Bayesian networks for decision analyses — an application to irrigation system selection

Abstract

Farmers are under continual pressure from Government and industry to change farm practices to meet productivity and environmental targets. In response to these pressures, farmers will make decisions to adopt practices that reflect their motivations and priorities. However, where the changes of practice are major, there may be considerable uncertainty associated with the decision-making process. Decision support tools are one method that may assist in reducing the uncertainty associated with decisions about changes in farm practices.

Bayesian networks provide a useful tool to assist in the structuring and analysis of decision problems. A Bayesian network is a decision analysis framework, based on Bayesian probability theory, which allows the integration of scientific and experiential knowledge, and the uncertainty associated with this knowledge. The approach involves describing a system in terms of variables and linkages, or relationships between variables, at a level appropriate to the decision making. This is achieved through representing linkages as conditional probability tables and propagating probabilities through the network to give the likelihood of variable outcomes. Therefore, the approach ensures that treatment of risks and uncertainties is an intrinsic part of the decision-making processes. The Bayesian network is dynamic and interactive, and hence if a network previously developed does not fit a user's conceptual understanding of the system, it can be adapted quickly and simply to the cognitive understanding of the user.

A case study Bayesian network has been developed for decisions associated with the selection of irrigation systems for irrigated dairy farms in Northern Victoria. This case study demonstrates that the most appropriate irrigation system for a dairy farm is dependent on factors including the amount of irrigation water available and soil types. Analysis of the Bayesian network indicates that the appropriate irrigation system is more sensitive to the income generated from pasture than to the price of water. The Bayesian network can demonstrate the impacts of decisions on the farmer's system and can allow the farmer to evaluate these impacts according to their own priorities and criteria. This information can then be used by the natural resource manager to assess the appropriate level of incentive or penalty required if the farmer is to adopt the preferred option that will also achieve preferable outcomes from a natural resource management perspective.