Classifying season long livestock grazing behavior with the use of a low-cost GPS and accelerometer

Abstract Remote tracking of livestock through the use of GPS technology has tremendous potential for the study of livestock use patterns on the landscape. The addition of high frequency accelerometers on GPS units has potential to give researchers and managers the capability of accurately partitioning GPS points into differing behaviors, providing further insight into livestock grazing selection, pasture use, and changes in forage preference through time. Most studies to date have examined the use of GPS collars with accelerometers on small pastures and over relatively short time periods; evaluation of this technology is needed on larger pastures over extended time periods. The objective of this study was to test the effectiveness of predicting season-long livestock behavior through the use of a low-cost, non-commercial, lab-constructed GPS collar outfitted with a high frequency 3-axis accelerometer. Collars were deployed on yearling steers continuously grazing native grassland pastures for 3 months. GPS devices were set to record a fix at 1-minute intervals. Accelerometers were programmed to record X, Y, and Z positions at 12 Hz (12 records per second). The accelerometer data was aggregated to 1 s intervals initially, and mean, minimum, maximum, and standard error of X, Y, and Z axes were calculated between the start and stop time of each GPS fix. Additional metrics were calculated from the GPS fix data to include rate of travel and stationary point identification to aid in classification. Direct visual observations of cattle wearing the GPS collars were recorded for numerous time periods each grazing season to provide data for classifying GPS data into graze and non-graze behaviors. Several classification algorithms were tested to assess misclassification rates and predict behavior of unobserved data. Overall misclassification rate was lowest for a random forest model (11.2%). Time spent grazing ranged from 8.67 to 10.49 h daily, and timing of grazing tended to be heaviest during morning and evening hours, both of which were expected from yearling steers grazing native pasture in the summer. These results show there is great promise in using accelerometer-equipped GPS collars to accurately identify livestock grazing locations, which could benefit researchers and land managers monitoring rangeland use.