 1 / r 2, where r is the distance the slip is found only in a small area near the station
to the fault, a small offset on an unmodeled fault near ONYX (the closest to the fault). The area of nonzero slip
to the geodetic station can be mapped into large slip on is located a considerable distance northeast of the hy-
the model fault. pocenter. This moment release is only a few percent of
There are two other patches of high slip which are the seismic moment, and it is not clear why the esti-
present in our models (Figs. 5 and 6) which are not pres- mated moment is so low. In the preferred model, there
ent in the surface slip, and are unlikely to be real. These is 300 times more moment release on the Landers faults
are on the southern Johnson Valley fault, south of the as on the Big Bear fault. We also tried models which
epicenter, and the northern end of the Camp Rock fault. used a single uniform dislocation to describe the fault
They probably also appear for the same reasons as the which ruptured in the Big Bear event, and these models
anomalous slip near Maumee, local deformation and an also had very little slip. More slip on the Big Bear fault
inadequacy in the model fault geometry. Both of these is found when data is included from Meeks (a GPS and
patches occur where a low-precision GPS site is located EDM site near the east end of the Big Bear aftershock
within roughly 1 km of the surface trace of the rupture. zone which is very poorly fit and was not used in the
The patch on the southern Johnson Valley fault is de- models presented here). However, including data from
termined primarily by the displacement of the site 6054, Meeks significantly increases the overall misfit, and the
located on Old Woman Springs Road within a few hundred slip on the Big Bear fault is concentrated at the extreme
meters of the fault trace (Fig. 1). The slip patch at the northeast end of the fault, far from the hypocenter.
northern end of the rupture is determined primarily by There are only two other sites (Bear and Onyx) in
the displacement of the site 7000, located on Camp Rock the near field of the Big Bear event. The Big Bear event
Road within about 1 km of the surface trace. The dis- did not rupture the surface and had a much lower mo-
placement of 7000 is poorly fit by the model, with a ment than the Landers event, so the surface displace-
large fault-normal motion component not explained by ments for this event are about an order of magnitude
the model (Fig. 2). smaller than those of the Landers event. To estimate the
656 J. Freymueller, N. E. King, and P. Segall

ability of our data to resolve slip on the Big Bear fault, a. Synthetic slip (5 m) placed on Johnson Valley fault
we generated synthetic data for a model with 1 m of left-
lateral slip from depths of 2 to 12 km on the Big Bear
fault, and inverted this synthetic data. The imaged slip
distribution was very similar to the input distribution.
We tried models with no slip on the Landers faults, and 18
with realistic slip distribution on the Landers faults, and
in all these synthetic models the estimated slip distri-
bution had about 1 m of left-lateral slip on the Big Bear
fault. So, it seems that an appropriate amount of slip on
the Big Bear fault should be in our estimated model if
there was actually slip on the assumed fault. However,
all that can be said with certainty is that the geodetic
b. Synthetic slip (5 m) placed on Emerson and Homestead Valley faults
data can be satisfied without significant slip on the Big 0
Bear fault. More stations close to the Big Bear fault,
especially to the west, would be required in order for the
geodetic data to constrain the amount of slip on this fault
and answer questions about the orientation of the fault. tSkm

The Eureka Peak Fault


Co-seismic or immediate postseismic surface slip on
the Eureka Peak fault totaled about 20 cm (Hough et al.,
1993). Hough et al. suggested that this fault ruptured
during an aftershock only minutes after the mainshock,
and that it may have ruptured only minutes previously
during the mainshock. The geodetic data cannot shed any '! ill
light on the question of timing, but slip on this fault is Right lateral slip (m)
required to adequately fit the EDM data from nearby sites
(chiefly Warren). This fault was modeled as a uniform Figure 8. Synthetic slip models. Synthetic slip
dislocation extending from the surface to 12 km. Our models were constructed to test the resolution of
model estimates 35 cm of slip on this fault. If the actual the model (a) near the southern slip patch on the
Johnson valley fault and (b) near the northern slip
fault ruptured to 18 kin, the estimated slip would drop patch on the Emerson and Homestead Valley faults.
to 25 cm, although the misfit is not significantly differ- Synthetic data were computed from input models
ent. We prefer the model with slip to 12-km depth be- which had 5 m of slip on all elements within the
cause this is the depth of slip observed on the other faults. thick solid lines. The models estimated from the
synthetic data are shown.
Model Resolution
We tested the resolution of the model by generating
synthetic data from two sample slip distributions and in- slip distribution. The better the imaged slip agrees with
verting those data. The slip distributions tested were a the input slip, the greater the resolution of the model in
model with 5 m of slip from 8- to 12-kin depth on the that region.
northernmost segments of the Johnson Valley fault (the The imaged slip distribution for the first synthetic
southern slip patch), and a model with 5 m of slip on model, for slip on the northern Johnson Valley fault, is
the northern Homestead Valley and Emerson faults at the shown in Figure 8a. The elements which had 5 m of slip
4- to 6-km depth (the northern slip patch). The inversion in the input model are outlined in heavy lines. The peak
of synthetic data is a good test of the resolution of the slip in the imaged distribution model is only 2.6 m, but
model, as it shows how a known slip distribution would the total moment of the imaged slip is slightly higher
be imaged given the actual data distribution and covari- than that of the input slip. The centroid of the imaged
ance. Taken together, the earth's elastic response, the slip patch is slightly deeper than that of the actual slip
measurement of geodetic data at a finite number of points, patch, suggesting that the data distribution systemati-
and the inversion process which gives some weight to cally images slip deeper than it actually is in this region.
the roughness of the slip distribution act as a filter which Another synthetic model with input slip placed closer to
maps the input "truth" slip distribution to the output es- the surface confirms this--slip on this part of the fault
timated slip distribution. Because of these effects, we do is imaged as being 1 to 2 krn deeper than it actually is.
not expect to recover exactly the input slip distribution. There is very little spread in the slip distribution along
Instead, slip will be smeared out to produce a smoother strike. The imaged slip for the Homestead Valley/Emer-
The Co-Seismic Slip Distribution of the Landers Earthquake 657

son synthetic model is shown in Figure 8b. The input There was significant slip on at least three separate
slip distribution had 5 m of slip on 8 elements, all at 4- faults: the Johnson Valley, Homestead Valley, and
to 6-km depth (outlined in heavy lines). Peak imaged Emerson/Camp Rock faults (the model does not distin-
slip was about 3 m, but again the moment of the imaged guish between the Emerson and the southern Camp Rock
slip patch was slightly greater than that of the input patch. faults, which are parallel and separated by only 1 to 2
The input slip patch had a line of elements with constant km at the surface). It is not clear from the geodetic model
slip, but the output model has two distinct areas of con- whether slip was transferred smoothly at depth from the
centrated slip. This shows that the model lacks resolu- Homestead Valley to the Emerson and Camp Rock faults.
tion in this area, so it cannot be used to infer the details It is clear that there was significant moment release on
of the slip transfer from the Homestead Valley fault to both faults, regardless of the nature of the overlap. The
the Emerson fault. Otherwise, there is no significant near-simultaneous rupture of all of these faults suggests
spread in depth or along strike. that they must have been close to brittle failure prior to
the earthquake. Along with the seismicity triggered by
Discussion the passage of waves from the Landers mainshock (Hill
et al., 1993), this may imply that broad areas of the crust
Our slip model has two distinct patches of high slip, were close to failure at the time of the Landers main-
and thus high moment release, on the faults that ruptured shock.
during the Landers event. The two patches correspond The slip distribution of our model is in good agree-
to the two peaks in the seismic moment release as a func- ment with the slip models of Cohee and Beroza (1994),
tion of time (Kanamori et al., 1992). The southern patch based on the inversion of strong-motion seismic data.
is located on the northern Johnson Valley fault, about 8 Both models image similar structures in similar loca-
km north of the hypocenter, at a depth of 8 to 12 km. tions, despite using different model geometries and com-
In the preferred model, the peak slip on this patch is pletely independent data. The Cohee and Beroza model
roughly 8 m, at a depth of 8 to 12 km, with surface slip has only two-thirds the total moment release of our model,
of 3 to 4 m. Based on the inversion of synthetic slip which may be due to the fact that they use only a portion
models, the actual slip may have been 1 to 2 km shal- of the frequency band while the geodetic model is es-
lower than the model indicates. To the north of this patch sentially a DC offset. Except for the difference in slip
the slip at all depths decreases until the northern half of magnitude, the two models are very complementary. Wald
the Homestead Valley fault. Slip on the Homestead Val- and Heaton (1994) used a similar geodetic data set in a
ley and Emerson faults was concentrated in the depth joint inversion with seismic data to determine the spatial
range 4 to 8 km, with a peak slip of 16 m on the southern distribution and time history of the rupture. They used
Emerson fault. Continuing to the north, the depth of the a slightly different model geometry than in our model,
slip maximum decreases--on the Emerson fault the slip and they gave all displacements equal weight rather than
maximum is at a depth of 2 to 6 km. The model does using the full covariance of the geodetic data. Despite
not have enough resolution to determine whether this slip these differences, their inversion of the geodetic data
patch is a single continuous rupture across the two faults produced a model which is generally similar to ours. The
or two separate ruptures separated in space. The model primary difference between the models is that their model
puts very little slip at the hypocenter, suggesting that the is much smoother than ours.
earthquake did not initiate solely as a result of a high Our model uses a similar fault geometry to the geo-
potential for slip at that point. This has been observed detic slip model of Murray et al. (1993), and provides
for a number of other earthquakes, including the 1966 a better fit to the geodetic data. They modeled the slip
Parkfield earthquake (Segall and Du, 1993) and the Loma distribution using 10 planar faults, each with a uniform
Prieta earthquake (e.g., Beroza, 1991). dislocation over its entire surface. The Johnson Valley,
Even in models with much more smoothing than those Homestead Valley, and Camp Rock faults were subdi-
presented here, the peak slip at depth is greater than vided into two segments, and the other three main faults
the slip observed at the surface. Of course, the magni- that make up the Landers rupture were modeled as single
tude of the peak slip is dependent on the weight given segments. Additional segments were used to represent
to smoothing. It is important to choose the optimum the Big Bear rupture and the Eureka Peak fault. Their
smoothing in an objective way, otherwise the resulting model did not agree with the surface slip in the southern
model will simply reflect our subjective prejudices about part of the rupture, but did place the largest amount of
what the slip distribution should look like. We have cho- slip on the northern Homestead Valley and Emerson faults.
sen the optimum smoothing based on two criteria--the They obtained 1 + 1.4 m of left-lateral slip on the Big
results of cross-validation and the trade-off curve of mis- Bear structure, which gives a more reasonable moment
fit versus roughness. It is possible that these models could for Big Bear than does our model. However, this amount
be biased toward a rougher solution because of the sim- of slip is not significantly different from zero. We sug-
plified model geometry used in this study. gest that the main shortcoming of their model is that it
658 J. Freymueller, N. E. King, and P. SegaU

did not allow for sufficient spatial variation in slip, es- References
pecially depth variation, and that uniform dislocations
were not representative of the true pattern of slip on the
Arnadottir, T. (1993). Earthquake dislocation models derived from
faults.
inversion of geodetic data, Ph.D. Thesis, Stanford University,
Murray et al. (1993) were unable to fit acceptably Stanford, California.
certain EDM lines, and increased the uncertainties of those Blewitt, G., M. Heflin, K. Hurst, D. Jefferson, F. Webb, and J.
lines in their model. Our model fits those lines very well, Zumberge (1993). Absolute far-field displacements from the 28
but was unable to fit two other lines which were fit well June 1992 Landers earthquake sequence, Nature 361, 340-342.
Beroza, G. (1991). Near-source modeling of the Loma Prieta earth-
by their model. They suggested that there could be sub- quake: evidence for heterogeneous slip and implications for
stantial sympathetic slip on faults off the main Landers earthquake hazard, Bull. Seism. Soc. Am. 81, 1603-1621.
structure, which could result in significant local defor- Boek, Y., D. Agnew, P. Fang, J. Genrich, B. Hager, T. Herring,
mation. This is certainly possible, but a comparison of K. Hudnut, R. King, S. Larsen, J. Minster, K. Stark, S. Wdow-
the misfit of our model and theirs suggests that limita- inski, and F. Wyatt (1993). Detection of crustal deformation from
the Landers earthquake sequence using continuous geodetic mea-
tions of the slip models may also explain the misfit of
surements, Nature 361, 337-340.
the data. There is no compelling evidence for more than Cohee, B. and G. Beroza (1994). An extended source model for the
10 cm of local deformation at any site. We expect that 1992 Landers mainshock from near-source seismograms, Bull.
later models will fit the data even better than ours, and Seism. Soc. Am. (in press).
may be able to fit the data which our model cannot ex- Dokka, R. (1983). Displacements on late CeCenozoic strike-slip faults
of the central Mojave Desert, California, Geology 11,305-308.
plain. Our model actually fits the data very well--the
Dokka, R. and C. Travis (1990a). Late Cenozoic strike-slip faulting
rms misfit is 2.4 or, and the postfit rms residuals are in the Mojave Desert, California, Tectonics 9, 311-340.
1/2870 of the prefit residuals (prefit residuals assume Dokka, R. and C. Travis (1990b). Role of the Eastern California Shear
no slip on any faults). Compared with the null model of Zone in accommodating Pacific-North American plate motion,
no deformation, our model explains 99.96% of the vari- Geophys. Res. Lett. 17, 1323-1326.
Du, Y., A. Aydin, and P. Segall (1992). Comparison of various in-
ance in the data.
version techniques as applied to the determination of a geo-
physical deformation model for the 1983 Borah Peak earthquake,
Bull. Seism. Soc. Am. 82, 1840-1866.
Conclusions
Georgiadou, Y. and A. Kleusberg (1988). On the effect on iono-
spheric delay on geodetic relative GPS positioning, Manuscripta
Our co-seismic slip models show significant mo- Geodetica 13, 1-8.
ment release on at least three separate (and separate at Harris, R. and P. Segall (1987). Detection of a locked zone at depth
depth) faults during the Landers rupture. The significant on the Parkfield, California, segment of the San Andreas fault,
patches of moment release were on the Johnson Valley J. Geophys. Res. 92, 7945-7962.
fault, and the Homestead Valley, Emerson, and Camp Hauksson, E., L. M. Jones, K. Hutton, and D. Eberhart-Phillips (1993).
The 1992 Landers earthquake sequence: seismological obser-
Rock faults. Our model has significant moment release vations, J. Geophys. Res. 98, 19835-19858.
on both the Homestead Valley and Emerson faults, but Hill, D. P., P. A. Reasenberg, A. Michael, W. J. Arabaz, G. Beroza,
lacks the resolution to determine the details of the trans- D. Brumbaugh, D. N. Brune, R. Castro, S. Davis, D. dePolo,
fer of slip between these two subparallel and overlapping W. L. Ellsworth, J. Gomberg, S. Marmsen, L. House, S. M.
faults. The release of significant moment on multiple faults Jackson, M. J. S. Johnston, L. Jones, R. Keller, S. Malone, L.
Munguia, S. Nava, J. C. Pechmann, A. Sanford, R. W. Simp-
suggests that the upper crust in this region was very close son, R. B. Smith, M. Stark, M. Stickney, A. Vidal, S. Walter,
to the point of failure. By similar reasoning, the stop- V. Wug, and J. Zollweg (1993). Seismicity in the western United
page of the rupture in the middle of the Camp Rock fault States remotely triggered by the M 7.4 Landers, California,
was due to a low pre-earthquake stress in that region, earthquake of June 28, 1992, Science 260, 1617-1623.
which could also explain the lack of aftershocks on the Hough, S. E., J. Mori, E. Sembera, G; Glassmoyer, C. Mueller, and
S. Lydeen (1993). Southern surface rupture associated with the
northern Camp Rock fault. The proximity to failure and 1992 M 7.4 Landers earthquake: did it all happen during the
the demonstrated ability to rupture multiple faults in a mainshock? Geophys. Res. Lett. 20, 2615-2618.
single event suggest that earthquake hazards in areas with Hudnut, K. W., Y. Bock, M. Cline, P. Fang, Y. Feng, J. Frey-
multiple active, but short, faults need to be reevaluated. mueller, X. Ge, W. K. Gross, D. Jackson, M. Kim, N. E. King,
J. Langbein, S. C. Lax'sen, M. Lisowski, Z. K. Shen, and J.
Svarc, and J. Zhang (1993). Coseismie displacements of the 1992
Acknowledgments Landers earthquake sequence, Bull. Seism. Soc. Am. 84, no. 3,
625-645.
Thanks are due to all of the institutions which collected raw GPS Jones, L. E. and S. Hough (1993). Rupture characteristics of the June
data in the Landers area, and graciously allowed others to analyze it. 28, 1992 Big Bear event inferred from TERRAscope data, EOS
Also, we thank Duncan Agnew (Scripps Institution of Oceanography) 74, Spring Meeting Suppl:, 214.
for collecting postearthquake data at some of the GIMMS sites. Ye- Jones, L. E. and D. V. Helmberger (1993). Source parameters of the
huda Bock (Scripps) and the Scripps GPS data archive helped us in 1992 Big Bear earthquake sequence, EOS 74, Spring Meeting
obtaining the Caltrans HPGN data (and, equally important, the site Suppl., 215.
documentation). Bob Simpson (USGS) provided software for plotting Kanamori, H., H.-K. Thio, D. Dreger, E. Hauksson, arid T. Heaton
the slip distributions. We thank Duncan Agnew for a helpful review. (1992). Initial investigation of the Landers, California. Earth-
The Co-Seismic Slip Distribution o f the Landers Earthquake 659

quake of 28 June 1992 using TERRAscope, Geophys. Res. Lett. geodetic data and the testing of geophysical deformation models,
19, 2267-2270. J. Geophys. Res. 93, 14954-14966.
Matthews, M. and P. Segall (1993). Statistical inversion of crustal Sieh, K., L. Jones, E. Hauksson, K. Hudnut, D. Eberhart-Phillips,
deformation data and estimation of the depth distribution of slip T. Heaton, S. Hough, K. Hutton, H. Kanamori, A. Lilje, S.
in the 1906 earthquake, J. Geophys. Res. 98, 12153-12163. Lindvall, S. McGill, J. Mori, C. Rubin, J. Spotila, J. Stock,
Murray, M. H., J. C. Savage, M. Lisowski, and W. K. Gross (1993). H.-K. Thio, J. Treiman, B. Wernicke, and J. Zachariasen (1993).
Coseismic displacements: 1992 Landers, California, Earthquake, Near-field investigations of the Landers earthquake sequence, April
Geophys. Res. Lett. 20, 623-626. to July 1992, Science 260, 171-176.
Ponti, D. (1993). Quaternary chronostratigraphy and deformation his- Wald, D. and T. Heaton (1994). Spatial and temporal distribution of
tory, Los Angeles Basin, California in National Earthquake slip for the 1992 Landers, California earthquake, Bull. Seism.
Hazards Reduction Program, Summaries of Technical Reports Soc. Am. 84, no. 3, 668-691.
Volume XXXIV, U.S. Geol. Surv. Open-File Rept. 93-195, 591.
Savage, J. C., M. Lisowski, and M. Murray (1993). Deformation Department of Geophysics
from 1973 through 1991 in the epicentral area of the 1992 Lan- Stanford University
ders, California, earthquake, J. Geophys. Res. 98, 19951-19958. Stanford, California 94305
Savage, J. C. and W. H. Prescott (1973). Precision of Geodolite dis- (J.F., P.S.)
tance measurements for determining fault movements, J. Geo-
phys. Res. 78, 6001-6008.
Savage, J. C., M. Lisowski, and W. H. Prescott (1990). An apparent U.S. Geological Survey
shear zone trending north-northwest across the Mojave Desert Menlo Park, California 94025
into Owens Valley, eastern California, Geophys. Res. Lett. 17, (N.K.)
2113-2116.
Segall, P. and M. Matthews (1988). Displacement calculations from Manuscript received 29 July 1993.

You might also like