Bayesian Network for Point and Diffuse Source Phosphorus Transfer from Dairy Pastures in South Otago, New Zealand.

Many factors affect the magnitude of nutrient losses from dairy farm systems. Bayesian Networks (BNs) are an alternative to conventional modeling that can evaluate complex multifactor problems using forward and backward reasoning. A BN of annual total phosphorus (TP) exports was developed for a hypothetical dairy farm in the south Otago region of New Zealand and was used to investigate and integrate the effects of different management options under contrasting rainfall and drainage regimes. Published literature was consulted to quantify the relationships that underpin the BN, with preference given to data and relationships derived from the Otago region. In its default state, the BN estimated loads of 0.34 ± 0.42 kg TP ha−1 for overland flow and 0.30 ± 0.19 kg TP ha−1 for subsurface flow, which are in line with reported TP losses in overland flow (0–1.1 kg TP ha−1) and in drainage (0.15–2.2 kg TP ha−1). Site attributes that cannot be managed, like annual rainfall and the average slope of the farm, were found to affect the loads of TP lost from dairy farms. The greatest loads (13.4 kg TP ha−1) were predicted to occur with above-average annual rainfall (970 mm), where irrigation of farm dairy effluent was managed poorly, and where Olsen P concentrations were above pasture requirements (60 mg kg−1). Most of this loading was attributed to contributions from overland flow. This study demonstrates the value of using a BN to understand the complex interactions between site variables affecting P loss and their relative importance.