You are on page 1of 14

Geophysical Journal International

Geophys. J. Int. (2011) 185, 1365–1378 doi: 10.1111/j.1365-246X.2011.05009.x

Is there a best source model of the Sumatra 2004 earthquake


for simulating the consecutive tsunami?

B. Poisson, C. Oliveros and R. Pedreros


BRGM, Land Use Planning and Natural Hazards, Orléans, France. E-mail: b.poisson@brgm.fr

Accepted 2011 March 7. Received 2011 February 25; in original form 2009 December 24

SUMMARY

Downloaded from https://academic.oup.com/gji/article/185/3/1365/603962 by guest on 27 April 2021


Many studies have attempted to invert the fault source of the Sumatra 2004 event. Whereas
they mostly consider the same fault geometry, they lead to a wide range of potential slip
distributions. Using tsunami modelling with GEOWAVE, a model based on fully non-linear
Boussinesq equations, we investigate the influence of five distinct source models of various
origins and the influence of rupture kinematics on the generated tsunami. The simulation
results are considered both at ocean-scale and at country-scale. We then compare the results
along the Jason-1 track to the corresponding sea surface height anomaly (SSHA) profile, and
examine the patterns of maximum sea surface elevation around the Bay of Bengal and on the
eastern and southern coastlines of Sri Lanka. In most cases, the synthetic SSHA profiles are
not very consistent with the observed one, although they all display a prevailing first wave.
The divergence in maximum sea surface elevation appears in particular along the Sri Lankan
coast, where three of the models lead to a clearly underestimated tsunami impact. The best
models derive from purely seismic and geodetic studies that did not consider any tsunami
modelling, thus suggesting that the methods to invert a fault model from tsunami data still
need some adjustment, especially concerning the handling of coastal data. In addition, it is
important to consider the rupture kinematics along such a long fault, as the generated tsunami
is more significant than when considering an instantaneous rupture. In the case of the Sumatra
2004 event, the tsunami impact on Sri Lanka is notably underestimated if modelled without

GJI Seismology
the rupture kinematics. We conclude that for tsunami modelling, a complex description of the
fault source model is not absolutely necessary, but some significant parameters such as rupture
kinematics should be taken into account.
Key words: Tsunamis; Earthquake dynamics; Indian Ocean.

et al. 2007; Chlieh et al. 2007), from seismic waveforms (Ammon


1 I N T RO D U C T I O N
et al. 2005; Song et al. 2005; Tsai et al. 2005; Vallee 2007), from tide
The 2004 December 26 Sumatra earthquake was one of the most gauge records (Tanioka et al. 2006; Piatanesi & Lorito 2007) and
disastrous recorded earthquakes, reaching a magnitude M w between from satellite altimetry data (Hirata et al. 2006; Sladen & Hébert
9.1 and 9.3 (Ammon et al. 2005; Park et al. 2005; Stein & Okal 2008). Many studies combined two or more data sets of different
2005). The earthquake generated a tsunami that propagated through nature to better constrain their model (e.g. Lay et al. 2005; Fujii &
the Indian Ocean and caused extreme flooding and destruction along Satake 2007; Pietrzak et al. 2007; Rhie et al. 2007). Some of these
the surrounding coasts (Chanson 2005). The resulting death toll studies included a check of the inverted source by simulating the
is estimated between 200 000 and 300 000. The most devastated generated tsunami (Lay et al. 2005; Song et al. 2005; Chlieh et al.
country was Indonesia, as it lies near the subduction zone where 2007; Pietrzak et al. 2007). In most cases, the simulated tsunami
the earthquake occurred, but the tsunami also severely affected Sri was then compared to the altimetric satellite records of Jason-1 and
Lanka, southern India and Thailand (Chanson 2005; Liu et al. 2005; Topex-Poseidon, both of which detected the tsunami waves in the
Narayan et al. 2005; Choi et al. 2006; Goff et al. 2006; Tsuji et al. Indian Ocean around 2 h after the earthquake (Gower 2005; Kulikov
2006; Yeh et al. 2006). et al. 2005; Smith et al. 2005). Generally, the tsunami simulation
Since 2005, numerous studies have attempted to model the source was carried out with a specific tsunami model chosen by the authors
of the earthquake that triggered the tsunami. Different methods were regardless of the choice of the other workers, which means that the
used: inversion of the slip distribution from GPS (Global Positioning results of different studies are not comparable, as they rely neither
System) offsets (Vigny et al. 2005; Subarya et al. 2006; Banerjee on the same data nor on the same tsunami modelling.


C 2011 The Authors 1365
Geophysical Journal International 
C 2011 RAS
1366 B. Poisson, C. Oliveros and R. Pedreros

In addition, the 2004 Sumatra earthquake involved an excep- Then, the propagation and inundation of tsunami waves are com-
tionally extended fault and the rupture propagated for around 500 s puted with FUNWAVE, a model based on fully nonlinear Boussi-
(e.g. Ishii et al. 2005; Tolstoy & Bohnenstiehl 2005). Some source nesq equations accounting for frequency dispersion (Wei et al.
models thus inverted the rupture velocity together with the slip dis- 1995).
tribution (e.g. Ishii et al. 2005; Vigny et al. 2005; Lambotte et al. The main advantages of using FUNWAVE are the following:
2007; Piatanesi & Lorito 2007; Rhie et al. 2007). However, numer-
(i) FUNWAVE takes into account the frequency dispersion,
ical studies of the tsunami usually do not take the rupture velocity
which can be important in the case of short-wave propagation
into account, or do so only by imposing a time lag between a few
through the deep ocean, as occurred west of the Sumatra–Andaman
large subfaults.
fault during the 2004 event. Grilli et al. (2007) estimated that the
In the light of these observations, some questions emerge.
dispersive effect reduces the maximum water elevation in deep wa-
(i) How can we account for such different models of the same ter by 23 per cent, compared to the results from non-linear shallow
seismic source whereas they all appear to fit some data? If we have water equations (NLSWE) models.
to choose one source model for this event, for example for studying (ii) The non-linearity of equations is crucial to the correct mod-
the regional tsunami hazard, which one should be preferred, and elling of propagation in shallow water and then, of coastal impact.
why? FUNWAVE thus is an appropriate model for simulating tsunami

Downloaded from https://academic.oup.com/gji/article/185/3/1365/603962 by guest on 27 April 2021


(ii) How does the rupture propagation influence the generation, propagation from the deep ocean to coastal regions. The GEOWAVE
propagation and impact of the tsunami? The 2004 Sumatra event methodology is described in Grilli et al. (2007) and the code has
provides an exceptional opportunity for studying this influence that already been validated on various tsunami studies (e.g. Watts et al.
has already been proved to be non-negligible (Song et al. 2005; 2003; Day et al. 2005; Ioualalen et al. 2007; Poisson et al. 2009).
Wang & Liu 2006; Piatanesi & Lorito 2007), but which is not taken When looking at the available satellite altimeter measurements
into account in most tsunami modelling studies. of the 2004 Sumatra tsunami, the frequency dispersion appears to
In this study, we simulate the tsunami generated by five different be actually significant (Kulikov et al. 2005; Ablain et al. 2006).
seismic source models from the literature. To this end, we use the High-frequency components in the tsunami signal are likely due
GEOWAVE tsunami generation and propagation model that was to variations in slip and bathymetry along the fault (Hanson et al.
elaborated by Watts et al. (2003) and validated on many tsunami 2007). The choice of the FUNWAVE model thus appears to be
events. At the same time, we simulate each tsunami both with the particularly appropriate for the 2004 Sumatra tsunami (Kulikov
inverted rupture velocity and with an infinite rupture velocity (in- et al. 2005).
stantaneous seafloor deformation). The simulation results are first
examined with respect to overall ocean and coastal observations,
2.2 Selected source models
after which we focus on Sri Lanka to compare the results at a
country-wide scale. In this study, we refer to five published earthquake source parameters
for the 2004 December 26 event resulting from data inversion. If
the authors inverted several models for the fault, we take the model
2 MODEL that they considered as the best one. Among all significant fault
parameters, the rupture velocity is not always estimated, depending
2.1 Tsunami generation and propagation on the method used. The main characteristics of the five models are
briefly summarized in Table 1.
We choose to use the tsunami generation, propagation and inun-
dation model GEOWAVE (Watts et al. 2003). The initial tsunami M1: Banerjee et al. (2007) deduced the coseismic slip distri-
(surface water deformation and velocity) is computed from the bution on the fault from GPS static offsets. The fault geometry
vertical seafloor displacements produced by an elastic dislocation is inferred from the aftershock focal mechanisms, the seismically
simulating the earthquake (Okada 1985). To do this computation, determined slab-depth contours, and near-field geodetic data in the
the fault or subfaults are usually subdivided into elementary areas Andaman and Nicobar islands (Banerjee et al. 2005). The coseismic
of 10 km × 10 km. In the simulation with a rupture velocity, we rupture area was divided into 15 subfaults; following this, the fault
impose smaller elementary areas of 1 km × 1 km whereby each one slip distribution was computed from the static surface displacement
breaks at a time computed from the distance to the hypocentre and data (near-field and far-field GPS offsets) through Green’s func-
from the rupture velocity. tions applied to a layered spherical earth. As they inverted the slip

Table 1. Main characteristics of the five source models tested in this study. The data used for inverting/checking the model may be: [defo], deformation data;
[tsun], tsunami data; [seis], seismological data. For each study, it is indicated if tsunami modelling was performed (for the inversion or checking), and—if
necessary—if it is based upon the (non)linear shallow water equations: (N)LSWE. Whether or not the rupture velocity was inverted or tested in the referred
study is also reported.
Model Reference Data used for inversion Data used for checking Tsunami modelling Rupture velocity
M1 Banerjee et al. [defo] GPS static offsets (near- and - No -
far-field)
M2 Fujii & Satake [tsun] SSHA + tide gauge data - Yes: LSWE a priori values tested
M3 Piatanesi & Lorito [tsun] tide gauge data - Yes: NLSWE inverted
M4 Rhie et al. [defo] GPS static offsets (near-field) - No inverted
[seis] long-period seismic waveforms
M5 Chlieh et al. [defo] GPS static offsets (near- and seismic data + very far-field Yes: NLSWE a priori values tested
far-field) + coral reefs uplift GPS data + SSHA


C 2011 The Authors, GJI, 185, 1365–1378

Geophysical Journal International 


C 2011 RAS
A best source for Sumatra 2004 tsunami? 1367

Downloaded from https://academic.oup.com/gji/article/185/3/1365/603962 by guest on 27 April 2021


Figure 1. Slip distribution on the discretized fault responsible for the 2004 tsunami, for each of the five considered models. The slip scale is in metres.

distribution from static surface deformation, Banerjee et al. (2007) (i) Many source models have been published for the 2004
did not obtain any indication about the rupture velocity. December 26 earthquake. However, we select models that were
M2: Fujii & Satake (2007) inferred their tsunami source from drawn up long after the event, so that they had the time to incorpo-
12 tide gauge records and the sea surface height anomaly (SSHA) rate validated data for inverting the source.
measured by three satellites (Jason-1, Topex-Poseidon and Envisat). (ii) The selected sources have a similar geometry, derived from
The fault geometry was slightly modified after Hirata et al. (2006) Banerjee et al. (2005), but they clearly show distinct slip distribu-
and several a priori rupture velocities were tested. They divided the tions (see Section 2.3, hereafter, and Fig. 1).
fault area into 22 subfaults for which tsunami waveforms (equiv- (iii) The rupture velocity has been inverted in four of the five
alent to Green’s functions) were computed. As they used linear models, with values ranging from 1 to 2.5 km s−1 , hence covering
shallow-water long-wave equations, they assumed that each tide a good part of the range of plausible velocities for this earthquake
gauge record was a linear combination of subfault waveforms. They (see Section 2.3 hereafter).
obtained one source model for each tested rupture velocity and fi- (iv) All five models were inverted from different data sets of
nally retained the one yielding the largest variance reduction. various origin (seismic, ground deformation and tsunami).
M3: Piatanesi & Lorito (2007) followed a similar approach for
inverting both the slip spatial distribution and the rupture velocity
from 14 tsunami waveforms recorded in the Indian Ocean. They
computed similar Green’s functions as Fujii & Satake (2007), but 2.3 Source parameters used in this study
also took into account the non-linearity of the tsunami propagation The required parameters of the fault geometry in each model are
by computing elementary sources for 10-m slip, considered as the shown in Table 2. The number of subfaults can be lower than in the
mean slip on the whole fault. After this, they inverted the rupture basic model of each study because subfaults with a null inverted
velocity at the same time as the slip distribution, using a non-linear slip were omitted.
inversion technique. The fault geometry in this case was the same The seismic moment corresponding to each model is also re-
as in Banerjee et al. (2005), involving 16 subfaults, just slightly ported. When the chosen rigidity was not specified in the published
different from the fault description by Banerjee et al. (2007). studies, it could be deduced from the given seismic moment and
M4: Rhie et al. (2007) computed a joint inversion of long period the sum of the products of slip on the subfaults and their area. It
seismic waveforms and near-field GPS static offsets. They started must be noted that all authors compare their computed M 0 to seis-
from the fault geometry model by Banerjee et al. (2005) and subdi- mic moments computed from waveforms (Ammon et al. 2005; Park
vided each fault segment so that the total number of subfaults was et al. 2005; Stein & Okal 2005). However, as all models concern
201, significantly higher than in the first three models. the same rupture event and then the same fault zone, it may be
M5: Chlieh et al. (2007) based their study on the work of Subarya interesting to compute their respective seismic moments with the
et al. (2006) who had inverted a coseismic slip distribution from same rigidity, considered as representative for the depth range of
near-field GPS surveys in northern Sumatra and in situ and remote the fault. That is why two seismic moments are indicated: M 0 from
observations of vertical displacements of coral reefs. Chlieh et al. each study, and M ∗0 recomputed with μ = 4.0 × 1010 N m−2 , which
(2007) carried out new inversions after adding near-field GPS data is a value close to the average rigidity in the first 80 km from surface
from the Nicobar-Andaman islands as well as far-field GPS data in the global earth model PREM (Dziewonski & Anderson 1981).
from Thailand and Malaysia. The fault geometry was taken from Through this little calculation, we note that the seismic moment M ∗0
Ammon et al. (2005). The model fault was then subdivided into shows more discrepancy than M 0 . The five studied models actu-
three main segments that were discretized into 661 smaller cells. ally consider more different amounts of released energy than the
displayed moments M 0 suggest.
We chose to test these five source models for the following Only one of the chosen models does not take into account the
reasons. rupture velocity, as it is purely inverted from static offsets (Banerjee


C 2011 The Authors, GJI, 185, 1365–1378
Geophysical Journal International 
C 2011 RAS
1368 B. Poisson, C. Oliveros and R. Pedreros

Table 2. Range of the source parameters used for the five tested models of the Sumatra 2004 earthquake. N s is the
number of subfaults with positive slip; the parameter range for all subfaults is reported for the slip (m), dip (◦ ), rake
(◦ ) and depth (m); V r is the rupture velocity (km s−1 ); M 0 is the seismic moment (N m) computed with the rigidity μ
(N m−2 ) imposed by the authors, and M ∗0 is the seismic moment (N m) recomputed with the same assumed rigidity μ =
4.0 × 1010 N m−2 for all models.
Simulation Model Ns Slip Dip Rake Depth Vr M0 μ M ∗0
S1 M1 10 0.5–19.4 11–35 90–139 0–90 - 7.6 5 6.1
S2 M2 16 1.0–24.6 10 85–130 3–37 1.0 6.0 5 4.8
S3 M3 14 4.0–30.0 11–35 90–139 0–50 2 5.7 3 7.6
S4 M4 200 0.2–35.3 11–18 90–138 0–40 2.5 7.1 4 7.1
S5 M5 661 0.1–17 12–17.5 70–150 2.0–54.7 2.2 6.9 6 4.6

et al. 2007). We thus complete the parameters of S1 by imposing Two simulations were carried out for each model: the first takes
a rupture velocity of 2.5 km s−1 , which corresponds to the most the parameters set with the corresponding non-null rupture velocity

Downloaded from https://academic.oup.com/gji/article/185/3/1365/603962 by guest on 27 April 2021


common value found in the studies of the 2004 Sumatra earthquake (kinematic model M K ), and the second assumes an infinite rupture
(Ammon et al. 2005; Guilbert et al. 2005; Krüger & Ohrnberge velocity (instantaneous model M I ).
2005; Lay et al. 2005; Ni et al. 2005; Lambotte et al. 2006; Vallee
2007). Besides, it approximately corresponds to the ratio of the
fault length (ca. 1200 km) to the rupture duration (ca. 480 s: Ishii 3 R E S U LT S
et al. 2005; Krüger & Ohrnberge 2005; Ni et al. 2005; Tolstoy & The results of our tsunami simulations are investigated through the
Bohnenstiehl 2005; Lambotte et al. 2006). following:
The spatial slip distribution projected on the surface is mapped
on Fig. 1 for the five cases. The location and global length of (i) the sea surface elevation (SSE) profile recorded along the
the fault is similar from one model to another, running along the space–time track of the Jason-1 satellite altimeter over the passage
subduction zone. The slip distribution however appears to differ of the 2004 tsunami; and
markedly between the models. All models display a region of large (ii) zmax, which is the maximum SSE reached at every point.
slip in the south, in front of the northern tip of Sumatra, but this
We cannot compare our results to real tide gauge records or to
region may be located in the deep or shallow part of the fault,
runup measurements because the simulations were run only on a
depending on the considered model. A rather large amount of slip
crude resolution grid so that the tsunami impact at the coast may
affects the middle part of the fault in four models (all but M2), and
well be underestimated. For a correct simulation of SSE time-series
the deep northern part in only one model (M3).
near the coast and inundation heights inland, a nested grid system
A rise time τ (slip duration before seafloor deformation) is taken
integrating detailed topobathymetric data should be used. However
into account in two of the five source models:
the aim of our study is to compare the simulated tsunami through
(i) Fujii & Satake (2007) find that a τ of 3 min for each 100 km × its global behaviour in the northeastern Indian Ocean, so that no
100 km subfault gives the largest variance reduction; focusing was undertaken.
(ii) Chlieh et al. (2007) impose a τ of 20 s to each 16 km ×
20 km subfault;
(iii) Rhie et al. (2007) neglect the rise time because their model 3.1 Overall influence of the seismic source
is based on long-period data; and To investigate the influence of the seismic source, we choose to
(iv) Banerjee et al. (2007) and Piatanesi & Lorito (2007) do not refer to M K simulations.
mention the rise time in their studies, so that we may suppose that
they did not take it into account.
3.1.1 In the deep ocean
In this study, we do not take into account the rise time in any
model. First, this choice is consistent with the three source models 3.1.1.1 Comparison to Jason-1 data. We use the Jason-1 record of
that neglect the rise time; and second, our model computes the the SSHA of cycle 1091 for checking the results of our simulations
tsunami generation on 1 km × 1 km subfaults so that the rise time (Fig. 2). Satellite motion was taken into account by extracting the
is actually negligible. computed SSE on our calculation domain along its ground track,
according to the measurement time sequence that spanned around
8.4 min.
2.4 Computational setting Some authors consider that the tsunami signal was better de-
scribed by the difference between the concerned cycle and the pre-
All simulations are carried out on a 1122 × 944 cartesian grid in ceding one (cycle 108, 10 d earlier), assuming that other oceano-
Mercator projection, covering the geographic extent [79 to 100◦ E, graphic effects than the tsunami are thus removed (Fujii & Satake
–5.5 to 19.5◦ N] with a pixel size of 2500 m. The bathymetric grid 2007). Gower (2006) chose to subtract the smoothed average of
is constructed from the ETOPO-2 data set (U.S. Department of heights observed on both cycles before and after cycle 109 instead.
Commerce & Atmospheric Administration 2006). The tsunami is Other authors used a specific ocean variability mapping technique
simulated during 2.5 real-time hours, so that the impact of the first
waves on the south and the east coasts of Sri Lanka can be estimated.
A 10-pixel-wide sponge-layer is imposed on each of the four grid 1ftp://podaac.jpl.nasa.gov / pub / sea_surface_height / jason / j1_ssha / data/
edges, where absorbing boundary conditions are imposed. c109/


C 2011 The Authors, GJI, 185, 1365–1378

Geophysical Journal International 


C 2011 RAS
A best source for Sumatra 2004 tsunami? 1369

Downloaded from https://academic.oup.com/gji/article/185/3/1365/603962 by guest on 27 April 2021


Figure 2. Map of SSE simulated at the beginning of the Jason-1 passage from south to north over the tsunami (around 2:54 UTC).

to distinguish the tsunami signal from the large-scale and mesoscale The first (and southernmost) part of Jason-1 data comprises the
ocean variability (Ablain et al. 2006). Here, we choose to consider largest and clearest crest wave (A). This most significant wave
directly the SSHA deduced from Jason-1 measurements (Fig. 3), corresponds to the front wave of the tsunami (Fig. 2), which also
for two reasons: furthermore results from the southern part of the fault where the
slip is the highest in most models (Fig. 1).
(i) a mean sea surface level is already taken into account in the All simulations but S5 lead to fit the presence of this high first
computation of the SSHA (GSFC00.1 model; Koblinsky et al. 1998) crest wave. Its amplitude in data and in simulation results is reported
as it is subtracted from the satellite measured SSH at the same time in Table 3. The half wavelength of this crest is 200 km on Jason-
as atmospheric and tide effects; 1 data and appears to be around 150–200 km in S2 and S4, its
(ii) the tsunami signal obtained in the cited studies is very similar amplitude being fitted by 80–100 per cent in these simulations.
to the SSHA on the main part of the record, where the two main The first wave simulated in S1 and S3 is of smaller amplitude and
waves are observed (parts A, B and C on Fig. 3, Ablain et al. 2006; wavelength.
Gower 2006). In the northern 2000 km of the profile (part D on The second peak is not well reproduced by any model, as it is
Fig. 3), the SSHA appears to be at worst 5–10 cm higher than the not of the same shape as the data (S1, S3), or is too small (S4) or
processed ones, where most models anyway have inadequate results missing (S2).
[see e.g. Sladen & Hébert (2008), for use of Ablain et al. (2006)’s From a temporal point of view, the first peak is matched in S1,
processed SSHA]. S3 and S4, but is late in S2.
When looking at snapshots of the tsunami propagation at the time In S5, the first crest appears to have a half wavelength of 400 km
of Jason-1’s passage (Fig. 2), wave patterns along this track seem to that is longer than the two first peaks together in the data. Besides,
be somewhat different according to the simulated source. The SSE its amplitude does not exceed a third of the observed crests.
profiles of the M K simulated tsunamis along the Jason-1 track are The following trough (B) is rather well reproduced in S4,
shown on Fig. 3 as red curves. but is badly reproduced in the other simulations. The rise of an


C 2011 The Authors, GJI, 185, 1365–1378
Geophysical Journal International 
C 2011 RAS
1370 B. Poisson, C. Oliveros and R. Pedreros

Downloaded from https://academic.oup.com/gji/article/185/3/1365/603962 by guest on 27 April 2021

Figure 3. Sea surface elevation profiles along the Jason-1 track, for all tested models. Black points are Jason-1 data, red profiles are for M K and blue ones for
M I simulations. Grey letters designate profile segments that are referred to in the text.

intermediate small crest is also visible in S1 and matches the data of this variability, most of the simulations results in this segment
in S4. are in the same range as the data. However, the best average fit is
The second rise (C) involves a high frequency signal with 0.2 obtained for S4, as the others show anomalous trends northwards,
m amplitude, the SSHA then extending from 0 to 0.4 m. Because with regard to the data.


C 2011 The Authors, GJI, 185, 1365–1378

Geophysical Journal International 


C 2011 RAS
A best source for Sumatra 2004 tsunami? 1371

Table 3. Comparison between: observed and simulated amplitude of the All models, except S5, show a strong impact on the northern tip
first crest wave on Jason-1 track profile; simulated zmax in the most impacted of Sumatra. Actually, the observed runups in the Aceh province are
places and maximum observed tsunami heights (in metres). the highest reported in the observations, reaching 20–30 m (Borrero
Data S1 S2 S3 S4 S5 et al. 2006; Jaffe et al. 2006).
Eastwards, the coast of Thailand is impacted in all simulations
Amplitude of first wave 0.59 0.42 0.59 0.34 0.46 0.20
on J-1 profile
but S2. The strongest impact there is computed in S1, where the
Indonesia (Aceh) 20–30 9.8 5.7 8.7 13 6.9 computed tsunami height reaches 5–6 m at some points, whereas
Thailand 5–15 6.7 3.6 6.4 6.2 6.3 the observations show that the highest tsunami heights ranged from
Myanmar (south) 2–3 2.6 1.0 3.3 1.9 2.9 5 to 15 m (Tsuji et al. 2006).
Myanmar (Ayeyarwaddy) 1.5–2 1.8 0.8 3.8 1.3 1.9 Northwards, on the Myanmar coast, the tsunami’s impact differs
India (Tamil Nadu) 3–5 3.4 1.7 3.5 3.4 3.1 according to the source model. In S2, no impact is observed, whereas
Sri Lanka (east) 5–7 7.1 2.0 4.9 5.5 5.2 in S3 the whole coast is hit by 3–4 m high waves. In the other cases,
the affected zones are the southern coastal areas that were actually
reached by 2–3 m high waves, and the Ayeyarwaddy Delta (northeast
The last part of profile (D) shows three successive bulges of
of our computation domain), where the observed tsunami heights
0.2–0.3 m amplitude that are not artefacts because they can also be
reached 1.5–2 m (Satake et al. 2006; Swe et al. 2006).

Downloaded from https://academic.oup.com/gji/article/185/3/1365/603962 by guest on 27 April 2021


observed on variously processed SSHA profiles (Ablain et al. 2006;
The impact of the simulated tsunamis on the southeastern coast
Gower 2006). None of our five simulations correctly reproduce these
of India is very similar for all simulations but S2. In the latter case,
bulges. However, the initiation of the rises is reasonably matched
zmax is almost everywhere less than 1 m high whereas the highest
on the S4 profile, and the global trend on D is roughly computed by
observed runups in this zone range mainly from 3 to 6 m (Sheth
S1, S3 and S4.
et al. 2006; Yeh et al. 2006). Apart from S2, the other simulations
It should be noted that referring to the sea level anomaly pro-
yield tsunami amplitudes of around 3 m.
cessed by Ablain et al. (2006) or Gower (2006) would lead to a
As for the Indian coast, the eastern coast of Sri Lanka is slightly
similar comparison with our simulation results.
impacted by the S2 tsunami, while it is almost twice as severely
3.1.1.2 Main zmax patterns. The zmax maps are shown on Fig. 4.
hit in the other simulations. Besides, S1 and S4 lead to an impact
Given the average direction of the fault, the mid-ocean zmax patterns
around 1 m higher than S3 and S5. The results on the coast of Sri
mainly extend west- and eastwards from the fault.
Lanka will be examined in detail hereafter.
As could be expected, the zmax patterns are strongly related to
the spatial slip distribution in both location and depth. Thus, the
trails of the highest tsunami waves start from the subfaults with
3.2 Overall influence of the rupture velocity
high slip, except where they are too deep. Then, the direction of
the high zmax areas varies notably from one simulation to another.
3.2.1 Comparison to Jason-1 data
In S1, S4 and S5, a high slip on the middle segments of the fault
leads to a high western zmax trail pointing to the south of Sri Lanka. M K and M I profiles from S1 to S4 are shifted about 20 to 75 km
Because of differences in the detailed slip distribution, the highest (Fig. 3). Taking into account the tsunami propagation speed of
waves directly hit Sri Lanka in S5, while they pass to the south in around 800 km h−1 at this depth, this shift corresponds to a propa-
S1 and S4. In the latter cases, however, the larger width of the high gation time lag of a few minutes (1.5–5.6 min). Considering the first
slip zone leads to a wider area of high zmax at sea, and the waves wave (A), the rise in M I is slightly in advance compared to the data.
that arrive on the coasts of southern India and Sri Lanka are of a M K simulations show a better temporal agreement with the data,
similar height. The slip distribution in S3 is longitudinally more except in S2 and S5. In S2, the first crest with M K is shifted about
homogeneous, and the resulting waves westward generate extended 40 km as compared with the data, which corresponds to a 3-min
areas of moderate zmax; in this case, the highest waves home in on delay or the rise time taken into account by Fujii & Satake (2007).
northern Sri Lanka. The zmax map for S2 appears to be markedly In S5, the first wave appears to be much spread and its amplitude is
distinct from the others, as the high slip only concentrates in the lower than the data by half.
south of the fault. Thus, the high waves go through the southern Besides the segment A, the other parts of the profiles are gen-
part of the domain only, far to the south of Sri Lanka. erally shifted following the same trend. However this is not a pure
East of the fault, high waves propagate from the middle segments translation, especially in the case of S5, where the M I and M K
in S1, S4 and S5. Again, the more homogeneous slip in S3 generates profiles do not look alike at all.
rather high waves to the east, until the northern part of the fault,
whereas the smaller slip in S2 strongly restricts the tsunami impact
eastwards. 3.2.2 Maximal SSE results
The difference between zmax computed by M K and M I simulations
(δ zmax) is mapped on Fig. 5. The difference is more important just
3.1.2 Coastal regions
above the fault, especially where the slip is high and not too deep;
To assess the tsunami impact at the coast, we need to refer to δ zmax is clearly positive on trails rising westwards from a more
some tsunami data, so we refer to tsunami height measurements or less wide northern part of the fault. These trails are due to the
(Table 3). As they mostly consist of surveyed runups, we have to northward propagation of rupture in M K simulations, so that they
keep in mind that computed zmax may be compared to runup values do not exhibit the same patterns depending on the slip distribution
only qualitatively, in terms of more or less impacted regions. of each case. However, all cases display δ zmax trails over 0.1 m
At first sight, the computed tsunami impact on the coasts depends crossing the deep ocean, where the front (and highest) wave is
upon the simulation in accordance with the different mid-ocean around 0.6–0.7 m high only. The ratio of δ zmax to zmax on this
zmax patterns. mid-ocean trail then exceeds 10 per cent for S4, 20 per cent for S1


C 2011 The Authors, GJI, 185, 1365–1378
Geophysical Journal International 
C 2011 RAS
1372 B. Poisson, C. Oliveros and R. Pedreros

Downloaded from https://academic.oup.com/gji/article/185/3/1365/603962 by guest on 27 April 2021


Figure 4. Maximum sea surface elevation (in metres) computed during the simulated Sumatra 2004 tsunami generation and propagation, for the five tested
earthquake source parameter sets through the kinematic model. Pink diamonds locate places where zmax was extracted from results and reported in Table 3.

and S3, reaching 50 per cent in S2 and S5 where zmax appears to Poisson et al. (2009) showed that the zmax computation on a
be quite low. 1600-m resolution grid around Sri Lanka represented on average 88
In S1, S4 and S5, δ zmax is positive towards Sri Lanka and per cent of actual tsunami heights with a standard deviation of 30
southern India (0.1–1 m), a slightly negative area being observed per cent. This leads to examine the zmax results on the Sri Lankan
further south. In S2, δ zmax is positive almost everywhere west of coast not only qualitatively but also quantitatively, as they should
the fault. In S3, δ zmax appears to be positive only towards India, not be much smaller than the actual tsunami heights.
north of Sri Lanka. East of the fault, δ zmax is lower but reaches
0.5 m in absolute value at some places. It is rather negative on the
coast of Thailand, and positive along Myanmar. On the northern tip 3.3.1 Influence of the seismic source
of Sumatra, δ zmax is slightly negative in S1, S4 and S5, except
The zmax maps obtained with the M K model on the eastern coast of
in some points of the coast. In S2 and S3, it is more heteroge-
Sri Lanka show a clear discrepancy between the outputs of the five
neous, and, when compared to zmax (8–10 m), can be considered as
simulations (Fig. 6). Letters a to f help to distinguish areas with
negligible.
distinct observations.
On the whole, S1 exhibits a much higher zmax at all annotated
points, whereas S2 leads to the lowest zmax values. At first sight, S3,
3.3 Focus on Sri Lanka
S4 and S5 produce the same zmax patterns intermediate between
As Sri Lanka lay squarely in the tsunami path, it was severely hit those of S1 and S2, but they differ in detail. S4 gives some higher
by the tsunami waves. For this reason, it is interesting to analyse estimations of zmax, especially in the southern part of the area
the simulated tsunami impact in this zone. Here we consider the where they are 1–2 m higher than in S3 and S5 (‘e’ and ‘f’). S3 and
results of the simulations on the eastern coast of Sri Lanka only, S5 differ locally, but with a discrepancy lower than 1 m. Almost
because of the presence of the western absorbing boundary of the everywhere along the coast, the S1 results are 2–3 m higher than
computational grid near the western coast. the S3 and S5 ones, whereas the S2 results are 2–3 m lower.


C 2011 The Authors, GJI, 185, 1365–1378

Geophysical Journal International 


C 2011 RAS
A best source for Sumatra 2004 tsunami? 1373

Downloaded from https://academic.oup.com/gji/article/185/3/1365/603962 by guest on 27 April 2021


Figure 5. Difference in maximum sea surface elevation (in metres) between the kinematic and instantaneous models, applied to the five tested parameters
sets.

Figure 6. Maximum sea surface elevation (in metres) computed on the eastern coast of Sri Lanka through the kinematic model with the five tested parameter
sets. The letters refer to areas mentioned in the text.


C 2011 The Authors, GJI, 185, 1365–1378
Geophysical Journal International 
C 2011 RAS
1374 B. Poisson, C. Oliveros and R. Pedreros

Downloaded from https://academic.oup.com/gji/article/185/3/1365/603962 by guest on 27 April 2021


Figure 7. Difference in maximum sea surface elevation (in metres) between the kinematic and instantaneous models, obtained along the eastern coast of Sri
Lanka with the five tested parameter sets. The letters refer to areas mentioned in the text.

To make a quantitative assessment of the simulations results, we the models lies in the spatial slip distribution. As the patches of
compare zmax at the coastline of Sri Lanka to published tsunami high slip are not located on the same fault segments, the high zmax
height data (Liu et al. 2005; Wijetunge 2006). More than 70 per patterns do not appear at the same places in all simulations.
cent of the measured tsunami heights on the eastern coast were The comparison of simulation results with the Jason-1 record in
comprised in the 3–6 m range. On the southern coast, the same ratio the deep ocean allows a first analysis of the discrepancy between
of reported measurements reached 5–10 m. models (Fig. 3). The most remarkable point is the first wave de-
The tsunami heights on the eastern coast were mostly comprised tected on the Jason-1 track. The fit with the first crest seems to be
in the 4–6 m range. On the southern coast, they rather reached closely related to the presence of a high slip patch in a not too deep
5–10 m. The two simulations that appear to be the more consistent southern segment of the fault (as in M2 and M4). If the high slip
with these data are S1 and S4, which lead to zmax ranging from is too moderate (M1) or too deep (M3), the first generated tsunami
4 to 7 m along the southeastern coast, whereas the others clearly wave appears to be notably smaller than the observed one, both in
underestimate the tsunami impact in this region. wavelength and amplitude.
However the discrepancy between models is only partly visible
through the comparison with Jason-1 data. For example, the S1 and
3.3.2 Influence of the rupture velocity S3 profiles along the Jason-1 track look alike but lead to clearly
The influence of the rupture velocity is illustrated in Fig. 7 which distinct zmax maps (Fig. 4). At the ocean scale, zmax patterns are
represents δ zmax in the same manner as Fig. 5. also closely related to the spatial slip distribution. When looking
δ zmax appears to be higher in the case of the S1, where zmax at the coastal impact of the simulated tsunami on the Sri Lanka,
was also higher than in the other cases. It reaches 1–2 m on the main however, the differences between simulations are less expectable.
part of the eastern coastline in S1, but is less significant in the other
cases, where a 1-m discrepancy appears in some spots (c and e in S3
4.2 Fault models involving tsunami modelling
to S5, f in S3 and d in S5). Taking into account the range of zmax,
the ratio of δ zmax to zmax amounts to 20–30 per cent at points Among the five tested earthquake source models, three are inverted
c, e and f in S3, as in S2. The S5 δ zmax map distinguishes itself or validated by their authors through tsunami modelling:
from the others as it shows a 0.5–1 m discrepancy over a more than
50-km-wide area before reaching the coast from a to e. However, M2: Fujii & Satake (2007) inverted the source with a linear
the resulting δ zmax at the coastline is lower than in the case of S1. tsunami model (Satake 1995);
The only case where δ zmax appears to be slightly negative M3: Piatanesi & Lorito (2007) used a NLSWE tsunami model for
(meaning that the M K model underestimates zmax in comparison computing elementary tsunami waveforms that were then linearly
to M I ) is S3, over a 150-km-long area in the middle of the coast. combined to invert the seismic source; and
M5: Chlieh et al. (2007) checked their slip model by simulating
the tsunami with a model based on a shallow-water approximation,
4 DISCUSSION without specifying if it is linear or not (both are referred: Hbert
et al. 2001; Sladen & Hbert 2005).
Here, we discuss the potential reasons for the observed discrepancy
between the different models, and between the model results and These three tsunami models not only differed among each other,
the tsunami data. but were also distinct from ours.

4.1 Influence of slip distribution 4.2.1 Tsunami generation


As could be expected when testing five distinct fault models, the Tsunami generation relies on Okada (1985)’s equations for all mod-
simulated tsunamis are distinct as well. The main difference between els. Fujii & Satake (2007) only added a correction due to the effect


C 2011 The Authors, GJI, 185, 1365–1378

Geophysical Journal International 


C 2011 RAS
A best source for Sumatra 2004 tsunami? 1375

of coseismic horizontal displacement in region of steep bathymetric diffracted waves along the southwestern coast. Specific configura-
slopes (Tanioka & Satake 1996), but the calculation of this correc- tions at these particular places may have induced site effects that
tion shows that it induces a change in the seafloor deformation of in turn might have generated specific tsunami characteristics that
3 per cent on average. Thus, the computation of the tsunami gen- would have biased the global behaviour.
eration cannot explain the observed divergences between the tested Moreover, some of the referred tide gauge stations are common
models and ours. to both, but the corresponding waveforms seem to differ from one
description to the other. We may presume that it is due to the re-
moving of the tidal component in the tsunami signals in M2, as M3
4.2.2 Frequency dispersion does not seem to consider the ocean tide effect. In any case, the
difference in the tide gauge data used by M2 and M3 compounds
The frequency dispersion appears to be obvious in Jason-1 data, as the differences of the tsunami models; M2 and M3 necessarily lead
high-frequency waves are clearly visible (see Figs 2 and 3, Kulikov to different slip distributions, and S2 and S3 are then expected to
et al. 2005). This observation confirms the importance of taking lead to different results.
this phenomenon into account in tsunami modelling, as it is done
in FUNWAVE. We observe that only S2 and S4 reproduce a high-
frequency component that is fairly similar in frequency to Jason-1 4.3 Seismic moments

Downloaded from https://academic.oup.com/gji/article/185/3/1365/603962 by guest on 27 April 2021


data (Fig. 3). This may be due to the higher spatial heterogeneity
The zmax patterns, on general maps as well as on our Sri Lanka
of the corresponding slip distributions. The other simulations lead
focus, show a restricted area affected by the tsunami impact in S2
to smoother SSHA profiles, probably because of their lower slip
and S5 compared to the other cases. This may be caused by the
gradients.
smaller seismic moments of the corresponding models (Table 2).
For the 2004 December earthquake, the seismic moment that is
most often obtained from seismic data amounts to 6.5 × 1022 N m
4.2.3 Linearity (Ammon et al. 2005; Banerjee et al. 2005; Park et al. 2005) or
We observe that both S2 and S3 involve an assumption of linearity more (Ishii et al. 2005; Stein & Okal 2005). A seismic moment
in their propagation model, either inside the equations, or through close to that value was computed by Fujii & Satake (2007) and
the way of inverting data. However, these models are both based Chlieh et al. (2007) with a higher rigidity than the value used
on inversion of tide gauge data (plus satellite data for S2), and tide here. However imposing the same rigidity in the computation of all
gauges are generally located in harbours or elsewhere at shallow seismic moments is justified, as all the models aim at describing
depth, in other words in places where the tsunami signal is highly the same fault at the same depth range. As an example, the slip
non-linear (Satake 1995; Zahibo et al. 2006). distribution obtained by Piatanesi & Lorito (2007) is exactly in
Concerning M2, Fujii & Satake (2007) also lead an inversion of the same depth range as the one inverted by Chlieh et al. (2007)
the source from tide gauge data only and one from satellite data (Table 2). However, they computed near seismic moments only
only. Their results show that the joint inversion of both tide gauge because they assumed very different rigidity values, respectively,
and satellite data (their ‘best model’ that we use here) leads to 3 and 6 × 1022 N m. Therefore, their slip distributions correspond
almost exactly the same results as the inversion of tide gauge data actually to clearly distinct amounts of released energy, and even
only. This means that the inverted slip distribution depends almost with a higher rigidity for all the seismic moments would remain
entirely on tide gauge records, confirming that this is questionable discrepant.
because of the linearity of the tsunami model used. On the contrary,
their inversion of satellite data only is likely more relevant as it
is consistent with the linear behaviour of the tsunami propagation 4.4 Validation of models
through the deep ocean. Among the five models considered here, the validity is treated as
follows.
M1: Banerjee et al. (2007) compared the inverted slip distribu-
4.2.4 Tide gauge data sets
tion to the results of other published inverted slip distributions.
In addition to the non-linearity question, sea level records at tide M2: Fujii & Satake (2007) compared the synthetic tsunami
gauge stations can be affected by the very local context around waveforms and SSHA profiles to the data that had been used in
them, so that numerical modelling on a grid of 1.8 km (S3) or even the inversion.
0.75 km (S2) spatial resolution may lead to incorrect results if the M3: Piatanesi & Lorito (2007) assessed the uncertainty of their
surroundings of the tide gauges are too roughly represented. To model by comparing the synthetic tsunami waveforms to the ob-
prevent local effects from affecting their inversion, Fujii & Satake served one that had been inverted, first on a synthetic data set and
(2007) and Piatanesi & Lorito (2007) used only the first oscillations then on the Sumatra 2004 data.
in each tide gauge record. This method tends to restrict the inver- M4: Rhie et al. (2007) first conducted tests of sensitivity to the
sion problem to the first wave impact, so that it is not surprising fault geometry and to the rupture velocity. After that, they carried
that the next waves appear to be poorly reproduced on the SSHA out an error analysis through the repeated inversion on a random
profile. selection of 50 per cent of the seismic and GPS stations.
The same effect arises from the rather small number of tide gauge M5: Chlieh et al. (2007)’s study is the only one involving a check
records used in these models (12 in M2 and 14 in M3). Consider- of the source model with other data than inverted ones. They exam-
ing the length and complexity of such a fault, more data would be ined the consistency of their model with the latitudinal variations
necessary to avoid over simplification of the resulting slip distribu- of the seismic moment and the energy radiated by T waves, with
tion. For instance, half of the tide gauges in India were located on the normal modes, with the amplitude of very long-period surface
the direct path of the tsunami, whereas the others were only hit by waves and with satellite-measured SSHA profiles.


C 2011 The Authors, GJI, 185, 1365–1378
Geophysical Journal International 
C 2011 RAS
1376 B. Poisson, C. Oliveros and R. Pedreros

All models but M5 lack a comparison with data that were not of each subfault. This difference may have induced a difference
taken into account in the inversion. Concerning the validation of between Piatanesi & Lorito (2007)’s results and our S3 results.
M5 on Jason-1 data, the first crest was already not reproduced in Hirata (2009) used the same approach as Piatanesi & Lorito
their tsunami simulation. However, the discrepancy with data was (2007) for demonstrating the importance of taking the rupture ve-
not so significant that here in S5 (M I , to conform to the choice in M5 locity into account in tsunami modelling. Our study, by refining the
not to take into account the rupture kinematics), as the amplitude discretization and by testing several slip distribution models, con-
of the second crest and the timing appeared to be consistent. This firms that the impact on Sri Lanka and eastern India can be notably
may be due to the difference of the tsunami models used. underestimated by neglecting the rupture kinematics.

4.5 Best models 5 C O N C LU S I O N

In our study, the simulations that best reproduce the SSHA profile The 2004 December Sumatra tsunami was simulated with the
and the impact on Sri Lanka are S4 followed by S1. The related tsunami code Geowave, as induced by five different published mod-
models were not actually validated on independent data, but both els of the seismic source and with or without taking into account
were based on the inversion of only geodetic and/or seismic data. the rupture velocity.

Downloaded from https://academic.oup.com/gji/article/185/3/1365/603962 by guest on 27 April 2021


As they did not take into account any actual tsunami data, it may ap- As expected, different slip distributions and different seismic
pear paradoxical that they fit tsunami observations so well through moments result in different zmax patterns, both when propagating
our simulations. In fact, tsunami data may still be less suitable for across the ocean and at the arrival near the coast, in the far field
inversion of a fault source model than geodetic and seismic data as well as in the near field. In most cases, the comparison with the
that have been studied for a long time. Models based on tsunami Jason-1 record leads to only a rough consistency. The shape and
data inversion commonly transpose the Green’s functions method amplitude of the waves are generally not reproduced by a same
from seismic/geodetic inversion, which is inappropriate to tsunami model at the same time, and in some cases both appear inconsistent.
inversion because of the linearity assumption. In fact, many mod- Some of the tested source models involved tsunami modelling. In
els refer to coastal tsunami data (runup measurements, tide-gauge our simulations, however, these models lead to some disagreement
records) although they cannot compute the tsunami behaviour in with Jason-1 data and zmax patterns. In some cases, the discrepancy
coastal areas because of the propagation model used or because of a was already notable in the original study, but in other cases it must be
lack in bathymetric data of high resolution. However, the inversion attributed to the tsunami models used. Contrary to other models used
of deep-ocean tsunami measurements is much easier to carry out in the concerned studies, Geowave uses fully non-linear Boussinesq
(e.g. Ritsema et al. 1995) and should enhance the development of equations and takes into account the frequency dispersion, which
fault source inversion from tsunami data as the DART (Deep-ocean may induce significant changes when compared with linear models,
Assessment and Reporting of Tsunamis) network grows up. especially near the coast.
Though M4 was not derived from satellite data and was com- Our simulation results show that tsunami modelling is very sen-
pletely disconnected from GEOWAVE, the agreement of the S4 sitive to spatial slip distribution. They also confirm that the 2004
synthetic SSHA profile with Jason-1 data tends to confirm the suit- December tsunami was a particularly complex event. Even, it was
ability of GEOWAVE for the correct modelling of tsunami genera- probably so complex that any detailed study intending to invert the
tion and propagation. If it was inadequate, it should not give such source from a specific data set almost certainly will not be con-
consistent results with any source of only seismic/geodetic origin. sistent with all tsunami observations. We observe that a simplified
Model M5 stands apart from the others. It takes into account approach, like the Grilli et al. (2007)’s model, gives just as correct
several distinct geodetic data sets, but S5 leads to a clear mismatch results as more complex models, though it describes the fault by
between synthetic and observed SSHA profiles. Chlieh et al. (2007) only five segments. This model would maybe not fit seismic and/or
mentioned that their best model underestimated the near-field dis- uplift data, but it leads to good consistency with tsunami data and
placements, likely because the data set used comprised post-seismic could be very useful for tsunami hazard studies.
displacement. The resulting slip distribution shows a gap of high Among the five tested models, the seismic/geodetic inversion by
slip in the southern part of the fault, implying an underestimated Rhie et al. (2007) leads to the best agreement with the Jason-1
and deformed first tsunami wave. SSHA profile and with the zmax patterns, especially those around
Sri Lanka. As it is independent of tsunami modelling, the observed
agreement supports the validity of GEOWAVE tsunami model.
Concerning the rupture kinematics, we observe that, when assum-
4.6 Rupture velocity ing an infinite rupture velocity, the simulation results are modified
Fujii & Satake (2007) discussed the influence of rupture velocity, with a systematic bias.
and concluded that it is negligible on the slip distribution but im- (i) The wavefront is a few minutes in advance against the sim-
portant on the tsunami impact on the coasts of eastern India and Sri ulations with finite rupture velocity. The time lag appears at the
Lanka. This agrees with our results, showing a significant modifi- tsunami generation, and seems to increase from mid-ocean to the
cation of zmax in these areas from M I to M K . coast.
Piatanesi & Lorito (2007) emphasized the importance of rupture (ii) The tsunami amplitude is notably underestimated (up to
velocity, as their inversion of tsunami data clearly determined a 50 per cent), especially along the Sri Lanka coastline.
strictly positive velocity. Nevertheless, this velocity was not taken
into account in the same manner as proposed by us, as they consid- We conclude that the rupture velocity, like the slip distribution,
ered the rupture velocity by considering a temporal gap between sub- may have a significant influence on the tsunami impact on the coast.
faults so that the rupture is instantaneous on each subfault, whereas Thus, when simulating tsunamis to feed a forecast database for a
we impose a finer kinematic rupture through further subsampling tsunami warning system, it should be essential to take into account a


C 2011 The Authors, GJI, 185, 1365–1378

Geophysical Journal International 


C 2011 RAS
A best source for Sumatra 2004 tsunami? 1377

non-infinite rupture velocity and the potential spatial heterogeneity Hbert, H., Heinrich, P., Schindel, F. & Piatanesi, A., 2001. Far-field sim-
of slip through several slip distributions. In fact, these parameters ulation of tsunami propagation in the Pacific Ocean: impact on the
could modify significantly the estimation of tsunami potential max- Marquesas Islands (French Polynesia), J. geophys. Res., 106, 9161–
imal amplitudes and arrival times. 9177.
Hirata, K., 2009. Tsunami amplification along the eastern coast of In-
dia and Sri Lanka due to earthquake rupture propagation in the
Sumatra–Nicobar–Andaman trenches, J. earthq. Tsunami, 3, 67–75.
AC K N OW L E D G M E N T S Hirata, K., Satake, K., Tanioka, Y., Kuragano, T., Hasegawa, Y., Hayashi, Y.
This study was funded by BRGM research projects. Philip Watts & Hamada, N., 2006. The 2004 Indian Ocean tsunami: tsunami source
model from satellite altimetry, Earth Planets Space, 58, 195–201.
(Applied Fluids Engineering Inc.) provided the initial GEOWAVE
Ioualalen, M., Asavanant, J., Kaewbanjak, N., Grilli, S.T., Kirby, J.T.
code version that we modified and used in this work. The authors
& Watts, P., 2007. Modeling the 26 December 2004 Indian Ocean
want to thank Mohamed Chlieh who kindly provided M5 slip dis- tsunami: case study of impact in Thailand, J. geophys. Res., 112, C07024,
tribution details. The authors are very grateful to John Douglas for doi:10.1029/2006JC003850.
improving the English of the manuscript and to three anonymous Ishii, M., Shearer, P.M., Houston, H. & Vidale, J.E., 2005. Extent, duration
reviewers for their helpful comments and suggestions. and speed of the 2004 Sumatra–Andaman earthquake imaged by the
Hi–Net array, Nature, 435, 933–936.

Downloaded from https://academic.oup.com/gji/article/185/3/1365/603962 by guest on 27 April 2021


Jaffe, B.E. et al., 2006. Northwest Sumatra and offshore islands field survey
REFERENCES after the December 2004 Indian Ocean tsunami, Earthquake Spectra, 22,
S105–S135.
Ablain, M., Dorandeu, J., Le Traon, P.Y. & Sladen, A., 2006. High resolution Koblinsky, C.J., Ray, R., Beckley, B.D., Wang, Y.M., Tsaoussi, L., Brenner,
altimetry reveals new characteristics of the December 2004 Indian Ocean A. & Williamson, R., 1998. NASA ocean altimeter Pathfinder project
tsunami, Geophys. Res. Lett., 33, L21602, doi:10.1029/2006GL027533. report 1: data processing handbook, NASA Tech. Memo. 1998-208605,
Ammon, C.J. et al., 2005. Rupture process of the 2004 Sumatra–Andaman NASA, Goddard Space Flight Center, Greenbelt, MD.
earthquake, Science, 308, 1133–1139. Krüger, F. & Ohrnberge, M., 2005. Tracking the rupture of the Mw 5 9.3
Banerjee, P., Pollitz, F.F. & Bürgmann, R., 2005. The size and duration of Sumatra earthquake over 1,150 km at teleseismic distance, Nature, 435,
the Sumatra–Andaman Earthquake from Far–Field Static Offsets, Science, 937–939.
308, 1769–1772. Kulikov, E.A., Medvedev, P.P. & Lappo, S.S., 2005. Satellite recording of
Banerjee, P., Pollitz, F., Nagarajan, B. & Bürgmann, R., 2007. Coseismic the Indian Ocean tsunami on December 26, 2004, Doklady Earth Sciences
slip distributions of the 26 December 2004 Sumatra–Andaman and 28 A, 401, 444–448.
March 2005 Nias earthquakes from GPS static offsets, Bull. seism. Soc. Lambotte, S., Rivera, L. & Hinderer, J., 2006. Rupture length and du-
Am., 97(1A), S86–S102. ration of the 2004 Aceh–Sumatra earthquake from the phases of the
Borrero, J.C., Synolakis, C.E. & Fritz, H., 2006. Northern Sumatra field Earth’s gravest free oscillations, Geophys. Res. Lett., 33, L03307,
survey after the December 2004 Great Sumatra earthquake and Indian doi:10.1029/2005GL024090.
Ocean tsunami, Earthquake Spectra, 22, S93–S104. Lambotte, S., Rivera, L. & Hinderer, J., 2007. Constraining the overall
Chanson, H., 2005. Le Tsunami du 26 décembre 2004: un phénoméne kinematics of the 2004 Sumatra and the 2005 Nias earthquakes us-
hydraulique d’ampleur internationale. Premiers constats, La Houille ing the Earth’s gravest free oscillations, Bull. seism. Soc. Am., 97(1A),
Blanche, 2, 25–32 (in French). S128–S138.
Chlieh, M. et al., 2007. Coseismic slip and afterslip of the Great Mw 9.15 Lay, T. et al., 2005. The Great Sumatra–Andaman earthquake of 26 Decem-
Sumatra–Andaman earthquake of 2004, Bull. seism. Soc. Am., 97(1A), ber 2004, Science, 308, 1127–1133.
S152–S173. Liu, P.L.F. et al., 2005. Observations by the International Tsunami Survey
Choi, B.H., Hong, S.J. & Pelinovsky, E., 2006. Distribution of runup heights Team in Sri Lanka, Science, 308, 1595, doi:10.1126/science.1110730.
of the December 26, 2004 tsunami in the Indian Ocean, Geophys. Res. Narayan, J.P., Sharma, M.L. & Maheshwari, B.K., 2005. Run–up and inun-
Lett., 33, L13601, doi:10.1029/2006GL025867. dation pattern developed during the Indian Ocean tsunami of December
Day, S.J., Watts, P., Grilli, S.T. & Kirby, J.T., 2005. Mechanical models of 26, 2004 along the coast of Tamil Nadu (India), Gondwana Res., 8, 611–
the 1975 Kalapana, Hawaii earthquake and tsunami, Mar. Geol., 215, 59– 616.
92. Ni, S., Kanamori, H. & Helmberger, D., 2005. Energy radiation from the
Dziewonski, A.M. & Anderson, D.L., 1981. Preliminary reference Earth Sumatra earthquake, Nature, 434, 582, doi:10.1038/434582a.
model, Phys. Earth planet. Inter., 25, 297–356. Okada, Y., 1985. Surface deformation due to shear and tensile faults in a
Fujii, Y. & Satake, K., 2007. Tsunami source of the 2004 Sumatra–Andaman half–space, Bull. seism. Soc. Am., 75, 1135–1154.
earthquake inferred from tide gauge and satellite data, Bull. seism. Soc. Park, J. et al., 2005. Earth’s free oscillations excited by the 26 December
Am., 97(1A), S192–S207. 2004 Sumatra–Andaman earthquake, Science, 308, 1139–1144.
Goff, J. et al., 2006. Sri Lanka field survey after the December 2004 Indian Piatanesi, A. & Lorito, S., 2007. Rupture process of the 2004
Ocean tsunami, Earthquake Spectra, 22, S155–S172. Sumatra–Andaman earthquake from tsunami waveform inversion, Bull.
Gower, J., 2005. Jason 1 detects the 26 December 2004 tsunami, EOS, Trans. seism. Soc. Am., 97(1A), S223–S231.
Am. geophys. Un., 86, 37–38. Pietrzak, J. et al., 2007. Defining the source region of the Indian Ocean
Gower, J., 2006. The 26 December 2004 tsunami measured by satellite tsunami from GPS, altimeters, tide gauges and tsunami models, Earth
altimetry, Int. J. Remote Sens., 28, 2897–2913. planet. Sci. Lett., 261, 49–64.
Grilli, S.T., Ioualalen, M., Asavanant, J., Shi, F., Kirby, J. & Watts, P., 2007. Poisson, B., Garcin, M. & Pedreros, R., 2009. The 2004 December 26 Indian
Source constraints and model simulation of the December 26, 2004 Indian Ocean tsunami impact on Sri Lanka: cascade modelling from ocean to
Ocean tsunami, J. Waterway Port Coast. Ocean Eng., 133, 414–428. city scales., Geophys. J. Int., 177, 1080–1090.
Guilbert, J., Vergoz, J., Schissel, E., Roueff, A. & Cansi, Y., 2005. Use Rhie, J., Dreger, D., Burgmann, R. & Romanowicz, B., 2007. Slip of the
of hydroacoustic and seismic arrays to observe rupture propagation and 2004 Sumatra–Andaman earthquake from joint inversion of long-period
source extent of the Mw = 9.0 Sumatra earthquake, Geophys. Res. Lett., global seismic waveforms and GPS static offsets, Bull. seism. Soc. Am.,
32, L15310, doi:10.1029/2005GL022966. 97(1A), S115–S127.
Hanson, J.A., Reasoner, C.L. & Bowman, J.R., 2007. High–frequency Ritsema, J., Ward, S.N. & González, F.I., 1995. Inversion of deep-ocean
tsunami signals of the Great Indonesian earthquakes of 26 December tsunami records for 1987 to 1988 Gulf of Alaska earthquake parameters,
2004 and 28 March 2005, Bull. seism. Soc. Am., 97(1A), S232–S248. Bull. seism. Soc. Am., 85, 747–754.


C 2011 The Authors, GJI, 185, 1365–1378
Geophysical Journal International 
C 2011 RAS
1378 B. Poisson, C. Oliveros and R. Pedreros

Satake, K., 1995. Linear and nonlinear computations of the 1992 Nicaragua rupture duration, length, and speed of the Great Sumatra–Andaman earth-
earthquake tsunami, Pure appl. Geophys., 144, 455–470. quake, Seism. Res. Lett., 76, 419–425.
Satake, K. et al., 2006. Tsunami heights and damage along the Myanmar Tsai, V.C., Nettles, M., Ekström, G. & Dziewonski, A.M., 2005. Multiple
coast from the December 2004 Sumatra–Andaman earthquake, Earth CMT source analysis of the 2004 Sumatra earthquake, Geophys. Res.
Planets Space, 58, 243–252. Lett., 32, L17304, doi:10.1029/2005GL023813.
Sheth, A., Sanyal, S., Jaiswal, A. & Gandhi, P., 2006. Effects of the December Tsuji, Y., Namegaya, Y., Matsumoto, H., Iwasaki, S.I., Kanbua, W.,
2004 Indian Ocean tsunami on the Indian mainland, Earthquake Spectra, Sriwichai, M. & Meesuk, V., 2006. The 2004 Indian tsunami in Thailand:
22, 435–473. surveyed runup heights and tide gauge records, Earth Planets Space, 58,
Sladen, A. & Hbert, H., 2005. Inversion of satellite altimetry data to recover 223–232.
the Sumatra 2004 earthquake slip distribution, EOS, Trans. Am. geophys. U.S. Department of Commerce, N.O. & Atmospheric Administration, N.G.
Un., 86 (Fall Meet. Suppl.), U22A–07. D.C., 2006. 2–minute Gridded Global Relief Data (ETOPO2v2).
Sladen, A. & Hébert, H., 2008. On the use of satellite altimetry to infer the Vallee, M., 2007. Rupture properties of the giant Sumatra earthquake imaged
earthquake rupture characteristics: application to the 2004 Sumatra event, by empirical Green’s function analysis, Bull. seism. Soc. Am., 97(1A),
Geophys. J. Int., 172, 707–714. S103–S114.
Smith, W.H.F., Scharroo, R., Titov, V.V., Arcas, D. & Arbic, B.K., 2005. Vigny, C. et al., 2005. Insight into the 2004 Sumatra–Andaman earthquake
Satellite altimeters measure tsunami, Doklady Earth Sciences A, 18, from GPS measurements in southeast Asia, Nature, 436(7048), 201–206.
11–13. Wang, X. & Liu, P.L.F., 2006. An analysis of 2004 Sumatra earthquake

Downloaded from https://academic.oup.com/gji/article/185/3/1365/603962 by guest on 27 April 2021


Song, Y.T., Ji, C., Fu, L.L., Zlotnicki, V., Shum, C.K., Yi, Y. & Hjorleifsdottir, fault plane mechanisms and Indian Ocean tsunami, J. Hydrau. Res., 44,
V., 2005. The 26 December 2004 tsunami source estimated from satel- 147–154.
lite radar altimetry and seismic waves, Geophys. Res. Lett., 32, L20601, Watts, P., Grilli, S.T., Kirby, J.T., Fryer, G.J. & Tappin, D.R., 2003. Landslide
doi:10.1029/2005GL023683. tsunami case studies using a Boussinesq model and a fully nonlinear
Stein, S. & Okal, E.A., 2005. Speed and size of the Sumatra earthquake, tsunami generation model, Nat. Hazards Earth Syst. Sci., 3, 391–402.
Nature, 434, 581–582. Wei, G., Kirby, J., Grilli, S. & Subramanya, R., 1995. A fully nonlinear
Subarya, C. et al., 2006. Plate–boundary deformation associated with the Boussinesq model for surface waves. I. Highly nonlinear, unsteady waves,
great Sumatra–Andaman earthquake, Nature, 440, 46–51. J. Fluid Mech., 294, 71–92.
Swe, T.L. et al., 2006. Myanmar Coastal Area Field Survey after the Decem- Wijetunge, J.J., 2006. Tsunami on 26 december 2004 : spatial distribution
ber 2004 Indian Ocean Tsunami, Earthquake Spectra, 22, S285–S294. of tsunami height and the extent of inundation in Sri Lanka, Sci. Tsunami
Tanioka, Y. & Satake, K., 1996. Tsunami generation by horizontal displace- Hazards, 24, 225–239.
ment of ocean bottom, Geophys. Res. Lett., 23, 861–864. Yeh, H., Chadha, R.K., Francis, M., Katada, T., Latha, G., Peterson, C.,
Tanioka, Y., Yudhicara, Kususose, T., Kathiroli, S., Nishimura, Y., Raghuraman, G. & Singh, J.P., 2006. Tsunami runup survey along the
Iwasaki, S. & Satake, K., 2006. Rupture process of the 2004 great southeast Indian coast, Earthquake Spectra, 22, S173–S186.
Sumatra–Andaman earthquake estimated from tsunami waveforms, Earth Zahibo, N., Pelinovsky, E., Talipova, T., Kozelkov, A. & Kurkin, A., 2006.
Planets Space, 58, 203–209. Analytical and numerical study of nonlinear effects at tsunami modeling,
Tolstoy, M. & Bohnenstiehl, D.R., 2005. Hydroacoustic constraints on the Appl. Math. Comput., 174, 795–809.


C 2011 The Authors, GJI, 185, 1365–1378

Geophysical Journal International 


C 2011 RAS

You might also like