Main content area

A Bayesian Network for Comparing Dissolved Nitrogen Exports from High Rainfall Cropping in Southeastern Australia

Nash, David, Hannah, Murray, Robertson, Fiona, Rifkin, Penny
Journal of environmental quality 2010 v.39 no.5 pp. 1699-1710
nitrogen, soil nutrient balance, losses from soil, Bayesian theory, nutrient management, farm management, prediction, runoff, fertilizer application, application rate, surface water, rainfed farming, sowing, soil transport processes, Australia
Best management practices are often used to mitigate nutrient exports from agricultural systems. The eff ectiveness of these measures can vary depending on the natural attributes of the land in question (e.g., soil type, slope, and drainage class). In this paper we use a Bayesian Network to combine experiential data (expert opinion) and experimental data to compare farmscale management for different high-rainfall cropping farms in the Hamilton region of southern Australia. In the absence of appropriate data for calibration, the network was tested against various scenarios in a predictive and in a diagnostic way. In general, the network suggests that transport factors related to total surface water (i.e., surface and near surface interfl ow) runoff, which are largely unrelated to Site Variables, have the biggest effect on N exports. Source factors, especially those related to fertilizer applications at planting, also appear to be important. However, the effects of fertilizer depend on when runoff occurs, and, of the major factors under management control, only the Fertilizer Rate at Sowing had a notable effect. When used in a predictive capacity, the network suggests that, compared with other scenarios, high N loads are likely when fertilizer applications at sowing and runoff coincide. In this paper we have used a Bayesian Network to describe many of the dependencies between some of the major factors affecting N exports from high rainfall cropping. This relatively simple approach has been shown to be a useful tool for comparing management practices in data-poor environments.