Relevancy and role of whole-farm models in supporting smallholder farmers in planning their agricultural season

Environmental Modelling & Software 68 (2015) 147e155

Contents lists available at ScienceDirect

Environmental Modelling & Software journal homepage: www.elsevier.com/locate/envsoft

Relevancy and role of whole-farm models in supporting smallholder farmers in planning their agricultural season A.W. Sempore a, c, *, N. Andrieu b, d, H.B. Nacro c, M.P. Sedogo e, P.-Y. Le Gal b Centre International de Recherche-d
eveloppement sur l'Elevage en zone Subhumide (CIRDES), 01 BP 454 Bobo-Dioulasso, Burkina Faso Centre International de Recherche Agronomique pour le D
evelomment (CIRAD), UMR Innovation, F-34398 Montpellier, France c Institut du D
eveloppement Rural (IDR), Universit
e Polytechnique de Bobo-Dioulasso (UPB), 01 BP 1091 Bobo-Dioulasso, Burkina Faso d Centro Internacional de Agricultura Tropical (CIAT) e Cali, Apartado A
ereo 6713, Colombia e Institut de l'Environnement et de Recherches Agricoles (INERA), Kamboinsin, Ouagadougou, Burkina Faso a

b

a r t i c l e i n f o

a b s t r a c t

Article history: Received 22 May 2014 Received in revised form 21 January 2015 Accepted 18 February 2015 Available online 7 March 2015

The way a model is designed to assist farmers in their decision-making may influence how it is understood and perceived by farmers and shape interactions between farmers and model users (researcher, advisor). This study compared the strengths and weaknesses of three types of whole farm models used by researchers to assist 18 crop-livestock farmers in Burkina Faso in planning the next agricultural season. Due to its simplicity, the static simulation tool of annual farm stocks and flows led to superior changes in the farmers' knowledge and practices. The rule-based dynamic simulation tool helped the researchers grasp farmers' decision-making processes but was difficult for farmers to understand due to the discrepancy between its multi-annual time step and the farmers' short-term planning horizon. The optimisation tool stimulated more strategic discussions regarding paths to improve farm income despite a design that was distant from the farmers' reality. © 2015 Elsevier Ltd. All rights reserved.

Keywords: Simulation tool Optimization Decision-making rules Farm Planning Burkina Faso

1. Introduction Faced with an increasingly complex and uncertain environment, fluctuating input and agricultural product prices, climate change impacts, and societal concern for the environmental impacts of agriculture (Thompson and Scoones, 2009), farmers worldwide are being forced to innovate. They do so by introducing technical changes or reorganizing how activities are combined on their farms. This innovation process can involve three decision levels: daily farm operations of field/animal components, seasonal planning, and long-term strategic choices (Cros et al., 2004; Martin et al., 2013). As proceeding by trial and error is a timeconsuming and risky process, modelling can be useful in assisting farmers to design, assess and implement innovative and sustainable production systems (Attonaty et al., 1999; Le Gal et al., 2011). Numerous modelling methods such as crop models


veloppement sur * Corresponding author. Centre International de Recherche-de l'Elevage en zone Subhumide (CIRDES), 01 BP 454 Bobo-Dioulasso, Burkina Faso. Tel.: þ226 76406786. E-mail address: [email protected] (A.W. Sempore). http://dx.doi.org/10.1016/j.envsoft.2015.02.015 1364-8152/© 2015 Elsevier Ltd. All rights reserved.

(Chatelin et al., 2005; Benjamin et al., 2010; Thorp et al., 2008), expert systems (Vandendriessche and van Ittersum, 1995; Snow and Lovatt, 2008) and information management tools (Jensen et al., 2000; Cornou and Kristensen, 2013) have been developed for daily farm operations. Seasonal planning and long term strategic choices require the support of holistic tools such as wholefarm models. Such tools render it possible to evaluate resource allocation decisions that farmers must make when designing and implementing change on their farms. While some of these models are mainly used by researchers to assess the merits of technical options (van Wijk et al., 2009; Whitbread et al., 2010), others are used with farmers to help them make their decisions or consider res et al., 2009b; potential changes on their farms (Vayssie Dogliotti et al., 2014). With the tremendous progress made in hardware and software development over the past 20 years, the use of modelling in research studies is now widespread. However, the use of modelling to assist farmers in their decision-making has been problematic (McCown, 2002; Jakku and Thorburn, 2010; Matthews et al., 2008). Rather than providing ready-made solutions, farm management models, used in interaction with a researcher or advisor, aim to help a farmer consider his or her

148

A.W. Sempore et al. / Environmental Modelling & Software 68 (2015) 147e155

options by comparing and discussing alternative production strategies (Le Gal et al., 2011; Rodriguez et al., 2014). To render the use of these models more effective, improved understanding of the interaction between model design, farmer needs, and the role of the model user (usually a researcher or an advisor) in management change is required. Several issues need to be addressed. The way a farmer understands and perceives a model helps shape the interaction between the farmer and the model user. The question which must be addressed is how does the design of a model influence this perception? To avoid a “black box” effect, the model should be transparent and produce outputs that make sense to the farmer (Rivington et al., 2007; Barnaud et al., 2008). An understanding of the kind of learning derived through the use of the model, both for farmers and the researcher or advisor working with them, also is critical (Matthews et al., 2011). The learning process that results from a model's iterative simulations includes (i) a better understanding of the farmer's current management practices, especially regarding interactions between the various components of the farming system, (ii) reflections regarding the potential alternatives to be simulated and (iii) evaluation of the consequences of each alternative on farm performance (McCown, 2012). A wide range of whole-farm models currently are available, which render difficult their selection in a decision support perspective. Three main types emerge from a review of the literature (Le Gal et al., 2011): (i) static simulation models, describing farm operations on the basis of stocks and flows over a single year (Martin et al., 2011; Andrieu et al., 2012; Rodriguez et al., 2014); (ii) rule-based dynamic simulation models with decision rules representing farmers' management modes in the form of “IF Conditions THEN Action” rules, simulating changes in the farm state over one or several years (Andrieu and Nogueira, 2010; Bergez et al., 2012; Moreau et al., 2013); and (iii) static linear programming models maximizing a utility function (income, for example) under constraints, representing the farm as a combination of linear activities, either over a single year (Groot
nchez et al., 2012) or over several et al., 2012; Rodríguez-Sa years (Naudin et al., in press). The specific objectives of a decision support process should determine the type of whole-farm model used. To date, however, scant attention has been paid to whether the different features of the various model types impact the interaction between the farmer and the model user. Ideally, model developers should consider whether their tool actually helps farmers in managing their production systems (Keating and McCown, 2001), yet few do so (Le Gal et al., 2011; Dogliotti et al., 2014). Moreover, in the decision support case studies reported, multiple decision support tools have rarely been used since researchers usually focus their studies on one model type, for instance optimization (Dogliotti et al., 2014) or a stock and flow simulation tool (Le Gal et al., 2013). Conducted in the frame of a two-year interaction between researchers and eighteen farmers in Burkina Faso, this study aims to investigate how farmers perceive models and what both farmers and researchers learn from the use of three modelling methods implemented in a decision support process at the farm level. The support process focussed on seasonal planning issues, which are critical for farmers due to a context characterized by deep uncertainty regarding the climate and economic environment. First, we present the study area, the three different tools used, the farm sample, and the approach followed. We then compare the use of the three tools using three criteria: assessment by farmers, facilitation of farmers' learning, and facilitation of researchers' learning. Lastly, we discuss the relevance of these types of tools in helping to design innovative production systems.

2. Materials and methods 2.1. The study area The study was conducted in the village of Koumbia in western Burkina Faso (latitude 12 420 20700 ; longitude 4 240 01000 ). Increasing demographic growth, with a current population density of 66 inhabitants/km2, and rising demand for plant and animal products is putting strong pressure on agro-pastoral resources (Vall and Diallo, 2009). The area also is characterized by a spatial-temporal rainfall variability, with an average of 900 mm/year, and three main cropping seasons: the rainy season, when biomass is produced (May to October); the cold dry season, (October to February), when crops are harvested, fodder stocks are replenished and animals are allowed to graze on fields after the harvest; and the hot dry season (MarcheApril), when herds consume the fodder stocks. During that period, herds can leave for transhumance, i.e. go to other village areas to access water and pasture. Farmers are operating in an economic environment marked by the rising cost of agricultural inputs and fluctuating global cotton fibre and local livestock prices. There are three main types of farmers in Koumbia: crop farmers (CF) cultivating cotton and cereals using animal traction; crop-livestock farmers (CLF) cultivating large areas and owning large herds; and semi-settled Peulh livestock farmers (LF) practicing cattle breeding and subsistence farming (Vall et al., 2006).

2.2. The three models The three tools used, Cik3 da, Simflex and Optimcik3 da, each belong to one of the three types of models noted in the introduction. All three represent a mixed croplivestock farm and are ad hoc tools (Affholder et al., 2012), meaning tools developed specifically for the study area. Each tool, along with the main characteristics of its model type, is briefly described below (see Table 1 for the main inputs and outputs of each model). Cik3 da is an static simulation model. This kind of tool represents decision outputs, such as cropping plans, crop management or herd diets, rather than decisions rules. This simple modelling structure is meant to be easier to understand, but it requires the use of some approximations (Le Gal et al., 2013). For instance, crop yields are not calculated based on mechanistic biophysical equations as in a crop model, but are directly entered by the model user. Cik3 da aims to support a farmer's assessment of the consequences of strategic (type and size of agricultural activities) and tactical (management of plant and animal production) choices on a farm's technical and economic performance (see Andrieu et al., 2012 for a detailed description). The balance between supply and demand for nutrients (nitrogen N, phosphorus P, potassium K), fodder, and cereals, as well as economic results, are calculated at the farm level for each configuration of the production system defined by the model user. Deficits in fodder result in the purchase of cotton meal which impact economic results. The biophysical processes considered are represented by static mean data obtained through surveys (average crop yield according to type of climate year), review of the literature (biomass mineral element content), and simplified calculations (exports of mineral elements). The simulation takes place over the course of one year. Simflex is a rule-based dynamic simulation model which explicitly represents farmers' decision rules. This kind of model is assumed to be a powerful tool for evaluating the consequences of management decision rules on farm performance since it mimics farmer behaviour (Cros et al., 2004; Chatelin et al., 2005 Andrieu et al., 2007). Decision rules can be either pre-established in the model based on res et al., 2009a) or entered by the model user based on a on-farm surveys (Vayssie meta-language (Romera et al., 2004). This modelling structure is supposed to be easily understood by farmers since their own rules are represented, but some simplifications are made to reduce the actual complexity of farmer decision rules (Romera et al., 2004). Simflex originally was developed for use in research exploring the impact of farmers' strategies to adapt to multiannual variability in economic and climatic conditions on their farms' technical and economic performance (Andrieu and Chia, 2012). Its direct use with farmers was tested in this study. Farmers' decision rules in response to changes in price and rainfall were represented using Python programming language. They were pre-established based on surveys conducted with another sample of farmers representing the three main farm types in the study area. These rules include (i) cropping plan choices based on the gross margin per hectare of cotton, (ii) mineral fertilisation of maize based on the purchase price of mineral fertilisers, (iii) the purchase of cotton meal or the start of transhumance based on the fodder stocks available and the purchase price of cotton meal, and (iv) the sale of animals when there is a negative economic balance. Simflex performs multi-annual simulations involving different climate and economic variables. Each year is independent of the year preceding it with the exception of the evolution of cattle herds and fodder stocks. Optimcik3 da is a linear programming model developed using GAMS software applications (Barbier, 1998; Dabire et al., 2011). This kind of optimization tool is based on a vision of a decision-maker being able to choose the solution maximizing his utility function from a large range of possibilities thanks to the completeness of the information available to him and his capacity to compare all of the solutions possible (Duke et al., 2012; Salassi et al., 2013). The tool is appreciated for its capacity to integrate biotechnical and economic variables, for instance to evaluate effects of policy decisions on farm incomes (Veysset et al., 2005) or potential value of

A.W. Sempore et al. / Environmental Modelling & Software 68 (2015) 147e155

149

Table 1 Main inputs and outputs of the three tools used in the course of the study. Modelling tool

Type

Input variable

Output variable

Cik3 da

Annual static simulation

 Farm structural characteristics: household size, family labour force, level of equipment, total farm surface area, maximum storage capacity for organic fertiliser and crop residues  Choice of cropping plan  Number of animals purchased and sold  Fraction of crop residues harvested  Number of animals from different animal mobs supplemented during the hot dry season  Amount of inputs purchased (fertiliser, animal feed supplements)

 Supply-demand balances for nutrient, fodder and cereals  Economic results (expenses, products, profits)

Simflex

Multi-annual rule-based dynamic

 Farm structural characteristics (see above)  Climate conditions: seasonal rainfall  Economic conditions: purchase price of fertiliser, purchase and sale price of cattle, purchase price of cotton meals

       

OptimCik3 da

Linear programming

 Farm resources: surface areas, inputs, labour force and animal mobs  Possible activities  Input consumption per activity unit

 Optimal cropping pattern and fattening unit  Maximized income

innovative cropping systems (Affholder et al., 2010). Since farmers' behaviour often is sub-optimal due to the incompleteness of the information available to them, their use with farmers has been increasingly oriented towards analysing the gap between the optimal solutions provided by the model and farmers' current practices (Attonaty et al., 1999). Optimcik3 da was designed at the request of extension services and policy-makers interested in addressing this gap. It calculates the optimal cropping plan and size of a cattle fattening unit maximizing farm income under resource constraints (total surface area, inputs, availability of manpower, availability of fodder, availability of feed supplements, number of cattle). The input data of this tool comes from the actual farms studied (farm resources) and data drawn from regional level surveys (yield of various crops, sale price of various productions) and from the literature (nutritive value of crop residues). 2.3. Choosing farmers Eighteen farms were selected based on the farm type (CF, CLF, LF) and the willingness of the farm head to participate in the study. Farmers were presented with the study's objectives and methodology before being asked to signal their interest in participating. Six farms were chosen for each of the three farm types (Table 2). CFs cultivated between 4 and 10 ha, were equipped with one or two pairs of draught oxen and had practically no cattle livestock. CLFs possessed large areas (35 ha on average), were well-equipped with animal traction (7 draught oxen on average) and had a cattle livestock unit of about 38 heads on average. LFs had large herds (50 heads on average) and cultivated 1e8 ha for home-consumption. Three sub-samples, each composed of two farms randomly selected per farm type, were then randomly attributed a tool. This protocol enabled us to distinguish the specific impact of each tool on the farmers' strategies and practices according to the farm type. It also helped each farmer better understand the tool used with him or her. 2.4. The approach A doctoral research candidate worked with each farmer to implement the research protocol. This researcher ensured the correct use and calibration of each tool that he himself helped to develop and adapt. He also cultivated the farmers' active participation throughout the support process through regular meetings. The involvement of the farmers was important for building credibility of both the researcher and the tools (Matthews et al., 2008). An initial diagnosis of each farm was followed up by alternating stages of planning and monitoring the crop year over the course of two years (2011 and 2012), and a final assessment of the whole process (Fig. 1). The initial diagnosis stage made it possible to identify the various problems farmers faced in managing agro-pastoral production and to assess farmers' initial knowledge of crop and livestock systems and their ability to estimate economic results. During this stage, the structure of the tool used and its modelling principles were presented to each farmer. The next stage was dedicated to the joint design of

Supply-demand balances for nutrient, fodder and cereals Economic results (expenses, products, profits) Cropping plan Amount of fertilisers purchased Amount of residues harvested Amount of cotton meals purchased Transhumance duration Number of animals sold

scenarios S0 and S1. S0 corresponded to the farmer's preceding crop year and S1 to the farmer's planning of the activities of the crop year to come. These two scenarios required the collection of data specific for the use of each tool as well as structural data about each farm (Table 3). The results of the simulations were presented to the farmers in the form of graphs to render them easier to understand. After discussions with the farmers, new planning scenarios were built through iteration (S2 to Sn) on the basis of the results of the preceding scenario and the farmer's production objectives. The agricultural activities were then monitored during the following crop year in order to measure the gaps between the scenarios discussed and the decisions actually taken. The last stage was dedicated to evaluating the approach and the tools at the end of the two years of work with the farmers. The assessment of both the tools and the support process followed a qualitative rather than quantitative methodology. Indeed, the intent of the support process was to enhance both farmers' and researchers' learning by providing quantitative trends related to each simulated scenario to foster participatory discussions. The tools consequently were assessed in

Table 2 Structural characteristics of the 18 farm partners in the approach. Farm

Number of Number of Number Tools Main Number Total breeding of breeding useda activityb of family area (ha) traction sheep cattle cattle workers

CF1Ci CF2Ci CF3Si CF4Si CF5Op CF6Op CLF1Ci CLF2Ci CLF3Si CLF4Si CLF5Op CLF6Op LF1Ci LF2Ci LF3Si LF4Si LF5Op LF6Op

Ci Ci Si Si Op Op Ci Ci Si Si Op Op Ci Ci Si Si Op Op

a b

C C C C C C CL CL CL CL CL CL L L L L L L

6 8 3 4 4 4 45 50 14 9 17 13 3 2 5 4 5 3

10 7.75 4.75 6.25 5.5 5 43 83.5 13 17 42 15.5 5.5 3.25 3 5.5 8 1.25

Ci: Cik3 da; Si: Simflex; Op: Optimcik3 da. C: Crop; CL: Crop-livestock; L: Livestock.

6 2 2 1 2 4 7 9 3 6 16 4 4 2 4 2 4 2

0 0 4 0 0 1 15 34 25 2 156 4 70 100 25 25 66 15

8 10 4 0 6 3 8 0 18 2 20 15 17 10 5 6 15 20

150

A.W. Sempore et al. / Environmental Modelling & Software 68 (2015) 147e155

Tool structure explained to the farmer

Diagnosis of mixed croplivestock farms

Simulations with tools and shaping of results

S0 and S1 co-design with farmers

----

Presentation of results to farmer and discussion

Evolution of S1 to Sn on the farmer request

April

Discussion of farmer’s reactions

January

February

2011

2012

June

---November

Monitoring agropastoral farm activities of farms

December January

Final evaluation of each tool with farmers

Fig. 1. Scheduling of the farmers' support process over two years (2011e2012). S0: preceding crop year scenario (initial situation); S1: planned activities for the coming crop year scenario; Sn: new planning scenarios emerging from the results of preceding scenarios.

terms of their capacity to facilitate farmers' learning processes rather than the accuracy of their predictive power. An evaluation form was designed to assess and compare the three tools based on farmers' opinions. The design drew from theories of qualitative model evaluation (Bennett et al., 2013) and learning, considering both change in representation € n, 1978) and learning from action (Kolb and Kolb, 2005). The form (Argyris and Scho included (i) the assessment of the tool by the farmers with a focus on its capacity to handle a theme and provide realistic results, (ii) the facilitation of farmers' learning in terms of improved knowledge and practices, with an evaluation based on the practices observed following the researcher's intervention, and (iii) the facilitation of the researcher's learning by better understanding farmers' strategies and decision-making processes. Points (i) and (ii) were based on the perceptions collected and questions asked during the preceding stage (interview guide). Five themes linked to farmers' concerns were selected: cropping plan for the coming crop year; influence of the climate/economic environment on agricultural

Table 3 Nature of the complementary data collected for each simulation tool. Tool

Specific inputs

Cik3 da

-

Simflex

Optimcik3 da

Doses/ha of herbicides, insecticides, NPK, Urea Purchase of different crop seeds Purchase of agricultural inputs Purchase and sale by animal mob/season Number of animals per mob given supplements during the hot dry season

- Critical rules: cotton/cereals cropping plan; purchases of mineral fertiliser; purchase of cotton meals; start of transhumance - Virtual series over 10 years: rainfall (favourable/ unfavourable); sale price per kg of cotton and maize; purchase price of beef and fattening cattle; sale price of beef and fattening cattle, litre of milk; purchase price of a bag of mineral fertiliser and of a kg of cotton meal - Costs related to purchase of seeds, mineral fertiliser, herbicides and insecticides - Costs related to veterinary services, purchase of salt, cotton meals - Costs related to salaried or temporary labour

Virtual series: Imaginary data built as a function of the agro-pastoral context.

production; fertiliser production and fertilisation; feeding animal mobs in hot dry season; cattle fattening.

3. Results 3.1. Scenarios according to farm type The scenarios simulated varied according to the farmers' concerns, which were in turn a function of their farm type (Table 4). All of the farmers were concerned with adapting cropping plans to fluctuating rainfall. However, the adjustment of the area under cotton to the economic context and the management of soil fertility only concerned CFs and CLFs, while animal feeding only concerned

Table 4 Topics of interest and themes simulated with the farmers in the support process according to their type. Type of farmer

Crop-livestock Livestock Crop farmers farmers (CLF) farmers (LF) (CF)

Concerns of farmers Adaptation of cropping plan to variable x rainfall Adaptation of the area under cotton to x the prices of cotton and inputs Management of soil fertility x Feeding herd

x

x

x x x

Themes taken into account in the simulated scenarios x Planning the cropping plan according to x the economic and climate context (tradeoffs between cotton and cereals) Modification of the area under cereals x Modification of the area under fodder x x Increase of organic fertiliser x applications Increase of mineral fertiliser applications

x

x x

x

A.W. Sempore et al. / Environmental Modelling & Software 68 (2015) 147e155

151

CLFs and LFs. Nevertheless, certain points of convergence appeared during the support process, notably around the introduction of fodder crops and the use of crop residues linked with the development of sheep and bovine fattening units.

also is a function of fodder resources and the water available in the area during the hot dry season.

3.2. Criterion 1: evaluation by the farmers

The majority of the farmers, regardless of the tool used, declared that their knowledge and practices had improved through the support process. The proportion of farmers impacted was nonetheless higher among those who used Cik3 da. However, there appears to be differences in the type of knowledge and practices generated by each tool.

All of the farmers deemed that the three tools correctly simulated cropping plans and the influence of climate and economic conditions. However, only Cik3 da seemed to provide relevant results for these two topics, since these results resembled the farmers' actual practices. Farmers did not understand the underlying logic of Simflex, which consists of analysing risks taken over the long term, because they had trouble projecting themselves beyond the next crop year. As one farmer remarks, “I cannot plan my crops at mid and long term. But the cereal productions calculated by the model correspond to my own productions for the former years”. For two of the six farmers who used Optimcik3 da, the gap between the cropping pattern that they planned and the solution provided by the tool left them doubtful about the realism of the model's outputs. They either had different objectives (“I do not aim to maximize my income but rather to be able to feed my family”) or the difference with their current farm results was too great (“The income calculated by the model is far from what I usually get”). Optimcik3 da follows a purely economic reasoning while the farmers' real planning processes integrate other elements such as the possibility of acquiring agricultural inputs on loan from the cotton company to spread on maize. All of the farmers who used Cik3 da found that the production of organic fertiliser and the application of fertilisers were represented in a relevant manner by the model because it allowed their current mineral fertiliser application and collection of animal waste and crop residue practices to be directly entered. None of the farmers who used Optimcik3 da found that this theme was adequately represented because Optimcik3 da optimizes the area planted with maize, over which organic fertiliser primarily is spread, according to the number of animals and the organic fertiliser stocks entered. The farmers who used Simflex had mixed opinions because the model only considers the potential production of organic fertiliser based on an estimation of the quantity of waste that various confined animal mobs could produce during the cold and hot dry seasons. Farmers considered that the three tools were capable of representing a cattle fattening unit, but that their outputs held little to no relevance. CLFs considered that this activity was simulated well by Cik3 da because it allowed them to assess the needs of the unit in terms of residues and supplements. In contrast, LFs considered that there was a gap between their current practices (selecting animals from their herd to be sold) and the way Cik3 da proceeds (purchase of lean animals to be fattened, which is the dominant practice in the area). No farmer found the results produced by Simflex and Optimcik3 da on this topic relevant. Simflex used the same inputs as Cik3 da but the decision rule used to simulate the purchase of supplements for fattened cattle was distant from their current decision rules. In Optimcik3 da, the number of animals for fattening was an output dependent on the crop residues available to the farmer. Lastly, only CLFs and LFS considered that the foddering of cattle livestock was correctly taken into account in Simflex. Indeed, Simflex simulates the number of days of transhumance for animal mobs when the fodder balance, calculated as a function of crop residues produced/stored on the farm and herd fodder needs, is negative and the price of supplements surpasses the farmers' purchase threshold. This output, which does not exist in Cik3 da and Optimcik3 da, corresponds to a very common practice among LFs, but its realism was considered weak because departure on transhumance

3.3. Criterion 2: facilitation of farmer learning

3.3.1. Planning of agro-pastoral activities The planning process, as an embodiment of a farmer's thinking about the organization of the coming crop year, lay at the core of the support process tested with them. Two thirds of the farmers thought that they had improved their planning process regardless of the tool used or the type of farm they operated. Of the four farmers impacted by the use of Cik3 da, there were two CFs, one CLF, and one LF. For the first three, using the tool raised their awareness of the need to anticipate the organic fertiliser applications of the next crop year by harvesting the crop residues from the preceding crop year. The CLF also recognized the need to prepare for the fattening year not, as he used to do, once the crop season had ended, but rather before it started by deciding on the size of the land under fodder and residue stocks. The LF also recognized that the ration of the animals in the hot dry season based on fodder crops should be planned when deciding the cropping plan at the beginning of the rainy season: As stressed by one farmer: “Cik3 da is a good planning tool since it allows to analyzing each farm activity in accordance with the other ones in order to improve it in the future”. CFs and CLFs benefited from using Simflex by better clarifying the mechanisms which they use to plan their areas under cotton in relation to the profit margin expected from this crop: “The tool has allowed me to compare different possibilities of crop plan with my current one according to the sale prices of cotton”. The planning exercises conducted with Optimcik3 da above all allowed two CLFs, one CF, and one LF to improve their capacity to assess the balance between their resources and the activities possible on their farms. It allowed them to identify, for example, the number of workers needed to correctly sow all of the surface areas allocated to crops. The other farmers were little affected due to the large gap between their practices and the optimal practices furnished by the tool. 3.3.2. Management of organic manure and soil fertility Regardless of the tool used, none of LFs acquired new knowledge about management of organic manure and soil fertility. Indeed, the production of fertiliser is not a constraint for them due to their high level of animal manure production resulting from their high stocking rate per hectare under cultivation. CFs and CLFs who used Cik3 da considered that their knowledge of fertiliser production and fertilizing crops had improved since they were able to quantify the size of the imbalances in terms of organic fertiliser applications and production: “Thanks to Cik3 da, I have been able to compare my production of manure in my cattle yard to what could be potentially produced and applied. This comparison encourages me to increase the manure production of my farm”. This knowledge acquisition was expressed by a 71%e300% increase of organic fertiliser stocks on 4 farms during the final stage of the process (Fig. 2a), and an increase of mineral fertiliser doses on two farms compared with initial practices (Table 5). A similar improvement of knowledge on imbalances between organic fertiliser supply and demand took place for CFs and CLFs who had used the other two tools. That said, an increase of organic fertiliser stocks actually only took place on two farms (Fig. 2b and c). The use of the tools had no

152

A.W. Sempore et al. / Environmental Modelling & Software 68 (2015) 147e155

a. Cikεda

b. Simflex LF4Si

LF2Ci LF1Ci

LF3Si

CLF2Ci

CLF4Si

CLF1Ci

CLF3Si

CF2Ci

CF4Si

CF1Ci

CF3Si 0

200

400 %

Fodder

600

800

0

50

Fodder

Organic manure

100 %

150

200

Organic manure

c. Optimcikεda LF6Op LF5Op

CLF6Op CLF5Op CF6Op CF5Op 0

50

100

Fodder

150 %

200

250

300

Organic manure

Fig. 2. Increase in fodder and organic manure production following the use of each tool in the support process with each farmer (% of the initial value).

impact on the farmers who already had some knowledge regarding the management of organic fertiliser and soil fertility. 3.3.3. Feeding animals on farm With the exception of one LF who was already aware of the issue, Cik3 da made it possible to improve farmers' knowledge about managing the feeding of different animal mobs. This particularly concerned the assessment of the quantities of residues to harvest to meet the animals' needs and limit their dependence on cotton meal. This better understanding of the balances between animals' needs and feed supply was concretely translated among CLFs and LFs into an increase of 28% to over 700% of crop residue stocks on their farms (Fig. 2a). This development was accompanied on two farms by the introduction of a fodder crop into the farmers' cropping plans and on one farm by the establishment of a cattle fattening unit (Table 5). Other farmers did not change their livestock activities because of the costs of purchasing cattle and cotton meal and labour constraints regarding the storage of crop residues and the monitoring of animals to be fattened. The use of Simflex advanced the knowledge of only two CLFs and one LF who were interested in ways to calculate fodder needs and the consequences of an imbalance at the farm level over the duration of transhumance. The increase of harvested crop residue stocks among five of the six farmers is less striking than in the preceding case, from 34% to 150%. No farmer started a fodder crop production or a fattening unit. This result, however, could be due to a sampling bias because three of them had already implemented

Table 5 Number of farmers who changed their practices following the use of each tool.

Introduction of fodder crop Increasing mineral fertiliser dose on maize Introduction of cattle fattening activity

Cik3 da

Simflex

Optimcik3 da

2 2 1

0 2 0

0 3 0

such changes. For those three, the use of the tool contributed nothing new (Table 5). The use of Optimcik3 da did not generate any additional knowledge on feeding animals on farm as it is limited to the calculation of the optimal number of fattening cattle as a function of available crop residues. Practically speaking, no farmer introduced a fodder crop or a cattle fattening unit, but an improvement in fodder stocks was observed on four farms. As in the preceding case, these results are strongly linked to the specific context of each farm. 3.3.4. Economic evaluation related to agro-pastoral activities CFs and CLFs growing cotton were familiar with the calculation of economic margins because they belonged to producer groups that provided accounting support services. LFs, on the other hand, were unaccustomed to calculating their margins in relation to cereal crops because their crops were home-consumed. Cik3 da and Simflex enabled them to familiarize themselves with this calculation, as stated by a farmer: “Cik3 da has enabled me to better assess my expenses and products for a complete agricultural year”. By providing the maximum income that theoretically could be attained at the farm level, Optimcik3 da had an impact on the knowledge of two CFs and two LFs who were surprised by the amount calculated. The gap between maximum and actual income was often related to the farmer's incorrect estimation of outside labour requirements for a given crop plan. The maintenance of certain crops may be neglected due to a lack of manpower, which results in lower yields and, consequently, lower real income compared with the income generated by Optimcik3da. 3.4. Criterion 3: facilitation of user learning The simple act of using and filling in a formal framework, that of the model, to describe a farm in a support process generates an ensemble of knowledge about the case studied. The construction, simulation, and discussion of prospective scenarios with the farmer

A.W. Sempore et al. / Environmental Modelling & Software 68 (2015) 147e155

is an additional way to throw light on, and better understand, the farmer's reasoning processes. While these observations were true for all three tools used, the tools differed in terms of the type of technical, strategic, and decision-related knowledge produced. Indeed, due to the process modelled, each tool oriented the researcher towards specific types of interactions with the farmer, such as calibrating the model or continuing discussions of simulated results. Knowledge about technical practices and the strategic directions of farms thus were acquired with Cik3 da, while decisionmaking processes were better understood with Simflex and slightly understood with Optimcik3 da. Due to its focus on the farmer's practices and the variety of scenarios possible, Cik3 da enabled the researcher to better understand the responses that farmers wished to give to problems encountered. For example, to better manage a reduction in soil fertility, the farmers tended to move the draught oxen onto the fields during the hot dry season so that they could consume the standing biomass stocks there and fertilize the soil with their waste. The calibration of Simflex required specific questions about the indicators that trigger decision rules such as cropping plan choices and the mobility of different herds of the farm, all sources of specific lessons. Thus, when the price of mineral fertiliser was very high, farmers preferred to reduce the area under maize rather than reduce the quantities of fertiliser applied per hectare in order to maintain the yield of this crop. The outputs also stimulated discussions on the evolution of transhumance practices among LFs and CLFs. Optimcik3 da enabled discussions to be held regarding the household's objectives at the end of the simulations. Indeed, the hypothesis that the farmer seeks to maximize his income led to discussions about the way households articulate different objectives such as CLFs increasing areas under cereals to generate income and the storage of harvests to limit climate risk. The role of cotton for farmers also was clarified through its triple objectives: (i) acquire mineral fertiliser on credit, (ii) maintain soil fertility by rotating with maize, (iii) provide cash flow on the farm at the beginning of the dry season. 4. Discussion 4.1. What is the purpose of whole-farm models? This study confirms the overall observation that whole-farm models facilitate discussion between researchers and farmers around the conception of innovative production systems, regardless of their type. Indeed, all of the tools used in this study provided a framework to guide the initial farm diagnosis at the beginning of the support process. However, they also helped ex-ante assessments of a range of alternatives and facilitated the acquisition of new knowledge by those participating in the approach (Matthews et al., 2008; Martin et al., 2013). Each type of tool has specific values and limits in a decision support process which are linked to its conceptual principles. Optimization tools feature relatively strict construction principles which are too distant from the reality of farmers's current decisionmaking processes. However, they enable potential solutions to be provided to farmers seeking to maximize their income according to their available resources (Bernet et al., 2001; Cabrera et al., 2008). The optimal solution tends to become a benchmark that farmers can compare with their current strategy (Attonaty et al., 1999), as mentioned by some of them: “The income calculated by Optimcik3 da has allowed me to get an idea about my farm potential of production”; “Thanks to Optimcik3 da I was able to know the crop plan adapted to my resources”. However, since optimisation models usually take

153

into account only one objective, this type of tool holds little interest for farmers who have multiple agricultural production objectives. Static stock and flow models appear to be more flexible and to better fit the reality of farmers. Moreover, outputs in terms of resource supply-demand balances over an agricultural year were well understood by farmers. They prove even more useful when the farmers are considering plans to develop their production systems. This tool type fully responds to the exploratory objective on which the prospective thinking of the farmer, in interaction with the researcher, is based (Kerr et al., 1999; Le Gal et al., 2013; Martin et al., 2013). It also allows certain technical knowledge of farmers to be improved and to situate innovations in the specific context of a farm. It thus can prove to be useful when a technological innovation is disseminated by allowing the consequences of farm results to be measured ex-ante. Their main limit lies in the kind of practices taken into account in the model. Multi-annual rule-based dynamic simulation tools, which examine the reliability of a farmer's decision making processes in response to climatic and economic risks, allow the researcher to better identify these processes. For farmers, their relevancy depends on their consistency with the farmers' planning horizon. Farmers who do not look beyond the next season as in this study will benefit less from these tools than ones who are more used to long-term planning. The perception of this type of model depends also on how a farmer understands the set of rules and interactions between rules supposed to mimic his behaviour. To enhance this comprehension process, role playing games have been used in combination with multi-stakeholder rule-based dynamic models (Bousquet et al., 2002), but rarely for farm support (Daatselaar and Tomson, 2011). Moreover our study shows that even when limited to a small agricultural region, the diversity of production systems encountered cannot be fully addressed by a single tool, and this diversity complicates the fit between the tools designed and the questions posed by farmers. This observation argues for the use of a range of tools which each have a structure and domain of validity that closely reflect a specific context. Using such a combination of tools is possible in a research setting. A researcher can then match the various farmers' contexts and concerns with the appropriate tool, and improve his or her own knowledge about farmers' behaviour. The value of certain tools for farmers is more limited when the tool has not been specifically designed to support their decisions, as was the case with Simflex and Optimcik3 da. These results confirm the need for designers to clarify the domain of utilisation of their models based on an evaluation from model users and the targeted audience, something rarely encountered in the literature on wholefarm models (McCown et al., 2009; Le Gal et al., 2011). 4.2. A qualitative evaluation methodology Assessing the impact of the use of a support process, with or without a formal tool, is a particularly delicate operation (Birner et al., 2009; Faure et al., 2011). In a context where the researcher res et al., 2009b) and of evaluator plays the role of advisor (Vayssie (Matthews et al., 2011), this interaction can only be correctly managed and analysed with small samples of farms (Siggelkow, 2007). As mentioned above, the assessment of the tools was essentially qualitative since the objective was to study their capacity to enhance both farmers' and researchers' co-learning processes. This assessment mainly was based on farmers' opinions, a process which is particularly adapted to a context with limited available quantitative data and co-learning objectives (Bennett et al., 2013; Krueger et al., 2012) based on learners' endogenous knowledge and reflections on their own experience (Kolb and Kolb, 2005).

154

A.W. Sempore et al. / Environmental Modelling & Software 68 (2015) 147e155

However, the methodology used in this study has some limitations which already have been mentioned by Sterk et al. (2011). As the researcher is no longer simply an observer, he must make a selfassessment of what he has drawn from the process, all the while knowing that his learning is a function of the manner in which he has steered the interaction with the various farmers encountered. The assessment of the impact of the tool then becomes complex because it simultaneously mixes the individual perceptions of the researcher and farmer regarding the process and its results, the existence of a question that is stated more or less clearly by the farmer, and the capacity of the process to respond to this question. In this respect, each tool might have been selected according to each farmer's specific problem situation to match the features of the tool to the needs of the farmer. Moreover, it was difficult to differentiate between farmers' increased confidence in their own knowledge resulting from their participation in the support process and new knowledge (learning per se) acquired through the process (Matthews et al., 2011). To overcome these limits, the introduction of a third person intervention, observing the implementation of the process in real time, or assessing it ex-post, would be useful but this is rarely done due to the skills and costs involved (Moumouni et al., 2009). 5. Conclusions This study allowed the application of three different wholefarm models to be tested and compared in the frame of a farmereresearcher interaction. The tools were used to support farmers in western Burkina Faso in planning their production activities. The static stock-flow simulation tool, which relied on a representation of farmers' practices, on-farm physical flows of material, and simple calculations based on resource balances, emerged as the tool closest to the reality and concerns of farmers. This type of tool facilitates farmers' learning, but requires researchers to go beyond the simple manipulation of quantitative variables to deepen his or her knowledge of a farmer's decision-making rationale. As an intermediary object, the simulation tool can stimulate discussions as long as the researcher knows how to go beyond simple data collection. The multi-annual rule-based dynamic simulation tool had the opposite effect in a context of short-term planning by farmers. Its multi-annual structure focussing on certain farm components renders it more difficult for farmers to understand. However, its use forces researchers to better grasp farmer decision-making processes, at least for the rules concerned. The optimisation tool at first appeared to be the least adapted to a decision support use due to its design principles, which distance it from the reality of farmers. However, a comparison between its internal logic and that of farmers can stimulate a more strategic discussion regarding paths leading to improved farm income. Taken together, these three tools thus constitute complementary discussion support aids that can help farmers and researchers improve their combined knowledge of agro-pastoral practices and of possible alternatives to achieve farm objectives. Acknowledgements This study was supported by a PhD grant from CIRAD and by funding from the Corus program of the French Ministry of Foreign Affairs (Grant agreement number 6057-2). The authors thank the farmers from Koumbia who allowed us to work with them throughout the study, the anonymous reviewers who greatly helped to improve a previous version of the paper, and Grace Delobel for translating the text into English.

References Affholder, F., Tittonell, P., Corbeels, M., Roux, S., Motisi, N., Tixier, P., Wery, J., 2012. Ad Hoc modeling in agronomy: what have we learned in the last 15 years? Agron. J. 104, 735e748. Affholder, F., Jourdain, D., Quang, D.D., Tuong, T.P., Morize, M., Ricome, A., 2010. Constraints to farmers' adoption of direct-seeding mulch-based cropping systems: a farm scale modeling approach applied to the mountainous slopes of Vietnam. Agric. Syst. 103, 51e62. Andrieu, N., Nogueira, D.M., 2010. Modeling biomass flows at the farm level: a discussion support tool for farmers. Agron. Sustain. Dev. 30, 505e513. Andrieu, N., Poix, C., Josien, E., Duru, M., 2007. Simulation of forage management strategies considering farm-level land diversity: Example of dairy farms in the Auvergne. Comput. Electron. Agric. 55, 36e48. le de simulation pluriannuelle des syste mes de Andrieu, N., Chia, E., 2012. Un mode
lisation, production d'Afrique subsaharienne: Simflex. In: Partenariat, mode
rimentations: quelles leçons pour la conception de l'innovation et expe
cologique ? Actes du se
minaire ASAP, Bobo-Dioulasso, Burkina l'intensification e Faso, 15-17 novembre 2011, 12 p.
, P., Le Gal, P.Y., Rueff, M., Schaller, N., Sempore, A., 2012. ValiAndrieu, N., Dugue dating a whole-farm modelling with stakeholders: evidence from a West African case. J. Agric. Sci. 4, 159e173. € n, D., 1978. Organizational Learning: a Theory of Action Perspective. Argyris, C., Scho Addison-Wesley, Reading, MA. Attonaty, J.-M., Chatelin, M.-H., Garcia, F., 1999. Interactive simulation modelling in farm decision making. Comput. Electron. Agric. 22, 157e170. Barbier, B., 1998. Induced innovation and land degradation: results from a bioeconomic model of a village in West Africa. Agric. Econ. 19, 15e25. Barnaud, C., Bousquet, F., Trebuil, G., 2008. Multi-agent simulations to explore rules for rural credit in a highland farming community of Northern Thailand. Ecol. Econ. 66, 615e627. Benjamin, L.R., Milne, A.E., Parsons, D.J., Lutman, P.J.W., 2010. A model to simulate yield losses in winter wheat caused by weeds, for use in a weed management decision support system. Crop Prot. 29, 1264e1273. Bennett, N.D., Croke, B.F.W., Guariso, G., Guillaume, J.H.A., Hamilton, S.H., Jakeman, A.J., Marsili-Libelli, S., Newham, L.T.H., Norton, J.P., Perrin, C., Pierce, S.A., Robson, B., Seppelt, R., Voinov, A.A., Fath, B.D., Andreassian, V., 2013. Characterising performance of environmental models. Environ. Modell. Softw. 40, 1e20. Bergez, J.-E., Charron, M.-H., Leenhardt, D., Poupa, J.-C., 2012. MOUSTICS: a generic dynamic plot-based biodecisional model. Comput. Electron. Agric. 82, 8e14. Bernet, T., Ortiz, O., Estrada, R.D., Quiroz, R., Swinton, S.M., 2001. Tailoring agricultural extension to different production contexts: a user-friendly farmhousehold model to improve decision-making for participatory research. Agric. Syst. 69, 183e198. Birner, R., Davis, K., Pender, J., Nkonya, E., Anandajayasekeram, P., Ekboir, J., Mbabu, A., Spielman, D.J., Horna, D., Benin, S., Cohen, M., 2009. From best practice to best fit: a framework for designing and analyzing pluralistic agricultural advisory services worldwide. J. Agric. Educ. Exten. 15, 341e355. Bousquet, F., Barreteau, O., d'Aquino, P., Etienne, M., Boissau, S., Aubert, S., Le Page, C., Babin, D., Castella, J.-C., 2002. Multi-agent systems and role games: collective learning processes for ecosystem management. In: Janssen, M.A. (Ed.), Complexity and Ecosystem Management. The Theory and Practice of Multi-agent Systems. Edward Elgar Publishing, Cheltenham, Great Britain, pp. 248e285. Cabrera, V., Breuer, N., Hildebrand, P., 2008. Participatory modeling in dairy farm systems: a method for building consensual environmental sustainability using seasonal climate forecasts. Clim. Change 89, 395e409. Chatelin, M.H., Aubry, C., Poussin, J.C., Meynard, J.M., Masse, J., Verjux, N., Gate, P., Le Bris, X., 2005. DeciBle, a software package for wheat crop management simulation. Agric. Syst. 83, 77e99. Cornou, C., Kristensen, A.R., 2013. Use of information from monitoring and decision support systems in pig production: collection, applications and expected benefits. Livest. Sci. 157, 552e567. Cros, M.J., Duru, M., Garcia, F., Martin-Clouaire, R., 2004. Simulating management strategies: the rotational grazing example. Agric. Syst. 80, 23e42. ^le avec les agriculteurs pour l'aide a  la Daatselaar, C., Tomson, N., 2011. Un jeu de ro
cision: l'expe
rience du LEI/Wageningen Co Daatselaar. In: Penot, E. (Ed.), de
Exploitations agricoles, strategies paysannes et politiques publiques. Les le Olympe, Paris, France, pp. 315e321. e
ditions Quae. apports du mode
vision Dabire, W.P.I., Barbier, B., Andrieu, N., 2011. Evaluation ex ante de la pre re climatique en petit paysannat burkinabe
. Rev. Elev. Med. Vet. Pay saisonnie 64, 43e50. Dogliotti, S., García, M.C., Peluffo, S., Dieste, J.P., Pedemonte, A.J., Bacigalupe, G.F., Scarlato, M., Alliaume, F., Alvarez, J., Chiappe, M., Rossing, W.A.H., 2014. Coinnovation of family farm systems: a systems approach to sustainable agriculture. Agric. Syst. 126, 76e86. Duke, J.M., Borchers, A.M., Johnston, R.J., Absetz, S., 2012. Sustainable agricultural management contracts: using choice experiments to estimate the benefits of land preservation and conservation practices. Ecol. Econ. 74, 95e103.
mique des dispositifs de Faure, G., Rebuffel, P., Violas, D., 2011. Une analyse syste  l'exploitation familiale en Afrique de l'Ouest. Cah. Agric. 20, 394e396. conseil a Groot, J.C.J., Oomen, G.J.M., Rossing, W.A.H., 2012. Multi-objective optimization and design of farming systems. Agric. Syst. 110, 63e77.

A.W. Sempore et al. / Environmental Modelling & Software 68 (2015) 147e155 Jakku, E., Thorburn, P.J., 2010. A conceptual framework for guiding the participatory development of agricultural decision support systems. Agric. Syst. 103, 675e682. Jensen, A.L., Boll, P.S., Thysen, I., Pathak, B.K., 2000. Pl@nteInfo®; d a web-based system for personalised decision support in crop management. Comput. Electron. Agric. 25, 271e293. Keating, B.A., McCown, R.L., 2001. Advances in farming systems analysis and intervention. Agric. Syst. 70, 555e579. Kerr, D.V., Chaseling, J., Chopping, G.D., Cowan, R.T., 1999. DAIRYPROda knowledgebased decision support system for strategic planning on sub-tropical dairy farms. II. Validation. Agric. Syst. 59, 257e266. Kolb, A.Y., Kolb, A.D., 2005. Learning styles and learning spaces: enhancing experiential learning in higher education. Acad. Manag. Learn. Educ. 4, 193e212. Krueger, T., Page, T., Hubacek, K., Smith, L., Hiscock, K., 2012. The role of expert opinion in environmental modelling. Environ. Modell. Softw. 36, 4e18.
, P., Faure, G., Novak, S., 2011. How does research address the Le Gal, P.Y., Dugue design of innovative agricultural production systems at the farm level? A review. Agr. Syst. 104, 714e728. Le Gal, P.-Y., Bernard, J., Moulin, C.-H., 2013. Supporting strategic thinking of smallholder dairy farmers using a whole-farm simulation tool. Trop. Anim. Health. Prod. 45, 1119e1129. Martin, G., Felten, B., Duru, M., 2011. Forage rummy: a game to support the participatory design of adapted livestock systems. Environ. Modell. Softw. 26, 1442e1453. Martin, G., Martin-Clouaire, R., Duru, M., 2013. Farming system design to feed the changing world. A review. Agron. Sustain. Dev. 33, 131e149. Matthews, K.B., Schwarz, G., Buchan, K., Rivington, M., Miller, D., 2008. Wither agricultural DSS? Comput. Electron. Agric. 61, 149e159. Matthews, K.B., Rivington, M., Blackstock, K., Mccrum, G., Buchan, K., Miller, D.G., 2011. Raising the bar? the challenges of evaluating the outcomes of environmental modelling and software. Environ. Modell. Softw. 26, 247e257. McCown, R.L., 2002. Changing systems for supporting farmers' decisions: problems, paradigms, and prospects. Agric. Syst. 74, 179e220. McCown, R.L., 2012. A cognitive systems framework to inform delivery of analytic support for farmers' intuitive management under seasonal climatic variability. Agric. Syst. 105, 7e20. McCown, R.L., Carberry, P.S., Hochman, Z., Dalgliesh, N.P., Foale, M.A., 2009. Reinventing model-based decision support with Australian dryland farmers: changing intervention concepts during 17 years of action research. Crop. Pasture. Sci. 60, 1017e1030. s, F., Baratte, C., Delaby, L., Faverdin, P., Gascuel-Odoux, C., Moreau, P., Ruiz, L., Verte Piquemal, B., Ramat, E., Salmon-Monviola, J., Durand, P., 2013. CASIMOD’N: an agro-hydrological distributed model of catchment-scale nitrogen dynamics integrating farming system decisions. Agric. Syst. 118, 41e51. Moumouni, I.M., Vodouhe, S.D., Streiffeler, F., 2009. What makes small-scale farmers participate in financing agricultural research and extension? Analysis of three case studies from Benin. J. Agric. Educ. Ext. 15, 301e316. Naudin, K., Bruelle, G., Salgado, P., Penot, E., Scopel, E., Lubbers, M., de Ridder, N., Giller, K.E., 2015. Trade-offs around the use of biomass for livestock feed and soil cover in dairy farms in the Alaotra lake region of Madagascar. Agr. Syst. (in press).

155

€ckle, C.O., Donatelli, M., Rivington, M., Matthews, K.B., Bellocchi, G., Buchan, K., Sto 2007. An integrated assessment approach to conduct analyses of climate change impacts on whole-farm systems. Environ. Modell. Softw. 22, 202e210. Rodriguez, D., Cox, H., deVoil, P., Power, B., 2014. A participatory whole-farm modelling approach to understand impacts and increase preparedness to climate change in Australia. Agric. Syst. 126, 50e61.
nchez, S.V., Pla -Aragone
s, L.M., Albornoz, V.M., 2012. Modeling tactical Rodríguez-Sa planning decisions through a linear optimization model in sow farms. Livest. Sci. 143, 162e171. Romera, A.J., Morris, S.T., Hodgson, J., Stirling, W.D., Woodward, S.J.R., 2004. A model for simulating rule-based management of cow-calf systems. Comput. Electron. Agric. 42, 67e86. Salassi, M.E., Deliberto, M.A., Guidry, K.M., 2013. Economically optimal crop sequences using risk-adjusted network flows: modeling cotton crop rotations in the southeastern United States. Agric. Syst. 118, 33e40. Siggelkow, N., 2007. Persuasion with case studies. Acad. Manag. J. 50, 20e24. Snow, V.O., Lovatt, S.J., 2008. A general planner for agro-ecosystem models. Comput. Electron. Agric. 60, 201e211. Sterk, B., van Ittersum, M.K., Leeuwis, C., 2011. How, when, and for what reasons does land use modelling contribute to societal problem solving? Environ. Modell. Softw. 26, 310e316. Thompson, J., Scoones, I., 2009. Addressing the dynamics of Agri-food systems: an emerging agenda for social science research. Environ. Sci. Policy 12, 386e397. Thorp, K.R., DeJonge, K.C., Kaleita, A.L., Batchelor, W.D., Paz, J.O., 2008. Methodology for the use of DSSAT models for precision agriculture decision support. Comput. Electron. Agric. 6 (4), 276e285.
, P., Blanchard, M., 2006. Le tissage des relations agriculture-e
levage Vall, E., Dugue au fil du coton. Cah. Agric. 15, 72e79. Vall, E., Diallo, M.A., 2009. Savoirs techniques locaux et pratiques : la conduite des ^turages (Ouest du Burkina Faso). Nat. Sci. Soc. 17, 122e135. troupeaux aux pa Vandendriessche, H.J., van Ittersum, M.K., 1995. Crop models and decision support systems for yield forecasting and management of the sugar beet crop. Eur. J. Agron. 4, 269e279. van Wijk, M.T., Tittonell, P., Rufino, M.C., Herrero, M., Pacini, C., Ridder, N.D., Giller, K.E., 2009. Identifying key entry-points for strategic management of smallholder farming systems in sub-Saharan Africa using the dynamic farmscale simulation model NUANCES-FARMSIM. Agric. Syst. 102, 89e101. res, J., Guerrin, F., Paillat, J.-M., Lecomte, P., 2009a. A global activity model for Vayssie evaluating the sustainability of dairy enterprises Part I e whole-farm dynamic model. Agric. Syst. 101, 128e138. res, J., Bocquier, F., Lecomte, P., 2009b. GAMEDE: a global activity model for Vayssie evaluating the sustainability of dairy enterprises. Part II e Interactive simulation of various management strategies with diverse stakeholders. Agric. Syst. 101, 139e151. Veysset, P., Bebin, D., Lherm, M., 2005. Adaptation to Agenda 2000 (CAP reform) and optimisation of the farming system of French suckler cattle farms in the Charolais area: a model-based study. Agric. Syst. 83, 179e202. Whitbread, A.M., Robertson, M.J., Carberry, P.S., Dimes, J.P., 2010. How farming systems simulation can aid the development of more sustainable smallholder farming systems in Southern Africa. Eur. J. Agron. 32, 51e58.