Pork farm odour has become an increasingly important problem for the pork industry because non-farming
rural residents object to the odours coming from pig facilities and insist that it disrupts enjoyment of their properties. The pork industry has thus faced strong opposition to farm expansion or the creation of new pork farms. Reducing pork farm odour requires an understanding of what causes the odour and the ability to measure the odour. The pork industry and researchers have attempted to model pork farm odour using single-component odour indicators, such as ammonia and hydrogen sulphide, with statistical models. Single component analysis refers to only one odour indicator being used to predict odour levels. In this paper, a neural network approach to the pork farm odour using single-component analysis with the consideration of other relevant factors, such as measurement location is proposed. Neural network models and statistical models for pork farm odour have been developed and compared for single-component models to determine which method produces superior results. In general, the use of neural networks to model the pork farm odour yields more accurate and precise odour intensity predictions than the statistical models. The measurement location for the pork farm odour was considered in several model comparisons. The neural network models significantly outperformed the statistical models in this comparison because the statistical models are not able to consider the measurement location. This indicates that measurement location is a relevant factor for modelling pork farm odour. This also demonstrates that factors other than odour components should be considered during modelling. It is hypothesised that a multiple-component (odour components) and multiplefactor
(environmental conditions and other human expert knowledge) analysis approach to the modelling of
pork farm odour using neural networks and other intelligent systems techniques will yield increased accuracy for odour prediction and a thorough understanding of this significant problem.