A Canopy Architectural Model to Study the ... - Oxford Journals

3 downloads 0 Views 387KB Size Report
S-ZAHRA-HOSSEINI CICI1–4,*, STEVE ADKINS3,4 and JIM HANAN1. 1Advanced ..... advances in grid computing (Berman et al., 2003) are quite promising, as ...
Annals of Botany 101: 1311– 1318, 2008 doi:10.1093/aob/mcn040, available online at www.aob.oxfordjournals.org

A Canopy Architectural Model to Study the Competitive Ability of Chickpea with Sowthistle S - Z A H R A - H O S S E I N I C I C I 1 – 4, *, S T E V E AD K IN S 3,4 and JI M HA NA N 1 1

Advanced Computational Modeling Center, University of Queensland, St Lucia, Brisbane, 4072, Australia, 2School of Agronomy and Plant Breeding, Shiraz University, Shiraz, Iran, 3School of Land and Food, University of Queensland, St Lucia, Brisbane, 4072, Australia and 4Cooperative Research Centre (CRC) for Australian Weed Management Received: 28 August 2007 Returned for revision: 27 September 2007 Accepted: 25 February 2008 Published electronically: 28 March 2008

† Background and Aims Improving the competitive ability of crops is a sustainable method of weed management. This paper shows how a virtual plant model of competition between chickpea (Cicer arietinum) and sowthistle (Sonchus oleraceus) can be used as a framework for discovering and/or developing more competitive chickpea cultivars. † Methods The virtual plant models were developed using the L-systems formalism, parameterized according to measurements taken on plants at intervals during their development. A quasi-Monte Carlo light-environment model was used to model the effect of chickpea canopy on the development of sowthistle. The chickpea– light environment– sowthistle model (CLES model) captured the hypothesis that the architecture of chickpea plants modifies the light environment inside the canopy and determines sowthistle growth and development pattern. The resulting CLES model was parameterized for different chickpea cultivars (viz. ‘Macarena’, ‘Bumper’, ‘Jimbour’ and ‘99071-1001’) to compare their competitive ability with sowthistle. To validate the CLES model, an experiment was conducted using the same four chickpea cultivars as different treatments with a sowthistle growing under their canopy. † Results and Conclusions The growth of sowthistle, both in silico and in glasshouse experiments, was reduced most by ‘99071-1001’, a cultivar with a short phyllochron. The second rank of competitive ability belonged to ‘Macarena’ and ‘Bumper’, while ‘Jimbour’ was the least competitive cultivar. The architecture of virtual chickpea plants modified the light inside the canopy, which influenced the growth and development of the sowthistle plants in response to different cultivars. This is the first time that a virtual plant model of a crop – weed interaction has been developed. This virtual plant model can serve as a platform for a broad range of applications in the study of chickpea– weed interactions and their environment. Key words: Plant architecture, virtual plant modelling, L-systems formalism, crop/weed competition, integrated weed management, chickpea, Cicer arietinum, sowthistle, Sonchus oleraceus.

IN TROD UCT IO N It is now widely accepted that the best approach to manage weeds is through the development and implementation of integrated weed management (IWM) systems (Swanton and Murphy, 1996). Choosing an appropriate individual crop plant architecture, integrated with an appropriate crop planting density is an important component of IWM to suppress weed growth (Lemerle et al., 2001). Evaluation of these effects is particularly important in view of the growing interest in reducing the reliance on chemical weed management (Blackshaw et al., 2006). A higher crop competitive ability is desirable as it decreases weed seed production, which in turn reduces weed infestation size in subsequent years (Jordan, 1993). Plant architecture is the size, number and spatial arrangement of plant organs upon the plant body. It determines the interaction of a plant with its environment and neighbouring plants (Aphalo and Ballare, 1995). A recent advance in the study of biological systems, such as plant architecture and crop– weed competition, is the use of computers to model and simulate such systems. In the last two decades different models have been developed to simulate * For correspondence. E-mail [email protected]

crop– weed competition (Bastiaans et al., 1997; Olesen et al., 2004) but none of them have taken into account the actual architecture of the individual crop plant and its effects on weed performance. Having a virtual plant model will help one to see the results of competition over time and combine different cultivars or different planting rules to analyse the competitive ability of that crop plant in silico (i.e. by running a computer simulation). A variety of modelling approaches can be used to implement a virtual model of plant development (Prusinkiewicz, 1998). The ability to express branching patterns and to describe relatively complex architectural structures based on rewriting rules makes the L-systems formalism (Lindenmayer, 1968) an appropriate candidate for modelling plant architectural development (Prusinkiewicz and Lindenmayer, 1996). In order to capture environmental influences, a virtual plant model can be interfaced with an environment program (Mech, 1997). The environment program takes the current structure of the plant as input, then processes it to determine certain environmental conditions (such as degree of shade or light availability), which are returned to the plant model and influence future development. Surface-based light models (Chelle and Andrieu, 1999) are one type of these environment

# The Author 2008. Published by Oxford University Press on behalf of the Annals of Botany Company. All rights reserved. For Permissions, please email: [email protected]

1312

Cici et al. — Canopy Architectural Modelling of Plant Competition

models that can compute light distribution over the 3-D canopy architecture of the virtual plants by taking the geometry of each organ into account. The present research study is the first attempt to use these 3-D virtual modelling techniques to help select crop cultivars that are more competitive with weeds or to select crop planting geometries that are also more conducive to weed-growth suppression. Chickpea (Cicer arietinum), as one of the important grain legumes and sowthistle (Sonchus oleraceus), as one of the common weeds found in this crop (Holm et al., 1988), were chosen for the study. A virtual plant model called the chickpea – light environment –sowthistle (CLES) was developed to study competitive ability of different chickpea cultivars with sowthistle. In this model, simulated development of sowthistle plants in response to the light environment under the chickpea plants serves as an indicator of the weed-suppression ability of chickpea cultivars. The model is also validated and its potential power in nonchemical IWM is discussed. This version of the CLES relies on the idea of size-asymmetric competition to simulate the effect of chickpea plants on sowthistle plants. Size-asymmetric competition (Schwinning and Weiner, 1998) suggests that increased crop density and spatial uniformity can have large effects on weed growth when the crop has a significant initial number/size advantage over the weed and also when enough water and nutrients are provided for plant growth and development. Here competition for light is important enough to dominate the crop– weed interaction. However, when below-ground competition is structuring the crop– weed interaction, a size-symmetric competition (Schwinning and Weiner, 1998) would be expected, which in turn reduces the potential for suppressing weeds by crops (Weiner et al., 2001). This second criterion has not yet been incorporated into the CLES model.

M AT E R I A L S A N D M E T H O D S Botany of chickpea and sowthistle

Chickpea (Cicer arietinum L.) is a legume with distichous (one leaf per node) phyllotaxy and its first two nodes have only scale leaves. The main stem is indeterminate and there is a delay in the outgrowth of the axillary buds (Singh, 1997). Leaves in most of the cultivars are pinnate (fern) with the number of leaflets differing by position on the main stem. Annual sowthistle (Sonchus oleraceus L.) is a broad-leaf weed from the Asteraceae. It is a rosette plant that first produces all its leaves and then produces a terminal flowering bud. Sowthistle has two different forms of leaves; rosette leaves grow in crowded circles from a common centre or crown, close to the ground, while cauline leaves grow on the upper part of the stem. The length of the stem internodes associated with cauline leaves increases dramatically after the appearance of the first flowering buds (bolting). All nodes on the main stem have an axillary bud that have the potential to grow out and form a branch (Holm et al., 1988).

Model development and parameterization

The virtual models of chickpea and sowthistle plants were developed using the L-systems formalism (Prusinkiewicz and Lindenmayer, 1996). The basic components of the virtual plant models are represented by these modules: apical meristem of the main stem, A; internodes, I; leaves, L; and axillary buds, B. The topology of the plants is captured by the order of the modules in the string (Prusinkiewicz and Lindenmayer, 1996). As an example, the basic apex (apical meristem) production rule used in the plant model is: A ! I½L½BA where apex (A) gives rise (!) to an internode (I ), a leaf (L), an axillary bud (B) and an ongoing apex (A). Square brackets are used for specifying branches. In order to quantify the properties of the plant components, numerical parameters (Hanan, 1992) are associated with the modules. The numerical parameters are real values expressed using arithmetic expressions, predefined functions and condition statements. Each module has two parameters; node number and age (represented as growing degree-days, 8Cd). To capture plant development, the L-systems formalism uses rewriting rules that are applied in parallel and simultaneously replace all previous modules in a given string. For details of the language syntax see the cpfg user’s manual in conjunction with the L-studio package from http://algorithmicbotany.org/. A visual example of the growth and development of a chickpea plant (Fig. 1) is shown using the L-studio software version 4.0 (Prusinkiewicz et al., 2000). In order to calculate the distribution of light under the chickpea canopy, plant models were interfaced with a model of light environment using the formalism of open L-systems (Mech and Prusinikiewicz, 1998). The calculation of light distribution is based on the quasi-Monte Carlo (QMC) method (Lemieux et al., 2004). Paths of photons are traced through the scene, reflected, absorbed and transmitted by the polygons representing the leaflets of chickpea (Fig. 2). Three L-systems were combined to simulate the ability of virtual chickpea plants to intercept light. The main L-system was for switching between chickpea development and the light modelling processes, the chickpea L-system was for plant development, and the light model L-system was for controlling the communication between the chickpea plant and the light computation program QMC. A communication module (Mech, 1997)

F I G . 1. Side view of a virtual chickpea plant 15 and 60 d after germination, with the in silico thermal time set to 20 8C per day.

Cici et al. — Canopy Architectural Modelling of Plant Competition

1313

TA B L E 1. Components of the virtual sowthistle parameterized for the light environment (%LE) Input Phyllochron Thermal time for cauline internodes Thermal time for leaves Final capitula numbers

F I G . 2. The communication of the two virtual plant models with the QMC light-environment model. Information on polygon positions (chickpea leaflets and sowthistle sensor) is passed to the QMC light-environment model (1), which calculates light availability for a virtual sensor at the sowthistle location (2); information is transmitted to the sowthistle through the virtual sensor (3).

Value

Units

66 – 0.38LE (r 2 ¼ 0.99) 398 – 1.5LE (r 2 ¼ 0.96)

Thermal time Thermal time

437 – 1.8LE (r 2 ¼ 0.85) 3.5LE – 157 (r 2 ¼ 0.87)

Thermal time –

(Table 1). In that study, sowthistle plants were grown under two densities of chickpea (to decrease the light availability to 70 and 50 %) and in full sun light (as a control). The phyllochron (Rickman and Klepper, 1995) and bolting time of sowthistle increased when the light availability decreased. In contrast, the number of branches, number of capitula and, consequently, seeds in sowthistle were significantly decreased under reduced light availability. CLES model parameterization

was used to send and receive environmental information. The light model processed the received information and returned the new values to the communication module parameters carrying the output from the light model to the virtual sensors. The direction of the light source in the model was set to the x, y and z co-ordinates of 0.5, – 0.86, and 0.0, which are the noon co-ordinates of the sun during the autumn (the time of the year when the experiments were conducted) in Brisbane, Queensland, Australia. It is possible to simulate the light direction for a whole day but this has not yet been built into the current version of the CLES model. The hemispherical reflectance and transmittance of the chickpea leaflets were set to 0.089 and 0.037, respectively, according to previously published data on this topic (Gausman and Allen, 1973). In order to simulate the growth of sowthistle plants over time, the amount of light that a sowthistle plant receives had to be calculated on a daily basis. Therefore the main L-system that controls the communication of the light model with the plant models (sub-L-systems 1 and 2) was created. Sub-L-system 1 was the chickpea plant model with its leaflets interfaced with the light model. Sub-L-system 2 was the sowthistle plant model with a virtual sensor. When the virtual sensor (rectangular polygon, Fig. 2) was under full light it sent 100 % light availability to the virtual sowthistle. The amount of light availability (%) would decrease if there was a canopy of chickpea above. This decrease depended on the canopy architecture of the virtual chickpea. When the chickpea canopy was denser, less light reached the virtual sensor and thus the growth of the virtual sowthistle was suppressed more. In other words, the virtual sensor reading was scaled so that it sent the value of 100 to the sowthistle when it was under full light and lower values when it was under the canopy of chickpea. On the basis of experiments conducted previously (Cici et al., 2006) some of the components of the virtual sowthistle were also parameterized for the light environment

The final CLES model was then parameterized based on data sets collected for various chickpea cultivars. To do this, an experiment was conducted using three replicates of four chickpea cultivars (‘Macarena’, ‘Bumper’, ‘Jimbour’ and ‘99071-1001’). Five chickpea seeds were planted at a depth of 3 – 4 cm in soil contained in plastic pots (16 cm diameter and 24 cm depth) and arranged adjacent to each other on a glasshouse bench. The soil used was a light-textured loam with a neutral pH. After emergence, only one seedling was kept in each pot and the remainder were removed. The soil was watered to field capacity and was fertilized every other week with 20 mL liquid fertilizer (2.5 gL – 1 water soluble NPK, 15:18: 12). Plants were monitored every other day using a plant mapping approach (Hanan and Room, 1997) to record their leaf production, number of leaflets per leaf, time of emergence of the branches and the appearance of the first flower. At the end of the experiment, 20 leaves from each cultivar were randomly sampled and their leaflet dimensions measured using a ruler. CLES model simulation

The simulated inter- and intra-row spacings between virtual chickpea plants were 20 and 12 cm, respectively. Five rows of virtual chickpea plants were simulated with a virtual sowthistle plant growing in the middle of the 2nd and 3rd rows. The exact position of the sowthistle plant in the simulation could be anywhere between the two rows of virtual chickpea plants, but it should be the same in all the simulations for the sake of comparison. The L-studio software version 4.0 (Prusinkiewicz et al., 2000) was used to interpret and visualize the CLES model. CLES model validation

The above-mentioned chickpea cultivars growing with a sowthistle were used for model validation. The experiment

1314

Cici et al. — Canopy Architectural Modelling of Plant Competition

was designed with four replicates. Seeds of all chickpea cultivars were sown 3 – 4 cm below the soil surface and seeds of sowthistle were planted onto the surface of the soil. The soil was a light-textured loam with a neutral pH. Sufficient amount of water was supplied to moisten the soil to field capacity and plants were fertilized as described above. The size of the pots used to grow the sowthistle and chickpea plants was 20 cm in diameter and 24 cm in depth. The crop density created was about 40 m – 2 with 20 cm inter-row spacing. Two chickpea plants were grown in each pot (12 cm apart) and a 5-d-old sowthistle seedling was placed in the middle of the rows of the chickpea seedlings in a separate pot. In addition, four sowthistle plants (in separate pots) were grown in full light and used as a control treatment with their growth responses being compared to those seen in the treatments (i.e. for features such as bolting time, area of rosette leaves, final height and number of capitula produced). The bolting time of sowthistle plants was recorded and their heights were measured every 10 d using a ruler. The leaf area was measured over time using a sonic digitizer and a ruler. The total number of all capitula on sowthistle plants was counted when plants were 11 weeks old. Both the results of the height of sowthistle plants (H ) over time and the number of capitula produced per plant were used to calculate the sowthistle plant growth index (SGI). This index was produced to interpret the results more conveniently: SGI ¼

pffiffi ðEI  CÞ

where EI is elongation index of sowthistle plants [(Hfinal – Hbolting)/Hbolting]) and C is the final number of capitula produced (Cici, 2007). Using the SGI gives the advantage of taking both height and capitula number into account at the same time. The means of the observed data were compared with the numerical results obtained from the simulations. The simulation ran three times, with the randomness being in the canopy distribution of virtual chickpea plants. Goodness-of-fit (Carver, 2004) of the observed data and simulated output were expressed as root-mean-squared error (RMSE).

Sensitivity analysis

To examine the importance of different canopy architectural traits of a chickpea plant in improving its competitive ability with a weed, a sensitivity analysis was performed by changing the structure and phyllochron of a chickpea cultivar (‘96033-1014’) in an in silico study. The structural modifications were the size of leaflets, the length of internodes, the number of branches produced per plant and the number of leaflets produced per leaf (Table 2). The sensitivity analysis was done using the same number of plants and row spacing as described in the CLES model simulation section.

TA B L E 2. Manipulation of the different architectural features of chickpea canopy plants in the CLES model (40 plants m – 2, inter-row spacing ¼ 20 cm, intra-row spacing ¼ 12 cm). LL, leaflet length (cm); LN, leaflet number; IL, internode length (cm); NB, number of main lateral branches; Phyllo, phyllochron (growing degree-days, 8Cd); R (reference line), the in silico changes were established using cultivar ‘96033-1014’ Treatment no.

Chickpea architecture

LL

LN

IL

NB

Phyllo

1 2 3 4 5 6 7 8 9 10 R

Large leaflets Small leaflets Few leaflets per leaf Many leaflets per leaf Long internodes Short internodes Sparsely branched Densely branched Rapid morphogenesis Slow morphogenesis ‘96033-1014’

2 0.5 1 1 1 1 1 1 1 1 1

* * 3 16 * * * * * * *

2 2 2 2 4 1 2 2 2 2 2

4 4 4 4 4 4 1 6 4 4 4

43 43 43 43 43 43 43 43 22 62 43



 LN ¼

5 þ 0:5node; 3  node  14 14 + 1; node  14

R E S U LT S Model development and validation

The CLES model developed here was parameterized for different chickpea cultivars (Table 3) to simulate their canopy effect on the growth and development of sowthistle plants (Fig. 3). It can be seen that in comparison with full light, the predicted average bolting time of sowthistle plants growing under cultivar ‘Jimbour’ (Fig. 3A) was increased by 5 % (Fig. 4) and the growth index of sowthistle plants was decreased by 58 % (Fig. 5). ‘Jimbour’ is a cultivar with a long delay in the production of branches and with a long phyllochron (Table 3). Whilst the cultivar ‘Bumper’ (Fig. 3 B) has a long phyllochron as well (Table 3), it has the shortest delay in the production of branches (Table 3). ‘Bumper’ increased the bolting time of sowthistle plants to 720 GDD (Fig. 4). The SGI for this treatment was 65 % lower than that of full light (Fig. 5). ‘Macarena’ (Table 3, Fig. 3C) was a unifoliate cultivar with a short phyllochron, a short delay in branching and large leaf size that increased the bolting time of sowthistle plants to 740 GDD (Fig. 4) and decreased the SGI by as much as 68 % (Fig. 5). In addition, this cultivar had a negative effect on the leaf area produced by sowthistle, as had ‘Bumper’ after 70 d of growth (Table 4). This negative effect on sowthistle was more obvious in the earlier stages in ‘Macarena’ than in ‘Bumper’. Cultivar ‘99071-1001’ (Table 3) had the shortest phyllochron observed and was the most competitive cultivar (Fig. 3D). The bolting time of sowthistle plants was increased by 15 % and its SGI decreased by 76 % in the presence of this cultivar (Figs 4 and 5). Figure 3E visualizes two growth stages of sowthistle plants under full light, with the plant producing more branches and capitula in comparison with those plants grown under a chickpea

Cici et al. — Canopy Architectural Modelling of Plant Competition

1315

TA B L E 3. Input of the CLES model related to phenology and growth of different chickpea cultivars. LL, average leaflet length (cm); LW, average leaflet width (cm); IL, average internode length (cm); DEL1, delay of branching at node 1 [growing degree-days (GDD), 8Cd]; DEL2, delay of branching at node 2 (GDD, 8Cd); DEL3, delay of branching at node 3 (GDD, 8Cd); Phyllo, phyllochron (GDD, 8Cd); LN, leaflet number LN* Chickpea cultivars

LL

LW

IL

‘Jimbour’ ‘Bumper’ ‘Macarena’ ‘99071-1001’

1.0 1.2 3.1 1.1

0.5 0.7 1.4 0.6

1.9 1.8 2.6 2.2

DEL1 6 4 4 5

Phyllo Phyllo Phyllo Phyllo

DEL2 6 4 4 5

Phyllo Phyllo Phyllo Phyllo

DEL3

Phyllo

Node 14

Node .14

7 Phyllo 5 Phyllo 6 Phyllo 6 Phyllo

45 47 39 38

b5 þ 0.6nodec r 2 ¼ 0.89 b5.4 þ 0.6nodec r 2 ¼ 0.85 1 (unifoliate, simple leaf trait) b5.6 þ 0.6nodec r 2 ¼ 0.83

14 14 1 14

*bc: Truncation real values to integer.

F I G . 4. Observed and simulated sowthistle bolting time (growing degreedays, GGD) as affected by different chickpea cultivars and full light. Values are means + s.e.

experiment show that the cultivar ‘99071-1001’ is the most competitive among those examined. The simulated results fell within the spread of the experimental data and the virtual model gave good estimates of the bolting time, the SGI and the leaf area of sowthistle plants growing under different cultivar canopies (Figs 4 and 5, Table 4). Sensitivity analysis

F I G . 3. Visualizations taken from the CLES model, showing the effect of the different chickpea cultivars (green) or full light on sowthistle (red) over time. The plants on the left are at 40 d after germination and those on the right are at 70 d. The average in silico temperature was set to 20 8C per day.

canopy. It can be seen that the bolting time of sowthistle plants (Fig. 4) is the earliest, the SGI (Fig. 5) the highest, and its leaf area (Table 4) also the highest in plants grown under full light. The results of this in silico

When the virtual chickpea in the CLES model was manipulated for different canopy traits it was found that the plants with the shortest phyllochron, the greatest branch number or the greatest leaflet number per leaf could suppress the growth and development of sowthistle plants (shown here as reductions in sowthistle height) more than chickpea plants with longer stem internodes (Fig. 6). It can be seen that by doubling the length of internodes in virtual chickpea (Table 2), the height of sowthistle was suppressed by 16 % (Fig. 6). In addition, when the length of internodes was shortened (Table 2) the suppressive ability of chickpea decreased the height of sowthistle by only 2.5 %. DISCUSSION In order to simulate the effect of different chickpea cultivars on the growth and development of sowthistle plants

1316

Cici et al. — Canopy Architectural Modelling of Plant Competition

F I G . 5. Observed and simulated sowthistle growth index (SGI) as affected by different chickpea cultivars and full light. Values are means + s.e.

over time, the architectural models of chickpea and sowthistle plants were combined with a light-environment program. The final CLES model integrates a large amount of experimental data (i.e. size of individual organs) measured at frequent time intervals to present the results as images that aid rapid interpretation. One of the advantages of this type of model is that it uses directly measurable morphological traits that can be easily scored and hence used by plant breeders. The CLES model was used to simulate the effects of different above-ground canopy architectures of chickpea cultivars on the growth and development of sowthistle plants. Sowthistle plant bolting time and growth index (which combines the effect of light on stem elongation and flower production) were used for identifying the most competitive chickpea cultivar among the four studied. Both the in silico simulated and empirical data sets suggested that the most important competitive traits in the chickpea plant are having a short phyllochron, a short TA B L E 4. Observed (mean + s.e.) and simulated area (cm2 plant – 1) of rosette leaves of sowthistle plants growing with different chickpea cultivars Days after germination Treatment* Observed, ‘Jimbour’ Simulated, ‘Jimbour’ RMSE Observed, ‘Bumper’ Simulated, ‘Bumper’ RMSE Observed, ‘Macarena’ Simulated, ‘Macarena’ RMSE Observed, 99071– 1001 Simulated, 99071–1001 RMSE Observed, Control (full light) Simulated, Control RMSE * RMSE: root mean-square error.

50

70

1028 + 14 1053 + 3 24.50 1025 + 32 1038 + 1 7.00 883 + 21 975 + 3 60.20 811 + 4.27 830 + 1 20.00 1245 + 41.7 1230 + 2 54.00

1378 + 7.7 1375 + 5 5.40 1350 + 12 1354 + 4 5.00 1346 + 10 1350 + 2 7.10 1290 + 10.8 1282 + 2 5.70 1562.5 + 66.9 1400 + 1 70.55

F I G . 6. The simulated height (70 d after germination) of sowthistle when growing with different types of chickpea architecture in the CLES model. R, reference line; treatment numbers indicate different simulated architectures as shown in Table 2.

delay in branching and a large leaflet size (Table 3 and Figs 3 – 5). Cultivars with longer internodes also would be more competitive (as reviewed by Lemerle et al., 2001). However, the sensitivity analysis test suggested that other changes (such as leaf size and phyllochron of chickpea) had more effect on sowthistle (Table 2 and Fig. 6). The performance of sowthistle plants was most affected by cultivar ‘99071-1001’ (Fig 3D), a cultivar with a short phyllochron and a short delay in its ability to branch (Table 3). This is also well supported by the results of the sensitivity analysis (Fig. 6). When the phyllochron of the virtual chickpea plant was decreased from 43 to 22 GDD the height of the virtual sowthistle plant was suppressed nine-fold. This reaffirms the significance of early canopy closure in chickpea crops to attain better sowthistle control. The second ranked cultivar for competitive ability in both simulated and observed experiments was ‘Macarena’ (Fig. 3C and Table 4). ‘Bumper’ (Fig. 3B), the third most competitive cultivar (Figs 4 and 5, Table 4), was a plant with a long phyllochron that was compensated for by having a short delay in its ability to produce branches; this was also the shortest cultivar in stature. Having the longest delay in its ability to produce branches, ‘Jimbour’ (Fig. 3A) was the leastcompetitive cultivar (Figs 4 and 5, Table 4). Thus, it could be seen that the competitive ability of a crop cultivar cannot be described by a single trait and the CLES model may be an invaluable computer tool in helping to find more competitive cultivars from those that are already present. Recently, a study on the weed-suppressing ability of several chickpea cultivars (Paolini et al., 2006) also showed that cultivars with a rapid early growth and an erect canopy architecture were those that were the most competitive. These conclusions were consistent with those of recently published reports on other crops (Williams et al., 2006). It is anticipated that the CLES model could be used as a tool to aid with the direction of plant breeding programs designed to develop more competitive chickpea crops. The model takes into account direct morphological traits and one can see the results immediately. It is worth noting that breeders are unlikely to select a new cultivar solely on its competitive ability, but the CLES will help them notice what the effect of changing a specific architectural characteristic will be on its weed-suppressing capacity.

Cici et al. — Canopy Architectural Modelling of Plant Competition Application of the CLES model

The potential power of the CLES model becomes more obvious when we consider using it to simulate the effect of different combinations of cultural practices, such as different relative times of weed emergence, crop seeding rate, crop row spacing and choice of cultivar (as shown above) on the performance of weeds in different environments (examples can be found in Cici, 2007). Crop density can be modelled by incorporating chickpea response to intra-specific density and changing the parameters that control the seeding rate and the row spacings in the model. The choice of cultivars can be modelled by changing model parameters such as the phyllochron, delay of branching, the size of leaflets and the length of internodes (as shown above). Environmental effects (such as those due to temperature) can be modelled by changing the growing degree-days the model uses. The sensitivity analysis can help to assess how much change in an architectural trait would be significant to improve its competitive ability. However, in real plants a single trait is determined by its physical links with others. This physical linkage of traits does not easily allow for an independent plant breeding changes to be made (Bastiaans et al., 1997). Models are simplified versions of the real world in which all the physical linkages are not well known, so models cannot completely account for certain kinds of side effects. This can be seen in the chickpea cultivars examined here; when the phyllochron was longer, plants had a shorter delay in branching (such as ‘Bumper’), and when they had larger leaflets their numbers were fewer (such as ‘Macarena’). The present work provides a computer-modelling approach to help improve our understanding of weed suppression by crop canopies through different in silico experiments. Selection of more competitive cultivars will help ensure better weed suppression in an IWM program and ultimately reduce the need for herbicide application. The long computation time required to run the simulations is a limiting factor in the present model (depending on the age of plants, each simulation takes about 2.5 – 4.0 h to produce the visualizations and numerical outputs) but advances in grid computing (Berman et al., 2003) are quite promising, as they allow larger number of simulations to be run in a shorter time. Indeed, by using a gridcomputing approach it may be possible to test hundreds of ‘what if’ scenarios in just a few hours. A program such as Nimrod (http://www.csse.monash.edu.au/~davida/ nimrod/index.htm; Lee et al., 2006) could be used to distribute the simulations over the grid. There is no limit to the application of computer models, in conjunction with field experiments, in addressing agricultural and weed science problems. However, developing realistic architectural models of plants is a tedious task since it needs extensive experimental data over time. This present research forms the basis of an architectural plant modelling approach that would enable a weed researcher or breeder to conduct in silico experiments before stepping out into the field (Hanan and Room, 1997). In other words, it can help one make a better experimental design and try different experimental combinations before conducting an actual experiment in the field.

1317

AC KN OW L E DG E M E N T S We thank M. Farri for supporting and helping S.Z.H. Cici during this work, M. Cieslak and T. Rudge for the light model, and the Ministry of Science in Iran and CRC for Australian Weed Management for a scholarship and some research funding for S.Z.H Cici, respectively.

L IT E RAT URE C IT E D Aphalo PJ, Ballare CL. 1995. On the importance of information acquiring systems in plant–plant interactions. Functional Ecology 9: 5–14. Bastiaans L, Kropff MJ, Kempuchetty N, Rajan A, Migo TR. 1997. Can simulation models help design rice cultivars that are more competitive against weed? Field Crop Research 51: 101–111. Berman F, Fox G, Hey A. 2003. Grid computing, making the global infrastructure a reality. Chichester, West Sussex, UK: Wiley. Blackshaw RE, O’Donovan JT, Harker KN, Clayton GW, Stougaard RN. 2006. Reduced herbicide doses in field crops: a review. Weed Biology and Management 6: 10– 17. Carver RH. 2004. Doing data analysis with Minitab 14. Victoria, Australia: Thomson. Chelle M, Andrieu B. 1999. Radiative models for architectural modelling. Agronomie 19: 225– 240. Cici SZH. 2007. Competitive ability of chickpea canopy with sowthistle: an architectural modeling study. PhD thesis, University of Queensland, Brisbane, Australia. Cici SZH, Hanan J, Sindel BM, Adkins S. 2006. Phenotypic plasticity of sowthistle and its offspring in response to light availability. In: Preston C, Watts JH, Crossman ND, eds. Proceedings of the Fifteenth Australian Weeds Conference. Weed Management Society of South Australia, Adelaide, 172– 175. Gausman HW, Allen WA. 1973. Optical parameters of leaves of 30 plant species. Plant Physiology 52: 57– 62. Hanan J. 1992. Parametric L-systems. PhD thesis, University of Regina, Regina, Canada. Hanan J, Room PM. 1997. Practical aspects of virtual plant research. In: Michalewicz MT, ed. Plants to ecosystems. Advances in computational life sciences. Melbourne, Australia: CSIRO Press, 28–44. Holm LG, Plucknett DL, Pancho JV, Herberger JP. 1988. The world’s worst weeds: distribution and biology. Ann Arbor, MI: University of Michigan Press. Jordan N 1993. Prospects for weed-control through crop interference. Ecological Applications 3: 84– 91. Lee B, Tang M, Zhang JH, Soon OY, Zheng C, Arzberger P, Abramson D. 2006. Analysis of jobs in a multi-organizational grid test-bed. In: Joita L, ed. 6th IEEE International Symposium on Cluster Computing and the Grid, Singapore. Los Alamitos, CA: IEEE, 1110– 1125. Lemerle D, Gill GS, Murphy CE, Walker SR, Cousens RD. 2001. Genetic improvement and agronomy for enhanced wheat competitiveness with weeds. Australian Journal of Agricultural Research 52: 527– 584. Lemieux C, Cieslak M, Luttmer K. 2004. RandQMC user’s guide. Calgary, Canada: University of Calgary. Lindenmayer A. 1968. Mathematical models for cellular interaction in development. Parts I and II. Journal of Theoretical Biology 18: 280– 315. Mech R. 1997. Modelling and simulation of the interaction of plants with the environment using L-Systems and their extensions. PhD thesis, University of Calgary, Calgary, Canada. Mech R, Prusinikiewicz P. 1998. User’s manual for environmental programs. Calguary, Canada: University of Calguary. Olesen JE, Hansen PK, Berntsen J, Christensen S. 2004. Simulation of above-ground suppression of competing species and competition tolerance in winter wheat varieties. Field Crops Research 89: 263– 280. Paolini R, Faustini F, Saccardo F, Crino P. 2006. Competitive interactions between chickpea genotypes and weeds. Weed Research 46: 335– 344.

1318

Cici et al. — Canopy Architectural Modelling of Plant Competition

Prusinkiewicz P. 1998. Modeling of spatial structure and development of plants: a review. Scientia Horticulturae 74: 113– 149. Prusinkiewicz P, Lindenmayer A. 1996. The algorithmic beauty of plants. New York: Springer-Verlag. Prusinkiewicz P, Karwowski R, Mech R, Hanan J. 2000. L-studio/cpfg: a software system for modeling plants. In: Nagl M, Schu¨rr A, Mu¨nch M., eds. Applications of graph transformations with industrial relevance. Berlin: Springer, 457–464. Rickman RW, Klepper BL. 1995. The phyllochron: where do we go in the future? Crop Science 35: 44–49. Schwinning S, Weiner J. 1998. Mechanisms determining the degree of size asymmetry in competition among plants. Oecologia 113: 447– 455.

Singh KB. 1997. Chickpea (Cicer arietinum L). Field Crops Research 53: 161–170. Swanton CJ, Murphy SD. 1996. Weed science beyond the weeds: the role of integrated weed management (IWM) in agroecosystem health. Weed Science 44: 437– 445. Weiner J, Griepentrog HW, Kristensen L. 2001. Suppression of weeds by spring wheat Triticum aestivum increases with crop density and spatial uniformity. Journal of Applied Ecology 38: 784–790. Williams MM, Boydston RA, Davis AS. 2006. Canopy variation among three sweet corn hybrids and implications for light competition. HortScience 41: 1449– 1454.