Incorporating Graph Automata into Plant Growth Simulation with Nutrients Transport

Lingqiu ZENG, Hongchun QU, Qingsheng ZHU, Youlan WANG

Incorporating Graph Automata into Plant Growth Simulation with

Nutrients Transport
1
Lingqiu ZENG, 2Hongchun QU, 1Qingsheng ZHU, 3 Youlan WANG
*1
College of Computer Science, Chongqing University, Chongqing 400044, China,
zenglqcqu@gmail.com
2
Key Laboratory of Network control & Intelligent Instrument (Chongqing University of Posts
and Telecommunications), Ministry of Education, Chongqing 400065, PR China,
quhc@cqupt.edu.cn
3
Chongqing Electric Power College, State Grid, Chongqing 400053, PR China
doi: 10.4156/jdcta.vol4.issue6.2

Abstract
In this paper we model the growth of living plants in a discrete approach, the graph automata based
plant model representing each metamer of the plant as node of the automata. The allocation and
transport of carbon, water and nitrogen among metamers, as an emergent property, resulted from the
strategy of local pressure equilibrium between adjacent neighbors using discrete pressure-flow kinetics
which were found to be suitable to describe mass flow in discrete space. Simulation results show that
the graph automata based parallel approach can deal with carbon transport and allocation effectively.

Keywords: Tree simulation, Discrete approach, Carbon transport, Automata

1. Introduction
The function of the nutrients transport in plant vasculature (both phloem and xylem) is one of the
most important aspects of interconnections among the individual plant components (i.e., the metamer)
[1]. The carbon transport is of fundamental importance to the survival of all higher plants [2].
Regarding the nutrients transport within plant, Münch pressure-flow hypothesis [3] has been regarded
as the general accepted model based on hydrodynamics principles. The Münch hypothesis is currently
understood that all vascular plants use this mechanism [4], or slight variants of it [5]. In past decades, a
large number of plant functional models tried to simulate the matter transportation within plant in
continuous manners, e.g., by solving partial differential equations. However, the mathematical
expression of continuous fluid properties itself is an approximation, which requires discrete method
employed by individual entities in computer simulation. In addition, digital computers handle discrete
systems in a naturally way, the mapping between discrete computational model and plant system can
be simpler and easier to understand, especially to those who are not plant physiologists.
From cell biology perspective, the long-distance transport behavior between roots and shoots organs
can be decomposed into nutrients interchange between direct neighboring cells which join into tubes
that form the vascular system of higher plant [2]. Such cell-to-cell based local interactions give rise to a
highly complex, dynamic system of nutrients transportation network in holistic plant level. These
locally adaptive behaviors work efficiently and intelligently without any central control mechanism.
In this paper, we take advantage of the discrete, parallel and local features of graph automata (GA) [6]
instead of complicated partial differential equations to mimic the long-distance transport within a plant. GA
is a special extended class of a Cellular Automata (CA). In GA, lattice structure is no longer necessary, but is
projected onto a planar graph (the discrete metamer structure of a plant) with each metamer having a
different number of neighbors according to the topology of the planar graph (Figure 1). From the functional
perspective, it inherits almost all advantages of a CA [7]. Moreover, 3D graphic and corresponding
deformations due to internal physiological status have been integrated into individual metamer.
The paper is structured as follows: first nutrients transport modeling for plant growth simulation,
including plant structure description and discrete pressure-flow model, is introduced in Sections 2. In
Section 3, the graph automata used in this work is presented. Section 4 is dedicated to the simulation
experiments of the graph automata based virtual trees. The 3D simulation results of the simulated tree
are also presented in Section 4. Finally, conclusions and lines for future research are presented.

18
 International Journal of Digital Content Technology and its Applications
Volume 4, Number 6, September 2010

Figure 1. The schematic description of plant branching network on different scales. (a) Plant consists of
discrete growth units identified by arabic numerals; (b) Growth unit consists of discrete metamer
identified by arabic numerals plus capital letter; (c) Metamer consists of lateral components (including
axillary bud, leaf and fruit) and internode; (d) Detailed geometric structure of the basic unit for nutrients
transport, the internode, which possess two channels: the xylem and the phloem.

2. The discrete nutrients transport model

2.1. Plant structure representation

Although the detailed discussion of botany and horticulture are beyond the scope of this paper, we
give a brief introduction to primary botanical entities of plant structure to keep the integration of this
paper and make it more users friendly. As reviewed by Barthelemy and Caraglio [8], plant is made up
of the repetition of many similar components at different scales, e.g. axes, growth units and metamers
as well as internodes, as shown in Figure1. Following are definitions of primary botanical entities and
markers getting involved in this paper.
(1) Node: segment of stem where leaves and lateral buds are attached.
(2) Internode: section of a stem between two nodes.
(3) Metamer: a set of organs made of leaves, buds, fruits and one internode.
(4) Growth unit: noted GU, is an elementary portion of an axis. It is a succession of metamers built up
between two resting phases, e.g. the elongation growth season.
In this work, we consider a plant as a Dynamic (growing) Branching Network (DBN) composed of
discrete components (metamers). For simplicity concern, we will focus on nutrients transport in
internodes, instead of lateral organs of metamer (such as axillary buds, leaves and fruits, etc.). These

19
 Incorporating Graph Automata into Plant Growth Simulation with Nutrients Transport
Lingqiu ZENG, Hongchun QU, Qingsheng ZHU, Youlan WANG

lateral organs attached to internode are regarded as source or sink components to perform nutrients
loading or unloading from xylem or phloem existed in internode, e.g., photosynthate (carbon) are
loaded from leaf to phloem and water are absorbed from soil to xylem.
DBN is represented by a triple consisting of a metamer set (nodes description) Nmet , a metamer
connectivity (topological description) Ctop and a set of evolutionary rules Pctrl controlling the dynamics of
DBN. Corresponding to the plant structure in the real world, Nmet is a partially ordered set of l metamers:
Nmet={m1, m2,…ml}, for any i<k, mk cannot be the parent of mi. Ctop enumerates all directed connections
between any two metamers in Nmet:Ctop = {conij(mi, mj)|mi∈Nmet, mj∈Nmet, mj grows from mi}.
For each metamer mi, two vectors of attributes are used to describe the metamer’s physiological
status and geometry: XphyA={Sp,Nx,Wp,Wx}, XgeoA={Pos(x,y,z), Vol(Rx,Rp,L)}. The former represents
nutrients content of current internode: carbon or sugar content (Cp) in phloem, nitrogen content (Nx) in
xylem, water content in phloem (Wp) and xylem (Wx), respectively. The latter depicts the metamer’s
absolute 3D position in Cartesian coordinate system, as well as volume information of its internode,
where Rx, Rp and L denote the radius and length of xylem and phloem, respectively.
For each connection con ij(mi, m j) (nonzero elements in adjacency matrix Aij of DBN), a set of
attribute vectors X topA={ORI ij[H,L,U]} picturing the relative rotation angle from parent metamer
to its direct children are attached. P ctrl refers to a rule set for controlling the branching pattern
of DBN during simulation. The rule extraction method as well as its implementation was
detailed in our previous work [9].

2.2. Discrete Pressure-flow Model

Considering the nutrients transport within plant, Münch pressure-flow hypothesis has been regarded
as the general accepted model based upon hydrodynamics principles. This hypothesis assume that the
loading process of photosynthates into the sieve-tube across the plasmalemma generates a high local
solute concentration but relative lower water potential, therefore, the water diffusion process from
xylem to phloem resulting in a high turgor pressure within the sieve-tubes of the source area, while
solute unloading at the sink area and consequently water osmosis generates a low hydrostatic pressure.
The hydrostatic pressure gradient between source and sink drives the bulk flow of phloem sap between
the source and sink through the sieve-tube conduit. The Münch hypothesis is currently understood that
almost all vascular plants use this mechanism, except little special cases.
Long-distance transport in xylem and phloem using pressure-flow hypothesis are decomposed into
nutrients interchange between direct neighboring internodes. For each internode, the xylem segment
contains some amount of water (Wx mol) and nitrogen ions (Nx mol), while the phloem segment
contains some amount of water (Wp mol) and carbon (Sp mol).
In an approximate manner, we compute the hydrostatic pressure in xylem segment as:

ε ⋅ (Wx + N x )
Px = = ε ⋅ (Wx + N x ) / Vx , (1)
π Rx2 L

where Vx is the volume of xylem segment in current internode. The water potential in xylem is:

Ψ x = Px + Π x , (2)

where Πx is the osmotic potential of solution of nitrogen ions computing as:

Nx
Π x = −R ×T × = −R ×T × ρN , (3)
Wx

where R is the universal gas constant and T is the absolute temperature, and ρN is the nitrogen
concentration.
Similarly, the hydrostatic pressure in phloem segment is:

20
 International Journal of Digital Content Technology and its Applications
Volume 4, Number 6, September 2010

ε ⋅ (W p + S p )
Pp = = ε ⋅ (W p + S p ) / V p , (4)
π R p2 L

where Vp is the volume of phloem segment. For the sake of simplicity, we ignore the thickness between
xylem and phloem. The water potential in phloem is:

Ψ p = Pp + Π p , (5)

where Πp is the osmotic potential of solution of sugar, it is computed as:

Sp
Π p = −R ×T × = −R × T × ρS , (6)
Wp

where and ρS is the sugar concentration. The water osmosis between xylem and phloem is determined
by the difference of water potential, according to Fick’s first law of diffusion, the water content
exchanged during diffusion process is:
1
Qw, xp = (Ψ x − Ψ p ) , (7)
rxp

where rxp is the apoplastic pathway resistance between xylem and phloem.
The turgor pressure difference between any connected internodes results in the bulk flow between
them. In xylem, the solution flow exchanged between mi and mj due to the trend of pressure equilibrium is
given:
1
Qx ,ij = ( Px ,i − Px , j ) , (8)
rx ,ij

where rx,ij is the average resistance of xylem segment between internode mi and mj. The same does for
phloem:
1
Q p ,ij = ( Pp ,i − Pp , j ) , (9)
rp ,ij

where rp,ij is the average resistance of phloem segment between internode mi and mj .
In each simulation step, a root (underground) can gain some amount of water through xylem
segment to compensate the water losing due to transpiration. The amount of water uptake (loading) per
simulation step can be captured by:
Vx / ε , (10)
I wx =
1 + exp(Vx / ε − τ ⋅ ξ ⋅Wx )

where Vx/ε specify the maximal amount of water a xylem segment can contain. τ and ξ are metamer age
(how many simulation steps it has been experienced) and soil moisture content, respectively. The water
evaporation (unloading) of shoot internodes can be specified by:

λWx , (11)
Owx =
1 + exp(λWx − ζ )

where ζ is the intensity of solar radiation, and λ is the user defined unloading coefficient scale.
The source of sugar in loading process comes from leaf photosynthesis, which is impacted by four
factors: the maximal volume of sugar storage (Vp/ε), the intensity of solar radiation (ζ ), current sugar
content (Sp) and the age (τ) to each metamer. The amount of sugar produced by leaf is calculated by:

21
 Incorporating Graph Automata into Plant Growth Simulation with Nutrients Transport
Lingqiu ZENG, Hongchun QU, Qingsheng ZHU, Youlan WANG

Vp / ε
I sp = , (12)
1 + exp(V p / ε − τ ⋅ ζ ⋅ S p )

Unloading of sugar usually appears on sink organs, such as roots, fruits, which consume sugar to
support their growth and development. We use the following equation to compute the amount of sugar
consumption per simulation step:
λS p , (13)
Osp =
1 + exp(λ S p − τ )

3. The graph automata model

After represent the pressure-flow model as discrete form, where the complex vascular system are
replaced by a large number of interconnected internodes, then, we therefore can utilize a powerful
discrete computational model, the graph automata, by taking advantage of its simple, local and parallel
features to mimic solution flow dynamics in a discrete way. GA can be regarded as a special class of
cellular automata CA. Its definition is extended from regular CA on cells topological distribution,
neighborhood structure as well as their state space.
In the proposed GA model, the regular lattice structure is no longer necessary, but is projected onto
a planar graph, where each nodes have different number of neighbors according to graph topology (we
substitute node for cell hereafter). We use the aforementioned DBN to represent the planar graph as a
backbone of GA lattice. Each primary nodes of DBN, metamers act as basic unit of GA. For metamer
mi, its neighbors are defined as direct connected metamers: Γ(i)={j| Aij≠0}, including both parent and
children metamers. Moreover, the state space of node in GA is extended to real number space in our
model. In order to run rapidly in the large-scale dynamical behavior simulation, CA cells usually have
only an integer state space (several bits). However, there is no intrinsic reason why CA-like
computational model should not have state values characterized as one or even several continuous-
valued numbers. There are four states for each node: water content Wx ,Wp, nitrogen content Nx and
sugar content Sp in xylem and phloem segment, respectively. They are all real numbers indicating
current nutrients statues of internode.
In each simulation step, individual node computes its states according to its current status as well as
its neighbors'. The nutrients flows are calculated from the pressure difference between node i and its all
neighbors Γ(i), because any fluids possess the tendency to pressure equilibrium in a closed space.
Moreover, we assume that the fluids are incompressible, which requires volume conservation for every
internode. This means that the sum of the fluxes in xylem and phloem segment of internode should
vanish:
∑ Qx,ij − Qw, xp = 0 ,
j∈Γ ( i )
(14)

∑Q
j∈Γ ( i )
p ,ij + Qw, xp = 0 , (15)

where Qw,xp, Qx,ij and Qp,ij are pure water flow between xylem and phloem, flows in xylem and phloem,
respectively.
Based on the principle of volume conservation (equations14,15), the pressure value for xylem and
phloem segment of each internode are updated so that this principle is satisfied in every internode:

1 1 ,
Px ,i (t + 1) = [ ∑ r
j∈Γ ( i ) x ,ij
Px , j (t ) + Qw, xp (t )] / ∑ r
j∈Γ ( i ) x ,ij
(16)

1 1 ,
Pp ,i (t + 1) = [ ∑
j∈Γ ( i ) rp ,ij
Pp , j (t ) − Qw, xp (t )] / ∑
j∈Γ ( i ) rp ,ij
(17)

where t and t+1 denote the current and the next simulation step, respectively.

22
 International Journal of Digital Content Technology and its Applications
Volume 4, Number 6, September 2010

Consequently, the nutrients content in xylem and phloem segment should be updated according to
the modification of pressure. Based on equations (1-6), we get new value of water, nitrogen and sugar
content:
1
Wx ,i (t + 1) = Wx ,i (t ) − {Px ,i (t ) − Pp ,i (t ) + R × T × [ ρ S ,i (t ) − ρ N ,i (t )]} , (18)
rxp ,i

1
W p ,i (t + 1) = W p ,i (t ) + {Px ,i (t ) − Pp ,i (t ) + R × T × [ ρ S ,i (t ) − ρ N ,i (t )]} , (19)
rxp ,i

1 [W (t + 1)] , 2
(20)
N x ,i (t + 1) = {Px ,i (t + 1) × Wx ,i (t + 1) − x ,i }
R ×T Vx ,i

1 [W p ,i (t + 1)]2
S p ,i (t + 1) = {Pp ,i (t + 1) × W p ,i (t + 1) − }. (21)
R ×T V p ,i

4. Simulation results

Figure 2. Graphical representation of plant growth dynamics based on the graph automata based
nutrients transport model. The age (number of iterations) of each plant is 20 (a), 40 (b), 80 (c), 120 (d)
and 160 (e), respectively.

In order to test the graph automata based tree simulation model with nutrients transport, we
developed a software system of graphical representation. It can produce pictures of plant branching
structure and simulate their internal behavior (i.e., the nutrients transport). This software is a C++
based application including a simple text editor for preparing parameters and a window for viewing the
progress of the simulation schematically. This view may be updated after each simulation step, helping
the user observe results, errors and debug the system. Moreover, this software allows the use of
external functions which make it possible to extend the flexibility of the arithmetic-based control
mechanism (i.e., the numerical variables and arithmetical rules).

23
 Incorporating Graph Automata into Plant Growth Simulation with Nutrients Transport
Lingqiu ZENG, Hongchun QU, Qingsheng ZHU, Youlan WANG

Figure 3. Simulated dynamic profiles of carbon concentration along the phloem channel of
a plant with complex branching structure.

Figure 2 shows an example of graphic representation of plant growth dynamics based on the graph
automata based model. The age (number of iterations) of each plant is 20 (a), 40 (b), 80 (c), 120(d) and
160 (e), respectively. In this example, internodes with different carbon concentration are distinguished
from each other using different colors, as shown in Figure 3.

5. Conclusion
In this work we model the plant growth dynamics integrating nutrients transport using a discrete
approach, a new variant of graph automata, which takes advantage of techniques of intelligent agents,
discrete pressure-flow theory as well as specific modular design pattern, etc. In addition, we also
developed a software system of graphical representation integrating the interpreter cpfg. It can render
agents (metamers) to produce pictures of plant branching structure and simulate the nutrients transport.
Simulation results show that our model can effectively simulate the plant branching structure and its
growth dynamics driven by internal nutrients allocation and transport.

6. Acknowledgements
This research is supported by the National Natural Science Foundation of China (60773082), the
National ‘863’ Hi-Tech Research and Development Project of China (2006AA10Z233).

7. References
[1] Room PM, Maillette L, Hanan JS, “Module and Metamer Dynamics and Virtual Plants”, In
Advances in Ecological Research, vol.25, no. 25, pp.105-157, 1994.
[2] Taiz, L., & Eduardo, Z., Plant Physiology, Fourth Edition: Wadsworth Publishing Co Inc., USA, 2004.
[3] Minchin, P.E.H., M.R. Thorpe and J.F. Farrar, “A simple mechanistic model of phloem transport
which explains sink priority”. Journal of Experimental Botany, vol.44, pp.947-955, 1993.
[4] Minchin, P.E.H. and A. Lacointe, “New understanding on phloem physiology and possible
consequences for modelling long-distance carbon transport”, New Phytologist, vol.166, pp.771–
779, 2005.

24
 International Journal of Digital Content Technology and its Applications
Volume 4, Number 6, September 2010

[5] Marcelis, L.F.M. and E. Heuvelink, Concepts of Modelling Carbon Allocation Among Plant
Organs. In: Workshop on Functional–Structural Plant Modelling in Crop Production, Vos, J.,
L.F.M. Marcelis, P.H.B. DeVisser, P.C. Struik, J.B. Evers (Eds.), ISBN: 978-1-4020-6033-5, pp:
103–111, Wageningen, Netherlands, 2006.
[6] Nichitiu, C., & Remila, E., “Simulations of Graph Automata”, In Proceeding of the Satellite
Workshop on Cellular Automata (MFCS’98), pp.330-342, 1998.
[7] Marr, C., & Hutt, M. T., “Topology regulates pattern formation capacity of binary cellular automata
on graphs”. Physica a-Statistical Mechanics and Its Applications, vol.354, pp.641-662, 2005.
[8] Barthelemy, D., & Caraglio, Y., “Plant architecture: A dynamic, multilevel and comprehensive
approach to plant form, structure and ontogeny”, Annals of Botany, vol.99, no.3, pp.375-407, 2007.
[9] Hongchun Qu, Qingsheng Zhu, Mingwei Guo and Zhonghua Lu, “An Intelligent learning
approach to L-grammer Extraction from Image Sequences of Real Plants”, International Journal on
Artificial Intelligence Tools, vol.18, no.6, pp.905–927, 2009.

25