Previous PageTable Of ContentsNext Page

Decision support for dynamic management on extensive beef properties

Richard Monypenny

Department of Economics, James Cook University, Townsville QLD 4811

Summary. Dynamic management strategies that respond to the current and expected conditions of a farm system are conceptually and intuitively attractive. One way to evaluate such strategies is to represent the strategies as rules and then use a model to evaluate the rules. Rules need to answer the following questions: (i) To what must management respond? (ii) Which are acceptable responses, for the individual decision maker? (iii) When to respond? Other problems are: how to best represent the farm system in a model and how to choose between the rules being evaluated?

Introduction

The search for appropriate solutions, to problems caused by unintended changes to atmosphere, soil, water, plants and animals, is of current interest to many people. One aspect of this search is the study of real world bio-economic farm systems in order to develop dynamic management strategies; that is, strategies that are usually responsive to the current and expected conditions of the farm system. One way to study dynamic management strategies is to use a model to evaluate the medium term outcome of various 'what if' decision alternatives.

Decision makers in the real world evaluate decision alternatives as if they were giving a single, unambiguous set of weights to the variables in their objective function, this being the desired outcome of managing the farm. In 'what if' studies using a model there are many possible sets of weights. Furthermore, these may change in each time period over the decision maker's time horizon. One way that the model can accommodate this complexity is to use a rule to represent the strategies.

The study of rules is easier when it is embedded in the provision of decision support. The two main reasons for this ease are that the strategies are more likely to be implemented; and that the decision maker takes responsibility for a large part of the burden of identifying valid decision alternatives. Useful rules have been developed for some specific strategies, as for example in drought management, but these are of little use in other management strategies. This presentation to the Agronomy Conference is seen as a way to stimulate the search for more appropriate rules.

Decision support

The idea of providing support to improve the quality of decisions is one that most people would think of as being 'good' but find difficult to define. Decision support has been practised under various names for many years, but to date there has been no generally accepted terminology or methodology. In this paper decision support is a process or an approach, rather than a specific model with which to formulate and evaluate alternatives and to arrive at decisions. Decision support usually involves skills of facilitation, problem formulation, analysis, synthesis, and of evaluation.

Complexity is an important aspect in providing decision support. Farm systems will always be complex by definition; decision support may be simple or complex. People choose which of these types of decision support they will use. We are well equipped for simple decision support, but as decision support becomes more complex we require more knowledge and more learning that is specific to the farm system in question.

A given decision situation can be divided basically into the following stages: problem formulation; evaluation of alternatives; problem owner choice; and follow up.

The actual implementation of decision support in a given decision support situation is as varied as the individuals and the decision situations involved. However, the one common feature of all decision support is that it's goal is the result of the decision. Thus decision support focuses on the constraints to decision making and their relationship with the constraints to the farm system about which decisions are to be made. These constraints need to be defined both in terms of one or more parameters or indicators (for a given time frame) and in terms of the available management responses or options for given changes in these parameters.

Verification and sensitivity analysis are important in providing a realistic expectation about the reliability of a decision support process. They can be planned in advance and carried out as the process develops from initial ideas through consolidation and into the closure of the whole process. They should only be problematic in situations where there is major conflict between stakeholders. On the other hand validation, that is the examination of the broader question of whether the intended decision support process is appropriate or should even be undertaken at all, is frequently and unavoidably very difficult.

Decision support for beef properties

In terms of decision support, an extensive beef cattle property in the semi-arid tropics on northeastern Queensland (2, 3, 4, 5) could be modelled as follows.

Topographically it consists of a large piece of land (say 30000 ha) which is partitioned into areas of (i) native pasture (in the hilly and/or poor country); (ii) cleared native pasture (in the more sustainable country); and (iii) over-sown legumes (in the better country). This partitioning represents the pasture type. For more details see (1).

As a commercial enterprise, the property grazes different cattle types on the various pasture types, usually in a fairly well-defined and stratified manner. In this case, the related interactions can be ignored. For management purposes the cattle types are partitioned into breeders (say 1500) and non-breeders and these are then partitioned by age. The role of the breeders is to maintain cattle numbers such that the need for restocking (purchase of breeders or of non-breeders) is minimized. During the year, older/fatter non-breeders are sold to generate revenue for the property, in the form of cash flow. The breeders, therefore, play a decisive role in the decision making but only a passive role in the generation of revenue.

In this context the aim of decision support is to evaluate alternative stocking (management) strategies, specifically to evaluate the number of breeders to retain each year in terms of net cash flow and within the constraints of expected pasture growth.

Due to the nature of the system (the property) there are insufficient data for either a purely inductive model or a fully deductive model. Thus a model was built that allows a decoupling of the inputs from the outputs: a deductive model related weight gain with revenue, and a part inductive and part deductive model related the biology (pasture types, cattle types and expected pasture growth) to weight gain (for details see (1)). Weight gain was defined as 'the annual weight gain of the cattle as a function of expected pasture growth'. Expected pasture growth was represented by a proxy called green weeks. Green weeks are an assessment of the growth potential of the pasture in the coming months. The number of green weeks was determined at the end of summer on the basis of the extent of that summer's rainfall.

This representation of the property is realistic since the property's activities are concerned with the weight gain of cattle. It therefore gives a natural, intuitive appeal to the model. It also provides a bridge between detailed agro-biological knowledge from research and the management alternatives of property decision making (for details see (5)).

This representation resulted from the frustration of not being able to formulate a model of a cattle property that was appropriate for decision support. The actual breakthrough followed the discovery of the inductive model relating pasture type, cattle type and expected pasture growth and its empirical validation.

The model, and the decision support system that has been built around this model, have been used in the evaluation of tree clearing (4) and of management alternatives during drought (3).

The work on tree clearing provided a description of the soil and salt relationships implied in the clearing of trees from the landscape on properties near Charters Towers. These were then used to put some figures on the whole property economic implications of tree clearing compared with various other development alternatives. From the point of view of the decision maker, the results of this work were to actually specify realistic alternatives for the property and to determine likely whole property implications or consequences of each of these alternatives. On a wider scale the results indicated that because of variability, often not apparent at first glance, each decision should be evaluated separately.

The work on management alternatives during drought evaluated the alternatives of feeding, agistment, and sale as related to properties near Charters Towers. The conclusion was that sale or agistment were the best alternatives. However the main contribution of this work was not the conclusion but rather the framework that it provided for the analysis of whole property implications of management alternatives.

Decision support for dynamic management

The development of dynamic management strategies is different in each of the following situations:

  • The real world of on-going property decision making: that is, the decision maker on an extensive beef property makes decisions. We and he/she may not know how, but decisions are made.
  • Property development: that is, the evaluation of a, usually discreet, set of medium term alternatives for property development. These alternatives will have varying degree of uncertainty associated with their description and their likely outcome.
  • Policy evaluation: that is, the evaluation of likely response to proposed policy measures. Response is usually at some aggregate level of beef properties, for example shire or State, and policies are usually government policies.

From the point of view of modelling dynamic management strategies, in the first situation (above), the problem is solved by the experience and intuition (however defined) of the property decision maker. In the third situation, it can be solved by using some relatively simple assumptions and then relying on the effect of aggregation to balance out errors. The second case is the most difficult, because the modeling needs to incorporate the detail of the specific situation, but without the experience and intuition of the on-going property decision making, or the approximation of aggregation. One of the reasons for the difficulty of this situation is that the system (the property) is not in a stable state, that is the results from one year can not be used as a proxy for the likely results for the next year. Furthermore, it is precisely the changes that occur during the transition from one unstable state to another that are of interest to the decision maker in the evaluation of development alternatives.

In this situation, rules for the number of breeders to retain on the property that have been successful to date were evolved through a long process of trial and error. They have been of little use when taken outside the specific question for which they are useful. For example in the case of drought management, three rules were found to be useful (only one rule was used at a time):

  • Set the number of breeders to be retained at a constant value, say 1200. This can also be expressed as a stocking rate of breeders of say 0.11 breeders/ha.
  • Set the number of breeders at different levels, say 1200 for the first three years followed by an increase to 1600.
  • Set the number of breeders so that the amount of feed per head is of a given value, say 900 green weeks*ha/breeder. Total feed is expressed as expected pasture growth, that is in turn expressed by the expected pasture growth proxy, green weeks; multiplied by the area available for grazing, measured in hectares.

The basic problem or difficulty of finding appropriate rules is that the system and the modelling of the system are not independent. Rather there are three sets of interdependences:

  • Interdependences between variables in the bio-ecological aspects of the farm system, for example between climate (rainfall), available pasture, number of cattle, weight gain of steers and branding rate.
  • Interdependences between management response to changes in the bio-ecological aspects of the farm system and the changes that actually occur because of management response, for example between feeding selected cohorts of animals or feeding all animals during drought.
  • Interdependences between the structure of the model and the implied assumptions about the farm system, for example between the discrete time periods used to represent a continuous system, and using linear programming or a spreadsheet.

The search for more appropriate rules needs to answer the following questions:

  • To what must management respond? The variable to which management responds needs to be available and to be a reliable predictor, for example, the rainfall over the last eight months. From the decision support point of view, we would expect the identity of this variable to vary between farm systems, if for no other reason than the fact that the decision makers are different.
  • How to respond or which are acceptable responses for the individual decision maker? For example in a drought situation, do you sell or do you feed?
  • When to respond? What is the latest date by which a decision must be implemented?
  • How to best represent the system in a model?
  • How to choose between the alternatives being evaluated? ('What if' outcomes or maximizing net farm income?)

For an individual case, the decision maker will be able to considerably reduce the number of feasible or acceptable answers to all these questions, especially to questions (a), (b) and (c). However, in the context of the evaluation of development alternatives, to a large degree the answers to questions (d) and (f) will be relevant.

References

1. Anderssen, R.S. and Monypenny, R. 1990. Proc. 1990 Mathematics in Industry Study Grou. (CSIRO Division of Maths and Stats). pp. 58-70.

2. Gillard, P. and Monypenny, R. 1988. Agric Syst. 26, 179-190.

3. Gillard, P. and Monypenny, R. 1990. Agric Syst. 34, 37-52.

4. Gillard, P. Williams, J. and Monypenny, R. 1989. J. Aust. Inst. Agric. Sci. 2, 34-39.

5. Monypenny, R. and McIvor, J. 1989. Proc. Simulation Soc. Aust. pp. 28-32.

Previous PageTop Of PageNext Page