2016 IEEE International Conference on Functional-Structural Plant Growth Modeling, Simulation, Visualization and Applications

Pattern-Oriented Modelling of Plant Architecture: A

New Approach for Constructing Functional-Structural
Plant Models
Ming Wang1, Grant Thorp2, Helen Hofman3, Neil White1,4, Ella Wherritt1 and Jim Hanan1
1
Queensland Alliance for Agriculture and Food Innovation (QAAFI), The University of Queensland, Brisbane, QLD, Australia
4072
2
Plant & Food Research Australia Pty Ltd, Brisbane, QLD, Australia 4000
3
Department of Agriculture and Fisheries, Bundaberg Research Facility, Kalkie, QLD, Australia 4670
4
Department of Agriculture and Fisheries, Toowoomba, QLD, Australia 4350
m.wang@uq.edu.au

Abstract—Computational modelling is becoming increasingly Keywords—pattern-oriented modelling; agent-based model;

significant in improving our understanding of natural systems,
and in making predictions to manage them. Many computational
models have been used in various areas for these purposes.
However, it has been suggested that models should not be either
too simple or too complex, if they are to be useful. Thus, it is of
importance to construct models with an optimised model
structure that sufficiently well represents their real world
counterparts. To do that, better modelling strategies are needed.
The pattern-oriented modelling (POM) strategy is an
approach that has been proposed to address these issues. It has
been used widely to develop agent-based models (ABMs), aiming
to make the models more comprehensive and rigorous, and
increasing their predictive power. Functional-structural plant
models (FSPMs) can be identified as ABMs, if organs/growth
units of a plant are considered as agents. To test the feasibility
and demonstrate the value of using the POM strategy for
functional-structural plant modelling, this study focuses on
modelling of avocado (Persea americana, cv. Hass), because of its
clear modular construction and its economic significance to
subtropical and tropical horticulture world-wide.
Our study focuses on the systematic development of
techniques to apply the POM strategy to functional-structural
plant modelling. The overall objective was to determine whether
the POM strategy could be used to construct FSPMs in order to
increase their predictive power. In the present study, a
functional-structural plant model of the annual growth module of
avocado was constructed using the POM strategy. The model was
able to reproduce multiple observed patterns of architecture and
shoot growth simultaneously, and to make independent
predictions providing insights into branching architecture, which
were consistent with independently generated findings of other
studies. Comparison of model outcomes to multiple observed
patterns of modular construction at different scales, e.g. metamer
level, growth unit level and branch level, increases our confidence
that the model performed well. Those independent predictions
can be strong indicators that the model is structurally realistic.
individual-based model; functional-structural plant model; model
analysis; model verification; model validation; avocado; Persea
americana; plant architecture
I. INTRODUCTION
With the rapid increase of computing power over recent
decades, computational modelling is now having a profound
effect on streamlining scientific research that would otherwise
require a large amount of work that is costly in both time and
resources. Many computational models have been used as
valuable tools in understanding the studied system and making
predictions for unknown scenarios [1-7]. However, when it
comes to models, a major issue is whether a given model
sufficiently well represents its real world counterpart, since a
model is intended to be a purposeful representation of a real
system [8]. A model certainly cannot contain every aspect of
the real system. Models should be neither too simple nor too
complex, if they are to be useful [9]. For example, if a model is
too simple, it may not be realistic enough and the essential
mechanisms of a real system could be ignored in model
structure. In such a case, the model cannot explain much, and
this will limit its potential to provide understanding and
testable predictions. On the other hand, if a model is too
complex, it may be very difficult to parameterize and model
analysis will be unmanageable due to too much detail. Thus,
better modelling strategies are needed to handle complexity
and uncertainty in model parameters and structure during
model development, so that the constructed model can be
structurally realistic [10]. Only models with realistic structures
will contain key structural elements of a real system and
produce independent/testable predictions. A realistic structure
will make model outcomes more reliable and allow them to be
transformed into better applications in relevant industries.
The pattern-oriented modelling (POM) strategy [11, 12] has
been proposed to improve handling of complexity and
uncertainty. The POM strategy is to use multiple observed
patterns at different scales to design, develop and test models,
because multiple patterns often emerge from different essential

We acknowledge funding for this research from ‘The Small Trees High
Productivity Initiative’ project – AI13004 ‘Transforming subtropical/tropical
tree crop productivity’ funded by Horticulture Innovation Australia Ltd (HIA)
and from the New Zealand Ministry of Business, Innovation and Employment
(MBIE) project: ‘Avocados for export – delivery on an industry version’
(C11X1305).

978-1-5090-1659-4/16/$31.00 © 2016 IEEE 204

Authorized licensed use limited to: University of Leeds. Downloaded on August 24,2021 at 22:39:55 UTC from IEEE Xplore. Restrictions apply.
properties of a real system. Simulation models often require
many parameters to simulate the complexity of the studied
system. For example, modelling plant growth usually requires
many parameters describing each biological component of
plant physiology (e.g. growth rate for each organ and their
priority for carbon allocation) as well as parameters describing
the environment, such as light and temperature. In general, it
can be very difficult or impossible to collect data on plant
growth for many parameters to allow models to be developed
in a timely manner. However, the POM strategy has been
suggested as an approach to overcome problems when there is
a lack of detailed biological knowledge [13, 14]. The
advantages of using the POM strategy are that (1) ‘inverse
modelling’ [1, 3, 5] can be achieved by adjusting a model to
reproduce multiple observed patterns to determine unknown
parameters, since each observed pattern can act as a ‘filter’ to
identify unrealistic parameters/processes or model structure,
and (2) alternative model versions/hypotheses can be tested
against multiple observed patterns representing different
aspects of a real system [9, 12]. Although many models may
correctly reproduce a single pattern, a model with a realistic
structure is more likely to reproduce multiple observed patterns
simultaneously [10]. In other words, a model capable of
reproducing multiple observed patterns will be most likely to
contain the right mechanisms to describe the studied system
and reveal emergent patterns. Hence, that model should be able
to generate reliable and testable predictions.
The POM strategy was originally developed in the field of
ecology and has been used widely to develop agent-based
models (ABMs) known as individual-based models (IBMs)
[10]. It has been applied in many fields in addition to ecology
[15, 16], including engineering [17], environmental science
[18], geography [19], microbiology [20] and social science [21,
22]. Functional-structural plant models (FSPMs) [23-25] can
be identified as IBMs, if organs/growth units of a plant are
considered as individuals. FSPMs use simple rules to simulate
plant behaviour, the interactions among and between
individuals in a branching structure, and the interactions
between individuals and their abiotic environment. Many
FSPMs have been used as tools to increase the understanding
of how different aspects of plant architecture and function
interact for various plant species in agriculture and horticulture,
e.g. for apple (Malus domestica) [26], cucumber (Cucumis
sativus) [27, 28], grapevine (Vitis vinifera) [29], kiwifruit
(Actinidia deliciosa) [30, 31], peach (Prunus persica) [32],
soybean (Glycine max) [33, 34], and tomato (Solanum
lycopersicum) [35]. However, there has been little application
of the predictive power of computational models in functional-
structural plant modelling (FSPM) research field, rather than
just using models to understand the studied system. The POM
strategy may be helpful in this respect, and the idea of using it
seems not to have been promoted for FSPM.
To test the feasibility and demonstrate the value of using
the POM strategy for FSPM, this study focused on modelling
of avocado (Persea americana, cv. Hass), because of its clear
modular construction [36, 37] and its economic significance to
subtropical and tropical horticulture in the world [38].
205
Authorized licensed use limited to: University of Leeds. Downloaded on August 24,2021 at 22:39:55 UTC from IEEE Xplore. Restrictions apply.
Generally, avocado trees grow in a rhythmic pattern with
periodic shoot extension in which a spring flush (mixed
reproductive and vegetative), a summer flush (vegetative) and
an autumn flush (vegetative) occur over an annual growing
period, and the canopy is made up of a mix of proleptic and
sylleptic shoot modules [36]. Proleptic shoots develop from
dormant or resting buds after their parent shoots have stopped
growing. In avocado, they can be identified by buds or a bud-
scar ring at their base and they can be vegetative or
reproductive. In contrast, sylleptic shoots develop from axillary
buds while the subtending parent shoots are still extending.
Cultivar differences can be clearly seen in the relative
frequency and dimensions of each shoot type which ultimately
determine the structure and shape of the mature tree.
Understanding these relationships and the natural growth habit
of trees, i.e. their architecture, is fundamental to the
development of more intensive growing systems for many fruit
crops including avocado [39]. Thus, development of such an
avocado architecture model is timely. Additional insight into
architecture and shoot growth patterns could translate into
applied outcomes, so that better canopy management strategies
can be conducted more effectively in avocado orchards.
Our overall objective was to determine whether the POM
strategy could be applied to construct FSPMs in order to
increase their predictive power. The aim of this paper was to
use the POM strategy to develop a functional-structural plant
model that simulated avocado annual growth modules as
described by Thorp and Sedgley [36] and Thorp et al. [37], and
to determine if the model could then accurately predict the
architecture and shoot growth patterns of avocado under
different orchard management strategies.
II. MATERIALS AND METHODS
A. The Model
A 3D functional-structural plant model was developed,
tested and parameterized in terms of the POM strategy [11, 12].
L-studio [40, 41] has been used to develop the model, and L-
systems [42] were used to create a 3D virtual plant
representation [43] of the avocado annual growth module [36],
that is formed from an indeterminate inflorescence that does
not set fruit over an annual growing period – one mixed
reproductive and vegetative (spring) and two vegetative
(summer and autumn) growth flushes (Fig. 1).
The model was designed to simulate the development of an
avocado annual growth module for a 4-year-old tree in the
absence of fruit set. Agents are individual organs (apices -
terminal buds, leaves, internodes, axillary buds and compound
inflorescences) represented as modules in the L-system, with
state variables appropriate to the organ (see Table 1). L-system
productions control the behaviour of agents. Production rules
for organs such as leaves and internodes capture extension
growth of the organs based on the parameters listed in Table 2.
Production rules for terminal buds and axillary buds control
when they produce a new shoot, based on probabilities in Table
2.
 TABLE I. OVERVIEW OF STATE VARIABLES

Organs (Agents) State Variables Description

Apex
Node The node number on the axis
Order The axis order
Degree days Age in thermal time
Shoot type flag Sylleptic type or proleptic type
Preformed nodes The number of preformed/neoformed nodes
Axillary shoot number Axillary shoot ID
Flush node number Node number within vegetative section of flush

Axillary bud
Node The node number on the parent axis
Delay Keeping axillary buds inactive
Order The axis order
Degree days Age in thermal time
Flush node number Node number within vegetative section of flush
Flush number Flush number of different flushes
Shoot type flag Sylleptic type or proleptic type
Proleptic bud
Node The node number on the subtending axis
ID The node number on the current axis
Order The axis order
Degree days Age in thermal time
Flush node number Node number within vegetative section of flush
Flush number Flush number of different flushes
Shoot type flag Proleptic type
Sylleptic bud
Node The node number on the subtending axis
ID The node number on the current axis
Order The axis order
Degree days Age in thermal time
Shoot type flag Sylleptic type
Internode
Length Current length of the internode
Degree days Age in thermal time
Flush number Flush number of different flushes
Leaf
Length Current length of a leaf
Width Current width of a leaf
Degree days Age in Thermal time
Inflorescence
Node The node number
Order The axis order
Inflorescence internode
Length Current length of the internode
Degree days Age in thermal time
Flush number Flush number of different flushes

206

Authorized licensed use limited to: University of Leeds. Downloaded on August 24,2021 at 22:39:55 UTC from IEEE Xplore. Restrictions apply.
 TABLE II. OVERVIEW OF INPUT PARAMETERS IN THE FUNCTIONAL-STRUCTURAL PLANT MODEL FOR AVOCADO ANNUAL GROWTH MODULE
Parameter Description Value and Units References
Avocado physiology
Thermal time – average degree days in a day 20 °C d *
Phyllochron 40 °C d *
Maximum length of leaf 60 cm *
Growth duration of leaf – maximum thermal time for leaf growth 100 °C d *
Maximum length of internode 30 cm *
Growth duration of internode – maximum thermal time for internode
50 °C d *
growth
Maximum length of the primary axis of the inflorescence 10 cm *
Growth duration of the inflorescence – maximum thermal time for
50 °C d *
inflorescence growth
Preformed node number
Number of nodes for leaves and leaf bracts in the spring flush 9.2 ± 0.4 (mean ± SE) [37]
Number of nodes for the secondary axes of the inflorescences in the
6.5 ± 0.3 (mean ± SE) [37]
spring flush
Number of nodes for bud scales in spring, summer or autumn flushes 5.7 ± 0.3 (mean ± SE) [37]
Number of nodes for the primary growth axis in summer or autumn
14.1 ± 0.3 (mean ± SE) [37]
flushes
Number of nodes for the second and third order growth axes including
11.9 ± 0.2 (mean ± SE) [37]
proleptic shoots and extension shoot axes in summer or autumn flushes
Neoformed node number
Number of nodes for sylleptic shoots in spring, summer or autumn flushes 5.9 ± 0.9 (mean ± SE) [37]
Location of second and
third order growth axes
4 (first basal four nodes
Based on [37] and
Location of sylleptic shoots excluding nodes for bud
model calibration
scales or inflorescences)
4 (last four nodes – Based on [37] and
Location of proleptic shoots
closest to the apex) model calibration
Probability
Probability of shoot extension in the spring flush 100% [45]
Based on [45]and
Probability of shoot extension in the summer flush 80%
model calibration
Based on [45] and
Probability of shoot extension in the autumn flush 30%
model calibration
Probability of a sylleptic shoot developing from an axillary bud on the Based on model
80%
primary growth axis in the position noted above calibration
Probability of a sylleptic shoot developing from an axillary bud on a Based on model
2%
second order growth axis in the position noted above calibration
Probability of a proleptic shoot developing from an axillary bud on the Based on model
90%
primary growth axis in the position noted above calibration
Probability of a proleptic shoot developing from an axillary bud on a Based on model
2%
second order growth axis in the position noted above calibration
Time
Start of spring flush 1 day *
Start of summer flush 92 day *
Start of autumn flush 183 day *
Total number of days 365 day [36]
*
Value was used only for visualization purposes

In our model, the terminal bud on a shoot produced during resting axillary buds in subterminal positions on the previous
the spring growth flush initiates a compound, indeterminate growth flush of the primary growth axis or second order
inflorescence. This inflorescence contains a vegetative bud in a growth axes. Sylleptic shoots may develop from axillary buds
terminal position with second and higher order axes producing formed during the current flush on the primary or second order
the branches of the compound inflorescence. The vegetative growth axes. Shoots stop extension growth when an
apex continues growth during spring by producing successive empirically determined maximum number of
metamers, each including an internode, a leaf, and an axillary preformed/neoformed nodes have developed.
bud. Those extending internodes become the primary
vegetative growth axis. Sylleptic shoots can develop from
axillary buds on this axis.
In subsequent summer and autumn growth flushes, the
terminal bud continues to develop as a vegetative shoot and
forms the primary growth axis. Proleptic shoots develop from

207

Authorized licensed use limited to: University of Leeds. Downloaded on August 24,2021 at 22:39:55 UTC from IEEE Xplore. Restrictions apply.

Fig. 1. Example of a 2D representation of an avocado annual growth module,
forming from an indeterminate inflorescence over an annual growing period –
one mixed reproductive and vegetative (spring – yellow) and two vegetative
(summer – green and autumn – red) growth flushes. The annual growth
module contains a primary growth axis, and second and third order growth
axes that can be either proleptic shoots (shown with a bud-scar ring at their
base) or sylleptic shoots. The current season’s shoot growth developed from a
terminal bud from the previous season with the shoot initially producing
lateral inflorescences without internode extension. Vegetative growth then
occurred in the three flushes with the formation of a terminal bud in between
each flush. The remains of the terminal buds turned into a bud-scar ring
visible between each flush later in the season.
The model parameters were chosen based largely on
multiple observed patterns of branching architecture and the
modular construction of trees. However, for visualization
purposes, parameters associated with plant physiology and the
start day of each flush used arbitrary values, and the growth
rate of inflorescences, leaves and internodes were assumed to
be linear. Specific values were not needed, since the current
study focuses on the branching architecture of an avocado
annual growth module. All model input parameters are listed in
Table 1. Simulation of the growth of an avocado annual growth
module in three flushes over an annual growing period is
shown in Fig. 2.
B. Model Assessment
The model was assessed using both qualitatively and
quantitatively observed patterns in terms of the POM strategy
[11, 12], and using a combination of model evaluation and
validation (the ‘evaludation’ approach [44]). The ‘evaludation’
approach is a protocol and guide for assessing a model’s
performance, reliability and quality in the field of ecological
modelling. The main purpose was to determine whether our
model reproduced patterns at multiple scales (metamer level,
208
Authorized licensed use limited to: University of Leeds. Downloaded on August 24,2021 at 22:39:55 UTC from IEEE Xplore. Restrictions apply.
growth unit level and branch level) that had been observed in
the field.
1) Model Output Verification: The term ‘verification’ is
defined as the comparison of model outcomes with patterns
that guided model design and were possibly used for
calibration. The model was run 50 times and outcomes used for
verification against observed patterns from field experiments
conducted by Thorp et al. [37] and Thorp and Sedgley [36].
This allowed us to evaluate the accuracy of model processes
and outcomes based on the data that the model was tuned for,
because correct properties (independent predictions) of a real
system can only emerge from a sound base of a verified model.
Fig. 2. Simulation of an annual growth module on a 4-year-old ‘Hass’
avocado tree. The annual growth module formed from an indeterminate
inflorescence that did not set fruit over an annual growing period. It shows
inflorescence positions at the base, the primary growth axis, and second and
third order growth axes at the end of each growth flush. A: the reproductive
and vegetative growth flush in spring (orange). B: the vegetative growth flush
in summer (brown). C: the vegetative growth flush in autumn (blue).
2) Model Calibration: Model calibration was performed to
identify the optimal values of several parameters. For example,
for the second/third order growth axis location parameters, it
was observed that the location for sylleptic shoots was
generally the first two or three nodes towards the base of a
shoot, whereas proleptic shoots developed from the last two or
three subterminal nodes at the apex of a shoot [37]. In addition,
it has been observed that the probability of shoot extension is
from 60% to 80% for summer flush and between 20% and 30%
for autumn flush in avocado orchards [45]. However, when
using the combination of those values for model parameters,
several observed patterns (e.g. the number of proleptic shoots
on the primary growth axis) were not able to be reproduced by
the model. To determine the best values, we followed the POM
strategy [11, 12] using observed patterns [36, 37] at multiple
scales as filters to determine the optimized value for calibrated
parameters that are capable of reproducing all observed
patterns for model output verification. In such a way, the
uncertainty in those parameters was reduced [9].
3) Model Output Corroboration: The term ‘corroboration’
refers to the comparison of model predictions with independent
patterns that were not used and preferably not even known,
C. Data Analysis
Statistical analysis of the comparison of model outputs with
multiple observed patterns was undertaken. The independent
samples t-test was used to compare model outputs with data
collected in the real world. The model was run 50 times. Each
run was set to stop when the time step reached 365 (equivalent
to 365 days in reality).
III. RESULTS
A. Model Output Verification
In order for the model to reproduce multiple observed
patterns [36, 37] at different scales, several parameters were
inversely determined by model calibration. We have tried a
range of values for those parameters and found optimized
values listed in Table 2, which led to the best correspondence
with what has been observed in reality, reproducing all
observed patterns from the experiments [36, 37].
Thirteen patterns observed by Thorp et al. [37] and Thorp
and Sedgley [36] were used for verification. Using those fine-
tuned parameters, our model outcomes match all observed
patterns. The model outcomes show:
• Pattern1: the mean number of nodes for leaves and leaf
bracts in spring flush was 9.56 (standard error of the
mean [SE] = 0.35), which was similar to 9.2 (SE = 0.4)
found by Thorp et al. [37] (t = 0.68, df = 89, P = 0.5).
• Pattern2: the mean number of nodes for axillary
inflorescences in spring flush was 6.66 (SE = 0.27),
which was similar to 6.5 (SE = 0.3) found in
experiments conducted by Thorp et al. [37] (t = 0.4, df
= 89, P = 0.69).
• Pattern3: the mean number of nodes for bud scales in
spring, summer or autumn flush was 5.86 (SE = 0.21),
which was similar to 5.7 (SE = 0.3) found by Thorp et
al. [37] (t = 0.44, df = 89, P = 0.66).
• Pattern4: the mean number of nodes for primary
growth axis in summer or autumn flush was 14.64 (SE
= 0.38), which was similar to 14.1 (SE = 0.3) found by
Thorp et al. [37] (t = 1.11, df = 100, P = 0.27).
• Pattern5: the mean number of nodes for the
second/third order growth axis including proleptic
shoots and extension shoot axis in summer or autumn
flush was 11.7 (SE = 0.19), which was similar to 11.9
(SE = 0.2) found in experiments conducted by Thorp et
al. [37] (t = 0.72, df = 100, P = 0.48).
• Pattern6: the mean number of nodes for sylleptic
shoots in spring, summer or autumn flush was 8.1 (SE
= 0.69), which was similar to 5.9 (SE = 0.9) found by
Thorp et al. [37] (t = 1.93, df = 100, P = 0.06).
209
Authorized licensed use limited to: University of Leeds. Downloaded on August 24,2021 at 22:39:55 UTC from IEEE Xplore. Restrictions apply.
when the model was developed, parameterised and verified.
Those independent predictions are emergent properties from
the model, as the model verification has been done to ensure
the accuracy of processes underlying the model.
• Pattern7: the mean number of sylleptic shoots on the
primary growth axis was 10.26 (SE = 0.2), which was
similar to 10.7 (SE = 1.5) found in experiments
conducted by Thorp and Sedgley [36] (t = 0.46, df =
66, P = 0.64).
• Pattern8: the mean number of proleptic shoots on the
primary growth axis was 6.56 (SE = 0.21), which was
similar to 7.2 (SE = 0.8) found by Thorp and Sedgley
[36] (t = 1.09, df = 66, P = 0.28).
• Pattern9: more sylleptic shoots than proleptic shoots on
the primary growth axis [36]: mean = 10.26 (SE = 0.2)
for sylleptic shoots and 6.56 (SE = 0.21) for proleptic
shoots (t = 12.82, df = 98, P < 0.001).
• Pattern10: the mean number of total nodes on the
primary growth axis was 51.36 (SE = 0.83), which was
similar to 53.4 (SE = 1.3) found by Thorp and Sedgley
[36] (t = 1.28, df = 66, P = 0.2).
• Pattern11: the mean number of growth units borne by
the primary growth axis was 26.48 (SE = 0.49), which
was similar to 25.5 (SE = 3.1) found in experiments
conducted by Thorp and Sedgley [36] (t = 0.49, df =
66, P = 0.63).
• Pattern12: the mean number of terminal growth units
was 19.2 (SE = 0.39), which was similar to 18.0 (SE =
2.2) found by Thorp and Sedgley [36] (t = 0.83, df =
66, P = 0.41).
• Pattern13: the mean number of nodes for proleptic
shoots was 11.7 (SE = 0.19) which was more than the
mean number of nodes for sylleptic shoots (mean =
8.1, SE = 0.69) and was consistent with the observed
pattern found by Thorp et al. [37] where (t = 5.0, df =
98, P < 0.001).
These model outcomes are thus consistent with
observations in the two studies.
B. Model Output Corroboration
After running simulation experiments, our model
predictions were compared with independently generated
findings of other studies on the architecture of avocado trees.
Six independent patterns have been used for model output
corroboration. The model predictions showed:
• Pattern14: the mean number of nodes (leaf nodes only)
per growth unit was 6.98 (SE = 0.28), which was
similar to 6.5 (SE = 0.24) on non-fruit-bearing shoots
found on 4-year-old ‘Hass’ trees in a commercial
orchard in central Queensland, Australia [46] (t = 0.93,
df = 64, P = 0.36).
 • Pattern15: the mean number of nodes (leaf nodes only)
on the primary growth axis was 27.12 (SE = 0.92),
which was similar to 24.23 (SE = 0.86) on non-fruit-
bearing shoots found on 4-year-old ‘Hass’ trees in a
commercial orchard in central Queensland, Australia
[46] (t = 1.69, df = 64, P = 0.1).
• Pattern16: the mean number of nodes including leaves
and leaf bracts nodes, axillary inflorescences nodes and
bud scales nodes in spring flush was 22.08 (SE = 0.48),
which was similar to 20.9 (SE = 0.4) found by Thorp et
al. [37] (t = 1.89, df = 100, P = 0.06).
• Pattern17: the proportions of sylleptic and proleptic
growth units were 42.52% and 57.48% respectively,
which was similar to 46% and 54% on shoots with
very few fruit found by Mickelbart et al. [47].
• Pattern18: third order growth axes occurred during the
simulation experiments as there were no fruit (see Fig.
2), which was consistent with the observation that
shoots with almost no fruit had third order growth axes
[47].
These model predictions were consistent with the
independent patterns noted in the referenced studies.
IV. DISCUSSION AND CONCLUSIONS
We have described, applied and demonstrated the possible
use of the POM strategy [11, 12] for FSPM. Therefore, we
constructed a functional-structural plant model of an avocado
annual growth module using the POM strategy to check how
well multiple observed qualitative and quantitative patterns can
be reproduced after model calibration, and to see how the POM
strategy could help refine model parameters and structure.
By comparing model outputs to multiple patterns observed
in the field or experiments we can increase our confidence that
the model performed well [9]. The model was able to predict
architecture and shoot growth patterns of the avocado annual
growth module that was consistent with independent data. The
model generated realistic outcomes observed in different
orchards using published data as model inputs. Our model was
able to identify the key elements of the modular growth or
branching architecture of avocado, e.g. the number of nodes,
and the proportion of sylleptic/proleptic growth units. Such
patterns, secondary or independent predictions, can be strong
indicators that the model is structurally realistic [9].
We used the POM strategy to guide construction of the
model. For model development and testing stages, we
compared the outcomes from the model with observed
patterns to refine the model and to determine uncertain
parameters. In this way, the model is most likely to capture the
essential elements/mechanisms of studied systems, because
model design and development are based on patterns that help
narrow down the scope of model structure and parameters in a
natural way. For instance, we used observed rhythmic growth
patterns to start constructing a fundamental structure of the
model, and then refine it by adding each biological component
(e.g. primary growth axis, second/third order growth axis and
shoot types) on the model structure in terms of multiple
observed patterns on branching architecture. Especially for the
210
Authorized licensed use limited to: University of Leeds. Downloaded on August 24,2021 at 22:39:55 UTC from IEEE Xplore. Restrictions apply.
third order growth axes, we did not know how many shoot
levels a non-fruit-bearing branch can be up to. In order for the
model to reproduce patterns on the number of terminal growth
units and growth units observed by Thorp and Sedgley [36],
the third order growth axes had to be considered in the model.
The third order growth axes indirectly emerged during the
construction of the model, which was later found it consistent
with the study conducted by Mickelbart et al. [47] because it is
unusual to see third order growth axes on trees with fruit
present in avocado orchards. Also, the uncertainty around
parameters such as the probability of shoot extension in
summer or autumn flushes, was inversely determined using
other observed architectural patterns on the number of nodes
and growth units. According to Grimm and Railsback [10], a
set of seemingly ‘weak’ patterns used for model construction
can be more efficient in capturing relevant parameter
combinations and identifying the most appropriate model
structure, than focusing on a single ‘strong’ pattern. Thus, it is
important for a model to reproduce multiple observed patterns
simultaneously, as the more patterns a model reproduces, the
greater the chance that those patterns are reproduced for the
right reason, instead of having the model tuned through
calibration to appear realistic.
For FSPM, it is common to build a model by tuning the
model through calibration, so that behaviours of the model can
look exactly like the plant that has been observed in the field or
lab experiments [48]. As a result, the model will tend to merely
appear realistic, rather than being structurally realistic. Such
models may help to increase the understanding of how
different aspects of plant architecture and function interact, but
are less likely to generate insightful perspectives (independent
predictions) or explanations and may lose their predictive
power. There are two main reasons for this. Firstly, plant
modellers may tend to identify plant structure empirically
without understanding underlying mechanisms before
constructing a model. Many parameters are determined by
specific field or lab experiments. The parameters used in the
model may not be all necessarily related to their scientific
research questions. This would lead to an over-parameterized
model with a complex model structure, so emergent properties
will not be able to fully arise from the model. Secondly, due to
the lack of multiple patterns in construction of the model, e.g.
verification and corroboration, wrong parameters or
inappropriate model structure could nevertheless reproduce
exactly what has been seen in the field or lab experiments.
Therefore, those models would not be suitable for providing
suggestions for decision-making or management purpose in
agriculture and horticulture, but certainly could be used by
scientists for the purpose of understanding.
Our model also generated specific predictions, which
provided insights into branching architecture of an avocado
annual growth module. During model calibration, we found
that the probability of shoot extension and the probability of an
axillary bud turning into a sylleptic/proleptic shoot, played an
important role in determining the whole structure of the annual
growth module. A study conducted by Thorp and Sedgley [36]
has shown that the mean number of growth units borne by the
primary main growth axis in a non-fruit-bearing annual growth
module is 25.5 (SE = 3.1), which is similar to the model
outcome (mean = 26.48, SE = 0.49). However, this is very
different from that found in central Queensland, Australia, by
the Queensland Department of Agriculture and Fisheries
(DAF), on 4-year-old ‘Hass’ trees in a commercial orchard of a
mean of 11.13 growth units (SE = 2.52) in the 2014-2015
season of growth and 2.29 (SE = 0.34) in the 2015-2016 season
[46]. This not surprising, because data about growth units were
collected from fruit-bearing branches with a particularly heavy
crop load in 2015-2016 by DAF. Furthermore, it has been
found that the excessive shoot development occurred when
there was little fruit presence on branches [47]. Mickelbart et
al. [47] found that the proportion of shoot module types was
46% for sylleptic growth units and 54% for proleptic growth
units on shoots with very few fruit, which is similar to model
outcomes where (sylleptic growth units: 42.52% and proleptic
growth units: 57.48%). In contrast, there was no more than 2%
sylleptic shoots in heavy crop load years. Third order growth
axes were also observed by Mickelbart et al. [47] in a light crop
load year, which is consistent with the model outcome,
whereas it is unusual to see the third order growth axes in
heavy crop load years. Thus, we hypothesize that those
differences can be ascribed to the repression of vegetative
growth by a heavy crop load. The lack of fruit growth results in
a higher probability for an axillary bud developing into a
sylleptic or proleptic shoot type in the model. This high
probability may represent maximum vigour or maximum
growth potential, which may be a result of the absence of
competition for carbon between vegetative components and
fruit.
In the current model, there was no fruit in the annual
growth module, so vegetative growth was very strong, which is
consistent with field experiments conducted by Thorp and
Sedgley [36]. The probability of shoot development might be
lower, if fruit were incorporated in the annual growth module.
While fruit may compete for carbon and thus suppress
vegetative growth, the presence of fruit may also have a more
direct effect on gene expression associated with vegetative
growth and floral development [49-52]. This phenomenon
could lead to biennial (‘on’/‘off’) patterns of avocado shoot
growth, because few branches would be generated if shoot-
development probability is lower. This argument can be
supported by studies [47, 49, 53] that have shown avocado
shoot growth follows such patterns. Thus, the model has
generated outcomes that relate to carbon allocation and gene
expression perspectives from model calibration in terms of the
POM strategy. The model has identified that fruit is a key
factor determining biennial patterns of avocado shoot growth
and yield [49].
Our model could be developed further to investigate carbon
competition and gene expression on the branching architecture
of the avocado annual growth module. In the current model,
physiological parameters like thermal time – average degree
days in a day, phyllochron and the size of leaf/internode have
been selected solely for visualisation, and the growth rate of
inflorescences, leaves and internodes have been assumed to be
linear. Specific measurements were not needed because the
current study focused on the branching architecture of the
avocado annual growth module. To test how fruit is important
to determine biennial patterns or involve carbon competition at
211
Authorized licensed use limited to: University of Leeds. Downloaded on August 24,2021 at 22:39:55 UTC from IEEE Xplore. Restrictions apply.
individual organ level for growth, a more sophisticated
functional-structural plant model can be achieved coupling
with the carbon transport-resistance allocation model (C-
TRAM) [54] or simulating the light environment where the
model exists (QuasiMC) [55], which can provide dynamic
growth patterns of architecture. Future research will be
conducted to develop such a model to investigate underlying
mechanisms of carbon competition on the avocado annual
growth module. A similar approach could be used to model
gene expression and the transition to flowering in avocado trees
to help explain biennial bearing.
Once a pattern-oriented model is constructed, however,
other adjustments are also possible. There may be cause to
adjust it further to study architecture of different cultivars other
than ‘Hass’. The model can be used as a generic model.
Adjustments for specific cultivar can be achieved by tuning
parameters based on targeted experimentally obtained data
[36]. Additionally, the strength of the pattern-oriented
modelling approach is that such pattern-oriented models can
produce comparative predictions which can be tested in the
field [11, 12]. As a result, time and resources can be better
focused on experimentation that shows the greatest likelihood
of success.
In summary, this study has demonstrated that it is feasible
to build a functional-structural plant model using the POM
strategy. We constructed a simple architecture model of the
avocado annual growth module, and applied the POM strategy
to identify parameter combinations that match with multiple
observed patterns from a number of field experiments.
Furthermore, the model was able to generate independent
predictions that match with independently generated findings
of other studies. The POM strategy cannot, of course, solve
everything for the FSPM research area, but it is a superior way
to construct a functional-structural plant model. Thus, we
strongly recommend that plant modellers follow the POM
strategy to construct models so that they can be structurally
realistic.
ACKNOWLEDGMENT
We would like to acknowledge Dr. David Doley at the
University of Queensland for his constructive comments, and
Dr. John Wilkie and Mr. Jarrad Griffin at the Queensland
Department of Agriculture and Fisheries for support of field
work in central Queensland.
REFERENCES
[1] K.R. Martin, D. Barua, A.L. Kauffman, L.M. Westrate, R.G. Posner,
W.S. Hlavacek, et al., “Computational model for autophagic vesicle
dynamics in single cells,” Autophagy, vol. 9, pp. 74-92, 2013.
[2] A. Malawska and C.J. Topping, “Evaluating the role of behavioral
factors and practical constraints in the performance of an agent-based
model of farmer decision making,” Agricultural Systems, vol. 143, pp.
136-146, 2016.
[3] J. Hanan, P. Prusinkiewicz, M. Zalucki, and D. Skirvin, “Simulation of
insect movement with respect to plant architecture and morphogenesis,”
Computers and Electronics in Agriculture, vol. 35, pp. 255-269, 2002.
[4] J.R. Karr, J.C. Sanghvi, D.N. Macklin, M.V. Gutschow, J.M. Jacobs, B.
Bolival, et al., “A whole-cell computational model predicts phenotype
from genotype,” Cell, vol. 150, pp. 389-401, 2012.
[5] N. White and J. Hanan, “A model of macadamia with application to
pruning in orchards,” in XXIX IHC – Proc. Int. Symp. on Nut Crops, M.
 Wirthensohn, Ed. Brisbane, Queensland, Australia: International Society
for Horticultural Science (ISHS), 2016, pp. 75-82.
[6] M. Wang, B. Cribb, I. Auzmendi, and J. Hanan, “Spatially explicit
individual-based modelling of insect-plant interactions: effects of level
of detail in Queensland fruit fly models,” in MODSIM2015, 21st
International Congress on Modelling and Simulation, T. Weber, M.
McPhee, and R. Anderssen, Eds. Gold Coast, Queensland, Australia:
Modelling and Simulation Society of Australia and New Zealand, 2015,
pp. 375–381.
[7] M. Renton, J. Hanan, B.J. Ferguson, and C.A. Beveridge, "Models of
longǦdistance transport: how is carrierǦdependent auxin transport
regulated in the stem?," New Phytologist, vol. 194, pp. 704-715, 2012.
[8] A. Starfield, K. Smith, and A. Bleloch, How to model it: problem
solving for the computer age, New York: McGraw-Hill, 1990.
[9] V. Grimm, E. Revilla, U. Berger, F. Jeltsch, W.M. Mooij, S.F.
Railsback, et al., “Pattern-oriented modeling of agent-based complex
systems: lessons from ecology,” Science, vol. 310, pp. 987-991, 2005.
[10] V. Grimm and S.F. Railsback, Individual-based Modeling and Ecology,
Princeton University Press, 2005.
[11] T. Wiegand, F. Jeltsch, I. Hanski, and V. Grimm, “Using pattern-
oriented modeling for revealing hidden information: a key for
reconciling ecological theory and application,” Oikos, vol. 100, pp. 209-
222, 2003.
[12] V. Grimm and S.F. Railsback, “Pattern-oriented modelling: a ‘multi-
scope’ for predictive systems ecology,” Philosophical Transactions of
the Royal Society B: Biological Sciences, vol. 367, pp. 298-310, 2012.
[13] G. Latombe, L. Parrott, and D. Fortin, “Levels of emergence in
individual based models: coping with scarcity of data and pattern
redundancy,” Ecological Modelling, vol. 222, pp. 1557-1568, 2011.
[14] C.J. Topping, T. Dalkvist, and V. Grimm, “Post-hoc pattern-oriented
testing and tuning of an existing large model: lessons from the field
vole,” PLoS ONE, vol. 7, pp. e45872, 2012.
[15] S. Bauduin, E. McIntire, M.-H. St-Laurent, and S. Cumming,
“Overcoming challenges of sparse telemetry data to estimate caribou
movement,” Ecological Modelling, vol. 335, pp. 24-34, 2016.
[16] M. Wang, B. Cribb, A.R. Clarke, and J. Hanan, “A generic individual-
based spatially explicit model as a novel tool for investigating insect-
plant interactions: a case study of the behavioural ecology of frugivorous
Tephritidae,” PLoS ONE, vol. 11, pp. e0151777, 2016.
[17] T. Jensen, G. Holtz, C. Baedeker, and É.J. Chappin, “Energy-efficiency
impacts of an air-quality feedback device in residential buildings: an
agent-based modeling assessment,” Energy and Buildings, vol. 116, pp.
151-163, 2016.
[18] S. Arnold, S. Attinger, K. Frank, and A. Hildebrandt, “Assessing the
structural adequacy of alternative ecohydrological models using a
pattern-oriented approach,” Ecological Modelling, vol. 316, pp. 52-61,
2015.
[19] N.R. Magliocca and E.C. Ellis, “Using pattern-oriented Modeling
(POM) to cope with uncertainty in multi-scale agent-based models of
land change,” Transactions in GIS, vol. 17, pp. 883-900, 2013.
[20] F.L. Hellweger, N.D. Fredrick, M.J. McCarthy, W.S. Gardner, S.W.
Wilhelm, and H.W. Paerl, “Dynamic, mechanistic, molecular-level
modelling of cyanobacteria: Anabaena and nitrogen interaction,”
Environmental Microbiology, 2016.
[21] M.A. Janssen, N.P. Radtke, and A. Lee, “Pattern-oriented modeling of
commons dilemma experiments,” Adaptive Behavior, vol. 17, pp. 508-
523, 2009.
[22] C. Yang, S. Kurahashi, I. Ono, and T. Terano, “Pattern-oriented inverse
simulation for analyzing social problems: family strategies in civil
service examination in imperial China,” Advances in Complex Systems,
vol. 15, pp. 1250038, 2012.
[23] C. Godin and H. Sinoquet, “Functional-structural plant modelling,” New
Phytologist, vol. 166, pp. 705-708, 2005.
[24] J. Vos, L. Marcelis, and J. Evers, “Functional-structural plant modelling
in crop production: adding a dimension,” in Functional-structural plant
modelling in crop production, J. Vos, L. Marcelis, P. De Visser, P.
Struik, J. Evers, Eds. Dordrecht, The Netherlands: Springer, 2007, pp. 1-
12.
212
Authorized licensed use limited to: University of Leeds. Downloaded on August 24,2021 at 22:39:55 UTC from IEEE Xplore. Restrictions apply.
[25] J. Vos, J.B. Evers, G. Buck-Sorlin, B. Andrieu, M. Chelle, and P.H. De
Visser, “Functional-structural plant modelling: a new versatile tool in
crop science,” Journal of Experimental Botany, vol. 61, pp. 2101-2115,
2010.
[26] M. Saudreau, A. Marquier, B. Adam, and H. Sinoquet, “Modelling fruit-
temperature dynamics within apple tree crowns using virtual plants,”
Annals of Botany, vol. 108, pp. 1111-1120, 2011.
[27] K. Kahlen and H. Stützel, “Simplification of a light-based model for
estimating final internode length in greenhouse cucumber canopies,”
Annals of Botany, vol. 108, pp. 1055-1063, 2011.
[28] D. Wiechers, K. Kahlen, and H. Stützel, “Dry matter partitioning models
for the simulation of individual fruit growth in greenhouse cucumber
canopies,” Annals of Botany, vol. 108, pp. 1075-1084, 2011.
[29] B. Pallas, A. Christophe, and J. Lecoeur, “Are the common assimilate
pool and trophic relationships appropriate for dealing with the observed
plasticity of grapevine development?,” Annals of Botany, vol. 105, pp.
233-247, 2009.
[30] M. Cieslak, A.N. Seleznyova, and J. Hanan, “A functional-structural
kiwifruit vine model integrating architecture, carbon dynamics and
effects of the environment,” Annals of Botany, vol. 107, pp. 747-764,
2010.
[31] M. Cieslak, A.N. Seleznyova, P. Prusinkiewicz, and J. Hanan, “Towards
aspect-oriented functional-structural plant modelling,” Annals of
Botany, vol. 108, pp. 1025-1041, 2011.
[32] M. Allen, P. Prusinkiewicz, and T. DeJong, “Using L-systems for
modeling source-sink interactions, architecture and physiology of
growing trees: The L-PEACH model,” New Phytologist, vol. 166, pp.
869-880, 2005.
[33] L. Han, P.M. Gresshoff, and J. Hanan, “A functional-structural
modelling approach to autoregulation of nodulation,” Annals of Botany,
vol. 107, pp. 855-863, 2010.
[34] L. Han, J. Hanan, and P.M. Gresshoff, “Computational
complementation: a modelling approach to study signalling mechanisms
during legume autoregulation of nodulation,” PLoS Computational
Biology, vol. 6, pp. e1000685, 2010.
[35] V. Sarlikioti, P.H. De Visser, G. Buck-Sorlin, and L. Marcelis, “How
plant architecture affects light absorption and photosynthesis in tomato:
towards an ideotype for plant architecture using a functional-structural
plant model,” Annals of Botany, vol. 108, pp. 1065-1073, 2011.
[36] T.G. Thorp and M. Sedgley, “Architectural analysis of tree form in a
range of avocado cultivars,” Scientia Horticulturae, vol. 53, pp. 85-98,
1993.
[37] T.G. Thorp, D. Aspinall, and M. Sedgley, “Preformation of node number
in vegetative and reproductive proleptic shoot modules of Persea
(Lauraceae),” Annals of Botany, vol. 73, pp. 13-22, 1994.
[38] B. Schaffer, B. Wolstenholme, and A. Whiley, The avocado: botany,
production and uses, 2nd ed. Wallingford, UK: CABI Pub, 2013.
[39] C.M. Menzel and M. Le Lagadec, “Increasing the productivity of
avocado orchards using high-density plantings: a review,” Scientia
Horticulturae, vol. 177, pp. 21-36, 2014.
[40] R. Karwowski and P. Prusinkiewicz, “The L-system-based plant-
modeling environment L-studio 4.0,” in Proceedings of the 4th
International Workshop on Functional-Structural Plant Models, C.
Godin, et al., Eds. Montpellier, France: UMR AMAP, 2004, pp. 403-
405.
[41] P. Prusinkiewicz, R. Karwowski, R. MČch, and J. Hanan, “L-
Studio/cpfg: a software system for modeling plants,” in Applications of
Graph Transformations with Industrial Relevance, M. Nagl, A. Schürr,
and M. Münch, Eds. Berlin: Springer-Verlag, 2000, pp. 457-464.
[42] P. Prusinkiewicz, M. Hammel, J. Hanan, and R. MČch. “L-systems: from
the theory to visual models of plants,” in Plants to Ecosystems.
Advances in Computational Life Sciences I, M.T. Michalewicz, Ed.
Melbourne, Australia: CSIRO Publishing, 1997, pp. 1-27.
[43] P. Room, J. Hanan, and P. Prusinkiewicz, “Virtual plants: new
perspectives for ecologists, pathologists and agricultural scientists,”
Trends in Plant Science, vol. 1, pp. 33-38, 1996.
[44] J. Augusiak, P.J. Van den Brink, and V. Grimm, “Merging validation
and evaluation of ecological models to ‘evaludation’: a review of
 terminology and a practical approach,” Ecological Modelling, vol. 280,
pp. 117-128, 2014.
[45] T.G. Thorp, unpublished.
[46] The Queensland Department of Agriculture and Fisheries, unpublished.
[47] M.V. Mickelbart, P.W. Robinson, G. Witney, and M.L. Arpaia, “‘Hass’
avocado tree growth on four rootstocks in California. II. Shoot and root
growth,” Scientia Horticulturae, vol. 143, pp. 205-210, 2012.
[48] W. Palubicki, K. Horel, S. Longay, A. Runions, B. Lane, R. MČch, et al.
“Self-organizing tree models for image synthesis,” ACM Transactions
on Graphics, vol. 28, 2009.
[49] C.J. Lovatt, Alternate bearing of ‘Hass’ avocado, vol. 93, California
Avocado Society Yearbook, 2011, pp. 125-140.
[50] D. Ziv, T. Zviran, O. Zezak, A. Samach, and V. Irihimovitch,
“Expression profiling of FLOWERING LOCUS T-Like gene in
alternate bearing ‘Hass’ avocado trees suggests a role for PaFT in
avocado flower induction,” PLoS ONE, vol. 9, pp. e110613, 2014.
213
Authorized licensed use limited to: University of Leeds. Downloaded on August 24,2021 at 22:39:55 UTC from IEEE Xplore. Restrictions apply.
[51] H.M. Smith and A. Samach, “Constraints to obtaining consistent annual
yields in perennial tree crops. I: heavy fruit load dominates over
vegetative growth,” Plant Science, vol. 207, pp. 158-167, 2013.
[52] A. Samach and H.M. Smith, “Constraints to obtaining consistent annual
yields in perennials. II: environment and fruit load affect induction of
flowering,” Plant Science, vol. 207, pp. 168-176, 2013.
[53] J. Dixon, Shoot growth of ‘Hass’ avocado tres in ‘on’ and ‘off’
flowering years in the western bay of plenty, Annual Research Report of
New Zealand Avocado Growers Association, 2007.
[54] P. Prusinkiewicz, M. Allen, A. Escobar-Gutierrez, and T. DeJong,
“Numerical methods for transport-resistance sink-source allocation
models,” in Functional-structural plant modelling in crop production, J.
Vos, L. Marcelis, P. De Visser, P. Struik, J. Evers, Eds. Dordrecht, The
Netherlands: Springer, 2007, pp. 123–137.
[55] M. Cieslak, C. Lemieux, J. Hanan, and P. Prusinkiewicz, “Quasi-Monte
Carlo simulation of the light environment of plants,” Functional Plant
Biology, vol. 35, pp. 837-849, 2008.