Prediction by decision tree modelling of the relative magnitude of functional group abundance in a pasture ecosystem in the south of Chile

Several pasture improvement methods were compared over three years in the humid temperate region of southern Chile. The control treatment was non fertilized naturalized pasture (NFP), and the improvement treatments included fertilized naturalized pasture (FP), sown with L. perenne - T. repens (Mixed), and sown with a four-species mixture (Diverse), in a randomized complete block design. Decision tree modelling was used to predict the relative abundance of the functional groups: high fertility response grasses (HFRG), low fertility response grasses (LFRG), legumes and flatweeds. The fitted model output was contrasted with regression models. Soil chemical and physical variables and pasture attributes were measured. The ranking of annual dry herbage mass by the third year was Diverse > FP > Mixed > NFP, with Diverse being dominated by HFRG (68%) and NFP by flatweeds (50%). The decision tree models indicated the main factors influencing the abundance of functional groups were soil Olsen-P, soil mineral nitrogen and soil pHwater. For HFRG, the increase of soil mineral nitrogen (>45.1 mg kg-1) was the strongest variable to stimulate their abundance, for LFRG and flatweeds it was low soil Olsen-P (<9.2 mg kg-1), while for legumes it was the increase in soil pHwater (>5.6). The decision tree model performed better than the regression model with respect to model fit. Decision tree modeling can be used to set soil fertility targets according to a determined objective for pasture functional group abundance. The successful integration of environmental variables influencing the abundance of functional groups with decision tree models provided a new approach to the sustainable management of pasture functional groups.