efita.org

Abstract: This paper argues in favor of a decision support approach relying on an explicit and rigorous modeling of the management strategy that underlies any farmer’s decision making behavior. A strategy is a roadmap of intended technical tasks over a management period. It is tied to an overall objective and specifies what to do depending of the encountered situations. An essential feature is its flexibility enabling to cope with stochastic fluctuations of the environment. In order to evaluate the worth of a strategy, the advocated approach relies on a simulation tool with which the effects of applying it are evaluated under different hypothetical weather conditions. The example of a rotational grazing dairy production system

Keywords: decision support system, management strategy, simulation, rotational grazing

Contents

Internationalization of markets, shifts in consumers’ demand and requirements, rapid evolution in technologies, greater concern of environmentally friendly production are among the recently appeared factors that make competitiveness much harder to achieve and maintain in the agricultural production industry. Unlike the rather stable context of the past decades, farmers must now strive for a dynamic competitive advantage that requires a well mastered understanding of their production processes so as to control them under various constraints and toward specific objectives that both may change from one year to the other. Consequently higher importance has been given to the ability of making as wise as possible decisions concerning configuration choices and day-to-day technical management. It is striking to see how profitability varies from one farm to the other just because of the differences in management skills of the farmers. The most successful ones usually operates on the basis of an anticipation of what situations could occur and what the appropriate reactions could be in order to ensure that their production system stays on the right track. In other words, driving profitably an agricultural production system requires a management strategy, that is, a conditional plan of actions specifying the intended courses of operations attached to the possible futures.

As part of a decision support project, the present paper argues in favor of a modeling and simulation approach of the management strategies elaborated to drive agricultural production systems. The systems addressed can be either a single farm enterprise (e.g. a crop on a set of fields managed in the same way), or a combination of highly interdependent or interlocking farm enterprises (e.g. a livestock enterprise that produce milk from different feeds and a forage enterprise that supplies the grazing grass to the dairy cow herd). The managerial task considered in this paper deals with the making and execution of decisions concerning timing, amount and mode of use of various resources (land, labor, machinery, inputs) in the production of a commodity such as milk, cereals, fruits, etc.. Hence we only deal with technical management aspects rather than marketing and financial aspects (organization and control of capital : when to invest, where to find capital, when to replace machinery).

The following section describes the management problem and management strategies in agricultural production systems. Section 3 emphasizes the need of decision support systems based on strategy simulation, and the reminder of the paper illustrates these concepts on a specific agricultural management problem concerning rotational grazing in dairy production.

2.1 Management problems in agriculture

The management problem has a dynamic nature due to changes from one year to the other (in available resources, in economic context and legislation) and due to unpredictable fluctuations within a production cycle (climate and sometimes prices). Thus management cannot be reduced to day-to-day running of a pre-established rigid set of actions, and must be seen as a temporally structured cognitive process. To some extent, farm enterprise management is similar to production management in the manufacturing industry. Essentially the complexity of the problem stems from the large number of uncertain data to deal with and the numerous decision steps and alternatives to consider. A classical approach to cope with such a complexity in industrial production management is to decompose the problem into different elementary functions like planning, scheduling and control of the production process [18]. In farm management, analogous decompositions of the decisional and technical activities have long been ignored, mainly because of the predominance of the concept of the farmer as the unique decision-maker and actor in his/her farm. Due to the nature of the processes to control and due to the level of uncertainty about the future, there are however important differences between farm management and industrial production management. More than in industrial processes, the response of crop yield and livestock output to inputs is subject to uncontrollable variations due to weather and disease. In agriculture the counter part of machines are not really optimized in regard to the production objectives, since they often are biophysical systems. Moreover, the socioeconomic environment is generally more multiform and less controllable by farmers. To sum up, farming systems seem to be more hazardous, more complex, and less standardized than industrial production systems.

Despite these inherent difficulties, farming system researchers [19] developed in the eighties a conceptual model of the management decision process for agricultural production systems. The framework has been studied in the setting of different production systems (e.g. sugar beet, wheat) and some implementations have been realized (see for instance [14], [16] , [10] and [3]), giving the concept of management strategy a more concrete content [4]. Current efforts in this line aim at further developing and formalizing the notion of strategy for more difficult management problems such as handling the various operations in a greenhouse production [17], [11]. Another example is rotational grazing that is briefly presented in the next subsection to show the kind of decision involved. This example is further developed in Section 4 to illustrate, in particular, what constitutes a management strategy in this problem.

2.2 An example of management problem: rotational grazing

Many dairy cow production systems (see Figure 1) rely strongly on a grassland feeding resource that is exploited through rotational grazing and completed by conserved feed (maize silage, concentrate and hay) in winter times when the herbage mass is insufficient [7]. The late winter to early summer period is a particularly crucial phase in which the diet must switch progressively from a fully maize-concentrate feeding to a predominantly or fully herbage-based feeding. The general objective of the farmer is to keep the milk production at its optimal level despite the uncontrollable fluctuations of some important factors such as weather. The main technical decisions that the manager (farmer) has to make concern: the set of fields definitely allocated to grazing, the set of fields set aside to cope with weather deviation and grazed only if necessary, the profiles of conserved feed distribution over the whole period, the fertilization policy, the cutting policy and the field rotation policy. She/he must find a coherent combination of choices such that an almost optimal production of milk is ensured over the whole production period for a sufficiently representative range of climatic conditions. The period considered in the management task covers nine months from the beginning of February until the end of October. The starting date corresponds to the change of the sward from the vegetative to reproductive stage, time at which the first fertilization operation may have to be performed. The ending date corresponds to the calving period and the strong decrease of the herbage production, time at which it becomes necessary to turn to a conserved feed diet that is not problematic from a decision point of view (some DDS tools exist for such a feed composition task).

HTML clipboard

The main difficulty in this management problem stems from the fact that the forage production process interacts strongly with its concomitant use through grazing. For rotational grazing to be successful, the forage supply must constantly match the demand as closely as possible. This requires anticipation (i.e. planning) as well as permanent adjustment to the growth state of forage. The underlying control problem is a complex one because it involves a multivariable optimization. An appropriate quantity/quality tradeoff of the available herbage should be maintained along the considered period given that the maize distribution profile can only be non-increasing and the grass growth rate is partially controllable by the fertilization but also partially uncontrollable due to the climatic influence. Some agronomists' results, mainly coming from studies on continuous grazing, have shown that in order to have herbage of good quality it is necessary that the grazing intensity be high and regular on rotational periodicity. The problem of strategic management of a herd fed predominantly by rotational grazing (henceforth we shall talk simply of rotational grazing management) has to be solved once every year because the stock of maize at the start of the period varies from one year to the other and the size and characteristics of the herd may change too.

From an economical point of view, the problem is crucial because the herbage resource is much cheaper than the maize-concentrate one and it gives a better public image to the produced milk. The concern to minimize risks has led to usage of conserved forage (maize silage) and concentrate being higher and grazing lower than what can be done from a profit maximization point of view. The conservative attitude is partly explainable by the imperfect knowledge of the farmers with respect to the mastering of the complex interaction between grass growth processes and the grazing and milk production of the livestock. Another reason relates to the European Union policy that has encouraged maize production by subsidizing it heavily. In France the rotational grazing management problem concerns almost 90% of the dairy producers.

2.3 Management strategy: a conceptual decision model

A key assumption of this study is that farmers, consciously or not and effectively or not, decide and act on the basis of management strategies which are decision and action trajectories defined conditionally to the important situations that might occur during the management process. A strategy can be seen as a set of planned tasks that incorporates capabilities to adapt to perturbations with respect to the average course of future events. The context-dependent adaptations are means to cope with the stochastic nature of the environment. Another way to characterize this conceptual decision model is to consider the temporal and hierarchical organizations of the management task. Even if the different cognitive and physical processes of an agricultural production system are managed by a single person (the farmer), it is worth considering a hierarchical decomposition of the management problem in order to better understand it and provide support for more effective and robust decisions.

The temporal structure of the management process along the decision cycle has the following features:

• one or few overall production objectives (e.g. maximizing milk production, minimizing herbage waste and consumed maize);
• a set of intermediate goal states (e.g. maintaining an appropriate quantity/quality tradeoff of the available herbage) which the farmer tries to achieve or trajectories which she/he tries to stick to by the day-to-day technical choices she/he makes;
• some planned rendezvous (e.g. the end of the first rotation cycle) between the farmer and the production system, where she/he makes some observations and diagnostics;
• a set of decision rules that allows the farmer to decides how to adapts its management trajectory to unexpected events or to significant deviations caused by the weather.

The management task can also be decomposed in different enterprises, each being itself decomposable into simpler subtasks that are relatively independent from the point of view of the resources they need. Of course, the different plans for these subtasks must be coordinated.

As far as the hierarchical representation of the decision processes is concerned three main functions can be isolated: planning, observing/monitoring and acting. The planning function determines a temporal organization of rather abstract subtasks on the basis of an anticipated course of future events (intermediate goals and instruction trajectories). It also constructs an initial plan and occasionally (at the rendezvous points) performs some adaptations when adverse trends (bad weather) or opportunities are noticed. The acting function expands the active subtasks specified in the plan by applying situation-dependent procedures (decision rules) that generate primitive actions to be performed on the controlled biophysical system. The acting function is invoked much more frequently than the planning function. Finally, the observing/monitoring function is responsible of getting relevant data about the biophysical system and external environment, in order to inform the planning and acting functions.

The level of detail in specifying a plan cannot be directly the level of primitive actions and physical variables because the plan would be too complicated to express and adapt due to the multitude of possible situations that might occur given the uncertainty about the future. Thus, the definition of a management procedure necessarily requires (essentially for planning and observing/monitoring functions) the use of abstract concepts that do not always correspond directly to tangible facts and decisions, and that are often the result of an important cognitive activity reflecting the empirical knowledge of the farmer or the community of farmers as accumulated through time.

Within this conceptual model of the management decision process, a strategy can be defined as a specification of this general model, which characterizes the way of deciding and acting of a particular farm manager. The definition of a strategy requires the specification of:

• how to plan, adapt and coordinate long term (with respect to the production horizon) management trajectories for the different tasks involved in the production process;
• how to generate every day the immediately executable actions expanding for each active task the planned trajectory in a way depending on the current situation;
• how should the important events that have to be monitored be defined;
• what interpretation should be made of the data in order to feed the decision process.

Since efficiency and risk control is directly dependent on the management strategy, it is necessary to study management strategies to improve the production results. A preliminary step is to express them formally. The benefits of explicit representation of strategies are undoubtedly important. First the concept of strategy is often seen with a very restricted view as a rigid sequence (or even a set) of decisions themselves reduced to the assignment of a value to a decision variable, whereas the truth is much more complex as different case studies have shown. Another reason to investigate strategies is to put down the complete picture of what constitutes the decision behavior of a farmer for the production system under consideration. It clarifies weak spots of the decision process and forces to face issues that are sometimes unconscious in the decision-maker's mind. Therefore it facilitates critics and thus improvement.

3.1 Simulation versus optimization

Two modeling options for the management strategies are possible depending on the kind of decision support one intends to provide. The first one, which is widespread today, amounts to consider a family of very simple strategies, classically some decision vectors, and develop an optimization approach to search for the "best" decisions according to a well-defined numerical criteria (see for instance [11], [5] and [15]). The second one, that is the object of this paper, consists in modeling as thoroughly as possible the strategies and biophysical processes, and to simulate their interactions on a computer. This approach can be used to support a trial-and-error learning process by exploring rapidly and at nearly no cost alternative management strategies. The simulation gives basic figures of the evolution of the biophysical system which enables to analyze the economic and technical efficiency of the strategy applied [1] (Figure 2).

3.2 Desirable features of a simulation tool

Despite its popularity in industrial contexts, simulation is still in its infancy in the agricultural management domain. In particular, the dynamic aspect of the decisional part has not been addressed in depth so far. Most simulators deal with crop response to uncontrolled inputs (e.g. solar radiation, temperature, rain) and controlled ones generated on the basis of rather static management rules. The later usually convey pre-established sequences of technical operations and sometimes support a reactive behavior when particular conditions occur. The farmers’ management abilities are crudely modeled in such systems since no provision is given for simulating the coherent anticipatory and adaptive decision trajectory that farmers should have in order to orient the enterprise production according to their objectives and to reduce as much as possible the impact of the fluctuations of the uncontrollable factors. Due to their oversimplified view of the management task and the strong hypotheses regarding the availability of information on the biophysical variables these simulation-based systems are too far from the real context of farmers' decision-making and their practical usefulness as decision support tools is questionable (arguably as limited as the current optimization tools).

The major aim of our research is consequently to promote and develop the use of structured languages for representing management strategies in interaction with biophysical systems (similar ideas about the interaction of decision plans, natural resources and weather condition have been investigated for management of winter wheat [13], conserved forage production [10] and greenhouse tomato production [17] ). In order to propose satisfying simulation-based DSS, it seems necessary to impose some design constraints on the modeling capabilities and computational framework of the simulators. These include:

• a proper level of detail and of precision of the biophysical model with respect to the intended use. The biophysical model must be able to respond dynamically to the actions determined by the decision system and it must be able to provide the kind of information used to make decisions. Although the predictive capabilities of the biophysical system need not be very high, it must exhibit a realistic behavior of the processes it encompasses;
• openness and flexibility of the formal language used to represent the production system and more especially the management strategies;
• usability of the simulation tool: ease of simulating the consequences of applying a strategy in different uncontrollable contexts;
• efficiency to cope with repeated simulations covering a range of hypothesis about the external (uncontrollable) environment.

The following section elaborates on the rotational grazing problem introduced in Subsection 2.2 and illustrates how strategies are represented in a specifically-developed formal language that is interpretable by the constructed simulator.

The Subsection 4.2 shows how a rotational grazing strategy can be structured and represented in an intelligible and rigorous language. Before turning to this level of detail, Subsection 4.1 presents the production system that is actually simulated so as to see where the strategy lies within this system.

The production system is constituted by three interacting subsystems represented in different colors in Figure 3: the decision system itself composed of the planning and acting systems, the information system composed of the monitoring and observing systems and, finally the biophysical system. The production system dynamics heavily depends on the external environment (weather) that is uncontrollable and only partly predictable. The objective of the production system is to optimize the milk production by a proper use of the resources given the material constraints on the biophysical system.

The biophysical system (see [8], for a complete description of its modeling) is the controlled system. It is modeled through a set of more or less empirical laws that express on a daily basis the dynamics of several interactive processes dealing with herbage production, cow intake and milk production. The pasture is divided into a certain number of fields having different sizes but producing the same kind of forage crop. The driving variables include, for climate, the average daily temperature, average incident solar radiation and daily rain and, for technical management interventions, the nitrogen level, grazing operation (moving the herd to a new grazed field), cutting operation and amount of conserved feeding in daily diet.

4.1.2 The decision system

The management actions are generated by the decision system that essentially performs the decision making task which the farmer is confronted to every day. As already mentioned, the complexity of the management task requires to decompose it into two simpler dependant modules: (i) the temporal planner of operations that produces a set of plans and operational constraints at an intermediate level (i.e. instructions that are not directly executable actions) so as to ensure a consistent temporal commitment over the production horizon and (ii) the generator of executable actions that tells what to do according to the current situation at execution time and given the general instructions of the planner. An example of instructions generated by the planning system is the specification of the set of fields to be used in the first cycle of rotational grazing, this set having to be elaborated consistently with the fertilization and feeding policy. On a given day within this cycle, the acting system is then responsible of the determination of the particular field to graze among the set fixed by the planner. The decision system must be responsive to the different situations that the production system is likely to encounter; from time to time the planning and acting systems must modify previously adopted commitments on the plans and action generator in order to adapt to weather fluctuations.

4.1.3 The information system

The role of the information system consists in providing access to the relevant data concerning the biophysical system and the external environment. What is relevant is highly subjective and is actually part of the decision making behavior adopted. Two functions must be performed by the information system: (i) monitor some expected events in the biophysical system or external environment and notify their occurrence to the decision system that uses them as decision making temporal landmarks, and (ii) interpret and store some data about the biophysical system and external environment and communicate the results to the decision system. This is respectively what the monitoring and observing systems are doing. An example of event that the decision system wants to be monitored is the earliest ending date of the first rotation which is defined as the first day after which the sum the average daily temperature since the beginning of February is greater than a given threshold over which the herbage intake decrease in relation to the quality degradation (for example, 600 degree-days for cocksfoot). The interpretation functions of the observation system are used to reproduce the real situation of a decision maker that, first, has only partial access to information (due to lack of time and sensing devices) and, second, relies on aggregated pieces of data for cognitive simplicity. For instance, the decision maker may plan, on the basis of a qualitative appraisal of the maize stock at the beginning of February; an interpretation function computes a qualitative value (above average, average or below average) by a simple translation of the number of days of feeding that can be covered with the available maize stock.

4.1.4 The components of a strategy

The management strategy fully specifies the decision making behavior of the farmer in charge of the control of the biophysical system. It tells in a structured way what to do conditionally to some states and events. Therefore, in order to define a strategy one has to state:

• planning rules that define trajectories for the different tasks involved in the production process;
• acting rule that expands, for each active task, the planned trajectory so as to generate situation-dependant actions;
• how the temporal landmarks involved in the planning and acting rules and associated to monitored events have to be defined;
• what interpretation or translation functions should be defined in order to inform the condition parts stated in the planning and acting rules.

The above items are the basic components used respectively by the planning, acting, monitoring and observing systems.

This subsection illustrates by few examples the main components of a strategy and how they are represented in the formal language created for this purpose. The reasons to develop a formal language for expressing management strategies are threefold:

• studying strategies requires a rigorous framework to support scientific experimentation and analysis;
• the writing of strategies by users of the simulator (research scientists and extension services agents) has to be facilitated by providing an easily learned and understandable environment incorporating the essential conceptual structures needed in formulating a strategy;
• the strategies have to be stated in a format lending itself to machine interpretation since they are fed into a simulator.

The basic conceptual structures that are used include the components mentioned in the previous subsection, in particular the planning rules and acting rules that use interpretation functions and the temporal landmarks. In addition to these, the rotational grazing strategies are decomposed into a set of management tasks that can be treated independently.

4.2.1 Tasks

So far five tasks have been identified as shown in Table 1. To each task is associated a set of plan variables and a set of action variables that are assigned values by the planning and acting rules respectively. For instance, the value of the ?FeedConcentrate plan variable of the ConservedFeed task may be the word no or a numerical value indicating either that no concentrate should be given or what percentage of the maximum potential amount should be provided daily (this value depending of the stage with respect to the last calving date). The value of the ?ConcentrateAmount action variable is the amount of concentrate given per cow. The plan variables are typically assigned for a period, whereas the temporal scope of an action variables assignment is only one day. Note that the last task in Table 1 is of different nature than the others since it serves to specify ordering on the tasks and actions.

Table 1. The tasks and the corresponding plan and action variables

4.2.2 Planning rules

For each task, the planning rules are responsible of assigning values to the corresponding plan variables for a given period over which these values remain constant, except if an adaptation is required (which would be realized by firing another planning rule). An example of planning rule defining the Grazing task from the beginning of February until the earliest date of turn-out to grass is shown in Figure 3. The decision part assigns a set of fields to the ?GrazingFields variable. This planning rule is fired at the beginning of the simulation and defines at this initial stage the set of fields that the farmer plans to use in the first grazing rotation (knowing this set of fields is useful for other tasks such as Fertilization and ConservedFeed and also to determine the effective turn-out date).

Figure 3. A planning rule specifying the set of fields allocated to grazing

In the above rule, the term !beginningD is a temporal landmark that is simply set to the initial date of the management period. The term !earliestTurn-outD is a temporal landmark specifying a day through the conditions that the production system should satisfy at that day. Consequently the !earliestTurn-outD gets a numerical value only when the conditions become satisfied, the value being the current date. Before that, the value is unknown. A planning rule is fired as soon as the landmarks in its triggering part are known. Some planning rules are declared to be usable several times; every use occurs when the landmarks in their triggering parts change from an unknown value to a known one. Figure 4 shows how the !earliestTurn-outD landmark is defined in the strategy language. Essentially the landmark is the date of the first day such that total herbage mass over the set of grazing fields is equivalent to more than 3 days of what the herd needs if fed only with grass.

Figure 4. The !earliestTurn-outD temporal landmark.

Besides the planning rules that set up nominal plans, a strategy normally contains planning rules that simply perform adaptation on these through modifications on plan variables or parameters involved in plans. An example of such a rule is given in Figure 5. It tells that at the end of the first grazing rotation, the set of grazing fields should be reduced (resp. enlarged) if the total herbage mass on the grazing fields initially planned is above what would correspond to 12 days of feed (resp. below 8 days). The reduction and enlargement consist in taking-out or adding as many fields (usually only one) to match as close as possible the required modification of grazing surface specified in the first parameter of the Reduce and Enlarge functions. The HerbageMassAvailability() function is a user-defined interpretation function.

Figure 5. A planning rule for adapting the set of grazing fields to specific situations

4.2.3 Acting rules

Another key component involved in the definition of a strategy is the set of acting rule that specifies completely for the current day what action to perform in the task under consideration. Figure 6 gives an example taken from the Cutting task. The rule specifies that if it has not rained during the past three days, then the fields to cut for silage are all those ensilable among the those planned for such a use.

Figure 6. An acting rule of the Cutting activity

The NoRain3D term in the above rule is defined by an interpretation function, as was the term HerbageMassAvailability mentioned in the figures 4 and 5. Its definition is given in Figure 7 where D stands for the date of the current day.

Figure 7. The code of the interpretation function of NoRain3D

The interpretation function are essentially used to provide past and present synthetic information about the external environment and the biophysical system. It can also be used as a predictor of future states.

The first step in using the simulator is to initialize the production system by describing, first, the various components of the biophysical system (the composition and initial state of the fields and the herd) and, second, the strategy as illustrated in the previous subsection. The user must also select an hypothetical climatic year in a database or construct one using a weather generator.

The simulation can then be run. For the first simulated day the planning system will try to fire all the planning rules in order to construct a nominal plan specifying the different tasks over the entire time horizon of the management problem. The planning rules also determine an ordering between the tasks. The ordered plans are then transmitted to the acting system that can start its job: consider each task in the specified order and determine the actions to execute this first day by using the acting rules. The actions are executed, causing together with the advance of time a change in the state variables of the biophysical system. The simulator considers then the next day. Similar treatments are made: detection of events corresponding to the landmarks used in the planning rules, application of planing rules to perform some plan adaptation if necessary, and generation of the actions to be executed this day according to the acting rules and the perception of the current situation (the interpretation functions), and finally updating of the current state of the biophysical system. The iterations are pursued until the end of the simulated period. The user can then perform other simulations by changing only the climate hypothesis. The results of the different simulation given as time series of the most significant variables selected by the user can then be analyzed and used to define new strategies which can be submitted to the same evaluation process.

The changes in the economic, technical and legal context require innovation or at least adaptation of the way to use the resources (how to produce) and consequently the way to manage the enterprise along the production period. In many cases, what was good is no longer sufficient or appropriate to new constraints and objectives. In order to support this necessary adaptation of the farm enterprise management and cope with the difficulty of finding new solutions to new problems we have advocated a modeling approach of the response of the biophysical processes to farmers’ technical operations and of the articulated logic underlying the choice of these operations. The dynamic simulation of this response can be used as a support tool for designing management strategies.

The main advantage of the simulation approach is that it enables to represent quite faithfully the types of strategies that might be encountered or conceived by farming system researchers. Nevertheless it is our belief that some optimization approaches are worth developing in combination with the type of simulation discussed in this paper. An example of such an approach producing robust decision rules by use of machine learning techniques is presented in [2]; it applies to a strategy structure consisting in a sequence of rules of the form If state then actions. The problem is more difficult with rotational grazing because the structure of the management strategies is more complex.

Presently the rotational grazing simulator discussed in Section 4 of this paper is still under development. The implementation of the first prototype is complete. The core of the program is written in C++ and the graphical interface is written in Tcl/Tk so that the application can run on various platforms (Linux, UNIX, Windows 95/NT, Macintosh). What remains to be done is to conduct a thorough evaluation process in collaboration with the intended users (farming system scientists and extension services). The underlying problem here is the one of validating a decision support system, an issue that is still in need of a methodology. Depending on the experimentation results obtained at the end of this validation work, we might consider the development of a general representation and simulation framework applicable to different agricultural production systems.

Although before having completed the validation stage we cannot report on the demonstrated practical usefulness of the simulation approach, it has shown to provide a profitable basis for future development of strategic decision support systems dealing with the management of technical aspects of agricultural production. The representation framework and simulation tools of the kind discussed in this paper can enhance creativity and intuition of those willing to explore new management strategies. It can also facilitate transportation of a solution to similar production configurations, as it has already been the case in a set of similar projects dealing with seasonal crop management (winter wheat, maize and rape seed).

In this paper the management of only a part of the farm (a single enterprise or a combination of few interdependent ones) has been considered. This excludes for instance the problem of finding the most profitable mix of crops and livestock products to produce from the available resources at the farm scale. The problem of managing a whole farm is more complex for several reasons: it requires to be addressed with a time horizon of several years, it involves many situations of concurrence on the use of resources (machinery and labor) and it is much harder to build a closed world in which to study biophysical, social, economical and managerial aspects. However, dealing successfully with the management of a combination of interdependent enterprises gives encouraging insight on the general principles to be taken into account in order to tackle the whole farm management problem.