Special Research Articles

The Great Sumatra-Andaman Earthquake of 26 December 2004

See allHide authors and affiliations

Science  20 May 2005:
Vol. 308, Issue 5725, pp. 1127-1133
DOI: 10.1126/science.1112250


The two largest earthquakes of the past 40 years ruptured a 1600-kilometer-long portion of the fault boundary between the Indo-Australian and southeastern Eurasian plates on 26 December 2004 [seismic moment magnitude (Mw) = 9.1 to 9.3] and 28 March 2005 (Mw = 8.6). The first event generated a tsunami that caused more than 283,000 deaths. Fault slip of up to 15 meters occurred near Banda Aceh, Sumatra, but to the north, along the Nicobar and Andaman Islands, rapid slip was much smaller. Tsunami and geodetic observations indicate that additional slow slip occurred in the north over a time scale of 50 minutes or longer.

The 26 December 2004 Sumatra-Andaman earthquake was the largest seismic event on Earth in more than 40 years, and it produced the most devastating tsunami in recorded history (1). Like other comparably sized great earthquakes—such as the 1952 Kamchatka, the 1957 Andreanof Islands in the Aleutians, the 1960 Southern Chile, and the 1964 Prince William Sound, Alaska, earthquakes—the Sumatra-Andaman event ruptured a subduction zone megathrust plate boundary. These giant earthquakes occur where large oceanic plates underthrust continental margins. They involve huge fault areas, typically 200 km wide by 1000 km long, and large fault slips of 10 m or more. Such events dwarf the contributions to plate motion of vast numbers of lower magnitude earthquakes. The high tsunami-associated death toll appears to have been due to the dense population of the affected region. The tsunami magnitude, Mt, of the earthquake was 9.1 (2), as compared to Mt = 9.1 for the 1964 Alaska and Mt = 9.4 for the 1960 Chile earthquakes (3). The event ruptured 1200 to 1300 km of a curved plate boundary, with variations in direction of interplate motion and age of subducting lithosphere apparently affecting the nature of the faulting. The 28 March 2005 event ruptured an adjacent 300-km-long portion of the plate boundary (4). These two events are the largest to occur after the global deployment of digital broadband, high-dynamic-range seismometers (5, 6), which recorded both the huge ground motions from the mainshocks and the tiny motions from ensuing free oscillations of the planet and from small aftershocks (7, 8). In this and two companion papers (9, 10), we report on the nature of faulting in these great earthquakes based on seismological analyses of the extensive, openly available seismogram data set from the international Federation of Digital Seismic Networks (FDSN) backbone network (5, 6).

Plate geometry and setting. The 2004 and 2005 earthquakes ruptured the boundary between the Indo-Australian plate, which moves generally northward at 40 to 50 mm/year, and the southeastern portion of the Eurasian plate, which is segmented into the Burma and Sunda subplates (Fig. 1). East of the Himalayas, the plate boundary trends southward through Myanmar, continuing offshore as a subduction zone along the Andaman and Nicobar Islands south to Sumatra, where it turns eastward along the Java trench (11). As a result of the highly oblique motion between the Indo-Australian plate and the Burma and Sunda subplates (Fig. 1), a plate sliver, referred to as the Andaman or Burma microplate, has sheared off parallel to the subduction zone from Myanmar to Sumatra (12). Oblique, but predominantly thrust, motion occurs in the Andaman trench with a convergence rate of about 14 mm/year (13, 14). The Andaman Sea ridge-transform system, an oblique back-arc spreading center, accommodates the remaining plate motion, joining with the Sumatra Fault to the south (15, 16). Underthrusting along the Sunda trench, with some right-lateral faulting on the inland Sumatra Fault, accommodates interplate motion along Sumatra.

Fig. 1.

Regional map showing earthquakes with magnitudes >5.0 from 1965 to 25 December 2004 from the earthquake catalog of the National Earthquake Information Center (NEIC). Red dots show events with depths <33 km; orange, depths of 33 to 70 km; yellow, depths of 70 to 105 km; and green, depths >105 km. Locations of previous large earthquake ruptures along the Sunda-Andaman trench system are shown in pink. Dashed box shows area of the map in Fig. 2. Green stars show the epicenters of the two recent great events; the green diamond shows the CMT centroid location for the 2004 Sumatra-Andaman event. The thick red arrows indicate the NUVEL-1 relative plate motions between the Indo-Australian and Eurasian plates.

Historic great earthquakes along this plate boundary occurred in 1797 [magnitude (M) ∼ 8.4], 1833 (M ∼ 9), and 1861 (M ∼ 8.5) (17, 18), providing the basis for the long-recognized potential for great earthquakes along Sumatra (11, 19). A smaller (M ∼ 7.8) event in 1907 just south of the 2004 rupture zone produced seismic and tsunami damage in northern Sumatra (11). These events all occurred to the southeast of the 2004 rupture zone (Fig. 1). The 28 March 2005 event ruptured the same region as the 1861 and 1907 events (Fig. 1). Smaller events in the Andaman trench, also presumed to involve thrusting motions, occurred beneath the Nicobar Islands in 1881 (M ∼ 7.9) and near the Andaman Islands in 1941 (M ∼ 7.9). There is no historical record of a previous tsunamigenic earthquake in the Bay of Bengal comparable to the 2004 event (12).

In the 40 years preceding the 2004 event, little seismicity occurred within 100 km of the trench in the region between the 2004 and 1881 epicenters (figs. S1 to S3). Similarly, seismicity was low in the source region of the great 1861 earthquake before the 28 March 2005 event and is still low in the 1833 rupture region (fig. S2). Numerous earthquakes occurred near the 2004 epicenter in recent years, including a seismic-moment magnitude (Mw) = 7.2 event in 2002. These features are consistent with long-term strain accumulation in the eventual rupture zone and stress concentration in the vicinity of the mainshock hypocenter.

The mainshocks. The 2004 mainshock rupture began at 3.3°N, 96.0°E, at a depth of about 30 km, at 00:58:53 GMT (1). The Harvard centroid-moment-tensor (CMT) solution indicates predominantly thrust faulting on a shallowly (8°) dipping plane with a strike of 329° (20, 21). The rake (110°) indicates a slip direction ∼20° closer to the trench-normal direction than to the interplate convergence direction, consistent with some long-term partitioning of right-lateral motion onto the Sumatra Fault, which is not reported to have ruptured during the 2004 event. The aftershock distribution (Fig. 2) gives a first-order indication of the extent of the mainshock ruptures. For the 2004 event, the distribution suggests a rupture length of 1300 km extending from northwestern Sumatra to the Andaman Islands. Along the northern extension of the aftershock zone, the strike of the fault rotates progressively clockwise (Fig. 2). The associated degree of right-lateral slip (rake >90°) on the megathrust fault should increase to the north unless that component of interplate motion is partitioned onto the back-arc transform system. Given the variation in the relative plate motion along the aftershock zone, it is surprising that the CMT solution is a nearly pure double-couple source, indicative of simple faulting geometry. The faulting solution favors a concentration of <500 s period seismic radiation in the southern portion of the aftershock zone, as does the location of the centroid, which lies about 160 km west of the epicenter (Fig. 2).

Fig. 2.

Map showing aftershock locations for the first 13 weeks after the 26 December 2004 earthquake from the NEIC (yellow dots, with radii proportional to seismic magnitude). Moment-tensor solutions from the Harvard CMT catalog (21) are shown for the 26 December 2004 and 28 March 2005 mainshocks (large solutions at bottom, with associated centroid locations) and aftershocks. Star indicates the epicenter for the 2004 rupture obtained by the NEIC. Dashed line shows the boundary between the aftershock zones for the two events.

The 2005 mainshock rupture began at 2.1°N, 97.0°E at a depth of about 30 km, at 16:09:36 GMT (4). Motion for this event was also predominantly dip-slip thrusting on a shallowly (7°) dipping plane with a strike of 329° (20, 21). The 300-km-long aftershock zone along Northern Sumatra (Fig. 2) suggests a relatively uniform rupture geometry.

Peak-to-peak ground motions for the 2004 event exceeded 9 cm in Sri Lanka, 15.5° from the epicenter, and long-period surface-wave motions exceeded 1 cm everywhere on Earth's surface (7) (fig. S4). This giant event produced motions (Fig. 3) that dwarf those of the 2005 event and the 23 June 2001 Peru earthquake (Mw = 8.4), the largest earthquake previously recorded by global broadband seismometers (22). The high-quality global recordings enable seismological quantification of these great earthquakes (23).

Fig. 3.

Vertical-component ground displacements for periods <1000 s observed for the three largest earthquakes of the past 40 years. The upper trace shows the seismogram from the 26 December 2004 Sumatra-Andaman earthquake observed 130° away in Pasadena, California, USA; the middle trace is for the 28 March 2005 Sumatra earthquake observed 131° away in Pasadena, California, USA; the lower trace shows a seismogram for the 23 June 2001 Mw 8.4 earthquake off the coast of Peru, observed 126° away in Charters Towers, Australia. Additional waveforms are shown in fig. S4.

Aftershock geometry. Harvard CMT focal mechanisms for aftershocks (Fig. 2) display a variety of geometries, including thrust faulting along the subduction zone and strike slip and normal faulting in the Andaman Sea back arc. These mechanisms are generally consistent with the expected slip partitioning along the boundary, with nearly arcnormal thrusting in the shallowly dipping subduction zone and right-lateral shearing in the back arc.

The most notable aftershock feature is a swarm of strike-slip and normal faulting events in the Andaman Sea back-arc basin (Fig. 2) involving more than 150 magnitude 5 and greater earthquakes that occurred from 27 to 30 January 2005 (20, 21). Previous swarms of events have occurred in this region (e.g., July 1984), but this is the most energetic earthquake swarm ever observed globally. This swarm activity can be seen as part of the overall interplate motion partitioning.

Although aftershock mechanism variability and location uncertainty make it difficult to constrain the fault geometry in detail using seismicity, the megathrust appears to be about 240 km wide along northwestern Sumatra, extending to a depth of about 45 km. Along the Nicobar and Andaman Islands, the megathrust fault plane appears to be no more than 160 to 170 km wide, extending to a depth of about 30 km.

Magnitude, source strength, and energy. The Harvard CMT solution for the 2004 earthquake, based on global FDSN recordings of 300- to 500-s period surface waves, has a seismic moment Mo = 4.0 × 1022 Nm (24), comparable to the cumulative seismic moment of all earthquakes for the preceding decade (Fig. 4). This moment yields Mw = 9.0, the widely quoted seismic magnitude for the mainshock. Uniform slip of about 5.0 m over a 1300-km-long fault varying in width from 240 to 160 km with rigidity μ = 3.0 × 1010 N/m2 would account for the CMT seismic-moment estimate. Larger slip on a smaller fault area in the south would also match the seismic moment. The CMT seismic moment for the 2005 mainshock is Mo = 1.1 × 1022 Nm (Mw = 8.6).

Fig. 4.

Plot of cumulative seismic moment as a function of time for the 29-year history of the Harvard CMT catalog, which contains results for global earthquakes of magnitude larger than ∼5.0, with great (Mw ≥ 8) earthquakes indicated by stars. The 300- to 500-s period seismic moment for the 2004 event is comparable to the cumulative global earthquake seismic moment release for the preceding decade.

Mw is intended to characterize the earthquake process in terms of its final, static offset but is usually based on measurements of long-period seismic waves. For very large earthquakes, these measurements are typically made at periods of 100 to 300 s, the range of commonly observed seismic surface waves like those in Fig. 3. Magnitudes of the 1960 Chile (Mw = 9.5), the 1964 Alaska (Mw = 9.2), and the 2004 Sumatra-Andaman (Mw = 9.0) earthquakes were all estimated from measurements around 300 s (2527). In this sense, their magnitudes can be directly compared. The 100- to 300-s period surface-wave amplitudes for the 1964 Alaska earthquake were about three times as large as those for the 2004 Sumatra-Andaman earthquake.

When an estimate of the seismic moment is made at periods too short to represent fully the earthquake source process, the result is an underestimate of the earthquake size. Given the high-quality seismic data available for the 2004 earthquake, the effective source strength (28) can be determined over a broad range of frequencies with relatively good confidence (Fig. 5). For an assumption of uniform faulting geometry, the strength of the seismic-wave excitation for periods >500 s was enhanced by a factor of 1.5 to 2.5 compared with that at 300 s (9, 10). The moment magnitude of the Sumatra-Andaman earthquake may thus be larger than 9.0 by 0.1 to 0.3 units (29). The data for the Chile and Alaska earthquakes are insufficient to determine the magnitude at very long periods. Even at 300 s, considerable uncertainty is involved in the values of Mw for old events because of data limitations. As a result, comparison of Mw for these events to one or two tenths of a magnitude unit is not meaningful. These three earthquakes should instead be compared with respect to all aspects of their source characteristics (e.g., source spectrum, radiated energy, slip distribution, and rupture speed).

Fig. 5.

Rayleigh-wave source-time function (STF) amplitude spectrum (solid and dotted line) computed by stacking more than 200 R1 observations obtained by deconvolution of propagation effects in the frequency domain. We used a signal duration ranging from 750 to 2000 s (depending on source-receiver distance) zero-padded to 6000 s to estimate each spectrum. The dotted portion of the spectrum indicates frequencies for which the R1 spectral amplitudes are not well resolved. White diamonds show estimates of moment from free oscillations (0S2, 0S3, 0S4, and 0S0, from left to right) made by Northwestern University. Black diamonds show spectral levels estimated by comparison of filtered time-domain signals with synthetics for the Harvard CMT solution. Effective source strengths from the Harvard CMT solution (300 to 500 s) and that estimated from 0S2, assuming the Harvard CMT faulting geometry, are indicated with dashed lines. The lowest frequency level, extrapolated to zero frequency, corresponds to the seismic moment, Mo. The approximate uncertainty for all of the low-frequency estimates is about a factor of 2. The bar indicates a corresponding range of uncertainty. The shaded region represents the composite source spectrum.

Short-period radiation is particularly important for ground accelerations and intensity of structural damage. Estimates of seismic intensity, based on relative measures of structural damage and vibration, indicate intensity IX in the vicinity of Banda Aceh, intensity VII in Port Blair in the Andaman Islands, and a relatively low intensity II to IV around the Bay of Bengal (12, 30). The overprinting devastation from the tsunami complicates the estimation of high-frequency effects, but intensity IX is consistent with the short-period magnitudes mb = 7.0 estimated by the USGS and b = 7.2 reported here, the latter being about 0.3 magnitude units lower than for the 1964 Alaska earthquake (31, 32) (fig. S5). These values imply that the 2004 event was not depleted in high-frequency radiation, unlike the notable 1992 Nicaragua tsunami earthquake (33).

The energy radiated by seismic waves, ER, is an important macroscopic seismic parameter, because the amount of potential energy partitioned to ER reflects the physical process of the source (34). Unfortunately, accurate estimation of ER, especially for great earthquakes with long source durations, is difficult. An estimate of ER = 1.1 × 1018 J is obtained for the 2004 event from P waves at 11 stations over a distance range of 45° to 95° (3537). Energy estimates for earlier giant earthquakes are based only on magnitude-energy relationships, so meaningful comparisons are difficult. The radiated energy estimated here is about 10 times that of the 1994 deep Bolivia earthquake (Mw = 8.3) and about 40 times that of the 2001 Peru earthquake (Mw = 8.4) (Fig. 3).

Slip process of the 2004 event. The 2004 Sumatra-Andaman earthquake had the longest known earthquake rupture. Short-period seismic body waves (0.5 to 0.25 s) show azimuthally varying durations that indicate that the seismic rupture front propagated to about 1200 km north of the epicenter with a rupture velocity of about 2.0 to 3.0 km/s and that short-period radiation was generated for at least 500 s (38). Array analysis of 1- to 2-s period seismic waves from Hi-net stations in Japan yields compatible results (39). Analysis of longer period body waves and surface waves demonstrates that most of the slip that generated seismic waves was concentrated in the southern half of the rupture zone, with diminishing, increasingly oblique slip toward the north on the fault (9). The seismic moment of models that successfully match the long-period body- and surface-wave data is about 1.5 times as large as the CMT moment, consistent with free oscillation observations (10).

The seismic model does not, however, account for all observations. Geodetic constraints require two to three times more slip in the north (40). This suggests rupture of the northern region with a long source-process time that generated little or no seismic waves. Well-documented tilting in the Andaman and Nicobar Islands (12), with the western margins of the islands being uplifted and the eastern margins being submerged, can be accounted for by substantial slip of about 10 m on a 160-km-wide thrust plane in the northern half of the rupture zone or by less slip on more steeply dipping splay faults. Such large slip must have occurred on time scales longer than 1000 s, because it did not generate strong seismic-wave radiation late in the rupture.

Arrival times of tsunami waves around the Sea of Bengal provide additional constraints on the slip distribution in the north. Bounds can be placed on the location of ocean-bottom uplift due to faulting by back-propagating the initial tsunami wavefront from tsunami recording locations to the source region. The source region for strong initial tsunami excitation extends 600 to 800 km north of the epicenter, terminating near the Nicobar Islands (41) (Fig. 6 and fig. S6). The northern third of the aftershock zone appears not to have produced rapid vertical ocean-bottom displacements capable of generating large tsunami waves (fig. S7), but delayed slip cannot be ruled out. This estimate of the tsunami source region is consistent with satellite altimetry observations of the deep-water waves obtained by fortuitous passage of two satellites over the Indian Ocean 2 to 3 hours after the rupture occurred (42) (Fig. 7).

Fig. 6.

Constraints on the tsunami source area obtained from the timing of tsunami arrivals at various locations around the Indian Ocean. Dark lines indicate the distance from the observing points from which tsunami might have generated at the event origin time. The tsunami source area outlined by these curves (brown region) appears to extend only 600 to 800 km north-northwest from the epicenter.

Fig. 7.

(Left) Tsunami model at a time of 1 hour 55 min after earthquake initiation, computed for a composite slip model with fast slip (50-s rise time) in the southern portion of the rupture and slow slip (3500-s rise time) in the north. The northward propagating rupture velocity is about 2 km/s for the first 745 km, then slows to 750 m/s. The amplitude of fast and slow slip on the six fault segments are indicated by white numbers and outlined numbers, respectively. The overall seismic moment of 8.8 × 1022 Nm (μ = 3.0 × 1010 N/m2) is divided fairly evenly between slow and fast contributions. Red colors in the map indicate positive ocean wave height, blue colors negative. The numbers along the wavefront give wave amplitudes in meters. Diagonal line is the track of the Jason satellite that passed over the region at about this time (10 min of actual transit time along the profile). The predicted (blue) and observed (red) tsunami wave are shown in the inset. The tsunami generated by the fast component of slip alone cannot explain the trough in the central Bay of Bengal (fig. S8 and Movie S1). (Right) Tsunami waveforms and estimated run-up heights for five locations around the Bay of Bengal. The first arrivals show water draw-down toward the east and inundation toward the west. Principal wave period is about 30 min.

The tsunami calculation shown in Fig. 7, which provides a generally satisfactory fit to the satellite observations, uses a finite-fault model with the same geometry as that providing a good fit to the seismic data but with somewhat different slips, rise times, and start times on each segment (Movie S1). A broad trough in the ocean surface in the central Sea of Bengal 2 hours after the earthquake can be modeled well if slow slip occurred over ∼1 hour under the Nicobar and Andaman Islands (fig. S8 shows the prediction for a model with no slow slip). This is currently the strongest constraint on the source-process time for the slow-slip component of the 2004 event. The tsunami energy computed for this composite model is 4.2 × 1015 J, less than 0.5% of the strain energy released by the faulting. Tsunami generation is not a very efficient process, but tremendous destruction can clearly result from this small component of the energy budget. No slow slip has yet been resolved for the 28 March 2005 event (9) (Movie S2).

Discussion. The 2004 Sumatra-Andaman earthquake rupture appears to have been a compound process of seismic-energy release, involving variable slip amplitude, rupture velocities, and slip duration. About 90 to 95% of the seismic observations can be accounted for with the rupture model (9, 10) depicted schematically in Fig. 8.

Fig. 8.

Summary rupture scenario for the 2004 Sumatra-Andaman earthquake. We subdivide the rupture zone into three segments according to the inferred rupture process, not because of clear physical fault segmentation. The rupture begins at the southeastern edge of the Sumatra segment, with the initial 50 s of rupture characterized by fairly low energy release and slow rupture velocity. The rupture front then expands to the north-northwest at about 2.5 km/s, extending about 1300 km. Short-period radiation tracks the rupture front, with a total duration of about 500 s and clear north-northwest directivity. Large, rapid slip occurs in the Sumatra segment, with some patches having slip as great as 20 m during the first 230 s. The Nicobar segment has weaker slip during the next 2 min, and the Andaman segment fails with little (<2 m) rapid slip. Slow slip appears to continue in the Nicobar and Andaman segments, with a total duration of about 1 hour. The precise amount of slip and total moment of the slow-slip component are not well resolved, but about 10 m of slip under the Andaman Islands is required to account for the tilt experienced by the islands.

The northern portion of the fault appears to have slipped 3 to 7 m more than accounted for by the seismic model, with a time scale of ∼1 hour or longer. The cause of the compound slip behavior is not well-understood. It appears that the slow slip occurred only along the Nicobar and Andaman Islands segments of the rupture zone, where the plate convergence is increasingly oblique and slip is strongly partitioned. For the low convergence rate of 14 mm/year in the region, it would take 700 years to accumulate 10 m of slip potential in the region, which is consistent with the lack of historical great events in the northern part of the subduction zone.

The age of the subducting oceanic plate increases from about 60 million to 90 million years between Sumatra and the Andaman Islands, and this change may also influence mechanical coupling on the thrust plane. Subduction of younger lithosphere tends to result in interplate faults with shallow dips and broad contact areas that generate great earthquakes, whereas in locations where older lithosphere is subducted and back-arc spreading is observed, great earthquakes are rare (4345). The strong lateral gradient in obliquity of interplate motion and the rapid increase in age of the subducting oceanic plate toward the north-northwest are distinctive features compared with the settings of previous great earthquakes. The most analogous tectonic environment may be the western Aleutians, where the 1965 (Mw = 8.7) Rat Island earthquake occurred along a curving plate boundary with increasing obliquity of interplate motion along the arc (46).

Logical regions for concern about future large earthquakes are along the Sumatra Fault and southeast of the 2005 event rupture; the adjacent region failed in 1833 and is likely to have accumulated substantial strain. International efforts to improve tsunami-warning capabilities in the Indian Ocean are warranted given the inevitability of future great thrust earthquakes along the Sumatra subduction zone.

Supporting Online Material


Figs. S1 to S8

Table S1

Movies S1 and S2

References and Notes

View Abstract

Navigate This Article