Informatica 28 (2004) 387-392 387
Multi-Attribute Modelling of Economic and Ecological Impacts of Cropping Systems
Marko Bohanec, Sašo Džeroski and Martin Žnidaršič
Institut Jožef Stefan, Department of Knowledge Technologies, Jamova 39, SI-1000 Ljubljana, Slovenia
marko.bohanec@ijs.si
Antoine Messéan
CETIOM, Centre technique interprofessionnel des Oléagineux métropolitains and
INRA Eco-Innov, POB BP1, Centre de Grignon, FR-78850 Thiverval, Grignon, France
Sara Scatasta and Justus Wesseler
Wageningen University, Environmental Economics and Natural Resources Group,
De Leeuwenborch, Hollandseweg 1, NL-6706 KN Wageningen, The Netherlands
Keywords: qualitative multi-attribute modelling, genetically modified plants, cropping system, ecology, economy
Received: June 27, 2004
Modelling of economic and ecological impacts of genetically modified crops is a demanding task. We present some preliminary attempts made for the purpose of the ECOGEN project "Soil ecological and economic evaluation of genetically modified crops". One of the goals of the project is to develop a computer-based decision support system for the assessment of economic and ecological impacts of using genetically modified crops, with special emphasis on soil biology and ecology. The decision support system will be based on a rule-based model incorporating both economic and ecological criteria. In this paper we present some preliminary results of developing the integral model and describe four specific sub-models. The first two sub-models are concerned with ecology and assess the ecological impacts of various types of weed and pest control, respectively. The other two sub-models assess the economic impacts of cropping systems at the farm and regional level, respectively. All the models were developed using a qualitative multi-attribute modelling methodology, supported by the software tool DEXi. Povzetek: članek opisuje modeliranje ekoloških problemov kmetijstva.
2. Provide an ecological risk assessment of a
1 Introduction
The possible use of genetically modified (GM) plants in agriculture needs in-depth investigations of ecological and economic consequences [1,2]. The investigations are important for both the European Commision (EC), who needs specifications for GM-plant risk assessment, and to farmers and the public who are concerned about the possible ecological and economic implications. Crop production involves complex decision-making processes, which require and justify the application of decision support systems [3].
The ECOGEN project [2] (Soil ecological and economic evaluation of genetically modified crops) is an EC-funded project aimed at combining simple lab tests, studies of multi-species model ecosystems, and field studies to acquire mechanistic and realistic knowledge about economic and ecological impacts of GM crops on the soil. Economic trade-offs are assessed and related to ecological effects. The economic and ecological knowledge gained in ECOGEN will be combined into a rule based model for a decision support tool. The goals of the ECOGEN project are to: 1. Provide ecological and economical assessment and comparison of integrated cropping systems using GM or conventional crops, respectively.
GM
cropping system and a conventional cropping system for the soil ecosystem based on single species tests, multispecies tests and long-term field investigations.
3.	Adapt existing ecotoxicity testing tools to GM plant material and validate their use.
4.	Provide economic assessment of GM crops and conventional crops with respect to a quantification of the expected trade-offs between the two and the implications for the EU Agriculture Policy.
Finally, we wish to incorporate ecological knowledge from single species tests, multispecies tests, and field investigations, as well as economic information from farming practices into a rule-based model to be used for predictions of economic decision-making processes and ecosystem behaviour.
In this paper, we present our preliminary attempts at this kind of modelling. We describe four qualitative multiattribute models. Two of these models assess the ecological impacts of using various cropping systems that differ in the applied weed-control and pest control mechanisms, respectively. The other two models assess the economic impacts of cropping systems; these are assessed at farm and regional level, respectively.
388 Informatica 28 (2004) 387-392
M. Bohanec et al.
2 Methodology
The goal of the project is to build an integrated rule-based model for assessing the sustainability of farming (GM and non-GM) taking into account ecological and economic aspects. On the ecological side, this includes a model of the impact of GM crops and pesticides on non-target organism and soil functions. The model will be hierarchically structured, with submodels for different aspects, e.g., a submodel for economic (ECONOMY) and a submodel for ecological (ECOLOGY) aspects. In general, then, the approach will involve the following components (Figure 1): 1. Cropping systems: Input items assessed by the model. Each cropping system is described by a vector of values, such as: crop type (e.g., corn), soil preparation (e.g., type of tillage), weed control (e.g., use of herbicides), pest control (e.g., use of pesticides), type and quantity of fertilization, soil characteristics, climate characteristics, economic indicators (e.g., involved yields and variable costs). Multi-attribute model: A model that aggregates the characteristics of cropping systems into overall ecological and economic evaluations. Outputs: Two assessments are obtained for each cropping system: ecological and economic impacts.
2.
3.
OUTPUTS
ECOLOGY
ECONOMY
MODEL
INPUTS
Figure 1: A general approach to multi-attribute assessment of cropping systems.
Using this schema, it will be possible to evaluate each cropping system and its impacts for several consecutive years, basically obtaining a chart as
sketched in Figure 2. In addition, the model will facilitate all analyses and reports typically available in multi-attribute modeling [4,5]: what-if analysis, sensitivity analysis, simulation, selective explanation, and various visualizations.
The integral multi-attribute model will be mainly rule-based and will contain further submodels, which will be both qualitative (using rules) and quantitative (numerical/equations). They will be developed by the soil biology experts in the respective subareas, and in intensive interaction and collaboration with the decision support/data analysis experts. Decision support methods that rely on manual knowledge acquisition from domain experts will be used to elicit existing knowledge. Techniques from the area of multiattribute decision-making and support will be used to support the construction of the overall model. Where enough data are available, some submodels will be generated in a (semi)automated fashion by data analysis. In particular, machine learning techniques will be used to construct some submodels by analysing available data. Some sub-models of this type have already been developed in this way [6]. Reasoning with the rule-based model for decision support is crisp by default, but can be extended to fuzzy reasoning with moderate effort.
ECONOMY
CSi initial
CSi year 2
CSi "year 3
CSi year 1
ECOLOGY
Figure 2: An example assessment of a cropping system CSi through three consecutive years.
So far, we have developed two models for the ecological assessment of cropping systems, dealing with weed-control and pest-control mechanisms, respectively. These are described in section 3. In addition, we have developed two models for the assessment of economic impacts of cropping systems. The first model assesses the impacts at the farm level. The second model is an extension and adaptation of the
MULTI-ATTRIBUTE MODELLING OF.
Informatica 28 (2004) 387-392 389
first one so as to assess the economic impacts at the regional level. These are presented later in section 4. All these models are hierarchical, qualitative and multiattribute. Thus, they are characterised by the following [4]:
•	Each model consists of a number of hierarchically structured variables called attributes.
•	Terminal nodes of the hierarchy represent input attributes; each cropping system is described by a vector of values of input attributes.
•	Input attributes are aggregated through several levels of aggregate attributes into the overall assessment, which is represented by a single root attribute.
•	All the attributes in the model are qualitative, meaning that they take symbolic values, described by words.
•	The aggregation of values in the model is defined
by rules.
The models were developed using the software tool DEXi [7]. DEXi facilitates the development of a tree of attributes, definition of aggregation rules (e.g., see Figure 4), evaluation of options (cropping systems in this case), what-if analysis and charting.
3 Ecological assessment
3.1 Weed-control model
With this model, cropping systems are assessed qualitatively using the five-value ordered scale: preferrable, acceptable, regular, poor, unacceptable. The model is hierarchical and has the structure of attributes as shown in Figure 3.
Figure 3: The hierarchical structure of the weed-control model.
390 Informatica 28 (2004) 387-392
M. Bohanec et al.
Input attributes. The assessment of cropping systems is based on six input attributes:
1.	Weedcontrolstrategy: the strategy of controlling weeds, either for conventional crops or for GM crops. The GM crops considered in this study are herbicide-tolerant (GMHT). There are six different strategies:
a.	Non-GM and simple pre-sowing herbicide application (pre_sowing trifluralin);
b.	GMHT with one fall application of herbicide;
c.	GMHT with fall + spring applications of herbicide;
d.	Non-GM with pre-sowing+pre-emergence applications of herbicide;
e.	Non-GM with pre-sowing+post-emergence applications of herbicide;
f.	Non-GM with post-emergence application of herbicide only.
2.	HTherbicide: the application of HT herbicide: glyphosate, none, glufosinate, or generic.
3.	N_fertilizer: the application of nitrogen fertilizer: high (>200 kg/ha), medium (150-200 kg/ha), or low (<150 kg/ha).
4.	Weeddiversity: the diversity of weed at the location studied: high, medium, or low.
5.	Other_costs: relative estimation of marginal costs other than weed control costs: high, medium, or low.
6.	Fixedcosts: relative estimation of the fixed cost of production: high, medium, or low.
Aggregate attributes. The aggregate (intermediate) attributes are grouped into three main subtrees:
1.	Weed profile is an aggregate sub-model that affects several other parts of the model. Basically, it defines the weed profile according to the abundance and diversity of weeds.
2.	Gross margin is estimated on the basis of Production value and Variable costs of production. Production value depends on yield and quality, which in turn depends on purity and technological properties of production. Variable costs are estimated on the basis of herbicide costs, seed costs and weed profile.
3.	Ecological benefit is estimated according to water quality, insects and weed diversity. The effect on insects is assessed through weed profile.
All the aggregate attributes in the model are assessed according to rules defined by an expert. Figure 4 shows two such rulesets, conveniently presented in a tabular form. The bottom ruleset defines the mapping between the input attribute Weed_control_strategy to the
aggregate attribute Abundance. The top ruleset defines the rules that combine Abundance and Weeddiversity into the aggregate attribute Weed_profile.
Abundance weed_diversity Weed_Profile
	22%	78%	
1	high	<=medium	high_potential_problems
2	<=medium	high	high potential problems
3	medium	medium	regular_problems
4	low	high	regular_problems
5	low	>=medium	low-problematic
6	<=medium	low	specific flora problems
weed_control_strategy_Abundance
100%	
1 simple pre-sowing	high
2 HT one fall application	medium
3 >=pre-sowing+pre-emergence	medium
4 HT fall + spring applications	low
Figure 4: Two tables of aggregation rules: for Weed profile and Abundance of weed.
3.2 Pest-control model
In addition to the weed-control model, we have also developed a pest-control model. Its structure is similar to the weed-control model, with the following differences:
•	The subtree Weed control is replaced by a subtree Pest control, having similar structure, but different attribute values and aggregation rules.
•	Two new attributes are added to the node
Ecological benefit: Greenhouse gasses and Soil. Soil is in turn composed of Soil fauna and Soil quality.
•	The attribute Purity is replaced by Damage.
4 Economic assessment
Figure 5 shows the structure of the ECONOMY submodel for the assessment of economical indicators. This model assesses gross margins at the level of a single farm for Bt-corn. [Bt-corn has been genetically engineered to produce an insecticide known as Bt-toxin, produced by a naturally occuring soil organism Bacillus thuringiensis (Bt).] In addition to the ECONOMY submodel, some possible links for the ECOLOGY submodel are also shown in Figure 5.
MULTI-ATTRIBUTE MODELLING OF.
Informatica 28 (2004) 387-392 391
Figure 5: Hierarchical structure of the ECONOMY model (Bt-corn, farm level).
Basically, the farm level impacts in terms of economy depend on Production value and Variable costs. The former can be determined on the basis of Price and Yields, where Price depends on the Quality of production.
Variable costs incorporate the costs of Pesticides, Seeds, Fuel, and Water. Each of these costs directly depends on the use of the respective item: pesticides, seeds, fuel, and costs. The used quantities of each of these items, as well as Yields and Quality, directly depend on the cropping system employed at the farm level. Thus, these four variables form a group referred to as Crop management. Notice that some of these indicators affect the ECOLOGY part of the model, too. For instance, Pesticide use influences Water quality and Soil biodiversity. The latter is affected by Seed use, too. Fuel use influences Greenhouse gasses. The lowest level of attributes form the group called Regional profile. These attributes describe the properties of a particular region in terms of: Productivity, Pest abundance, Purpose of maize (which can be used for grain or feed). Productivity depends on the characteristics of Soil and Climate in the region. Notice that all the three main attributes of Regional
characteristics affect the Crop management group, and that two of them (Pest abundance and Purpose) additionally affect Soil biodiversity. When we move from the farm level to the regional level, a new important factor comes into play: Adoption rate. Namely, when assessing a cropping system at the farm level, it is clear whether Bt-corn has been adopted in that system or not. It can only be adopted or not adopted, there are no intermediate choices. However, when assessing a cropping system at the regional level, it becomes important which proportion of the farms have adopted Bt-corn, because the adoption rate influences the Regional yields and Regional costs of the crop. Also, the adoption rate itself can be influenced by the Price in the market, which introduces a cycle into the model.
All this is reflected in the strucure of the ECONOMY model for Bt-corn at the regional level (Figure 6). The structure is very similar to the structure of the model at the farm level (Figure 5), except that there is an additional block appearing above the Crop management block. This new block assesses Adoption rate based on Field trial yields and Field trial costs. Both of these variables depend on the indicators of
392 Informatica 28 (2004) 387-392
M. Bohanec et al.
Crop management, which in turn depend on the assessed cropping system. Once determined, Adoption rate directly influences Regional yields and Regional costs.
Figure 6: Hierarchical structure of the ECONOMY model (Bt-corn, regional level).
Second, the developed models are truly hierarchical (as opposed to traditional tree-like structure) and involve some very complex relationships between attributes, even cycles. These characteristics exceed the capabilities of currently available supporting software, which will have to be accordingly modified and extended. Last but not least, the developed models provide just a part of the final integral model, which will address additional cropping system control mechanisms and additional GM crops.
Acknowledgement
This work was supported by the ECOGEN project, funded by the Fifth European Community Framework Programme: Quality of life and management of living resources, under contract QLK5-CT-2002-01666.
References
[1]	Birch, A.N.E., Krogh, P.H., Cortet, J., Tabone, E., Griffiths, B.S., Dzeroski, S., Wesseler, J., Gomot de Vaufleury, A., Badot, P-M., Andersen, M.N., Messean, A. (2003). ECOGEN: Soil ecological and economic evaluation of genetically modified crops. Poster at "Biodiversity Implications of Genetically Modified Plants", September 7-12, 2003 Monte Verita , Ascona, Switzerland Centro Stefano Franscini, Swiss Federal Institute of Technology Zürich.
[2]	ECOGEN: Soil ecological and economic evaluation of genetically modified crops. http://www.ecogen.dk
[3]	Recio, B., Rubio, F., Criado, J.A. (2002): A decision support system for farm planning using AgriSupport II. Decision Support Systems 36, 189— 203.
5 Conclusion
The modeling of economic and ecological impacts of genetically modified crops is inherently difficult. It requires knowledge from different fields and disciplines, which is scarce and largely unknown. It also requires complementary approaches, such as a combination of data mining and expert modeling, which has been attempted in ECOGEN. The benefits of modeling, however, are manifold, as it facilitates various computer-based assessments, evaluations, analyses and simulations. The results are eagerly awaited by European administration, politicians, ecologists, farmers and the interested public. The models presented in this paper provide a preliminary step in this direction. They are in an early development stage and a lot of further work is expected. First, the models should be tested using real field data which is being collected. The models and their results should be evaluated by relevant experts.
[4]	Bohanec, M. (2003): Decision support. In: Mladenič, D., Lavrač, N., Bohanec, M., Moyle, S. (eds.) Data mining and decision support: Integration and collaboration. Kluwer Academic Publishers.
[5]	Triantaphyllou, E. (2000): Multi-Criteria Decision Making Methods: A Comparative Study. Kluwer Academic Publishers.
[6]	Demšar, D., Džeroski, S., Larsen, T., Krogh, P.H. (2003): Identifying most important agricultural factors for soil community of microarthropods. In Proc. of the 12th International Electrotechnical and Computer Science Conference. 25.-26. September 2003, ERK'2003, Ljubljana.
[7]	Jereb, E., Bohanec, M., Rajkovič, V. (2003): DEXi: Računalniški program za večparametrsko odločanje [DEXi: Computer program for multi-attribute decision making]. Moderna organizacija, Kranj, Slovenia.