An application of belief networks to future crop production

Bayesian belief networks are shown to be natural and efficient knowledge representation tools for modelling and manipulating uncertainties in developing expert systems. They provide a basis for probabilistic inference, to calculate the changes in probabilistic belief as new evidence is obtained. However, their use in real problem domains is hampered by the difficulties facing the construction of such belief networks, particularly in domains where neither sufficient data nor human expertise is available. In this paper, we show that this problem can be circumvented by exploiting knowledge from existing mathematical models. An application of belief networks to assess the impact of climate change on potato production is used as an illustration. We show how the uncertainty of future climate change, variability of current weather and the knowledge about potato development can be combined in a belief network, which provides an aid for policy makers in agriculture. The model is tested using synthetic weather scenarios. The results are compared with those obtained from a conventional mathematical model