Volume 13, (1994) number 1, pp.

57-67

Strands, Gravity and Botanical Tree Imagery

Matthew Holton

School of Computing and Mathematics, University of Teesside, Middlesbrough, UK

Abstract
This paper presents a technique for the modelling and rendering of realistic botanical tree images. A strand
model is used that is analogous to the internal vascular structure of a tree. The model is “grown” under the
simulated influence of gravity and light. The strand densities at each branching point are used to determine
branching angles, branch lengths and branch thicknesses, taking into account stored, user definable parameters
that characterize the species of tree being modelled. These parameters address such factors as gravimorphism,
phototropism, orthotropism, plagiotropism, planartropism and phyllotaxis, and are distributed according to a
branch ordering system.
Branch segments andjoints are modelled by Bézier splines, with an assumed circular cross-section. Leaves are
made up from numbers of sample ranges from vector plane equations. The trees are rendered using a surface
sampling algorithm with a light Z buffer for shadows and autoregression textures for tree bark and grass.

Keywords: Trees, strands, gravity, branching model

1. Introduction To this end, it takes into account a wide range of factors

that may influence a tree’s appearance. Many previous
With the rapid improvements in computer graphics
tree models have been highly specific to a single tree
technology and techniques over recent years, attention
has been focused more and more on the natural world as species, or have been constrained to a finite library of
possible tree species. Many potential users may wish for
the inspiration for the work being done. The synthesis of
realistic images of natural scenes is an immensely difficult greater flexibility, where a model will allow them to
rapidly create new tree models or modify existing ones.
task, due to the fact that natural objects such as
The technique described here accommodates a high
mountains, rivers, shrubs and, above all, trees are very
degree of flexibility and is intended for a diverse range of
hard to model using conventional techniques.
possible applications.
A great deal of effort has been spent in recent years
addressing the problems of modelling natural scenes and,
2. The Strand Model
in particular, the problem of producing realistic images of
trees. These efforts have included work on fractals1-5, The essential tree model is based on the internal vascular
geometric models6, particle systems’,stochastic models8, structure of a botanical tree. All trees have a hierarchical
textured surface9,10, L systems11-13, combinatorial ana- arrangement of elongated vascular cells for the dis-
lysis14 and developmental botanical models15. These have tribution of nutrients. The concentration of these vascular
produced mixed results, chiefly due to the immense elements at any point in the branching structure is in
problem of realistically simulating the extremely complex direct proportion to the number of leaves above the
structure of a tree without incurring prohibitive expen- point. The vascular density in a branch is also pro-
diture of rendering time and memory. This difficulty is portional to the branch’s cross-sectional area, and there-
common to most natural objects, but is particularly fore its thickness. This all seems quite natural, thicker
severe with trees. The problem cannot easily be evaded as branches carry more leaves.
trees are the dominant feature of most natural scenes.
The strand model is based on the ‘pipe model’16-18, see
The objective of the technique described in this paper is Figure 1, which has been used as the basis for explaining
to produce images of trees that are as realistic as possible. the branching of rivers, bronchial passages in lungs and

© The Eurographics Association 1994. Published by Blackwell

Publishers, 108 Cowley Road, Oxford OX4 1JF, UK and 238
Main Street, Cambridge, MA 02142, USA.
58 Matthew Holton / Strands, Gravity and Botanical Tree Imagery

Certain features of the branching process have to be

governed by parameters set by the user. These parameters
are used to determine specific branching angles, overall
tree size, specific branch lengths, gravitational strength
and the strength of growth tendencies such as photo-
tropism and orthotropism. The effects of these para-
meters are distributed about the branching process
according to a system of branching orders.

2.1. Branch Ordering

Figure 1 : The ‘Pipe’ Model
The strand model makes use of two branch ordering
systems simultaneously. These two systems together
in botany as an underlying process in plant and tree provide a quite specific addressing regime for particular
architecture16, 19 . branches, making the application of branching para-
meters much more precise.
The strand model attempts to represent this process by
imagining the tree to be made up entirely of a number of
2.1.1. Gravelius Ordering
thin, primitive strands. These strands are indivisible in
themselves and run unbroken from the base of the trunk, Gravelius20 proposed an ordering method for binary
through the branching structure to an individual leaf or branching systems intended for use in the description of
bunch of leaves. The number of strands in a branch will rivers and streams. This method designates one sub-
determine that branch’s thickness and the number of its branch at a branching point as being the major one and
sub-branches, twigs and leaves. is given the same order value as the parent branch, see
Binary branching is assumed, but there is no reason Figure 2, the secondary sub-branch is assigned a value
why any rate of branching could not be accommodated. one order higher. This implies that one sub-branch may
At each branching point the parent branch will contain a be thought of as being a continuation of the parent
certain number of strands. These strands must be divided branch while the other is a departure from it.
between two sub-branches. Each sub-branch must con- This scheme fits in well with the strand model as it
tain at least one strand, or else it will simply not exist. If naturally follows that the thicker sub-branch, containing
a parent branch came to a branching point containing the most strands, is given the same order value as the
only one strand then, as a strand is indivisible, it cannot parent. The role of strands in the determination of
bifurcate and must therefore be terminated. It is at this branching angles, explained below, also fits in well with
point that leaves, fruit and other organs may be defined. this ordering scheme.
It is clear from this that the strand model is a self limiting
process. It cannot branch on and on forever. Given a
finite number of strands there will be a finite number of
branches in the tree.
S = Number of strands in the model
B = Number of branch segments in the model

The use of strands yields a number of immediate benefits.

As the strand distribution is finite, there is no need to
apply artificial recursive limits on the branching model.
The extent of the tree will arise naturally from the number Figure 2 : The Gravelius Ordering Scheme
of strands used. Numerous features of the branching 2.1.2. Weibull Ordering
model, such as branching angles, branch thicknesses,
branch lengths, gravitational effects and growth ten- Weibull ordering21 was intended as a description of the
dencies, may be controlled or influenced by the strand branching structures of lungs. This scheme is a straight-
distribution process. Increasing the number of the strands forward hierarchic method where there is no differen-
in the model increases the size and complexity of the tree. tiation between the sub-branches and they are each
This allows for a very simple approach to the rep- assigned a value one order higher than that of the parent
resentation of growth over time. branch, see Figure 3.

© The Eurographics Association 1994

 Matthew Holton / Strands, Gravity and Botanical Tree Imagery 59

For modelling purposes a global characteristic branch

thickness parameter is used which acts as a simple scaling
factor for the thickness of all branches.

Diameter of parent branch

Diameter of child branches
Characteristic branch thickness

Figure 3 : The Weibull Ordering Scheme 2.4. Branching Angles

A formula for branching angles is a particularly useful
For the purposes of the strand model Weibull ordering result of the strand model. In certain bronchial branching
acts as a useful separator for the distribution of branching models a relationship has been proposed between the
model parameters, cutting across the flow of Gravelius relative thicknesses of a pair of sub-branches and the
orders. angles at which they diverge from the path of the parent
branch18.
2.2. Branching Probabilities If two sub-branches contain equal numbers of strands
then they will be of equal thickness, and will branch from
The distribution of strands through the branching model
their parent at equal angles. Where one sub-branch
is governed by a set of branching probabilities. These
contains the vast majority of available strands and the
probabilities control the number of strands allocated to other contains the small remainder, then the smaller of
each sub-branch at a branching point. The particular
the two will branch at much the greater angle. Indeed, it
probability used depends upon both and the Gravelius may even by that the larger sub-branch appears to be a
and the Weibull orders of a branch. Each child branch continuation of the parent with no discernible deviation
most contain at least one strand. from the parent’spath. Where the distribution of strands
Branching probability (G, W) lies somewhere in-between these two cases then the
Number of strands in the parent branch branching angles may be determined from the branches
Number of strands in child branches relative thicknesses, see Figure 4. The effect of this is to
produce branching structures that have excellent space-
filling properties.
In the current model, the calculation of branching
angles is done using a simple linear expression involving
the ratio of the numbers of strands in the two sub-
branches at a branching point (7). This is quick, simple
2.3. Branch Thickness
and produces realistic results.
The most obvious and immediate product of the strand
Different tree species have different characteristic
model is a consistent description of branch thickness. The
branching angles and this is catered for in the model by a
cross-sectional area of a branch may be reasonably held
set of definable angles distributed according to both
to be directly proportional to the number of strands it
Gravelius and Weibull orders. This permits different
contains. The diameter of the branch is therefore pro-
branching angles throughout the branching structure,
portional to the square root of the number of strands.
but is usually of most use when differentiating the
The relationship between parent and child branch branching angles of branches departing the main trunk of
thicknesses is discussed at length by Mandelbrot3, in- the tree and those much higher in the Gravelius ordering
cluding the original observations made by Leonardo da scheme. For extreme ratios of strand numbers where the
Vinci on the apparent relationship between parent branch one sub-branch contains a negligible number of strands
thickness, sub-branch thickness and leaf density. The compared with the other, then the branching angle
accepted thickness relationship, for botanical as opposed calculation becomes superfluous as the results would be
to biological branching processes, is that the sum of the almost imperceptible. These situations may be avoided by
squares of the diameters of the two sub-branches is equal using a threshold ratio value beyond which the whole
to the square of the diameter of the parent branches. This characteristic branching angle is used for the bifurcation
is a simple Pythagorean equation. angle of the minor branch.

© The Eurographics Association 1994

60 Matthew Holton / Strands, Gravity and Botanical Tree Imagery

Figure 4: Brunch Thickness and Branching Angle

Branching angle (G, W) that are distributed according to the Gravelius and
Major branching angle Weibull order of the parent branch and according to the
Minor branching angle branch’s type as a major or minor path. The separate
branch lengths allow the visible length of branch in the
model to be independent of the inherited branch length.
This is a very useful feature, particularly when controlling
branch lengths in and around the main trunk of the tree.
The trunk is clearly one homogeneous process and i s
largely independent of the lengths of the branches it gives
2.5. Branch Length
rise to.
The use of strands in the determination of branch lengths
is considerably more involved and less obviously derived Characteristic length of parent branch
than for branching angles. Branch length is an extremely segment
important component of branching models. Many mod- Characteristic lengths of child branch
elling approaches in the past have struggled for realism segments
due to the difficulty in controlling the effect of branch Characteristic branch length ratios
length on the branching structure. Branch length is Active lengths of child branch segments
determined by some proportion which is applied to the Active branch length ratios
length of the parent branch. Initially, this proportion is Maximum characteristic branch length (type,
calculated by taking the ratio (9), (10) between the square G, W)
root of the numbers of strands in the sub-branches and Minimum characteristic branch length (type,
the parent branch. G, W)
Maximum active branch length (type, G, W)
Acting as a limit on this process are two separate Minimum active branch length (type, G, W)
parameter sets. These are used to differentiate two branch
length processes. The first of these i s the characteristic
branch length, a hierarchic property of the branching
structure as each branch will pass on some proportion of
its characteristic length to its sub-branches. The second
process is the active branch length. This controls the
length of the branch as it appears in the final model. The
active length of a branch is independent of the active
lengths of other branches, but does make use of the
characteristic length of the parent branch.
For every branch, a characteristic branch length and
an active branch length are calculated. These lengths are
constrained by certain pre-defined maxima and minima

© The Eurographics Association 1994

 Matthew Holton / Strands, Gravity and Botanical Tree Imagery 61

3. Gravity, Light and Growth Tendencies

The appearance of a tree has a great deal to do with its
environment. As a tree grows it is buffeted by the elements
and dragged down by gravity. The direction of sunlight
and the inherent growth characteristics of the tree itself
also contribute. Gravitational, phototropic and general
growth tendencies are incorporated in the model by
treating them as forces quantifiable as vectors, which are
used to modify branch paths.

3.1. Gravity
Figure 5: Gravitational Influences
Botanical trees may be thought of as natural archi-
tecture22-24. They grow and develop in direct conflict with
multiplier may be positive or negative, thus inducing
gravity, and must be strong enough and grow in such a
upward or downward tendencies.
way as to support a large branching structure over long
periods of time, balanced on a single stem. Within these
structural constraints, trees must also aim to distribute 3.2. Phototropism
their leaf surface area as efficiently as possible to facilitate
The influence of light on the growth of plants is well
photosynthesis.
established. Incident light on plant forms induces that
Trees are able to modify their growth by producing plant form to grow in the direction of the light. In
auxins25. These are growth stimulating hormones which botanical trees, the effect of light is distributed over many
are generated in certain areas of the tree in response to years and acts as a subtle attractor for the branching
specific stimuli. These stimuli may come from many structure, an incentive for it to take on the most efficient
external sources, including light and gravity. The effect of light capturing form.
these auxins may be to straighten a tree trunk that is
Phototropism is incorporated in the model by a vector
leaning too far to one side, or to promote growth in the
quantity that is unique to each branch it affects. The
direction of sunlight. Gravity is included in the model in
vector is calculated according to the position of the
two ways. Firstly, a gravitational effect is included that
branch relative to the tree centre. The resulting outward
acts only on the central trunk of the tree. This effect,
facing vector may be modified slightly to aim at the sky
referred to in the model as gravicentralism, operates by
within certain limits of elevation. The effect of this vector
striving to correct any drift in the trunk’s path away
is to gently force the branching structure to develop
from the central axis of the branching structure. This is
consistently outwards from the centre of the tree. The
accomplished by a vector quantity that is calculated
best results are achieved with a relatively weak photo-
according to the trunk’s position relative to the central
tropic vector.
axis. This vector is added into the trunk’sbranching path,
thus bringing it back towards the centre, see Figure 5. The
vector quantity should not be too large, a relatively gently 3.3. Orthotropism and Plagiotropism
modification of the trunk’spath is most desirable. Too
Orthotropism and plagiotropism are growth tendencies
large a correction and the trunk will appear to writhe
that are characteristic of certain tree species and are often
back and forth about the central axis. The trunk of the
variably distributed within the branching structure16.
tree is defined as being the sequence of branches with the
Orthotropism is the tendency of a branch to grow
lowest Gravelius order.
vertically upward. This is usually a feature of the tree
The second gravitational effect is gravimorphism, the trunk but may extend to higher Gravelius orders. An
general gravitational effect on all the branches of the tree. extreme example of this is the Lombardy Poplar which
The response of a branch to the gravitational force exhibits orthotropism over, practically, its entire struc-
dragging on it depends upon its thickness, its position in ture. Orthotropic effects are included in the model as
the branching structure and the species of tree it belongs vector quantities and are distributed according to both
to. Some trees have their thin, terminal branches and Gravelius and Weibull ordering. Plagiotropism is the
twigs hanging down under the pull of gravity, while complement of orthotropism. It refers to the tendency of
others have theirs turning upward. The gravimorphic a branch to grow in a horizontal direction, and may be
effect is incorporated in the model by a vertically oriented characteristic of any order of branch except the trunk.
vector. A scalar multiplier, acting on this vector, controls Plagiotropism is likewise included in the model as a
the influence of the vector on the branch in question. This vector quantity, see Figure 6.

© The Eurographics Association 1994

62 Matthew Holton Strands, Gravity and Botanical Tree Imagery

branch. The model generally uses a simple random angle

in range 0 to 360 degrees to define the position of the sub-
branch plane. The exception to this is planartropism.
Planartropism is the tendency of the sub-branches to
grow in a plane that is perpendicular to the axis/parent
plane described above, see Figure 7. This growth pattern
is incorporated in the model by simply keeping the
phyllotaxis angle constant at a relative 90 degrees to the
axis/parent plane. The effect can be toggled on or off and
is distributed about the model according to Gravelius
Figure 6: Growth Tendencies ordering only.

3.5. Branch Structure and the Writhe Coefficient

The branches themselves are constructed as cubic Bézier
curves, the control points of which are distorted by the
growth tendencies described above, see Figure 8. Included
in the distortion is a random noise value known as the
writhe coefficient. This has the effect of introducing an
element of individuality to the tree. The larger the
coefficient, the more the branch curve control points are
twisted about. The coefficient is used to create a random
vector which is included in the vector branching model
below. The initial branch vector is defined by the
phyllotaxis and branching angles and has a length of
either or depending on type.
Figure 7: Planartropism
Orthotropism strength (G, W)
Plagiotropism strength (G, W)
Phototropism strength
3.4. Planartropism and Phyllotaxis
Gravimorphism strength (G, W)
At a branching point the two sub-branches together Writhe coefficient
define a plane. It is usual for the parent branch to lie in the Orthotropism unit vector
same plane. In the branching structure of a tree, Plagiotropism unit vector
phyllotaxis refers to the angle of this plane relative to the Phototropism unit vector
plane defined by the central axis of the tree and the parent Gravimorphism unit vector

Figure 8: Branch Segment Distortion

© The Eurographics Association 1994

 Matthew Holton / Strands, Gravity and Botanical Tree Imagery 63

= Writhe unit vector limb. This operates in conjunction with the overall
b = Initial branch vector branching angle.
q = Intermediate branch vector
B = Final branch vector
There are active constraints on the limb length close to
the trunk. This is to avoid the creation of open areas. The
rapid bifurcation close to the trunk leads to a dense
branching distribution.

3.6. Example: A Fir Tree

Further active constraints are placed on major limb
The fir tree models in Figure 14 are fairly simple examples length to control the cone like shape of the tree.
of the technique. Each of the three trees shown was
generated using the same set of parameters. Differing In addition to the above, some orthotropism is applied
seed values for the randomly generated writhe vectors in to the tree trunk and some positive (upwards) gravi-
each tree ensure that no two trees will be quite alike. The morphism is applied to the branch extremities. This
essential parameters for this model may be summarised ensures that the tree grows very straight, and that the
as follows: terminating branch segments are turned, realistically,
away from the ground.

The branches departing the tree trunk are much thinner 4. Modelling and Rendering
than the trunk itself, so the vast majority of strands
should remain in the trunk. The Gravelius order 0 (tree To achieve a realistic result, it is inevitable that the tree
trunk) branching probability is therefore set high, at over model must possess some measure of the complexity of a
ninety eight percent. The remaining branching proba- real tree. This presents problems for the construction of
bilities are set to ensure a moderate (sixty percent) the 3D model in the computer. The number of strands
majority for the major sub-branch at each branching necessary to achieve the desired level of realism depends
point. upon the nature of the specific tree being modelled, but,
in general, a realistic model should contain at least 1000
s = 2500 strands. This means that the model will contain around
This number of strands will produce around five thou- 2000 branches. In practice, strand numbers in the range
sand branch segments and is representative of a mature 2000 to 8000 have proved the most satisfactory. The
tree. models produced have a pleasing, natural degree of
complexity. Each branch of the model is a curved sweep
with circular cross section and smoothly diminishing
The branching angle of branches departing the trunk is diameter. Building a model of this complexity using
set at over ninety degrees to ensure that they have an traditional polygonalization techniques would involve
initial downward sweep away from the trunk. Other prohibitive storage and processing overheads, therefore
branching angles are set at thirty five degrees to ensure a alternatives must be used.
narrow spread of branches around each departing limb.
4.1. Branches
The characteristic length of the tree trunk is set ensure All the branches in the tree model are assumed to have a
that there is little reduction at each branching point, no circular cross-section. This allows look-up tables to be
active limits are used as no alteration is required here. created of co-ordinates and unit normals for varying
Where no active limits are specified, the active length is sample rates around a circle. The branching paths are
the same as the characteristic length. defined by cubic Bézier curves, the control points of
which are extracted from the nodes of the wire frame
branching model. At the branching point, an extra joint
The remaining major branch elements are maintained at object is created using a quadratic Bézier curve. This
average reduction to match the reduction in strand curve makes use of interpolated control points lying
numbers. between the last two control points of the parent branch,
the branching point itself and the first two control points
of the sub-branch, see Figure 9. The branches and joints
The minor branches are set at a much more severe are modelled and rendered as sweeps with a circular cross
reduction to constrain them around the path of the major section. The renderer makes use of a surface sampling

© The Eurographics Association 1994

64 Matthew Holton / Strands, Gravity and Botanical Tree Imagery

be rendered very quickly and simply and are quite

accurately realistic, except when viewed at very close
quarters. For such close views a texturing solution may
be used to improve realism.

4.3. Shadows, Atmospheric Haze and Textures

To improve the realism of the final rendered result, a few
extra features are included. Shadows are generated using
a conventional shadow Z-buffer and atmospheric haze is
added by simple depth cueing. Tree bark is simulated
Figure 9: Joint Curve Control Points using an autoregressive stochastic texture, the parameters
of which may be subtly altered to vary the effects
produced. The bark texture is created as two separate bit
routine that samples using the information from the maps and by interpolating between them along the length
cross-section look up tables. The sampling rate is de- of each branch segment a smooth continuity of pattern is
termined by the number of pixels being addressed by the maintained, see Figure 12. A “softer” texture is used on
full extent of the branch segment. A bounding box is used the flat terrain to simulate grass.
to fix maxima and minima for the range and the sampling
is derived as being the minimum number of samples along
5. Conclusion
the length of the sweep and around the cross-section that
satisfy the Nyquist limit, i.e. sampling at twice the highest The system used to produce the images included in this
frequency at pixel resolution. paper was implemented on a Hewlett Packard Apollo 400
and requires around three hours to render a scene
The geometry of the branching point is determined by
containing five trees each with 6000 branch segments and
the number of strands in each sub-branch. An equal
around 20000 leaves. Despite the large number of
number of strands will create a smooth merging of parent
parameters used to define a tree, new tree models can be
and both sub-branches, see Figs 11 and 12. Unequal
created quite quickly by addressing large numbers of
strand numbers will cause the major sub-branch to
parameters simultaneously via their Gravelius and
appear to be a continuation of the parent branch with the
Weibull orders. The creation of a specific tree type
minor sub-branch as an off-shoot.
requires some practice and the ability to isolate, visually,
those elements in the target tree that are addressed by
4.2. Leaves each parameter. A new tree model typically takes around
twenty to thirty minutes to create, from scratch.
Leaves are positioned on the terminating branches of the
tree structure. At these points, each branch contains only The parameters used to define a specific tree model lend
one strand and is at its thinnest. A small library of leaf themselves to various forms of procedural manipulation.
types and patterns is included in the model and may be At present, one of the most interesting possibilities
accessed by the user. The leaf shape is a quadrilateral or involves treating the parameter sets as if they were genes,
combination of quadrilaterals which are stored as sample where species would be “evolved” by repeated mutation
areas of vector plane equations, see Figure 10. The leaf is and/or mating. More mundane operations on the par-
rendered by sampling across this area using sample look- ameter sets would include tools for the rapid re-definition
up tables in a similar manner to the branch rendering of a tree’s structure, making the technique more user-
technique described above. friendly.
The result is a trade of speed for detail. The leaves may The tree model described here has been successful in
producing images of trees that achieve a quite satisfying
degree of realism. Refinements in the rendering tech-
niques, particularly in the surface sampling area, have
yielded rendering times that permit the model to be of
practical use in animation. There are inevitable problems
in producing scenes containing upwards of a million
branch segments. The applications of this model would,
most probably, include architectural and landscape
visualisation, animation and, when the processing tech-
nology becomes available, flight simulation and virtual
Figure 10: Examples of Library Leaves reality.

© The Eurographics Association 1994

 Matthew Holton / Strands, Gravity and Botanical Tree Imagery 65

13. P. Prusinkiewicz and A. Lindenmayer, The Algo-

References
rithmic Beauty of Plants, Springer Verlag (1990).
1. S. Demko, L. Hodges and B. Naylor, “Construc-
14. X.B. Viennot, G. Eyrolles and N. Janey, “Com-
tion of Fractal Objects with Iterated Function
binatorial Analysis of Ramified Patterns and Com-
Systems”, Computer Graphics, 19(3) (1985).
puter Imagery of Trees ”, Computer Graphics, 23(3),
2. M.D. Holton, “The Quick Computer Generation of pp. 31-40 (1989).
Realistic Tree Images”, Proc. Computer Graphics
15. P. deReffye, C. Edelin, J. Francon, M. Jaeger and
90, Blenheim Online, pp. 173-192 (1990).
C. Puech, “Plant Models Faithful to Botanical
3. B.B. Mandelbrot, The Fractal Geometry of Nature, Structure and Development ”, Computer Graphics,
W.H. Freeman (1982). 22(4), pp. 151-158 (1988).
4. P. Oppenheimer, “Real Time Design and Ani- 16. F.E. Fritsch, and E. Salisbury, Plant Form and
mation of Fractal Plants and Trees”, Computer Function, G. Bell & Sons Ltd (1965).
Graphics, 20(4), pp. 55-64 (1986).
17. M.H. Zimmerman, P.B. Tomlinson, J. Leclaire and
5. A.R. Smith, “Plants, Fractals and Formal F.L.S. Leclaire, “Vascular Construction and De-
Languages”, Computer Graphics, 18(3), pp. 1-10 velopment in the Stems of Certain Pandanaceae”,
(1984). Bot. J. Linn. Soc., 68, pp. 21-41, (1974).
6. M. Aono and T.L. Kunii, “Botanical Tree Image 18. N. Macdonald, Trees and Networks in Biological
Generation”, IEEE CG&A, 4, pp. 16-34 (1984). Models, J. Wiley & Sons (1983).
7. W.T. Reeves and R. Blau, “Approximate and 19. M.D. Holton, “StrandTracking and Botanical Tree
Probabilistic Algorithms for Shading and Rendering Imagery”, Proc. Eurographics UK, pp. 97-1 10
Structured Particle Systems”, Computer Graphics, (1992).
19(3), pp. 313-322 (1985).
20. H. Gravelius, Flusskunde, Goschen, Berlin, (1914).
8. A. Fournier and D.A. Grindal, “The Stochastic
21. E.R. Weibull, Morphometry of the Human Lung,
Modeling of Trees”, Graphics Interface 86, pp.
Springer Verlag (1963).
164-172 (1986).
22. T.A. McMahon, and R.E. Kronauer, “Tree Struc-
9. J. Bloomenthal, “Modeling the Might Maple”,
tures: Deducing the Principle of Mechanical De-
Computer Graphics, 19(3), pp. 305-31 1 (1985).
sign”,J. Theor. Biol., 59, pp. 443-466 (1976).
10. G.Y. Gardner, “Simulation of Natural Scenes
23. F. Halle, R.A.A. Oldeman and P.B. Tomlinson,
Using Textured Quadric Surfaces”, Computer
Tropical Trees and Forests, Springer Verlag (1978).
Graphics, 18(3), pp. 11-20 (1984).
24. P.B. Tomlinson, “Tree Architecture”, American
11. P. Prusinkiewicz and J. Hanan, Lindenmayer
Scientist, 71(March-April), pp. 141-149 (1983).
Systems, Fractals and Plants, Springer Verlag
(1991). 25. M.H. Zimmerman, and C.L. Brown, Trees, Struc-
ture and Function, Springer Verlag (1971).
12. P. Prusinkiewicz, A. Lindenmayer and J. Hanan,
“Developmental Models of Herbaceous Plants for

(See pages 66 and 67 for colour figures 11,12,13 and 14).

© The Eurographics Association 1994

66 Matthew Holton / Strands, Gravity and Botanical Tree Imagery

Figure 11: Joint Geometry

Figure 12: Rendered Joint with Texture

© The Eurographics Association 1994

Matthew Holton / Strands, Gravity and Botanical Tree Imagery 67

Figure 13: Maple

Figure 14: Firs

© The Eurographics Association 1994

5-2