Combining simulation models and genetic algorithms to improve design of arable cropping systems

Aim

Evaluation of efficiency and effectivity of various optimization methods based on Evolutionary Algorithms for designing promising crop rotations. The methodological issue is elaborated in a major case study for north-east Australia. Minor case studies for western European conditions are developed with MSc students.

Cooperation

Agricultural Production Systems Research Unit, Toowoomba, Qld. Australia

Plant Research International
WU-Computer Science

Progress

The interest in sustainability of agricultural systems (e.g. Agenda21) signals a change in the role of agriculture in society. Aspects such as food quality, water utilisation, maintenance of attractive landscapes and equal opportunities for men and women have gained a place in the public debate besides the volume produced. In this debate, focus often is more on 'means' than on 'ends', contributing to muddled processes and polarisation. Tools are needed to distinguish objectives related to sustainability and to explore consequences of pursuing these at different spatial scales. The development of such a tool at the cropping system level is subject of this paper.

At the cropping system level, models that capture key bio-physical components of agro-ecosystems are well suited for exploring more sustainable systems and are being used successfully as part of processes in which stakeholders redesign cropping systems. In our research groups, two types of models have been used: mechanistic, dynamic simulation models as part of the Agricultural Production Systems Simulator APSIMand interactive multiple goal linear programming models based on MGOPT. Conceptually, these models belong to different domains, APSIM being a process based simulation model and MGOPT an optimisation model. In APSIM, carbon, nitrogen and water dynamics are represented by differential equations that are solved numerically with time steps of one day or less. Management options, such as crop and cultivar choice, sowing date, rate and density, and nitrogen fertiliser application rates are defined by the user. The model can be used to analyse field experiments or to conduct 'what-if' explorations of management systems. In MGOPT, the linear programming algorithm assembles rotations of crops and inter-crop measures from building blocks, called activities, that describe all inputs and outputs needed to grow a particular crop or to apply a particular inter-crop measure in a particular environment. The environment includes the carry-over effects of previous crops or inter-crop measures. The approach is static, and implies that states are stationary over subsequent rotation cycles.

In contributing to learning and design processes, process based simulation and linear programming exhibit complementary features. While in linear programming the input-output relations describing crop management are not readily transparent, results of process based simulation can be understood from underlying bio-physical processes. On the other hand, linear programming enables evaluation of a potentially very large set of alternative cropping systems to reveal optimal designs and trade-offs among objective functions, while simulation does not address the question of optimality. Both features, understanding of the causes of observed phenomena and benchmarking, are important components of learning and design processes. Aim of this project is to present a method that combines the strengths of a process based simulation model with optimisation to enhance the design of sustainable cropping systems.

Two evolutionary algorithms described in literature on computer design and industrial assembly line design are adapted to optimise cropping systems design problems that are characterised by multiple objectives and large search spaces. The case study used to illustrate and compare the methods and their results, addresses crop choice and sowing rule in north-east Australian cropping systems. Sustainability of these systems is evaluated in terms of economic performance (gross margin, financial risk) and resource use (erosion). Computational intensity of the method is addressed by parallel model evaluation with central control. The project has been started during a sabbatical leave of Rossing, and is continued with financial support of the Australian Grains Research Development Corporation (GRDC).

Publications

• Coomans, E., 2000. Zijn genetische algoritmen toepasbaar voor het optimaliseren van gewasrotaties: een case studie met het model Rotask. MSc Thesis Wageningen University, 60 p.
• Rossing, W.A.H., P.S. Carberry, P. Devoil, G.L. Hammer & H. Meinke, 2001. In search of sustainability: combining simulation models and genetic algorithms to improve design of arable cropping systems in north-east Australia. In prep.

Poster

Optimising profitability and environmental trade-offs in managing cropping systems using differential evolution (download poster in PDF-format)