Predictive modelling of hill-pasture productivity: integration of a decision tree and a geographical information system

Abstract

One challenge in predictive modelling of productivity for pastures varying in topography, soils or management is to achieve the prediction over space with acceptable accuracy. As a new modelling approach, the decision tree has been shown to have high predictive accuracy; while geographical information systems (GISs), with their strong ability to deal with spatial factors, have been widely used in environmental modelling. Integration of a decision tree approach with a GIS offers a potential solution in meeting this challenge. In this study, decision tree models were developed for annual and seasonal pasture productivity (aboveground dry matter in kg/ha) using environmental and management variables and the outputs of these decision trees were integrated with a GIS to get predictions of pasture productivity in a hill-pasture grazing system. Results showed that the decision tree model for annual pasture productivity adequately predicted 91% of cases in the model validation, and the GIS-based prediction for annual pasture productivity was verified in three of four test farmlets. The decision tree models also revealed the relative importance of environmental and management variables and their interaction in influencing pasture productivity. Hill slope, soil Olsen P and annual P fertiliser input were the most significant variables influencing annual pasture productivity, while hill slope, annual P fertiliser input, autumn rainfall and soil Olsen P were the most significant variables influencing spring, summer, autumn and winter pasture productivity, respectively. The successful integration of the decision tree model with a GIS in this study provided a platform to predict pasture productivity for pastures with heterogeneous environmental variables and management features, and to present model predictions over space for further application and investigation. This modelling approach can be used as, or incorporated in, decision support systems to improve pasture management, and to investigate the interrelationship between pasture productivity and environmental and management variables.