Physics-based algorithm to simulate tree dynamics

under wind load

Lifeng Xu, Zhongzhu Yang, Weilong Ding, Gerhard G. Buck-Sorlin

To cite this version:

Lifeng Xu, Zhongzhu Yang, Weilong Ding, Gerhard G. Buck-Sorlin. Physics-based algorithm to simu-
late tree dynamics under wind load. International Journal of Agricultural and Biological Engineering,
2020, 13 (2), pp.26-32. �10.25165/j.ijabe.20201302.4967�. �hal-02886776�

HAL Id: hal-02886776

https://hal.inrae.fr/hal-02886776
Submitted on 1 Jul 2020

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 26 March, 2020 Int J Agric & Biol Eng Open Access at https://www.ijabe.org Vol. 13 No. 2

Physics-based algorithm to simulate tree dynamics under wind load

Lifeng Xu1, Zhongzhu Yang1, Weilong Ding1, Gerhard Buck-Sorlin2*

(1. College of Computer Science & Technology, Zhejiang University of Technology, Hangzhou 310026, China;
2. IRHS, INRAE, AGROCAMPUS OUEST, Universitéd'Angers, SFR 4207 QUASAV, 42 rue Georges Morel,
49071 Beaucouzécedex, France)

Abstract: Rapid development in the different computer science fields during the recent decades has facilitated the creation of
new applications in the area of dynamic simulation of plant development. Among these new applications, simulation of trees
swaying in the wind is of great importance, as those computer graphics related areas, e.g., computer games, tree cultivation and
forest management simulations, help a lot in revealing the mechanisms of tree dynamics under wind load. However, it is a big
challenge to balance the effect of visualization in real time and calculation efficiency for any simulation algorithm. A
physics-based algorithm to simulate tree dynamics under wind load was proposed in this study. A mechanistic model
simulating the bending of a cantilever beam was used within the algorithm to simulate deformation of stems, and the algorithm
was integrated with a landscape model in which different types of trees were constructed with an L-system-based formalism.
Simulation results show that realistic dynamic effects can be achieved with reasonably high computational efficiency.
Keywords: physics-based algorithm, tree model, L-system, Cantilever beam
DOI: 10.25165/j.ijabe.20201302.4967

Citation: Xu L F, Yang Z Z, Ding W L, Buck-Sorlin G. Physics-based algorithm to simulate tree dynamics under wind load.
Int J Agric & Biol Eng, 2020; 13(2): 26–32.

theory. Virtual plant modeling approaches include two major

1 Introduction categories, i.e. graphical-based approaches[5] and image-based
Trees are an essential component of the natural environment, approaches[6]. Aono and Kunii[7] made a detailed estimate of the
and virtual tree models have been widely used in many aspects distribution of the stem and the leaves on the parent by using the
such as digital games, movies, urban modeling, plant biology, tree as biological object model, and developed an interactive
forestry, and agriculture. Given such complex structures, and the synthetic tree operating system called A-system. Oppenheimer[8]
large amounts of data coming along with them, it is difficult to used fractals to construct a tree model, adjusting the parameters to
construct a model with instant rendering and interaction between change the geometric and topological shapes of the simulated
trees and the environment. With the continuous development of object, considering the random effects of the environment to avoid
virtual reality theory and technology, more and more researchers the self-similarity of the fractal, resulting in greater diversity and
are focusing on the dynamic simulation of trees in the wind[1]. As realism. In order to solve the problems inherent to plant image
a non-linear flexible system, it is relatively difficult to simulate the processing, Shlyakher et al.[9] used image segmentation and
deformation of a tree in the wind accurately and quickly, thus image-based three-dimensional reconstruction techniques to
making it a challenging topic in computer graphics. Previous produce the skeleton (trunk and branch) of the plant and L
methods[2] can only be used within relatively simple tree models, in systems[10] to generate the 3D model of the plant. Based on image
which the control nodes and shape fitting patches are relatively rendering technology (IBR, Image-Based Rendering), the rapid
small, mostly counted in hundreds of control nodes and thousands visual modeling of plant images was well realized. However, the
of shape fitting patches. Obviously, the use of such a simple reconstruction of skeletons mainly depended on the approximate
model to generate the natural sway of complex trees is difficult; the 3D shape of the plant, and the result was different from the actual
dynamic deformation of the calculation accuracy and imaging of skeletons of the plants. Okabe et al.[11,12] designed and
the realism will be less precise. implemented a 3D model system using hand-drawn sketches and
Virtual plant technology uses computers to simulate the growth instance-based editing operations that could be used to generate 3D
process of plants, and represents the morphological structure of tree graphics quickly and easily based on the 2D tree sketches,
plants with a 3D model. Virtual plants can help us to research which were also manually drawn by the user.
plants with a new perspective. It can not only reveal the complex A number of authors have been working on the simulation of
morphological structure and life characteristics of plants but also swaying of trees. Stam et al.[13] obtained the turbulence effect by
promote the development of computer modeling and animation filtering the white noise in the frequency domain space, and then
used the finite-element model analysis method to calculate the
Received date: 2019-02-03 Accepted date: 2020-03-02
natural frequency of the branches, to obtain the vibration effect of
Biographies: Lifeng Xu, PhD, research interests: virtual plant modeling. Email: the branches under wind load. This method simulated various
[email protected]; Zhongzhu Yang, Master Student, research interests: plant types of trees, but using the white noise filtering simulation model
modeling and virtual simulation, Email: [email protected]; Weilong made it difficult to represent the natural wind characteristics
Ding, PhD, research interests: virtual plant modeling, Email: [email protected]. faithfully. Shin et al.[14] simulated the wind field by using the 1/fβ
*Corresponding author: Gerhard Buck-Sorlin, PhD, research interests: plant
modeling. IRHS, INRAE, AGROCAMPUS OUEST, Universitéd'Angers, SFR
noise function. The spring model was used to simulate branch
4207 QUASAV, 42 rue Georges Morel, 49071 Beaucouzécedex, France. Tel: motion, while the leaves were regarded as rigid bodies. The
+33(0)241225548, Email: [email protected]. motion was the vector sum of translation and rotation. With this
 March, 2020 Xu L F, et al. Physics-based algorithm to simulate tree dynamics under wind load Vol. 13 No.2 27

method, the wind field was simulated relatively faithfully, but there A ternary formula can be used to represent the syntax of an
was not enough space for the spring model to move with the rigid L-system[25], as follows:
body, rendering the simulated effect far from elaborated. Akagi et G = < V, ω, P >
al.[15] developed a wind model using the Navier-Stokes equation, where, V is the alphabet; ω is the axiom or start word, i.e. a symbol
and used the “boundary condition map” method to analyze the or string from V+ (the set of all not empty words that can be
occlusion of the foliage. Hu et al.[16] presented an interactive generated with the alphabet) which represents the initial state of the
approach for the modelling and animation of outdoor trees from tree model; P represents the set of growth or production rules of
photographs and videos. The fast Fourier transform was used in trees, as in the following example:
that study to estimate physical parameters of branches and leaves, V : {A, B, C}
which were then applied to a simplified physics-based model to ω:A
generate animations of trees. In the simulation of a large-scale P1 : A  BC
scene, Fan et al.[17] used empirical equations from meteorology P2 : B  AC
instead of physical equations to accelerate the computing process. P3 : C  B
Different kinds of trees could be simulated by changing There are three symbols in the alphabet V. The object A is a
coefficients in the equations. Ancelin et al.[18] introduced a model start word. Three rules (P1, P2 and P3) are defined in this
of wind damage using the Transfer Matrix Method of beams. L-system rule set: A will be replaced by BC, B be replaced by AC,
This model could be used to explore how damage is related to wind and C be replaced by B. The rule set will be applied repeatedly
speed. For the uncertainty about the appropriate exponent and and in parallel to a series of strings (ω – σ1 – σ2 – …– σn), which
reliability of the conventional form, Cullen[19] reported on can be further interpreted with Turtle geometry[26], thereby
modeling of aerodynamic drag equation forms and concluded that rendering a developmental series of a virtual tree.
conventional form could be appropriate for trees. Gaffrey et 2.2.1 Surface modeling
al.[20,21] developed a complete 3D model to simulate the elastic To enhance the realism of the tree model, Bezier surfaces[27]
mechanical behavior of trees under wind and gravity. The stress were used to simulate the trunk surface. The Bezier surface was
and strain distribution at the surface of stem could be estimated defined as:
using this model. Sellier et al.[22] presented a dynamic model to m n

analyze the mechanical response of trees in turbulence. The 3D P(u, v)   Pij Bi , m (u ) B j , n (v) u , v  [0,1] (1)
i 0 j 0
description of tree architecture was integrated in this finite-element
model. Bi , m (u)  Cmi ui (1  u)mi (2)
n j
In this study, we proposed a new physics-based approach for Bj , n (v)  C v (1  v)
n
j j
(3)
calculating deformation of trees subject to different ranges of wind where, Pij is the space point range; u and v are the parameters of the
loads. Furthermore, a method to develop a 3D scene within the Bezier surface. Equations (2) and (3) are the Bernstein basis
simulation was introduced, as well as a modeling method for functions. The space grid (also called feature grid) is formed by
different types of trees. The deformation algorithms proposed connecting the two adjacent points in Pij. The matrix of the
were implemented and realized together with the modeling Bezier surface is calculated with Equation (4), as follows:
approach in a landscaping context for validation.
 P00 P01    P0 m   B0, m (v) 
 P P    P   B (v ) 
2 Materials and methods   1, m 
P(u, v)   B0, n (u ), B1, n (u ),   , Bm, n (u )  
10 11 1m

        

2.1 Simulation of the scene   
The 3D virtual scene contains three modules: a sky model, a  Pn 0 Pn1    Pnm   Bn , m (v) 
terrain model, and a radiation model. (4)
For the sky model, SkyBox[23] was used to construct the 2.2.2 Topology modeling
background module of the 3D virtual scene. Representing the 3D Ranks were used to specify the branches on an individual tree
scenes as a cube, SkyBox does not have the ability to bend texture model (as shown in Figure 1). A frustum of a cone was used as
coordinates. Thus the standard UV coordinate mapping method the basic unit for tree branches. The branch unit model was
was applied to handle the corresponding texture, together with defined with the following structure,
Cubic texture technology which was applied to simulate the details B(rt, rb, l, r, o) (5)
of the sky. Unlike normal textures, the six sides (i.e. Back, Front, where, rt represents the top radius of the branch; rb is the bottom
Up, Down, Left, and Right) of the skybox can be tiled into the radius of the branch; l indicates the length of the branch; o denotes
entire space using cube texture technology. The skybox texture is the sequence of the branch; r indicates the position of the branches
generally produced by a panorama, making the whole scene look with the same order.
continuous and realistic.
A rectangle was used to represent the ground to improve fusion
with the Terrain model into the background. Then
texture-mapping technology was used to map the ground texture
image to the Terrain model to enhance simulation of the entire 3D
scene.
The virtual sun was simulated as a directional light source.
The direction of the incident light was determined by the location Figure 1 Structure of branches with different ranks designed
of the sun, i.e. its altitude and azimuth [24]. within the topological models
2.2 Plant modeling Since the growth direction of each branch is not necessarily the
The L-system formalism was used to construct the tree model. same from one branch to the next, parameters to control growth
 28 March, 2020 Int J Agric & Biol Eng Open Access at https://www.ijabe.org Vol. 13 No. 2

directions for the branches were needed. Three turtle commands, velocity in this direction. The output wind velocity V1 could be
named RU(a1), RH(a2), RL(a3) applying geometry interpretation obtained after calculating the changes of Vn.
based on Turtle geometry[28], were used to accomplish this task.
2.2.3 Branch modeling
The tree model established with L-systems has a modular
structure that is recursively generated from simple structures
according to the growth and development rules of the tree in
question. Taking advantage of this, calculation and realization of
deformation for the branches under wind load can be made with
low cost by an algorithm representing the relationship between a. Quadrangle in the middle of the voxel b. Circle as the cross section of the
wind forces and branch bending. In this algorithm, a single as a projection of a branch segment, on stem within the voxel, with the
which S represents the direction of the arrows representing the wind
branch was divided into multiple segments that were connected
branch segment and N perpendicular to S velocities around the circle
with several joints. Each segment was considered as a rigid body.
Figure 4 Diagram of a voxel with branch segment and wind load
Therefore, the bending of the branches can be obtained by
upon the stem
calculating the deflection angle of the nodes. A schematic
diagram of the branch model is shown in Figure 2. Calculation of Vn can be transformed into calculation of a
cylinder with circular cross-section (Figure 4b). Ideal fluid
around the cylinder is composed of uniform flow and doublet flow.
The complex potentials of the uniform flow and doublet flow (W1
and W2, respectively) could be calculated with Equations (6) and
(7):
W1  Vn r (cos  i sin  ) 

M  (r  r0 ) (6)
W2  (cos  i sin  ) 
2 r 
Figure 2 Schematic diagram showing the structure of a branch
and its reaction to the wind force in the model. Points refer to M  2 r02Vn (7)
joints connecting different segments, while arrows represent the where, M is the dipole moment, and r represents the distance
wind force, which has an impact on the joint to bend the branch between P(x, y) and the origin O. The radius of the cylinder is r0.
θ is the angle between OP and the X axis. According to the
2.3 Wind
principle of superposition of potential flows, the complex potential
The wind field was simulated as a uniform flow field.
W of a new flow formed by the superposition of uniform flow and
Movement of deformable solid with the effect of fluid loads could
doublet flow could be obtained as follows:
exert an effect on the uniform flow field to change the distribution
of fluid loads. This phenomenon was named fluid-structure W = W1 + W2 (8)
interaction[29]. Interactions between wind load and trees were Then the velocity potential function Φ and the stream function
considered as a similar phenomenon. Thus, a similar approach[30] ψ can be calculated as:
 M x
  V1 x  2 x  y
could be used to simulate the effect on trees from different wind
directions and speeds. Two modules, branch and leaf, were  2 2
 (9)
designed in this study, both were considered within the calculation   V1 y  M y
of interaction between tree modules and wind load.  2 x2  y2
The virtual 3D space was divided into many voxels to calculate The velocity components at position P in the flow field are:
the influence of tree modules on the wind field. Assuming the
 
entering wind velocity is V, then it would decay to Vi after having Vx  x
travelled through i continuous voxels (Figure 3).  (10)
V y  
 y
Vn1 is obtained from Equation (10) and then combined with Vs1 to
get the output wind velocity V1.
2.3.2 Occlusion of leaves
Figure 3 Diagram of squares each represents a voxel. The wind Compared to the rigidity of the branches, the rigidity of leaves
velocity V decays to V1 after having crossed the first voxel. As is low and the variation in the wind is rather complicated. In this
before, V1 decays to V2 after having crossed the second voxel study, a simplified method was used to approximately simulate the
effect of the blade on the wind.
2.3.1 Occlusion of branches
As shown in Figure 5, the two-dimensional projection of the
Only one single branch segment was considered per voxel
blade in the direction of the wind was calculated first; then the
(Figure 4a). The particles of the fluid around a branch, which
input wind velocity V of the voxel interacts with the blade. The
latter was regarded as a cylinder, were disturbed. The input wind
output wind velocity V1 was calculated as follows:
velocity V of this voxel was decomposed into two components (Vs
 S  S
being parallel to the direction of that branch, Vn perpendicular to V1  1  sin  l V  k sin  l V , 0  k  1 (11)
the direction of that branch). Since Vs was not obstructed by the  Sv  Sv
branches, the component Vs1 of the output wind velocity in this Sl is the area of the leaf in the voxel in question. Sv represents the
direction would be equal to the component Vs of the input wind projected area of the voxel in the direction of the wind velocity.
 March, 2020 Xu L F, et al. Physics-based algorithm to simulate tree dynamics under wind load Vol. 13 No.2 29

force translation theorem to move F2 to O.

m = F2×LBO (13)
where, LBO is the center of mass of the branch segment relative to
the end point B of the branch segment. The relationship between
Figure 5 Schematic diagram of occlusion of leaves. The rectangle wind velocity and wind force can be calculated with Equation (14).
in the middle represents the projection of a leaf. θ is the angle F  Fd  Ft
between the direction of leaf projection and input wind velocity 1
Fd  Cd V 2 S (14)
2.3.3 Application 2
1
Depending on the accuracy of voxel partitioning, the number Ft  Ct V 2 S
of branches and leaves in a voxel were different. In actual 2
calculation, there might be multiple branches and leaves in one where, Cd represents the resistance coefficient; Ct, the lift
voxel. The branches were discretized (Figure 6), i.e. each branch coefficient; ρ, the air density; V, the velocity of wind; Fd, the
segment was considered as a node and then could be analyzed resistance of the branch, and Ft the drag force of the branch.
using the above-mentioned circular cylinder model. Similarly, the
leaves were also discretized into nodes. Assuming that the
number of branch segments is m and the number of leaves is n, the
output wind velocity V1 will be:
 S m S  S
V1  1   i  0 sin i li   i  0 bi V  k  i  0 sin i li V 
n n

 Sv Sv  Sv Figure 7 Force analysis for a branch segment. The force F2 was

(12) exerted by the rear segments at the B end in the case of a child
m Sbi
 i  0 Sv V  branch attached to B
2.4.2 Calculation of deformation
where, Sb is the projected area of branch segment in the direction of
wind velocity. V′ represents the average wind velocity. The end of a branch which attached to its father branch was
considered as fixed on the upper branch and the other end was free
to move. Thus, deformation of the cantilever beam from the
mechanics of materials can be applied on this matter. Since
uniformly distributed loads were acting upon the beam, the bending
moment on the full length of the beam could be described by a
function. As shown in Figure 8, it was cut off from any cross
a. b.
Figure 6 (a) Branch segments and leaves within a voxel; section, with a uniformly distributed load permanently present on
(b) Discretized nodes within a voxel, a green node indicates the leaf the left side of the section, and a reactive force FRO on the fulcrum
while a black node indicates the branch generated by the wind load q, but only a uniformly distributed load
existed on the right side of the section. The balance of all the
2.4 Dynamic simulation
forces on the right part was used to calculate the bending moment,
Deformations can be observed on branches when the wind
as follows:
blows through the trees. In doing so, the deflection angle was
1
found to be the key parameter during simulation of the processes. M ( x)   q( L  x) 2 (15)
Factors affecting the deflection angle of the branches were 2
mainly of three aspects, the wind velocity, branch diameter, and where, q is uniformly distributed load; L is the length of the beam;
branch rigidity. The rigidity of the branch is in the Young's M represents the bending moment. The deflection differential
modulus: the higher it is, the less deformation the branch would be equation is:
subjected to. As a complex system, the values of Young’s d 2W ( x)
EI   M ( x) (16)
modulus for different tree parts vary. To simplify it in the dx 2
simulation, the branches of a tree were divided into three parts: where, E represents the Young’s modulus of the branch, and I is the
trunk, branch, and twig. The range of values of Young’s modulus area moment of inertia. Equation (15) and Equation (16) are
for the parts are given in Table 1 (according to US Department of integrated in the following equation:
Agriculture[31]). dW ( x) 1
EI   q ( L  x )3  C (17)
Table 1 The value range of Young’s modulus for a tree[31] dx 6
Different parts of the tree Young’s modulus value/GPa where, C is an arbitrary constant. The constraints at the fixed end
are as follows:
trunk 5-12
dW ( x)
branch 1-5 0 (18)
twig 0.5-1 dx x 0

2.4.1 Force analysis of branches Then C is calculated as Equation (19):

As shown in Figure 7, the branch segment was connected with qL3
C (19)
its father branch at point A, and there might be a child branch at the 6
end B. The branch segment was affected by the wind force F1 and Combining (17) with (19), the following equation can be
the force F2 exerted by the rear segments at the end B if there was a obtained:
child branch attached on end B. The point of action of the wind
dW ( x) q
force F1 is at the center of mass of the branch segment. A  [ L3  ( L  x)3 ] (20)
dx 6 EI
moment of couple m is attached to the branch according to the
 30 March, 2020 Int J Agric & Biol Eng Open Access at https://www.ijabe.org Vol. 13 No. 2

Finally, the deflection angle can be calculated as:

dW ( x)
 (21)
dx xL

a. 0 m/s b. 4 m/s c. 8 m/s d. 12 m/s

Figure 9 Maximum curvatures of the branch (Koelreuteria
paniculata) bending under different wind velocities
The dynamics of two tree species (Koelreuteria paniculata and
Ulmus pumila) under four different wind loads, i.e. 0 m/s, 4 m/s,
8 m/s and 12 m/s, were simulated as the second step. The
maximum curvatures were captured and compared (as shown in
Figure 8 Diagram showing the stress analysis of cantilever beam. Figure 10). Different dynamics of bending and swaying of the
Uniformly distributed loads q are applied to the beam. FRO is a branches between two trees and among different wind velocities on
reactive force on the fulcrum generated by the wind load q the same tree could be observed. Physical properties of the two
3 Simulation Results trees are different, including tree architecture, Young’s modulus,
etc. This is the main reason why the two trees reacted differently
Using the method described above, a dynamic simulation under the same wind load. Concerning the same tree under
system with 3D tree model was implemented in the Eclipse different wind velocities, it was clear that it became more bent as
development environment using Java programming language and the wind velocity increased. Finally, the method proposed by
JOGL graphics engine. The computer hardware used in the Akagi and Kitajima[15] was applied to the same tree models, as
simulation was configured with Intel (R) Core (TM) i7-6700HQ shown in Figure 11, and the result shows that there was no
CPU, 8GB memory, and NVIDIA GeForce GTX 970M graphics significant difference between dynamic reactions from their
card. method and the algorithm proposed in this study.
Since the entire 3D space was divided into multiple voxels, the
division accuracy would have a direct influence on the simulation
results. With higher division accuracy or more voxels designed,
the results would be more accurate, but also be more time
consuming. Therefore, it was necessary to find a division model
with a relatively high computational efficiency and which at the
same time would meet the requirement of simulation accuracy.
For a tree model, the total number of discretization nodes was set to
a fixed value, which meant the number of nodes within a certain
voxel would be different when the total voxel number was changed.
For a tree model with 23,684 triangles, Table 2 shows the
relationship between the division accuracy and the calculation time
for each frame. It can be clearly observed that the calculation
took much longer when the space was divided into more voxels,
among which the 80×80×80 voxels took the longest time to Figure 10 Snapshots of the maximum bending state of the two
calculate, while the 40×40×40 only took about one third of the tree models, Koelreuteria paniculata (a1-d1) and Ulmus pumila
former, which, compared to those lower division settings, meant (a2-d2), under four different wind loads with velocities of: a1, a2:
that it gained enough accuracy but with acceptable calculation time. 0 m/s; b1, b2: 4 m/s; c1, c2: 8 m/s; d1, d2: 12 m/s, with the
Table 2 Calculation time cost within the algorithm under physics-based algorithm proposed in this study
different division accuracies
Division accuracy Calculation time/ms
5×5×5 5.6253
10×10×10 8.3641
20×20×20 13.8216
40×40×40 34.2537
80×80×80 113.2165
The simulation results were verified on three aspects: 1)
movement of a single branch under different wind velocities; 2)
movement of the same tree under different wind velocities; 3)
movement of different trees under the same wind velocity.
A branch from Koelreuteria paniculata was used for
simulation to demonstrate the results of the algorithm described for
a single branch, of which the bending was affected with four Figure 11 Snapshots of the maximum bending state of the two
settings of wind velocity values, i.e. 0 m/s, 4 m/s, 8 m/s, and 12 m/s. tree models, Koelreuteria paniculata (a1-d1) and Ulmus pumila
The Young’s modulus of the branch is 3.4 GPa, and the maximum (a2-d2), under four different wind loads with algorithm discussed
curvatures of the branch bending under different wind velocities in the paper[15]. Velocity of wind: a1, a2: 0 m/s; b1, b2: 4 m/s; c1,
are shown in Figure 9. c2: 8 m/s; d1, d2: 12 m/s
 March, 2020 Xu L F, et al. Physics-based algorithm to simulate tree dynamics under wind load Vol. 13 No.2 31

Admittedly, distribution of the wind field in a 3D space is

4 Discussion relatively complicated, not to mention movement of a tree in the
A physics-based algorithm to simulate tree dynamics of two wind. To simplify the question, we assumed that the power of the
different tree species under various wind loads was proposed in this wind within the space stayed the same, in an attempt to control
study. By considering the physical properties of the trees, i.e. calculation cost, however, simulation results were also less
branches and leaves, the wind field and forces on trees can be convincing. In order to get more realistic results, analysis of
simulated. Leaf shedding under high wind load has not yet been spatial distribution of wind field will be added in the future.
considered in our model, which should be considered in the
simulation in further study. Acknowledgements
One of the applications of the algorithm lies in tree cultivation This work was supported by the National Natural Science
and forest industry. In the forest, the strong wind can affect the Foundation of China (Grant No. 31301230, 61571400, 31471416);
growth process of the trees, resulting in structural deformation of and the Scientific Research Fund of Zhejiang Provincial Education
the trees, even affecting the structural layout of the forests and the Department (Grant No. Y201224867). The authors are grateful to
tree species diversity[32]. In urban areas, trees on both sides of a the anonymous reviewers whose comments helped to improve this
road are always at risk to be ruptured by strong winds, often paper.
causing harm and damage to people, vehicles or houses[33], all this
representing substantial financial risks for insurance companies. [References]
In a forestry context, the simulation algorithm can be used to [1] Ancelin P, Courbaud B, Fourcaud T. Development of an individual
predict storm damage or to optimize planting density and row tree-based mechanical model to predict wind damage within forest stands.
Forest Ecology Management, 2004; 203(1-3): 101–121.
orientation in a stand in certain geographical locations, and it can
[2] Sakaguchi T, Ohya J. Modeling and animation of botanical trees for
also be used to simulate and quantify the financial risks for urban interactive virtual environments. ACM Symposium on Virtual Reality
landscaping. Currently, only the morphological effects of the Software and Technology, London, United Kingdom, 1999; pp. 139–146.
trees under wind load were considered in the algorithm, while the [3] Ono H. Practical experience in the physical animation and destruction of
impacts on physiological processes of trees were not. It is also trees. Eurographics Workshop on Computer Animation and Simulation,
Vienna, Austria, 1997; pp. 149–159.
essential to combine the effects of the wind load on biomechanics
[4] Giacomo T, Capo S, Faure F. An interactive forest. Eurographics
of the trees in order to better simulate its dynamics. We will do Workshop on Computer Animation and Simulation, Vienna, Austria, 2001;
this in the near future. pp. 65–74.
Another application of our model could be in the context of [5] Cannon J, Mandelbrot B. The fractal geometry of nature. The American
ecological modelling, more particularly bioclimatology and Mathematical Monthly, 1984; 91(9): 594.
[6] Teng C, Chen Y. Image-based tree modeling from a few images with
photosynthesis models: it is well known that the rate of very narrow viewing range. The Visual Computer, 2009; 25(4): 297–307.
photosynthesis depends on the concentration of CO2 inside the leaf, [7] Aono M, Kunii T. Botanical tree image generation. IEEE Computer
which in turn depends on stomatal conductance[34]. The latter Graphics and Applications, 1984; 4(5): 10–34.
process is controlled by a number of plant and biophysical [8] Oppenheimer, Peter E. Real time design and animation of fractal plants
and trees. ACM SIGGRAPH Computer Graphics, 1986; 20(4): 55–64.
processes, amongst which features wind velocity. Detailed
[9] Shlyakhter I, Rozenoer M, Dorsey J, Teller S. Reconstructing 3D tree
information (at a fine spatial resolution) about the speed and models from instrumented photographs. IEEE Computer Graphics and
direction of air movement around the leaf would greatly improve Applications, 2001; 21(3): 53–61.
estimations of the CO2 concentration and relative humidity in the [10] Lindenmayer A. Mathematical models for cellular interactions in
leaf boundary layer, which latter serves as the base for estimating development I. Filaments with one-sided inputs. Journal of Theoretical
Biology, 1968; 18(3): 280–299.
the amount of CO2 available for photosynthesis, and which
[11] Okabe M, Owada S, Igarashi T. Interactive design of botanical trees
currently cannot be very well quantified. using freehand sketches and example-based editing. Computer Graphic
Forum, 2005; 24(3): 487–496.
5 Conclusions [12] Ijiri T, Owada S, Okabe M, Igarashi T. Floral diagrams and
inflorescences: interactive flower modeling using botanical structural
A method for simulating the movement of trees in the wind
constraints. ACM Transactions on Graphics, 2005; 24(3): 720–726.
based on physics was presented in the study. The L-system [13] Jos S. Stochastic Dynamics: Simulating the effects of turbulence on
formalism was used to construct a tree model and a whole branch flexible structures. Computer Graphics Forum, 2010; 16(3): 159–164.
was divided into multiple segments. Considering the wind load [14] Ota S, Tamura M, Fujita K. A hybrid method for real-time animation of
acting on each branch, we calculated the deflection of each branch trees swaying in wind fields. The Visual Computer, 2004; 20(10): 613–623.
[15] Akagi Y, Kitajima K. Computer animation of swaying trees based on
of the tree in the wind, and then superimposed the deflection of all
physical simulation. Computers and Graphics, 2006; 30(4):529–539.
the small segments to obtain the movement of a whole branch. [16] Hu S, Zhang Z, Xie H, Igarashi T. Data-driven modeling and animation
The experimental results show that the method is more realistic and of outdoor trees through interactive approach. Visual Computer, 2017;
greatly reduces the computation time. 33(6-8): 1017–1027.
It is true that L-systems are not necessarily the only method to [17] Fan J, Xiao S. The study of real-time animation of forest scene in wind
projection. ACM SIGGRAPH International Conference on Virtual
simulate trees with static architecture exposed to wind loads, as Reality Continuum and its Applications in Industry, Kobe, Japan, 2015; pp.
other methods of architecture reconstruction would be more 101–104.
efficient in some way. L-systems were used in this study, as it is [18] Ancelin P, Courbaud B, Fourcaud T. A population approach to study
only the first step to consider the static structure of the simulated forest stand stability to wind: individual tree-based mechanical modeling.
International Conference Wind Effects on Trees, University of Karlsruhe,
trees. Our next step will include the effect of wind load on growth
Germany, 2013.
of a tree, and eventually consider the influence on the whole [19] Cullen S. Trees and wind: A practical consideration of the drag equation
growth period and not only on structure, but also on functional velocity exponent for urban tree risk management. Journal of
physiological processes. Arboriculture, 2005; 31(3): 101–113.
 32 March, 2020 Int J Agric & Biol Eng Open Access at https://www.ijabe.org Vol. 13 No. 2

[20] Gaffrey D, Kniemeyer O. The elasto-mechanical behaviour of Douglas [27] Hu S, Zuo Z, Sun J. Approximate degree reduction of triangular bezier
fir, its sensitivity to tree-specific properties, wind and snow loads, and surfaces. Tsinghua Science and Technology, 1998; 3(2): 55–58.
implications for stability - a simulation study. Journal of Forest Science, [28] Prusinkiewicz P. Graphical applications of L-systems. Proceedings -
2002; 48(2): 49–69. Graphics Interface, 1986; pp. 247–253.
[21] Gaffrey D, Sloboda B. Modifying the elastomechanics of the stem and [29] Pei J, Yuan S, Yuan J. Fluid-structure coupling effect on periodically
the crown needle mass distribution to affect the diameter increment transient flow of a single-blade sewage centrifugal pump. Journal of
distribution: A field experiment on 20-year old Abies grandis trees. Mechanical Science and Technology, 2013; 27(7): 2015–2023.
Journal of Forest Science, 2004; 50(5): 199–210. [30] Li F. Realistic Simulation of three-dimensional trees swaying in the wind.
[22] Sellier D, Brunet Y, Fourcaud T. A numerical model of tree aerodynamic Computer Science, 2012; 39(11): 254–260.
response to a turbulent airflow. Forestry, 2008; 81(3): 279–297. [31] Concepts R. Wood handbook - wood as an engineering material.
[23] Ping S, Tao Y. Research and realization of modeling method for virtual Agriculture handbook / United States. Dept. of Agriculture (USA). No. 72.
geographic scenes. International Conference on Information Science and 2013, 1.
Engineering, 2011; pp. 2297–2299. [32] Dupont S, Pivato D, Brunet Y. Wind damage propagation in forests.
[24] Kolivand H, Rhalibi A, Sunar M, Saba T. ReVitAge: Realistic virtual Agricultural and Forest Meteorology, 2015; 214-215(3): 243–251.
heritage taking shadows and sky illumination into account. Journal of [33] Kennethr J, Nicholas H, Peterk A. Mechanical stability of trees under
Cultural Heritage, 2018; 32: 166-175. dynamic loads. American Journal of Botany, 2006; 93(10): 1522.
[25] Prusinkiewicz P, Lindenmayer A. The algorithmic beauty of plants. [34] Nikolov N, Massman, W, Schoettle A. Coupling biochemical and
New York: Springer-Verlag, 1990. biophysical processes at the leaf level: an equilibrium photosynthesis
[26] Abelson H, Disessa A. Turtle geometry. MIT Press, Cambridge, model for leaves of C3 plants. Ecological Modelling, 1995; 80(2-3):
Mass.-London, 1980. 205–235.