Nankai–Tonankai megathrust earthquakes and tsunamis pose significant risks to coastal communities in western and central Japan. Historically, this seismic region hosted many major earthquakes, and the current national tsunami hazard assessments in Japan consider megathrust events as those having moment magnitudes between 9.0 and 9.1. In responding to the lack of rigorous uncertainty analysis, this study presents an extensive tsunami hazard assessment for the Nankai–Tonankai Trough events, focusing on the southwestern Pacific region of Japan. A set of 1000 kinematic earthquake rupture models is generated via stochastic source modelling approaches, and Monte Carlo tsunami simulations are carried out by considering high-resolution grid data of 10 m and coastal defence structures. Significant advantages of the stochastic tsunami simulation methods include the enhanced capabilities to quantify the uncertainty associated with tsunami hazard assessments and to effectively visualize the results in an integrated manner. The results from the stochastic tsunami simulations can inform regional and local tsunami risk reduction actions in light of inevitable uncertainty associated with such tsunami hazard assessments and complement conventional deterministic tsunami scenarios and their hazard predictions, such as those developed by the Central Disaster Management Council of the Japanese Cabinet Office.

Nankai–Tonankai megathrust earthquakes and tsunamis pose an imminent
threat to people living in western and central Japan. Historically, the
Nankai–Tonankai Trough hosted major earthquakes (Ando, 1975; Garrett et al.,
2016; Fujino et al., 2018; Fujiwara et al., 2020; see Fig. 1). The most
recent events in 1944 and 1946 ruptured the eastern (Tonankai) and western
(Nankai) parts of the trough, respectively, and similar segmented ruptures
occurred during the 1854 Ansei earthquakes on 23 and 24 December with 32 h apart. The moment magnitudes (

Nankai–Tonankai Trough source region and history of tsunamigenic megathrust ruptures. The segment boundaries Z to E are based on Garret et al. (2016). The numbers in the square brackets correspond to the numbers of sub-faults of the 2012 Central Disaster Management Council tsunami source model falling within the segments. The rupture history is based on Garret et al. (2016), Fujino et al. (2018), and Fujiwara et al. (2020).

The 2012 Central Disaster Management Council tsunami source models.

The 2012 CDMC tsunami source models take into account different rupture and asperity patterns across the entire region of the Nankai–Tonankai Trough through expert judgement (Fig. 2). The ruptures reach the accretionary wedge of the Nankai–Tonankai Trough, and the largest slip values of circa 52 m are considered. The significant movements of the accretionary wedge at the shallow part of the subduction interface are capable of generating massive tsunamis, as inferred from the 2011 Tohoku earthquake and tsunami (Lotto et al., 2019). The earthquake slip distributions of the 2012 CDMC models are spatially broader than inverted source models for the 2011 Tohoku earthquake and tsunami that show more concentrated slips (Iinuma et al., 2012; Satake et al., 2013) and are consistent with current geodetic estimates of inter-seismic coupling (e.g., Baranes et al., 2018; Watanabe et al., 2018). It is noteworthy that the 2012 CDMC models are intended for representing extreme rupture scenarios at national scale and are not sufficient for encompassing the variability of tsunami hazards at regional and local scales. Such variability is of critical importance when tsunami hazards at municipality and township levels are concerned. In other words, local extreme situations may not be captured by such national-level earthquake rupture scenarios. These considerations are essential for developing effective tsunami risk reduction and evacuation strategies for local municipalities.

Since one cannot predict how future megathrust events will unfold, evaluating tsunami hazards based on a broad set of possible earthquake rupture scenarios is a viable strategy for better disaster preparedness. Regional and local tsunami risk reduction actions should be informed by rigorous uncertainty analysis (e.g., Fukutani et al., 2015; Mueller et al., 2015; Park et al., 2017). However, there is no study for the Nankai–Tonankai megathrust events that evaluates the variability of tsunami profiles and inundation extents by considering numerous rupture scenarios and high-resolution elevation data. A study by Goda et al. (2018) may be considered as an exception which employs a stochastic source modelling method of Mai and Beroza (2002) for tsunami hazard assessments of the Nankai–Tonankai megathrust; however, that study is based on a coarse grid resolution of 90 m and 100 source models alone.

This study presents an extensive tsunami hazard assessment for the
Nankai–Tonankai Trough events, focusing on the southwestern Pacific region
of Japan. Numerical simulations are conducted for two magnitude ranges

The Nankai–Tonankai Trough is the primary source of major offshore
subduction earthquakes and tsunamis in western and central Japan. The
driving tectonic mechanism of these seismic activities is the Philippines
Sea Plate subducting underneath the Eurasian Plate with slip rates between
40 and 55 mm yr

The Nankai–Tonankai Trough region can be divided into six segments (Garret et al., 2016) as shown in Fig. 1. The geological evidence of the Hyuga-nada segment (Z in Fig. 1) is limited, and the current seafloor geodetic observations and numerical results indicate relatively low coupling in this segment (Yokota et al., 2016; Kimura et al., 2019). The Nankai segments (A–B in Fig. 1) ruptured many times in the past, and abundant evidence has shown that they are capable of producing high tsunami waves and large-scale coastal inundations (e.g., Tanigawa et al., 2018). The rupture history of the Tonankai segments (C and D in Fig. 1) is also well studied (e.g., Fujino et al., 2018); ruptures have resulted in widespread shaking and tsunami effects in the central Pacific region of Japan. Seismic imaging surveys conducted by Kodaira et al. (2006) indicated that the shallow fractured portion of the crust off the Kii Peninsula (between segment B and segment C) can serve as a physical boundary between the Nankai and Tonankai segments (e.g., the 1854 Ansei and 1944 and 1946 Showa events). On the other hand, the same study also indicated that the deeper portion of the crust underneath the Kii Peninsula is strongly coupled, and thus a large-scale synchronized rupture, such as the 1498 Meio and 1707 Hoei events, is possible, which would be analogous to the case of the 2011 Tohoku event. The Tokai segment (E in Fig. 1) ruptured synchronously with the Tonankai segments in the past and generated widespread geological evidence of large seismic events (e.g., Fujiwara et al., 2020).

For tsunami hazard mapping of a future Nankai–Tonankai megathrust event, the
CDMC developed 11 tsunami source models by considering that the synchronized
rupture over multiple segments is possible and that the magnitude of a future
Nankai–Tonankai earthquake can be as large as

The entire fault plane of the 2012 CDMC models is represented by a set of
5773 sub-faults, each sub-fault having a size of 5 km by 5 km. The fault
plane consists of the main subduction interface (5669 sub-faults) and the
Kumano splay fault (104 sub-faults, which are located in segment C
around a depth contour of 5 km), and geometrical parameters (i.e., strike,
dip, and rake) of these sub-faults are variable over the curved, steepening
fault plane along the dip direction. The total fault-plane area is approximately

To determine the slip distributions of the 2012 CDMC source models, two
types of asperity, i.e., large and very large, are considered. The large slip
areas take up circa 20 % of the entire fault plane with an average
slip value twice as large as the average slip over the fault plane, whereas
the very large slip areas take up circa 5 % of the entire fault plane with
an average slip value 4 times as large as the overall average slip. The
large slip areas are positioned at depths shallower than 20 km, and the very
large slip areas are positioned at depths less than 10 km within the
large slip areas. Among the 11 models (Fig. 2), models 1 to 5 have a
single asperity region (i.e., large and very large slip areas) with slips
concentrated in segments C–E, B–D, B and C, A–B, and Z–B,
respectively. Models 6 and 7 take into account the splay fault in Kumano
Sea with the asperity in segments D and E or in segment B. Models 8 to 11 consider two asperities in segments C–E, B–D, A and C, and Z–B, respectively. The rupture nucleation is set to occur near the center of an asperity region at a depth of about 20 km, and it varies for individual cases. The slip values for individual sub-faults are determined based on spatially varying convergence rates at the sub-faults given that the total slip amount (or seismic moment) within the large slip and very large slip areas is conserved according to the above-mentioned average slip values. In the tsunami source models, a kinematic rupture is represented by a set of earthquake slip distributions (with temporal interval of 10 s) by considering the rupture propagation velocity equal to 2.5 km s

Stochastic tsunami simulations for the Nankai–Tonankai megathrust events
require the generation of multiple earthquake rupture models that are
suitable for representing geometrical characteristics and spatial slip
distributions for the target region. The rupture geometry, such as length
and width, and average and maximum slips over a fault plane typically scale
with earthquake magnitude (e.g., Murotani et al., 2013; Thingbaijiam et al.,
2017), whereas the spatial slip distributions can be specified by wavenumber
spectral functions (e.g., Mai and Beroza, 2002). To determine the main source
parameters, statistical scaling relationships of the eight source parameters
(i.e., fault length, fault width, mean slip, maximum slip, power
transformation parameter for the marginal slip values, along-strike
correlation length, along-dip correlation length, and Hurst number) that are
developed based on a large number of inverted source models of the past
major earthquakes can be employed (Goda et al., 2016). Subsequently, the
stochastic simulation of constrained random slip distributions is performed
to generate 500 earthquake rupture models for the two magnitude ranges

Numerical steps of stochastic tsunami simulations.

In generating the stochastic rupture models for the Nankai–Tonankai megathrust events, the 2012 CDMC fault plane (Fig. 1) is adopted as a baseline. Both strike and dip angles of sub-faults, each having a dimension of 5 km by 5 km, are variable and reflect the current tectonic setting in the Nankai–Tonankai Trough region (Hirose et al., 2008). These features are retained in the stochastic rupture models. Specifically, the sub-faults of the 2012 CDMC model are mapped onto a 2D (rectangular) matrix, noting that sub-faults for the Kumano splay fault are excluded. The size of the 2D matrix is 153 (along-strike) by 53 (down-dip), and its origin is set to the southwestern corner (upper-left) of the fault plane.

Because the fault length

For a given fault plane geometry, the earthquake slip distribution is
determined based on marginal slip statistics and spatial slip distribution
parameters. A candidate slip distribution is first simulated from an
anisotropic 2D von Kármán wavenumber spectrum with its amplitude
spectrum being parameterized by along-strike correlation length, along-dip
correlation length, and Hurst number and its phase being randomly
distributed between 0 and 2

To ensure that the simulated earthquake slip distribution has realistic
characteristics for the Nankai–Tonankai megathrust events, several checks on
the simulated slip distribution are conducted. The maximum slip of the
simulated earthquake slip distribution is required to be less than 77 and
57 m for

In modelling the Nankai–Tonankai megathrust events, a kinematic rupture is
taken into account (Goda et al., 2018). The probability density functions
for hypocenter locations are defined based on the statistical models
developed by Mai et al. (2005). The preliminary probability density
functions for hypocenter locations are specified based on the fault
dimensions and the mean and maximum slip ratios. Subsequently, further
constraints are placed to exclude unlikely hypocenter locations for a given
slip distribution using the empirical findings of Mai et al. (2005). By
combining the preliminary probability density functions and the constraints,
the final probability density function for hypocenter locations is obtained.
This is used to sample the location of hypocenters. Using the randomly
generated rupture propagation velocity and rise time, the kinematic rupture
process of the synthesized slip distribution can be simulated. In this
kinematic rupture modelling, the rupture propagation velocity is
characterized by a truncated normal variable with mean equal to 2.5 km s

Finally, by repeating the above-mentioned procedure for earthquake rupture
modelling 500 times for the two magnitude ranges

For a given earthquake rupture model, the following calculation steps are implemented to evaluate various tsunami hazard characteristics and parameters, such as wave profiles at offshore locations, maximum tsunami heights along coastal lines, and maximum inundation depths at onshore locations (see Sect. 4). The numerical tsunami inundation analysis follows a standard procedure, namely, computing a vertical dislocation of seawater triggered by an earthquake rupture (Okada, 1985; Tanioka and Satake, 1996) and solving nonlinear shallow water equations for tsunami wave propagation and run-up (Goto et al., 1997).

To represent a computational domain of the tsunami inundation simulations
accurately, a complete dataset of bathymetry and elevation, coastal and riverside
structures (e.g., breakwater and levees), and surface roughness is obtained
from the Japanese Cabinet Office (same as the 2012 CDMC models). The data
are provided in the form of six nested grids following a

The ocean-floor topography data are based on the

The elevation data of the coastal/riverside structures are provided by
municipalities and supplemented by the national coastline database. In the
coastal/riverside structural dataset, structures having dimensions less than
10 m are represented, whereas those having dimensions greater than 10 m are
included in the elevation data. In the tsunami simulation, the
coastal/riverside structures are represented by vertical walls at one or two
sides of the computational cells. To evaluate the volume of water that
overpasses these walls, overflowing formulae of Honma (1940) are employed for
coastal breakwater modelling at sub-grid scales. The failures of the
coastal/riverside structures are not considered in the simulations. On the
other hand, the bottom friction is evaluated using Manning's formula
following the Japan Society of Civil Engineers standard (JSCE, 2002). The
Manning's coefficients are assigned to computational cells based on national
land use data in Japan: 0.02 m

The vertical dislocation profile of seawater due to an earthquake rupture is computed using Okada equations (Okada, 1985) evaluated at a 810 m grid resolution. To account for the effects of horizontal movements of steep seafloor on the vertical dislocation of seawater, a method proposed by Tanioka and Satake (1996) is implemented. To alleviate the abrupt changes in the vertical dislocation of seawater, a spatial smoothing filter of 9 cells by 9 cells is employed in a similar way to the 2012 CDMC models. To represent the kinematic rupture process in a tsunami simulation, these vertical dislocation profiles are evaluated at every 10 s and are used as input in the simulation.

The tsunami modelling is carried out using a well-tested TSUNAMI code (Goto et al., 1997) that solves the nonlinear shallow water equations using a leap-frog staggered-grid finite difference scheme which is capable of generating offshore tsunami propagation and onshore run-up. The run-up calculation is based on a moving boundary approach (JSCE, 2002), in which a dry/wet condition of a computational cell is determined based on total water depth relative to its elevation. The numerical tsunami calculation is performed for a 3 h duration which is sufficient to model the most critical phase of tsunami waves for the Nankai–Tonankai scenarios. The multi-domain nesting from coarse to fine resolution is conducted to consider large to small scale tsunami waves, depending on water depth. The time-stepping intervals for the regional (90 m) and local (10 m) tsunami simulations are set to 0.5 and 0.1 s, respectively, to satisfy the Courant–Friedrichs–Lewy condition. Moreover, the effects of ground deformation are taken into account by adjusting the elevation data prior to the tsunami simulation run. However, the effects of tidal variations (e.g., mean high water level) are not considered in the simulations.

This section presents the uncertainty quantification of tsunami inundation
assessments for Shikoku Island and Kochi Prefecture (regional focus) and for Kuroshio (local focus) subjected to stochastic rupture events originating from
the Nankai–Tonankai Trough. The magnitude ranges of the stochastic rupture
models are

Elevation maps of

In Sect. 4.1, offshore wave profiles and maximum tsunami heights of the stochastic tsunami simulations are discussed in comparison with the counterparts based on the 2012 CDMC models. In Sect. 4.2, tsunami inundation areas in Shikoku and Kuroshio are adopted as regional and local tsunami hazard metrics, respectively, and their relationships with earthquake source characteristics, such as moment magnitude and regional concentration of earthquake slip, are investigated. In Sect. 4.3, tsunami inundation characteristics in the Ogata and Saga districts are focused on. In particular, the severity of local tsunami hazards in these districts is examined by evaluating critical tsunami scenarios.

To evaluate the regional tsunami hazard characteristics for Shikoku Island
(mainly Kochi Prefecture), offshore tsunami wave profiles at three locations, indicated as P1–P3 in Fig. 4a, are examined in Fig. 5. These locations are near the city of Shimanto, the city of Kochi, and Cape Muroto and are at sea depths of 36, 45, and 70 m, respectively. For each of the two
magnitude ranges

Offshore tsunami wave profiles at P1–P3 (Fig. 4a) for the two magnitude ranges

The largest first tsunami heights reach 16.8, 9.5, and 16.6 m at P1–P3, respectively, for the

Subsequently, to compare the maximum tsunami heights at near-shore locations
along the coast of Shikoku Island based on the stochastic tsunami models and
the 2012 CDMC models, 154 observation points are set up starting from Sukumo
in Kochi Prefecture to Naruto in Tokushima Prefecture (Fig. 6a). The
extracted near-shore maximum tsunami height results shown in Fig. 6 are
based on a 90 m resolution. The water depths at these observation points are
at sea depths of approximately 5 to 10 m. Figure 6b and c plot the maximum
tsunami heights along the coast for the two magnitude ranges

Maximum tsunami heights along the coast of Kochi and Tokushima
prefectures

To facilitate the comparisons of the stochastic tsunami simulation results
and the 2012 CDMC results, the ratios of the three percentiles of the
maximum tsunami heights based on the stochastic tsunami simulations and the
average of the 2012 CDMC results are shown in Fig. 7. The maximum tsunami
heights based on the 2012 CDMC models are closer to the 50th percentiles of
the stochastic tsunami simulation results for both

Ratio of maximum tsunami heights between three percentiles of the
stochastic source models and the average of the 11 CDMC models along the coast of Kochi and Tokushima prefectures for the two magnitude ranges

The extent of onshore tsunami inundation is a useful hazard metric
for damage and loss (e.g., Goda et al., 2019). It is thus essential
to investigate the characteristics of regional and local inundation area
metrics and to relate these to the corresponding earthquake characteristics.
In this section, we study the relationships between inundation area
parameters and three types of earthquake source parameters. The first source
parameter is the moment magnitude, which captures the macroscopic feature of
released energy and is proportional to the total slip over a fault plane.
The second source parameter is the slip ratio, which is calculated as the
summed slip within a specified segment divided by the total slip over the
entire fault plane. For instance, the slip ratio can be calculated for
individual or combined segments of the Nankai–Tonankai fault plane model
(Fig. 1). The third source parameter is the tsunami potential energy

Figure 8 shows histograms of the tsunami inundation areas in Shikoku and Kuroshio for the two magnitude ranges

Histograms of tsunami inundation areas in Shikoku

To examine the relationships between regional and local inundation areas and
earthquake slip distributions in segments, tsunami inundation areas in
Shikoku and Kuroshio are plotted against slip ratios in segments Z (Hyuga-nada), A–B (Nankai), and C–E (Tonankai–Tokai) in Fig. 9. It is
noted that for a given earthquake source model, the slip ratios in segments Z, A–B, and C–E sum to 1. From the scatter plots, positive and negative trends can be recognized for segments A–B and C–E, respectively, whereas no clear trends can be identified for segment Z (see also the Pearson's linear correlation coefficient

Scatter plots of slip ratios in segments Z (Hyuga-nada), A–B (Nankai), and C–E (Tonankai–Tokai) versus tsunami inundation areas in Shikoku

For investigating the effects of spatial earthquake slip distribution on the severity of tsunami hazard, the tsunami potential energy (Eq. 1) is an ideal metric because it consolidates the effects of various features of earthquake sources into a single parameter. Figure 10 presents scatter plots of moment magnitude versus tsunami inundation area, as well as those of tsunami potential energy versus tsunami inundation area. Figure 10a and b exhibit moderate positive relationships between moment magnitude and inundation areas (which can also be inspected in Fig. 8). In contrast, Fig. 10c and d show stronger positive relationships (i.e., linear correlation coefficients of 0.6 to 0.7) between logarithmic tsunami potential energy (with base 10) and inundation areas. The positive correlation is more pronounced for the regional inundation area in Shikoku than for the local inundation in Kuroshio because there is a greater chance that elevated tsunami potential energy will affect the target region or location.

Scatter plots of moment magnitude versus tsunami inundation area in Shikoku and Kuroshio for the two magnitude ranges

Although not presented in this study, similar correlation analyses are also carried out for other earthquake source parameters, such as fault length and width and maximum slip. These results indicate that tsunami potential energy shows high degrees of correlation with regional and local inundation areas and thus can be used as an effective tsunami source predictor of regional and local inundation extents.

Assessing tsunami inundation hazards and quantifying their uncertainty at local community levels require an integrated modelling and analysis of stochastic earthquake sources and tsunami run-up simulations. This section focuses on (1) site-specific evaluations of tsunami inundation depths at two vertical evacuation towers in the Ogata and Saga districts (Fig. 4) and (2) the identification of critical tsunami scenarios based on local inundation metrics and their comparisons with the 2012 CDMC model cases. The former is crucially important to ensure that the current evacuation plans for the local population in these districts are effective in the case of extreme situations beyond the current tsunami hazard scenarios by the CDMC. The latter facilitates the risk communication of tsunami hazard threat by recognizing different possible tsunami hazard scenarios explicitly in light of inevitable uncertainty associated with tsunami hazard assessment.

The evacuation space of the tower (usually the top floor or the roof) in the
Ogata district is at 18.2 m a.m.s.l. (above mean sea level) (Fig. 4d); by taking into account the land elevation of 3.87 m at the tower, the critical inundation depth at the tower is 14.33 m. On the other hand, the evacuation space of the tower in the Saga district is at 25.3 m (Fig. 4e); therefore, the critical inundation depth at the tower is 21.92 m when considering the land elevation of 3.38 m. These critical water depths are used for evaluating the sufficiency of these two towers as vertical refuge facilities. To carry out such assessments, histograms of maximum inundation depth at the vertical evacuation towers in the Ogata and Saga districts are shown in Fig. 11 for the two magnitude ranges

Histograms of maximum inundation depth at the vertical evacuation
towers in the Ogata and Saga districts (see Fig. 4) for the two magnitude
ranges

Next, to derive critical tsunami hazard scenarios based on local inundation
areas, inundation areas in the Ogata and Saga districts are investigated,
and histograms of these inundation parameters are shown in Fig. 12. Note
that the areas that are considered for these local communities are smaller
than those considered for Kuroshio (i.e., Fig. 8b and d). Based on
the histograms shown in Fig. 12 and by taking the 50th and 90th percentiles as critical scenario levels (note that other percentiles can be adopted), the critical inundation areas obtained are 2.20 and 3.71 km

Histograms of inundation areas in the Ogata and Saga districts for the two magnitude ranges

Critical tsunami hazard scenarios for the Ogata and Saga districts for the

Once the earthquake rupture models that correspond to the identified critical inundation areas have been completed, various stochastic tsunami simulation results, such as regional maximum tsunami heights and local maximum inundation depths, can be extracted. For such purposes, the earthquake slip distribution, the maximum tsunami height distribution, the maximum inundation depth in the Ogata district, and the maximum inundation depth in the Saga district for the 50th and 90th percentile inundation levels are displayed in Fig. 13 for the

Critical tsunami hazard scenarios for the Ogata and Saga districts for the

Significant advantages of the stochastic tsunami simulation methods and their use in deriving critical tsunami scenarios and related tsunami hazard maps are the enhanced capabilities to quantify the uncertainty associated with tsunami hazard assessments and to visualize the results in an integrated manner more effectively. It is also important to point out that the deterministic tsunami scenarios, such as the 2012 CDMC models, and the stochastic tsunami scenarios, such as the 1000 rupture models, are complementary. Typically, deterministic scenarios are derived based on current geodetic and available geological data, whereas stochastic scenarios are more inclined to statistical features of earthquake source models of past events. The systematic comparison of these results will improve the understanding of different modelling approaches and their key assumptions and will allow hazard modellers to gain confidence in the derived results.

Tsunami hazard assessments based on stochastic tsunami simulations offer
valuable insights into the degree of uncertainty associated with such
investigations. To evaluate tsunami inundation hazards and quantify their
variability at local community levels, this study presented an integrated
modelling and analysis of stochastic earthquake sources and tsunami run-up
simulations for future Nankai–Tonankai megathrust events. For this
purpose, 1000 kinematic tsunami rupture models were generated for the
moment magnitudes ranging from 8.7 to 9.1, and tsunami simulations were
performed with high-resolution grid data of 10 m by taking into account the
effects of existing coastal defence structures in Japan. To benchmark the
results from the stochastic tsunami simulations, a set of tsunami source
models developed by the CDMC of the Japanese Cabinet Office was employed.
Our motivations in comparing the stochastic tsunami inundation results with
the deterministic 2012 CDMC models were to quantify the variability of
tsunami hazards at municipality and township levels and to account for local
extreme situations. To enable a consistent comparison with the CDMC tsunami
source models (

The numerical investigations focused on the southwestern Pacific region of Japan, i.e., Shikoku Island and Kuroshio, for regional and local viewpoints. The comparisons of the offshore wave profiles and maximum tsunami heights along the coast of Shikoku Island based on the stochastic tsunami simulations and the 2012 CDMC models indicated that the 2012 CDMC results are consistent with the typical stochastic simulation results for the same magnitude range but are unable to capture extreme scenarios of local tsunami hazards and their variability.

To relate the regional and local inundation extents to earthquake source characteristics, correlations between inundation area metrics and moment magnitude, slip ratio in segments, and tsunami potential energy were examined. The results indicated that tsunami potential energy could be used as an effective tsunami source predictor of regional and local inundation extents. Moreover, to evaluate the sufficiency of the two existing vertical evacuation towers in the local communities of Ogata and Saga in Kuroshio, the critical inundation depths of these towers were compared to the stochastic tsunami simulation results. Since the exceedance of the critical inundation depths was rare (i.e., 5 and 1 out of 1000 cases for the towers in the Ogata and Saga districts, respectively), the existing two vertical evacuation towers were judged to be satisfactory. Finally, critical tsunami scenarios for the local communities in Ogata and Saga were identified based on the local inundation metrics, and their tsunami inundation extents were compared with those of the 2012 CDMC models. The use of stochastic tsunami simulation methods improved the quantification and visualization of uncertain tsunami hazard assessments and was able to complement the tsunami hazard predictions based on conventional deterministic tsunami scenarios. Future investigations should extend the hazard assessments into tsunami evacuation problems, as well as into multi-hazard risk assessments.

The bathymetry and elevation data were obtained from the Central Disaster Management Council of the Japanese Cabinet Office. The tsunami simulation code is not available because its copyright does not belong to the authors.

KG carried out the analysis and writing. KG, NM, TY, RDR, AM, and FDL all participated in the study site visit and a series of meetings and verified the analysis set-up, as well as the results.

The authors declare that they have no conflict of interest.

The tsunami simulations were performed using BlueCrystal Phase 3, which was supported by the Advanced Computing Research Centre of the University of Bristol.

This research has been supported by the Leverhulme Trust (grant no. RPG-2017-006), the Canada Research Chairs (grant no. 950-232015), the Natural Sciences and Engineering Research Council of Canada (grant no. RGPIN-2019-05898), and the Japan Society for the Promotion of Science Joint Research Project (grant no. 19039901-000829).

This paper was edited by Maria Ana Baptista and reviewed by Thomas Ulrich, Reza Amouzgar, and four anonymous referees.