by J. M. Go´mez, R. Madariaga, A. Walpersdorf, and E. Chalard · 2008-03-27 · 739 Bulletin of the Seismological Society of America, 90, 3, pp. 739–751, June 2000 The 1996 Earthquakes

739

Bulletin of the Seismological Society of America, 90, 3, pp. 739–751, June 2000

The 1996 Earthquakes in Sulawesi, Indonesia

by J. M. Gomez, R. Madariaga, A. Walpersdorf, and E. Chalard

Abstract We study the rupture process of the 1 January (Mw � 7.9), 16 July (Mw

� 6.6), and 22 (Mw � 7) July 1996 earthquakes in Sulawesi, Indonesia. A teleseis-mic body waveform inversion of very broadband records shows that these eventsoccurred beneath the accretionary prism. The main shock was due to a well deter-mined fault with a strike of 53�N and a very shallow dip of 7�. From its source-timefunction duration (30 sec) and the aftershock distribution we estimate a rupture areaof 90 � 60 km2 and an average slip of 1.80 m. The surface displacement computedwith our best model fits well the displacement vector at the only available GPS stationat Tomini. The tsunami generated by the mainshock had an approximate sourceradius of 45 km which roughly agrees with the rupture size estimated above. Theearthquakes took place in a relay zone between the trench and the Palu-Koro trans-current fault. They ruptured a shallow dipping thrust fault which corresponds to thesubduction interface under the North Sulawesi arm. The slip vectors of all the eventshave a NNW orientation parallel to the direction of convergence between the NorthSulawesi arm and the Celebes Sea. Among the peculiarities of the rupture process,we found that the July 22 aftershock was the only event to have a well definedprecursor 1.8 sec before the main P-wave onset.

Introduction

The Minahasa, Indonesia earthquake of 1 January 1996is the largest event (Mw � 7.9) that has struck NorthwesternSulawesi in this century (Fig. 1). The epicenter is locatedapproximately 25 km from the Tonggolobibi village and 180km north of the Palu-Koro fault (see Table 1 and Figures 1and 2). The earthquake was followed by a tsunami (Peli-novsky et al., 1996) which affected a zone of about 100 kmof the island.

Sulawesi is located near the triple junction of the Eu-rasian, Australian, and Philippines Sea Plates. The NorthernSulawesi arm overrides the Celebes Sea plate at a relativelyslow convergence rate of 4 cm/yr. This rate has been de-duced from geological (Silver et al., 1983), paleomagnetic(Otofuji et al., 1981; Surmont et al., 1994), and GPS obser-vations (Walpersdorf et al., 1998a). The finite rotation of theNorth Sulawesi arm, welded to the Sula block since 15 Ma,is less than 25� (Walpersdorf et al., 1998a) during the last 5Ma. This implies that some 250 km of the Celebes oceaniccrust have been subducted southward.

Three large earthquakes of magnitudes between 6.6 and7.9 took place between January and July 1996 inside a regionthat had been surveyed in the framework of the programGEODYSSEA (Wilson et al., 1998). In association with thatprogram, a linear array of GPS sites has been repeatedly sur-veyed across the Palu-Koro strike-slip fault that cuts throughthe Eastern arm of Sulawesi (Walsperdorf et al., 1998c) nearthe site of the 1996 events. These events are the largest well-

recorded earthquakes in the region, so they are of particularinterest for understanding the seismicity and tectonics of theNorthern Sulawesi subduction zone.

In this article we use a variety of data sets in order todetermine the seismic rupture of the events of 1996. We usebathymetric and gravity data to verify the location and shapeof the trench. Then we model in detail the 1 January 1996mainshock (Mw � 7.9) and the aftershocks of 16 July (Mw

� 6.6) and 22 July 1996 (Mw � 7.0). Using teleseismicbody-waveform inversion we determine their centroiddepths and moment tensors. From the distribution of after-shocks relocated using a master event technique we con-strain the geometry of the rupture zone and outline the sub-ducting plate. We show that the earthquakes were due to ashallow-dipping thrust fault that roughly coincides with theinterface between Sulawesi and the Celebes Sea subductionzone.

Tectonic Setting

The Sulawesi Island lies near the triple junction betweenthe Eurasian, Australian, and Philippine Sea Plate (Fig. 1).The geodynamic evolution of this complex region showsthat intraplate shortening is still taking place along severalweak zones like the Sulu trench (in Northern Kalimantan),the North Sulawesi trench (NST), and the Eastern arm ofSulawesi (Silver et al., 1983; Rangin et al., 1990). Part of

740 J. M. Gomez, R. Madariaga, A. Walpersdorf, and E. Chalard

116˚E

116˚E

118˚E

118˚E

120˚E

120˚E

122˚E

122˚E

124˚E

124˚E

126˚E

126˚E

6˚S

4˚S

2˚S

0˚

2˚N

4˚N

01/01/96

07/16/96

07/22/96

Kal

iman

tan

Celebes sea

Palu-K

oro fault

North Sulawesi arm

East arm

Southeast armSouth arm

Tolo Gulf

Bone G

ulfS U

L A

W E

S I

Sula block

Nor

th

Mak

assa

r ba

sin

km

Matano fault

Celebes sea

Tolo Gulf

NST

Gorontalo basinTomini Gulf

0 100 200

km

-6.0

-5.5

-5.0

-4.5

-4.0

-3.5

-3.0

-2.5

-2.0

-1.5

-1.0

-0.5

0.0

Australian Craton

Indian Ocean Plate

Caroline Plate

PhilippineSea Plate

EurasianPlate South

ChinaSea

Kalim

anta

nSulu

CelebesSea

Figure 1. Tectonic setting of Sulawesi, Indonesia in the area where the three largeearthquakes of 1996 occurred. The inset shows the location of Sulawesi in the contextof plate tectonics of the area. The trench is well defined by the bathymetry. The three1996 events took place far from the trench close to the Palu-Koro fault. North SulawesiTrench (NST).

Table 1Reported Solutions for 1 January 1996 Sulawesi Earthquake

Source Lon. (�) Lat. (�) � d k M0 (1020 N m) h (km)

HRV 119.93 0.74 36� 6� 54� 7.78 15USGS 119.93 0.73 36� 12� 32� 5.10 14ERI 119.84 0.69 35� 14� 58� 1.55 49Ruff 119.90 0.70 32� 19� 68� 3.36 18

01/01/96 53� � 8� 7� � 3� 68� � 8� 3.53 � 0.3 16 � 216/07/96 51�* 5� � 5� 59� � 8� 0.10 � 0.02 18 � 222/07/96 51� � 5� 10� � 3� 56� � 8� 0.24 � 0.03 24.5 � 2

�, d, k, M0, and h are strike, dip, slip, the scalar-seismic moment, and depth, respectively. ERI is a japanese CMT rapid determination, while L. Ruffsolution is a Michigan University report for large earthquakes.

The 1996 Earthquakes in Sulawesi, Indonesia 741

118˚E

118˚E

120˚E

120˚E

122˚E

122˚E

2˚S

0˚

2˚N01/01/96

(b) mgal

TonggolobibiPangalasean Is.

Palu town

0 50 100

km

-100

-80

-60

-40

-20

0

20

40

60

80

100

01/01/96

(a)

Figure 2. Seismicity in North Sulawesi reported by the USGS during January 1996.Raw seismicity is shown in (a), events relocated by a Master event method are shownas black circles in (b). The aftershocks roughly define a zone parallel to the Northerncoast line of Sulawesi and are not parallel to the trench. The shading in the oceanindicate free air gravity that follow the trench near the subduction front. Focal mech-anisms correspond to the three largest earthquakes of 1996.

the deformation is accommodated by the Matano and Palu-Koro faults located at the southern and southwestern limitsof the Sula block, respectively (Silver et al., 1983; Rangin,1989). The NST absorbs a large part of the collision of theSula Block with the Eurasian Plate. The North Sulawesi armis a north-facing island arc that was once east facing but thathas rotated to form the north and southern arms of Sulawesi(Hamilton, 1979). Silver et al. (1983) considered evidencein favor of clockwise rotation of Sulawesi around a polelocated at the eastern end of the arm, which was later locatedby Walpersdorf et al. (1998b) at 126.2� E and 2.9� N. Sur-mont et al. (1994) using paleomagnetic data and Walpers-dorf et al. (1998b) from GPS measurements estimated thatthe rotation angle is about 20�–25�. Additional evidence infavor of rotation is the increasing width of the accretionarywedge westward along the Northern Sulawesi margin (Silveret al., 1983; Kopp et al., 1999). Rotation probably occurredas a consequence of NNW movement of Sulawesi along theleft-lateral Palu-Koro transcurrent fault (Silver et al., 1983;

Surmont et al., 1994), produced by a northwestern motionof the Sula platform relative to South Sulawesi. During therotation some 250 km of the Celebes Sea were subductedbeneath North Sulawesi during the last 5 Ma. Because noarc volcanism is associated with this southward subduction(Rangin and Silver, 1991), it is possible that the slab has notyet reached the depth where partial melting occurs (Surmontet al., 1994).

Bathymetry and Gravity

Figure 1 shows the bathymetry and topography aroundSulawesi based on ETOPO-5 (NGDC, 1988) data. The to-pography shows that the NST is parallel to the northern armof Sulawesi, in which water depths reach nearly 5500 m.The bathymetry defines a clear EW orientation for the trench.Hamilton (1979) proposed that the NST was slightly curvednear its intersection with the seaward extension of the Palu-


Figure 3. (a) Teleseismic hypocenters and mo-ment tensors from 1977 to 1996 as reported by USGSand Harvard. Earthquake focal mechanism solutionsare shown as equal area projection of the back hemi-sphere of the focal sphere. Solid and open quadrantsrepresent compressional and dilatational first motion,respectively. (b) Cross section AB. The trench originis located at point A. The beginning of the plate lookslike very sub-horizontal, and the dip steepens beyondthe hypocenters of the 1996 events.

Koro fault. Unfortunately the bathymetry is not accurateenough to resolve this kind of detail.

Figure 2 shows the free-air anomalies around the NorthSulawesi arm derived from satellite measurements by Sand-well and Smith (1997). As in most trenches, a gravity troughclosely mimics the bathymetry of the Southern Celebes seabasin. In Figure 2, the NST is associated with gravity anom-alies from �90 mgal to �15 mgal. There is no evidence ingravity anomalies nor in the bathymetry for any change oforientation of the NST near its intersection with the Palu-Koro fault (Figs. 1 and 2).

Seismicity

Indonesia has a seismic network concentrated aroundSumatra and Java, so local earthquake locations in Sulawesiare not very good; hypocentral depth control is particularlypoor because all the Indonesian stations are in a narrow-azimuthal range. For these reasons we could only use theaftershock activity reported by the USGS, even though theseobservations have high uncertainties, with most of the hy-pocenters having a quality C or D. We relocated these eventsusing the master event technique (Fitch, 1975). Many after-shocks could not be relocated because of their poor stationdistribution or the bad quality of the readings, so finally wecould only relocate 21 events. Figure 2 shows the horizontaldistribution of aftershocks in the first four weeks followingthe main shock. The aftershocks form an elongated cloudroughly oriented N55� E stretching for more than 90 kmalong the trench.

Figure 3 shows lower hemisphere projection of the shal-low depth focal mechanisms listed in the Centroid MomentTensor (CMT) catalog by Harvard (HRV) and USGS since1977. The epicentral determinations from USGS solutionsare shown in Figure 3a. Even though the aftershock distri-bution of Figure 2a is oblique with respect to the trench, thecompressional axis of most of the moment tensors is sub-horizontal and trends almost NS. The CMT seismicity is pre-dominantly strike-slip around the Palu-Koro fault andchanges to under-thrusting toward the east (Figure 3a). Fig-ure 3b shows a cross-section of the centroid depths fromHRV-CMT. These events are clearly not sufficient to deter-mine the geometry of the Celebes sea Wadati-Benioff zone,but it is in very good agreement with the subduction ge-ometry proposed by Cardwell et al. (1980) and Jarrard(1986). Cardwell et al. (1980) used data reported by severalseismological centers from 1959 up to 1977. They couldmake a number of seismic cross-sections across the NTS, butdue to the scarcity of well located events in the area theycould not determine the geometry of the Benioff zone inNorthwestern Sulawesi. By extrapolation, Cardwell et al.(1980) proposed a shallow dipping plane with a rapid bend-ing geometry. Figure 3b seems to indicate a low dippingplate interface with a rapid bending beyond 130 km land-ward from the trench, which confirms the geometry pro-posed by Cardwell et al. (1980).

The Tsunami

Three destructive tsunamis (1927, 1938, and 1968) oc-curred in North Sulawesi during this century (Tsuji et al.,1995; Pelinovsky et al., 1996). The 1 January 1996 earth-quake generated a tsunami that affected the coast line formore than 100 km. Unfortunately no tidal gauges were avail-able in the area, so the only available observations of thetsunami are those made by Pelinovsky et al. (1996) after theearthquake. According to eyewitness reports, the maximumobserved tsunami run-up were of the order of 1.6–3.4 mabove the mean sea level. After corrections for tidal com-ponents these values were reduced to 1.0–2.8 m with thehighest values located to the west of the epicenter near theTonggolobibi village (Figure 2).

On the Pangalasean Island (119.9� E, 0.40� N; Fig. 2),land subsidence was better determined from the vertical dis-placements of the coral wave break. The corrected values of


Figure 4. Point source modeling of P andSH waveforms of the 1 January 1996 Sulawesiearthquake. Solid and dashed lines are ob-served and synthetic displacement seismo-grams, respectively. The resulting source timefunction, shown at the center of the figure, hasa large and relatively simple pulse of momentrelease.

uplift were close to 30–50 cm. Using these observations andthe HRV-CMT solution, Pelinovsky et al. (1996) inferred anequivalent radius of the tsunami source of 45 km. This valueagrees with our relocated aftershock distribution observedduring January 1996 (Fig. 2).

Pelinovsky et al. (1996) assumed that the earthquakefault plane was the subvertical one. As discussed later in thisarticle, this choice disagrees with the low-dipping angle ofsubduction proposed by Cardwell et al. (1980), with thebody-waveform modeling presented later in this article, andwith the GPS observation at Tomini.

Body-Wave Analysis

We performed a body-wave inversion of digital P- andSH-wave seismograms recorded by the IRIS and GEOSCOPEdigital networks. The inversion was done by minimizing thesum of the squares of the residuals using an iterative pro-cedure based on a gradient method with a positivity con-

straint. We inverted for the best double-couple focal mech-anism (strike, dip, slip), the scalar moment, the source-timefunction (STF), and the centroid depth using the algorithmproposed by Nabelek (1985).

In order to avoid phases coming from the Earth’s coreand upper mantle, we used data recorded only between 30�and 92� for P waves, and between 34� and 68� for SH waves.In order to have the same reference instrumental response atall stations, individual instrument responses were removedfrom the seismograms and all were convolved with the in-strument response of the MAJO instrument in Japan. Themain-shock seismograms were integrated to displacementand bandpassed with a third-order butterworth filter between5 and 125 sec. The July shocks were processed similarly andbandpass-filtered between 2.5 and 111 sec. In order to givethe same weight to all the waveforms, an equalization cor-rection was applied to bring all seismograms to the samesource-station distance of 40�. We gave a lower relativeweight to SH waves (� 0.3). For lack of a local model, a


Figure 5. Variation of normalized rms residual for the hypocentral depth for January1 Sulawesi earthquake. Panel (a) shows residuals as a function of depth. The starsindicate the scalar seismic moment associated. The broad minimum is due to the verylow dipping angle of the fault plane. Panel (b) shows residuals as a function of azimuthfor a propagating source. The minimum residual (N56�E) indicates a weak directivityeffect, so that the source is probably distributed symmetrically about the centroid.

half-space with Vp � 6.40 km/sec, q � 2.86 gr/cm3, and aPoisson coefficient of 0.25 was used as a crustal model forthe source area. We consider an average attenuation factor(t*) of 1 sec for P waves and 4 sec for SH waves.

The 1 January 1996 Earthquake

The 1 January 1996 earthquake (Mw � 7.9) shows arelatively simple rupture process. The STF has a total du-ration of about 30 sec of roughly triangular shape with alarge initial moment release during 20 sec, followed by atail of about 10 sec. Thanks to a very good azimuthal stationcoverage, we got a very good point-source moment-tensioninversion (Fig. 4). Waveforms were very well fitted.

Figure 5a shows the relative residual of the waveformfit as a function of depth. The minimum is rather broad,indicating that several solutions have similar waveform fits.Outside the 16- to 24-km depth interval the misfit increasesrapidly, which indicates that these depths can no longer ex-plain observed characteristics of P and SH waveforms. Al-though the Benioff zone is poorly defined, we fixed the depthat 16 km below the sea floor because this fits well with therest of the seismicity shown in Figure 3. The correspondingnodal plane solution has a strike (�) of 53� � 8�, dip (d) of7� � 3�, and a rake (k) of 66� � 8� (see Table 1). Theorientation of the steeply north dipping fault plane is welldetermined by the data, whereas the strike of the south dip-ping fault plane may vary substantially.

The low-dip angle obtained from body-wave modelingis in good agreement with the low angle of the dipping plate(Cardwell et al., 1980; Lewis, 1991), suggesting that theevent took place along the thrust interface. This low anglecomplicates the determination of the fault strike but, fortu-nately, the very good azimuthal station coverage by P andSH waves strongly constrains the solution (Table 1, Fig. 4).

It is interesting to remark that the fault strike coincides withthe orientation of the relocated aftershock distribution shownin Figure 2.

Because the aftershocks are distributed mainly along thefault strike in the NE direction, we attempted an inversionfor a finite source. Looking for a directivity effect, we founda weak residual variation along the N56�E � 5� direction(Fig. 5b). We also tested different propagation directionsalong the azimuth of the fault plane, unilaterally and bilat-erally, along the dip, and along the rake. None of these prop-agation directions improved the waveform fit. Therefore, weconclude that a point-source approximation describes thisearthquake well. This apparent absence of directivity is con-sistent with a relatively small source region and could bedue to a highly symmetric rupture propagation. We areaware, of course, that the shallow dip of the fault does notfacilitate the determination of the preferential rupture prop-agation.

From the total STF duration (Figure 4), and assuminga typical rupture velocity (vs � 0.8b � 3 km/sec, we esti-mate a rupture length of 90 km, similar to that estimatedfrom the aftershock distribution and in very good agreementwith the diameter of the tsunami source area inferred byPelinovsky et al. (1996). The seismic moment obtained fromour inversion is M0 � 3.53 � 1020 N m, half of that obtainedby HRV (Table 1). This discrepancy is a typical problemin several shallow subduction earthquakes (Hagerty andSchwartz, 1996; Escobedo et al., 1998), and it is discussedfurther in the following sections.

The July 1996 Aftershocks

The Sulawesi region was particularly active during1996. The mainshock was followed by intense seismic ac-tivity for a period of several weeks, and the activity in-


Figure 6. P and SH waveform modeling ofthe 16 July 1996 Sulawesi earthquake. Solidand dashed lines are observed and synthetic ve-locity seismograms, respectively. The invertedsource time function shows a two step momentrelease.

creased again in late July; two late aftershocks occurredabout 50 km to the NE of the January earthquake on 16 July1996 (Mw � 6.6) and 22 July 1996 (Mw � 7.0). We carriedout a similar waveform inversion for the aftershocks as forthe January event, although in order to preserve waveformdetails we inverted velocity records instead of displacement.In this way we preserve the small precursor observed a fewseconds before the 22 July 1996 earthquake.

For the 16 July 1996 aftershock, the model convergedpoorly toward the fault-plane solution shown in Figure 6.We observe a well-defined subvertical nodal plane, whereasthe subhorizontal one is practically unconstrained. Becauseof its small dip angle the waveforms are insensitive to thestrike of the plane; for this reason we decided to reduce thespace of solution by introducing an additional assumption.Given the aftershock distribution and the fault-plane solutionof the January earthquake, we forced the fault strike to beas close as possible to that of the main shock, the nodalstations around the subvertical plane are well fitted when we

do this. Under this assumption, we obtained a solution thatconverged rapidly: � � 51�, d � 5� � 5� and k � 59� �8�, with a scalar moment of M0 � 0.1 � 1020 N m and acentroid depth of 18 � 2 km (see Table 1). The STF ischaracterized by an episodic history and a short duration ofabout 20 sec (Fig. 6); most of the moment release occurs inthe first 12 sec. Although the fit of P waves is good, theinitial onset is difficult to fit at most of the stations. Similarwaveform fitting problems were found for the SH waves,especially for stations located in the NE quadrant.

The 22 July 1996 earthquake had a duration of about26 sec, and its STF is more complex than that of the Januaryearthquake due to the presence of a precursory phase. Thecentroid depth was found to be 24.5 � 2 km, the fault ori-entation was � � 51� � 5�, d � 10� � 3� and k � 56�� 8� (Fig. 7), and the seismic moment was M0 � 0.24 �1020 N m (Table 1). The dip angle is slightly steeper thanthat of the January main shock, but the difference is wellwithin the range of uncertainty of the solution (Table 1).


Figure 7. P and SH waveform modeling ofthe the 22 July 1996 Sulawesi earthquake.Solid and dashed lines are observed and syn-thetic velocity seismograms, respectively. Theinverted source time function shows a three-step moment release with a clear initial pre-cursor.

Even if this event was deeper, it had a similar fault orien-tation as that of the two other events (Table 1). Because ofthe general similarity of the three events, we propose thatthey all took place on the interplate fault zone (Fig. 3).

Seismic Waveform Signature

We compared the displacement waveforms of the threeevents at several stations in order to determine whether thesimilarity of focal solutions is also reflected in the wave-forms generated during the rupture processes. For the com-parison, we removed the instrument response, and all theseismograms were integrated to displacement, bandpass fil-tered, and normalized. Figure 8 shows the P waveforms ob-tained for CTAO station (in Australia). They are similar forthe three earthquakes, though a small difference appears onthe 22 July signals, which are somewhat more complex dueto the presence of a clear precursory phase. The same pro-cedure with similar results was carried out at different sta-

tions, indicating a similar seismogenic signature for all theevents.

Modeling Surface Displacements

We compare our modeled displacements with those ob-tained from geodetic measurements by Walpersdorf et al.(1998b). They estimated the displacements produced by theearthquake eliminating the regional deformation producedby the rotation of the North Sulawesi arm. Among the bench-marks located in Northern Sulawesi, only that of Tomini(TOMI) recorded significant displacements. We model thesurface displacements using the best fault-plane solution forthe 1 January 1996 earthquake (Table 1). We assume thatseafloor deformation is caused by uniform slip on a rectan-gular fault of 90 km � 60 km embedded in an elastic half-space (Okada, 1985). Considering a centroid depth of 16 km,the upper and lower edges of the fault are 12.3 km and


0 20 40 60 80 100 120 140

01/01

07/16

07/22

CTAO/1996

Time (s)Time (s)

Figure 8. Comparison of the displacement of Pwaveforms of the three large 1996 Sulawesi earth-quakes as recorded at the station CTAO. The threehave a similar seismic signature.

20 km, respectively. Taking a rigidity of 3.6 � 1010 N/m2,we obtain a dislocation (D � M0/lS) of about 1.80 m (Fig-ure 9).

Our synthetic displacements (Fig. 9, Table 2) fit quitewell those observed by Walpersdorf et al. (1998b). Accord-ing to our model, the central Sulawesi stations (WATA,PALU, and TOBO) around the Palu-Koro fault, may have alsorecorded weak displacements produced by the earthquake(Fig. 9); unfortunately, the displacement of these stations ispredicted to be as weak as the uncertainties of measurementin the GPS data.

Finally, we computed the slip vector at TOMI assumingthat the earthquake took place on the subvertical plane asproposed by Pelinovsky et al. (1996). We found that for thisfault model the slip vector at TOMI would have been smallerand in the opposite direction to the regional slip vector overthe fault. Thus, a displacement field produced by subhori-

zontal plane, obtained by body waves, is in better agreementwith the regional slip direction.

Seismic Precursor

It is very important to study the rupture process of largeevents in as much detail as possible in order to properlycharacterize the rupture process. Among the three earth-quakes studied here only the 22 July event presented a com-plex history including a clear P-wave precursor. We do notknow whether this kind of phase is common or rare in theregion, but its observation is important because it may beassociated to the initiation of earthquake rupture (Ellsworthand Beroza, 1995; Abercrombie and Mori, 1994). We veri-fied that the precursor of the 22 July event was not an artifactdue to the noncausal finite-impulse-response (FIR) filtersused in modern seismic recording systems (Scherbaum andBouin, 1997).

In Figure 10 we show the stack of several very broad-band (VBB) seismograms of the 22 July 1996 event recordedat increasing distances along a common azimuth. The stackshould suppress multipathing and site effects, so we expectthat the stacked signal contains mostly source information.We estimated the seismic moment of the precursor as about1.3% of the main moment release. Thus, with a scalar mo-ment of 5.0 � 1017 N m, the precursor equivalent to an eventof magnitude of about M � 5.9. Using the master eventtechnique (Fitch, 1975), we relocated the origin of the pre-cursor at 1.8 sec prior the largest moment release of theearthquake 4.3 km seaward in the N26�W direction.

In Figure 11 we show deconvolved velocity waveformsfor the three events as recorded at two stations, ARU in thedilation quadrant and KMBO in the compressional one. Weobserve that the 1 January earthquake seems to have a smallphase preceding the P-wave onset that is observed at severalazimuths. For this event, the main moment release arrivesbefore the end of the precursor; its duration is somewhatshorter than 1.5 sec. The deconvolved velocity waveformsin Figure 11 are a typical example of the variability of pre-cursors of large earthquake observed using VBB data.

Results and Discussion

The study of the 1 January 1996 earthquake in the NWSulawesi region shows that this event had a relatively simplemoment release. Thanks to a very good azimuthal stationcoverage, the orientation of the steeply dipping nodal planeof the three earthquakes was well constrained. The strike ofthe almost horizontal shallow fault plane was more difficultto resolve, because the stations are concentrated near thecenter of the focal sphere. In spite of this problem, when thestrike of the plane is outside the range from 46� to 66�,the misfit between observed and synthetics increases rapidly,so that only a small range of solutions can simultaneouslyexplain the inverted body waveforms. The dip (around 7�)determined here is similar to that of many subducting plates

Table 2Displacements Determined by a Dislocation of 1.8 m due to

Sulawesi Earthquake of 1 January 1996

Station Lon. (�) Lat. (�) E (mm) N (mm) Z (mm)

TOMI* 121 0.45 �51.0 44.0 ?TOMI 121 0.45 �47.6 28.2 �10.1TOBO 120.09 �0.7 �6.5 32.8 �1.4PALU 119.91 �0.91 �2.1 22.9 �0.8WATA 119.57 �0.87 1.4 20.0 �1.3

*Value obtained by Walpersdorf (1997) from GPS measures.


-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

119.0 119.5 120.0 120.5 121.0 121.5 122.0

A

B

TOMI

TOBO

PALUWATA

(a)

TOBO

PALUWATA

TOMI

-0.2

0.0

0.2

0.4

0.6

0.0 0.5 1.0 1.5 2.0

Vertical Displacement

A B

(b)

� (o)

Amp (m)

1.723 m

91.4 mm91.4 mm

0 50 100

km

Figure 9. (a). Predicted surface displace-ments for the Sulawesi earthquake of 1 January1996. Fault geometry was determined frombody-wave modeling. The computed displace-ment field is that of a dip-slip fault with a smallstrike-slip component. Only the GPS observa-tions at TOMINI can be compared directly toour fault model (see table 2 and Figure inset).The agreement is very good. (b) The curve be-low the map shows the vertical displacementsalong the cross section. Black and whiteABdiamonds identify the Tonggolobibi island(119.98, 0.483) and the Pangalasean village(119.9, 0.489), respectively.

at trenches in the western Pacific (Watts and Talwani, 1974;McCaffrey and Nabelek, 1984; Becker and Lay, 1995).These earthquakes involve an active thrusting seawards ofthe North Sulawesi arm. We suggest that the south-dippingnodal plane is the active one because this angle agrees wellwith the dip estimated for the subduction of the Celebes Sea(Hamilton, 1979; Silver et al., 1983; Kopp et al., 1999).

As shown in the cartoon of Figure 12, the orientation ofthe fault plane is oblique to the North Sulawesi trench. Theearthquake took place here near the intersection of the sub-duction zone with the Palu-Koro fault. The oblique distribu-tion of seismicity seems to be just a peculiarity of the after-shock distribution, the slip vector of most events being invery good agreement with the direction of subduction. Mostof the fault plane solutions combine a shallow underthrustwith 15–20% of left-lateral motion at the subduction inter-face (Table 1). We hypothesize that this strike-slip compo-nent may be due to an interaction between subduction alongthe NST and strike-slip faulting along the Palu-Koro fault.

Even though the waveforms of the main shock werewell matched, the difference between our scalar moment andthat obtained by Harvard is large (Table 1). From inversion,a large seismic moment is not compatible with body wavesfor a source at 16-km depth and 7� of fault dip. We mustnote, however, that for shallow thrust events like these, thereis a large trade-off between fault depth and moment estimate(Fig. 5). We do not know why there is such a large differencebetween moment estimated from body waves with energycentered around 20 sec and the moment determined at lowerfrequencies by Harvard from long-period body waves (T �45 sec) and surface waves (T � 135 sec). We suspect thatthe difference in seismic moment may reflect a lower fre-quency component of the source that we can not resolvefrom body waves. In any case, we can not settle this questionwithout a simultaneous inversion of surface waves, as it isdone by Harvard CMT solution (Ekstrom, 1989); see Hagertyand Schwartz (1996) and Escobedo et al. (1998) for a relateddiscussion.


Figure 10. Waveform stacking along an Asian path for the 22 July 1996 earthquake.The precursor is observed at all stations on the path. The stack at the bottom shows itvery clearly.

Figure 11. Comparison of broadband ve-locity waveforms observed for the three 1996earthquakes at two stations, ARU in the dila-tational quadrant and KMBO in the compres-sional one. The precursor is clear for the 22July earthquake (up arrow), it is absent on 11July and the main shock on 1 January presentsa small phase which could be a precursor too.Down arrows indicate the P-wave arrival.


slip slip

North Sulawesi Trench

S U L A W E S I

Palu fault

slip

Convergence

Sula Block

Convergence

Celebes Sea

Figure 12. A schematic model proposed to de-scribe the geometry rotation at the northwestern partof the North Sulawesi arm. The fault-plane geometryproposed in this study would result from a complexmechanical interactions between the NST and thePalu-Koro fault. The slip direction is well coincidingwith the regional one and the movement of the SulaBlock.

The 1 January 1996 earthquake is a typical example ofan important trade-off between STF and seismic momentthat we usually attribute to waveform compensations at lowfrequencies due to opposite signs of P, pP and sP phases atmost stations. This is probably not the case for the mainshock because when we analyzed the surface waves (notshown in this article) for this event, we obtained a momentM0 � 6.3 � 1020 N m, but we also found a source timeduration about twice as long as that inferred from bodywaves. Thus there seem to be a genuine difference in lowfrequency and higher frequency moments. We must caution,however, that, as discussed by Kanamori and Given (1981),earthquakes of shallow depths and shallow dips as that of1 January 1996 excite surface waves very inefficiently, caus-ing some moment-tensor elements (Mxz and Myz) to be dif-ficult to resolve.

Determination of source mechanism was more difficultfor the July earthquakes. The strike of the fault of the 16July 1996 event was poorly determined: the range of pos-sible solutions is large due to a lower signal-to-noise ratio.For the three earthquakes the dip angle of the fault variesbetween 5� and 10�. The rake is consistent among all threeearthquakes. Although the centroid of the 22 July 1996earthquake was deeper, its seismogenic characteristics aresimilar to those of the other earthquakes.

As we mentioned earlier, Pelinovsky et al. (1996) pro-posed that this earthquake took place on the almost vertical-fault plane. We rule out this possibility in favor of shallow-dipping fault for at least two main reason. First, the largedisplacement vector at TOMI can not be matched by slip ona semivertical-fault plane. Second, a magnitude 7.9 earth-quake on a subvertical fault plane would have penetratedwell below the brittle-ductile transition zone, having a cen-troid beyond the 30-km depth. Body-waveform modeling isincompatible with source depths greater than 24 km (seeTable 2).

Even though the three 1996 earthquakes look similar,the 22 July event had a clearly identifiable precursor and amore complex STF, suggesting that local asperities on thefault plane played a more important role for this earthquake.This precursor of about 1.8 sec is a good example that thiskind of nucleation phase is not a universal feature of largeearthquakes. Unfortunately, our data does not allow us todetermine whether this event belongs to a preslip or to acascade model as proposed by Ellsworth and Beroza (1995),neither are we able to infer whether this phase is a particularfeature in the local rupture process. The main difficulty en-countered in this work was the lack of a good local seismicdata. A more accurate relocation of aftershocks would haveprovided a better definition of the three-dimensional geom-etry of the fault zone.

Conclusions

We analyzed the rupture process of three earthquakesthat occurred in the Northwestern part of the Sulawesi Islandduring 1996. We used body-waveform inversion, tsunamireports, aftershock distribution, and surface displacementsin order to define the spatial and temporal evolution of therupture associated with these subduction events.

The 1 January 1996 earthquake had a simple seismic-moment release. A point-source approximation is enough tofit the observed teleseismic body-wave signals, whose weakNE directivity coincides with the orientation of the after-shock distribution.

The strong similarity of the waveforms of the threeearthquakes and their similar seismogenic characteristics in-dicate these shocks ruptured along the same interface. Thisseismicity is located far from the trench and it has a trendthat is oblique with respect to the direction of the trench.The rupture geometry and the estimated surface displace-ments are in good agreement with the single available geo-detic measurement. The slip vector obtained is orientedNNW and corresponds well with the direction of subductionof the South Celebes sea plate under Northern Sulawesi. Theposition of the earthquake hypocenters between the NST andthe island arc and the fault-plane solutions suggest that the1996 earthquakes are analogous to thrust events in other sub-duction zone and represent slip between the subducting andoverriding plates. Thus, the Celebes sea subducts with a very


shallow dip angle of 7� beneath the accretionary prism ofthe North Sulawesi arm.

Finally, a clear precursor was detected only for the 22July 1996 event, whereas the main shock of 1 January hada very small precursor practically hidden by the triggeringof the main energy release. These observations demonstratethat in spite of the similar characteristics of the rupture pro-cess for the three events, precursors are highly variable fromevent to event.

Acknowledgments

This research was supported by the Institut Universitaire de France.J. M. Gomez acknowledges a Ph.D. fellowship in France from CONACyT,Mexico. We thank R. Gaulon and M. Pubellier, whose comments and sug-gestions have contributed greatly to this work. We also acknowledge fruit-ful discussions with P. Bernard, A-G. Bader, and J. Ramos. J-C Rueggkindly furnished programs for the modeling of deformation. Thanks toE. Pelinovsky and C. Kopp for providing preprints.

References

Abercrombie, R., and J. Mori (1994). Local Observations of the Onset ofa Large Earthquake: 28 June 1992 Landers, California, Bull. Seism.Soc. Am. 84, 725–734.

Becker, L., and T. Lay (1995). Very broadband analysis of the 1992 Flores,Indonesia, earthquake (Mw � 7.9), J. Geophys. Res. 100, 18,179–18,193.

Cardwell, R. K., B. L. Isack, and D. E. Karig (1980). The spatial distributionof earthquakes, focal mechanism solutions, and subducted lithospherein the Philippine and Northeastern Indonesian Islands, in The Tectonicand Geologic Evolution of Southeast Asian Seas and Islands, D. E.Hayes (Editor), American Geophysical Monograph 23, 1–35.

Ekstrom, G. (1989). A very broad band inversion method for the recoveryof earthquake source parameters, Tectonophysics 166, 73–100.

Ellsworth, W. L., and G. C. Beroza (1995). Seismic evidence for a seismicnucleation phase, Science 268, 851–855.

Escobedo, D., J. Pacheco, and G. Suarez (1998). Teleseismic body-waveanalysis of the 9 October, 1995 (Mw � 8.0), Colima-Jalisco, Mexicoearthquake, and its largest foreshock and aftershocks, Geophys. Res.Lett. 25, 547–550.

Fitch, T. (1975). Compressional velocity in source regions of deep earth-quakes: an application of the master earthquakes technique, EarthPlanet. Sci. Lett. 26, 156–166.

Hagerty, M. T. and S. Y. Schwartz (1996). The 1992 Cape Mendocinoearthquake: broad-band determination of source parameters, J. Geo-phys. Res. 101, 16,043–16,058.

Hamilton, W. (1979). Tectonics of the Indonesian region, U.S. Geol. Surv.Profess. Pap. 1078, 340 pp.

Jarrard, R. D. (1986). Relations among subduction parameters, Rev. Geo-phys. 24, 217–284.

Kanamori, H., and J. W. Given (1981). Use of long-period surface wavesfor rapid determination of earthquake-source parameters, Phys. EarthPlanet. Inter. 27, 8–31.

Kopp, C. E. R. Flueh, and S. Neben (1999). Rupture and accretion of theCelebes Sea crust related to the North-Sulawesi subduction: combinedinterpretation of reflection and refraction seismic measurements, J.Geodynamics 27, 309–325.

Lewis, S. D. (1991). Geophysical setting of the Sulu and Celebes Seas, inProc of the Ocean Drilling Program, Sci. Results 124, 65–73.

McCaffrey, R. and J. Nabelek (1984). The geometry of back arc thrustingalong the eastern sunda arc, Indonesia: constraints from earthquakeand gravity data, J. Geophys. Res. 89, 6171–6179.

Nabelek, J. (1985). Geometry and mechanism of faulting of the 1980 El

Asnam, Algeria, earthquake from inversion of teleseismic body wavesand comparison with field observations, J. Geophys. Res. 90, 12,713–12,728.

National Geophysical Data Center, ETOPO-5 (1988). Bathymetry/topog-raphy data, Natl. Oceanic and Atmos. Admin. Data Announc. 88-MGG-02, U.S. Dep. Commer, Boulder, Colo.

Okada, Y. (1985). Surface deformation to shear and tensile faults in a halfspace, Bull. Seism. Soc. Am. 75, 1135–1154.

Otofuji, Y., S. Sasajima, S. Nishimura, A. Dharma, and F. Hehuwat (1981).Paleomagnetic evidence for a clockwise rotation of the northern armof Sulawesi, Indonesia, Earth Planet. Sci. Lett. 54, 272–280.

Pelinovsky, E., D. Yuliadi, G. Pratseya, and R. Hidayat (1996). The 1996Sulawesi tsunamis, Russian Academy of Sciences IAP, 392, 1–36.

Rangin, C. (1989). The Sulu arc, a back-arc basin setting with a Neogenecollision zone, Tectonophysics 161, 119–141.

Rangin, C., and E. A. Silver (1991). Development of Celebes Sea Basin inthe context of the Western Pacific marginal basin history, in Pro-ceedings of the Ocean Drilling Program Sci. Results 124, 311–321.

Rangin, C., L. Jolivet, M. Pubellier, and the Tethys Pacific working group(1990). A simple model for the tectonic evolution of southeast Asiaand Indonesia region for the past 43 m.y., Bull. Soc. Geol. Fr. 6, 889–905.

Sandwell, D. T. and W. H. F. Smith (1997). Marine gravity anomaly fromGeosat and ERS1 satellite altimetry, J. Geophys. Res. 102, 10,039–10,054.

Scherbaum, F. and M. P. Bouin (1997). FIR filter effects and nucleationphases, Geophys. J. Int. 101, 8437–8455.

Silver, E. A., R. McCaffrey, and R. B. Smith (1983). Collision, Rotation,and the Initiation of Subduction in the Evolution of Sulawesi, Indo-nesia, J. Geophys. Res. 88, 9407–9418.

Surmont, J., C. Laj, C. Kissel, C. Rangin, H. Bellon, and B. Priadi (1994).New paleomagnetic constraints on the Cenozoic tectonic evolution ofthe North Arm of Sulawesi, Indonesia, Earth Planet. Sci. Lett. 121,629–638.

Tsuji, Y., F. Imamura, H. Matsumoto, C. E. Synolakis, P. T. Nanang,J. Jumadi, S. Harada, S. S. Han K. Arai, and B. Cook (1995). Fieldsurvey of the east Java earthquake and tsunami of June 3, 1994, PureAppl. Geophys. 144, 839–854.

Walpersdorf, A., C. Rangin, and C. Vigny (1998a). GPS compared to long-term geologic motion in the north arm of Sulawesi, Earth Planet. Sci.Lett. 159, 47–55.

Walpersdorf, A., C. Vigny, P. Manurung, C. Subarya, and S. Sutisna(1998b). Determining the Sula block kinematics in the triple junctionarea in Indonesia by GPS, Geophys. J. Int. 135, 351–361.

Walpersdorf, A., C. Vigny, C. Subarya, and P. Manurung (1998c). Moni-toring of the Palu-Koro fault (Sulawesi) by GPS, Geophys. Res. Lett.25, 2313–2316.

Watts, A. B., and M. Talwani (1974). Gravity anomalies seaward of deep-sea trenches and their tectonic implications, Geophys. J. R. Astr. Soc.36, 57–90.

Wilson, P., J. Rais, Ch. Reigber, E. Reinhart, B. A. C. Ambrosius, X. LePichon, M. Kasser, P. Suharto, D. Abdul Majid, D. Paduka Awang,R. Almeda, and C. Boonphakdee (1998). Study provides data on ac-tive plate tectonics in Southeast Asia region, EOS Trans. AGU 79,545–549.

Unidad de Investigacion en Ciencias de la TierraCampus Juriquilla-UNAMQueretaro 76230, Qro. MexicoApdo. Postal 1-742, C.P. [email protected]

Ecole Normale Superieure, Lab. de GeologieU.R.A. 1316, 24, Rue Lhomond,75231 Paris Cedex 05, France

Ecole des Mines de Paris, Centre de Recherche en Geophysique35, rue Saint HonoreF-77305 Fontainebleau Cedex, France

Manuscript received 16 April 1999.

Documents

by J. M. Go´mez, R. Madariaga, A. Walpersdorf, and E. Chalard · 2008-03-27 · 739 Bulletin of the Seismological Society of America, 90, 3, pp. 739–751, June 2000 The 1996 Earthquakes