Life-time Selection and
Self-Organization in Tree Growth
An Artificial Life Paper
|
|
Simple rules of growth can lead to interesting patterns
in the way trees grow and branch. No two trees grow exactly alike. Within
any tree some branches are thicker and some form odd shapes and
wiggles. The leaves may form an umbrella-like cluster to get the most
out of sunlight. The way a tree grows and branches is determined partly
by its genome, but mostly by its interactions with the environment. The
growth of such a tree is self-organized, and the shape it takes is an
emergent property of the tree-environment system. The technique we
present here may be classified as a morphogenetic algorithm which is
based on complex interactions. It does not need to resort to
probabilistic methods to generate variety. Variety stems from the
complexity of the tree's environment. A slight variation in the
position of a shade may cause an entire branch to disappear or new
branches and turns to appear. Two trees growing side by side affect
each other's growth indirectly. The branching patterns result from
complex interactions, branch efficiencies and a principle we call
life-time selection. Unlike natural selection which works over
evolutionary time, life-time selection involves applying a
"survival of the fitter" rule to the components of the
growing tree, and selecting those that are successful, for reproduction
(branching). This simple principle, combined with the complexity of the
environment, is able to generate complex branching patterns akin to
those found in nature.
February 1997
Rekesh John, Senior Project Officer: Innovative
techniques in computer science
Indian Institute of Technology
Madras, India.
C.R.Muthukrishnan
Dept. of Computer Science and Engineering
Indian Institute of Technology
Madras, India.
|
|
Introduction
|
|
Simple rules of growth can lead to interesting patterns
in the way trees grow and branch. No two trees grow exactly alike.
Within any tree some branches are thicker and some form odd shapes and wiggles.
The leaves may form an umbrella-like cluster to get the most out of
sunlight. The way a tree grows and branches is determined partly by its
genome, but mostly by its interactions with the environment. The growth
of such a tree is self-organized, and the shape it takes is an emergent
property of the tree-environment system. The technique we present here
may be classified as a morphogenetic algorithm which is based on
complex interactions. It does not need to resort to probabilistic
methods to generate variety. Variety stems from the complexity of the
tree's environment. A slight variation in the position of a shade may
cause an entire branch to disappear or new branches and turns to
appear. Two trees growing side by side affect each other's growth indirectly.
The branching patterns result from complex interactions, branch
efficiencies and a principle we call life-time selection. Unlike
natural selection which works over evolutionary time, life-time
selection involves applying a "survival of the fitter" rule
to the components of the growing tree, and selecting those that are
successful, for reproduction (branching). This simple principle,
combined with the complexity of the environment, is able to generate
complex branching patterns akin to those found in nature.
Back to Index
|
|
Basics
|
|
The components of the artificial tree are its branches
and leaves. A branch or a leaf interacts only with the components that
are connected to it. Sunlight is the only criteria used for growth.
Gravity, phototaxis, and other factors are
not considered so as to make the simulation easier to study. The idea
is not to create a visually perfect and attractive tree. Unlike the
work described in [1] this is an attempt to test the effect of a primary criteria, namely sunlight and shade, on
the structure of a growing tree when simple rules are applied to its
components.
The rules involved are the following:
1.
Only leaves process sunlight. Every leaf adds
a contribution of energy to its connected branch.
2.
The energy acquired by a branch is the sum
total of the energies it receives from its sub-branches/leaves. The
more energy that a branch processes, the longer and thicker it becomes.
3.
A branch has an associated efficiency value,
which is the ratio of the energy it actually processes to that it could
have processed at best (with its sub-components also at their best
efficiency). A leaf is at its peak efficiency when fully exposed to
light. Thus the efficiency of a branch is defined recursively. When the
efficiency of a branch falls below a threshold for some specified time,
it dies. Branching takes place only if the efficiency is above a
specified branching threshold.
Back to Index
|
|
Simulation
|
|
We used the OpenInventorTM programming
system on Silicon Graphics workstations to model the artificial trees.
The branches are idealized as long, thin cylinders. The leaves are
idealized as flat discs. A tree data structure is used to represent
connectivity, with nodes storing properties of the individual
components. In this simulation, branches that once formed sub-branches
do not branch again. Simple branching rules are used to define how a
branch forms sub-branches. Between consecutive branching, a simple form
of phylotaxis is used by rotating the
sub-branches about its mother branch through an angle.
The energy contribution of a leaf is dependent on its
orientation and any obstacles in the environment that shade it from
sunlight. This contribution is calculated as follows. Rays are traced
along an inverted vertical cone with its apex at the center of the leaf
surface. A ray originates from the apex at the leaf and is sent out
into the "sky". The set of rays so traced are limited by the
three dimensional angle of the cone. Each ray has associated with it,
some quanta of energy. The more slant a ray
has from the vertical, the less energy it contributes. If a ray meets
any object in the world (including the leaf itself, depending on its
orientation), then its contribution is taken as nil. The energy
contribution of a leaf in a single iteration of the simulation run is
computed as the sum of energies of all the rays which did not meet any
obstacles in the world.
Figures 1 through 14 show a tree growing in the shade.
The area to the right is completely blacked out from sunlight. In this
simulation, whenever branching occurs, three new sub-branches are
formed. The angle that a sub-branch makes with its mother branch is set
to 40 degrees. The energy of a vertical ray is set to one energy unit.
The cone angle is 120 degrees at the apex. About 25 rays are traced per
leaf through the cone. The cut-off efficiency level for a branch to die
is set to 30% of the maximum efficiency possible. Even if a branch's
efficiency is below the cut-off, it is allowed to survive for a few
more iterations of growth before it is pruned. The thickness of a
branch is approximated based on that of its sub-branches, such that the
sum of the cross-sectional areas of the sub-branches equal that of
their mother branch. The length of a branch is increased by 5% for 100%
branch efficiency.
It can be seen that the tree "bends" without incorporting any phototaxis
mechanism. Note that the bends here are restricted in their geometric
configuration by our graphics implementation and should actually be
thought of as curves. Variation in thickness along various paths
through the tree can be observed. Some paths show sharp turns as well.
Clusters of leaves can be seen, each cluster forming an umbrella over
the branches (fig 17). Depending on variations in the parameters and
branching rules, the resulting patterns vary, but share the above
properties. Figures 16 shows two trees growing next to each other. The
tree on the right was grown under a shade as before. The one on the
left is younger than the other by 20 iterations of growth. It can be
seen that the younger tree also "bends" and forms similar
branching patterns. Figure 15 shows where the branches of the two trees
come together and spread out.
Back to Index
|
|
Conclusions
|
|
Here we see that complex branching patterns can be
induced by a simple set of rules that are applied to the components of
the tree. The pattern is dependent on mutual interactions between the
components and the environment as well as on life-time selection. The
pattern dynamically modifies itself based on these interactions
(self-organization) and leads to emergent properties (bends, clusters,
umbrellas, varying thickness paths). If two trees are grown side by
side, they affect each other's growth indirectly. Like in most complex
systems, the best way to predict how exactly a cluster of such trees
will grow will be to actually run the system and observe!
Back to Index
|
|
References
|
|
[1] Prusinkiewicz, P., Lindenmayer, A. The Algorithmic Beauty of Plants. New York:
Springer-Verlag 1990
Back to Index
Back to Index
|
|
|
|