A belief network-based system for predicting future crop production

Impact studies of future climate change on crop production have been hampered by the many uncertainties involved. Belief networks have been proved to be a very useful tool in dealing with uncertainties. However, the construction of a belief network for a complex application domain can be very difficult. Monte Carlo simulation can be used to exploit knowledge from existing mathematical models and hence ease the problem of belief network construction. A system was designed to show how the uncertainty of future climate change, variability of current weather, knowledge of human experts, and knowledge contained in crop simulation models can be integrated in a belief network, and to provide an aid for policy makers in agriculture. The constructed system was applied to simulate current and future potato growth in Kinless, Scotland, using synthetic weather data. Predictions given by our system agreed with those obtained from a conventional Monte Carlo simulation, and the system produced its predictions in a more flexible and efficient manner.