In search of an optimal bio-logger epoch and device combination for quantifying activity budgets in free-ranging cattle

Precision livestock management incorporates new technologies, including bio-loggers, to remotely monitor livestock health and behavior. Despite the potential benefits in extensive cattle systems, limited adoption of these sensors has occurred potentially due to cost, technical, or processing challenges. We resampled high-resolution GPS and accelerometer data across multiple epochs (spanning 10 s to 15 min) to evaluate which combinations of devices, data features, and epochs might be considered optimal for assessing resting, grazing, travel, and rumination behavior in free-ranging cattle. We used random forest models to predict cattle behavior across the growing season to assess how variations in model accuracy were reflected in inference of activity budgets. Classification accuracy was greatest (>0.90) when GPS information was combined with at least one accelerometer metric. Epochs of 30–90 s provided the greatest classification accuracy, although epochs up to 300 s had similar classification accuracies, but with increased variability in accuracy. Classification accuracies decreased when we included rumination, but similarly had the greatest performance when both GPS and a full suite of accelerometer features was used (accuracy of ∼0.90). Average daily grazing time (8.3 h day-1) was within 2 h across devices, epochs, and behavioral schemes. Rumination time was again similar across devices and epochs, averaging 6.5 h day-1. Daily travel distance decreased by ∼4 km as the GPS fix interval increased from 10 s to 15 min. This study provides guidance for balancing fine-scale data collection with data processing and battery limitations for assessing cattle behaviors in extensive rangelands.