Rapid characterisation of large earthquakes by multiple seismic broadband arrays

Nat. Hazards Earth Syst. Sci., 10, 923932, 2010www.nat-hazards-earth-syst-sci.net/10/923/2010/ Author(s) 2010. This work is distributed underthe Creative Commons Attribution 3.0 License.

Natural Hazardsand Earth

System Sciences

Rapid characterisation of large earthquakes by multiple seismicbroadband arrays

D. Roessler, F. Krueger, M. Ohrnberger, and L. Ehlert

Department of Earth and Environmental Studies, University of Potsdam, K.-Liebknecht-Str. 24/H27, 14476 Potsdam,Germany

Received: 1 December 2009 Revised: 25 March 2010 Accepted: 26 March 2010 Published: 27 April 2010

Abstract. An automatic procedure is presented to retrieverupture parameters for large earthquakes along the Sunda arcsubduction zone. The method is based on standard arrayanalysis and broadband seismograms registered within 30100 epicentral distance. No assumptions on source mech-anism are required. By means of semblance the coherencyof P waveforms is analysed at separate large-aperture arrays.Waveforms are migrated to a 1010 wide source region tostudy the spatio-temporal evolution of earthquakes at each ar-ray. The multiplication of the semblance source maps result-ing at each array increases resolution. Start, duration, extent,direction, and propagation velocity are obtained and pub-lished within 25 min after the onset of the event. First prelim-inary results can be obtained even within 16 min. Their rapiddetermination may improve the mitigation of the earthquakeand tsunami hazard. Real-time application will provide rup-ture parameters to the GITEWS project (German IndonesianTsunami Early Warning System). The method is applied tothe twoM8.0 Sumatra earthquakes on 12 September 2007, totheM7.4 Java earthquake on 2 September 2009, and to ma-jor subduction earthquakes that have occurred along Sumatraand Java since 2000. Obtained rupture parameters are mostrobust for the largest earthquakes with magnitudesM 8.The results indicate that almost the entire seismogenic part ofthe subduction zone off the coast of Sumatra has been rup-tured. Only the great Sumatra event in 2004 and theM7.7Java event on 17 July 2006 could reach to or close to the sur-face at the trench. Otherwise, the rupturing was apparentlyconfined to depths below 25 km. Major seismic gaps seem to

Correspondence to:D. Roessler([email protected])

remain off the coast of Padang and the southern tip of Suma-tra.

1 Introduction

Earthquakes that are sufficiently small compared to the con-sidered wavelengths of the radiated wavefield are describedas point sources. Larger earthquakes with finite extent canbe described as the superposition of point sources that rup-ture at different stages of the event. Kinematic source pa-rameters describe the orientation, extent, duration, and rup-ture propagation of such earthquakes. Their knowledge isimportant to assess and to mitigate societal risk in the after-math of destructive events. They are often modelled in detailby inversion of seismic waveforms, aftershock distributionand geodetic observations. Results are, however, often highlynonunique or require considerable amounts of computationaleffort or time to record aftershocks (hours to days).

Seismic arrays are used to detect and locate earthquakesand seismic structures by increasing the signal-to-noise ratioas compared to single seismic stations (seeRost and Thomas,2009, for an overview). Since the megathrust tsunamigenicM9.1 Sumatra earthquake in 2004 (Kruger and Ohrnberger,2005; Kruger and Ohrnberger, 2005; Ishii et al., 2005) arraymethods have also found application to determine kinematicparameters of large earthquakes such as rupture extent, direc-tion, duration, and velocity (Kruger et al., 2006; Ishii et al.,2007; Roler et al., 2008; Zhang et al., 2008; Roler et al.,2009; Xu et al., 2009). Here we propose a fast and robustmethod to retrieve such rupture parameters within a few min-utes after the event using multiple seismic broadband arrays.

Published by Copernicus Publications on behalf of the European Geosciences Union.

http://creativecommons.org/licenses/by/3.0/

924 D. Roessler et al.: Rapid characterisation of large earthquakes

The method is operational and routinely applied in near-real-time. It was initially developed for the German-IndonesianTsunami Early Warning System (GITEWSRudloff et al.,2009). Therefore, we put emphasis on the robustness as wellas on computational speed. We demonstrate results for recentlarge earthquakes along the Sunda arc which are obtainedfrom an ongoing automatic real-time application and previ-ous manual data processing.

1.1 Theory

Classical array beamforming assumes coherent signals butuncorrelated noise, plane wave propagation, and equal sta-tion timing. Waveformsx(t) may be composed of coherentsignalss(t) and uncorrelated noisen(t), x(t)= s(t)+n(t).The beamb(t) is formed by stacking seismograms at timetfor a specific slowness vectoru which is a property of theconsidered signal

b(t)= s(t)+1

N

Ni=1

ni(t+r iu), (1)

with the coordinate vectorr i of station with indexi. In (1) tis the time at a reference station or a reference point thatr iis referred to. For uncorrelated noise beamforming increasesthe signal-to-noise ratio by a factor of

N .

The beampowerE(t) expresses the energy of the beamwithin a time window centred aroundt

E(t)=1

K+1

K/2k=K/2

1NN

i=1

xi(t+k1t+r iu)

2

, (2)

wherek is the sample index and1t is the increment of thetime within the time window. Ifu is unknown it can be de-termined by a grid search maximisingE(t).

The plane-wave assumption will be violated with increas-ing dimension of the array. As a consequence, slowness,u,is not equal for all stations. However, large arrays are desiredbecause the spatial resolution of an array increases with aper-ture (seeRost and Thomas, 2009, for a summary). Thereforewe account for curved wavefronts in the global Earth. Thatis, the term(t+ r iu) in (1) and (2) is replaced by(t0+ ti),wheret0 is the time at any given hypothetical hypocentre andti is the traveltime of the considered phase from the hypocen-tre to stationi. Travel timesti are computed using a standardspherical Earth model such as ak135 (Kennett et al., 1995)and corrections for ellipticity. Replacing the argument beampower (2) takes the form

E(t0)=1

K+1

K/2k=K/2

1NN

i=1

xi(t0+k1t+ ti)

2

. (3)

BeampowerE(t0) is an absolute measure of the ampli-tudes at an array. It depends on hypocentral distance, az-imuth as well as on magnitude and mechanism of an earth-quake. For large earthquakes the radiated seismic energy

may even vary significantly during different parts of the rup-ture process (e.g. start and stop of the rupture). Therefore, weprefer to use semblance (Neiddell and Turhan Taner, 1971)as a measure of coherency of the wavefield for rupture track-ing. Semblance,S, is simply beampower (3) normalised bythe energy summed over all considered seismograms

S(t0)=1

N

K/2k=K/2

Ni=1xi(t0+k1t+ ti)2K/2k=K/2

Ni=1x

2i (t0+k1t+ ti)

. (4)

Due to the normalisation, 0 S(t) 1 and is independentof individual signal amplitudes. For uncorrelated noiseS 1/N but significantly higher for coherent signalsS (Douzeand Laster, 1979).

By Eqs. (3) and (4) seismograms are migrated to any hy-pothetic source position and time. The semblance is non-centrally F-distributed (Douze and Laster, 1979). Therefore,semblance relates to probability of seismic energy being ra-diated from a given source location and att0. Semblance re-lates to probability viaF -statistics (Douze and Laster, 1979).It therefore provides a probability that, seen at the given ar-ray, seismic energy was radiated at the given source point andtime. If the considered phase was generated at the assumedlocation and time, semblance will be high andS > 1/N , butlow otherwise.

The semblance calculation is easily extended from a sin-gle source point and time to a large area or volume and manytime steps giving time-dependent semblance source maps.For such an area or volume the traveltimesti are calculatedfrom any given source point to all stations. Semblances(t0)is computed for all points. By shiftingt0 a time series of sem-blance source maps is obtained. If an earthquake occurredwithin the assumed area or volume this directly allows toimage the spatio-temporal evolution and to track the ruptureduring an earthquake. In principle this method can be ap-plied continuously to the whole Earth. Computational effortdecreases by constraining region and time. A rough hypocen-tre estimate must be provided to define a source region. Forearthquakes in many regions hypocentres are available withina few minutes (< 10 min) after an event initiated (GEOFON,http://geofon.gfz-potsdam.deHanka et al., 2000).

The spatial resolution of a seismic array increases withnumber of stations, array aperture and the frequency band-width of the signal (seeRost and Thomas, 2009, for asummary). However, as array aperture increases, the fre-quency, above which waveforms loose coherency, decreases.This can be attributed to increasing structural inhomogeneity,the effect of radiation pattern and source directivity whichchanges the waveforms (Aki and Richards, 2002). As a re-sult image resolution of arrays that are too large (e.g. the fullglobal network) may be reduced compared to smaller arrays(Xu et al., 2009).

The method is applied to large and very large earthquakeswhich typically show dominant frequencies of the source

Nat. Hazards Earth Syst. Sci., 10, 923932, 2010 www.nat-hazards-earth-syst-sci.net/10/923/2010/

http://geofon.gfz-potsdam.de

D. Roessler et al.: Rapid characterisation of large earthquakes 925

spectra below 0.1 Hz forP waves. We found by manual in-spection of theP waveforms of such events that arrays aslarge as 4000 km aperture can be used.

For a single array the distance/azimuth resolution dependsmainly on aperture but depending on the actual number ofstations and the interstation distances aliasing may be a prob-lem for a specific array configuration. The time resolution ofa single array is directly linked with its slowness resolutionand typically is smeared along the great circle arc connectingthe source and the receiver array.

The polarity of aP wave is determined by the earthquakesource mechanism and the take-off direction of the wave. Po-larity reversals destroy coherency of waveforms and reducesemblance if waveforms are stacked. They occur near thenodal planes of the radiatedP waves. A single regional arrayby chance may lay at or near the nodal plane of the specificmechanism.

To overcome the above mentioned problems of single ar-rays we use a distributed set of large aperture regional ar-rays surrounding the hypothetical source region (compareFig. 1). Each array will then give an individual semblancemap. Semblance is a one sided positive function of wave-form coherency independent of the overall polarity of theincoming waveforms. We can therefore avoid the polarityproblem due to the in general unknown source mechanismby summing or multiplying the semblance source maps ofthe different arrays for a specific source time. Because sem-blance can be related to likelihood we feel that multiplicationof the semblance values of different arrays is more appropri-ate than summation. The multiplication also corresponds toa multiplication of the transfer functions of the arrays. Ide-ally, their central lobe should be at the same position but theside lobes are at different positions. The multiplication leadstherefore to a more distinct central lobe (that refers to the ac-tual source signal) and to the suppression of side lobes. Dueto heterogeneity of the Earth, the single array localisationsmay deviate slightly from actual hypocentres. By multiplica-tion this error is reduced.

Taken all facts together the combined semblance

S(t0)=

Mi=1

Si(t0) (5)

has increased resolution in space and time and more stabilitythan the single array estimates (M is the number of arrays).The resulting combined semblance maps are related to a jointprobability that an earthquakes occurred in the area at thegivent0.

1.2 Station selection

For rapid rupture tracking of large earthquakes, semblanceanalysis is applied assuming the first-arriving directP wavesoriginate from distances between 30 and 100. Withinthis distance rangeP waves are well separated from strong

Fig. 1. Map of stations used for semblance analyis of theM7.4 Javaearthquake on 2 September 2009 (centre of the map). Stations arelocated within epicentral distances of 30100. Colours indicatethe different arrays used for data processing.

later body wave phases and surface waves. At smaller dis-tances the wavefield is more complicated by triplications inP due to upper-mantle discontinuities. At greater distancesP turns intoPdiff which is connected with waveforms distor-tion and rapid amplitude decay. The restriction to teleseismicP waves determines the minimum time needed for computa-tion. UsingSH waves which arrive much later thanP wouldsignificantly increase waiting time. Therefore,SH waves arenot considered for automatic near-real-time application.

We use seismograms of high quality seismic broadbandstations. All stations were manually selected and checkedupon their broadband character and data quality in theprepraratory phase of the automatic online processing. Theyare merged to six virtual large-aperture regional arrays whichare combined during processing (Fig.1).

1.3 Data processing and implementation

Broadband data are acquired continuously using SeedLink,saved to disk, and processed automatically. Arrays are pre-defined to optimise azimuthal coverage for earthquakes alongthe Sunda arc. We associate stations to six separate arrays inregions of Europe, around the eastern mediterenean sea, eastAsia, Alaska and eastern Russia, Africa, and Australia andNew Zealand (compare Fig.1).

We continuously receive earthquake alerts including mag-nitude and preliminary hypocentre from GEOFON via emailand the SeisComp system. Semblance analysis is routinelystarted for all events with magnitudeM 5.5. Based on the

www.nat-hazards-earth-syst-sci.net/10/923/2010/ Nat. Hazards Earth Syst. Sci., 10, 923932, 2010


preliminary hypocentres, stations are selected within epicen-tral distances of 30 to 100. A 1010 wide area aroundthe epicentre at the hypocentre depth defines the target area.Source depth is kept constant at hypocentre depth due to lim-ited depth resolution (teleseismicP waves leave the sourceregion with steep incidence angles). Data acquisition starts at19 min and 46 s (1186 s) after source time to account for max-imum travel times of theP waves (826 s), up to 300 s rup-ture duration, and 60 s data latency. The data latency is thetime needed to transmit seismogram data from the recordingstation to the processing computer. It varies between secondsand sometimes minutes.

We only consider vertical component broadband seismo-grams where theP waves are most prominent. Incomingraw seismograms have different sampling frequencies in therange of 20100 Hz which need to be homogenised. For ef-ficient semblance analysis data are downsampled to 10 Hzand bandpass filtered between 0.01 and 4 Hz. Starting 200 sbefore actual source time, semblance is calculated for the se-lected area and for 100 time steps (t0) in intervals of 5 s. Forthe separate arrays all data are processed and the semblancesource maps are computed on separate client nodes on a high-performance computer cluster.

A master node collects the semblance source maps of theindividual arrays. They are combined by pointwise multipli-cation (see Fig.2 andOhrnberger and Kruger, 2005; Roleret al., 2008) as mentioned above.

1.4 Parameter extraction

The maxima of the combined semblance source maps areconsidered to image the most likely position of the majorseismic energy release at a given timet0. They are used toretrieve rupture parameters such as start, duration, length, di-rection, and velocity. This is especially useful in the caseof unilateral rupturing but may be non-unique for bilateralruptures after the semblance computations are finished.

Start and duration of the rupture are determined usinga conventional sta/lta trigger applied to the time seriesof the maximum semblance value over all grid points.Sta and lta define the moving short-term and long-termsemblance averages within 15 s and 100 s, respectively.Rupture length is calculated from the greatest distancebetween two maxima within the time of rupturing. Dura-tion and rupture length give an average rupture velocity.The direction is approximated from mean azimuth of thesemblance maxima with respect to epicentre. The hori-zontal rupture area is also estimated. However, accuracystrongly depends on the actual data coverage, rupturepropagation, and source mechanism. Data processing wasdeveloped and tested manually. Since 2008 the algorithmhas been implemented for automatic real-time applica-tion. A summary image of the results is published in theinternet (http://www.geo.uni-potsdam.de/arbeitsgruppen/

8S

6Sa b

8S

6Sc d

8S

6Se f

106E 108E 110E

8S

6Scombined

Fig. 2. Combination for semblance source maps by multiplicationat the onset of theM7.4 Java earthquake on 2 September 2009. As-suming source time, semblance source maps are computed at arraysin (a) Europe,(b) Africa, (c) East Asia,(d) eastern MediterraneanSea(e) Alaska, and(f) Australia, New Zealand (compare Fig.1).Colours indicate normalised semblance.

geophysikseismologie/forschung/ruptrack/index.php) forthe largest earthquakes.

Although semblance computations are relatively time ef-ficient, computational effort increases with the number ofavailable seismograms. Typically, the computation of sem-blance source maps and rupture parameters takes about5 min. If earthquake alerts are received before theP wavesarrive at 100 epicentral distance then the results are pub-lished within about 25 min after the origin time of the earth-quake.

2 Examples

Although rupture tracking is routinely performed for earth-quakes worldwide we have optimised the performance tosubduction-zone earthquakes along the Sunda arc near Javaand Sumatra. Here we concentrate on the largest, mostrecent earthquakes that occurred off the coast of Sumatra,the two adjacentM8.0 Bengkulu events on September 12,2007, 11:10 and 23:49 (Figs.3, 4) and on theM7.4 Java


http://www.geo.uni-potsdam.de/arbeitsgruppen/geophysik_seismologie/forschung/ruptrack/index.phphttp://www.geo.uni-potsdam.de/arbeitsgruppen/geophysik_seismologie/forschung/ruptrack/index.php


25 km

50 km

75 km

100E 101E 102E

5S

4S

3S

2S0 40 80 120

rupture time [s]

Time [s] from start at 12Sep2007_11:10:23

0.000

0.123

Sem

blan

ce

0 50 100 150 200 250 300 350 400 4500

50

100

150

200

250

300

Dis

tanc

e [k

m]

0

90

180

270

360

Azi

mut

h [d

egre

e]

Fig. 3. Results for theM8.0 Sumatra earthquake on 12 September2007, 11:10. Top: epicentral area with maxima (dots) and 0.9 con-tour lines of normalised semblance source maps. Colours indicaterupture time. Star: epicentre (GEOFON). The focal mechanismshows the centroid moment tensor solution (Ekstrom and Nettles,2009). Grey contour lines indicate slab depth (Gudmundsson andSambridge, 1998). Bottom: time series of maximum of semblancemaxima over all grid points (black/red line), distance (crosses) fromand azimuth (circles) to epicentre from multiplied semblance sourcemaps. Origin time (GEOFON) is a 200 s. Values at times outsidethe interval marked in red are attributed to noise and phases foll-wing theP waves. The red line and area: time interval when sta/ltatrigger indicates rupturing.

earthquakes on 2 September 2009 (Fig.6). The two Benkuluevents caused minor local tsunamis. The Java event is locatednear the tsunami earthquake on 17 July 2006. We also sum-marise the results for other major events of the region sincethe year 2000 (Fig.7 and Table1). Their rupture parame-ters where retrieved applying semblance analysis and similararray configurations as in Fig.1.

The Sunda arc is a major subduction zone at which theIndo-Australian plate is subducted underneath the Sundaplate at a convergence rate of up to 6 cm/yr (Subarya et al.,2006). Within this region a number of majorM > 7 earth-quakes have occurred since the destructive megathrustM9.1earthquake on 26 December 2004. Two other events on

99E 100E 101E

3S

2S

1S

0 40 80 120rupture time [s]

Time [s] from start at 12Sep2007_23:49:01

0.000

0.081

0.162

Sem

blan

ce

0 50 100 150 200 250 300 350 400 4500

50

100

150

200

250

300

Dis

tanc

e [k

m]

0

90

180

270

360

Azi

mut

h [d

egre

e]

25 km

50 km

Fig. 4. As in Fig. 3 but for the nearbyM8.0 earthquake on 12September 2007, 23:49.

28 March 2005 and on 17 July 2006 caused tsunamis thatwere destructive along nearby coastlines. Although most ofthese very large earthquakes show dip in the direction of thesubduction there are also examples for rotated mechanisms(Fig. 7).

Both Benkulu events are typical subduction zone thrustearthquakes with strike being parallel to the subduction front.The onset of the firstM8.0 event at 11:10 is well definedby a rapid semblance increase. It is caught by the sta/ltatrigger at t0 = 195 s close to the GEOFON origin time att0= 200 s (Fig.3). The onset location given by the first trig-gered semblance maximum is located near the GEOFON epi-centre which lies well within the 0.9 combined semblancecontour line. Combined semblance is highest near the on-set but decreases as the rupture progresses. This decreasemay be caused by simultaneous rupture of different parts ofthe fault. After the onset, semblance remains high for 125 s.This is interpreted as the total rupture duration. The rupturepropagates over at least 200 km to the north at a velocity ofabout 2 km/s. Interestingly, the centroid location (Ekstromand Nettles, 2009) is located at the centre of the rupture.Assuming that rupturing occurs along the slab interface itseems to reach progressively greater depth of 5060 km. It



Table 1. Results for selected major earthquakes along the Sunda arc since 2000. Hypocentre parameters and magnitudes are obtained fromGEOFON ( http://geofon.gfz-potsdam.de) or USGS ( http://earthquake.usgs.gov/eqcenter).

hypocentre latitude longitude depthMw duration length velocity direction dominanttime [N] [E] [km] [s] [km] [km/s] character

04/06/2000 4.7 102.1 33 7.9 100 200 2 SE unilateral24/12/2004 3.3 95.9 30 9.1 500 1200 2.8 N unilateral28/03/2005 2.1 97.0 30 8.7 130 200 2 trench-parallel bilateral17/07/2006 9.2 107.3 34 7.7 170 150 1 E unilateral08/08/2007 6.1 107.7 301 7.6 < 50 stationary12/09/2007 11:10 4.6 101.3 32 8.0 120 200 2 N unilateral12/09/2007 23:49 2.6 100.7 28 8.0 110 150 2 N unilateral02/09/2009 8.0 107.3 62 7.4 < 40 stationary

is important to note that we have no indication that rupturingoccurred shallower than 25 km depth. This slab depth inter-val coincides at the surface approximately with the trench-parallel Mentawai fault. This poses the question whether ornot the shallower part of the subduction interface was loadedby the event. The latter would increase the potential for an-other shallow very large event with great tsunami risk.

The second very largeM8.0 Benkulu earthquake hap-pened only a few hours later on 23:49 of the same day. Itoriginated at the northern edge of the first event and con-tinued to propagate northwards (Fig.4). As before the on-set is well defined by high semblance. Rupture duration andlength are 100110 s and 160 km, respectively. The northern-most semblance maximum is interpreted as an outlier. As be-fore, rupturing seems to be bounded between depths of 25 to75 km. Contour lines concentrate near and downdip from thestart and at the end of the rupture. In addition, semblance isincreased towards the end of the event att0=260 s followinga local minimum. This indicates relatively smooth rupturingwith only little variations in slip. The observations as wellas the extent of the rupture retrieved for both Benkulu eventsare compatible with finite source modelling published by theUSGShttp://earthquake.usgs.gov/eqcenter/eqinthenews.

The M7.4 Java earthquake on 2 September 2009 seemsremarkable as it shows reverse faulting with strike perpen-dicular to the subduction trench (Fig.6). The size and focaldepth of the event are appropriate to study some limitationsof the multiple array semblance analysis.

The German Grafenberg array, located at about 100 epi-central distance, has recorded high-qualityPdiff and Sdiffphases (see Fig.5). SinglePdiff phases show seismic en-ergy concentrated near the phase onset but coherent wave-forms for about 60 s with clear additional onsets after about30 s and 50 s. The stack of the 13 single traces (beam) showsthat these phases have the same or very similar slowness. Incontrast, theSdiff wavetrain is shorter and can be followedfor only 40 s with an additional coherent phase about 30 safter the arrival ofSdiff . The different lengths of the wave-trains indicate that the rupture duration is not longer than

40 s. Therefore, the phases at 25 s and 55 s on the verticaland the transverse component seismograms in Fig.5 indicatethe start and the termination of the rupture. Given the depthof the event (Table1) the depth phasespPdiff andsPdiff ar-rive 17.5 and 24.5 s, respectively, afterPdiff . They interferewith the ongoing wavetrain ofPdiff and the phase at 75 s onthe vertical component can be interpreted as the depth phaseof the stopping phase.sSdiff arrives 30.5 s afterSdiff but isnot clearly visible on the transverse components or interfereswith Sdiff .

Since only vertical components are used for real-time sem-blance analysis their long duration is likely to be caused bylate structure related phases originating in the source region.Semblance clearly increases and the event onset is triggeredat 195 s (Fig.6), 5 s before origin time (GEOFON). After theonset, semblance remains high for 65 s where the automaticanalysis defines the event termination. The location of thesemblance maximum at onset is offset by about 0.5 fromthe epicentre. This mislocation reduces within the following30 s. According to the seismogram interpretation, the actualrupture duration is about 30 s. This period appears too shortand the rupture extent too small to resolve other rupture pa-rameter than the event onset. Thereafter, areas of high sem-blance scatter around the epicentre and spread in western andeastern direction. The greatest apparent extent is reached inWNW direction, 50 s after the start of the rupture and 88 kmaway from the epicentre. Such scatter is typical for bilateralearthquakes. Here, it seems to be an artefact due to depthphases.

The interpretation of results is generally difficult for eventsof this size. From our experience smaller earthquakes (M 7.5rupture areawith epicentre

Fig. 7. Map of rupture areas for events in Table1 along with cen-troid moment tensor solutions (Ekstrom and Nettles, 2009). Areasin red show the rupture areas of the events discussed in this paper.Areas in pink show the results for selected events since 2000. Bluestars show epicentres of other large earthquakes,M > 7.5, in thisarea since 1900.

By the combination of well-placed separate arrays insteadof one global array, such polarity correction can be avoided.The considered stations are chosen such that polarity rever-sals within the individual arrays are unlikely for subductionearthquakes along wide regions of the Sunda arc (compareFig. 1). If such polarity reversals occur, semblance will below and uncorrelated at the single array. Even in this case,rupture tracking will remain feasible by the array combina-tion if the polarities within the remaining arrays are consis-tent. The resolution, however, may be reduced. Therefore,the combination of single arrays seems more robust for real-time application than the single global array if the sourcemechanism is not accounted for.

For traveltime calculations,P waves and constant depth asgiven by the hypocentre estimate are assumed. Since depthresolution is insufficient, the automatic analysis is only ap-plied to resolve the horizontal extent of the rupture. Dur-ing beamforming, the signals of phases that have differentslownesses than theP waves are suppressed. Depending onsource depth and epicentral distance the slowness of thePwave are similar to thepP andsP depth phases and differ-ential timesP pP , P sP may be small. Therefore, depthphases may not be sufficiently suppressed. Furthermore, am-plitudes ofpP andsP waves can be large compared to di-rect P . This sometimes may give rise to locating energies

of depth phases using slowness and traveltimes ofP wavesinto the source region. The resulting bias is reflected in shiftsof the occurrence time of a particular source point, less inthe location. This may lead to increased semblance values atlater times corresponding to the traveltime differences. Fordeeper earthquakes the depth phases are well separated fromtheP waves resulting in distinct semblance peaks. This sepa-ration may be used to discriminate between shallow and deepearthquakes. On the other hand, increased semblance due todepth phases may lead to apparently longer rupture duration.This is most important for shallow earthquake were the slow-ness differences are smallest, for small earthquakes, and forearthquakes with a duration similar to the traveltime differ-ences (compare example in Fig.6). Both effects at presentrequire manual inspection.

Rapid rupture characterisation of shallowM 8 earth-quakes along the Sunda arc is of prime importance fortsunami early warning. The automatic retrieval of start time,duration, extent, direction, and propagation velocity of therupture is found reliable for such events. Single semblancemaxima give a spatial resolution of about 0.51. The seriesof maxima provides an image of the rupture that can be eval-uated. The estimation of the rupture area depends on a cali-bration of the shape of the semblance source maps. Resultsfor smaller earthquakes in the range 7M < 8 strongly de-pend on hypocentre as well as on data availability and qual-ity. A high-quality image of the rupture was achieved forthe M7.7 Java earthquake on 17 July 2006 (Kruger et al.,2006) by manual data processing, at that time state-of theart. For the smallerM7.4 event on 2 September 2009, theonset and upper estimates of the extent and the duration wereobtained. Unique event detection is often possible forM 6earthquakes but usually fails below. The application of themethod to other regions of the world requires a different ar-ray configuration.

The computations also improve with data availabilitywhich is limited by the data latency of individual stations.Owing to the goal for rapid rupture characterisation only datalatencies of up to 60 s are considered. However, seismogramsfor some stations are available first at a later time. Therfore,computations are often repeated. This generally improves theresults.

So far the automatic results are obtained and publishedwithin about 25 min. Latest developments aim to increasecomputational efficiency. After coding the algorithm asan integrated part of the new Seiscomp3 software package(Weber et al., 2007) rupture parameters are obtained within22 min after the onset of the event. Preliminary parameterswith lower resolution can be achieved even within 17 minwhen only stations up to 60 epicentral distance are consid-ered. After final testing, results will be made online in thenear future.

Based on our results (Fig.7) we find that almost the en-tire coastal strip to the west of Sumatra has been ruptured bylarge subduction-related thrust earthquakes (M 7.5) since



2000. Three areas that obviously form seismic gaps remain:between the 2000 earthquake and the tip of the Sumatra faultoff the coast of southern Sumatra, off the coast of Padangbetween the 2005 Nias earthquake and the 2007 Bengkuluearthquakes, and the whole stretch between the trench andthe up-dip extent of the aforementioned earthquakes to thesouth of theM9.1 Sumatra-Andaman earthquake in 2004.The seismic gap off the coast of Padang was possibly reducedby theM7.7 event on 30 September 2009 (GEOFON). Thisraises the question for the state of the remaining slab surfaceclose to the trench (compare Fig.7). Brittle behaviour andincreased load of this part of the subduction would increasethe tsunami threat due to large shallow rupture.

It seems that the tsunamigenic potential of an earthquakedepends also on the location and the rupture mechanism.Both the Sumatra-Andaman and the Java earthquakes in 2004and 2007 could reach to or close to the trench. This proximityto the trench and the exceptional low rupture velocity of theJava earthquake in 2006 possibly supported the generation ofa destructive local tsunami. On the contrary, the twoM8.0Sumatra earthquakes in September 2007 apparently rupturedonly at depths below 25 km and could not generate majortsunamis.

4 Conclusions

We have presented a method to retrieve rupture paramtersof large earthquakes along the Sunda arc automatically in areal-time fashion. The method is based on directP waves atseparate large-aperture seismic broadband arrays within tele-seismic epicentral distances of 30 to 100. It evaluates thesemblance as a measure of coherency of waveforms to de-termine the location and the evolution of earthquake ruptur-ing. The semblance source maps resulting from individualarrays are combined by multiplication. This increases res-olution. Results are less dependent on source mechanismsthan for a single global array. For shallowM 8.0 eventsstart, duration, extent, direction, and velocity of the ruptureare determined with confidence. Results for smaller earth-quakes depend on rupture properties, location, and data. Theretrieved rupture parameters can help to evaluate and miti-gate the earthquake and tsunami hazard.

We have presented automatic applications of the method tothe twoM8 earthquakes on 12 September 2007, off the coastof Sumatra and theM7.4 earthquake on 2 September 2009,off the coast of Java. For the two Sumatra events, their start,duration, extent, propagation direction, and velocity of therupture could be retrieved. They show almost trench-parallel,northward rupturing at about 2 km s1 that lasted more than100 s extending up to 200 km. For the smaller Java eventonly the start and an upper limit of the duration (< 40 s) wasfound with confidence. Analysis of the other large (M > 7.5)thrust earthquakes of the region show that almost the entiresubduction zone off the coast of western Sumatra has under-

gone rupturing since 2000. Seismic gaps, however remain tothe south of Nias island and at the southern tip of Sumatraprobably leaving the region at high earthquake risk. Only thegreat Sumatra earthquake in 2004 and the Java earthquakein 2007, both tsunamigenic, could reach to or close to thetrench. Comparing rupture locations and the RUM model(Gudmundsson and Sambridge, 1998) we find that the otherevents where confined to depths below 25 km.

Retrieved parameters are published and can be used toconstrain seismic and tsunami hazard along the Sunda arc.Therefore, the application will be an intergrated part ofGITEWS (Rudloff et al., 2009). At present, the results areavailable within 25 min after event onset. Future develop-ments will reduce computation effort to less than 22 min. Bylimiting the observations to 3060 epicentral distance firstestimates can be provided even within 17 min.

Acknowledgements.Real-time data were provided by GE-OFON (Deutsches GeoForschungszentrum GFZ) and IRIS(http://www.iris.edu). This work received funding through theGITEWS project (German IndonesianTsunami Early WarningSystem), Grant 03TSU01. We thank the editor and the reviewersfor their comments and very stimulating reviews that helped toimprove the manuscript.

Edited by: A. RudloffReviewed by: A. Hutko and another anonymous referee

References

Aki, K. and Richards, P.: Quantitative seismology, University Sci-ence Books, Sausalito, 2002.

Douze, E. J. and Laster, S. J.: Statistics of semblance, Geophysics,44, 19992003, doi:10.1190/1.1440953, short Note, 1979.

Ekstrom, G. and Nettles, M.: The Global CMT Project, online avail-able at: http://www.globalcmt.org, last access: October 2009,2009.

Gudmundsson, O. and Sambridge, M.: A regionalized upper mantle(RUM) seismic model., J. Geophys. Res., 103, 71217136, 1998.

Hanka, W., Heinloo, A., and Jaeckel, K.-H.: Networked Seismo-graphs: GEOFON Real-Time Data Distribution, Orfeus Newslet-ter, 2, 24, http://www.orfeus-eu.org/Organization/Newsletter/vol2no3/geofon.html, 2000.

Ishii, M., Shearer, P., and Vidale, J.: Extent, duration and speedof the 2004 Sumatra-Andaman earthquake imaged by the Hi-Netarray, Nature, 435, 933936, doi:10.1038/nature03675, 2005.

Ishii, M., Shearer, P. M., Houston, H., and Vidale, J. E.: Tele-seismicP wave imaging of the 26 December 2004 Sumatra-Andaman and 28 March 2005 Sumatra earthquake ruptures usingthe Hi-net array, J. Geophys. Res., 112, B11307, doi:10.1029/2006JB004700, 2007.

Kennett, B., Engdahl, E., and Buland, R.: Constraints on seismicvelocities in the Earth from travel times, Geophys. J. Int., 122,108124, 1995.

Kruger, F. and Ohrnberger, M.: Tracking the rupture of theMW =9.3 Sumatra earthquake over 1150 km at teleseismic distance,Nature, 435, 937939, doi:10.1038/nature0369kao, 2005.


http://www.iris.eduhttp://www.globalcmt.orghttp://www.orfeus-eu.org/Organization/Newsletter/vol2no3/geofon.htmlhttp://www.orfeus-eu.org/Organization/Newsletter/vol2no3/geofon.html


Kruger, F., and M. Ohrnberger: Spatio-temporal source characteris-tics of the 26 December 2004 Sumatra earthquake as imaged byteleseismic broadband arrays, Geophys. Res. Lett., 32, L24312,doi:10.1029/2005GL023939, 2005.

Kruger, F., Roler, D., and Ohrnberger, M.: Rupture propagationof the July 17, 2006, Mw=7.7 TsE off-coast Java, in: Eos Trans.AGU, vol. 87 of Fall Meet. Suppl., Abstract S21A-0127, 2006.

Neiddell, N. and Turhan Taner, M.: Semblance and other coherencymeasures for multichannel data, Geophysics, 36, 482497, 1971.

Ohrnberger, M. and Kruger, F.: Imaging of Large Earthquake Rup-ture Processes Using Multiple Teleseismic Arrays: Applicationto the Sumatra-Andaman Island, in: Eos Trans. AGU, vol. 86 ofFall Meet. Suppl., Abstract U11A-0819, 2005.

Roler, D., Kruger, F., and Ohrnberger, M.: Rupture Propagationof the 2008/05/12 Ms8.0 Wenchuan Earthquake Using MultipleTeleseismic Arrays, in: Eos Trans. AGU, vol. 89 of Fall Meet.Suppl. Abstract S31B-1908, 2008.

Roler, D., Kruger, F., Ohrnberger, M., and Ehlert, L.: Teleseismicrupture tracking of large earthquakes, online available at:http://www.uni-potsdam.de, 2009.

Rost, S. and Thomas, C.: Improving Seismic Resolution ThroughArray Processing Techniques, Surv. Geophys., 30, 271299, doi:10.1007/s10712-009-9070-6, 2009.

Rudloff, A., Lauterjung, J., Munch, U., and Tinti, S.: Preface TheGITEWS Project (German-Indonesian Tsunami Early WarningSystem), Nat. Hazards Earth Syst. Sci., 9, 13811382, 2009,http://www.nat-hazards-earth-syst-sci.net/9/1381/2009/.

Subarya, C., Chlieh, M., Prawirodirdjo, L., Avouac, J.-P., Bock,Y., andA. J. Meltzner, K. S., Natawidjaja, D. H., and Mc-Caffrey, R.: Plate-boundary deformation associated with thegreat Sumatra Andaman earthquake, Nature, 440, 4651, doi:10.1038/nature04522, 2006.

Weber, B., Becker, J., Hanka, W., Heinloo, A., Hoffmann, M.,Kraft, T., Pahlke, D., Reinhardt, J., Saul, J., and Thoms, H.:SeisComP3 automatic and interactive real time data process-ing, in: Geophysical Research Abstracts, vol. 9 of EGU GeneralAssembly, 2007.

Xu, Y., Koper, K. D., Sufri, O., Zhu, L., and Hutko, A. R.: Rup-ture imaging of theMw7.9 12 May 2008 Wenchuan earthquakefrom back projection of teleseismicP waves, Geochem. Geophy.Geosy., 10, Q04006, doi:10.1029/2008GC002335, 2009.

Zhang, H., Xu, L., Chen, Y., Li, C.-l., and Stammler, K.: Frequency-domain array technique analysis for the rupture duration time andgeometrical characteristics of the 2001 Kunlun Mountain Passearthquake, Acta Seismologica Sinca, 21, 1123, doi:10.1007/s11589-008-0011-0, 2008.


http://www.uni-potsdam.dehttp://www.uni-potsdam.dehttp://www.nat-hazards-earth-syst-sci.net/9/1381/2009/

Rapid characterisation of large earthquakes by multiple seismic broadband arrays

Documents

Transcript of Rapid characterisation of large earthquakes by multiple seismic broadband arrays