You are on page 1of 13

Engineering Geology 284 (2021) 105920

Contents lists available at ScienceDirect

Engineering Geology
journal homepage: www.elsevier.com/locate/enggeo

A computational design optimization method for rockfall


protection embankments
Hasuka Kanno a, *, Shuji Moriguchi b, Shunsuke Hayashi c, Kenjiro Terada b
a
Graduate School of Engineering, Tohoku University, Aza-Aoba 6-6-04, Aramaki, Aoba-ku, Sendai, Japan
b
IRIDeS, Tohoku University, Aza-Aoba 468-1, Aramaki, Aoba-ku, Sendai, Japan
c
Department of Industrial and System Engineering, Faculty of Science and Engineering, Hosei University, Kajino-cho 3-7-2, Koganei-shi, Tokyo, Japan

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

Keywords: In this study, an optimization method based on numerical rockfall simulation is proposed to determine the layout
Rockfall design of a protection embankment, including its position and length on a construction site. Stopping rockfalls
Protection embankment farther away from the slope toe requires a lower embankment since the movement of the rockfall decreases with
Optimization Problem
the runout distance and, at the same time, a longer embankment due to the lateral deviation from the central
Discrete Element Method
Risk Reduction
rockfall path. This complicated relation makes the layout design difficult. The proposed method mainly focuses
Disaster Prevention on two design parameters: the distance from the slope toe and the embankment length. With regard to these two
decision variables, an optimization problem is formulated, with the aim of minimizing the installation cost of an
embankment, which is subject to the global performance requirement for arresting rockfalls. Solving this
problem leads to the identification of an optimal layout plan for an embankment at the site of interest. The
optimization result is objective and quantitative; therefore, it allows us to choose the best one from several
embankment types with different design conditions and costs. To evaluate the performance of the proposed
method, an application example was conducted using a virtual rock-slope model. In this application, we
employed the discrete element method for the rockfall trajectory simulations. The optimization process was then
conducted on 50 different groups of computed trajectories to determine the most economical and stable con­
struction plan among several embankment types and all possible layouts. The reliability of the result was also
supported by the fact that the variations in the optimal solutions were reduced with the increase in the sample
size.

1. Introduction Chiaro et al. 2015).


The implementation of protection or mitigation measures is there­
A rockfall is one of the most significant natural disasters for ground fore important in rockfall-prone areas (Jaboyedoff et al. 2005; Agliardi
structures and people in mountainous regions. Blocks crumble from et al. 2009; Volkwein et al. 2011). Rockfall countermeasures are pri­
rocky cliffs, and the detached blocks tumble down the slopes with high marily categorized into two: active measures and passive measures. The
energy and mobility. The occurrence of these events is triggered by the former category aims to permanently anchor unstable rocks on a slope
following natural phenomena: earthquakes (Kobayashi et al. 1990; face using concrete, wire ropes, bolts, or drapery systems (Thoeni et al.
Chiaro et al. 2015), rainfall (Wei et al. 2014), freeze–thaw cycles 2014; Japan-Road-Association 2017). Conversely, the latter category
(Matsuoka and Sakai 1999), and high winds, as well as anthropogenic aims to obstruct rockfalls and absorb kinetic energy using protection
activities (Macciotta et al. 2015). The occurrence frequency of these structures located at the bases of the mountain slopes (Peila et al. 1998;
events is higher than those of other slope disasters. The rocks also vary in Gentilini et al. 2013; Lambert and Bourrier 2013; Moon et al. 2014;
size, from boulders as large as cars to small rocks and gravels (Hungr Bourrier et al. 2015). Such implementation requires the following tasks:
et al. 1999). Regardless of the block size, collisions with passengers, the identification of rockfall-prone areas and potential sources, analysis
vehicles, or structures cause casualties or significant economic losses of rockfall hazards, risk assessment of the areas, and determination of
(Bunce et al. 1997; Lan et al. 2010; Mavrouli and Corominas 2010; countermeasures.

* Corresponding author.
E-mail address: hasuka.kanno.r7@dc.tohoku.ac.jp (H. Kanno).

https://doi.org/10.1016/j.enggeo.2020.105920
Received 2 June 2020; Received in revised form 9 November 2020; Accepted 17 November 2020
Available online 21 November 2020
0013-7952/© 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
H. Kanno et al. Engineering Geology 284 (2021) 105920

This study focuses on the design of reinforced soil embankments. trajectories, flying heights, velocities, and other valuable information
Embankments are typical passive protection structures, such as protec­ for the planning of protection measures can be computed (Crosta and
tion fences, nets, and rock sheds. Protection embankments currently Agliardi 2004; Volkwein et al. 2011; Lambert and Bourrier 2013). In
include a variety of geometries and constituent materials (i.e., com­ addition, probabilistic approaches enable us to consider the aleatory
pacted ground, rocks, and geosynthetic materials). Among those, rein­ uncertainties involved in the numerical modeling of rockfalls, as pre­
forced soil embankments are most widely used. In this method, an earth sented in Fig. 1 (Li and Lan 2015). Aleatory uncertainty refers to the
fill is strengthened by wood and steel bars, geotextiles, wire meshes, or a variability arising from the lack of observation data or knowledge about
mixture of waste materials (e.g., shredded tires). The embankment ab­ the phenomenon.
sorbs the energy through its own deformation caused by the impact of Several papers have used numerical models to evaluate the move­
rockfalls. Usually, the embankment has a trapezoidal cross section and ments of rockfalls, considering protection embankments (Agliardi et al.
includes steep dip angles of up to ] 2009; Masuya et al. 2009; Lambert et al. 2013). Additionally, the design
on the front side face to prevent rocks with high rotational energy of other protection structures (i.e., barriers or nets) based on reliability-
from passing over (Plassiard and Donzé 2010). based analysis or statistical approaches, addressing the abilities of
It has been reported that in Japanese local municipalities, the de­ structures to resist impact and adequately act on the block propagation,
cisions regarding the construction of rockfall protection structures are has been improved (Bourrier et al. 2015, 2016; Mentani et al., 2016; Toe
made based on field survey results, recent events, or requests from res­ et al. 2018). These improvements have been achieved by utilizing
idents. The use of those current approaches results in problems, such as simulation results to validate a specific, prearranged design plan for
the lack of a uniform method and absence of objectivity. Designers must protection structures. However, the trial-and-error approaches by en­
have an accurate estimation of the magnitude of energy that will be gineers are still required in the decision making during the design pro­
sustained by the protection structure. In addition, the position of the cess. There have been no agreements on how engineers determine the
protection structure on a construction site significantly affects its layout design, which involves the identification of the position and
capability to successfully obstruct rockfalls. The structure should be length of the protection structures on the construction site, in an
designed in a manner that ensures its global performance, considering objective and quantitative way.
the trajectories, passing heights, and kinetic energies of possible rockfall When a rockfall reaches a subhorizontal area, hereafter referred to as
events. As a lucid and easy-to-use approach for estimating the kinetic the “resting area,” its kinetic energy and passing height progressively
energies exhibited by falling blocks, a frictional model based on the “sled decrease with the runout distance. Consequently, stopping the boulder
model” first proposed by Heim (1932) has been employed in practice (e. far from the slope toe requires a lower embankment with a reduced
g., the handbook in Japan (Japan-Road-Association 2017)). This sled height and thus reduced allowable energy (keeping in mind that the
model originally deals with the motion of mass sliding; therefore, some latter depends on the embankment height). Meanwhile, due to the
limitations of the use of the model for rockfalls are recognized. A major lateral deviation, obstructing rockfalls with an embankment at larger
limitation is that the model often overestimates the kinetic energy when distances from the slope toe requires an embankment with longer
a falling block has a radius of several meters or when a block tumbles length. Therefore, there is a balance between the required embankment
down a slope with length on the order of tens of meters. This is due to the height and required embankment length. They change inversely with
lack of consideration of characteristic mechanisms, for example, the respect to the distance from the slope toe, and both relate to the
energy loss of the block resulting from ground deformation during col­ installation cost of an embankment. Thus, optimization of the
lisions with the slope surface. Moreover, the information obtained embankment design with respect to a cost–benefit analysis is not
without analysis of the temporal and spatial movements of rockfalls is straightforward.
insufficient to examine the design of protection structures. In this study, an optimization method for determining the layout
Rockfall simulations with numerical modeling are more valid than design of an embankment based on numerical rockfall simulation is
the aforementioned approach, because in these simulations, the proposed. Our method mainly focuses on two design parameters: the

Fig. 1. Classification of uncertainties involved in the numerical modeling of rockfall phenomena. Adopted from Li and Lan (2015).

2
H. Kanno et al. Engineering Geology 284 (2021) 105920

distance between the slope toe and the embankment and the length of 2. Optimization method
the embankment. Handling of the distance from the slope toe inevitably
involves the consideration of the embankment height and allowable The spatial layout of a protection embankment must be determined
energy (Bourrier et al. 2016). To minimize the cost function subject to to obstruct passing rocks and absorb their kinetic energy in almost all
the required global performance for arresting rockfalls, the distance and predictable cases while maintaining a low total installation cost. The
length are adjusted. This allows us to quantitatively determine the best total installation cost is explicitly related to the embankment construc­
among all selective embankment designs and all possible layouts. tion cost and land cost. It is easy to determine whether an optimization
Finally, in this study, an example of the proposed optimization method approach is applicable to this problem and whether it supports finding a
using simulation results with a simple, virtual rock-slope model is pre­ reasonable solution more rapidly than manual procedures. In the field of
sented. Among several numerical analysis methods, we selected the rock engineering, only a few attempts have been made to utilize such an
discrete element method (DEM) to simulate the large number of rockfall optimization approach; however, a study on the planning of the loca­
trajectories. tions of temporary facilities on construction sites (El-Rayes and Khala­
fallah 2005) can be used as a good reference. Based on the previous
studies, the design plans of an embankment are evaluated in two aspects:

Fig. 2. Judgment algorithm for the embankment design to estimate the global performance against computed rockfall trajectory.

3
H. Kanno et al. Engineering Geology 284 (2021) 105920

global performance and cost efficiency. obtained through a judgment process, a continuous function is required
As discussed in detail below, a global performance function, a cost to easily solve our optimization problem. Therefore, a relational
function, and an optimization problem are formulated in terms of two expression of G(x, l) is described by employing the response surface
decision variables, namely, the distance x between an embankment and method. The response surface model is constructed by a polynomial
a slope toe and the length l of an embankment. The slope toe is defined as function with the help of multiple linear regression analysis. The poly­
the location where the longitudinal inclination of the slope significantly nomial function model has several advantages, such as flexible variable
changes (i.e., the boundary with the resting area). Solving the problem selection and its simple estimation using the linear least squares method.
enables us to identify an optimal layout plan for an embankment at the We assume the following global performance function of polynomial
site of interest. However, construction planners must determine not only form:
the layout but also the height, width, and other design conditions of the
̂ l|θ )
protection structure (Lambert and Bourrier 2013; Moon et al. 2014; De G(x, l) ≈ G(x,
(2)
Biagi et al. 2020). The method for deciding on the latter conditions is = α0 + α1 x + α2 l + α3 x2 + α4 xl + α5 l2
expressed at the end of this section.
where θ denotes the set of unknown parameters:
θ = (α0 , α1 , α2 , α3 , α4 , α5 ) (3)
2.1. Global performance function
The second-order terms are required to express the nonlinearity of
Here, we determine the lowest probabilities that predictable rock­ the original data. Estimation of the maximum–likelihood values of θ is
falls will bypass, clear, or collapse a designed embankment; we use these performed based on the backward elimination approach, in which the
probabilities to evaluate the global performance of the embankment; Akaike information criterion (Akaike 1974) is used as a model fit
and we estimate them quantitatively based on a sample of simulated criterion.
block trajectories. The global performance function is defined as
/
G(x, l) = Nsafe Nall (1) 2.2. Cost function

where Nall denotes the sample size, and Nsafe denotes the number of cases Optimization of only the protection scenario in terms of the global
among the samples for which the embankment successfully obstructs the performance function may result in the excessive specification of the
rockfall. protection structure against infrequent severe hazards. Another
Judging whether an embankment can obstruct a simulated rockfall is perspective in terms of economic efficiency to prevent this problem
conducted algorithmically, as presented in Fig. 2. In the figure, the su­ exists. Economic efficiency quantification also enables us to compare the
perscript t indicates the time in the numerical simulation. The point p scenarios of different types of protection structures.
denotes the center of mass coordinate of the block. E denotes the The cost function is modeled to express the penalty in an uneco­
computed kinetic energy of the block, and Eemb denotes the allowable nomical situation. In this study, two specific situations are considered.
energy of the embankment. The constants γ E and γ h are safety factors in The first situation is an unnecessary distance from the slope toe. In this
terms of the allowable energy and height, respectively, which are case, we must pay a greater land cost than necessary. The other unrea­
considered on the basis of national regulations (Ronco et al. 2009). The sonable situation is an unnecessary structure length. This indicates that
embankment height should be designed as the sum of the maximum high construction costs must be paid. The total cost is assumed to be the
block passing height and a free board (Lambert and Bourrier 2013). sum of the following two factors: the construction cost, which is pro­
Instead of a direct approach that determines both the maximum block portional to the embankment length, and the land cost of the rectangular
passing height and the required free board, we employ the safety factor, area in front of the embankment. Therefore, the total cost C(x, l) is
γ h, to confirm that a block cannot pass over the specified embankment calculated as
height. In addition, through this process, we can comprehensively judge C(x, l) = W1 l + W2 xl (4)
whether the block does not bypass both sides of the embankment and
whether the block energy does not exceed the allowable energy of the where W1 denotes the construction cost of the embankment per unit
embankment. The process is performed at the point where a trajectory length, and W2 denotes the land cost per unit area. W1 denotes the
intersects the frontside face of the embankment. If the block does not embankment technology and type as well as its dimensions. In this
bypass, clear, or collapse the embankment, it is assumed to be arrested study, only the dimensions are considered. W2 depends on the
safely. geographical context and is constant for a specified site, whereas W1 is
There are several reasons why direct numerical simulations of not. Contrary with W2, W1 denotes the principal and most variable
block–embankment interaction should be avoided. First, many diffi­ parameter for a specified site and depends on the required allowable
culties can be encountered in the deformation analysis of protection energy, height, and length of an embankment. These factors all depend
structures in modeling the block–structure interaction, such as material on the embankment position with respect to the slope toe, and they are
property identification, material nonlinearity, global structure response all considered in this study.
representation, or calibration with numerous experimental results
(Ronco et al. 2009; Bertrand et al. 2012; Gentilini et al. 2013; Lambert 2.3. Optimization problem
and Bourrier 2013; Moon et al. 2014). Second, the computational cost
significantly increases as the direct simulation needs to be restarted for The following optimization problem aims to determine the most
all rockfall trajectories each time the protection wall is redesigned. efficient layout design consisting of two decision variables: x and l.
These difficulties discourage engineers from practical use; in addition, Again, the variable x denotes the distance between the embankment and
the great variability in numerical modeling may spoil the objectivity of the slope toe, and the variable l denotes the length of the embankment.
the judgment. Hence, we decide to avoid the incorporation of the direct The cost function is selected as an objective function to be minimized.
simulation into our method. The method also completely separates the The problem to be solved is mathematically written using Eqs. (2) and
propagation simulation process and performance assessment process, (4) as follows:
thus improving the independence of each process. Another great
advantage of this approach is that any tool for rockfall trajectory min C(x, l) { ⃒ }
⃒ (5)
simulation can be employed. ̂ l) ,
subject to S = (x, l) ⃒ b ≤ G(x,
Although the values of G(x, l) with different layouts are discretely

4
H. Kanno et al. Engineering Geology 284 (2021) 105920

where 0 ≤ b ≤ 1. The constant b denotes the global performance et al. 2014; Bourrier et al. 2015; Shen et al. 2019). It provides a detailed
requirement that needs to be satisfied. Setting b = 1 indicates that the description of the parameters that control the motion of rocks, such as
embankment is not allowed to miss the arrest of all possible rockfalls, friction coefficient, viscous damping coefficient, and moment of inertia.
and it often results in overdesign. In practice, protection structures are Moreover, the block movement (i.e., flying, rolling, or sliding) naturally
often designed considering the 95% percentile (i.e., b = 0.95). In addi­ shifts between forms without providing any specific conditions in the
tion, S indicates the feasible region, the set of all possible points satis­ mathematical descriptions. Hence, the DEM is thought to accurately
fying the constraints of the problem, on the x-l plane. represent the behavior of blocks more than the lumped mass model.
Fig. 3 presents the procedure for solving the above optimization However, it should be noted that the proposed optimization method
problem. First, the feasible region S is defined. The outline of S denotes allows any other tool to be utilized in the trajectory simulation, as
mentioned in Section 2.1.
the contour line on the global performance function G(x,̂ l), the joining
The following simulation model and its parameters are provided
points of Ĝ = b, and it is analytically derived from Eqs. (2) and (5).
with limited justification. There is a limitation in the sense that many
Subsequently, the minimum value of the cost function C(x, l) is numer­ data are available from the literature to validate the model. However, in
ically determined in the feasible region. Eventually, the optimal com­ the context of this study, the main objective is to propose a concept for
binations of the two variables x and l can be obtained. optimization and its procedures to be applied, considering the results of
Furthermore, we conduct a parallel layout optimization for several the rockfall trajectory simulation. Thus, we can easily replace the
embankment types with different design conditions. The change in the simulation tool.
design conditions indeed changes the allowable energy, total cost, and
optimal layout of the embankment. The optimal solutions obtained by
3.1. Summary of the DEM
the proposed method are quantitative and comparable; thus, the most
economical design conditions and layout for embankment construction
The DEM is a numerical technique employed to address the behavior
have become obvious in the end.
of rigid-body particles (Cundall and Strack 1979). The sequential motion
of each particle is represented by solving the contact force between
3. Application example
particles subject to Newton’s motion equations. Since the DEM solves
the equations using an explicit time-integration method for all particles,
We apply the optimization method defined in the previous section to
it requires a small time interval and a large computational load. Spe­
a virtual rock-slope model. Such an application primarily aims to eval­
cifically, the use of a larger number of particles results in a higher
uate the performance of the proposed method. First, a three-dimensional
computational cost in the simulation. However, the increases in the
rockfall simulation is repeated 5000 times. Then, the design optimiza­
speed and capacity of computers have expanded the use of DEM.
tion of the protection embankment is repeated while changing the
The interaction between two particles is determined by the contact
sample data or sample size. Finally, the variations in the optimal solu­
model presented in Fig. 4. This model includes a spring and a dashpot in
tions are quantified, the reliability is discussed, and a necessary and
both the normal and tangential directions, along with a slider in the
sufficient number of analysis cases are examined. This application
tangential direction. To achieve an efficient contact detection and sim­
example is related to the release of a single rock whose geometry can be
ple implementation, spherical particles are considered in this study. The
as simple as possible. However, in such simplified conditions, the block
contact force vectors are represented as follows:
trajectory still exhibits wide spatial variability. Therefore, we hypothe­
size that the use of the numerical simulation and proposed optimization Fn + Cn u̇ + Kn u = 0 (6)
method could be sufficient for this application example.
Among the three-dimensional rockfall simulation methods (e.g., ss + Cs v̇ + Ks v = 0 (7)
lumped mass model method and discontinuous deformation analysis
where Fn and Fs denote the contact force vectors in the normal and
method), the DEM is selected to be employed in this study. In the field of
tangential directions, respectively; u and u̇ denote the relative
rockfall engineering, the DEM has been utilized in several studies to
displacement vector and relative velocity vector in the normal direction,
analyze the characteristic motion of blocks (Masuya et al. 2009; Wang
respectively; v and v̇ denote the relative displacement vector and relative
and Tonon 2011; Zhao et al. 2017) or the block–protection structure
velocity vector in the tangential direction, respectively; Kn and Ks denote
interaction (Plassiard and Donzé 2010; Bertrand et al. 2012; Thoeni
the spring constants in the normal and tangential directions,

Fig. 3. Procedure for solving the optimization problem.

5
H. Kanno et al. Engineering Geology 284 (2021) 105920

Fig. 4. DEM contact model.

respectively; and Cn and Cs denote the viscosity coefficients of the particle does not change the results of the calculation, it is assumed to be
dashpot. We assume that in both directions, the spring constants and equal to the contacting particle for the simple calculation algorithm.
viscosity coefficients are the same. The viscosity coefficients have
theoretical solutions based on the vibration equation under a single- 3.2. Numerical rock-slope model
degree-of-freedom system as follows:
√̅̅̅̅̅̅̅̅̅ We consider the free fall of a boulder or a mass of rock, for which the
Cn = 2h mK n (8)
position and mass are specified. The boulder is hereafter referred to as
√̅̅̅̅̅̅̅̅̅ the “block.” To consider the uncertainty of the phenomenon, the block
Cs = 2h mK s (9)
has zero initial velocity and randomly undergoes a three–dimensional
where m denotes the mass of the block, and h denotes the attenuation rotation before falling; that is, it is rotated around each axis at certain
coefficient. The attenuation coefficient is obtained from the restitution degrees, as sampled from a uniform probability distribution in the range
coefficient ec and the circle ratio π as follows: of ±20∘. This randomness provides strong variability in the estimated
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ trajectories despite the simplified geometric features of the block and
(lnec )2 slope. In the simulation, the block has a mass of approximately 110 t,
h= (10) and a height difference exists between the departing area and resting
(lnec )2 + π2
area of approximately 23 m.
In the tangential direction, the slider limits the magnitude of the Fig. 5 presents the numerical model of the block in the simulation.
contact force subject to Coulomb’s law of friction as follows: The rock has a square bipyramidal body, which represents the shape of
( an angular boulder. It has been known that modeling the shape of rocks
Fs (if |Fs |≤ μ|Fn |)
Fs = (11) highly characterizes their falling motion (Leine et al. 2014) and the
μFs ∣Fn ∣/∣Fs ∣ (if |Fs |> μ|Fn |)
maximum impact force (Breugnot et al. 2016; Shen et al. 2019). The
where μ denotes the friction coefficient. The summation of the total complicated shapes also provide great variability in rockfall trajectories,
force on each particle is computed by solving Eqs. (6) and (7); then, the whereas the extreme cases, such as a block with a major axis signifi­
motion parameters of each particle are updated over the small time cantly longer than the minor axes, impede the rotational movement of
interval. To describe the collision with slope surfaces, the contact be­ blocks and causes attenuated mobility. This block geometry is defined to
tween a particle and a polygonal element is also considered. As pre­ pose difficulty with regard to correctly predicting where the block would
sented on the right of Fig. 4, another particle of which the material pass. In the DEM, clumped spherical particles can approximately
properties correspond to the slope material is virtually set on the represent arbitrary shapes. The clump behaves as a rigid body since the
opposite side of the contact surface; hence, the contact force is computed relative positions of the particles do not change. It should be noted that
to be the same for the particles. Although the diameter of the virtual the clumped model does not strictly satisfy Eqs. (8)–(10); nevertheless,
we are employing the model only as a simple calculation algorithm.

Fig. 5. Rock model in DEM simulation.

6
H. Kanno et al. Engineering Geology 284 (2021) 105920

Fig. 6 presents a cross section of the slope surface model, which has a Table 1
width of 40 m in the y-direction. The surface is represented by a set of DEM input parameters.
polygonal elements of a right-angled triangle with sides of approxi­ Rock volume 38 (m3)
mately 3 to 4 m, which are arranged without space. We define point d as 5
Rock mass 1.1×10 (kg)
the slope toe. Point g corresponds to the center of gravity coordinates of Particle radius 0.60 (m)
the block to be released. Although the initial direction of the block is Restitution coefficient 0.30 (− )
changed for each simulation, this gravity point is stable due to the Friction angle 30 (deg)
symmetric property. Spring coefficient 1.0×108 (N/m)
Time step 1.0×10− 6 (sec)
The DEM input parameters are presented in Table 1. The restitution
coefficient on the natural slope surface is commonly determined to be
within the range of 0.1–0.2 (Masuya et al. 2009; Thoeni et al. 2014);
however, we have defined the higher restitution coefficient, with the Table 2
aim of emphasizing the spatial dispersion of block trajectories in this Dimensions of protection embankments.
application example. Conversely, the friction angle of 30∘ is derived Type-1 Type-2 Type-3
from previous studies which performed rockfall simulations on rough, Bottom width L1 (m) 4.47 6.00 7.52
rocky slopes (Japan-Road-Association 2002; Zhao et al. 2017). The Top width L2 (m) 0.90 0.90 0.90
spring coefficient and time step are adjusted within general definitions Height H (m) 4.20 6.00 7.80
Dip angle (deg) 67 67 67
to prevent the DEM particles from penetrating deep into the slope sur­ α
Cross-sectional area A (m2) 11.28 20.70 32.84
face (Japan-Road-Association 2002; Thoeni et al. 2014; Zhao et al. Allowable energy Eemb (kJ) 2,819 5,175 8,210
2017). Construction cost W1 (USD/m) 1,887 3,774 5,660
Land cost W2 (USD/m2) 943 943 943

3.3. Embankment types


embankment. It has been known that large downhill face displacements
indicate embankment collapse. Hence, we employ this criterion and
In this study, three embankment types are considered to compare the
roughly estimate the allowable energy of embankments from the cross-
results of layout optimization among various embankments with
sectional area. It should be noted that the above estimation does not
different sizes. The dimensions of the embankments are presented in
consider the properties of constitutive materials. In addition, the crite­
Table 2. All structures are assumed to exhibit a trapezoidal cross section,
rion is sometimes applied in cases outside of its validity domain, in
which is a typical shape of reinforced soil embankments (Peila et al.
particular, high-impact energies. Therefore, in this application, we
2007; Ronco et al. 2009; Lambert and Kister 2018). L1 denotes the
carefully define the safety factor for embankment resistance γ E = 1.3.
bottom width; L2, the top width; H, the height; α, the dip angle of both
the uphill face and downhill face; A, the cross-sectional area; and Eemb,
the allowable energy of embankments. The construction cost W1 is 3.4. Results
determined, assuming that a positive correlation exists between the
cross-sectional area and construction cost, whereas the land cost W2 is 3.4.1. Results of the rockfall trajectory simulations
understandably constant regardless of the embankment type. It should Fig. 7 presents the plane view of three groups of rockfall trajectories
be noted that although W1 also depends on the structure technology, we that differ in the number of cases. It also summarizes the histograms of
consider a single type of constituent materials and a similarity shape of the longitudinal dispersion of the trajectories on the resting area. The
cross-sectional geometry in this study. Moreover, the cost per unit length comparison of the three groups is performed, with the aim of avoiding
is arbitrary, and thus, the following results are limited to this unnecessary computational costs for trajectory simulation. One of the
application. three groups represents all 5000 calculation cases, whereas the other
From Table 2, it can be observed that Type-3 has exhibits the highest two represent the 500 and 100 cases that are respectively sampled from
energy dissipation capacity, but it is the most expensive, followed by all cases in a random manner. In the plane view, each line traces the
Type-2 and then Type-1. The value of Eemb is derived from the impact centroid position of the block on the slope surface, and the line color
strength criterion for reinforced embankments proposed by Lambert and changes from purple to yellow in terms of the magnitude of kinetic
Kister (2018). Such a criterion defines the magnitude of kinetic energy energy exhibited by the block. Significant differences obviously exist in
that can statistically avoid significantly large downhill face displace­ the density of the lines drawn on the plane; nevertheless, the histograms
ments by the inequality expression E/A < 250. Here, E denotes the ki­ indicate quite similar longitudinal dispersion. In the same vein, the
netic energy of a rockfall, and A denotes the cross-sectional area of the lateral dispersion is observed in Fig. 8. The figure presents the histo­
grams of the lateral dispersion at two locations: the slope toe (x = 0 m)
and the distant area (x = 15 m). At x = 0 m (on the left side of Fig. 8), the
histograms indicate a roughly similar lateral dispersion. However, at
x = 15 m (on the right side), the reduction of calculation cases resulted
in the narrowing of the tail and a bias to one side that would not have
been true otherwise. These drawbacks worsen as the sample size is
reduced, thus affecting the reliability of the following optimization
process.
The use of numerical simulations allows an appropriate assessment
of kinetic energy, regardless of the magnitude and characteristics of
rockfalls; in addition, it enables us to consider the variability contained
in the phenomenon in the hazard assessment. As presented in Fig. 8, the
lateral dispersion of approximately 20 m occurred at the slope toe by
changing the initial direction of the block. This result indicates the
strong variability of trajectories and the importance of three-
dimensional modeling for managing the risks of rockfall (Crosta and
Fig. 6. Cross section of the slope surface model. Agliardi 2004).

7
H. Kanno et al. Engineering Geology 284 (2021) 105920

Fig. 7. Plane view of the three groups of rockfall trajectories that differ in the number of cases and the histograms of their longitudinal dispersion.

8
H. Kanno et al. Engineering Geology 284 (2021) 105920

Fig. 8. The histograms of the lateral dispersion at two locations, with regard to the three groups of rockfall trajectories.

3.4.2. Results of optimization


The most economical and stable construction plan is identified for 50
groups of sample trajectories and for all 5000 trajectories. The 50 groups
consist of five different sample sizes (100, 200, 300, 400, and 500), each
with 10 random sampling iterations. Based on the results to be obtained
from them, we will discuss the necessary and sufficient number of
analysis cases for this application at the end of this section.
To estimate a response surface model G(x, ̂ l) of the global perfor­
mance function, the judgments demonstrated in Fig. 2 that present Nsafe
and G(x, l) are first processed for 96 different layout designs, with x
ranging from 0 to 15 in increments of 1 and l ranging from 10 to 30 in
increments of 4. Among those judgment processes, the safety factor for
an embankment height γh of 1.1, the safety factor for the allowable
energy γ E of 1.3, and the global performance requirement b of 0.90 are
applied. Although the requirement of b = 0.95 is more general, we
intentionally set the lower value to avoid having no solutions in the
following optimization problem. Based on the values of G(x, l), the or­
dinary least squares regression, and the model selection to determine
G(x,
̂ l), the deletion of each variable in Eq. (3) using AIC is then tested
using R (R Core Team 2017). Fig. 9. Response surface model of the global performance function obtained via
multiple linear regression analysis.
Fig. 9 presents an example of a response surface model in which a
sample of 500 trajectories and the Type-2 embankment are considered.
It determines two statistics, namely, the mean and the standard devia­
The blue-green surface indicates the response surface model G(x,
̂ l), and
tion of the residual error, which can be calculated as follows:
the magenta points (which partially appear to be different colors as the
surface overlays them) indicate the values of G(x, l) from the judgments 1 ∑ m ∑ n (
( ) ( ))
on 96 embankment layouts. To evaluate the accuracy of the response MeanErr = ̂ xi , lj
G xi , lj − G (12)
m⋅n i=1 j=1
surface model, the leave-one-out cross-validation method is employed.

9
H. Kanno et al. Engineering Geology 284 (2021) 105920

√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
√∑ problem. Fig. 11 presents the relationship between the number of
n ( ( ) ( ))2
√m ∑ ̂ xi , lj computed trajectories and variation in the optimal solutions. The means
√ G xi , lj − G
√i=1 j=1 and standard deviations are plotted in the line graphs; the horizontal
SdErr = − Mean2Err (13)
m⋅n axis indicates the group sample size, and the vertical axis indicates the
This example results in MeanErr = 2.03 × 10− 4 and SdErr = 0.0227. optimized value of x or l. For comparison, the dotted lines indicate the
We have confirmed that these statistics are sufficiently small, and thus, optimized values obtained by using 5000 simulation cases as a true
the response surface model accurately describes the global performance value. We confirmed that the variations in the optimal solutions caused
function. Moreover, the cumulative distribution of the difference be­ by the change in the sample data are reduced with the increase in sample
size. Focusing on the average values, it also appears that even a few
tween G(x, l) and G(x,
̂ l) is presented in Fig. 10. The probability of this
hundred simulation cases can produce an optimal solution that is suf­
model overestimating the global performance is less than 50%, which
ficiently close to the true value.
indicates that the model is globally conservative. This tendency can also
Therefore, to confirm whether a few hundred simulation cases could
be observed in other samples or structure types. Fig. 9 demonstrates that
have produced a reliable solution, the following examination is con­
the global performance increases monotonically and asymptotically
ducted. In this examination, an optimal solution from Sample 1 and
approaches 1 as the distance and length of the protection embankment
Type-3 embankment is utilized as an example, which results in the
increase. This tendency is common to all types of embankments pre­
lowest cost in Table 3. Fig. 12 presents the optimized protection scenario
sented in Table 2.
virtually constructed for the 500 trajectories in the sample. This figure
Then, each optimization problem is numerically solved using R,
demonstrates that the trajectories are visually cut on the front side of the
evaluating all values of C(x, l) on grid points that finely divide the
embankment according to the judgment algorithm (see Fig. 2), which
feasible region on the x-l plane. Table 3 presents some of the obtained
results in some trajectories not being arrested when passing through or
solutions, selecting 10 cases with a sample size of 500. It should be noted
bypassing the embankment. The embankment has a distance of 7.3 m
that the blank spaces in the table indicate that the solution that satisfies
from the slope toe, which is 0.3 m below the true value; it also has a
the global performance requirement does not exist. The variation in the
length of 16.7 m, which is 0.3 m below the true value. To confirm the
total cost C among different samples originates from the changes in both
reliability of this optimal solution, we recalculate the global perfor­
the embankment length l, which directly affects the construction cost
mance of this designed embankment for all 5000 simulation cases.
and land cost, and the embankment position x, which affects not only
Fig. 13 presents a cross-sectional view cut out at the front side of the
the land cost but also the required embankment length. However, C
embankment, which represents the centroid positions of the blocks on
varies among different scenarios using a certain type of embankment,
the y-z plane. The black dots indicate the blocks below the allowable
since the construction cost per unit length increases with the increase in
energy of the Type-3 embankment, whereas the red points are above the
embankment height, whereas the required x and l decrease. Table 4
allowable energy (i.e., unsafe cases for the embankment). Additionally,
presents the mean, maximum, and minimum values of the total costs for
the dots outside the hatched area indicate unsafe cases of embankments.
the 10 samples. In addition to the cost itself, the differences between the
The recalculated global performance G(x, l) is 0.88, which is slightly
maximum and minimum values (i.e., variation in the optimal solutions)
below the required value of 0.90 in this application. This error is thought
are smaller according to the ascending order of the allowable energy.
to be caused by the underestimation of the lateral dispersion of trajec­
Thus, a greater allowable energy obviously corresponds to a smaller
tories when the number of calculations is reduced (see Fig. 8).
allowed distance from the slope toe to satisfy the global performance
In addition, there is a limitation to the embankment height design.
requirement. A smaller distance from the slope toe reduces the lateral
Essentially, the height is not a decision variable of this optimization
deviation of rockfall trajectories and further reduces the variation in the
problem; thus, it is determined corresponding to the embankment type,
required embankment lengths. Hence, it is believed that the variation in
as presented in Table 2. The embankment presented in Fig. 12, the cross
the total cost consequently decreases according to the allowable energy.
section of which is derived from the specified dimensions of the Type-3
Thus, the construction of the Type-3 embankment is more economical
embankment, seems to be oversized for the passing height of the tra­
than the construction of the other two types in all groups of sample
jectories. However, it has at least been confirmed that the total cost is
trajectories. Moreover, the change in the sample selection does not
lowest among the three types and that the allowable energy is sufficient.
significantly affect the optimization in terms of embankment types.
Indeed, a necessary and sufficient number of analysis cases varies
Finally, we investigate how much trajectory simulation computa­
depending on the variations in the slope topography and block shape.
tional cost is required to obtain a reliable solution for this optimization
However, the estimation of the minimum number of trajectory simula­
tions that sufficiently represent the variability of the phenomenon is not
straightforward. For instance, although a complex topography and block
shapes make it difficult to predict trajectories, the holistic variations in
trajectories can be rather small. The reliability of an optimal solution can
be enhanced by increasing the number of trajectory simulations in ex­
change for the computational cost. Thus, avoidance of an unnecessary
computation becomes an important issue when using the proposed
optimization problem in practice. Further work is therefore required to
improve the robustness of the optimization with respect to the
embankment length from the viewpoint of engineering as well as to
establish an estimation approach for the minimum number of simula­
tions that can provide a stable optimal solution.

4. Conclusion

In this study, a novel method for determining the layout design of a


rockfall protection embankment based on a numerical simulation was
proposed. Our method mainly focused on two design parameters,
Fig. 10. Cumulative distribution of the model error of the global performance namely, the distance between the slope toe and the embankment and the
function in an example. length of the embankment. With regard to these two decision variables,

10
H. Kanno et al. Engineering Geology 284 (2021) 105920

Table 3
Optimal solutions for different types of embankment and different samples consisting of 500 trajectory simulation cases.
Type-1 Type-2 Type-3

Sample x l C x l C x l C

(m) (m) (k USD) (m) (m) (k USD) (m) (m) (k USD)

1 15.0 17.7 283.1 14.9 13.6 242.9 7.3 16.7 209.3


2 15.0 18.3 293.8 14.1 15.4 263.1 7.6 16.7 214.8
3 15.0 21.8 349.8 15.0 13.6 244.2 8.2 16.4 220.7
4 15.0 17.4 278.8 13.9 16.8 285.2 7.8 17.7 230.2
5 15.0 19.3 307.0 13.7 13.2 221.3 7.6 16.8 216.5
6 14.7 13.9 245.7 8.1 16.4 218.4
7 15.0 19.0 304.6 15.0 13.4 239.4 7.8 16.8 218.9
8 15.0 19.5 313.2 14.9 13.8 245.7 8.6 16.0 221.4
9 15.0 19.7 315.4 14.9 14.2 252.5 8.1 16.2 214.8
10 15.0 19.8 317.5 14.9 14.8 265.2 8.5 16.0 219.8

change with respect to the distance from the slope toe, in the context of a
Table 4
cost–benefit analysis. Since the optimization was conducted in an
The mean, maximum, and minimum values of C(x, l) in 500 trajectory simulation
objective and quantitative manner, it could also provide the best choice
cases.
from among the several types of embankments that differ in design
Type-1 Type-2 Type-3
conditions and costs. Additionally, the proposed method completely
Mean (k USD) 307.0 250.5 218.5 separated the propagation simulation process and performance assess­
Max. (k USD) 349.8 285.2 230.2 ment process, thus improving the independence of each process. This
Min. (k USD) 278.8 221.3 209.3
approach has exhibited the great advantage that any tool for the rockfall
trajectory simulation could be employed.
we have formulated the optimization problem with the aim of mini­ An application example using a virtual rock-slope model was con­
mizing the installation cost of an embankment subject to the global ducted to confirm the performance of the proposed method. For the
performance requirement. By solving this optimization problem, an rockfall trajectory simulations in this application, we employed the
optimal layout design for an embankment at the site of interest could be DEM. This numerical analysis method not only enabled an accurate
determined. This method allowed us to balance the required embank­ assessment of kinetic energies but also allowed us to consider the spatial
ment height and the required embankment length, which inversely variability contained in the rockfall phenomenon. Subsequently, the

Fig. 11. Relationship between the number of computed trajectories and variation in the optimal solutions. The dotted lines indicate the optimized values obtained by
using 5000 simulation cases.

11
H. Kanno et al. Engineering Geology 284 (2021) 105920

Fig. 12. Visualization of an optimized protection scenario.

Fig. 13. Cross-sectional view cut out at the front side of a designed embankment. The centroid positions of the blocks on the y-z plane are represented with respect to
5000 simulations.

optimization processes were conducted on 50 groups of computed tra­ the work reported in this paper.
jectories consisting of five different sample sizes, each with 10 iterations
of random sampling. In this application example, we could successfully Acknowledgments
identify the most economical and reliable construction plan among
several types of embankments and all possible layouts. The reliability of This work is supported by Grant-in-Aid for JSPS Research Fellow
the result was supported by the fact that the variations in optimal so­ (18 J20339). We thank Professor I. Yoshida at Tokyo City University for
lutions were reduced with the increase in the sample size. his helpful comments on this work.
Although the rock-slope model in this application was very simpli­
fied, the proposed framework of optimization methods should be References
applicable for models exhibiting complex geometries. Further work
should include follow-up studies to achieve a more proper optimization Agliardi, F., Crosta, G.B., Frattini, P., 2009. Integrating rockfall risk assessment and
countermeasure design by 3D modelling techniques. Nat. Hazards Earth Syst. Sci. 9,
in terms of embankment height and width, in addition to establishing an 1059–1073. https://doi.org/10.5194/nhess-9-1059-2009.
estimation approach for the minimum number of simulations, which can Akaike, H., 1974. A new look at the statistical model identification. IEEE Trans. Autom.
provide a stable optimal solution. Control 19, 716–723. https://doi.org/10.1109/TAC.1974.1100705.
Bertrand, D., Trad, A., Limam, A., Silvani, C., 2012. Full-scale dynamic analysis of an
innovative rockfall fence under impact using the discrete element method: from the
Declaration of Competing Interest local scale to the structure scale. Rock Mech. Rock. Eng. 45, 885–900. https://doi.
org/10.1007/s00603-012-0222-5.
Bourrier, F., Lambert, S., Baroth, J., 2015. A reliability-based approach for the design of
The authors declare that they have no known competing financial rockfall protection fences. Rock Mech. Rock. Eng. 48, 247–259. https://doi.org/
interests or personal relationships that could have appeared to influence 10.1007/s00603-013-0540-2.

12
H. Kanno et al. Engineering Geology 284 (2021) 105920

Bourrier, F., Baroth, J., Lambert, S., 2016. Accounting for the variability of rock Li, L., Lan, H., 2015. Probabilistic modeling of rockfall trajectories: a review. Bull. Eng.
detachment conditions in designing rockfall protection structures. Nat. Hazards 81, Geol. Environ. 74, 1163–1176. https://doi.org/10.1007/s10064-015-0718-9.
365–385. https://doi.org/10.1007/s11069-015-2084-0. Macciotta, R., Martin, C.D., Edwards, T., Cruden, D.M., Keegan, T., 2015. Quantifying
Breugnot, A., Lambert, S., Villard, P., Gotteland, P., 2016. A discrete/continuous coupled weather conditions for rock fall hazard management. Georisk 9, 171–186. https://
approach for modeling impacts on cellular geostructures. Rock Mech. Rock. Eng. 49, doi.org/10.1080/17499518.2015.1061673.
1831–1848. https://doi.org/10.1007/s00603-015-0886-8. Masuya, H., Amanuma, K., Nishikawa, Y., Tsuji, T., 2009. Basic rockfall simulation with
Bunce, C.M., Cruden, D.M., Morgenstern, N.R., 1997. Assessment of the hazard from rock consideration of vegetation and application to protection measure. Nat. Hazards
fall on a highway. Can. Geotech. J. 34, 344–356. https://doi.org/10.1139/t97-009. Earth Syst. Sci. 9, 1835–1843. https://doi.org/10.5194/nhess-9-1835-2009.
Chiaro, G., Kiyota, T., Pokhrel, R.M., Goda, K., Katagiri, T., Sharma, K., 2015. Matsuoka, N., Sakai, H., 1999. Rockfall activity from an alpine cliff during thawing
Reconnaissance report on geotechnical and structural damage caused by the 2015 periods. Geomorphology 28, 309–328. https://doi.org/10.1016/S0169-555X(98)
Gorkha Earthquake, Nepal. Soils Found. 55, 1030–1043. https://doi.org/10.1016/j. 00116-0.
sandf.2015.09.006. Mavrouli, O., Corominas, J., 2010. Vulnerability of simple reinforced concrete buildings
Core Team, R., 2017. R: A Language and Environment for Statistical Computing. R to damage by rockfalls. Landslides 7, 169–180. https://doi.org/10.1007/s10346-
Foundation for Statistical Computing, Vienna, Austria. URL: https://www.r-project. 010-0200-5.
org/. Mentani, A., Govoni, L., Gottardi, G., Lambert, S., Bourrier, F., Toe, D., 2016. A new
Crosta, G.B., Agliardi, F., 2004. Parametric evaluation of 3D dispersion of rockfall approach to evaluate the effectiveness of rockfall barriers. Proc. Eng. 158, 398–403
trajectories. NHESS 4, 583–598. https://doi.org/10.5194/nhess-4-583-2004. vI Italian Conference of Researchers in Geotechnical Engineering, CNRIG2016 -
Cundall, P.A., Strack, O.D.L., 1979. A discrete numerical model for granular assemblies. Geotechnical Engineering in Multidisciplinary Research: from Microscale to
Géotechnique 29, 47–65. https://doi.org/10.1680/geot.1979.29.1.47. Regional Scale, 22–23 September 2016, Bologna (Italy). https://doi.org/10.1016/j.
De Biagi, V., Marchelli, M., Peila, D., 2020. Reliability analysis and partial safety factors proeng.2016.08.462.
approach for rockfall protection structures. Eng. Struct. 213, 110553. https://doi. Moon, T., Oh, J., Mun, B., 2014. Practical design of rockfall catchfence at urban area
org/10.1016/j.engstruct.2020.110553. from a numerical analysis approach. Eng. Geol. 172, 41–56. https://doi.org/
El-Rayes, K., Khalafallah, A., 2005. Trade-off between safety and cost in planning 10.1016/j.enggeo.2014.01.004.
construction site layouts. J. Constr. Eng. Manag. 131, 1186–1195. https://doi.org/ Peila, D., Pelizza, S., Sassudelli, F., 1998. Evaluation of behaviour of rockfall restraining
10.1061/(ASCE)0733-9364(2005)131:11(1186). nets by full scale tests. Rock Mech. Rock. Eng. 31, 1–24. https://doi.org/10.1007/
Gentilini, C., Gottardi, G., Govoni, L., Mentani, A., Ubertini, F., 2013. Design of falling s006030050006.
rock protection barriers using numerical models. Eng. Struct. 50, 96–106. https:// Peila, D., Oggeri, C., Castiglia, C., 2007. Ground reinforced embankments for rockfall
doi.org/10.1016/j.engstruct.2012.07.008. protection: design and evaluation of full scale tests. Landslides 4, 255–265. https://
Heim, A., 1932. Bergsturz und menschenleben (Landslides and human lives), 20. Fretz & doi.org/10.1007/s10346-007-0081-4.
Wasmuth. Plassiard, J.P., Donzé, F.V., 2010. Optimizing the design of rockfall embankments with a
Hungr, O., Evans, S.G., Hazzard, J., 1999. Magnitude and frequency of rock falls and rock discrete element method. Eng. Struct. 32, 3817–3826. https://doi.org/10.1016/j.
slides along the main transportation corridors of southwestern British Columbia. engstruct.2010.08.025.
Can. Geotech. J. 36, 224–238. https://doi.org/10.1139/t98-106. Ronco, C., Oggeri, C., Peila, D., 2009. Design of reinforced ground embankments used for
Jaboyedoff, M., Dudt, J.P., Labiouse, V., 2005. An attempt to refine rockfall hazard rockfall protection. Nat. Hazards Earth Syst. Sci. 9, 1189–1199. https://doi.org/
zoning based on the kinetic energy, frequency and fragmentation degree. Nat. 10.5194/nhess-9-1189-2009.
Hazards Earth Syst. Sci. 5, 621–632. https://doi.org/10.5194/nhess-5-621-2005. Shen, W., Zhao, T., Dai, F., Jiang, M., Zhou, G.G., 2019. DEM analyses of rock block
Japan-Road-Association, 2002. Survey Research Data of Rockfall Simulation Method: shape effect on the response of rockfall impact against a soil buffering layer. Eng.
Reference Material for the Rockfall Measure Handbook. Maruzen Publisher, Tokyo. Geol. 249, 60–70. https://doi.org/10.1016/j.enggeo.2018.12.011.
Japan-Road-Association, 2017. Manual for Anti-Impact Structures against Falling Rocks. Thoeni, K., Giacomini, A., Lambert, C., Sloan, S.W., Carter, J.P., 2014. A 3D discrete
Maruzen Publisher, Tokyo. element modelling approach for rockfall analysis with drapery systems. Int. J. Rock
Kobayashi, Y., Harp, E.L., Kagawa, T., 1990. Simulation of rockfalls triggered by Mech. Min. Sci. 68, 107–119. https://doi.org/10.1016/j.ijrmms.2014.02.008.
earthquakes. Rock Mech. Rock. Eng. 23, 1–20. https://doi.org/10.1007/ Toe, D., Mentani, A., Govoni, L., Bourrier, F., Gottardi, G., Lambert, S., 2018. Introducing
BF01020418. meta-models for a more efficient hazard mitigation strategy with rockfall protection
Lambert, S., Bourrier, F., 2013. Design of rockfall protection embankments: a review. barriers. Rock Mech. Rock. Eng. 51, 1097–1109. https://doi.org/10.1007/s00603-
Eng. Geol. 154, 77–88. https://doi.org/10.1016/j.enggeo.2012.12.012. 017-1394-9.
Lambert, S., Kister, B., 2018. Efficiency assessment of existing rockfall protection Volkwein, A., Schellenberg, K., Labiouse, V., Agliardi, F., Berger, F., Bourrier, F.,
embankments based on an impact strength criterion. Eng. Geol. 243, 1–9. https:// Dorren, L.K., Gerber, W., Jaboyedoff, M., 2011. Rockfall characterisation and
doi.org/10.1016/j.enggeo.2018.06.008. structural protection - a review. Nat. Hazards Earth Syst. Sci. 11, 2617–2651.
Lambert, S., Bourrier, F., Toe, D., 2013. Improving three-dimensional rockfall trajectory https://doi.org/10.5194/nhess-11-2617-2011.
simulation codes for assessing the efficiency of protective embankments. Int. J. Rock Wang, Y., Tonon, F., 2011. Discrete element modeling of rock fragmentation upon
Mech. Min. Sci. 60, 26–36. https://doi.org/10.1016/j.ijrmms.2012.12.029. impact in rock fall analysis. Rock Mech. Rock. Eng. 44, 23–35. https://doi.org/
Lan, H., Martin, C.D., Zhou, C., Lim, C.H., 2010. Rockfall hazard analysis using LiDAR 10.1007/s00603-010-0110-9.
and spatial modeling. Geomorphology 118, 213–223. https://doi.org/10.1016/j. Wei, L.W., Chen, H., Lee, C.F., Huang, W.K., Lin, M.L., Chi, C.C., Lin, H.H., 2014. The
geomorph.2010.01.002. mechanism of rockfall disaster: a case study from Badouzih, Keelung, in northern
Leine, R.I., Schweizer, A., Christen, M., Glover, J., Bartelt, P., Gerber, W., 2014. Taiwan. Eng. Geol. 183, 116–126. https://doi.org/10.1016/j.enggeo.2014.10.008.
Simulation of rockfall trajectories with consideration of rock shape. Multi. Syst. Dyn. Zhao, T., Crosta, G.B., Utili, S., De Blasio, F.V., 2017. Investigation of rock fragmentation
32, 241–271. https://doi.org/10.1007/s11044-013-9393-4. during rockfalls and rock avalanches via 3-D discrete element analyses. J. Geophys.
Res. Earth Surf. 122, 678–695. https://doi.org/10.1002/2016JF004060.

13

You might also like