You are on page 1of 17

International Journal of Disaster Risk Reduction 54 (2021) 102050

Contents lists available at ScienceDirect

International Journal of Disaster Risk Reduction


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

Multi-hazard earthquake-tsunami loss estimation of Kuroshio Town, Kochi


Prefecture, Japan considering the Nankai-Tonankai megathrust
rupture scenarios
Katsuichiro Goda a, b, *, Raffaele De Risi c, Flavia De Luca c, Ario Muhammad c,
Tomohiro Yasuda d, Nobuhito Mori e
a
Department of Earth Sciences, Western University, London, Ontario, Canada
b
Department of Statistical and Actuarial Sciences, Western University, London, Ontario, Canada
c
Department of Civil Engineering, Bristol University, Bristol, United Kingdom
d
Kansai University, Suita, Osaka, Japan
e
Disaster Prevention Research Institute, Kyoto University, Kyoto, Japan

A R T I C L E I N F O A B S T R A C T

Keywords: The Nankai-Tonankai Trough is the primary source of major subduction earthquakes in western and central
Multi-hazard loss estimation Japan, and the preparedness against future Nankai-Tonankai megathrust subduction events is of critical
Earthquake importance. This study develops a novel multi-hazard catastrophe model for the future Nankai-Tonankai meg­
Tsunami
athrust subduction earthquake and tsunami based on a stochastic earthquake source modeling approach, and
Nankai-Tonankai trough
Critical hazard-risk map
applies it to Kuroshio Town, Kochi Prefecture, Japan, that faces significant seismic-tsunami hazards due to the
Nankai-Tonankai subduction earthquakes. The earthquake-tsunami catastrophe model incorporates the sto­
chastic rupture sources, spatially correlated ground motion fields, tsunami inundation simulations, detailed
building portfolio data, seismic and tsunami fragility models, and building damage cost estimation. It produces
the multi-hazard and single-hazard loss distributions, accompanied by detailed earthquake rupture scenarios,
shaking-tsunami hazard intensity distributions, and building damage distributions. The earthquake-tsunami loss
estimation results for Kuroshio Town indicate that the tsunami loss tends to increase with the magnitude more
significantly than the shaking loss and contributes more towards the increase in the multi-hazard total loss with
the earthquake magnitude. Importanly, the new multi-hazard tool facilitates the identification of critical multi-
hazard loss scenarios and leads to integrated hazard-risk maps that are particularly useful for disaster risk
reduction and management purposes.

1. Introduction secondary peril. Earthquake loss due to secondary peril is not negligible
and can exceed the primary shaking loss [6]. In evaluating the earth­
People and assets in active subduction regions are exposed to sig­ quake impact due to a future megathrust subduction event, it is essential
nificant seismic and tsunami risks due to possible megathrust subduc­ to consider multi-hazard loss generation processes because spatial pat­
tion earthquakes. In the last two decades, moment magnitude (M) 9- terns for shaking and tsunami damage/loss are different.
class megathrust events caused catastrophic consequences in Indian Probabilistic multi-hazard analysis is the fundamental step for
Oceans, Chile, and Japan (e.g. Refs. [1–3]. Globally, there are other assessing natural disaster risk accurately [7]. A multi-peril methodology
subduction regions, where M9-class subduction earthquakes are ex­ that allows the risk quantification in economic terms is necessary to
pected to occur in the future, including the Cascadia region off the Pa­ evaluate the cost-benefit effectiveness of available risk mitigation op­
cific Northwest [4] and the Makran region off Pakistan and Iran [5]. The tions [8–10]. For earthquake-related perils, recent advances in seismi­
fault rupture is the main and common trigger, resulting in strong c/tsunami science and engineering have led to the development of new
shaking as primary peril and tsunami, landslide, and liquefaction as multi-hazard catastrophe models for earthquakes and tsunamis (e.g.

* Corresponding author. Department of Earth Sciences, Western University, London, Ontario, Canada.
E-mail address: kgoda2@uwo.ca (K. Goda).

https://doi.org/10.1016/j.ijdrr.2021.102050
Received 28 November 2020; Received in revised form 12 January 2021; Accepted 12 January 2021
Available online 20 January 2021
2212-4209/© 2021 Elsevier Ltd. All rights reserved.
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

Fig. 1. (a) Nankai-Tonankai Trough source region and (b) location of the target case-study site Ogata district in Kuroshio Town, Kochi Prefecture. The fault plane
geometry of the 2012 CDMC models is shown in panel (a).

Refs. [11,12]. Herein, the term multi-hazard is used to emphasize that Nankai-Tonankai earthquake are generated from empirical scaling re­
common earthquake source features of strong shaking and tsunami are lationships of earthquake source parameters and stochastic synthesis of
captured through physical parameters of earthquake rupture. In the new earthquake slip distributions [17]. Shaking intensity modeling is con­
multi-hazard catastrophe models, simulations of shaking and tsunami ducted based on a ground motion model for Japanese subduction
hazard processes are conducted sequentially by maintaining the source earthquakes [20] together with intra-event spatial correlation [21],
dependency. More specifically, a multi-hazard loss model for Japanese whereas tsunami inundation modeling is performed by solving
subduction earthquakes and tsunamis developed by Goda and De Risi nonlinear shallow water equations [22]. To enable realistic and accurate
[11] is based on stochastic modeling of earthquake ruptures with het­ estimation of shaking- and tsunami-induced losses, local site conditions
erogeneous slip distribution and can be used for different earthquake for amplifying shaking intensity (250-m grids from the J-SHIS, http
magnitudes. In their model, seismic-tsunami loss of wooden buildings is ://www.j-shis.bosai.go.jp/en/) and 10-m grid elevation data, which
evaluated using empirical fragility functions that were derived from also reflect in-situ coastal defense structures, are used. The
seismic/tsunami damage data in Japan. high-resolution elevation data are provided by the Japanese Cabinet
The Nankai-Tonankai Trough region of western and central Japan Office and are based on a 5-m grid digital terrain model derived from
(Fig. 1a) is anticipating a recurrent megathrust subduction event in the airborne laser surveys and aerial photographic surveys by the Geospatial
near future. Since 684 A.D., the Nankai-Tonankai Trough hosted 9 Information Authority of Japan (see Ref. [17]. The results are combined
events of M8.0+ [13,14], leading to a mean recurrence period of circa with empirical fragility functions that are applicable to conventional
150 years. The most recent events were the 1944 Tonankai earthquake wooden buildings in Japan for shaking damage [23–25] and for tsunami
and the 1946 Nankai earthquake, rupturing the eastern and western damage [26]. The main results from the developed catastrophe model of
parts of the Nankai-Tonankai Trough, respectively. When the trough a building portfolio subject to megathrust subduction earthquakes and
segments rupture separately, the earthquake magnitude tends to result tsunamis are the multi-hazard and single-hazard loss distributions,
in smaller magnitudes (typically less than M8.5). On the other hand, accompanied by detailed earthquake rupture scenarios,
geological evidence from the past earthquakes indicates that synchro­ shaking-tsunami hazard intensity distributions, and building damage
nous ruptures of the Nankai and Tonankai segments occurred, such as distributions. The new multi-hazard tool facilitates the identification of
the 887 Ninna, 1361 Shohei, and 1707 Hoei events [15,16], resulting in critical multi-hazard loss scenarios and leads to integrated hazard-risk
greater moment magnitudes (M8.6 and above). For preparing against maps that are particularly useful for disaster risk reduction and man­
future earthquake-tsunami disasters in western and central Japan, the agement purposes. The paper is organized as follows. Section 2 in­
Central Disaster Management Council (CDMC) of the Japanese Cabinet troduces the key characteristics of the Nankai-Tonankai megathrust
Office developed national tsunami source models (hereafter, 2012 subduction events. In Section 3, the multi-hazard portfolio loss esti­
CDMC models) by considering the synchronous M9 rupture scenarios mation model for the Nankai-Tonankai earthquakes and tsunamis is
involving multiple segments and produced tsunami inundation maps for developed. In Section 4, numerical loss estimation results for the Ogata
municipalities in the affected regions. Recently, Goda et al. [17] district of Kuroshio Town are discussed by focusing upon the
generated a realistic set of 1000 stochastic earthquake source models of multi-hazard and single-hazard loss distribution characteristics and the
M8.7 to M9.1 based on the fault plane geometry of the 2012 CDMC identification and visualization of critical loss scenarios. This is followed
models by using empirical source scaling relationships [18] and sto­ by a list of conclusions and pending work for the future studies.
chastic synthesis of heterogeneous slip distribution [19]. Subsequently,
they performed Monte Carlo-based tsunami inundation analyses by 2. Nankai-Tonankai megathrust earthquake and tsunami
focusing upon coastal communities in Kuroshio Town of Kochi Prefec­
ture, Japan (Fig. 1b). Their results provided valuable insights into the The Nankai-Tonankai Trough is the primary source of major sub­
degree of uncertainty associated with regional and local tsunami hazard duction earthquakes in western and central Japan, and their seismic
assessments. activities are induced by tectonic movements of the Philippine Sea Plate
This paper presents the first multi-hazard portfolio loss estimation and the Eurasian Plate with slip rates between 40 and 55 mm/year [27].
study for the future Nankai-Tonankai megathrust subduction earth­ Current estimates of the spatial distribution of slip deficit rates indicate
quake and tsunami by considering 1000 stochastic earthquake rupture that the potential of a future megathrust event is high [28,29], and thus
scenarios of M8.7 to M9.1. The study site is the Ogata district of Kur­ enhancing the preparedness against future Nankai-Tonankai megathrust
oshio Town, Kochi Prefecture (Fig. 1b), which is one of the most exposed subduction events is of critical importance.
municipalities subject to the Nankai-Tonankai events in south-western The Nankai-Tonankai Trough region can be broadly divided into the
Pacific region of Japan. The earthquake rupture models for the Nankai and Tonankai segments, as shown in Fig. 1a. The Nankai

2
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

Fig. 2. Multi-hazard earthquake-tsunami loss estimation framework.

segment spans from Hyuga-nada, Shikoku Island, and western Kii and local tsunami disaster preparedness, such as construction of coastal
Peninsula and ruptured many times in the past. There is abundant evi­ defense structures and planning of tsunami evacuation. Using the fault
dence that earthquakes originating from the Nankai segment are capable plane geometry of the 2012 CDMC models, Goda et al. [17] generated
of producing intense ground shaking and large-scale coastal inundations 1000 stochastic source models for future Nankai-Tonankai megathrust
due to tsunamis (e.g. Ref. [30]. The rupture history of the Tonankai events.
segment, which spans from eastern Kii Peninsula, Enshu-nada, and
Tokai, is also well studied (e.g. Refs. [15,16]. The past Tonankai events 3. Multi-hazard earthquake-tsunami loss model for the Nankai-
caused widespread shaking and tsunami effects in the central Pacific Tonankai megathrust
region of Japan.
For tsunami hazard mapping of a future Nankai-Tonankai mega­ Building upon the stochastic source models and tsunami inundation
thrust event, the CDMC developed eleven M9-class earthquake source analyses conducted by Goda et al. [17] for future Nankai-Tonankai
models by considering that the synchronized rupture across both the megathrust events, an earthquake-tsunami loss model for the
Nankai and Tonankai segments is possible. The eleven source models Nankai-Tonankai events is developed in this section. Although a
reflect (a limited number of) different rupture scenarios in terms of lo­ comprehensive earthquake impact assessment due to multiple hazards
cations of large slip concentrations (i.e. asperities) within the rupture has been conducted by the Japanese Cabinet Office for national disaster
plane (see Goda et al. [17] for more details). The main subduction management and mitigation purposes, these estimates are derived for a
interface of the 2012 CDMC models can be represented by a set of 5669 limited number of worst-case earthquake scenarios and are insufficient
sub-faults, each sub-fault having a size of 5 km by 5 km, with the total to quantify the uncertainty of the seismic-tsunami damage and loss of a
fault plane area of 1.44 × 105 km2 (Fig. 1a). The geometrical parameters coastal community. The developed multi-hazard loss model for future
(i.e. strike, dip, and rake) of the sub-faults are variable over the curved, Nankai-Tonankai earthquakes addresses this research gap by consid­
steepening fault plane along the dip direction. The models are intended ering numerous stochastic earthquake rupture scenarios (in terms of
to represent extreme scenarios that can be used for improving regional fault geometry and earthquake slip distribution) and by propagating the

Fig. 3. Building distribution in the Ogata district of Kuroshio Town. (a) Elevation, (b) average surface shear-wave velocity based on J-SHIS, (c) zoomed area near the
vertical evacuation tower, and (d) photo of the vertical evacuation tower.

3
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

Fig. 4. (a) Stochastic source modeling, (b) shaking hazard footprint modeling, and (c) tsunami hazard footprint modeling.

uncertainty from earthquake rupture, shaking-tsunami hazard intensity, 6500 people live in 2700 households. Along the two local rivers, sed­
to multi-hazard damage-loss assessment. The multi-hazard loss estima­ imental deposits are prevalent and thus larger ground shaking intensity
tion is based on the methodology proposed by Goda and De Risi [11]; is expected in the river plain areas. This can be seen in Fig. 3b, which
which was developed for the Tohoku region of Japan, and consists of shows the average shear-wave velocity in the upper 30 m (Vs30) ob­
four major modules: (i) exposure modeling, (ii) seismic-tsunami foot­ tained from J-SHIS (250-m mesh grids). Values of Vs30 at locations of
print modeling, (iii) seismic-tsunami fragility modeling, and (iv) risk the houses and local facilities are typically less than 300 m/s (e.g.
assessment. A computational procedure of the multi-hazard loss esti­ NEHRP site class D), indicating that local site conditions are relatively
mation is illustrated in Fig. 2, by linking with major modules that are soft. The beach areas of the Ogata district are mainly used for recrea­
described in subsequent subsections. Since the multi-hazard risk tional and agriculture purposes, and many of the residents live in inland
assessment methodology that is implemented for the Nankai-Tonankai areas, which are behind vegetated sand dune/hills having 10-m eleva­
megathrust is largely identical to the framework presented in Goda tion. This can be seen in Fig. 3c. In this zoomed area, a tsunami evac­
and De Risi [11]; the computational modules are described succinctly in uation tower was constructed for local residents living in low-lying areas
the following subsections. Furthermore, because detailed descriptions of (Fig. 3d). The evacuation space is at the tower’s top floor and is 18.2 m
the stochastic source modeling and tsunami inundation analyses for the above the mean sea level. By taking into the tower’s land elevation of
Nankai-Tonankai earthquakes can be found in Goda et al. [17]; expla­ 4.0 m, the critical inundation depth at the tower for safe evacuation is
nations for this aspect are kept brief as well. 14.2 m. This critical depth can be used to evaluate the adequacy of the
tsunami evacuation tower under extreme situations.
To develop a building exposure model for the Ogata district, a na­
3.1. Exposure model
tional GIS database maintained by the Geospatial Information Authority
of Japan (https://www.gsi.go.jp/kiban/) is utilized. The dataset in­
Kuroshio Town is a municipality in Kochi Prefecture (Fig. 1b) that
cludes geographical coordinates and property plot boundary data of
faces significant seismic-tsunami hazards due to the Nankai-Tonankai
individual buildings, and thus building footprint areas of individual
subduction earthquakes. As of September 2020, the population and
buildings can be calculated. Since the current version of the earthquake-
household number of Kuroshio Town are 10,905 and 5,475, respec­
tsunami catastrophe model is developed for low-rise wooden buildings
tively. According to the 2012 national seismic-tsunami hazard assess­
(note: this limitation is due to the availability of suitable seismic fragility
ments conducted by the Japanese Cabinet Office, shaking and tsunami
functions; Section 3.3), screening of the building information contained
hazard levels that need to be considered for Kuroshio’s earthquake
in the original dataset is performed in two stages. First, general building
disaster risk reduction strategy and planning are the Japan Meteoro­
area statistics of Japanese wooden buildings are consulted. By assuming
logical Agency Seismic Intensity scale VII (highest, equivalent to the
that the average building footprint area is lognormally distributed with
Modified Mercalli Intensity of X to XII) and the assessed tsunami wave/
mean = 100 m2 and coefficient of variation (CoV) = 0.32 [32], the in­
run-up height of 34 m [31].
terval of the mean minus/plus two standard deviations can be obtained
The Ogata district is one of the main populated areas in Kuroshio
as 51 m2 and 178 m2. In the first screening stage, all buildings that have
Town along the coast (Fig. 3a). Houses and facilities (see black polygons
the footprint areas less than 50 m2 and more than 200 m2 are excluded.
in Fig. 3a,c) are distributed on an alluvial plain surrounded by Fukigami
In the second screening stage, visual inspection of the selected buildings
River and Kakize River. Within the area shown in Fig. 3a, approximately

4
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

Fig. 5. (a) Shaking fragility functions, and (b) tsunami fragility functions.

is conducted in comparison with Google Earth images of the area, and 56 slip and maximum slip. In this study, stochastic synthesis of constrained
buildings are manually identified as non-residential. The building random slip distributions is performed to generate 1000 stochastic
database for the Ogata district after the 2-stage screening procedure source models having earthquake magnitudes in the range between
contains 3018 buildings and is used for the multi-hazard loss estimation M8.7 and M9.1 [17]. The synthesized earthquake source models reflect
as described in the following sections. The spatial distribution of the possible variability of tsunamigenic earthquakes from the
selected buildings is shown in Fig. 3a and c. It is noted that the calcu­ Nankai-Tonankai Trough in terms of geometry, fault location, and slip
lated statistics of the building footprint area of the selected buildings in distribution (Fig. 4a). These rupture models are used to perform Monte
Ogata (mean = 101 m2 and CoV = 0.36) are consistent with the assumed Carlo simulations of shaking intensity and tsunami inundation in the
statistics in the first screening stage (i.e. mean = 100 m2 and CoV = Ogata district of Kuroshio Town.
0.32). For the loss estimation purposes, the unit construction cost for For a given earthquake source model, shaking and tsunami hazard
typical wooden buildings is obtained from the Japanese building cost intensities at building locations are evaluated by using a ground motion
information handbook [33] and is assumed to be lognormally distrib­ model and by solving non-linear shallow water equations (NSWE) for
uted with mean = 1600 US$/m2 and CoV = 0.32 (assuming 1 US$ = 100 initial boundary conditions of sea surface caused by an earthquake
yen). In calculating the total building replacement cost (= total footprint rupture, respectively. In this study, the peak ground velocity (PGV) is
area times unit construction cost), the total footprint area is estimated by selected as shaking hazard parameter and the maximum inundation
multiplying the building footprint area of each individual property (as in depth is adopted as tsunami hazard parameter. The choice of PGV as
the original building database) by a factor of 1.3, which is considered to seismic hazard measure is owing to its compatibility with empirical
be applicable based on the building/construction statistics from Con­ seismic fragility functions in Japan (Section 3.3). The local site condi­
struction Research Institute [33] and the Japanese Ministry of Land, tions are based on the J-SHIS average shear-wave velocity database
Infrastructure, Transport and Tourism (http://www.mlit.go.jp/toukeijo (Fig. 3b). The PGV ground motion model by Morikawa and Fujiwara
uhou/chojou/stat-e.htm). Based on the above building cost information, [20] together with the intra-event spatial correlation model of Goda and
the expected total cost of the 3018 buildings in Ogata is 628 million US$. Atkinson [21] is used to generate spatially correlated ground motion
fields for the 1000 stochastic sources (Fig. 4b).
On the other hand, tsunami propagation and inundation simulations
3.2. Hazard model are performed using a well-tested NSWE TUNAMI computer code by
Goto et al. [22]. The initial water surface dislocation due to an earth­
A stochastic source model captures the spatial uncertainty of earth­ quake rupture is evaluated based on the analytical equations by Okada
quake rupture for a given earthquake magnitude. In generating the [34] and Tanioka and Satake [35]. The computational domains are
stochastic rupture models for the Nankai-Tonankai megathrust events, nested with 2430-m, 810-m, 270-m, 90-m, 30-m, and 10-m grids.
the 2012 CDMC fault plane (Fig. 1a) is adopted as baseline. The sub- High-resolution elevation data together with coastal/riverside structural
faults of the 2012 CDMC model are mapped onto a 2D rectangular data (e.g. breakwater and levees) and surface roughness data are taken
matrix with the size of 153 (along-strike) by 54 (down-dip). The source into account to ensure the accuracy of local tsunami inundation analyses
uncertainty is characterized by empirical prediction equations of (note: the mentioned data are the same as the 2012 CDMC models). The
earthquake source parameters and stochastic synthesis of earthquake run-up calculation is based on a moving boundary approach [36], where
slip [18]. For a magnitude value, eight source parameters, i.e. fault a dry/wet condition of a computational cell is determined based on total
width, fault length, mean slip, maximum slip, Box-Cox power param­ water depth relative to its elevation. The maximum inundation depths at
eter, correlation length along dip, correlation length along strike, and the building locations are determined by subtracting the land elevations
Hurst number, are generated. Once the geometry and position of a sto­ from the maximum inundation heights. Tsunami simulations are con­
chastic source model are determined, a random heterogeneous slip ducted for all 1000 stochastic sources by considering a time step of 0.1 s
distribution is generated using a Fourier integral method, where to satisfy the Courant–Friedrichs–Lewy condition and tsunami simula­
amplitude spectrum is represented by von Kármán spectrum and tion duration of 3 h. Moreover, the effects of ground deformation are
random phase [19]. To generate a slip distribution with realistic taken into account by adjusting the elevation data prior to the tsunami
right-heavy tail features, the synthesized slip distribution is converted simulation run. However, the effects of tidal variations (e.g., mean high
via Box-Cox power transformation. The transformed slip distribution is water level) are not considered in the simulations. For each of the
then adjusted to achieve the suitable slip characteristics, such as mean

5
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

Fig. 6. (a) Shaking damage map, (b) tsunami damage map, and (c) combined shaking-tsunami damage map. The background color map shows to local eleva­
tion data.

stochastic source models, a detailed local tsunami inundation map can and tsunami are estimated by applying seismic and tsunami fragility
be obtained (Fig. 4c). functions. For shaking damage, the adopted seismic fragility models are
three empirical functions for low-rise wooden buildings in Japan
3.3. Fragility and damage model [23–25], which were developed based on shaking damage data from the
1995 Kobe earthquake, seven crustal earthquakes that occurred between
A vulnerability model relates the hazard intensity and the building 2003 and 2008, and the 2011 Tohoku earthquake, respectively. The
damage via an impact severity metric (e.g. damage ratio). In the three seismic fragility functions are combined by considering equal
developed multi-hazard catastrophe model, damage ratios for shaking

Fig. 7. Histogram of the combined shaking-tsunami loss for the building Fig. 8. Scatter plot and histograms of the shaking loss and tsunami loss for the
portfolio in the Ogata district. building portfolio in the Ogata district.

6
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

weights. Using the lognormal model, the adopted equations for the three the zoomed area (i.e. closer to the coast) than the western side of the
damage states, i.e. partial damage dsPD, half collapse dsHC, and total area (Fig. 6b). This is consistent with the tsunami inundation pattern
collapse dsTC, are: shown in Fig. 4c, where the eastern side of the zoomed area experiences
[ ( ) ( )] flow depths of 3–4 m, while the western side experiences flow depths of
1 ln(PGV) − 4.340 ln(PGV) − 4.650
P(DS > dsPD ) = 2 × Φ +Φ 0–2 m. With flow depths of 3–4 m, the probability of major tsunami
3 0.358 0.400
damage becomes significantly higher in comparison with flow depths of
(1) 0–2 m (Fig. 5b).

[ ( ) ( ) ( )]
1 ln(PGV) − 4.650 ln(PGV) − 4.771 ln(PGV) − 4.930
P(DS > dsHC ) = Φ +Φ +Φ (2)
3 0.382 0.394 0.415
[ ( ) ( ) ( )]
1 ln(PGV) − 4.950 ln(PGV) − 5.071 ln(PGV) − 5.630
P(DS > dsTC ) = Φ +Φ +Φ (3)
3 0.429 0.402 0.576

Subsequently, for each building, a greater of the estimated shaking


and tsunami damage ratios is adopted as the final damage ratio of the
Note that PGV is in cm/s, and Φ(•) is the standard normal distribu­ building. A multi-hazard loss value is calculated by sampling a value of
tion function. For each equation (i.e. damage state), the first, second, the building replacement cost (Section 3.1) and then by multiplying it
and third terms on the right-hand side of the equations correspond to the by the final damage ratio. For the illustrative cases shown in Fig. 6a and
equations suggested by Yamaguchi and Yamazaki [23]; Midorikawa b, the combined damage ratios of the buildings in the zoomed area of the
et al. [24]; and Wu et al. [25]; respectively. Midorikawa et al. [24] did Ogata district are displayed in Fig. 6c. The final seismic-tsunami damage
not develop a seismic fragility function for the partial damage state. ratio map reflects the estimated damage patterns from both shaking and
Therefore, the weight for the seismic fragility function for Yamaguchi tsunami risk assessments. The shaking damage is distributed more
and Yamazaki [23] is doubled with respect to that for Wu et al. [25] (see randomly over the area but the tsunami damage is systematically
Equation (1)). The weighted seismic fragility functions for the partial distributed along the coastal side. It is noteworthy that the above-
damage, half collapse, and total collapse damage states are shown in mentioned approach of calculating the combined shaking-tsunami
Fig. 5a (note: the individual functions vary significantly; see Goda and damage ratio does not account for the interaction between shaking
De Risi [11]. The seismic damage ratios for the partial damage, half and tsunami damage (including tsunami debris effects) explicitly. For
collapse, and total collapse are assigned as 0.03–0.2, 0.2–0.5, and the present study, this limitation is alleviated because the tsunami
0.5–1.0, respectively [37]. fragility model by De Risi et al. [26] is based on the tsunami damage
For tsunami damage, the tsunami fragility model developed by De data from the 2011 Tohoku event that include the effects due to shaking
Risi et al. [26] based on the tsunami damage data from the 2011 Tohoku damage. Nevertheless, when a multi-hazard loss model is developed for
event is adopted. The fragility model is expressed in terms of tsunami a different region, it is desirable to consider such shaking-tsunami
flow depth, and the following five tsunami damage states are consid­ damage accumulation effects (e.g. Ref. [38].
ered: minor, moderate, extensive, complete, and collapse. The equations Finally, by repeating the risk assessments for all selected buildings
for the tsunami fragility functions are not shown here because the and all earthquake scenarios, multi-hazard and single-hazard loss sam­
multinomial logistic regression model, i.e. M3 model with the model ples can be obtained for the building portfolio of interest (e.g. Fig. 3). By
parameters listed in Table 3 of De Risi et al. [26]; is relatively compli­ plotting the loss percentiles of the single-hazard and multi-hazard loss
cated and repeating the same information and explanations that can be samples against the sample size of the earthquake scenarios, the stability
found in De Risi et al. [26] is unnecessary. The tsunami fragility func­ of the simulated loss data is examined. In the post-processing stage of the
tions for the five damage states are shown in Fig. 5b. The tsunami multi-hazard loss estimation, associations of the synthesized earthquake
damage ratios for the damage ratio ranges of 0.03–0.1, 0.1–0.3, 0.3–0.5, scenario, simulated PGV distribution, inundation depth distribution,
0.5–1.0, and 1.0, respectively (http://www.mlit.go.jp/toshi/toshi-h and single-hazard as well as multi-hazard damage ratio distributions (i.
ukkou-arkaibu.html). e. Figs. 4 and 6) are retained. This facilitates the development of critical
hazard-risk scenario maps that can be utilized in managing multi-hazard
disaster risks. Such new multi-hazard risk management tools will be
3.4. Risk assessment demonstrated in Sections 4.3 and 4.4.

A damage ratio of each building due to ground shaking is determined 4. Multi-hazard loss estimation results
by identifying the shaking damage state for a given PGV value and then
by simulating a uniform random variable within the lower and upper This section presents numerical loss estimation results for the Ogata
bounds of the respective damage state. The same procedure can be district of Kuroshio Town using the 1000 stochastic source models for
adopted for determining the tsunami damage ratio. The shaking and the M8.7 to M9.1 Nankai-Tonankai Trough megathrust rupture sce­
tsunami damage ratio maps of the buildings in the zoomed area of the narios. Section 4.1 focuses upon the characteristics of the simulated loss
Ogata district (Fig. 3c) are illustrated in Fig. 6a and Fig. 6b, respectively. samples for the multi-hazard and single-hazard cases. In Section 4.2, the
The earthquake source for the shaking-tsunami risk assessment is shown shaking-tsunami hazard intensity distributions of the vertical evacua­
in Fig. 4a, and the corresponding PGV and inundation distributions for tion tower in Ogata are investigated. By linking the calculated portfolio
this scenario are shown in Fig. 4b and c, respectively. It can be observed loss percentiles with other multi-hazard characteristics, two types of
from Fig. 6a and b that the shaking damage is more uniformly spread critical hazard-risk scenario maps can be derived. The first type is based
among the buildings in the zoomed area because the PGV intensity for on an individual event that corresponds to a certain loss percentile and
this scenario is relatively high (exceeding 100 cm/s2; Fig. 4b) and thus thus various results, such as the earthquake source, hazard intensity
the half-collapse and collapse damage states can be induced (Fig. 5a). On footprint, and damage severity, can be associated directly. These results
the other hand, the tsunami damage is more severe on the eastern side of

7
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

Fig. 9. Scatter plot and histograms of peak ground velocity and inundation depth at the vertical evacuation tower in the Ogata district (Fig. 3d).

are discussed in Section 4.3. As the second type of critical loss scenario the increase in earthquake magnitude itself does not result in the
maps, the averaged scenario loss results of the source, intensity, and increased ground motion intensity due to the saturation of magnitude
damage characteristics that cause specified ranges of the multi-hazard scaling as incorporated in the ground motion model by Morikawa and
loss can also be useful for disaster risk management purposes. Such Fujiwara [20]. This is the main reason that the effect of earthquake
multi-hazard loss estimation results are explored in Section 4.4. magnitude on the shaking loss is limited, as seen on the top panel of
Fig. 8.
4.1. Loss distributions for building portfolio in Ogata In contrast, the tsunami loss is more influenced by the increase in
tsunami potential energy due to larger earthquakes [39]. The tsunami
The building portfolio loss samples for the 1000 stochastic source potential energy is the sum of the squared sea surface dislocation values
models of the Nankai-Tonankai megathrust events are obtained from the caused by an earthquake rupture and accounts for the combined effects
developed multi-hazard catastrophe model. Fig. 7 shows a histogram of of earthquake magnitude and other source characteristics, such as slip,
the combined shaking-tsunami loss for the building portfolio in the depth, strike, dip, and rake, on the sea surface dislocation. For the same
Ogata district, whereas Fig. 8 shows a scatter plot and histograms of the set of the Nankai-Tonankai stochastic source models, Goda et al. [17]
shaking loss and tsunami loss for the building portfolio in the Ogata demonstrated that the inundation area, which is an accurate predictor
district. In these two figures, the loss samples are distinguished for two for local tsunami loss [40], is strongly correlated with tsunami potential
magnitude ranges (M8.7-8.9 and M8.9-9.1). From Fig. 7, it can be energy. Therefore, a more significant increase in the tsunami loss is
observed that the combined loss tends to increase with the earthquake expected with the increase in earthquake magnitude, as seen in the right
magnitude. By inspecting single-hazard loss histograms shown in Fig. 8 panel of Fig. 8, than the shaking loss. It is also noticeable that the
(the top and right panels of the scatter plot), the tsunami loss tends to tsunami loss histograms exhibit bimodal loss distributions, i.e. a lower
increase with the magnitude more significantly than the shaking loss, mode at tsunami losses less than 200 million US$ and a higher mode at
and thus contributes more towards the increase in the combined loss tsunami losses greater than 200 million US$. When a large loss occurs in
with the earthquake magnitude as shown in Fig. 7. In other words, Ogata, the earthquake asperities tend to be located off Shikoku Island
tsunami risk is more nonlinearly affected by the earthquake size than and thus massive tsunami waves are directed towards the target coastal
shaking risk, and this observation is consistent with the results reported areas.
in Goda and De Risi [11] for the building portfolios in the Tohoku region Lastly, from the scatter plot shown in Fig. 8, it is not possible to
of Japan. identify a clear dependency between the shaking loss and the tsunami
With regard to the estimated shaking loss based on the developed loss. This may be due to (at least) two reasons. One reason is that the
multi-hazard catastrophe model, the increase in earthquake magnitude current ground motion model does not account for locations of asperities
results in a larger earthquake rupture area (as per empirical scaling re­ in computing the ground shaking intensity (i.e. the site-to-source dis­
lationships), and thus there is more chance that the distance from the tance depends only on the overall fault model and does not reflect local
rupture plane to the building sites becomes closer. On the other hand, features of the earthquake rupture) and its predicted intensity is

8
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

Fig. 10. Earthquake source models (a,c,e,g) and initial vertical deformation distributions (b,d,f,h) of the 16th, 50th, 84th, and 97.5th percentile loss scenarios
identified based on the combined shaking-tsunami loss for the building portfolio in the Ogata district (Fig. 7).

saturated with respect to earthquake magnitude. In addition, the ground 4.2. Multi-hazard shaking-tsunami hazard intensity at vertical evacuation
motion variability term, such as between-event variability, does not tower in Ogata
depend on the (average) earthquake slip. An obvious refinement of the
developed multi-hazard catastrophe model is to implement hybrid Before investigating the source, hazard, and damage characteristics
ground motion simulations (e.g. Refs. [41,42] to capture more local at different combined loss levels in Sections 4.3 and 4.4, it is interesting
source characteristics in the shaking footprint estimation. The other to examine the shaking and tsunami hazard intensities at the vertical
reason for the lack of clear dependency between the shaking loss and the evacuation tower in Ogata (Fig. 3d). For this purpose, information on
tsunami loss is the consideration of local target areas for the loss esti­ PGV and inundation depth at the vertical evacuation tower is extracted
mation in this study. When the multi-hazard loss estimation is per­ and is shown as a scatter plot and histograms in Fig. 9 by distinguishing
formed for a wider region (e.g. Shikoku or western Japan), the bimodal the two magnitude ranges. The PGV distributions (top panel) have
characteristics of the regional tsunami loss distribution will become less similar unimodal shapes for the two magnitude ranges with the repre­
noticeable, and the regional shaking loss distribution depend more on sentative PGV values of 80 cm/s and the extreme PGV values exceeding
the earthquake magnitude. These are future research topics and are 250 cm/s. The inundation depth distributions (right panel) exhibit the
outside of the scope of the current study. right skewed distributions with large concentrations at inundation depth
of 0 m (i.e. the site is not inundated, noting that the tower is at 4.0 m

9
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

Fig. 11. Peak ground velocity (a,c,e,g) and inundation depth (b,d,f,h) of the 16th, 50th, 84th, and 97.5th percentile loss scenarios identified based on the combined
shaking-tsunami loss for the building portfolio in the Ogata district (Fig. 7).

above the mean sea level). There are two cases where tsunami depth 4.3. Critical hazard-risk maps based on individual multi-hazard loss
becomes greater than the critical depth of 14.2 m; the chance of expe­ scenarios
riencing such incidence may be regarded as sufficiently small given that
extreme rupture scenarios are simulated in this study [17]. Importantly, Each of the multi-hazard loss results shown in Fig. 7 can be directly
the multi-hazard catastrophe model can produce useful outputs that feed associated with the corresponding earthquake source, hazard intensity
into advanced engineering analyses. For instance, the joint multi-hazard footprint, and damage severity. In other words, critical hazard-risk
intensity values are particularly useful for examining the structural scenario maps can be identified based on the combined loss distribu­
integrity of the vertical evacuation tower under extreme tion. To demonstrate how such joint scenario maps can be produced
seismic-tsunami loading conditions. from the multi-hazard catastrophe model based on stochastic earth­
quake sources, four loss levels are selected: 16th, 50th, 84th, and 97.5th.
These four loss levels are indicated in Fig. 7. In the following, the
earthquake source models and initial vertical deformation profiles, PGV

10
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

Fig. 12. Shaking damage ratios (a,d,g,j), tsunami damage ratios (b,e,h,k), and combined damage ratios (c,f,i,l) of the 16th, 50th, 84th, and 97.5th percentile loss
scenarios identified based on the combined shaking-tsunami loss for the building portfolio in the Ogata district (Fig. 7). The background color map shows the
elevation for the range of 0 m (gray) and 20 m (white).

and inundation depth hazard intensity distributions, and shaking, and 50th percentile scenarios), intense levels of shaking hazards (100
tsunami, and combined damage distributions at the selected four port­ cm/s to 200 cm/s in terms of PGV) can cause moderate-to-extensive
folio loss percentiles are mainly focused upon to investigate how the shaking to the buildings (Fig. 5a). On the other hand, for the cases
above-mentioned hazard-risk characteristics change with the combined shown in Fig. 11, the extent of tsunami inundation is limited to the
portfolio loss levels in Ogata. coastal areas where not many people live (Fig. 11b and d). As the loss
Fig. 10 shows the earthquake source models and initial vertical levels increase (e.g. 84th and 97.5th percentile scenarios), the extent of
deformation distributions corresponding to the 16th, 50th, 84th, and the tsunami inundation expands rapidly as shown in Fig. 11f and h. By
97.5th percentile loss scenarios. Generally speaking, when the loss levels recognizing that the majority of the wooden buildings can be severely
are low, an earthquake source model has either broader spatial distri­ damaged or washed away at the water depth of 2–4 m (Fig. 5b),
bution of earthquake slip (e.g. 16th percentile scenario) or large as­ buildings in inundated areas with light blue color will experience severe
perities but not directly facing the target site (e.g. 50th percentile damage or collapse. In contrast, the shaking intensity levels for the 84th
scenario). When the loss levels are high, the locations of large asperities and 97.5th percentile scenarios are similar to that of the 50th percentile
are directly off the target site (e.g. 84th percentile scenario), and the scenario.
spatial size of the asperity tends to become larger (e.g. 97.5th percentile Fig. 12 shows the building-by-building damage ratios due to shaking,
scenario). Essentially the same trends can be seen in the initial vertical tsunami, and combined shaking-tsunami effects, corresponding to the
deformation distributions. Note that the initial vertical deformation 16th, 50th, 84th, and 97.5th percentile loss scenarios. To show the local
distributions can be used to calculate the tsunami potential energy and variations of the simulated damage ratios at individual buildings, the
thus are useful as a more accurate predictor of tsunami loss [17,39]. maps zoom in the vicinity of the vertical evacuation tower in the Ogata
Fig. 11 shows the PGV and inundation depth distributions in the district (Fig. 3c). The background filled contour map displays the ground
Ogata district that correspond to the 16th, 50th, 84th, and 97.5th elevation, which can be viewed as local topographical indicators (i.e.
percentile loss scenarios. The locations of the buildings are also dis­ low elevation locations tend to experience higher local shaking ampli­
played on the maps. When the loss levels are relatively low (e.g. 16th fication and higher tsunami inundation). As inferred from the PGV

11
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

Fig. 13. Average shaking-tsunami hazard-risk maps for the 16th (13.5th to 18.5th) percentile loss scenarios for the building portfolio in the Ogata district (Fig. 7). (a)
Earthquake source model, (b) initial vertical deformation distribution, (c) peak ground velocity, (d) inundation depth, (e) shaking damage ratio, (f) tsunami damage
ratio, and (g) combined damage ratio. The background color map in (e), (f), and (g) shows the elevation for the range of 0 m (gray) and 20 m (white).

distribution maps shown in Fig. 11a, c, 11e, and 11g, the shaking within the area because the intra-event spatial correlation of ground
damage extent of the buildings in the zoomed area is relatively low for motion variability decays rapidly over a short distance; for instance, the
the 16th percentile loss, whereas the damage extent is moderate for the intra-event correlations at 1 km and 5 km distances are 0.74 and 0.57
50th, 84th, and 97.5th percentile loss cases. It is also noted that severely based on the average model by Goda and Atkinson [21]. The tsunami
damaged buildings due to strong shaking are distributed more randomly damage ratios of the buildings in the zoomed area are initially zero for

12
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

Fig. 14. Average shaking-tsunami hazard-risk maps for the 50th (47.5th to 52.5th) percentile loss scenarios for the building portfolio in the Ogata district (Fig. 7). (a)
Earthquake source model, (b) initial vertical deformation distribution, (c) peak ground velocity, (d) inundation depth, (e) shaking damage ratio, (f) tsunami damage
ratio, and (g) combined damage ratio. The background color map in (e), (f), and (g) shows the elevation for the range of 0 m (gray) and 20 m (white).

13
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

Fig. 15. Average shaking-tsunami hazard-risk maps for the 84th (81.5th to 86.5th) percentile loss scenarios for the building portfolio in the Ogata district (Fig. 7). (a)
Earthquake source model, (b) initial vertical deformation distribution, (c) peak ground velocity, (d) inundation depth, (e) shaking damage ratio, (f) tsunami damage
ratio, and (g) combined damage ratio. The background color map in (e), (f), and (g) shows the elevation for the range of 0 m (gray) and 20 m (white).

14
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

Fig. 16. Average shaking-tsunami hazard-risk maps for the 97.5th (95th to 100th) percentile loss scenarios for the building portfolio in the Ogata district (Fig. 7). (a)
Earthquake source model, (b) initial vertical deformation distribution, (c) peak ground velocity, (d) inundation depth, (e) shaking damage ratio, (f) tsunami damage
ratio, and (g) combined damage ratio. The background color map in (e), (f), and (g) shows the elevation for the range of 0 m (gray) and 20 m (white).

15
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

the 16th and 50th percentile loss scenarios (Fig. 12b and e), while they 5. Conclusions
increase rapidly to nearly 1.0 at higher loss levels (Fig. 12h and k). Since
the combined damage ratios of the buildings are defined based on the A new multi-hazard catastrophe model for a coastal building port­
larger of the shaking damage ratios and the tsunami damage ratios for folio exposed to the future Nankai-Tonankai megathrust subduction
the individual buildings, the characteristics of the combined damage earthquake and tsunami was developed. The earthquake-tsunami ca­
ratios for given loss levels are affected by the more dominant hazard tastrophe model incorporated the stochastic rupture sources, spatially
characteristics at the considered loss levels. correlated ground motion fields, tsunami inundation simulations,
detailed building data, seismic and tsunami fragility models, and dam­
age cost estimation. The main results from the developed catastrophe
4.4. Average critical hazard-risk maps model were obtained as the multi-hazard and single-hazard loss distri­
butions, accompanied by earthquake rupture scenarios, shaking-
It is noteworthy that the earthquake source models, hazard intensity tsunami hazard intensity distributions, and building damage distribu­
maps, and damage ratio maps shown in Figs. 10–12 are based on indi­ tions. The model was demonstrated by focusing upon the Ogata district
vidual events for specific loss levels. The results for similar loss levels (e. of Kuroshio Town, Kochi Prefecture, which is one of the most exposed
g. results for the 500th and 501th scenarios as ranked based on the municipalities subject to the Nankai-Tonankai events in south-western
multi-hazard portfolio loss metric) can exhibit different characteristics Pacific region of Japan.
in terms of earthquake rupture, hazard intensity distributions, and loss The case study application of the new earthquake-tsunami catas­
contributions. To capture more general features of the critical hazard- trophe model highlighted the following observations and conclusions.
risk scenario maps, the average characteristics of the critical hazard-
risk scenario maps that correspond to a range of loss levels are investi­ • The tsunami loss tends to increase with the magnitude more signif­
gated in this section. In the following, the four selected loss levels are icantly than the shaking loss, and contributes more towards the in­
still focused upon, but respective ranges of the loss levels are broadened crease in the multi-hazard total loss with the earthquake magnitude.
by taking the representative loss level minus and plus 2.5%. For The multi-hazard loss distribution has a wide variation in the esti­
instance, in developing the average critical hazard-risk scenario maps mated loss values and thus it is important to consider a wide range of
for the 84th percentile loss level, all loss events that cause the combined possible earthquake rupture scenarios to quantify this uncertainty.
portfolio loss between 81.5th and 86.5th levels (51 cases for the 1000 • The multi-hazard catastrophe model can produce useful input data
stochastic source models). To attempt a different way of organizing and for more advanced earthquake-tsunami engineering analyses. The
visualizing the results, seven hazard-risk metrics, as shown in joint multi-hazard intensity values (peak ground velocity and
Figs. 10–12, i.e. earthquake source model, initial vertical deformation maximum tsunami inundation depth) at the vertical evacuation
profile, PGV distribution, inundation depth distribution, shaking dam­ tower in Ogata were used for illustrating such applications.
age ratio, tsunami damage ratio, and combined damage ratio, for the • The major advantage of the multi-hazard catastrophe model based
same loss levels are consolidated into one figure. Such integrated on the stochastic source modeling approach is that critical hazard-
average critical hazard-risk scenario maps for the 16th, 50th, 84th, and risk scenario maps can be developed by linking the calculated port­
97.5th percentile loss levels are shown in Fig. 13, 14, 15, and 16, folio loss percentiles with other multi-hazard characteristics, such as
respectively. earthquake source, hazard intensity footprint, and damage severity.
An important observation from all average critical hazard-risk sce­ In this study, two types of such critical hazard-risk scenario maps
nario maps shown in Figs. 13–16 is that notable general trends of the were demonstrated. The individual-scenario-based critical hazard-
earthquake sources, hazard intensity, and damage severity emerge. For risk maps are useful to maintain a one to one relationship among
instance, the earthquake source model for the representative 16th the loss results, hazard estimations, and damage assessments,
percentile loss scenario (Fig. 13a) shows a broad slip concentration in whereas the average-scenario-based critical hazard-risk maps can
the middle of the Nankai and Tonankai segments. This broad slip con­ capture notable general trends of the earthquake sources, hazard
centration is moved to the Nankai segment off Shikoku Island when the intensity, and damage severity that emerge at specified portfolio loss
representative 50th percentile loss scenario is considered (Fig. 14a). levels. The latter is particularly useful for disaster risk management
With the increase of the representative loss level, the maximum value of purposes.
the earthquake slip increases from 20 m for the 50th percentile
(Fig. 14a), to 30 m for the 84th percentile (Fig. 15), and to 40 m for the The developed multi-hazard loss model can be extended in various
97.5th percentile (Fig. 16a). Similar gradual increases of the initial ways. Two important improvents for the future development can be
deformation profile, hazard intensity, and damage ratio can be observed mentioned. One is to implement hybrid ground motion simulations to
by comparing the respective figure panels shown in Figs. 13–16. capture more local source characteristics in the shaking footprint esti­
From earthquake-tsunami disaster risk management perspectives, mation. This will improve the earthquake shaking prediction of mega­
using the average critical hazard-risk scenario maps may be more thrust subduction events by incorporating more detailed features of the
effective in conveying the uncertainty associated with the predictions of earthquake ruptures, such as strong motion generation areas and as­
earthquake-tsunami consequences, because scenario-to-scenario vari­ perities. This is likely to reveal the dependency between the shaking
ability of the hazard-risk maps can be largely smoothed or suppressed. damage/loss and the tsunami damage/loss, which is masked by the
This is one of our motivations to compare individual critical hazard-risk limitations of the current multi-hazard catastrophe model. The other is
scenario maps (Figs. 10–12) and averaged critical hazard-risk scenario to extend the stochastic rupture scenarios to a wider range of earthquake
maps (Figs. 13–16). Moreover, it should be emphasized that the average magnitudes by considering different rupture patterns (e.g. individual
earthquake source models are different from national earthquake source versus synchronized rupture scenarios of the Nankai and the Tonankai
models, such as the 2012 CDMC models (see Ref. [17]. The former has segments). The latter will lead to probabilistic earthquake-tsunami
the connections to the multi-hazard loss distribution of the target hazard-risk analysis.
building portfolio, whereas the latter does not have such linkage. In this
regard, the critical earthquake scenarios and related hazard and risk Declaration of competing interest
maps are more useful for disaster risk management purposes and can be
utilized for other more advanced analyses, such as earthquake-tsunami The authors declare that they have no known competing financial
evacuation planning based on agent-based simulation and stochastic interests or personal relationships that could have appeared to influence
earthquake modeling [43]. the work reported in this paper.

16
K. Goda et al. International Journal of Disaster Risk Reduction 54 (2021) 102050

Acknowledgements [19] P.M. Mai, G.C. Beroza, A spatial random field model to characterize complexity in
earthquake slip, J. Geophys. Res.: Solid Earth 107 (2002), https://doi.org/
10.1029/2001JB000588.
The work was supported by the Leverhulme Trust (RPG-2017-006), [20] N. Morikawa, H. Fujiwara, A new ground motion prediction equation for Japan
the Canada Research Chair program (950–232015), the NSERC Dis­ applicable up to M9 mega-earthquake, J. Disaster Res. 8 (2013) 878–888.
covery Grant (RGPIN-2019-05898), and the Japan Society for the Pro­ [21] K. Goda, G.M. Atkinson, Intraevent spatial correlation of ground-motion
parameters using SK-net data, Bull. Seismol. Soc. Am. 100 (2010) 3055–3067.
motion of Science Joint Research Project (19039901-000829). The [22] C. Goto, Y. Ogawa, N. Shuto, F. Imamura, Numerical Method of Tsunami
tsunami simulations were performed using BlueCrystal Phase 3, which Simulation with the Leap-Frog Scheme, vol. 35, IOC Manual, Paris, France, 1997.
was supported by the Advanced Computing Research Centre of the [23] N. Yamaguchi, F. Yamazaki, Estimation of strong motion distribution in the 1995
Kobe earthquake based on building damage data, Earthq. Eng. Struct. Dynam. 30
University of Bristol. (2001) 787–801.
[24] S. Midorikawa, Y. Ito, H. Miura, Vulnerability functions of buildings based on
References damage survey data of earthquakes after the 1995 Kobe earthquake, Journal of
Japan Association of Earthquake Engineering 11 (2011) 34–47.
[25] H. Wu, K. Masaki, K. Irikura, S. Kurahashi, Empirical fragility curves of buildings in
[1] S. Murata, F. Imamura, K. Katoh, Y. Kawata, S. Takahashi, T. Takayama, Tsunami –
northern Miyagi Prefecture during the 2011 off the Pacific coast of Tohoku
to Survive from Tsunami, World Scientific, 2010, p. 302.
earthquake, J. Disaster Res. 11 (2016) 1253–1270.
[2] W. Siembieda, L. Johnson, G. Franco, Rebuild fast but rebuild better: Chile’s initial
[26] R. De Risi, K. Goda, T. Yasuda, N. Mori, Is flow velocity important in tsunami
recovery following the 27 February 2010 earthquake and tsunami, Earthq. Spectra
empirical fragility modeling? Earth Sci. Rev. 166 (2017) 64–82.
28 (2012) 621–641.
[27] Y. Yokota, T. Ishikawa, S. Watanabe, T. Tashiro, A. Asada, Seafloor geodetic
[3] S. Fraser, A. Pomonis, A. Raby, K. Goda, S.C. Chian, J. Macabuag, M. Offord,
constraints on interplate coupling of the Nankai Trough megathrust zone, Nature
K. Saito, P. Sammonds, Tsunami damage to coastal defences and buildings in the
(2016), https://doi.org/10.1038/nature17632.
March 11th 2011 Mw9.0 Great East Japan earthquake and tsunami, Bull. Earthq.
[28] H. Baranes, J.D. Woodruff, J.P. Loveless, M. Hyodo, Interseismic coupling-based
Eng. 11 (2013) 205–239.
earthquake and tsunami scenarios for the Nankai Trough, Geophys. Res. Lett. 45
[4] H. Park, D.T. Cox, M.S. Alam, A.R. Barbosa, Probabilistic seismic and tsunami
(2018) 2986–2994.
hazard analysis conditioned on a megathrust rupture of the Cascadia subduction
[29] H. Kimura, K. Tadokoro, T. Ito, Interplate coupling distribution along the Nankai
zone, Frontiers in Built Environment 3 (2017) 32, https://doi.org/10.3389/
Trough in southwest Japan estimated from the block motion model based on
fbuil.2017.00032.
onshore GNSS and seafloor GNSS/A observations, J. Geophys. Res.: Solid Earth 124
[5] P. Momeni, K. Goda, M. Heidarzadeh, J. Qin, Stochastic analysis of tsunami hazard
(2019) 6140–6164.
of the 1945 Makran subduction zone Mw 8.1-8.3 earthquakes, Geosciences 10
[30] K. Tanigawa, M. Shishikura, O. Fujiwara, Y. Namegaya, D. Matsumoto, Mid- to
(2020) 452, https://doi.org/10.3390/geosciences10110452.
late-Holocene marine inundations inferred from coastal deposits facing the Nankai
[6] J.E. Daniell, A.M. Schaefer, F. Wenzel, Influence of secondary hazards in
Trough in Nankoku, Kochi Prefecture, southern Japan, Holocene 28 (2018)
earthquake loss, Frontiers in Built Environment 3 (2017) 30, https://doi.org/
867–878.
10.3389/fbuil.2017.00030.
[31] Central Disaster Management Council (CDMC), Working Group Report on Mega-
[7] J. Selva, R. Tonini, I. Molinari, M.M. Tiberti, F. Romano, A. Grezio, D. Melini,
Thrust Earthquake Models for the Nankai Trough, Japan. Cabinet Office of the
A. Piatanesi, R. Basili, S. Lorito, Quantification of source uncertainties in seismic
Japanese Government, Tokyo, 2012. http://www.bousai.go.jp/jishin/nankai/nan
probabilistic tsunami hazard analysis (SPTHA), Geophys. J. Int. 205 (2016)
kaitrough_info.html.
1780–1803.
[32] K. Goda, J. Song, Uncertainty modeling and visualization for tsunami hazard and
[8] W. Marzocchi, A. Garcia-Aristizabal, P. Gasparini, M.L. Mastellone, A. Di Ruocco,
risk mapping: a case study for the 2011 Tohoku earthquake, Stoch. Environ. Res.
Basic principles of multi-risk assessment: a case study in Italy, Nat. Hazards 62
Risk Assess. 30 (8) (2016) 2271–2285.
(2012) 551–573.
[33] Construction Research Institute, Japan Building Cost Information 2011, 2011,
[9] A. Mignan, S. Wiemer, D. Giardini, The quantification of low-probability–high-
p. 547. Tokyo, Japan.
consequences events: part I. A generic multi-risk approach, Nat. Hazards 73 (2014)
[34] Y. Okada, Surface deformation due to shear and tensile faults in a half-space, Bull.
1999–2022.
Seismol. Soc. Am. 75 (1985) 1135–1154.
[10] K. Mitchell-Wallace, M. Jones, J. Hillier, M. Foote, Natural Catastrophe Risk
[35] Y. Tanioka, K. Satake, Tsunami generation by horizontal displacement of ocean
Management and Modelling, Wiley, 2017, p. 506.
bottom, Geophys. Res. Lett. 23 (1996) 861–864.
[11] K. Goda, R. De Risi, Multi-hazard loss estimation for shaking and tsunami using
[36] Japan Society of Civil Engineers (JSCE), Tsunami Assessment Method for Nuclear
stochastic rupture sources, International Journal of Disaster Risk Reduction 28
Power Plants in Japan, 2002. https://www.jsce.or.jp/committee/ceofnp/Tsunam
(2018) 539–554.
i/eng/JSCE_Tsunami_060519.pdf.
[12] H. Park, M.S. Alam, D.T. Cox, A.R. Barbosa, J.W. van de Lindt, Probabilistic seismic
[37] A. Kusaka, H. Ishida, K. Torisawa, H. Doi, K. Yamada, Vulnerability functions in
and tsunami damage analysis (PSTDA) of the Cascadia Subduction Zone applied to
terms of ground motion characteristics for wooden houses evaluated by use of
Seaside, Oregon, International Journal of Disaster Risk Reduction 35 (2019)
earthquake insurance experience, Architectural Institute of Japan Journal of
101076.
Technology and Design 21 (2015) 527–532.
[13] M. Ando, Source mechanisms and tectonic significance of historical earthquakes
[38] N. Attary, J.W. van De Lindt, A.R. Barbosa, D.T. Cox, V.U. Unnikrishnan,
along the Nankai Trough, Japan, Tectonophysics 27 (1975) 119–140.
Performance-based tsunami engineering for risk assessment of structures subjected
[14] E. Garrett, O. Fujiwara, P. Garrett, V.M.A. Heyvaert, M. Shishikura, Y. Yokoyama,
to multi-hazards: tsunami following earthquake, J. Earthq. Eng. (2019), https://
A. Hubert-Ferrari, H. Brückner, A. Nakamura, M. De Batist, the QuakeRecNankai
doi.org/10.1080/13632469.2019.1616335.
team, A systematic review of geological evidence for Holocene earthquakes and
[39] D. Melgar, A.L. Williamson, E.F. Salazar-Monroy, Differences between
tsunamis along the Nankai-Suruga Trough. Japan, Earth Sci. Rev. 159 (2016)
heterogeneous and homogenous slip in regional tsunami hazards modelling,
337–357.
Geophys. J. Int. 219 (2019) 553–562.
[15] S. Fujino, H. Kimura, J. Komatsubara, D. Matsumoto, Y. Namegaya, Y. Sawai,
[40] K. Goda, N. Mori, T. Yasuda, Rapid tsunami loss estimation using regional
M. Shishikura, Stratigraphic evidence of historical and prehistoric tsunamis on the
inundation hazard metrics derived from stochastic tsunami simulation,
Pacific coast of central Japan: implications for the variable recurrence of tsunamis
International Journal of Disaster Risk Reduction 40 (2019) 101152.
in the Nankai Trough, Quat. Sci. Rev. 201 (2018) 147–161.
[41] A. Pitarka, R. Graves, K. Irikura, H. Miyake, A. Rodgers, Performance of Irikura
[16] O. Fujiwara, A. Aoshima, T. Irizuki, E. Ono, S.P. Obrochta, Y. Sampei, Y. Sato,
recipe rupture model generator in earthquake ground motion simulations with
A. Takahashi, Tsunami deposits refine great earthquake rupture extent and
Graves and Pitarka hybrid approach, Pure Appl. Geophys. 174 (2017) 3537–3555.
recurrence over the past 1300 years along the Nankai and Tokai fault segments of
[42] A. Frankel, E. Wirth, N. Marafi, J. Vidale, W. Stephenson, Broadband synthetic
the Nankai Trough, Japan, Quat. Sci. Rev. 227 (2020) 105999.
seismograms for magnitude 9 earthquakes on the Cascadia megathrust based on 3D
[17] K. Goda, T. Yasuda, N. Mori, A. Muhammad, R. De Risi, F. De Luca, Uncertainty
simulations and stochastic synthetics, Part 1: methodology and overall results, Bull.
quantification of tsunami inundation in Kuroshio Town, Kochi Prefecture, Japan
Seismol. Soc. Am. 108 (2018) 2347–2369.
using the Nankai-Tonankai megathrust rupture scenarios, Nat. Hazards Earth Syst.
[43] A. Muhammad, R. De Risi, F. De Luca, K. Goda, N. Mori, T. Yasuda, Agent-based
Sci. 20 (2020) 3039–3056.
modelling of existing tsunami evacuation plan in Kuroshio Town, Kochi, Japan, in:
[18] K. Goda, T. Yasuda, N. Mori, T. Maruyama, New scaling relationships of
17th World Conference on Earthquake Engineering, Sendai, Japan, Paper
earthquake source parameters for stochastic tsunami simulation, Coast Eng. J. 58
C003862, 2020.
(2016) 1650010.

17

You might also like