A Bayesian network approach to knowledge integration and representation of farm irrigation: 2. Model validation

For models to be useful, they need to adequately describe the systems they represent. The probabilistic nature of Bayesian network models has traditionally meant that model validation is difficult. In this paper we present a process to validate Inteca‐Farm, a Bayesian network model of farm irrigation that we described in the first paper of this series. We assessed three aspects of the quality of model predictions, namely, bias, accuracy, and skill, for the two variables for which validation data are available directly or indirectly. We also examined model predictions for any systematic errors. The validation results show that the bias and accuracy of the two validated variables are within acceptable tolerances and that systematic errors are minimal. This suggests that Inteca‐Farm is a plausible representation of farm irrigation system in the Shepparton Irrigation Region of northern Victoria, Australia.