- Full paper
- Open Access

# Should tsunami simulations include a nonzero initial horizontal velocity?

- Gabriel C. Lotto
^{1}Email author, - Gabriel Nava
^{1, 3}and - Eric M. Dunham
^{1, 2}

**Received:**18 May 2017**Accepted:**8 August 2017**Published:**23 August 2017

## Abstract

## Keywords

- Tsunami
- Tsunami modeling
- Tsunami generation
- Offshore earthquake
- Subduction zone
- Shallow water wave equations

## Introduction

Tsunamis from earthquake sources pose a significant threat to coastlines around the world and have killed hundreds of thousands of people in the last few decades. The Earth science community often relies on numerical models to better understand tsunami physics (e.g., Titov and Synolakis 1998; Tanioka and Satake 1996; Kowalik 2003), perform hazard assessments (e.g., Geist and Parsons 2006; Borrero et al. 2006; González et al. 2009), and plan for early warning (e.g., Titov et al. 2005; Melgar and Bock 2013; Behrens et al. 2010).

For numerical modelers, the tsunami life cycle is often divided into three parts: generation, propagation, and inundation (Bernard et al. 2006; Mori et al. 2011; Titov and Gonzalez 1997; Watts et al. 2003; Song and Han 2011). This paper is entirely focused on the first two parts, though modeling inundation poses its own unique set of challenges and is far from being completely understood.

*x*and

*y*the horizontal and vertical spatial coordinates, respectively,

*t*is time, \(h = h(x,t)\) is the depth of the ocean, \(v = v(x,t)\) is the depth-averaged horizontal velocity, and

*g*is the gravitational acceleration. As in Fig. 1, we define the sea surface perturbation about \(y = 0\) as \(y = \eta (x,t)\) and the position of the seafloor as \(y = -H(x)\), so that \(h(x,t) = \eta (x,t) + H(x)\). Linearization is justified assuming \(|\eta (x,t)| \ll |H(x)|\). The equations describe nondispersive wave propagation at speed \(\sqrt{gH}\).

Time-dependent seafloor displacement, which generates the tsunami, can be accounted for through an additional term in the mass balance Eq. (1). However, it is more common to account for seafloor displacement indirectly through initial conditions. This is justified when the tsunami propagation distance over the rupture duration is much smaller than horizontal scales of interest. This condition is almost always satisfied because earthquake ruptures propagate at elastic wave speeds that are an order of magnitude faster than the tsunami propagation speed (Kajiura 1970).

*H*when translating the seafloor uplift to the sea surface (Kajiura 1963; Saito and Furumura 2009; Saito 2013).

We briefly note an important consequence of choosing a nonzero value for \(v_0(x)\), which might be testable with tsunami waveform observations if there is sufficient uniqueness in the solution of the inverse problem of determining initial conditions from data. A nonzero initial velocity introduces asymmetries in the sea surface profile as the tsunami propagates, as illustrated by the simple example shown in Fig. 2. A tsunami model with an initial condition other that (7) must justify this asymmetry. To the best of our knowledge, this asymmetry is not seen in tsunami observations, though it might be challenging to detect if the initial \(v_0(x)\) is rather small and bathymetric effects complicate the tsunami waveform. We return to this point subsequently.

## Modeling approach

As mentioned above, we perform two parallel sets of tsunami simulations, one with a depth-resolved, coupled solid Earth-ocean model and the other with a depth-averaged shallow water wave model.

The fully-coupled, or full-physics, model is a 2-D depth-resolved model introduced by Lotto and Dunham (2015), building on earlier work by Maeda and Furumura (2013). We solve the elastic wave equation in the solid Earth and the acoustic wave equation in the compressible, inviscid ocean, and incorporate gravity via a modified free surface boundary condition. Small wave amplitudes, relative to ocean depth, are assumed to justify a linearization of the free surface condition. However, no assumption is made regarding horizontal wavelengths relative to ocean depth; the ocean response becomes nonhydrostatic at short wavelengths, giving rise to well-known effects such as the filtering of short wavelength features in the initial sea surface height and dispersion during tsunami propagation (Kajiura 1963, 1970). In the full-physics model, the earthquake source and all waves are simultaneously generated, unlike the typical approach of first modeling the earthquake and then using the final displacement from the earthquake model to set initial conditions for the tsunami. We use this model to simulate the full seismic, ocean acoustic, and tsunami wavefields. This model contains the full physics of tsunami generation; the seafloor moves dynamically in response to fault rupture and that motion translates directly to motion and wave propagation in the water. Forces exerted by the solid Earth on the ocean generate both ocean acoustic waves and tsunamis.

We employ two structural models in this study. The first is an idealized subduction zone geometry with a homogeneous solid Earth, and the second is based on our previous studies of the 2011 Tohoku earthquake (Kozdon and Dunham 2013, 2014). In the idealized model, an ocean with a sloping seafloor rests above a homogeneous solid Earth, with a planar thrust fault intersecting the seafloor at its deepest point (Fig. 3). The fault dips at \(10^{\circ }\). This shallowly dipping fault is typical of subduction zone plate boundaries, especially near the trench (e.g., Kopp and Kukowski 2003; Gutscher et al. 2001). The seafloor bathymetry to the right of the trench is flat, for simplicity. To the left of the trench, the bathymetry is set by a hyperbolic curve that asymptotes to horizontal away from the trench and to a slope of angle \(\theta\) near the trench. We vary \(\theta\) across simulations, which changes the depth of the trench and the slope of the seafloor. Results with different \(\theta\) are all qualitatively similar, so only simulations with \(\theta = 10^{\circ }\) are reported here. The vertical position of the seafloor is defined as \(y = -H(x)\); \(y = -H_1 = -1\) km at the landward (left) side of the domain and \(y = -10\) km at the trench for \(\theta = 10^{\circ }\). The trench is slightly deeper than in most subduction zones, but this hyperbolic seafloor shape is a simple way to parameterize the geometry while avoiding too many free parameters.

*T*is the shear strength of the fault, \(\bar{\sigma }\) is the effective normal stress,

*V*is the slip velocity, \(l=0.8\) m is the state evolution distance, \(f_0 = 0.6\) is the friction coefficient for steady sliding at reference velocity \(V_0 = 1\,\upmu {\mathrm {m/s}}\), \(b=0.02\), and \(a=0.016\). The dimensionless parameter \(b-a\) determines whether friction increases or decreases with changes to

*V*. Since \(b-a = 0.004\) for our simulations, friction is velocity-weakening and unstable slip can occur.

*B*is Skempton’s coefficient.

Initial effective normal stress \(\bar{\sigma }_0\) is calculated as the difference between normal stress in the Earth and pore pressure. Normal stress on the fault is set to lithostatic, and pore pressure is hydrostatic; effective normal stress increases with depth to a maximum of \(\bar{\sigma }_0 = 40\) MPa, at which point pore pressure increases lithostatically. Initial shear stress on the fault, \(T_0\), is set to \(T_0 = 0.6 \bar{\sigma }_0\).

We compare the results of the full-physics model to those of a 2-D linearized shallow water wave model (i.e., Eqs. 1, 2), for which we use the same bathymetry as the full-physics model and set initial conditions based on the results of the full-physics model as follows. To set the initial sea surface height, \(\eta _0(x)\), we first run a simulation using the full-physics model where we set the gravitational acceleration to \(g=0\) so that surface gravity waves do not propagate. After seismic and ocean acoustic waves propagate out of the domain, the sea surface is statically perturbed in a profile that resembles the seafloor displacement but with short wavelength variations filtered out (Kajiura 1963; Lotto and Dunham 2015; Saito 2013). This static sea surface perturbation becomes the initial condition on \(\eta _0(x)\) for the shallow water model.

We set the initial condition on depth-averaged velocity \(v_0(x)\) using the expressions provided by Song et al. (2008) and are summarized in Eqs. (8) and (9) and Fig. 4. Here, \(u_x(x)\) is taken from the final displacement of the seafloor in the zero gravity version of the full-physics simulation, \(R_x\) comes from the model geometry, and the rise time \(\tau\) is a free parameter that we vary to control the amplitude of the initial velocity.

## Comparing full-physics to shallow water simulations

Here, we compare the results of the shallow water model to those of the full-physics model, using the idealized material structure and geometry for both models. In order to make a proper comparison, we must first account for a time delay that comes about because the full-physics model begins with an earthquake rupture that generates the tsunami over tens of seconds whereas the shallow water model is initialized at time \(t=0\) with a tsunami profile already in place. We calculate the time delay by comparing \(\eta (x,t)\) at each time step of the full-physics simulation to the initial profile \(\eta _0(x)\) of the equivalent shallow water simulation (which, as described previously, is set by a zero gravity full-physics simulation). We calculate the misfit with respect to \(\eta _0(x)\) at each time step using the \(L_2\) norm, with the time of minimum misfit is taken to be the time delay between the full-physics model and the shallow water model. Alternatively, we can calculate the time delay by selecting some time step of \(\eta (x,t)\) in the full-physics model after which all acoustic waves have exited the domain and matching \(\eta (x,t)\) of the shallow water model, subject to some time delay, to that in the full-physics model. The time delay from this alternate approach yields basically the same results as the approach previously described. Though the time delay changes slightly between simulations, the offset is typically \(\sim\)85 s, meaning that it takes about 85 s (roughly, the rupture duration) for the full-physics model to generate a tsunami profile that matches the initial condition of the shallow water code.

We run two sets of shallow water simulations with initial horizontal velocity set by Eqs. (8) and (9), one using \(L_H = 1.5H_\mathrm{{max}}\) and the other using \(L_H = 0.5H(x)\). The only remaining free parameter is the rise time \(\tau\), for which Song et al. use a value of \(\tau \sim 5\)–20 s for their Sumatra and Tohoku simulations (Song and Han 2011; Xu and Song 2013; Song et al. 2017). We vary \(\tau\) over a much larger range, as we find that using \(\tau \sim 5\)–20 s leads to tsunami waves of unrealistically large amplitudes, as evident in the bottom rows of Fig. 5.

Rather, we find that the shallow water tsunami best fits the full-physics tsunami when rise time \(\tau\) approaches infinity (top rows of Fig. 5), i.e., when \(v_0(x) = 0\). Note that even when tsunami amplitudes match, the waveforms do not align perfectly because the tsunami experiences dispersion of short wavelengths in the full-physics model but not in the shallow water model. With \(\tau\) interpreted as the local rise time, setting \(\tau = \infty\) is inconsistent with the assumption that the seafloor completes its deformation over a timescale smaller than the tsunami propagation time over horizontal wavelengths characterizing the deformation profile. However, more generally \(\tau\) can be viewed simply as a scalar parameter that controls the amplitude of the initial velocity, with the shape or profile still given by Eqs. (8) and (9).

Shallow water models with non-negligible initial velocity (i.e., those with finite rise times) produce the same sort of asymmetry in tsunami waveform as illustrated in the simple example in Fig. 2. In contrast, the full-physics simulation yields a tsunami profile that is nearly symmetrical in amplitude between seaward and landward waves, though this is complicated somewhat by the nonuniform ocean depth.

We can quantify the misfit between the tsunamis produced by the shallow water and full-physics models by taking the \(_2\) norm of the difference between the final sea surface heights plotted in Fig. 5. Those misfits are shown in Fig. 8a. The misfit is never zero because the full-physics model accounts for the finite source duration, surface gravity wave dispersion, and ocean compressibility, which are neglected in the shallow water model. For either choice of \(L_H\), the smallest misfits occur in the limit as rise time approaches infinity, i.e., \(\tau \rightarrow \infty\) so \(v_0 \rightarrow 0\). When using \(L_H = 0.5 H(x)\), we find that misfits are nearly at their minimum for \(\tau \ge 40\) s. In Fig. 9a, we plot kinetic energy (KE) as a portion of the total energy in the ocean at the initial time step of each shallow water simulation. We define energy and discuss it more thoroughly in “Energy considerations” section, but for now it is sufficient to view KE as a quantitative scalar measure of particle velocities in the ocean. By comparison with Fig. 8a, we observe that models with minimum misfit are the same as those with negligible initial KE. In other words, even though models with \(L_H = 0.5 H(x)\) and \(\tau \ge 40\) s produce small misfits, they only do so because they essentially have zero initial horizontal velocities.

## Tohoku simulations

To supplement the idealized set of simulations discussed in the previous section, we perform a similar analysis on a 2-D simulation of the 2011 Tohoku-oki earthquake and tsunami. The Tohoku simulation, whose seafloor geometry is pictured in Fig. 6, is nearly identical to the simulation performed by Kozdon and Dunham (2013, 2014) with velocity-strengthening friction near the trench and a heterogeneous material structure. Unlike that prior work, the Tohoku simulation presented here accounts for gravity and tsunami generation and propagation, which allows us to test the Song et al. momentum transfer hypothesis for a more realistic problem.

Note that, as seen in Fig. 6, the ocean layer rises above and covers the land in an unphysical way. This is an artifact of the way we break our full-physics model into curvilinear computational “blocks,” but it does not affect our results as long as we are interested in tsunami propagation prior to wave arrival at the coast.

Just as we did for the simplified geometry in the previous section, we run two shallow water simulations setting initial horizontal velocity with Eqs. (8) and (9), one using \(L_H = 1.5H_\mathrm{{max}}\) and the other using \(L_H = 0.5H(x)\). The bathymetry of the full-physics simulation determines *H*(*x*) in the shallow water models, and the initial condition \(\eta _0(x)\) comes from the final displacement of the sea surface in a full-physics Tohoku simulation with no gravity. As with the idealized geometry simulations, the time delay between the full-physics and shallow water simulations is calculated by finding the time step of least misfit between \(\eta (x,t)\) of the full-physics model and \(\eta _0(t)\) of the shallow water model. For the Tohoku geometry, that misfit is \(\sim\)70 s.

The results, shown in Fig. 7, demonstrate once again that the best choice for rise time using the Song et al. approach in Eq. (8) is \(\tau = \infty\), i.e., \(v_0(x) = 0\). We quantify the misfits between the final tsunamis produced by the shallow water and full-physics models in Fig. 8b, and we observe that misfits are lowest as \(\tau \rightarrow \infty\) and \(v_0 \rightarrow 0\). Just as in the previous section, misfits are at their minimum when initial kinetic energy (KE) in the ocean is negligible (Fig. 9b), i.e., when \(v_0(x) = 0\). These results do not indicate that \(v_0(x) = 0\) is the correct, or even most self-consistent, choice for an initial condition on horizontal velocity for shallow water models, as this experiment explores only two possible initial velocity profiles. However, the results do imply that with these velocity profiles, the Song et al. theory makes inconsistent predictions of the tsunami, especially when choosing \(\tau = 5{-}20\) s. We furthermore see no evidence arguing against using \(v_0(x)=0\) as the initial velocity for most tsunami modeling problems.

## Adjoint problem

Next, we use an alternative approach to determine tsunami initial conditions that avoids any assumptions regarding the initial velocity profile. In concept, the optimal way to constrain the initial condition on horizontal velocity is through direct examination of the particle velocity field of our full-physics model. If we integrate horizontal velocity over the ocean depth at an appropriate “initial” time, we could simply use that horizontally variable, depth-averaged velocity profile as an initial condition for a shallow water simulation. However, this approach is unfeasible: Setting aside the question of how to select an initial time, our full-physics model produces large amplitude seismic and ocean acoustic waves that interfere with the wavefield of the tsunami for the first few hundred seconds. This makes it impossible to isolate the tsunami particle velocity from the total particle velocity for use in setting the initial condition on \(v_0(x)\).

However, appropriate (i.e., self-consistent) tsunami initial conditions can be determined using the following adjoint wavefield procedure. We first run a full-physics simulation (with the same idealized structure as in Fig. 3) for 1000 s of simulation time, long enough for all seismic and acoustic waves to exit through the boundaries of the domain, leaving only the slower-propagating tsunami in the ocean. Having isolated the tsunami, we then use the stress and velocity fields at the end of this first simulation as an initial condition on a time-reversed adjoint simulation. In addition to running backward in time, the adjoint simulation involves flipping the sign of the particle velocity. Since the wave propagation problem is self-adjoint, no changes are required to the code to perform an adjoint simulation. In the adjoint simulation, the tsunami propagates backward in time, and landward and seaward propagating tsunami waves converge above the earthquake source into the initial tsunami profile. We determine the initial condition by finding the time where the sea surface profile from the adjoint simulation best matches the sea surface profile of the equivalent forward simulation with zero gravity (as in the top of Fig. 10). The sea surface height and depth-averaged horizontal velocity fields at this time are identified as \(\eta _0(x)\) and \(v_0(x)\), the initial conditions that are most consistent with the full-physics simulations.

Not surprisingly, we can find a time step in the adjoint problem where the tsunami profile very closely matches that of the equivalent zero gravity simulation. At this time step, the depth-averaged horizontal velocity in the ocean is of order \(10^{-2}\) m/s and appears in a profile that does not match either of those profiles suggested by Song et al. The adjoint calculation suggests that the self-consistent initial condition is one that has nonzero, but very small, horizontal velocity. This result does not corroborate the shape of either of the profiles proposed by Song et al., nor is the velocity amplitude consistent with predicted initial velocity using the rise times that Song et al. recommend.

## Energy considerations

Song et al. (2008, 2017), Song and Han (2011), and Titov et al. (2016) have frequently brought the concept of energy into their discussion of tsunamis, making estimates of kinetic energy (KE) and gravitational potential energy (GPE) in the ocean, immediately after the earthquake prior to tsunami propagation, and attributing the kinetic energy to momentum transfer from the horizontal motion of the seafloor during tsunami generation. In this section, we calculate the energy balance of the ocean for the full-physics model using the idealized geometry discussed in a previous section.

*E*, as the sum of these three components:

*p*is the pressure, \(K = \rho c_p^2\) is the bulk modulus of water, and \(D=5000\) km is the horizontal distance from the trench to either end of the domain. Energy can escape out of the side boundaries as waves pass through the edges of the domain.

Figure 11 tracks the components of *E* over time for the same full-physics simulation discussed in “Adjoint problem” section. An earthquake nucleates at time \(t=0\), generating seismic waves and deforming the elastic solid. The solid Earth does work on the ocean through normal tractions on the seafloor, transferring energy into the ocean in the form of acoustic waves and surface gravity waves. The fast-moving acoustic waves propagate out of the domain within the first few hundreds of seconds of simulation time, carrying with them much of the KE and AE. After about 300 s, only the tsunami remains within the domain, with relatively constant KE and GPE of approximately equal amplitudes and negligible AE.

At \(t=1000\) s, we begin to run the adjoint problem with reversed time. In the adjoint problem, there is no source of acoustic waves and so the AE continues to be negligible. Both GPE and KE remain relatively constant until the tsunamis that were propagating in opposite directions begin to interfere with each other in the source region. This interference establishes the tsunami initial conditions, with the dashed line in Fig. 11 corresponding to the initial conditions shown in Fig. 10. At this time, virtually all of the energy in the ocean is GPE, with negligible KE, such that KE/(KE + GPE) = 0.002. This observation argues against the claim of Song et al. that the KE of the initial tsunami is higher (or even significantly higher) than the GPE.

However, it is important to note that our energy balance calculation provides quantitative support for the impulse-momentum theory promoted by Song et al. There is indeed a considerable amount of KE imparted to the ocean by the solid Earth during the tsunami generation process. However, almost all of this KE is coupled with AE in the form of acoustic waves, rather than remaining in the vicinity of the tsunami source region to provide a non-negligible contribution to initial velocity.

## Discussion and conclusions

We have taken two unique approaches in our attempt to determine the appropriate initial condition on horizontal velocity \(v_0(x)\) for shallow water tsunami models. The first approach involves running comparable tsunami simulations with a full-physics model and a linearized shallow water wave model, and selecting the initial condition on horizontal velocity for the shallow water model using the expressions presented by Xu and Song (2013), Titov et al. (2016) and Song et al. (2017). We then vary the rise time parameter \(\tau\), which scales the amplitude of \(v_0(x)\), and compare the tsunamis predicted by the shallow water and full-physics simulations.

This comparison is made for two model geometries: an idealized subduction zone geometry (Fig. 3) and a 2011 Tohoku earthquake geometry (Fig. 6). For both geometries, our simulations unequivocally demonstrate that the initial condition on \(v_0(x)\) (Eqs. 8, 9) using \(\tau =\) 5–20 s, as Song et al. have suggested, overestimates tsunami amplitudes by several times. Tsunami amplitudes are nearly an order of magnitude too large when choosing \(L_H = 1.5H_\mathrm{{max}}\), as Song et al. have done in more than one publication (Song and Han 2011; Xu and Song 2013). If we use larger values for \(\tau\) – which may be more physically relevant because the rupture process for typical tsunamigenic earthquakes lasts several tens of seconds—we find that the shallow water model produces results closer to the full-physics simulations, especially when \(L_H = 0.5 H(x)\). But when \(\tau \ge 80\) s and \(L_H = 0.5 H(x)\), the initial horizontal velocity profile is nearly flat, virtually indistinguishable from choosing \(v_0(x) = 0\), i.e., \(\tau = \infty\). Quantitative measures of misfit between the two models (Fig. 8) support the conclusion that shallow water simulations best match their full-physics counterparts as rise time \(\tau\) approaches infinity. Indeed, comparing Figs. 8 and 9 reveals that any shallow water simulation that reasonably matches its full-physics counterpart does so because initial KE—and, accordingly, initial horizontal velocity—is negligible.

The second approach requires us to run a full-physics simulation long enough for all seismic and acoustic waves to leave the domain, and then to run it backward in time to solve the adjoint problem. This adjoint simulation method allows us to use the full-physics simulation to directly determine \(v_0(x)\). In doing so, we again find that \(v_0(x)\) is so small in magnitude so as to be practically equivalent to the null hypothesis, \(v_0(x) = 0\). This method produces a velocity profile that does not qualitatively match that suggested by Song et al. Most notably, the initial velocity profile determined by the adjoint method does not reach its maximum at the trench (\(x=0\)) and is not uniformly positive. We also find that at the time of this adjoint-based initial condition, almost all the energy in the ocean is in the form of gravitational potential energy and virtually none is kinetic energy.

We conclude that Song et al.’s proposed choice of \(v_0(x)\) for modeling tsunamis will consistently overestimate tsunami heights. Furthermore, we have found no convincing evidence of any better choice than \(v_0(x) = 0\), which is the default initial condition on horizontal velocity used by most tsunami modelers.

In a study published after initial submission of our manuscript, Song et al. tested their tsunami generation theory by comparing simulations using initial condition (8 and 9) to tide gauge and DART data from the 2011 Tohoku event (Song et al. 2017). They argued that their tsunami generation theory matches the data better than the conventional tsunami generation theory. However, many other authors have successfully used the standard initial condition \(v_0(x) = 0\) to fit data both landward and seaward of the rupture area (Bletery et al. 2014; Gusman et al. 2012; Hooper et al. 2013; Satake et al. 2013). The fits are equally good. This suggests that there is sufficient nonuniqueness with regard to the tsunami initial conditions given the available tsunami data from the Tohoku event. It follows that we cannot equivocally disambiguate between Song’s initial conditions and the conventional initial conditions from even this well-recorded tsunami.

Song et al. have also used laboratory wave basin tests to create an empirical scaling relationship between the ratio of initial kinetic to gravitational potential energy (KE/GPE) and a normalized version of tsunami amplitude (Song et al. 2017). Using characteristic values for ocean depth, fault slip, rise time, and tsunami amplitude, they fit the 2011 Tohoku and 2004 Sumatra tsunamis into this paradigm. A full discussion of these laboratory experiments is beyond the scope of this paper, but we note certain flaws in the scaling relations of the experiments. First, tsunami amplitudes in the open ocean are much smaller than ocean depths (\(|\eta | / H \ll 1\)), but in these experiments \(|\eta |/H \approx 0.1 - 0.7\). Nonlinearity is therefore essential in the experiments, while negligible in the real ocean. Second, water compressibility effects are vastly less important in the experiments than in the real ocean. Two dimensionless parameters quantifying compressibility effects are \(gH/c^2\) and \(H/c \tau\), where *c* is the sound speed in the ocean and \(\tau\) is again the rise time or source duration. These dimensionless parameters, and therefore compressibility effects, are several orders of magnitude more important in real tsunamis than in the experiments. These differences in scaling suggest that waves in these laboratory experiments are generated and propagate in a different manner from real tsunamis.

What accounts for the stark difference between the results in this paper and in the studies of Song and collaborators? In our full-physics simulations, we observe large amplitude acoustic waves in the ocean, likely due in part to the transfer of horizontal momentum from the solid Earth to the ocean during earthquake rupture. Our full-physics model, which includes a compressible ocean, captures these acoustic waves that a model using an incompressible ocean would miss. In an incompressible ocean with a finite domain, horizontal momentum transferred from the solid Earth to the ocean would be left in the vicinity of the source region rather than being carried away by acoustic waves. However, in a full-physics, compressible ocean model most of the kinetic energy injected into the ocean by the seafloor displacement is carried away by acoustic waves and does not significantly contribute to tsunami height. Overall, we conclude that the standard practice of setting initial horizontal velocity to zero remains the best, and most consistent, procedure for tsunami modeling.

## Declarations

### Authors’ contributions

Lotto and Dunham identified the research question and developed an initial plan to answer it. Nava ran and compared the numerical simulations presented in section 3. Lotto ran the simulations in sections 4 and 5, analyzed the results, and wrote the majority of the manuscript. Dunham edited and wrote some sections of the manuscript, and supervised the entire project. All authors read and approved the final manuscript.

### Acknowlegements

We thank the two anonymous reviewers for constructive comments and Tatsuhiko Saito for insightful discussions regarding tsunami generation.

### Competing interests

The authors declare that they have no competing interests.

### Availability of data and materials

All modeling results, along with MATLAB scripts to read them, will be shared freely upon request.

### Consent for publication

Not applicable.

### Ethics approval and consent to participate

Not applicable.

### Funding

This work was supported by the National Science Foundation (EAR-1255439).

### Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Open Access**This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

## Authors’ Affiliations

## References

- Behrens J, Androsov A, Babeyko AY, Harig S, Klaschka F, Mentrup L (2010) A new multi-sensor approach to simulation assisted tsunami early warning. Nat Hazards Earth Syst Sci 10(6):1085–1100View ArticleGoogle Scholar
- Bernard EN, Mofjeld HO, Titov V, Synolakis CE, González FI (2006) Tsunami: scientific frontiers, mitigation, forecasting and policy implications. Philos Trans R Soc Lond A Math Phys Eng Sci 364(1845):1989–2007View ArticleGoogle Scholar
- Bletery Q, Sladen A, Delouis B, Vallée M, Nocquet J-M, Rolland L, Jiang J (2014) A detailed source model for the Mw9. 0 Tohoku-Oki earthquake reconciling geodesy, seismology, and tsunami records. J Geophys Res Solid Earth 119(10):7636–7653View ArticleGoogle Scholar
- Borrero JC, Sieh K, Chlieh M, Synolakis CE (2006) Tsunami inundation modeling for western Sumatra. Proc Natl Acad Sci 103(52):19673–19677View ArticleGoogle Scholar
- Cocco, M, Rice, JR (2002) Pore pressure and poroelasticity effects in Coulomb stress analysis of earthquake interactions. J Geophys Res Solid Earth 107(B2):ESE 2-1–ESE 2-17View ArticleGoogle Scholar
- Fujii Y, Satake K, Sakai S, Shinohara M, Kanazawa T (2011) Tsunami source of the 2011 off the Pacific coast of Tohoku Earthquake. Earth Planets Space 63(7):55. doi:10.5047/eps.2011.06.010 View ArticleGoogle Scholar
- Geist EL, Parsons T (2006) Probabilistic analysis of tsunami hazards. Nat Hazards 37(3):277–314View ArticleGoogle Scholar
- González FI, Geist EL, Jaffe B, Kânoğlu U, Mofjeld H, Synolakis CE, Titov VV, Arcas D, Bellomo D, Carlton D, et al (2009) Probabilistic tsunami hazard assessment at Seaside, Oregon, for near-and far-field seismic sources. J Geophys Res Oceans 114:C11023View ArticleGoogle Scholar
- Gusman AR, Tanioka Y, Sakai S, Tsushima H (2012) Source model of the great 2011 Tohoku earthquake estimated from tsunami waveforms and crustal deformation data. Earth Planet Sci Lett 341:234–242View ArticleGoogle Scholar
- Gutscher M-A, Klaeschen D, Flueh E, Malavieille J (2001) Non-coulomb wedges, wrong-way thrusting, and natural hazards in Cascadia. Geology 29(5):379–382View ArticleGoogle Scholar
- Hooper A, Pietrzak J, Simons W, Cui H, Riva R, Naeije M, Terwisscha A, van Scheltinga E, Schrama G Stelling, Socquet A (2013) Importance of horizontal seafloor motion on tsunami height for the 2011 Mw = 9.0 Tohoku-Oki earthquake. Earth Planet Sci Lett 361:469–479View ArticleGoogle Scholar
- Kajiura K (1963) The leading wave of a tsunami. Bull Earthq Res Inst 43:535–571Google Scholar
- Kajiura K (1970) Tsunami source, energy and the directivity of wave radiation. Bull Earthq Res Inst Univ Tokyo 48:835–869Google Scholar
- Kopp H, Kukowski N (2003) Backstop geometry and accretionary mechanics of the Sunda margin. Tectonics 22(6):1072View ArticleGoogle Scholar
- Kowalik Z (2003) Basic relations between tsunamis calculation and their physics-ii. Sci Tsunami Hazards 21(3):154–173Google Scholar
- Kozdon JE, Dunham EM (2013) Rupture to the trench: dynamic rupture simulations of the 11 March 2011 Tohoku earthquake. Bull Seismolog Soc Am 103(2B):1275–1289View ArticleGoogle Scholar
- Kozdon JE, Dunham EM (2014) Constraining shallow slip and tsunami excitation in megathrust ruptures using seismic and ocean acoustic waves recorded on ocean-bottom sensor networks. Earth Planet Sci Lett 396:56–65View ArticleGoogle Scholar
- Lotto GC, Dunham EM (2015) High-order finite difference modeling of tsunami generation in a compressible ocean from offshore earthquakes. Comput Geosci 19:327–340View ArticleGoogle Scholar
- Lotto GC, Dunham EM, Jeppson TN, Tobin HJ (2017) The effect of compliant prisms on subduction zone earthquakes and tsunamis. Earth Planet Sci Lett 458:213–222View ArticleGoogle Scholar
- Maeda T, Furumura T (2013) FDM simulation of seismic waves, ocean acoustic waves, and tsunamis based on tsunami-coupled equations of motion. Pure Appl Geophys 170(1–2):109–127View ArticleGoogle Scholar
- Melgar D, Bock Y (2013) Near-field tsunami models with rapid earthquake source inversions from land-and ocean-based observations: the potential for forecast and warning. J Geophys Res Solid Earth 118(11):5939–5955View ArticleGoogle Scholar
- Mori N, Takahashi T, Yasuda T, Yanagisawa H (2011) Survey of 2011 Tohoku earthquake tsunami inundation and run-up. Geophys Res Lett 38(7):L00G14View ArticleGoogle Scholar
- Saito T (2013) Dynamic tsunami generation due to sea-bottom deformation: analytical representation based on linear potential theory. Earth Planets Space 65(12):1411–1423. doi:10.5047/eps.2013.07.004 View ArticleGoogle Scholar
- Saito T, Furumura T (2009) Three-dimensional tsunami generation simulation due to sea-bottom deformation and its interpretation based on the linear theory. Geophys J Int 178(2):877–888View ArticleGoogle Scholar
- Satake K (2002) Tsunamis. Int Geophys 81:437–451View ArticleGoogle Scholar
- Satake K, Fujii Y, Harada T, Namegaya Y (2013) Time and space distribution of coseismic slip of the 2011 Tohoku earthquake as inferred from tsunami waveform data. Bull Seismol Soc Am 103(2B):1473–1492View ArticleGoogle Scholar
- Sato M, Ishikawa T, Ujihara N, Yoshida S, Fujita M, Mochizuki M, Asada A (2011) Displacement above the hypocenter of the 2011 Tohoku-Oki earthquake. Science 332(6036):1395–1395View ArticleGoogle Scholar
- Song YT, Fu L-L, Zlotnicki V, Ji C, Hjorleifsdottir V, Shum CK, Yi Y (2008) The role of horizontal impulses of the faulting continental slope in generating the 26 December 2004 tsunami. Ocean Model 20(4):362–379View ArticleGoogle Scholar
- Song YT, Han S-C (2011) Satellite observations defying the long-held tsunami genesis theory. In: Tang DL (ed) Remote sensing of the changing Oceans, Springer, pp 327–342Google Scholar
- Song YT, Mohtat A, Yim SC (2017) New insights on tsunami genesis and energy source. J Geophys Res Oceans 122(5):4238–4256View ArticleGoogle Scholar
- Tanioka Y, Satake K (1996) Tsunami generation by horizontal displacement of ocean bottom. Geophys Res Lett 23(8):861–864View ArticleGoogle Scholar
- Titov V, Song YT, Tang L, Bernard EN, Bar-Sever Y, Wei Y (2016) Consistent estimates of tsunami energy show promise for improved early warning. Pure Appl Geophys 173(12):3863–3880View ArticleGoogle Scholar
- Titov VV, Gonzalez FI (1997) Implementation and testing of the method of splitting tsunami (MOST) model. Technical Report NOAA Tech. Memo. ERL PMEL-112 (PB98-122773), NOAA/Pacific Marine Environmental Laboratory, Seattle, WAGoogle Scholar
- Titov VV, Gonzalez FI, Bernard EN, Eble MC, Mofjeld HO, Newman JC, Venturato AJ (2005) Real-time tsunami forecasting: challenges and solutions. In: Bernard EN (ed) Developing tsunami-resilient communities, Springer, pp 41–58Google Scholar
- Titov VV, Synolakis CE (1998) Numerical modeling of tidal wave runup. J Waterw Port Coast Ocean Eng 124(4):157–171View ArticleGoogle Scholar
- Watts P, Grilli ST, Kirby JT, Fryer GJ, Tappin DR (2003) Landslide tsunami case studies using a Boussinesq model and a fully nonlinear tsunami generation model. Nat Hazards Earth Syst Sci 3(5):391–402View ArticleGoogle Scholar
- Xu Z, Song YT (2013) Combining the all-source Green’s functions and the GPS-derived source functions for fast tsunami predictions–illustrated by the March 2011 Japan tsunami. J Atmos Ocean Technol 30(7):1542–1554View ArticleGoogle Scholar