1

Probabilistic Analysis of Strong Ground Motion and Tsunami

Hazards in Southeast Asia

Hong Kie Thio*, Paul Somerville*#, and Gene Ichinose*

*URS Corporation, 566 El Dorado Street, Pasadena, CA, USA

#Risk Frontiers, Macquarie University, Sydney, NSW, Australia

Abstract

The December 2004 Sumatra-Andaman earthquake emphasized the need for a

consistent and comprehensive assessment of tsunami hazard. We have developed a

method for Probabilistic Tsunami Hazard Analysis (PTHA) based on the traditional

Probabilistic Seismic Hazard Analysis (PSHA) and therefore completely consistent with

standard seismic practice. In lieu of attenuation relations, it uses the summation of finite-

difference Green’s functions that have been pre-computed for individual subfaults, which

enables us to rapidly construct scenario tsunami waveforms from an aggregate of

subfaults that comprise a single large event. For every fault system, it is then possible to

integrate over sets of thousands of events within a certain magnitude range that represents

a fully probabilistic distribution. Because of the enclosed nature of ports and harbors,

effects of resonance need to be addressed as well, which is why we have extended this

method to not only analyze exceedance levels of maximum wave height, but also of

spectral amplitudes. As in PSHA, these spectral amplitudes can be matched with the

spectral response of harbors, and thus allow a comprehensive probabilistic analysis of

tsunami hazard in ports and harbors.

1 Introduction

The recent tsunami disaster caused by the 2004 Sumatra-Andaman earthquake

(Ammon et al, 2005; Bilek et al., 2007) has focused our attention on the hazard posed by

large subduction zone earthquakes and the tsunamis that they generate. Even before this

destructive event, a significant amount of work was carried out in this field, primarily

through deterministic modelling of tsunami scenarios (e.g. Borrero et al., 2005). Such

studies often address worst-case scenarios or some type of maximum credible event. In

order to put these types of studies on a firm basis, it is necessary to conduct a

comprehensive review of tsunamigenic sources that can affect a certain locality, and

determine the probabilistic hazard level based on this set of sources. Also,

notwithstanding the great usefulness of individual scenario maps, in order to assess the

hazard for a certain region, it may be more appropriate to start with a map of the tsunami

hazard, analogous to the seismic hazard maps that are published by government agencies

such as the United States Geological Survey and the Global Seismic Hazard Assessment

Program. Even though events like the Sumatra-Andaman earthquake and tsunami are

rare, the very large loss of life (more than 200,000 dead or missing) and the tremendous

material destruction over large geographical areas warrant a significant effort towards the

mitigation of the tsunami hazard. In recent years, the risk posed by tsunami to coastal

communities from a variety of sources has become apparent, but it has received much

less attention than has been paid to the hazard from strong ground motions. The lack of a

2

consistent framework to evaluate tsunami hazard has given rise to unnecessarily

conservative estimates of the hazard, which can result in an economic barrier to

development of coastal communities and facilities. The need for a comprehensive and

consistent methodology to evaluate tsunami hazard is clear.

Given the maturity and widespread acceptance of Probabilistic Seismic Hazard

Analysis (PSHA) in seismic hazard mitigation, we believe it is most beneficial to cast a

methodology for tsunami hazard mitigation in a similar framework. By exploiting the

commonality between tsunami and seismic hazard models, such as the earthquake

recurrence models, we assure maximum consistency across the two hazards, facilitating

the evaluation of the combined hazard posed to coastal communities, facilities and

infrastructure.

2 Probabilistic Seismic Hazard Analysis

Probabilistic Seismic Hazard Analysis (PSHA) has become standard practice in the

evaluation and mitigation of seismic hazard to populations in particular with respect to

structures, infrastructure and lifelines. Its ability to condense the complexities and

variability of seismic activity into a manageable set of parameters greatly facilitates the

design of effective seismic resistant buildings and the planning of infrastructure projects.

Probabilistic Tsunami Hazard Analysis (PTHA) achieves the same goal for hazards posed

by tsunami. Although this field is at an early stage of development, we believe there are

great advantages of implementing such a method to evaluate the total risk (seismic and

tsunami) to coastal communities, facilities and infrastructure. The PTHA method that we

have developed is based on traditional PSHA and is therefore completely consistent with

standard seismic hazard practice. It provides a synopsis of the tsunami hazard along

entire coastlines, and helps identify vulnerable locations along the coast and the specific

tsunami source regions to which these vulnerable locations on the coastline are sensitive.

The methodology behind PSHA is well known (e.g. McGuire, 2004) and here we

will only briefly describe the adaptations that are made for PTHA. Whereas in PSHA we

are usually interested in the exceedance of some ground motion measure such as Peak

Ground Acceleration (PGA) or Spectral Acceleration (SA) in PTHA a parameter of

interest (not necessarily the only one) is the maximum tsunami height that is expected to

be exceeded at sites along the coast. The earthquake occurrence model used in the two

methods is the same, and the only difference is that in PTHA we are not concerned about

earthquakes that occur completely inland. The main difference between the two methods

lies in the way in which the wave propagation is treated. In traditional PSHA, attenuation

relations relate earthquake magnitude to the ground motion parameters as a function of

distance. Because of the strong laterally varying nature of tsunami propagation, we have

adopted a waveform excitation and propagation approach instead of trying to develop

analogous tsunami attenuation relations. In fact, current developments in traditional

PSHA include the replacement of the attenuation relations with ensembles of numerically

generated ground motions (Graves et al., 2006), which is entirely analogous to the

approach that we propose here.

Probabilistic Seismic Hazard Analysis (PSHA) is based on methodology originally

proposed by Cornell (1968) and we will present a brief overview of the method as well as

specific information on the parameters and models used in our analysis. The probability

3

that a ground motion, such as Spectral Acceleration (SA), exceeds a certain value (s) in a

time period t is given by:

P(A > s) =1 e (s )t

where (s) is the annual mean number of events (also known as “annual frequency

of exceedance”) in which the ground motion parameter of interest exceeds the value s.

For engineering purposes, we are interested in computing s for a certain probability of

occurrence, P, in a time period t. For this project, the targets are probabilities of 2%, 5%,

10% and 50% occurrence in a time span of 50 years, which is equivalent to annual

frequencies of exceedance of 1/2475, 1/975, 1/475 and 1/72 per year respectively. We

usually refer to the latter in terms of return period, e.g. 475 and 72 years.

The annual frequency of exceedance is calculated as follows:

(s) = ( f (m)(P(A > s | m,r)P(r | m)dmdr)im,ri=1

Faults

where:

f(mi) = probability density function for events of magnitude mi

P(A>s|m,r) = probability that amplitude A exceeds s given magnitude m and

source at r

P(r|m) = probability for a source at r, given a source of magnitude m.

Whereas in traditional PSHA the r term represents source to site distance, in our case

it refers more generally to the source location r with the propagation from the source to

the site computed explicitly through the Green’s function approach instead of an

attenuation relation. Uncertainties associated with seismic source parameters, such as

geometry, location, rupture scenario and recurrence rates were incorporated using a logic

tree approach.

3 Earthquake Sources in Southeast Asia

Southeast Asia is a tectonically complex region (Figure 1), consisting of several

different micro-plates that are often bounded by convergent margins, reflecting the

overall compressive regime that is characterized by the northward motion of the

Australian plate (60-80 mm/yr relative to the Sunda plate), and eastward motion of the

Philippine Sea plate (100-110 mm/yr relative to the Sunda Plate). The eastern end of the

Sunda trench marks the transition of pure oceanic subduction environment to a more

continental collision type environment. Part of the convergence is taken up along the

northern coastline of the Sunda islands, and there is thought to be an EW trending zone of

deformation along the northern coast of the lesser Sunda islands. The Philippine Islands

form a complex zone of tectonic deformation flanked on the east and west by convergent

margins. This complexity is reflected in the number of source zones for this region, some

of which are quite short. The Philippine trench is one of the longest of the convergent

margins in the Philippines, extending from the southern end of Mindanao to central

Luzon. The Luzon trench can be regarded as a northern continuation of the Philippine

trench that has been offset to the west. It is relatively short, and has a lower convergence

4

rate since the convergence from the Philippines Trench is partly taken up along the

Manila trench on the western side of Luzon. The Manila trench forms the convergent

margin between the Philippines Sea plate and the Sunda plate from central Luzon to

Taiwan. The Negros trench is a short convergence zone along the western margin of the

central Philippines, and the Cotabato trench is another short trench system, located along

the south-western coast of Mindanao.

In order to obtain a comprehensive overview of the tsunami hazard in our target

region, we considered tsunami sources from all around the Pacific Rim. Although in

many cases this may appear redundant, since some source regions seem unlikely to pose

any tsunami hazard, we include them since the effort is primarily a computational one.

Given the extensive distribution of our target region, it is likely that any source region

affects at least part of our target region. With our deterministic based approach,

conservatism in choosing sources is warranted since low hazard regions will

automatically be interpreted correctly in the process.

For any probabilistic hazard analysis, the statistical properties of the source model

are essential. There have been many recent studies on the recurrence rates of earthquakes

along the subduction zone interfaces of the Circumpacific region (e.g. Gusev et al., 2005,

Ikehara et al 2001, Nelson et al. 2003; Nishenko, 1985). In general, we use tectonically

based recurrence rates for the source models. These can be checked against tsunami

records at locations where they are available. The source zones were digitized using the

Plate model of Bird (2003) and the RUM model of Gudmundsson and Sambridge (1998).

Overall convergence rates were taken from Bird (2003) with some details for the

Philippines and Indonesia from Kreemer et al. (2000). We checked the subduction zone

geometries, in particular with respect to the dip angle, by comparing the digitized models

with hypocenter depths from the relocated ISC catalog (Engdahl et al., 1998). In the

present analysis, we have not included local offshore crustal faults.

Based on our experience with the tsunami modeling of the 2004 Sumatra earthquake

and other events, we have parameterized the rupture planes following the subduction

zone model with a cut-off depth of 50 km, so ruptures are not allowed to propagate

deeper than 50 km. This boundary is consistent with seismic studies of interface

earthquakes. The exact depth is not very important for tsunami generation, since their

amplitude is predominantly dictated by shallow slip, but it is important for recurrence

models. The extent of shallow slip is very important, and we include random variability

in the upward extent of the rupture in our analysis.

3.1 Uncertainties in Earthquake Source Models

Earthquake source recurrence models form the basis of the probabilistic analysis, and

these models are of critical importance in tsunami hazard analysis. In seismic hazard

analysis, recurrence models are often a mix of Gutenberg-Richter type and characteristic

type models, with an emphasis on characteristic models for the larger earthquake faults.

Since in tsunami hazard we are only concerned with large earthquakes (M > 7.5), we

have opted for characteristic earthquake models for all our sources. Since almost all

sources are subduction zones, for which we have well-constrained convergence rates, we

have used slip rates (and maximum magnitudes) as the primary input that governs the

recurrence of earthquakes rather than a-values or recurrence intervals. Empirically based

5

rates for these large events are poorly defined in most circumstances due to the short

historical record relative to the recurrence intervals.

Figure 1. Subduction source zones (red lines) in East Asia.

6

A crucial element in PTHA is the estimation of the frequency of occurrence and

maximum magnitudes of large earthquakes in each source region. Due to the very short

historical record for mega-thrusts and other large earthquakes in relation to their

recurrence times, it is not possible to base such constraints directly on the observed

seismicity. We therefore need to use models that are at least partly based on earthquake

mechanics, which can be as simple as magnitude/area relations but can also include

physics based constraints in addition to empirical data such as earthquake locations.

Uncertainties in source parameters, such as maximum earthquake and slip rate, are

included using logic tree analysis. Other approaches toward PTHA often use a limited

range of deterministic scenarios with associated probabilities or return periods,

sometimes in combination with historical tsunami records (Berryman, 2006; Imamura et

al., 2006; Geist and Parsons, 2006).

3.2 Time Dependent Earthquake Hazard

A probabilistic hazard analysis of peak ground acceleration for Sumatra and the

southern Malay Peninsula was published by Petersen et al. (2004). They assumed that

earthquakes occur randomly in time, following a Poisson process. It is possible that their

estimates are low, because their study was completed before the occurrence of the Mw

9.15 December 26, 2005 and Mw 8.7 March 28, 2006 Sumatra earthquakes. Based on the

concept of stress transfer, it has been proposed (e.g. Nalbant et al., 2005) that the

occurrence of these two earthquakes has increased the likelihood of occurrence of the

following two earthquakes during the next few years:

A repeat of the 24 November 1833 earthquake on the Sumatra subduction zone. This

subduction earthquake would have a magnitude of about 9.0.

A repeat of the 17 May 1892 earthquake on the Sumatra fault. This strike-slip

earthquake would have a magnitude of about 8.0.

4 Ground Shaking Hazard from Earthquakes

The broad features of ground shaking hazard in Southeast Asia are portrayed by the

GSHAP seismic hazard map in Figure 2 (Giardini, 1999), which shows the level of peak

acceleration on soft rock sites for an annual probability of exceedance of 1/500. This

map was compiled from several regional studies that used different probabilistic hazard

analysis methods, so the results are expected to be somewhat heterogeneous.

Nevertheless, the strong contrast in hazard levels between plate margin regions and plate

interior regions is clear in this map.

Reflecting uncertainty in the attenuation of ground motion from crustal earthquakes

in Southeast Asia region, Petersen et al. (2004) used a combination of ground motion

models for stable and tectonically active regions. Strong motion simulation methods

have been used to estimate ground motion levels and develop ground motion models for

the tectonically stable regions of Southeast Asia (e.g. Megawati and Pan, 2002; Megawati

et al., 2005, 2006). Strong motion recordings in Singapore over the past decade have

been used to check these ground motion models (e.g. Pan et al., 2001, 2005). Balendra et

al. (2002, 2004) have used these ground motion estimates to assess the seismic capacity

of buildings in Singapore.

7

Figure 2. GSHAP probabilistic peak acceleration hazard map for an annual probability of exceedance of 1/500. Source: Giardini (1999).

The 24 November 1833 earthquake, which occurred on the Sumatra subduction

zone, produced ground-shaking intensities of MMI II-III, corresponding to a peak

acceleration of less than 1.5%g in Singapore. The May 17, 1892 earthquake, which

occurred on the Sumatra fault, caused the strongest felt ground motion in Singapore, with

MMI intensity IV-V, in Singapore. MMI intensity IV corresponds approximately to a

peak acceleration of 1.5 – 4.0%g, and MMI intensity V to 4.0 – 9.0%g. Structures

present in Singapore at the time of both of these earthquakes were probably insensitive to

long period ground motions, and the ground motions that did occur caused practically no

damage. However, if either of these events were to reoccur, they might cause long period

ground motions that could have a significant impact on buildings in Singapore, especially

those located on the soft soils of the Kallang Formation. The largest published peak

acceleration recorded in Singapore in the past decade, since strong ground motions were

first installed, is about 0.5%g (Pan et al., 2005), which occurred in the Mw 9.15 2005

Sumatra subduction earthquake. Petersen et al. (2004) estimated the peak acceleration in

Singapore for an annual probability of exceedance of 1/500 to be 5%g on NEHRP site

category B/C boundary (soft rock or very dense soil).

5 Tsunami Hazard from Earthquakes

Even though the occurrence of large tsunamis is rare, the potential for extensive loss

of life and damage, as demonstrated in the 2004 Sumatra-Andaman earthquake, merits a

strong commitment toward tsunami hazard mitigation issues. The vast majority of

tsunamis are caused by earthquake-induced displacement of the seafloor. Most of the

world’s largest tsunamis, which have caused damage at locations thousands of miles

away, have been caused by megathrust (subduction interface) earthquakes around the

Pacific Rim. These include the 1960 Chile earthquake, the 1964 Alaska earthquake and

the 2004 Sumatra-Andaman earthquake. On a local scale, smaller earthquakes can cause

8

significant tsunamis as well, but usually the hazard from these events is lower because of

their longer recurrence rate and localized impact.

Earthquakes are by no means the only sources of tsunami. Submarine landslides,

whether triggered by earthquakes or not, are another important source for tsunamis

(Satake, 2001; Watts, 2004), and although their effects tend to be more localized (e.g.

1996 Flores earthquake), there is speculation that major submarine landslides (e.g. in

Hawaii, Satake, 2001) and asteroid impacts (Ward and Aphaug, 2000) could generate

giant tsunamis that could devastate coastal regions thousands of miles away. We have

limited the scope of this paper to include only tsunamis that are directly generated by

earthquakes, but the method that we describe is also able to include submarine landslides

and catastrophic volcanic collapse. Based on the much lower recurrence rates for these

latter sources, we feel confident that the earthquake model includes all the significant

sources for return periods of up to 2500 years.

5.1 Probabilistic tsunami hazard analysis

Downes and Stirling (2001) proposed the use of empirical tsunami attenuation

relation similar to ground motion attenuation relations. They recognize that such

attenuation relations would have to be source and site specific, and it appears doubtful

whether there would ever be enough data available for such attenuations relations to be

derived consistently. Sato et al. (2003) developed a simplified method of tsunami risk

assessment. Annaka et al. (2004) and Geist and Parsons (2005) developed methods that

use the full linear tsunami calculations for a limited number of scenarios for earthquakes

near the site. The main difference between our work and theirs is that through the Green’s

function summation, we can generate many more fault scenarios and at arbitrary

distances including teleseismic, which allows us to run full probabilistic analyses over a

much wider area. Also, our method is very efficient for the analysis of many sites

simultaneously, which allows us to quickly identify areas at elevated risk. Such

information is indispensable for the effective allocation of funds for tsunami hazard

mitigation work.

5.2 Tsunami Modeling

In modeling a tsunami the particle motion of the fluid is described using an Eulerian

approach. Using this approach, the velocity changes of the fluid are only described at a

given point and at a given instant of time, rather than describing its absolute

displacement. We consider a wave that is a propagating disturbance from an equilibrium

state. When the horizontal scale of motion is much larger than the water depth, then the

vertical acceleration of water is much smaller than the gravity acceleration and thus

negligible. This means that the whole water mass from the bottom to the surface is

assumed to move uniformly in a horizontal direction. This kind of gravity wave is also

known as a “long-wave.” Long wave approximations are appropriate when the water

depths (< 5 km) are much smaller than the length of the disturbance (fault lengths ~ 10-

1000 km).

The excitation and propagation of tsunami in deeper water can be modeled using the

shallow water wave approximation, which, for amplitudes that are significantly smaller

than the water depth are linear (Satake, 1995). We can solve the equation of motion

numerically using a finite-difference method. The numerical method has been validated

9

to produce accurate tsunami heights for propagation through the oceans. The assumption

of linearity is not valid for tsunamis where the amplitudes are comparable to the water

depth. Also, the detailed bathymetry near the shoreline is important to estimate the final

run-up heights. For these cases, a non-linear method is necessary to compute the run-up

heights correctly. Nevertheless the linear approach provides a very good first

approximation of tsunami propagation taking into account the effects of lateral variations

in seafloor depth. Accordingly, our approach to probabilistic tsunami hazard analysis is

to use fast linear calculations to compute the tsunami response at a particular depth

contour off the coastline (15 m), as demonstrated for the December 2004 and March 2005

Sumatra earthquakes in Figures 3 and 4.

Peak Amplitudes

Figure 3. Peak calculated wave heights for the December 2004 (left) and March 2005 (right) Sumatra tsunamis.

Peak Shoreline Wave Heights

Figure 4. Peak calculated shoreline wave heights for the December 2004 (left) and March 2005 (right) Sumatra tsunamis.

10

5.3 Tsunami Green’s function summation

The underlying principle of our PTHA approach is the validity of the linear

behaviour of tsunami waves. This enables us to deconstruct a tsunami that is generated by

an earthquake into a sum of individual tsunami from a set of subfaults that adequately

describe the earthquake rupture. The slip on the fault is translated into vertical

deformation using a numerical model (Wang et al., 2003) of a multi-layered crustal

model (Kopp et al., 2002). By pre-computing and storing the tsunami waveforms at

points along the coast generated by each subfault for a unit slip (Figure 5), we can

efficiently synthesize tsunami waveforms for any slip distribution by summing the

individual subfault tsunami waveforms (weighted by their slip). The same principle is

used in the inversion of tsunami waves for earthquake rupture (e.g. Satake, 1995). This

efficiency makes it feasible to use Green’s function summation instead of attenuation

relations to provide very accurate estimates of tsunami height for probabilistic

calculations, where one typically needs to compute thousands of earthquake scenarios.

For instance, in this study the probabilistic tsunami height results are based on several

thousands of scenarios, which were computed on a Linux cluster. The assumption of

linearity is not valid for tsunami where the amplitudes are comparable to the water depth.

Also, the detailed bathymetry near the shoreline is important to estimate the final runup

height. For these cases, a non-linear method is necessary to compute the runup heights

correctly.

Figure 5. Peak wave height of tsunami Green’s functions calculated for one meter of slip on a 50 x 50 km fault element at the top edge of the Sumatra subduction plate interface.

11

5.4 Bathymetry

We have evaluated several bathymetric models for modeling the tsunami waveforms

in East Asia (and across the Pacific). Our initial intention was to use the ETOPO2 global

topography, which currently is most often used by tsunami modelers and which we have

used in the past. However, upon close inspection with the high resolution nearshore

bathymetry in several regions, we have concluded that this model has large errors close to

the shore (with occasional differences of up to 200 m compared to the high-resolution

bathymetry, at actual depths of 20 m). Because the hazard is determined at shallow

depths, and because of the high sensitivity of tsunami amplitude to water depth, we

decided to explore different models. We subsequently tested several models for accuracy:

ETOPO2 – based on satellite gravity data

GEBCO – global bathymetric model based on ship-track data

S2004/ETOPO2v2 – a mix of etopo2 for deep oceans and GEBCO for nearshore

bathymetry and terrestrial topography

SRTM+ - a mix of etopo2-based bathymetry and space shuttle radar mapping

It appears that the GEBCO model is more accurate at shallow depths, but has poor

resolution in the oceans, because of the scarcity of ship-track data. ETOPO2 has much

better and uniform resolution in the open oceans, but, as mentioned, suffers from poor

accuracy near the shore. It may be that the strong gravity contrast across the ocean-

continent boundary introduces erroneous depths because of insufficient correction of the

gravity data. The SRTM+ model, although based on gravity data as well, produces more

accurate near shore bathymetry, but still shows some serious discrepancies in the same

areas where ETOPO2 fails. S2004 was developed specifically to address these issues and

combine the strengths of GEBCO and the gravity-derived model, and this philosophy was

subsequently adapted in the second iteration of the ETOPO2 model, ETOPO2v2 that

became available during the course of this study. Based on our experiences in different

parts of the world, where were able to compare the global models with local high-

resolution data, we now conclude that the ETOPO2v2 model provides the best

compromise between deep and shallow water accuracy and resolution and we have

adapted this model in our tsunami waveform simulations.

5.5 Tsunami Hazard Maps

We have computed the probabilistic tsunami wave heights for many areas around the

Pacific Ocean based on earthquake recurrence models of all the subduction zones and

some offshore crustal faults. An example of such a map is presented in Figure 6. It shows

the peak tsunami wave height that is exceeded in 475 years, a typical hazard level for

engineering purposes. The wave height patterns show the expected high hazard levels in

the Philippines and parts of Indonesia and lower levels along the Pacific sides of

Malaysia and Thailand.

12

Figure 6. Probabilistic tsunami hazard map for Southeast Asia showing the peak wave height exceedance levels for 475 year return period. These results are based on earthquake sources in the Pacific Ocean only. The hazard on the Indian Ocean side of the Indonesian archipelago is therefore underestimated. Source: Report to Tokio Marine & Nichido Risk Consulting Co. Ltd., 2006.

13

5.6 Tsunami Hazard Curves

We have plotted corresponding hazard curves for several localities in southeast Asia

in Figure 7. It is obvious that there are strong variations between sites, not only in the

general level of the hazard, but also in the dependence of waveheight on return period,

which reflects the very different recurrence rates on different sources.

Figure 7. Tsunami wave height hazard curves for different localities in Southeast Asia. Source: Report to Tokio Marine & Nichido Risk Consulting Co. Ltd., 2006.

5.7 Uniform tsunami hazard spectra

Since the hazard calculations are based on actual waveform simulations, we have the

ability to compute the hazard for any tsunami waveform characteristic, not just the peak

wave height. For instance, in cases where withdrawal of water is of importance, we could

compute the hazard for minimum wave height. For application to ports and harbors, we

suggest that spectral amplitudes are of interest because of the problem of harbor

resonance. This phenomenon is well known and extensively studied in the context of

storm waves but is equally important for tsunami waves that span very wide period range.

14

Analogous to PSHA, we can determine uniform hazard spectra as shown in Figure 8.

The spectra represent the spectral amplitude that is exceeded for that particular hazard

level, say 475 years, for a set of periods. Note that these spectra do not represent any

single tsunami simulation, but instead for every individual period a complete probabilistic

calculation. We have only plotted the response at a few periods because of time

considerations, but cancompute more continuous uniform hazard spectra.

The spectra in Figure 8 are simple Fourier spectra, as opposed to the response spectra

used in PSHA. If the response spectrum for a particular harbor is known, we can

determine actual response spectra that are tailor made for that particular location,

including specific source characterizations, propagation effects and local response.

5.8 Disaggregation of tsunami hazard

The probabilistic peak and spectral wave heights are very useful tools in assessing

the hazard posed by tsunamis. However, as mentioned before, the current approach that

allows compute thousands of scenario tsunamis efficiently is not suited for detailed (non-

linear) inundation studies. Disaggregation of tsunami hazard into source areas (and other

relevant source parameters such as magnitude) allows identifying all the high-hazard

sources that need to be studied in more detail through scenario modeling. In contrast to

earlier modeling efforts, these scenarios are firmly based on a probabilistic analysis, and

do not represent some poorly defined maximum credible event, but instead an event with

a specific hazard level.

The disaggregation map shown in Figure 9 illustrates how the tsunami hazard at a

given location, Singapore, stems from several source regions with the highest hazard

from the Manila Trench. In some situations, the subduction zone that is closest to the site

may pose a very low hazard to the site, due to the orientation of the subduction zone

Figure 8. Uniform tsunami hazard spectra for sites along the California coast and Hawaii.

15

relative to the direct path to the site. The largest amplitudes for distant tsunami are

registered in the direction perpendicular to the strike of the fault.

Figure 9. Source disaggregation for the tsunami hazard in Singapore (red square) for peak waveheight. The blue bars represent the relative contribution of each element to the tsunami hazard in Singapore. Source: Report to Tokio Marine & Nichido Risk Consulting Co. Ltd., 2006.

6 Discussion

The method for probabilistic tsunami hazard analysis is very versatile, and can be

tailor made for very specific purposes, such as ports and harbors, while still being

consistent with common practice in hazard analysis. Because of the similarities with

PSHA, which is firmly established as standard practice in ground motion hazard analysis,

we believe that PTHA, and our implementation in particular, can be a very useful tool in

tsunami hazard mitigation for a wide range of facilities and projects. A probabilistic

analysis enables us to evaluate the tsunami hazard in a more objective framework and

help identify scenarios of particular concern, rather than proceeding directly with

scenario modeling with arbitrary likelihoods of occurrence.

16

Our current results show that this approach is very feasible. We have identified areas

with elevated tsunami hazard, and through our spectral analysis, we can also address the

hazard in relation to harbor resonance, and identify particular source areas that warrant

further study. For longer recurrence times (> 1000 years), the current set of source zones

needs to be augmented with local earthquake sources, which typically have much larger

return periods, and can cause local tsunamis that are more severe than distant tsunamis.

The large uncertainties that are encountered with landslide sources in particular (e.g. slide

velocity, recurrence rates and dimensions) can be taken into account using the logic tree

approach used to treat uncertainties in earthquake source parameters.

7 Acknowledgment

The authors would like to thank Mr Takashi Mikajiri and Mr Ichiro Sato of Tokio

Marine & Nichido Risk Consulting Co. Ltd. for their sponsorship of the probabilistic

tsunami hazard analysis for the East Asia region that is described in this paper, and their

permission to publish Figures 6, 7 and 9.

8 References

Ammon, C.J., J. Chen, H.K. Thio, D. Robinson, S. Ni, V. Hjorleifsdottir, H. Kanamori,

T. Lay, S. Das, D. Helmberger, T. Lay, S. Das, G. Ichinose, J. Polet, D. Wald (2005).

Rupture process of the great 2004 Sumatra-Andaman earthquake. Science 308, 1133-

1139.

Annaka, T, K Satake, N Shuto, T Sakakiyama, K Yanagisawa, 2004. Logic-tree

Approach for Probabilistic Tsunami Hazard Analysis and its Applications to the

Japanese Coasts, Abstract: 2004 Fall AGU meeting.

Balendra, T., M.J. Perry, J.L. Wilson, and N.K.T. Lam (2004). Seismic capacity of

buildings in Singapore designed primarily for gravity loads. Proceedings of the 13th

World Conference on Earthquake Engineering, Vancouver, B.C., Canada, Paper No.

1920.

Balendra, T., N.K.T. Lam, J.L. Wilson, and K.H. Kong (2002). Analysis of long-distance

earthquake tremors and base shear demand for buildings in Singapore. Engineering

Structures 24, 99-108.

Berryman, K., 2006. Review of Tsunami hazard and risk in New Zealand, report by the

Institute of Geological and Nuclear Sciences, New Zealand, 140 pp.

Bilek, S.L., K. Satake and K. Sieh (2007). Introduction to the Special Issue on the 2004

Sumatra-Andaman earthquake and the Indian Ocean Tsunami. Bull. Seism. Soc. Am

97, S1-S5.

Borrero, J., Cho, S., Moore, J.E., Richardson, H.W., and C. Synolakis, 2005. Could it

happen here? Civil Eng., 75, 54-65.

Bird, P., 2003. An updated digital model of plate boundaries, Geochem. Geophys.

Geosyst., 4, doi:10.1029/2001GC000252.

Cornell, C. A.: 1968, Engineering seismic risk analysis, Bull. Seism. Soc. Am 58, 1583-

1606.

17

Downes, G.L. and M.W. Stirling, 2001. Groundwork for development of a probabilistic

tsunami hazard model for New Zealand, International Tsunami Symposium

Proceedings, session 1, 293-301.

Geist, E. and T. Parsons, 2006. Probabilistic analysis of tsunami hazards, Natural

Hazards, 37, 277-314.

Giardini, (1999). The Global Seismic Hazard Assessment Program, 1993-1999. Annali

di Geofisica.

Graves, R., Maechling, P., Zhao, L., Mehta, G., Gupta, N., Mehringer, J., Deelman, E.,

Kesselman, C., Callaghan, S., Cui, Y., Field, E., Gupta, V., Jordan, T., Okaya, D., and

Vahi, K., 2006. SCEC Cybershake Platform: Incorporating Deterministic 3D

Waveform Modeling into Probabilistic Seismic Hazard Curves.

Gudmundsson and Sambridge 1998. A regionalized upper mantle (RUM) seismic model, J. of Geophys. Res., No. B4, 7121-7136, 1998.

Gusev, A. A., and L. S. Shumilina. 2004. Recurrence of kamchatka strong earthquakes on

a scale of moment magnitudes. Izvestiya - Russian Academy of Sciences.Physics

of the Solid Earth 40, (3) (Mar): 206-215.

Ikehara, Ken, Masataka (president) Ando, Akira (prefacer) Asada, Takashi (prefacer)

Ikeda, Yuzo (president) Ishikawa, Yoshiyuki (prefacer) Kaneda, and Makoto

(prefacer) Murakami. 2001. Recurrence interval of large earthquakes along the

eastern nankai trough inferred from deep-sea turbidites; preparations for the

coming large earthquakes at the nankai trough. Chigaku Zasshi = Journal of

Geography 110, (4): 471-478.

Kopp, H., D. Klaeschen, E.R. Flueh, J. Bialas and C. Reichert, 2002. Crustal structure of

the Java margin from seismic wide-angle and multichannel reflection data, J.

Geophys. Res., 107, B2 doi:10.1029/2000JB000095.

Kreemer, C., W.E. Holt, S. Goes and R. Govers, 2000. Active deformation in the eastern

Indonesia and the Philippines frm GPS and seismicity data, J. Geophys. Res., 105,

663-680.

Latief, H., D.H. Natawidjaya, H. Sunendar, A.R. Gusman and Y. Tanioka, 2006.

Accounts and Modeling of the Old and Modern Sumatra Tsunamis for Mitigation in

the Future, Third International Conference on Urban Earthquake Engineering,

Tokyo, March 6-7, 2006.

McGuire, R.K., 2004. Seismic hazard and risk analysis, EERI, 240 pp.

Megawati, K., W.T. Wong, L.S. Chan, A.M. Chandler, and Y.W. Chan (2006).

Simulation of distant earthquakes affecting Hong Kong. Preprint.

Megawati, K., T.-S. Pan, and K. Koketsu (2005). Response spectral attenuation

relationships for Sumatran subduction earthquakes and the seismic hazard

implications to Singapore and Kuala Lumpur. Soil Dynamics and Earthquake

Engineering 25 (2005), 11-25.

18

Megawati, K., and T.-S. Pan (2002). Prediction of maximum credible ground motion in

Singapore due to a great Sumatra subduction earthquake: the worst case scenario.

Earthquake Engineering and Structural Dynamics 31, 1501-1523.

Nalbant, S.S., S. Steacy, K. Sieh, D. Natawidjaja, and J. McCloskey

(2005). Earthquake

risk on the Sunda trench. Nature 435, 756-757 (9 June 2005) | doi:

10.1038/nature435756a.

Nelson, Hans, Chris Goldfinger, Julia Gutierrez Pastor, Joel Johnson, and Texas A&M

University, College Station, TX, United States, performer (USA). 2003. Holocene

seismicity of the cascadia subduction zone based on precise dating of turbidite

event record.

Nishenko, S.P. (1985). Seismic potential for large interplate earthquakes along the

Chilean and southern Peruvian margins of South America: a quantitative appraisal.

Journal of Geophysical Research 90, 3589-3615.

Pan, T.-S., K. Megawati, J.M.W. Brownjohn, and C.L. Lee (2001). The Bengkulu,

Southern Sumatra, earthquake of 4 June 2000 (Mw = 7.7): another warning to remote

metropolitan areas. Seismological Research Letters 72, 171-185.

Pan, T.-S. and K. Megawati (2002). Estimation of peak ground accelerations of the

Malay Peninsula due to distant Sumatra earthquakes. Bull. Seism. Soc. Am. 92,

1082-1094.

Pan, T.-S. and C.L. Lee (2002). Site response in Singapore to long-distance Sumatra

earthquakes. Earthquake Spectra 18, 347-367.

Pan, T.-S. (1997). Site-dependent building response in Singapore to long-distance

Sumatra earthquakes. Earthquake Spectra 13, 475-488.

Pan, T.-S. and C.K. Wah (2004). Site response to maximum credible earthquakes from

Sumatra. Civil Engineering Research January 2004, 100-103.

Pan, T.-S., K.R. Karim, X. You, C.L. Lee and C.L. Leong (2005). The great Sumatra-

Andaman Islands earthquake (Mw = 9.0) of 26 December 2004: far-field response in

Singapore.

Pan, T.-S., X. You, and K.R. Karim (2005). A brief report on Republic Plaza response to

the great Sumatra-Andaman Islands earthquake (Mw = 9.0) of 26 Dec 2004.

Pan, T.-S. and J. Sun (1996). Historical earthquakes felt in Singapore. Bull. Seism. Soc.

Am. 86, 1173-1178.

Petersen, M.D., J. Dewey, S. Hartzell, C. Mueller, S. Harmsen, A. Frankel, and K.

Rukstales (2004). Probabilistic seismic hazard analysis for Sumatra, Indonesia and

across the Southern Malaysian Peninsula. Tectonophysics 390, 141-158.

Polet, J. and H. Kanamori, 2000. Shallow subduction zone earthquakes and their

tsunamigenic potential, Geophys. J. Int., 142, 684-702.

Prasetya, G.S., W.P. de Lange and T.R. Healy, 2001. The Makasser Strait tsunamigenic

region, Indonesia, Nat. Haz., 24, 295-307.

19

Rangin, C., X. Le Pichon, S. Mazotti, M. Pubellier, N. Chamot-Rooke, M. Aurelio, A.

Walpersdorf and R. Quebral, 1999. Plate convergence measured by GPS across the

Sundaland/Philppine Sea Plate deformed boundary: the Philippines and eastern

Indonesia, Geophys. J. Int., 139, 296-316.

Ruff, Larry J. 1996. Large earthquakes in subduction zones; segment interaction and

recurrence times; subduction top to bottom. Geophysical Monograph 96, : 91-104.

Satake, K. (2001). Tsunami modeling from submarine landslides. Proceedings of the

International Tsunami Symposium, Session 6, Number 6-4, p. 665-674. Seattle.

Satake, K., 1995. Linear and nonlinear computations of the 1992 Nicaragua earthquake

tsunami, Pageoph, 144, 455-470.

Sato, H., Murakami, H, Kozuki, Y., Yamamoto, N., 2003. Study on a Simplified Method

of Tsunami Risk Assessment, Natural Hazards 29: 325–340, 2003.

Thio, H.K., Somerville, P., and G. Ichinose, 2006. Probabilistic Tsunami Hazard

Analysis, Geophys. Res. Lett., in prep.

Thio, H.K., G. Ichinose, P. Somerville and J. Polet (2005a). Source process and tsunami

generation of the 2004 Sumatra-Andaman earthquake. Proceedings of the Second

Annual Meeting of the Asia-Oceania Geosciences Society, Singapore, June 20-24.

Thio, H.K., G. Ichinose, and C. Saikia (2005b). Slip model of the 2004 Aceh-Nicobar

earthquake from teleseismic body waves, surface waves, and tsunami data.

Seismological Research Letters, 76, p. 221.

Wang. R., F.L. Martin and F. Roth, 2003. Computation of deformation induced by

earthquakes in a multi-layered elastic crust-FORTRAN programs EDGRN EDCMP.

Comp. & Geosc., 29, 195-207, inc. erratum: February 2006.

Ward, S.N. and E. Asphaug, 2000. Asteroid impact tsunami: a probabilistic hazard

assessment, Icarus, 245, 64-78.

Watts, Philip, 2004. Probabilistic predictions of landslide tsunamis off Southern

California, Marine Geology, 203, 281-301.

Youngs, R.R., S.-J. Chiou, W.J. Silva, and J.R. Humphrey (1997). Strong ground motion

relationships for subduction earthquakes, Seismological Research Letters 68, 58-73.

Zhao, J.X., J. Zhang, A. Asano, Y. Ohno, T. Oouchi, T. Takahashi, H. Ogawa, K. Irikura,

H.K. Thio, P. Somerville, Y. Fukushima, and Y. Fukushima (2006). Attenuation

Relations of Strong Motion in Japan using Site Classification based on Predominant

Period. Bull. Seism. Soc. Am. 93, 898-913.