Utilizing model to optimize crop plant density: a case study on tomato

Transcription

1 Utilizing model to optimize crop plant density: a case study on tomato LiLi Yang 1 QiaoXue Dong 1 Daoliang Li 1 * College o Inormation and Electrical Engineering, China Agricultural University, Beijing, , China. * li.daoliang@163.com Abstract. Based on the unctional-structural plant model GreenLab, a common plant category tomato was selected as a test case in order to analyze the impact o plant density on the yield. A yield optimization model was set up based on minimum ripe ruit weight constraints, ruit weight serve as the objective unction or the bound-constrained optimization problem. Particle Swarm Optimization (PSO) algorithm was applied or maximizing the crop production, which will provide a theoretical reerence or agronomy measures based on plant spacing. Keywords: Tomato, Plant density, GreenLab model, Yield optimization, PSO 1 Introduction Tomato s economic beneit is related to ruit yield and quality, and these two actors are considerably aected by light condition [1][2]. Manipulation o plant spacing is a possible means to change light interception or peasant. Traditionally, plant density is based on empirical knowledge rom iled experiments, and this is quite time consuming and expensive. While plant growth models provide a very important tool or realizing optimal yields o crops with an eicient means [3]. Functional-structure model GreenLab can simulate plant growth based on organ size, so it provides possibility or plant density optimization [4]. Combining plant mathematical model with optimization method can provide guiding or the optimization o planting density and horticultural practice such as pruning and environmental control in special constrained environment. 2 The description o optimization problem 2.1 Greenlab model In GreenLab model, time scale, called a growth cycle, is oten the phyllochron between appearances o two successive leaves in main stem [5]. Plant total product

2 biomass at growth cycle n is Q(n). S( n) A Q( n) AB 1 (1) A is the projection surace o single plant (m 2 /m 2 ), which is in response to plant density. B is plant growth potential multiplied by an empirical resistance, which can be computed by environment data. Sn ( ) is total leaves surace at n th growth cycle. D(n) is all organ demands at the n th growth cycle n D( n) P ( t) N ( n t 1) (2) o o o o t 1 Where P o (unitless) is the sink o organ o ( o=internode (i), blade (b), petiole (p), ruit ()), which is deined as the ability o organs to compete or biomass. N ( 1) o n t is the number o organ o in t growth cycle. O (t) is sink variation o organ o at t th growth cycle, which is described by an empirical Beta unction. α o and β o are the coeicients o the Beta unction, and α o is set constant. A, P o (unitless) and β o (unitless) are hidden parameter. These parameters were identiied by itting the periodical destructively-measured data and environment data with the corresponding model output. 2.2 Fruit weight calculation At the end o n th growth cycle, biomass or one chronological aged j (growth cycle) ruit can be calculated as below: j Q( n j k) q ( n, j) p ( k) (3) k 1 D ( n j k ) k is ruit growth cycle, all organs including ruit compete or biomass in a common pool. The ormula or calculating the total ruit weight o an aged n (growth cycle) plant as ollows: Q Y p ( k) (4) D tx_ n ( tx_ k ) k 1 n ( tx_ k ) t x_ is ruit extension time, which is an observation value. 2.3 Model parameter estimation o Greenlab or tomato For establishing the relationships between GreenLab model parameters and plant density, our gradient density tomatoes named by D1,D2,D3,D4 rom low density to high density were planted in solar greenhouse at the Chinese Academy o Agricultural Science in Beijing (39.55 N, E)[6]. The generalized least square method was used or itting [6]. Four plant density model parameters are shown in table 1:

3 Table 1. Estimated parameter values o model Parameters Values D1 D2 D3 D4 P i P p P β i β p β b β A An equation is used to describe model parameter variation trend with plant density. The supreme itting degree unction is shown as below: a bexp( cx) (5) x is plant density, a, b,c are constants which can be got by itting, Ωis GreenLab parameters as table1 listed. 3 Tomato optimal plant density solution 3.1 Tomato growth simulation at dierent plant density Dierent density plants growth and development can be simulated using GreenLab model with parameters listed in table 1. resh weight(g) D1 sim D2 sim D3 sim D4 sim D1 obs D2 obs D3 obs D4 obs ruit growth cycle Fig. 1. Fitting results on ruit resh weight evolution or a ground area plant o our densities symbols- measurement, lines- model Model outputs show that increasing plant density results in a reduction o single plant

4 ruit resh weight (Figure is not shown) but in an increase o ruit resh weight on a ground area (Fig.1), which is consistent with measured data. 3.2 Mathematical solution or the tomato plant density optimization problem Considering market economic beneit, we deine a minimum harvestable ruit weight as constraint. A single object non-linear optimization unction is built as ollow, max Y S( d, ) d R G gmin Optimal object is maximum yield Y. Plant density d is an optimizing object variable and it varies in real ield. All the GreenLab parameters Ωare control variables. g min is a threshold value o tomato ruit reerred to practice(100g). G is single ruit weight. The PSO(Particle Swarm Optimization) algorithm was implied to optimize. The ruit weight output with respect to dierent plant density is shown as Fig 2. Fruit yield variation with plant density can be divided into three phases. During the irst phase (1 to 8.2 plants per square meter) ruit yield grow bigger with density increasing. During the second phase (8.2 to 10.5 plants per square meter) ruit yield grow less with density increasing. In the ollowing, ruit yield drop to zero. Because ruits get smaller and smaller with later density increasing, even it can t up to harvestable standard. Optimal plant density is 8.2 plants per square meter, and the ruit weight is close to 5kg per square meter. Fig. 2. Tomato yield evolution with planting density 4 Conclusion Model output is well conormity with measurement. Results illustrate that model has the ability o relecting plant density eect on yield in aspect o biomass production, distribution and organ number changing. We can iner model parameter,

5 plant growth and development process in any environment rom model calibration and parameter interpolation. Combining with optimization algorithm, we can get optimum plant conditions (in this paper plant density as an example). It provides a reerence or decision-makers using limited experiment, and it can be used in plant growth prediction, manuacture management and so on. Acknowledgments. This research was inancially supported by Hi-Tech Research and Development (863) Program o China (#2008AA10Z218), NSFC ( ) and the Start Fund or Scientiic Research, CAU (2011JS156). Reerences [1] Bertin N, Competition or Assimilates and Fruit Position Aect Fruit Set in Indeterminate Greenhouse Tomato. Annals o botany, 1995, 75:55-65 [2] Papadopoulos AP, Pararajasingham S. The inluence o plant spacing on light interception and use in greenhouse tomato (Lycopersicon esculentum Mill.): A review. Scientia Hortticulturae, 2003, 69:1~29. [3] J. Vos, L.F.M. Marcelis, P.H.B. de Visser, et al. Functional-Structural Plant Modeling in Crop Production. In:J.B. Evers eds. Netherlands: Springer, [4] Rui Qi, Véronique Letort, Mengzhen Kang et al. Application o the GreenLab Model to Simulate and Optimize Wood Production and Tree Stability: A Theoretical Study. Silva Fennica, 2009, 43(3):465~487. [5] Guo Y, Ma YT, Zhan ZG, Li BG, Dingkuhn M, Luquet D, De Reye P. Parameter optimization and ield validation o the unctional structural model GREENLAB or maize. Annals o Botany, 2006,97: [6] Yang L L, Wang Y M, Kang M Z, et.al. Relation between Tomato Fruit Set and Trophic Competition-a Modelling Approach.Plant Growth Modeling and Application Procedding[A]. Third International Symposium on Plant Growth Modeling, Simulation, Visualization and Applications, PMA 09[C]. Beijing, IEEE CPS, 2009,