International Journal of Disaster Risk Reduction 10 (2014) 406–418

Contents lists available at ScienceDirect

International Journal of Disaster Risk Reduction

journal homepage: www.elsevier.com/locate/ijdrr

Global tsunami hazard and exposure due to large

co-seismic slip
Finn Løvholt a,b,c,n, Sylfest Glimsdal a,b,c, Carl B. Harbitz a,b,c, Nick Horspool d,f,
Helge Smebye a,c, Andrea de Bono e, Farrokh Nadim a,c
a
NGI, PO Box 3930, N-0806 Ullevål Stadion, Oslo, Norway
b
University of Oslo, Department of Mathematics, PO Box 1072, Blindern, Oslo 0316, Norway
c
International Centre for Geohazards, co NGI, Oslo, Norway
d
Geoscience Australia, GPO Box 378, Canberra, ACT, Australia
e
UNEP/DEWA/GRID-Geneva, International Environment House, 11 Chemin des Anémones, 1219 Châtelaine, Switzerland
f
GNS Science, PO Box 30-368, Lower Hutt, New Zealand

a r t i c l e in f o abstract

Article history: Tsunamis are infrequent events with the power to cause massive loss of life, large
Received 5 November 2013 economic losses, and cascading effects such as destruction of critical facilities. Historical
Received in revised form tsunamis and paleotsunami evidence indicates indirectly that massive megathrust earth-
14 March 2014
quakes leads to the majority of the losses due to tsunamis. There is a need to quantify the
Accepted 2 April 2014
tsunami hazard from megathrust events in order to compare tsunamis with other natural
Available online 13 April 2014
hazards on a global level, as previous attempts have been lacking. Here, we determine the
Keywords: earthquake induced tsunami hazard for a return period of 500 years using both a
Tsunamis deterministic scenario based approach as well as a probabilistic tsunami hazard assess-
Hazard analysis
ment method (PTHA). The resulting hazard level for a set of selected areas in South and
Global study
South East Asia are compared quantitatively for both methods. The comparison demon-
Population exposure
Economic loss strate that the accuracy of the analysis is rather rough, which is expected given the global
Cascading effects character of the analysis. Globally, the exposed elements at risk such as population,
Critical facilities produced capital, and nuclear power plants are determined for each nation affected. It is
shown that populous Asian countries account for the largest absolute number of people
living in tsunami prone areas, more than 50% of the total exposed people live in Japan.
Smaller nations like Macao and the Maldives are among the most exposed by population
count. Exposed nuclear power plants are limited to Japan, China, India, Taiwan, and USA.
& 2014 Elsevier Ltd. All rights reserved.

1. Introduction tsunamis having a low probability. The 2011 Tohoku tsunami

Recent megatsunamis have changed our understanding of
how to deal with these rare and high consequence events
[1,2]. The spotlight was put on the risk to human lives, but
also on critical facilities that should withstand destructive
n
Corresponding author.
E-mail address: finn.lovholt@ngi.no (F. Løvholt).
caused remote and indirect economic consequences such as
reduced industrial production in countries not hit by the
tsunami, as well as resulting in the phasing out of nuclear
power plants in Germany [3]. These events have highlighted
the global repercussions. Such events are reinforcing the need
for a comparable basis for assessing the risk posed by
tsunamis worldwide. Previously, most hazard assessments
have been spatially constrained to local sites or regions
[4–8]. Following Løvholt et al. [9], this study is the first to

http://dx.doi.org/10.1016/j.ijdrr.2014.04.003
2212-4209/& 2014 Elsevier Ltd. All rights reserved.
 F. Løvholt et al. / International Journal of Disaster Risk Reduction 10 (2014) 406–418
cover the major earthquake sources worldwide, and to
quantify both exposed critical facilities and population living
in inundated areas.
The Hyogo Framework for Action (HFA) was adopted by
the United Nations International Strategy for Disaster Risk
Reduction (UN-ISDR) in January 2005 in order to reduce
disaster risk. As an instrument to compare the risk due to
different natural hazards worldwide, an integrated world-
wide study was implemented and published in several
bi-annual Global Assessment Reports (GAR) by UN-ISDR [3].
This paper concerns the global tsunami hazard and expo-
sure for a return period of 500 years. Contrary to a local
assessment, the present study therefore aims to capture
average trends on a country level across the globe.
Using the NGDC tsunami database [10] it is found that 54
of the reported historical tsunamis have caused a thousand or
more fatalities, with these 3% generating 97% of the total
fatalities. A majority of these were reported to occur in the last
400 years, as older records are more sparse. Hence, it is clear
that the next large tsunami disaster is linked to a low
probability. Furthermore, the sources of these tsunamis are
dominated by earthquakes (the NGDC tsunami database [10]
reports that more than 80% are due to earthquakes). Gonzalez
et al. [7] conducted a probabilistic tsunami hazard assessment
(PTHA) for the city of Seaside (USA), and found virtually no
inundation for the 100-year return period. Using a character-
istic magnitude-frequency distribution, they found that the
predicted tsunami run-up increased abruptly when the return
period was increased. As a consequence, an almost complete
inundation of Seaside was found for the 500-year return
period, and that this was predominantly due to local mega-
thrust sources. Paleotsunami information from Chile indicates
a relatively regular recurrence of major tsunamis in the order
of 500 years [11]. Based on the damage of the 2004 Indian
Ocean tsunami, Nadim and Glade [12] postulated that the
damage characteristics of the tsunami is highly non-linear,
with a rapid transition from low to high fatality for an
increasing return period. Recent damage characteristics on
buildings derived from the 2004 Indian Ocean and 2011
Tohoku tsunamis [13] also indicate a rapid transition from
small damage to total destruction as a function of the tsunami
flow depth.
The findings from the brief discussion above [7,10–13]
suggest that the tsunamis driving the risk are associated with
low probability of occurrence, and that the risk may change
rapidly from one return period to another. The findings from
Seaside and Chile are partly based on paleotsunami data,
indicating that a return period of 500 years may constitute the
main risk driver. Clearly however, this is site dependent, and
in other areas such as Portugal (see e.g. the study of Matias
et al. [8]) earthquakes having longer return periods are likely
to contribute more to the risk. In Japan, the last event of size
similar to the 2011 Tohoku tsunami is believed to be the 869
Jogan tsunami [14], indicating a somewhat longer inter event
time. Meanwhile, Japan has been hit by at least three other
earthquake induced events each with reported fatalities
exceeding 10000 [10] which are each comparable to the
losses caused by the 2011 Tohoku tsunami. To this end, Kagan
and Jackson [2], based on McCaffrey [15], indicate a return
period close to 500 years for a Mw 9.0 earthquake for Japan.
Clearly, it is difficult to define a characteristic return period
407
representative for major events globally. Yet, the indicated
500-year return period represent the order of magnitude for
the lower bound return period of major subduction zone
events. Hence, we choose to quantify the global tsunami
hazard at the 500-year return period.
Previous models for linking earthquake potential to
parameters such as lithospheric plate age, thickness, and
convergence rate to the earthquake potential has recently
been refuted [1,2]. Hence, it is presently difficult to rule
out the occurrence of large earthquakes for any major
subduction zone or fault worldwide.
2. Methods
The hazard results reported here comprises both those
originating from a scenario based method, supplemented
with tsunami hazard maps using a probabilistic tsunami
hazard assessment (PTHA) method [6] for the Indian
Ocean and the South West Pacific. However, the quantifi-
cation of the run-up and exposure are comprised by a joint
method. The different methods are briefly outlined below.
2.1. Description of the scenario based hazard assessment
method
The earthquake scenarios are confined to those with the
potential for tsunami generation due to co-seismic dip-slip
motion. For the scenario earthquakes, earthquake faults of
uniform width, length and slip are established, and in turn
converted to seabed displacement using the standard analy-
tical formula of [16]. The hydrodynamic response from the
seabed dislocation is smoothed using the model of Kajiura
[17,18]. For most of the subduction zones, new scenarios were
constructed assuming fault locking over a period of 500 years
(the exception is South America, where scenarios taken from
Løvholt et al. 2012 [9] were used). Convergence rates obtained
from Bird [19] are used. Related magnitudes were deduced
from the scaling relations of Blaser et al. [20]. By making
assumptions on the fault shear strengths, related fault lengths
and widths were in turn derived from the scaling. Typically
the shear strengths were in the range of 20–40 GPa. In the
more tectonically complex regions, including the Adriatic
Sea, Cascadia subduction zone, Sicily, eastern Indonesia, and
Portugal, worst case scenarios from the literature are used
directly or adapted [7–9,21–23]. Altogether this gives rela-
tively conservative estimates for the scenario earthquakes.
However, as discussed below, there are several other
assumptions in the overall methodology that are non-
conservative.
Near source and regional tsunami propagation are mod-
elled using a linear dispersive wave model GloBouss [24–26],
on publicly available ETOPO1 grids. For convergence, the grids
are refined to the desired resolution by bi-linear interpolation.
The maximum water level obtained from the time series at
the control points is used to compute the further amplification
to the shoreline. The near shore control points have an
approximate spacing ranging from 20 to 50 km. The control
points are extracted automatically by a contouring algorithm
(GMT [27]) at a small reference depth of 50 m.
408 F. Løvholt et al. / International Journal of Disaster Risk Reduction 10 (2014) 406–418
2.2. Description of the probabilistic tsunami hazard
assessment (PTHA) method
The probabilistic tsunami hazard assessment (PTHA)
approach describes the probability of exceedence for a given
tsunami metric. Relatively recent studies have applied the
PTHA method to different regions, e.g. [6,28–31]. A short
description of the employed PTHA method follows here, for a
more complete description we refer to [32].
The PTHA framework can be summarized as:

Define tsunami sources (earthquake faults) to be
included in the analysis.

For each source discretize the fault into smaller sub-
faults.

For each source create a synthetic earthquake catalogue
based on a recurrence model of choice (e.g. Gutenberg–
Richter or Characteristic), which has probabilities asso-
ciated with each earthquake.

For each sub fault, calculate the unit seafloor deforma-
tion and propagate the tsunami from source to the
control points at the reference depth.

For each event in the catalogue, estimate the maximum
water level at the near shore control point by summing
the waves from all the individual sub faults that make
up that event, and then scale by the amount of slip for
that event.

Combine the maximum water level from all sources to
estimate the probability of exceedence.
The subduction zone geometry and recurrence rates
were taken from the PTHA for Australia [8], which uses
plate velocity vectors from GPS data to estimate the
magnitude frequency distribution assuming full coupling
on the plate interface. Sub faults for local crustal sources
are 20 km  10 km, whereas sub faults that are distant
only are 100 km  50 km.
In deep water the tsunami is linear, meaning that any
tsunami can be constructed by the summation of the
responses from the sub faults. Hence, the simulations are
only carried out ones for the sub faults, and the hazard is
determined by superpositioning. In the PTHA a linear finite
difference model allowing for nesting formulated in geogra-
phical coordinates [33] is used for the simulation of the
tsunami propagation for the unit sources. As for the worst
case scenario simulations the maximum water elevation was
extracted at the 50 m reference depth.
Sources of epistemic uncertainty (uncertainties due to lack
of knowledge) that are included in the PTHA are slip rate,
earthquake recurrence model type, and maximum magnitude
(Fig. 1). Maximum magnitudes are constrained by scaling laws
[20]. The maximum magnitude from the mean of the scaling
laws is given a weighting of 0.6, and two alternative max-
imum magnitudes that are þ0.2 magnitude units and  0.2
magnitude units from the best estimate, are given a weighting
of 0.2. For each source, a truncated Gutenberg–Richter Mag-
nitude Frequency Distribution (MFD) was given a weighting of
0.66 and a Characteristic Earthquake distribution was given a
weighting of 0.34. A b-value of 1.0 is used for both MFD’s.
Main sources of aleatory uncertainty (inherent uncertainty) in
Fig. 1. Outline of PTHA logic tree.
the PTHA come from modelling uncertainties in source
geometry, and random slip. The aleatory uncertainty was
accounted for by summing up different variances from model
errors, fault dip, and random fault slip. The uncertainties in
dip and random slip were obtained from Monte Carlo
simulations by varying the dip angle and employing the
different slip realizations, respectively. Aleatory uncertain-
ties were included by integrating across probability density
functions.
Combining all the information from the sources and logic
trees, a synthetic catalogue is generated which represents the
full integration over earthquake magnitudes, locations and
sources for every logic tree branch. The catalogue was
generated by iterating through each magnitude in the MFD,
and calculating the rupture dimensions using the scaling laws
[20]. The rupture is then iteratively moved across the fault one
sub fault at a time until that magnitude has occurred on every
possible location within the fault dimensions. For M7.0 earth-
quakes on the subduction interface, the rupture dimensions
are equal to one sub fault; therefore the number of ruptures
would be equivalent to the number of sub faults. The
maximum magnitude earthquake would occur once and
rupture the whole fault if scaling laws have been used to
constrain the maximum magnitude. This iterative process
ensures that all magnitudes could occur at any possible
location on the fault plane. For each event the probability of
that magnitude was then weighted by one over the number of
earthquakes represented by that magnitude. This ensures that
the sum of the events of the same magnitude equals the
annual probability of one event of that magnitude.
For each event in the synthetic catalogue, the tsunami
hazard is calculated at each control point along the coast
by summing the contributions from the sub faults that
make up that event, and by scaling the tsunami height by
the event slip. For each site, this results in a list of tsunami
heights and associated annual probabilities. For a coherent
description of the hazard compared to the worst case
scenario simulations, maximum surface elevation from the
PTHA for a return period of 500 years is given at near shore
control points at the reference depth of 50 m. The further
amplification to run-up is accounted for using the ampli-
fication factor method (Section 2.3). The present model
 F. Løvholt et al. / International Journal of Disaster Risk Reduction 10 (2014) 406–418
assumes a Poisson process where earthquakes are inde-
pendent and occur at a fixed rate over time.
2.3. Run-up estimation and inundation mapping using
amplification factors
To estimate tsunami run-up globally refined numerical
inundation simulations are too time consuming (yet,
examples of locally refined simulations are covering an
ocean wide scale have previously been accomplished
by means of adaptive unstructured meshes, see e.g. [34]).
A faster procedure is to relate the near-shore surface
elevations to the maximum shoreline water levels by using
a set of amplification factors based on the parameters of
the incident wave and the bathymetric slope. This proce-
dure is described and validated in detail by [9], and only a
part of it is reviewed here. The method of amplification
factors is sketched in Fig. 2. The basis for the procedure is a
large set of pre-computed plane wave numerical simula-
tions in idealized two dimensional (2D) transects with an
Fig. 2. Principles of the amplification factor method. Upper panel,
regional tsunami simulation and locations of the time series gauges at
the reference depth contour. Mid panel, sketch of an idealized bathy-
metric profile. The amplification factor is defined as the ratio between the
water surface elevation at the shoreline over the water surface elevation
at 50 m water depth. Lower panel, maximum shoreline water level
obtained from superimposing results from a series of simulations.
409
incident sine shaped N-wave as input. Parameters such as
bathymetric profile, leading trough or leading peak, and
wave period are varied. The plane wave simulations are all
run on idealized plane bathymetric configurations [9]
where the shelf is broken up into two linear segments.
From the plane wave simulations, factors for amplification
that relate the surface elevation at time series gauges
located at water depths of 50 m to the maximum shoreline
water level are computed and stored in lookup tables.
To determine the amplification factors along the idealized
bathymetric profiles we apply a linear hydrostatic plane
wave model. For smaller islands the plane wave assump-
tion is severely violated, a 2HD model (GloBouss) is
applied. Both models apply a no-flux boundary condition
at the shoreline leading to a doubling of surface elevation
due to reflection. An idealized bathymetric profile is
manually assigned to each point. To estimate the max-
imum shoreline water level from the offshore time series
gauges in a tsunami simulation, the amplification factor for
a set of parameters is extracted from the lookup tables and
in turn multiplied with the maximum surface elevation
measured at the time series gauges. Although the models
do not include dry land inundation, the surface elevation
on the boundary close to the shoreline (at 0.5 m water
depth) with a no-flux condition yields a good approxima-
tion. For long non-breaking waves, the linear solution for
the run-up height at the shoreline and the non-linear
solution for the run-up height on land are identical [35].
The validation of the procedure is presented in [9]. Based
on the findings of Pedersen [36], we may further assume
that the procedure should also provide reasonably accu-
rate results for waves of moderately oblique incidence.
For reviews of numerical methods for run-up estimation,
see [37–39].
2.4. Mapping the inundation
Resulting run-up heights were projected from the offshore
points to the shoreline by nearest neighbor interpolation. In
turn, an inverse distance weighted method was used to
determine the inundated area, using the SRTM topography
by [40]. It also turned out that for some very flat near shore
locations, the inundation distance could be unreasonably high.
To limit the inundation, a crude formula taking into account
the head loss due to bottom friction was used. We represent
the wave load at the shoreline by a constant surface elevation
η and choose a friction coefficient f¼10  2. This friction is
relatively high, slightly counterbalancing the conservative
assumption of fault locking, yet providing reasonable inunda-
tion distances compared to real events. By assuming a
quadratic friction law and a constant drop of hydraulic head
loss along the inundation path, a simple formula for the
maximum inundation distance Lmax was obtained, propor-
tional to the ratio of the surface elevation over the friction:
η
Lmax o
f
The 500-year return period inundation maps are over-
lain by population exposure data [41], urban produced
capital data [42], and critical facility data for nuclear
power plants [43] to integrate total exposed elements in
410 F. Løvholt et al. / International Journal of Disaster Risk Reduction 10 (2014) 406–418

each country. The status of the nuclear power plants in
Japan has been updated manually, as most of them have
been shut down following the 2011 Tohoku earthquake.
The exposed population is taken as the cell population,
times the inundated cell area over the total Landscan cell
area. The exposed urban produced capital is quantified as
the cell urban produced capital times the inundation cell
area. The produced capital is the total value of machinery,
equipment, structures and urban land. The urban pro-
duced capital is defined as the capital in urban areas with
more than 2000 inhabitants [3].
It should be noted that inundation maps are based on
coarse topographic data (SRTM) hampered with inaccuracies
and falsely elevated land. This may lead to an underestimation
of the inundation and therefore also the exposure. The effect
is particularly pronounced in tropical areas [44], but may also
play an important role in urban areas. Moreover, the effects of
countermeasures such as breakwaters which are expected to
decrease the exposure are not considered. Breakwaters are for
instance common in Japan.
3. Comparison of the scenario based method and PTHA
for the tsunami hazard and population exposure in the
Indian Ocean
Hazard maps in the Indian Ocean using both the
scenario based and the PTHA method are displayed for
four different areas, namely the Arabian Sea (Fig. 3), the
coastlines of the western Bengal Bay (Fig. 4), the coastlines
of the eastern Bay of Bengal and the Andaman Sea (Fig. 5),
and along the western coast of Sumatra. Furthermore,
the model sensitivity (scenario based vs PTHA) on the
computed exposed population is demonstrated for four
different countries in Table 1. The sensitivity to using the
(biased) STRM topographic data is also briefly addressed
by uniformly subtracting a value of 2 m and comparing the
exposed population with those obtained by the regular
SRTM data.
The Makran fault zone is located along southern coast-
lines of Iran and Pakistan facing the Arabian Sea. Fig. 3
shows the simulated run-up along the northern part of the

Fig. 3. Upper panel, simulated run-up using the method of amplification factors along the Makran trench using the scenario based approach (blue bars),
and PTHA (red bars). Lower panel, PTHA run-up normalized by the scenario based run-up (black bars). Here, perfect agreement is defined for a unitary
normalized run-up (red curve). It is noted that extent of the domains for the two different methods (scenario based and PTHA) deviate, and that
comparison cannot be obtained for these points. The coloured contours depicts the bathymetric and topographic isolines. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)
 F. Løvholt et al. / International Journal of Disaster Risk Reduction 10 (2014) 406–418 411

Fig. 4. Upper panel, simulated run-up using the method of amplification factors along the western coastlines of Bengal using the scenario based approach
(blue bars), and PTHA (red bars). Lower panel, PTHA run-up normalized by the scenario based run-up (black bars). Here, perfect agreement is defined for a
unitary normalized run-up (red curve). It is noted that extent of the domains for the two different methods (scenario based and PTHA) deviate, and that
comparison cannot be obtained for these points. The coloured contours depicts the bathymetric and topographic isolines. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)

Arabian Sea. Due to the vicinity of the coastlines to the
Makran trench, the maximum run-up in this region
expected to be from this fault zone. Near the fault zone,
large run-up often exceeding 5 m and sometimes 10 m is
found. In this region near the source, there is a clear
tendency that the scenario based method yields higher
run-up than for PTHA. Typically, the run-ups using the
scenario based method are factors 1.5–2 larger than the
run-ups obtained from PTHA. In certain regions such as in
Iranian, the difference may be even larger along short sections
of the coastline. In the far-field coastal stretches along Oman
and India however, the agreement between the two models
are close to unitary. The exception is the northernmost section
of the Oman coastline, where the scenario based method
again produces much higher run-ups than the PTHA. The
reason for the large model discrepancies here and along the
western Pakistan, the Iranian, and the northern Oman coast-
lines, is that the westernmost Makran trench scenario are
extending further west than the corresponding unit sources
from the PTHA method. The observed discrepancy in the
computed run-up is also reflected different population expo-
sure as exemplified in Table 1 for Pakistan, with the scenario
412 F. Løvholt et al. / International Journal of Disaster Risk Reduction 10 (2014) 406–418

Fig. 5. Upper panel, simulated run-up using the method of amplification factors along the eastern coastlines of the Bay of Bengal and the Andaman Sea
using the scenario based approach (blue bars), and PTHA (red bars). Lower panel, PTHA run-up normalized by the scenario based run-up (black bars). Here,
perfect agreement is defined for a unitary normalized run-up (red curve). It is noted that extent of the domains for the two different methods (scenario
based and PTHA) deviate, and that comparison cannot be obtained for these points. The coloured contours depicts the bathymetric and topographic
isolines. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

based method providing about twice as large exposure as the
PTHA. This is not unexpected as the differences in run-up are
substantial.
Figs. 4 and 5 show the simulated run-up along western
and eastern part of the Bengal Bay and the Andaman Sea.
Here, the maximum run-up typically originate from the
northern section of the Sunda Arc (the subduction zone
extending from the Andaman Islands to Sumbawa in
Indonesia), as well as from the Burma trench. The Indian
and Sri-Lankan coastlines (Fig. 4) typically exhibit 500-
year return period run-up ranging from 3 to 5 m, with the
highest run-up in the South, gradually reducing north-
wards. Along the eastern Bay of Bengal and Andaman Sea
coastlines, the run-up may extend up 10 m. In Sri-Lanka
and along the southern part of India, a close model
agreement is obtained. Despite this overall good agree-
ment, Table 1 shows that the PTHA method yet provide a
much larger exposure. This is interpreted as a result of the
larger run-up obtained by the PTHA method near impor-
tant urban areas such as Trincomalee and Batticaloa on the
 F. Løvholt et al. / International Journal of Disaster Risk Reduction 10 (2014) 406–418
Table 1
Compared population exposure (exposed population in thousand people)
using PTHA and the scenario based method. The effect of uniformly
subtracting 2 m from the topographic SRTM data on the estimated is also
estimated.
Scenario PTHA Scenario based PTHA
based SRTM SRTM SRTM—2 m SRTM—2 m
Myanmar 284 253 371 312
Sri-Lanka 109 209 222 335
Pakistan 47 24 70 49
India 209 529 357 851
east Sri-Lankan coast. Further north along the Indian
coastline, the PTHA model provide the highest run-up,
up to factors of 1.5–2 times that of the scenario based
approach. The larger run-up obtained along by the PTHA is
also reflected in a higher population exposure as exempli-
fied in Table 1 for India. The reason for the discrepancy
here is that the northernmost extension of the PTHA unit
sources along the Sunda Arc are further north than the
location of scenario events in the same region. A similar
trend is found for the Burmese coastline, yet here the
population exposure compare favourably for the two
methods. At the coastlines of the Andaman Islands and
Nicobar, relatively similar results are found using the two
methods. Along the Thailand coastline, a good agreement
is found for run-ups in the regional north of Phuket,
whereas the scenario based method is typically factors
1.5–2 larger than the run-ups obtained from PTHA south of
Phuket.
Fig. 6 shows the simulated run-up along western
Sumatra as well as at the coastlines of the nearby forearc
islands. The typical simulated run-up is 10 m. As shown,
there is a tendency that the scenario based method
provides larger run-up compared to the PTHA. Exceptions
are the south part of Sumatra, where PTHA provides the
higher results, and on northern Sumatra and on the forearc
islands where a relatively close agreement is found.
The simulated run-up over the different sections of
coastlines discussed above suggests that the two methods
provide a comparable hazard, however with a clear ten-
dency that the scenario-based method provides the higher
run-up. In some regions such as the Bay of Bengal, the
agreement between the methods are very close given
the uncertainties in questions. In other regions such as
Makran and Sumatra, there seems to be a tendency that
the scenario based method systematically provides larger
run-up. This is interpreted as due to the assumption of
fault locking. In certain areas, large model discrepancies
may appear as a result of different extents or position of
the source locations using the two models. It may seem
as the population exposure values exhibit larger model
related differences than the individual hazard. This is
interpreted as due to the alongshore sensitivity to the
run-up representation in densely populated areas. Using
Sri-Lanka as the example, we see that although the mean
run-up is generally comparable, the deviations in simu-
lated run-up close to two of the largest coastal commu-
nities on the East coast give rise to a large effect on the
calculated exposure. As demonstrated, the exposure is
413
expected only to be within the same order of magnitude.
By subtracting uniformly 2 m from the SRTM data, we find
that the population exposure increase up to a factor 2 for
Sri-Lanka, and somewhat smaller factors for the other
countries. Knowing that the SRTM data has a particularly
positive bias in tropical and urban areas [44], we are led to
the conclusion that the present exposure based on SRTM
are lower bound estimates.
4. Results—Global hazard and exposure
Fig. 7 shows the predicted earthquake-induced tsunami
run-up height for a 500-year return period at the global
scale. In the Pacific Ocean, run-up exceeding 5 m is found
in South America, southern Central America, the states of
Oregon, Washington, and Alaska, Kamchatka, Japan, parts
of China, the Philippines and Papua New Guinea. In the
Indian Ocean, we find run-up exceeding 5 m in parts of
the Bay of Bengal and Andaman Sea, Sumatra, Java, north-
western Australia, Pakistan, Oman, and eastern Iran.
The hazard in the above mentioned areas is related to major
subduction zones including the Pacific Ring of Fire, the Sunda
Arc and the Makran trench. Run-up in this range is also
found adjacent to the smaller and more complex fault zones
in the Mediterranean as well as in eastern Indonesia.
Populous Asian countries, most prominently Japan, but
also China and Indonesia account for a large absolute
number of people living in tsunami prone areas (Fig. 8).
Although large run-up heights are also predicted in USA,
South America, and parts of Central America, the popula-
tion exposure is smaller. Normalizing the exposure by the
total population, smaller countries such as Macao and the
Maldives have relative population exposures similar to
Japan (Fig. 8). High, relative exposure is further dominated
by small island states (in the Indian Ocean, southwest
Pacific nations, and the Caribbean) and South American
countries.
Fig. 9 shows that Japan accounts for the majority of the
total exposed urban produced capital. Countries such as
Greece, Italy, and Australia are ranked relatively high in
terms of economic exposure despite being ranked lower in
terms of population exposure. The exposed urban pro-
duced capital is normalized against the countries total,
with a several small island states having large relative
exposure. The amount of steel and reinforced masonry
buildings subject to major tsunami damage increases from
20–30% to 80–90% when the tsunami water level is
increased from 2 to 5 m [13]. Although this study do not
address probability of loss, it is apparent that water levels
exceeding 5 m will result in almost total economic losses
locally.
Statistics covering the nuclear power plants within the
inundation zone as well as facilities located within 1 km
from the coastline is provided. The statistics for the
facilities located less than 1 km from the coastline is
included mainly because the employed methodology for
computing the inundation is considered less accurate than
inundation maps obtained by using local inundation
models. Nuclear power plants should withstand loads of
much larger return periods than 500 years. Non-seismic
sources such as submarine landslides have source
414 F. Løvholt et al. / International Journal of Disaster Risk Reduction 10 (2014) 406–418

Fig. 6. Upper panel, simulated run-up using the method of amplification factors along coastlines Sumatra using the scenario based approach (blue bars),
and PTHA (red bars). Lower panel, PTHA run-up normalized by the scenario based run-up (black bars). Here, perfect agreement is defined for a unitary
normalized run-up (red curve). It is noted that extent of the domains for the two different methods (scenario based and PTHA) deviate, and that
comparison cannot be obtained for these points. The coloured contours depicts the bathymetric and topographic isolines. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)

characteristics that enable larger local run-up than tsuna-
mis induced by an earthquake [45], and therefore repre-
sent lower probability tsunami hazard, but with
potentially higher consequences [46].
Fig. 10 depicts the number of inundated and near shore
nuclear power plants per country. Japan and China have
the largest number of nuclear power plants within the
inundated area. The Japan Sea is omitted for this study; it
is therefore possible that the exposure in Japan and the
Republic of Korea should be higher. However, the majority
of the nuclear power plants in Japan are presently sus-
pended or being shut down, whilst in other Asian coun-
tries such as China, Taiwan, and India, there are nuclear
power plants currently under construction at locations less
 F. Løvholt et al. / International Journal of Disaster Risk Reduction 10 (2014) 406–418
Fig. 7. The global tsunami hazard map displaying the run-up height for a return period of.500 years. Panels A–D shows close ups of Europe, North America,
East Asia, and South America, respectively.
than 1 km from the shoreline according to the nuclear
power plant database [43]. Only 2 out of 13 of the nuclear
power plants in USA located less 1 km from the coastline
are exposed, as most are located along the Atlantic shore
and subject to lower earthquake induced tsunami hazard.
Similarly, the UK has seven nuclear power plants located
less than 1 km from the coastline, of which none lies
within the inundation zone. It is noted that evidence
of past large submarine landslides having tsunamigenic
potential exists offshore both USA [47] and offshore Norway
[48], which may indicate a potential high consequence low
probability tsunami hazard.
5. Concluding remarks
This analysis concludes that the population exposure
due to large earthquake induced tsunamis is close to 29
million people globally, which is about twice as much as
415
reported previously [9]. The increased exposure is due to a
more comprehensive coverage of both earthquake sources
as well as inclusion of a larger section of the shorelines.
The majority of the exposed populations are located in
East and Southeast Asia, with Japan accounting for more
than half of the exposed population worldwide. South
American and smaller island countries are among the
highest ranked for exposure relative to their population.
Despite recent focus on cascading risks, this analysis
demonstrates that the number of nuclear power plants
exposed to large earthquake induced tsunamis are limited.
One remaining question surrounds hazard related to non-
seismic sources, and whether critical facilities outside the
identified hazard zone may be subject to tsunami threat
from such sources.
For most of the areas, a relatively simple scenario based
method is used on the expense of the more sophisticated
PTHA method. The examples from the Indian Ocean shows
416 F. Løvholt et al. / International Journal of Disaster Risk Reduction 10 (2014) 406–418

Fig. 8. Left panel; number of people living in areas potentially affected by tsunamis for a 500-year return period. Right panel; number of exposed persons
divided by the total population in each country in percent.

Fig. 9. Left panel; exposed urban produced capital in MUSD due to tsunamis for a 500-year return period. Right panel; produced urban capital divided by
the total population in each country in percent.

that resulting 500-year run-up using the two methods are
roughly comparable in the average sense. In certain
regions such as in the Bay of Bengal, close to a one-to-
one agreement is found. In other regions however, there
seems to be a systematic tendency that the scenario based
method provides a higher run-up, typically providing run-
up that is a factor 1.5–2 higher than for the PTHA. We assume
that this is partly due to the assumption of fault locking using
the scenario based method. Locally, larger deviations are also
found, but these are related to different extensions of the fault
zones comprising the tsunami sources. The population expo-
sure in four different countries compared for the scenario
based and the PTHA methods is model sensitive, and devia-
tions of more than a factor two occur even if the average run-
up exhibit fairly similar value. A possible explanation for the
apparently large sensitivity may be that the coastal fluctua-
tions in run-up become crucial when correlated with the
coastal population pattern. Based on the comparison, it is
expected that a PTHA method would be able to provide
refined accuracy, and should be attempted in the future. This
would also enable hazard quantification at different return
periods. Yet, as the tsunami risk is dominated by events with
high return periods, a larger uncertainty should still be related
to establishing the source probability at high return periods.
 F. Løvholt et al. / International Journal of Disaster Risk Reduction 10 (2014) 406–418 417

Fig. 10. Total number of all nuclear power plants located within 1 km from the coastline (both exposed and not exposed). Left panel; red coloured bars
denotes exposed power plants, whereas blue denotes power plants not within the simulated inundation zone. Right panel; operational status of the
nuclear power plants. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

It is further noted that the sensitivity due to the topographic
data is found to be prominent, and that the SRTM data used to
derive inundation maps is expected to provide a significant
bias [44]. The present analysis indicate that use of refined
topographic data should expectedly produce larger exposure
numbers.
It should also be noted that tsunamis generated by
volcanoes, submarine landslides, and smaller earthquakes
are not addressed in the present study. Non-seismic
sources contribute to the generation of about one fifth of
all tsunamis globally, and there are several examples of
such tsunamis causing devastation, a recent example is the
1998 Papua New Guinea tsunami caused by a submarine
landslide. In areas like eastern Indonesia [23] and the
Caribbean [30,49] tsunamis due to landslides and volca-
noes are relatively more frequent, and contribute to a
significant portion of the risk. It has also recently been
claimed that large run-up in northern Japan following the
2011 Tohoku tsunami was induced by a huge submarine
slump [50]. Unlike earthquakes, landslides are not con-
strained to the major subduction zones and may strike
more surprisingly. Due to their source characteristics, they
may generate larger run-up locally compared to earth-
quakes, but are generally less dangerous for the far field
propagation (for a discussion of their hazard, see e.g. [46]).
However, addressing their return periods is difficult.
Acknowledgements
This work has been co-funded funded by UN-ISDR
(Grant No. UNISDR/GR/2013/022) and the Australian
Agency for International Development (AusAID). Andrew
Maskrey and Manuela di Mauro at UN-ISDR are thanked
for co-ordinating the tsunami hazard assessment with the
risk assessment for the other natural hazards. We thank
UNEP-GRID Geneva for providing the critical facility data-
bases for the exposure calculations. Dr. Phil Cummins and
Dr. Victoria Miller at Geoscience Australia are thanked for
constructive comments improving the manuscript. The
applied methods were reviewed by independent experts
from UNESCO, including Jö rn Behrens (Professor, Univer-
sity of Hamburg, Germany), Stefano Tinti (Professor, Uni-
versity of Bologna, Italy), and Kenji Satake (Professor,
University of Tokyo, Japan), which are all thanked for their
constructive suggestions. We also thank one anonymous
reviewer for helpful and constructive comments. This
work was published with the permission of the CEO of
Geoscience Australia.
References
[1] Stein S, Okal E. Ultralong period seismic study of the December 2004
Indian Ocean earthquake and implications for regional tectonics
and the subduction process. Bull Seismol Soc Am 2007;97(1A):
S279–95 (S).
[2] Kagan YY, Jackson DD. Tohoku earthquake: a surprise? Bull Seismol Soc
Am 2013;103(2B):1181–94, http://dx.doi.org/10.1785/0120120110.
[3] UN-ISDR. Global assessment report on disaster risk reduction—from
shared risk to shared value: the business case for disaster risk
reduction. Report (2013).
[4] K Berryman, (editor). Review of tsunami hazard and risk in New
Zealand. Report (2005).
[5] Lorito S, Tiberti M, Basili R, Piatanesi A, Valensise G. Earthquake-
generated tsunamis in the Mediterranean Sea: scenarios of potential
threats to southern Italy. J Geophys Res 2008;B01301.
418 F. Løvholt et al. / International Journal of Disaster Risk Reduction 10 (2014) 406–418
[6] Burbidge D, Cummins P, Mleczko R, Thio H. A probabilistic tsunami
hazard assessment for Western Australia. Pure Appl Geophys
2008;165:2059–88.
[7] Gonzalez F, Geist E, Jaffe B, Kanoglu U, Mofjeld H, Synolakis C, et al.
Probabilistic tsunami hazard assessment at Seaside, Oregon, for near-
and far-field seismic sources. Pure Appl Geophys 2009;114:C11023.
[8] Matias LM, Cunha T, Annunziato A, Baptista MA, Carrilho F. Tsunami-
genic earthquakes in the Gulf of Cadiz: fault model and recurrence.
Nat Hazard Earth Syst Sci 2013;13:1–13.
[9] Løvholt F, Glimsdal S, Harbitz C, Zamora N, Nadim F, Peduzzi P, et al.
Tsunami hazard and exposure on the global scale. Earth Sci Rev
2012;110:58–73.
[10] NOAA/WDC Historical Tsunami Database at NGDC. URL: 〈www.ngdc.
noaa.gov/seg/hazard/tsu db.html〉. (2013).
[11] Cisternas M, Cisternas M, Atwater B, Torrej´on F, Sawai Y, Machuca
G, et al. Predecessors of the giant 1960 Chile earthquake. Nature
2005;437:404–7.
[12] F Nadim, T Glade, On tsunami risk assessment for the west coast of
Thailand. In Nadim, F., Pö ttler, R., Einstein, H., Klapperich, H. &
Kramer, S. editors. In: ECI Symposium Series, 7; (2006).
[13] Suppasri A, Mas E, Charvet I, Gunasekera G, Imai FY, Abe K,Y, et al.
Building damage characteristics based on surveyed data and fragility
curves of the 2011 great east Japan tsunami. Nat Hazard 2013;66:
319–41.
[14] Sugawara D, Imamura F, Goto K, Matsumoto H, Minoura K. The 2011
Tohoku-oki earthquake tsunami: similarities and differences to the
869 Jogan tsunami on the Sendai plain. Pure Appl Geophys
2013;170:831–43.
[15] McCaffrey R. Global frequency of magnitude 9 earthquakes. Geology
2008;36(3):263–6.
[16] Okada Y. Surface deformation due to shear and tensile faults in a
half-space. Bull Seismic Soc Am 1985;74(4):1135–54.
[17] Kajiura K. The leading wave of a tsunami. Bull Res Inst 1963;41:
535–71.
[18] Glimsdal S, Pedersen G, Harbitz CB, Løvholt F. Dispersion of
tsunamis; does it really matter? Nat Hazard Earth Syst Sci 2013;13:
1507–26.
[19] Bird P. An updated digital model of plate boundaries. Geochem Geo-
phys Geosyst 2003;4(3):1027, http://dx.doi.org/10.1029/2001GC000252
(52 pp.).
[20] Blaser L, Krüger F, Ohrnberger M, Scherbaum F. Scaling relations of
earthquake source parameter estimates with special focus on
subduction environment. Bull Seis Soc Am 2010;100:2914–26.
[21] Tiberti M, Lorito S, Basili R, Kastelic V, Piatanesi A, Valensise G.
Scenarios of earthquake-generated tsunamis for the Italian coast of
the Adriatic Sea. Pure Appl Geophys 2010;165:2117–42.
[22] Tinti S, Armigliato A, Zaniboni F, Pagnoni G. Influence of the
heterogeneity of the seismic source on the timely detectability of
a tsunami: implications for tsunami early warning in the central
Mediterranean. In: Proceedings of the 22nd international and polar
engineering conference; ISOPE-2012.
[23] Løvholt F, Kühn D, Bungum H, Harbitz CB, Glimsdal S. Historical
tsunamis and present tsunami hazard in eastern Indonesia and the
southern Philippines. J Geophys Res 2012;117(B09310). http://dx.
doi.org/10.1029/2012JB009425.
[24] Pedersen G, Løvholt F. Documentation of a global Boussinesq solver,
preprint series in applied mathematics 1. Norway: Dept. of Mathe-
matics, University of Oslo; 2008.
[25] Løvholt F, Pedersen G, Gisler G. Oceanic propagation of a potential
tsunami from the La Palma Island. J Geophys Res 2008;113:
C09026, http://dx.doi.org/10.1029/2007JC004603.
[26] Løvholt F, Pedersen G, Glimsdal S. Coupling of dispersive tsunami
propagation and shallow water coastal response. In: Zahibo, N.,
Pelinovsky, E., Yalciner, A., and Titov, V. editors. In: Proceedings of
the “Caribbean Waves 2008” workshop in Guadeloupe Dec. 2008.
The Open Oceanography Journal special volume; 2010.
[27] GMT, The generic mapping tool; (2011) URL: 〈http://gmt.soest.
hawaii.edu/〉.
[28] Geist E, Parsons T. Probabilistic analysis of tsunami hazards. Nat
Hazard 2006;37:277–314.
[29] Annaka T, Satake K, Sakakiyama T, Yanagisawa K, Shuto N. Logic-tree
approach for probabilistic tsunami hazard analysis and its applica-
tions to the Japanese coasts. Pure Appl Geophys 2007;164:577–92.
[30] Parsons T, Geist E. Tsunami probability in the Caribbean Region.
Pure Appl Geophys 2009;165:2089–116.
[31] Thio HK, Somerville P, Polet J. Probabilistic tsunami hazard in
California, PEER report 2010/108 Pacific Earthquake Engineering
Research Center, 2010.
[32] Horspool N, Pranantyo NI, Griffin J, Latief H, Natawidjaja DH, Kongko
W, et al. A Probabilistic Tsunami Hazard Assessment for Indonesia.
Manuscript in preparation.
[33] Satake K. Linear and non-linear computations of the 1992 Nicaragua
earthquake tsunami. Pure Appl Geophys 1995;144:455–70.
[34] Harig S, Chaeroni, Pranowo W, Behrens J. Tsunami simulations on
several scales. Ocean Dyn 2008;58(5-6):429–40.
[35] Carrier GF, Greenspan HP. Water waves of finite amplitude on a
sloping beach. J Fluid Mech 1958;4:97–109.
[36] Pedersen G. Oblique runup of non-breaking solitary waves on an
inclined plane. J Fluid Mech 2011;668:582–606, http://dx.doi.org/
10.1017/S0022112010005343.
[37] Synolakis CE, Bernard EN, Titov VV, Kânoglu U, Gonzaléz, F. F.
Validation and verification of tsunami numerical models. Pure Appl
Geophys 2007;165:2197–228.
[38] Pedersen G. Modeling run-up with depth integrated equation
models. In: Liu PL-F, Yeh H, Synolakis C, editors. Advanced numerical
models for simulating tsunami waves and run-up. World Scientific;
2008. p. 3–41.
[39] Løvholt F, Lynett P, Pedersen G. Simulating run-up on steep slopes
with operational Boussinesq models; capabilities, spurious effects
and instabilities. Nonlinear Processes Geophys 2013;20:379–95,
http://dx.doi.org/10.5194/npg-20-379-2013.
[40] Jarvis A, Reuter H, Nelson A, Guevara E. Hole-filled seamless SRTM
data v4, International Centre for Tropical Agriculture, 2013.
[41] LandScan. High resolution global population dataset ©Operator of
Oak Ridge National Laboratory, USA, Dataset is available upon
demand to ONRL; (2012).
[42] World Bank. The changing Wealth of Nations: measuring sustain-
able development in the new millennium. Washington DC: World
Bank; 2011.
[43] UNEP/GRID-Geneva. Nuclear power stations of the world, geodata-
set generated from various sources, situation as of December 2010;
(2011).
[44] Rö mer H, Willroth P, Kaiser G, Vafeidis AT, Ludwig R, Sterr H, et al.
Potential of remote sensing techniques for tsunami hazard and
vulnerability analysis—a case study from Phang-Nga province, Thai-
land. Nat Hazard Earth Syst Sci 2012;12:2103–26.
[45] Okal EA, Synolakis CE. Source discriminants for near-field tsunamis.
Geophys J Int 2004;158:899–912.
[46] Harbitz CB, Løvholt F, Bungum H. Submarine landslide tsunamis:
how extreme and how likely? Natural Hazard 2013:1–34,
http://dx.doi.org/10.1007/s11069-013-0681-3 (URL).
[47] Chaytor J, ten Brink US, Solow A, Andrews B. Size distribution of
submarine landslides along the U.S. Atlantic margin. Mar Geology
2009;264:16–27 (URL).
[48] Bryn P, Berg K, Forsberg C, Solheim A, Kvalstad T. Explaining the
Storegga slide. Mar Pet Geol 2005;22:11–9.
[49] Harbitz CB, Glimsdal S, Bazin S, Zamora N, Løvholt F, Bungum H,
et al. Tsunami hazard in the Caribbean: regional exposure derived
from credible worst case scenarios. Cont Shelf Res 2012;8:
1–23, http://dx.doi.org/10.1016/j.csr.2012.02.006.
[50] Grilli ST, Harris JC, Tajali Bakhsh TS, Tappin DR, Masterlark, T. T, Kirby
JT, et al., Recent progress in the nonlinear and dispersive modelling
of tsunami generation and coastal impact: application to Tohoku
2011, 13èmes Journées de l´Hydrodynamique, 21–23 November 2012,
Chatou, France.