해저 산사태 쓰나미의 최대 초기 파동 진폭 추정: 3차원 모델링 접근법

Ramtin Sabeti ^a, Mohammad Heidarzadeh ^ab

^aDepartment of Architecture and Civil Engineering, University of Bath, Bath BA27AY, UK
^bHydroCoast Consulting Engineers Ltd, Bath, UK

https://doi.org/10.1016/j.ocemod.2024.102360

Highlights

• •Landslide travel distance is considered for the first time in a predictive equation.
• •Predictive equation derived from databases using 3D physical and numerical modeling.
• •The equation was successfully tested on the 2018 Anak Krakatau tsunami event.
• •The developed equation using three-dimensional data exhibits a 91 % fitting quality.

Abstract

Landslide tsunamis, responsible for thousands of deaths and significant damage in recent years, necessitate the allocation of sufficient time and resources for studying these extreme natural hazards. This study offers a step change in the field by conducting a large number of three-dimensional numerical experiments, validated by physical tests, to develop a predictive equation for the maximum initial amplitude of tsunamis generated by subaerial landslides. We first conducted a few 3D physical experiments in a wave basin which were then applied for the validation of a 3D numerical model based on the Flow3D-HYDRO package. Consequently, we delivered 100 simulations using the validated model by varying parameters such as landslide volume, water depth, slope angle and travel distance. This large database was subsequently employed to develop a predictive equation for the maximum initial tsunami amplitude. For the first time, we considered travel distance as an independent parameter for developing the predictive equation, which can significantly improve the predication accuracy. The predictive equation was tested for the case of the 2018 Anak Krakatau subaerial landslide tsunami and produced satisfactory results.

Keywords

Tsunami, Subaerial landslide, Physical modelling, Numerical simulation, FLOW-3D HYDRO

1. Introduction and literature review

The Anak Krakatau landslide tsunami on 22nd December 2018 was a stark reminder of the dangers posed by subaerial landslide tsunamis (Ren et al., 2020; Mulia et al. 2020a; Borrero et al., 2020; Heidarzadeh et al., 2020; Grilli et al., 2021). The collapse of the volcano’s southwest side into the ocean triggered a tsunami that struck the Sunda Strait, leading to approximately 450 fatalities (Syamsidik et al., 2020; Mulia et al., 2020b) (Fig. 1). As shown in Fig. 1, landslide tsunamis (both submarine and subaerial) have been responsible for thousands of deaths and significant damage to coastal communities worldwide. These incidents underscored the critical need for advanced research into landslide-generated waves to aid in hazard prediction and mitigation. This is further emphasized by recent events such as the 28th of November 2020 landslide tsunami in the southern coast mountains of British Columbia (Canada), where an 18 million m³ rockslide generated a massive tsunami, with over 100 m wave run-up, causing significant environmental and infrastructural damage (Geertsema et al., 2022).

Physical modelling and numerical simulation are crucial tools in the study of landslide-induced waves due to their ability to replicate and analyse the complex dynamics of landslide events (Kim et al., 2020). In two-dimensional (2D) modelling, the discrepancy between dimensions can lead to an artificial overestimation of wave amplification (e.g., Heller and Spinneken, 2015). This limitation is overcome with 3D modelling, which enables the scaled-down representation of landslide-generated waves while avoiding the simplifications inherent in 2D approaches (Erosi et al., 2019). Another advantage of 3D modelling in studying landslide-generated waves is its ability to accurately depict the complex dynamics of wave propagation, including lateral and radial spreading from the slide impact zone, a feature unattainable with 2D models (Heller and Spinneken, 2015).

Physical experiments in tsunami research, as presented by authors such as Romano et al. (2020), McFall and Fritz (2016), and Heller and Spinneken (2015), have supported 3D modelling works through validation and calibration of the numerical models to capture the complexities of wave generation and propagation. Numerical modelling has increasingly complemented experimental approach in tsunami research due to the latter’s time and resource-intensive nature, particularly for 3D models (Li et al., 2019; Kim et al., 2021). Various numerical approaches have been employed, from Eulerian and Lagrangian frameworks to depth-averaged and Navier–Stokes models, enhancing our understanding of tsunami dynamics (Si et al., 2018; Grilli et al., 2019; Heidarzadeh et al., 2017, 2020; Iorio et al., 2021; Zhang et al., 2021; Kirby et al., 2022; Wang et al., 2021, 2022; Hu et al., 2022). The sophisticated numerical techniques, including the Particle Finite Element Method and the Immersed Boundary Method, have also shown promising results in modelling highly dynamic landslide scenarios (Mulligan et al., 2020; Chen et al., 2020). Among these methods and techniques, FLOW-3D HYDRO stands out in simulating landslide-generated tsunami waves due to its sophisticated technical features such as offering Tru Volume of Fluid (VOF) method for precise free surface tracking (e.g., Sabeti and Heidarzadeh 2022a). TruVOF distinguishes itself through a split Lagrangian approach, adeptly reducing cumulative volume errors in wave simulations by dynamically updating cell volume fractions and areas with each time step. Its intelligent adaptation of time step size ensures precise capture of evolving free surfaces, offering unparalleled accuracy in modelling complex fluid interfaces and behaviour (Flow Science, 2023).

Predictive equations play a crucial role in assessing the potential hazards associated with landslide-generated tsunami waves due to their ability to provide risk assessment and warnings. These equations can offer swift and reasonable evaluations of potential tsunami impacts in the absence of detailed numerical simulations, which can be time-consuming and expensive to produce. Among multiple factors and parameters within a landslide tsunami generation, the initial maximum wave amplitude (Fig. 1) stands out due to its critical role. While it is most likely that the initial wave generated by a landslide will have the highest amplitude, it is crucial to clarify that the term “initial maximum wave amplitude” refers to the highest amplitude within the first set of impulse waves. This parameter is essential in determining the tsunami’s impact severity, with higher amplitudes signalling a greater destructive potential (Sabeti and Heidarzadeh 2022a). Additionally, it plays a significant role in tsunami modelling, aiding in the prediction of wave propagation and the assessment of potential impacts.

In this study, we initially validate the FLOW-3D HYDRO model through a series of physical experiments conducted in a 3D wave tank at University of Bath (UK). Upon confirmation of the model’s accuracy, we use it to systematically vary parameters namely landslide volume, water depth, slope angle, and travel distance, creating an extensive database. Alongside this, we perform a sensitivity analysis on these variables to discern their impacts on the initial maximum wave amplitude. The generated database was consequently applied to derive a non-dimensional predictive equation aimed at estimating the initial maximum wave amplitude in real-world landslide tsunami events.

Two innovations of this study are: (i) The predictive equation of this study is based on a large number of 3D experiments whereas most of the previous equations were based on 2D results, and (ii) For the first time, the travel distance is included in the predictive equation as an independent parameter. To evaluate the performance of our predictive equation, we applied it to a previous real-world subaerial landslide tsunami, i.e., the Anak Krakatau 2018 event. Furthermore, we compare the performance of our predictive equation with other existing equations.

2. Data and methods

The methodology applied in this research is a combination of physical and numerical modelling. Limited physical modelling was performed in a 3D wave basin at the University of Bath (UK) to provide data for calibration and validation of the numerical model. After calibration and validation, the numerical model was employed to model a large number of landslide tsunami scenarios which allowed us to develop a database for deriving a predictive equation.

2.1. Physical experiments

To validate our numerical model, we conducted a series of physical experiments including two sets in a 3D wave basin at University of Bath, measuring 2.50 m in length (W_L), 2.60 m in width (W_W), and 0.60 m in height (W_H) (Fig. 2a). Conducting two distinct sets of experiments (Table 1), each with different setups (travel distance, location, and water depth), provided a robust framework for validation of the numerical model. For wave measurement, we employed a twin wire wave gauge from HR Wallingford (https://equipit.hrwallingford.com). In these experiments, we used a concrete prism solid block, the dimensions of which are outlined in Table 2. In our experiments, we employed a concrete prism solid block with a density of 2600 kg/m³, chosen for its similarity to the natural density of landslides, akin to those observed with the 2018 Anak Krakatau tsunami, where the landslide composition is predominantly solid rather than granular. The block’s form has also been endorsed in prior studies (Watts, 1998; Najafi-Jilani and Ataie-Ashtiani, 2008) as a suitable surrogate for modelling landslide-induced waves. A key aspect of our methodology was addressing scale effects, following the guidelines proposed by Heller et al. (2008) as it is described in Table 1. To enhance the reliability and accuracy of our experimental data, we conducted each physical experiment three times which revealed all three experimental waveforms were identical. This repetition was aimed at minimizing potential errors and inconsistencies in laboratory measurements.

Table 1. The locations and other information of the laboratory setups for making landslide-generated waves in the physical wave basin. This table details the specific parameters for each setup, including slope range (α), slide volume (V), kinematic viscosity (ν), water depth (h), travel distance (D), surface tension coefficient of water (σ), Reynolds number (R), Weber number (W), and the precise coordinates of the wave gauges (WG).

⁎

The acceptable ranges for avoiding scale effects are based on the study by Heller et al. (2008).⁎⁎

The Reynolds number (R) is given by g^0.5h^1.5/ν, with ν denoting the kinematic viscosity. The Weber number (W) is W = ρgh²/σ, where σ represents surface tension coefficient and ρ = 1000kg/m³ is the density of water. In our experiments, conducted at a water temperature of approximately 20 °C, the kinematic viscosity (ν) and the surface tension coefficient of water (σ) are 1.01 × 10⁻⁶ m²/s and 0.073 N/m, respectively (Kestin et al., 1978).

Table 2. Specifications of the solid block used in physical experiments for generating subaerial landslides in the laboratory.

Solid-block attributes	Property metrics	Geometric shape
Slide width (bs)	0.26 m
Slide length (ls)	0.20 m
Slide thickness (s)	0.10 m
Slide volume (V)	2.60 × 10−3 m3
Specific gravity, (γs)	2.60
Slide weight (ms)	6.86 kg

2.2. Numerical simulations applying FLOW-3D hydro

The detailed theoretical framework encompassing the governing equations, the computational methodologies employed, and the specific techniques used for tracking the water surface in these simulations are thoroughly detailed in the study by Sabeti et al. (2024). Here, we briefly explain some of the numerical details. We defined a uniform mesh for our flow domain, carefully crafted with a fine spatial resolution of 0.005 m (i.e., grid size). The dimensions of the numerical model directly matched those of our wave basin used in the physical experiment, being 2.60 m wide, 0.60 m deep, and 2.50 m long (Fig. 2). This design ensures comprehensive coverage of the study area. The output intervals of the numerical model are set at 0.02 s. This timing is consistent with the sampling rates of wave gauges used in laboratory settings. The friction coefficient in the FLOW-3D HYDRO is designated as 0.45. This value corresponds to the Coulombic friction measurements obtained in the laboratory, ensuring that the simulation accurately reflects real-world physical interactions.

In order to simulate the landslide motion, we applied coupled motion objects in FLOW-3D-HYDRO where the dynamics are predominantly driven by gravity and surface friction. This methodology stands in contrast to other models that necessitate explicit inputs of force and torque. This approach ensures that the simulation more accurately reflects the natural movement of landslides, which is heavily reliant on gravitational force and the interaction between sliding surfaces. The stability of the numerical simulations is governed by the Courant Number criterion (Courant et al., 1928), which dictates the maximum time step (Δt) for a given mesh size (Δx) and flow speed (U). According to Courant et al. (1928), this number is required to stay below one to ensure stability of numerical simulations. In our simulations, the Courant number is always maintained below one.

In alignment with the parameters of physical experiments, we set the fluid within the mesh to water, characterized by a density of 1000 kg/m³ at a temperature of 20 °C. Furthermore, we defined the top, front, and back surfaces of the mesh as symmetry planes. The remaining surfaces are designated as wall types, incorporating no-slip conditions to accurately simulate the interaction between the fluid and the boundaries. In terms of selection of an appropriate turbulence model, we selected the k–ω model that showed a better performance than other turbulence methods (e.g., Renormalization-Group) in a previous study (Sabeti et al., 2024). The simulations are conducted using a PC Intel® Core™ i7-10510U CPU with a frequency of 1.80 GHz, and a 16 GB RAM. On this PC, completion of a 3-s simulation required approximately 12.5 h.

2.3. Validation

The FLOW-3D HYDRO numerical model was validated using the two physical experiments (Fig. 3) outlined in Table 1. The level of agreement between observations (O_i) and simulations (S_i) is examined using the following equation:(1)�=|��−����|×100where ε represents the mismatch error, O_i denotes the observed laboratory values, and S_i represents the simulated values from the FLOW-3D HYDRO model. The results of this validation process revealed that our model could replicate the waves generated in the physical experiments with a reasonable degree of mismatch (ε): 14 % for Lab 1 and 8 % for Lab 2 experiments, respectively (Fig. 3). These values indicate that while the model is not perfect, it provides a sufficiently close approximation of the real-world phenomena.

In terms of mesh efficiency, we varied the mesh size to study sensitivity of the numerical results to mesh size. First, by halving the mesh size and then by doubling it, we repeated the modelling by keeping other parameters unchanged. This analysis guided that a mesh size of ∆x = 0.005 m is the most effective for the setup of this study. The total number of computational cells applying mesh size of 0.005 m is 9.269 × 10⁶.

2.4. The dataset

The validated numerical model was employed to conduct 100 simulations, incorporating variations in four key landslide parameters namely water depth, slope angle, slide volume, and travel distance. This methodical approach was essential for a thorough sensitivity analysis of these variables, and for the creation of a detailed database to develop a predictive equation for maximum initial tsunami amplitude. Within the model, 15 distinct slide volumes were established, ranging from 0.10 × 10⁻³ m³ to 6.25 × 10⁻³ m³ (Table 3). The slope angle varied between 35° and 55°, and water depth ranged from 0.24 m to 0.27 m. The travel distance of the landslides was varied, spanning from 0.04 m to 0.07 m. Detailed configurations of each simulation, along with the maximum initial wave amplitudes and dominant wave periods are provided in Table 4.

Table 3. Geometrical information of the 15 solid blocks used in numerical modelling for generating landslide tsunamis. Parameters are: l_s, slide length; b_s, slide width; s, slide thickness; γ_s, specific gravity; and V, slide volume.

Table 4. The numerical simulation for the 100 tests performed in this study for subaerial solid-block landslide-generated waves. Parameters are a_M, maximum wave amplitude; α, slope angle; h, water depth; D, travel distance; and T, dominant wave period. The location of the wave gauge is X=1.030 m, Y=1.210 m, and Z=0.050 m. The properties of various solid blocks are presented in Table 3.

In all these 100 simulations, the wave gauge was consistently positioned at coordinates X=1.09 m, Y=1.21 m, and Z=0.05 m. The dominant wave period for each simulation was determined using the Fast Fourier Transform (FFT) function in MATLAB (MathWorks, 2023). Furthermore, the classification of wave types was carried out using a wave categorization graph according to Sorensen (2010), as shown in Fig. 4a. The results indicate that the majority of the simulated waves are on the border between intermediate and deep-water waves, and they are categorized as Stokes waves (Fig. 4a). Four sample waveforms from our 100 numerical experiments are provided in Fig. 4b.

The dataset in Table 4 was used to derive a new predictive equation that incorporates travel distance for the first time to estimate the initial maximum tsunami amplitude. In developing this equation, a genetic algorithm optimization technique was implemented using MATLAB (MathWorks 2023). This advanced approach entailed the use of genetic algorithms (GAs), an evolutionary algorithm type inspired by natural selection processes (MathWorks, 2023). This technique is iterative, involving selection, crossover, and mutation processes to evolve solutions over several generations. The goal was to identify the optimal coefficients and powers for each landslide parameter in the predictive equation, ensuring a robust and reliable model for estimating maximum wave amplitudes. Genetic Algorithms excel at optimizing complex models by navigating through extensive combinations of coefficients and exponents. GAs effectively identify highly suitable solutions for the non-linear and complex relationships between inputs (e.g., slide volume, slope angle, travel distance, water depth) and the output (i.e., maximum initial wave amplitude, a_M). MATLAB’s computational environment enhances this process, providing robust tools for GA to adapt and evolve solutions iteratively, ensuring the precision of the predictive model (Onnen et al., 1997). This approach leverages MATLAB’s capabilities to fine-tune parameters dynamically, achieving an optimal equation that accurately estimates a_M. It is important to highlight that the nondimensionalized version of this dataset is employed to develop a predictive equation which enables the equation to reproduce the maximum initial wave amplitude (a_M) for various subaerial landslide cases, independent of their dimensional differences (e.g., Heler and Hager 2014; Heller and Spinneken 2015; Sabeti and Heidarzadeh 2022b). For this nondimensionalization, we employed the water depth (h) to nondimensionalize the slide volume (V/h³) and travel distance (D/h). The slide thickness (s) was applied to nondimensionalize the water depth (h/s).

2.5. Landslide velocity

In discussing the critical role of landslide velocity for simulating landslide-generated waves, we focus on the mechanisms of landslide motion and the techniques used to record landslide velocity in our simulations (Fig. 5). Also, we examine how these methods were applied in two distinct scenarios: Lab 1 and Lab 2 (see Table 1 for their details). Regarding the process of landslide movement, a slide starts from a stationary state, gaining momentum under the influence of gravity and this acceleration continues until the landslide collides with water, leading to a significant reduction in its speed before eventually coming to a stop (Fig. 5) (e.g., Panizzo et al. 2005).

To measure the landslide’s velocity in our simulations, we attached a probe at the centre of the slide, which supplied a time series of the velocity data. The slide’s velocity (v_s) peaks at the moment it enters the water (Fig. 5), a point referred to as the impact time (t_Imp). Following this initial impact, the slides continue their underwater movement, eventually coming to a complete halt (t_Stop). Given the results in Fig. 5, it can be seen that Lab 1, with its longer travel distance (0.070 m), exhibits a higher peak velocity of 1.89 m/s. This increase in velocity is attributed to the extended travel distance allowing more time for the slide to accelerate under gravity. Whereas Lab 2, featuring a shorter travel distance (0.045 m), records a lower peak velocity of 1.78 m/s. This difference underscores how travel distance significantly influences the dynamics of landslide motion. After reaching the peak, both profiles show a sharp decrease in velocity, marking the transition to submarine motion until the slides come to a complete stop (t_Stop). There are noticeable differences observable in Fig. 5 between the Lab-1 and Lab-2 simulations, including the peaks at 0.3 s . These variations might stem from the placement of the wave gauge, which differs slightly in each scenario, as well as the water depth’s minor discrepancies and, the travel distance.

2.6. Effect of air entrainment

In this section we examine whether it is required to consider air entrainment for our modelling or not as the FLOW-3D HYDRO package is capable of modelling air entrainment. The process of air entrainment in water during a landslide tsunami and its subsequent transport involve two key components: the quantification of air entrainment at the water surface, and the simulation of the air’s transport within the fluid (Hirt, 2003). FLOW-3D HYDRO employs the air entrainment model to compute the volume of air entrained at the water’s surface utilizing three approaches: a constant density model, a variable density model accounting for bulking, and a buoyancy model that adds the Drift-FLUX mechanism to variable density conditions (Flow Science, 2023). The calculation of the entrainment rate is based on the following equation:(2)�������=������[2(��−�����−2�/���)]1/2where parameters are: V_air, volume of air; C_air, entrainment rate coefficient; A_s, surface area of fluid; ρ, fluid density; k, turbulent kinetic energy; g_n, gravity normal to surface; L_t, turbulent length scale; and σ, surface tension coefficient. The value of k is directly computed from the Reynolds-averaged Navier-Stokes (RANS) (k–w) calculations in our model.

In this study, we selected the variable density + Drift-FLUX model, which effectively captures the dynamics of phase separation and automatically activates the constant density and variable density models. This method simplifies the air-water mixture, treating it as a single, homogeneous fluid within each computational cell. For the phase volume fractions f₁and f₂​, the velocities are expressed in terms of the mixture and relative velocities, denoted as u and u_r, respectively, as follows:(3)��1��+�.(�1�)=��1��+�.(�1�)−�.(�1�2��)=0(4)��2��+�.(�2�)=��2��+�.(�2�)−�.(�1�2��)=0

The outcomes from this simulation are displayed in Fig. 6, which indicates that the influence of air entrainment on the generated wave amplitude is approximately 2 %. A value of 0.02 for the entrained air volume fraction means that, in the simulated fluid, approximately 2 % of the volume is composed of entrained air. In other words, for every unit volume of the fluid-air mixture at that location, 2 % is air and the remaining 98 % is water. The configuration of Test-17 (Table 4) was employed for this simulation. While the effect of air entrainment is anticipated to be more significant in models of granular landslide-generated waves (Fritz, 2002), in our simulations we opted not to incorporate this module due to its negligible impact on the results.

3. Results

In this section, we begin by presenting a sequence of our 3D simulations capturing different time steps to illustrate the generation process of landslide-generated waves. Subsequently, we derive a new predictive equation to estimate the maximum initial wave amplitude of landslide-generated waves and assess its performance.

3.1. Wave generation and propagation

To demonstrate the wave generation process in our simulation, we reference Test-17 from Table 4, where we employed Block-7 (Tables 3, 4). In this configuration, the slope angle was set to 45°, with a water depth of 0.246 m and a travel distance at 0.045 m (Fig. 7). At 0.220 s, the initial impact of the moving slide on the water is depicted, marking the onset of the wave generation process (Fig. 7a). Disturbances are localized to the immediate area of impact, with the rest of the water surface remaining undisturbed. At this time, a maximum water particle velocity of 1.0 m/s – 1.2 m/s is seen around the impact zone (Fig. 7d). Moving to 0.320 s, the development of the wave becomes apparent as energy transfer from the landslide to the water creates outwardly radiating waves with maximum water particle velocity of up to around 1.6 m/s – 1.8 m/s (Fig. 7b, e). By the time 0.670 s, the wave has fully developed and is propagating away from the impact point exhibiting maximum water particle velocity of up to 2.0 m/s – 2.1 m/s. Concentric wave fronts are visible, moving outwards in all directions, with a colour gradient signifying the highest wave amplitude near the point of landslide entry, diminishing with distance (Fig. 7c, f).

3.2. Influence of landslide parameters on tsunami amplitude

In this section, we investigate the effects of various landslide parameters namely slide volume (V), water depth (h), slipe angle (α) and travel distance (D) on the maximum initial wave amplitude (a_M). Fig. 8 presents the outcome of these analyses. According to Fig. 8, the slide volume, slope angle, and travel distance exhibit a direct relationship with the wave amplitude, meaning that as these parameters increase, so does the amplitude. Conversely, water depth is inversely related to the maximum initial wave amplitude, suggesting that the deeper the water depth, the smaller the maximum wave amplitude will be (Fig. 8b).

Fig. 8a highlights the pronounced impact of slide volume on the a_M, demonstrating a direct correlation between the two variables. For instance, in the range of slide volumes we modelled (Fig. 8a), The smallest slide volume tested, measuring 0.10 × 10⁻³ m³, generated a low initial wave amplitude (a_M= 0.0066 m) (Table 4). In contrast, the largest volume tested, 6.25 × 10⁻³ m³, resulted in a significantly higher initial wave amplitude (a_M= 0.0319 m) (Table 4). The extremities of these results emphasize the slide volume’s paramount impact on wave amplitude, further elucidated by their positions as the smallest and largest a_M values across all conducted tests (Table 4). This is corroborated by findings from the literature (e.g., Murty, 2003), which align with the observed trend in our simulations.

The slope angle’s influence on a_M was smooth. A steady increase of wave amplitude was observed as the slope angle increased (Fig. 8c). In examining travel distance, an anomaly was identified. At a travel distance of 0.047 m, there was an unexpected dip in a_M, which deviates from the general increasing trend associated with longer travel distances. This singular instance could potentially be attributed to a numerical error. Beyond this point, the expected pattern of increasing a_M with longer travel distances resumes, suggesting that the anomaly at 0.047 m is an outlier in an otherwise consistent trend, and thus this single data point was overlooked while deriving the predictive equation. Regarding the inverse relationship between water depth and wave amplitude, our result (Fig. 8b) is consistent with previous reports by Fritz et al. (2003), (2004), and Watts et al. (2005).

The insights from Fig. 8 informed the architecture of the predictive equation in the next Section, with slide volume, travel distance, and slope angle being multiplicatively linked to wave amplitude underscoring their direct correlations with wave amplitude. Conversely, water depth is incorporated as a divisor, representing its inverse relationship with wave amplitude. This structure encapsulates the dynamics between the landslide parameters and their influence on the maximum initial wave amplitude as discussed in more detail in the next Section.

3.3. Predictive equation

Building on our sensitivity analysis of landslide parameters, as detailed in Section 3.2, and utilizing our nondimensional dataset, we have derived a new predictive equation as follows:(5)��/ℎ=0.015(tan�)0.10(�ℎ3)0.90(�ℎ)0.10(ℎ�)−0.11where, V is sliding volume, h is water depth, α is slope angle, and s is landslide thickness. It is important to note that this equation is valid only for subaerial solid-block landslide tsunamis as all our experiments were for this type of waves. The performance of this equation in predicting simulation data is demonstrated by the satisfactory alignment of data points around a 45° line, indicating its accuracy and reliability with regard to the experimental dataset (Fig. 9). The quality of fit between the dataset and Eq. (5) is 91 % indicating that Eq. (5) represents the dataset very well. Table 5 presents Eq. (5) alongside four other similar equations previously published. Two significant distinctions between our Eq. (5) and these others are: (i) Eq. (5) is derived from 3D experiments, whereas the other four equations are based on 2D experiments. (ii) Unlike the other equations, our Eq. (5) incorporates travel distance as an independent parameter.

Table 5. Performance comparison among our newly-developed equation and existing equations for estimating the maximum initial amplitude (a_M) of the 2018 Anak Krakatau subaerial landslide tsunami. Parameters: a_M, initial maximum wave amplitude; h, water depth; v_s, landslide velocity; V, slide volume; b_s, slide width; l_s, slide length; s, slide thickness; α, slope angle; and ����, volume of the final immersed landslide. We considered ����= V as the slide volume.

⁎

Geometrical and kinematic parameters of the 2018 Anak Krakatau subaerial landslide based on Heidarzadeh et al. (2020), Grilli et al. (2019) and Grilli et al. (2021): V=2.11 × 10⁷ m³, h= 50 m; s= 114 m; α= 45°; l_s=1250 m; b_s= 2700 m; v_s=44.9 m/s; D= 2500 m; a_M= 100 m −150 m.⁎⁎

a_M= An average value of a_M = 134 m is considered in this study.⁎⁎⁎

The equation of Bolin et al. (2014) is based on the reformatted one reported by Lindstrøm (2016).⁎⁎⁎⁎

Error is calculated using Eq. (1), where the calculated a_M is assumed as the simulated value.

Additionally, we evaluated the performance of this equation using the real-world data from the 2018 Anak Krakatau subaerial landslide tsunami. Based on previous studies (Heidarzadeh et al., 2020; Grilli et al., 2019, 2021), we were able to provide a list of parameters for the subaerial landslide and associated tsunami for the 2018 Anak Krakatau event (see footnote of Table 5). We note that the data of the 2018 Anak Krakatau event was not used while deriving Eq. (5). The results indicate that Eq. (5) predicts the initial amplitude of the 2018 Anak Krakatau tsunami as being 130 m indicating an error of 2.9 % compared to the reported average amplitude of 134 m for this event. This performance indicates an improvement compared to the previous equation reported by Sabeti and Heidarzadeh (2022a) (Table 5). In contrast, the equations from Robbe-Saule et al. (2021) and Bolin et al. (2014) demonstrate higher discrepancies of 4200 % and 77 %, respectively (Table 5). Although Noda’s (1970) equation reproduces the tsunami amplitude of 134 m accurately (Table 5), it is crucial to consider its limitations, notably not accounting for parameters such as slope angle and travel distance.

It is essential to recognize that both travel distance and slope angle significantly affect wave amplitude. In our model, captured in Eq. (5), we integrate the slope angle (α) through the tangent function, i.e., tan α. This choice diverges from traditional physical interpretations that often employ the cosine or sine function (e.g., Heller and Hager, 2014; Watts et al., 2003). We opted for the tangent function because it more effectively reflects the direct impact of slope steepness on wave generation, yielding superior estimations compared to conventional methods.

The significance of this study lies in its application of both physical and numerical 3D experiments and the derivation of a predictive equation based on 3D results. Prior research, e.g. Heller et al. (2016), has reported notable discrepancies between 2D and 3D wave amplitudes, highlighting the important role of 3D experiments. It is worth noting that the suitability of applying an equation derived from either 2D or 3D data depends on the specific geometry and characteristics inherent in the problem being addressed. For instance, in the case of a long, narrow dam reservoir, an equation derived from 2D data would likely be more suitable. In such contexts, the primary dynamics of interest such as flow patterns and potential wave propagation are predominantly two-dimensional, occurring along the length and depth of the reservoir. This simplification to 2D for narrow dam reservoirs allows for more accurate modelling of these dynamics.

This study specifically investigates waves initiated by landslides, focusing on those characterized as solid blocks instead of granular flows, with slope angles confined to a range of 25° to 60°. We acknowledge the additional complexities encountered in real-world scenarios, such as dynamic density and velocity of landslides, which could affect the estimations. The developed equation in this study is specifically designed to predict the maximum initial amplitude of tsunamis for the aforementioned specified ranges and types of landslides.

4. Conclusions

Both physical and numerical experiments were undertaken in a 3D wave basin to study solid-block landslide-generated waves and to formulate a predictive equation for their maximum initial wave amplitude. At the beginning, two physical experiments were performed to validate and calibrate a 3D numerical model, which was subsequently utilized to generate 100 experiments by varying different landslide parameters. The generated database was then used to derive a predictive equation for the maximum initial wave amplitude of landslide tsunamis. The main features and outcomes are:

• •The predictive equation of this study is exclusively derived from 3D data and exhibits a fitting quality of 91 % when applied to the database.
• •For the first time, landslide travel distance was considered in the predictive equation. This inclusion provides more accuracy and flexibility for applying the equation.
• •To further evaluate the performance of the predictive equation, it was applied to a real-world subaerial landslide tsunami (i.e., the 2018 Anak Krakatau event) and delivered satisfactory performance.

CRediT authorship contribution statement

Ramtin Sabeti: Conceptualization, Methodology, Validation, Software, Visualization, Writing – review & editing. Mohammad Heidarzadeh: Methodology, Data curation, Software, Writing – review & editing.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Funding

RS is supported by the Leverhulme Trust Grant No. RPG-2022-306. MH is funded by open funding of State Key Lab of Hydraulics and Mountain River Engineering, Sichuan University, grant number SKHL2101. We acknowledge University of Bath Institutional Open Access Fund. MH is also funded by the Great Britain Sasakawa Foundation grant no. 6217 (awarded in 2023).

Acknowledgements

Authors are sincerely grateful to the laboratory technician team, particularly Mr William Bazeley, at the Faculty of Engineering, University of Bath for their support during the laboratory physical modelling of this research. We appreciate the valuable insights provided by Mr. Brian Fox (Senior CFD Engineer at Flow Science, Inc.) regarding air entrainment modelling in FLOW-3D HYDRO. We acknowledge University of Bath Institutional Open Access Fund.

Data availability

• All data used in this study are given in the body of the article.

References

1. Baptista et al., 2020M.A. Baptista, J.M. Miranda, R. Omira, I. El-HussainStudy of the 24 September 2013 Oman Sea tsunami using linear shallow water inversionArab. J. Geosci., 13 (14) (2020), p. 606View in ScopusGoogle Scholar
2. Bolin et al., 2014H. Bolin, Y. Yueping, C. Xiaoting, L. Guangning, W. Sichang, J. ZhibingExperimental modeling of tsunamis generated by subaerial landslides: two case studies of the Three Gorges Reservoir, ChinaEnviron. Earth Sci., 71 (2014), pp. 3813-3825View at publisher CrossRefView in ScopusGoogle Scholar
3. Borrero et al., 2020J.C. Borrero, T. Solihuddin, H.M. Fritz, P.J. Lynett, G.S. Prasetya, V. Skanavis, S. Husrin, Kushendratno, W. Kongko, D.C. Istiyanto, A. DaulatField survey and numerical modelling of the December 22, 2018, Anak Krakatau TsunamiPure Appl. Geophys, 177 (2020), pp. 2457-2475View at publisher CrossRefView in ScopusGoogle Scholar
4. Chen et al., 2020F. Chen, V. Heller, R. BrigantiNumerical modelling of tsunamis generated by iceberg calving validated with large-scale laboratory experimentsAdv. Water Resour., 142 (2020), Article 103647View PDFView articleView in ScopusGoogle Scholar
5. Courant et al., 1928R. Courant, K. Friedrichs, H. LewyÜber die partiellen Differenzengleichungen der mathematischen PhysikMath. Ann., 100 (1) (1928), pp. 32-74View in ScopusGoogle Scholar
6. Ersoy et al., 2019H. Ersoy, M. Karahan, K. Gelişli, A. Akgün, T. Anılan, M.O. Sünnetci, B.K. YahşiModelling of the landslide-induced impulse waves in the Artvin Dam reservoir by empirical approach and 3D numerical simulationEng. Geol., 249 (2019), pp. 112-128View PDFView articleView in ScopusGoogle Scholar
7. Flow Science 2023Flow Science. (2023). Flow-3D Hydro version 12.0 user’s manual. Santa Fe, NM, USA. Retrieved from https://www.flow3d.com/. 6 Aug 2023.Google Scholar
8. Fritz, 2002H.M. FritzDoctoral DissertationETH Zurich (2002)Google Scholar
9. Fritz et al., 2003H.M. Fritz, W.H. Hager, H.E. MinorLandslide generated impulse waves. 1. Instantaneous flow fieldsExp. Fluids, 35 (2003), pp. 505-519View in ScopusGoogle Scholar
10. Fritz et al., 2004H.M. Fritz, W.H. Hager, H.E. MinorNear field characteristics of landslide generated impulse wavesJ. Waterw. Port Coastal Ocean Eng., 130 (6) (2004), pp. 287-302View in ScopusGoogle Scholar
11. Geertsema et al., 2022M. Geertsema, B. Menounos, G. Bullard, J.L. Carrivick, J.J. Clague, C. Dai, D. Donati, G. Ekstrom, J.M. Jackson, P. Lynett, M. PichierriThe 28 Nov 2020 landslide, tsunami, and outburst flood – a hazard cascade associated with rapid deglaciation at Elliot Creek, BC, CanadaGeophys. Res. Lett., 49 (6) (2022)Google Scholar
12. Grilli et al., 2021S.T. Grilli, C. Zhang, J.T. Kirby, A.R. Grilli, D.R. Tappin, S.F.L. Watt, J.E. Hunt, A. Novellino, S. Engwell, M.E.M. Nurshal, M. AbdurrachmanModeling of the Dec. 22nd, 2018, Anak Krakatau volcano lateral collapse and tsunami based on recent field surveys: comparison with observed tsunami impactMar. Geol., 440 (2021), Article 106566View PDFView articleView in ScopusGoogle Scholar
13. Grilli et al., 2019S.T. Grilli, D.R. Tappin, S. Carey, S.F. Watt, S.N. Ward, A.R. Grilli, S.L. Engwell, C. Zhang, J.T. Kirby, L. Schambach, M. MuinModelling of the tsunami from the Dec. 22, 2018, lateral collapse of Anak Krakatau volcano in the Sunda Straits, IndonesiaSci. Rep., 9 (1) (2019), p. 11946 View at publisher This article is free to access.View in ScopusGoogle Scholar
14. Heidarzadeh et al., 2023M. Heidarzadeh, A.R. Gusman, I.E. MuliaThe landslide source of the eastern Mediterranean tsunami on 6 Feb 2023 following the Mw 7.8 Kahramanmaraş (Türkiye) inland earthquakeGeosci. Lett., 10 (1) (2023), p. 50 View at publisher This article is free to access.View in ScopusGoogle Scholar
15. Heidarzadeh et al., 2022M. Heidarzadeh, A.R. Gusman, A. Patria, B.T. WidyantoroPotential landslide origin of the Seram Island tsunami in Eastern Indonesia on 16 June 2021 following an Mw 5.9 earthquakeBull. Seismol. Soc. Am., 112 (5) (2022), pp. 2487-2498View at publisher CrossRefView in ScopusGoogle Scholar
16. Heidarzadeh et al., 2020M. Heidarzadeh, T. Ishibe, O. Sandanbata, A. Muhari, A.B. WijanartoNumerical modeling of the subaerial landslide source of the 22 Dec 2018 Anak Krakatoa volcanic tsunami, IndonesiaOcean. Eng., 195 (2020), Article 106733View PDFView articleView in ScopusGoogle Scholar
17. Heidarzadeh et al., 2017M. Heidarzadeh, T. Harada, K. Satake, T. Ishibe, T. TakagawaTsunamis from strike-slip earthquakes in the Wharton Basin, northeast Indian Ocean: March 2016 M w7. 8 event and its relationship with the April 2012 M w 8.6 eventGeophys. J. Int., 211 (3) (2017), pp. 1601-1612, 10.1093/gji/ggx395 View at publisher This article is free to access.View in ScopusGoogle Scholar
18. Heller et al., 2008V. Heller, W.H. Hager, H.E. MinorScale effects in subaerial landslide generated impulse wavesExp. Fluids, 44 (2008), pp. 691-703View at publisher CrossRefView in ScopusGoogle Scholar
19. Heller and Hager, 2014V. Heller, W.H. HagerA universal parameter to predict subaerial landslide tsunamis?J. Mar. Sci. Eng., 2 (2) (2014), pp. 400-412View at publisher CrossRefView in ScopusGoogle Scholar
20. Heller and Spinneken, 2015V. Heller, J. SpinnekenOn the effect of the water body geometry on landslide–tsunamis: physical insight from laboratory tests and 2D to 3D wave parameter transformationCoastal Eng., 104 (2015), pp. 113-134View PDFView articleView in ScopusGoogle Scholar
21. Heller et al., 2016V. Heller, M. Bruggemann, J. Spinneken, B.D. RogersComposite modelling of subaerial landslide–tsunamis in different water body geometries and novel insight into slide and wave kinematicsCoastal Eng., 109 (2016), pp. 20-41View PDFView articleView in ScopusGoogle Scholar
22. Hirt, 2003C.W. HirtModeling Turbulent Entrainment of Air at a Free SurfaceFlow Science, Inc (2003)Google Scholar
23. Hu et al., 2023G. Hu, K. Satake, L. Li, P. DuOrigins of the tsunami following the 2023 Turkey–Syria earthquakeGeophys. Res. Lett., 50 (18) (2023)Google Scholar
24. Hu et al., 2022G. Hu, W. Feng, Y. Wang, L. Li, X. He, Ç. Karakaş, Y. TianSource characteristics and exacerbated tsunami hazard of the 2020 Mw 6.9 Samos earthquake in Eastern Aegean SeaJ. Geophys. Res., 127 (5) (2022)e2022JB023961Google Scholar
25. Iorio et al., 2021V. Iorio, G. Bellotti, C. Cecioni, S.T. GrilliA numerical model for the efficient simulation of multiple landslide-induced tsunamis scenariosOcean Model, 168 (2021), Article 101899View PDFView articleView in ScopusGoogle Scholar
26. Kestin et al., 1978J. Kestin, M. Sokolov, W.A. WakehamViscosity of liquid water in the range− 8 C to 150 CJPCRD, 7 (3) (1978), pp. 941-948View at publisher CrossRefView in ScopusGoogle Scholar
27. Kim et al., 2020G.B. Kim, W. Cheng, R.C. Sunny, J.J. Horrillo, B.C. McFall, F. Mohammed, H.M. Fritz, J. Beget, Z. KowalikThree-dimensional landslide generated tsunamis: numerical and physical model comparisonsLandslides, 17 (2020), pp. 1145-1161View at publisher CrossRefView in ScopusGoogle Scholar
28. Kirby et al., 2022J.T. Kirby, S.T. Grilli, J. Horrillo, P.L.F. Liu, D. Nicolsky, S. Abadie, B. Ataie-Ashtiani, M.J. Castro, L. Clous, C. Escalante, I. Fine, J.M. González-Vida, F. Løvholt, P. Lynett, G. Ma, J. Macías, S. Ortega, F. Shi, S. Yavari-Ramshe, C. ZhangValidation and inter-comparison of models for landslide tsunami generationOcean Model., 170 (2022), Article 101943View PDFView articleView in ScopusGoogle Scholar
29. Li et al., 2019G. Li, G. Chen, P. Li, H. JingEfficient and accurate 3-D numerical modelling of landslide tsunamiWater, 11 (10) (2019), p. 2033View at publisher CrossRefView in ScopusGoogle Scholar
30. MathWorks 2023MathWorksMATLAB User ManualThe MathWorks Inc., MA, USA (2023), p. 282Google Scholar
31. McFall and Fritz, 2016B.C. McFall, H.M. FritzPhysical modelling of tsunamis generated by three-dimensional deformable granular landslides on planar and conical island slopesProc. R. Soc. A. Math. Phys. Eng. Sci., 472 (2188) (2016), Article 20160052View at publisher CrossRefGoogle Scholar
32. Mulia et al., 2020aI.E. Mulia, S. Watada, T.C. Ho, K. Satake, Y. Wang, A. AditiyaSimulation of the 2018 tsunami due to the flank failure of Anak Krakatau volcano and implication for future observing systemsGeophys. Res. Lett., 47 (14) (2020), Article e2020GL087334 View at publisher This article is free to access.View in ScopusGoogle Scholar
33. Murty, 2003T.S. MurtyTsunami wave height dependence on landslide volumePure Appl. Geophys., 160 (2003), pp. 2147-2153View in ScopusGoogle Scholar
34. Mulia et al., 2020bI.E. Mulia, S. Watada, T.C. Ho, K. Satake, Y. Wang, A. AditiyaSimulation of the 2018 tsunami due to the flank failure of Anak Krakatau volcano and implication for future observing systemsGeophys. Res. Lett., 47 (14) (2020)Google Scholar
35. Mulligan et al., 2020R.P. Mulligan, A. Franci, M.A. Celigueta, W.A. TakeSimulations of landslide wave generation and propagation using the particle finite element methodJ. Geophys. Res. Oceans, 125 (6) (2020)Google Scholar
36. Najafi-Jilani and Ataie-Ashtiani, 2008A. Najafi-Jilani, B. Ataie-AshtianiEstimation of near-field characteristics of tsunami generation by submarine landslideOcean Eng., 35 (5) (2008), pp. 545-557View PDFView articleView in ScopusGoogle Scholar
37. Onnen et al., 1997C. Onnen, R. Babuška, U. Kaymak, J.M. Sousa, H.B. Verbruggen, R. IsermannGenetic algorithms for optimization in predictive controlControl Eng. Pract, 5 (10) (1997), pp. 1363-1372View PDFView articleView in ScopusGoogle Scholar
38. Panizzo et al. 2005A. Panizzo, P. De Girolamo, M. Di Risio, A. Maistri, A. PetacciaGreat landslid event in Italian artificial reservoirsNat. Hazards Earth Syst. Sci., 5 (5) (2005), pp. 733-740View at publisher CrossRefView in ScopusGoogle Scholar
39. Ren et al., 2020Z. Ren, Y. Wang, P. Wang, J. Hou, Y. Gao, L. ZhaoNumerical study of the triggering mechanism of the 2018 Anak Krakatau tsunami: eruption or collapsed landslide?Nat. Hazards, 102 (2020), pp. 1-13View in ScopusGoogle Scholar
40. Robbe-Saule et al., 2021M. Robbe-Saule, C. Morize, Y. Bertho, A. Sauret, A. Hildenbrand, P. GondretFrom laboratory experiments to geophysical tsunamis generated by subaerial landslidesSci. Rep., 11 (1) (2021), pp. 1-9Google Scholar
41. Sabeti and Heidarzadeh, 2022aR. Sabeti, M. HeidarzadehNumerical simulations of water waves generated by subaerial granular and solid-block landslides: validation, comparison, and predictive equationsOcean Eng., 266 (2022), Article 112853View PDFView articleView in ScopusGoogle Scholar
42. Sabeti and Heidarzadeh, 2022bR. Sabeti, M. HeidarzadehA new empirical equation for predicting the maximum initial amplitude of submarine landslide-generated wavesLandslides, 19 (2) (2022), pp. 491-503 View at publisher This article is free to access.CrossRefView in ScopusGoogle Scholar
43. Sabeti et al. 2024R. Sabeti, M. Heidarzadeh, A. Romano, G. Barajas Ojeda, J.L. LaraThree-Dimensional Simulations of Subaerial Landslide-Generated Waves: Comparing OpenFOAM and FLOW-3D HYDRO ModelsPure Appl. Geophys. (2024), 10.1007/s00024-024-03443-x View at publisher This article is free to access.Google Scholar
44. Si et al., 2018P. Si, H. Shi, X. YuA general numerical model for surface waves generated by granular material intruding into a water bodyCoastal Eng., 142 (2018), pp. 42-51View PDFView articleView in ScopusGoogle Scholar
45. Sorensen, 2010R.M. SorensenBasic Coastal Engineering(3rd edition), Springer Science & Business Media (2010), p. 324Google Scholar
46. Syamsidik et al., 2020Benazir Syamsidik, M. Luthfi, A. Suppasri, L.K. ComfortThe 22 December 2018 Mount Anak Krakatau volcanogenic tsunami on Sunda Strait coasts, Indonesia: tsunami and damage characteristicsNat. Hazards Earth Syst. Sci., 20 (2) (2020), pp. 549-565View in ScopusGoogle Scholar
47. Synolakis et al., 2002C.E. Synolakis, J.P. Bardet, J.C. Borrero, H.L. Davies, E.A. Okal, E.A. Silver, D.R. TappinThe slump origin of the 1998 Papua New Guinea tsunamiProc. R. Soc. Lond. A Math. Phys. Eng. Sci., 45 (2002), pp. 763-789View in ScopusGoogle Scholar
48. Wang et al., 2022Y. Wang, H.Y. Su, Z. Ren, Y. MaSource properties and resonance characteristics of the tsunami generated by the 2021 M 8.2 Alaska earthquakeJ. Geophys. Res. Oceans, 127 (3) (2022), Article e2021JC018308 View at publisher This article is free to access.View in ScopusGoogle Scholar
49. Wang et al., 2021Y. Wang, M. Heidarzadeh, K. Satake, G. HuCharacteristics of two tsunamis generated by successive M w 7.4 and M w 8.1 earthquakes in Kermadec Islands on March 4, 2021Nat. Haz. Earth Sys. Sci., 22 (2021), pp. 1073-1082, 10.5194/nhess-22-1073-2022View at publisher View in ScopusGoogle Scholar
50. Watts et al., 2005P. Watts, S.T. Grilli, D.R. Tappin, G.J. FryerTsunami generation by submarine mass failure. II: predictive equations and case studiesJ. Waterw. Port Coast. Ocean Eng., 131 (6) (2005), pp. 298-310View in ScopusGoogle Scholar
51. Watts et al., 2003P. Watts, S.T. Grilli, J.T. Kirby, G.J. Fryer, D.R. TappinLandslide tsunami case studies using a Boussinesq model and a fully nonlinear tsunami generation modelNat. Hazards Earth Sys. Sci., 3 (5) (2003), pp. 391-402View at publisher CrossRefView in ScopusGoogle Scholar
52. Watts, 1998P. WattsWavemaker curves for tsunamis generated by underwater landslidesJ. Waterw. Port. Coast. Ocean. Eng., 124 (3) (1998), pp. 127-137Google Scholar
53. Zhang et al., 2021C. Zhang, J.T. Kirby, F. Shi, G. Ma, S.T. GrilliA two-layer non-hydrostatic landslide model for tsunami generation on irregular bathymetry. 2. Numerical discretization and model validationOcean Model., 160 (2021), Article 101769View PDFView articleView in ScopusGoogle Scholar