Predictive models for grazing distribution: A GIS approach

Abstract

Grazing distribution and forage use patterns are important influences on rangeland ecosystems. Spatial patterns of grazing by domestic cattle (Bos taurus) were observed over 2 consecutive years under 2 grazing systems, intensive-early stocking and season-long stocking. The purposes were to determine factors influencing observed patterns and develop predictive models for grazing distribution and forage removal. Field-collected data on grazing distribution were linked with associated geophysical properties of pastures utilizing a GIS. Separate models were developed to predict grazing distribution and forage utilization using a backward stepwise regression procedure. The forage utilization model was linked with grazing distribution by utilizing Tobit analysis. Nineteen independent variables were used to interpret the observed variation in grazing distribution. Comparison of predicted probability of grazing values from the model with the observed grazing distribution in a hold-out data set yielded a close fit (R=.99). Eighteen independent variables were included in the forage removal model. Comparison of predicted forage removal with observed values in a hold-out data set yielded a poor fit (R=.28). Lack of forage quality variables probably accounts for the poor performance of the forage removal model. Differences in the success of the 2 models support the hypothesis that grazing distribution and forage utilization operate at different spatial scales and parameters. The use of GIS holds promise as a technique for developing useful predictive models for range management.