Inference of Animal Activity From GPS Collar Data on Free-Ranging Cattle

Abstract

Global positioning systems (GPSs) enable continuous and automatic tracking of an animal’s position. The value of such spatial-temporal information can be improved if the corresponding activity of the animal is known. We evaluated the potential of Lotek GPS collars to predict activity of beef cattle on extensive rangeland in 2 contrasting foraging environments. Collars were configured to record animal location at intervals of 20 minutes (United States) or 5 minutes (Israel), together with counts from 2 motion sensors. Synchronized field observations of collared cows were conducted in 1999 (United States) and in 2002 and 2003 (Israel). Grazing, traveling (without grazing), and resting activities were recorded as minutes out of 20 for each category (United States), or as a single category (Israel). For the US data, stepwise regression models of grazing, traveling, and resting time accounted for 74%-84% of the variation, on the basis of the motion sensor counts for the left-right axis and the distances between GPS fixes. Regression tree analysis of grazing time yielded a simple model (4 splits) that accounted for 85% of the variation. For the Israeli data, the misclassification rates obtained by discriminant analysis and classification tree analysis of animal activity were 14% and 12%, respectively. In both analyses, almost all grazing observations were correctly classified, but other activities were sometimes misclassified as grazing. Distance alone is a poor indicator of animal activity, but grazing, traveling,and resting activities of free-ranging cattle can be inferred with reasonable accuracy from data provided by Lotek GPS collars.