Validation of the RumiWatch System to Monitor Feeding and Locomotive Behaviors

By Glenda M. Pereira, PhD Animal Science student, Kirsten T. Sharpe, graduate student, and Bradley J. Heins, Associate Professor, Dairy Science

December 2019


Many precision technologies may accurately record feeding behavior; however, grazing behavior may be difficult to define because grazing may be considered both active and eating behavior because cows may graze while standing or while walking. Because many precision technologies do not monitor grazing behavior, grazing has not been researched as extensively. Recently in Ireland, a halter and pedometer system (RumiWatch, Itin and Hoch GmbH, Liestal, Switzerland) were validated for grazing behavior with 92% accuracy. Alternative grazing environments may provide opportunity for variation in monitoring behaviors. The objective of the study was to validate a halter and pedometer for monitoring feeding and locomotive behaviors by direct visual observation in a grazing dairy herd in Minnesota, USA. 

Validation of RumiWatch

The study was conducted at the University of Minnesota West Central Research and Outreach Center organic dairy in Morris, Minnesota from May to June 2018. Lactating crossbred dairy cows (n = 12) were offered pasture for 22 hours per day and milked twice per day. The pastures were comprised of grasses and legumes that included smooth bromegrass, orchardgrass, meadow fescue, alfalfa, red clover, and kura clover. Cows were stocked at a rate of 3 cows per hectare and rotated to a new paddock every 2 days, with 4,834 kg of DM/ha available at the initiation of grazing. 

The halter system can classify data as feeding behaviors, including ruminating, eating, drinking and other. In addition, the halter can classify jaw movements as grazing bites or rumination chews. The pedometer, a 3-axis accelerometer, monitors locomotive behaviors such as standing, lying and walking. Data from the halter and pedometer were collected in 10 Hz resolution, and the RumiWatch Converter V. transformed data into minute and hour summaries. Observational data were recorded by 3 trained observers on Samsung tablets, using the Pocket observer app (The Observer XT, Version 14.0, Noldus Information Technology, Leesburg, VA). Data from the visual observations were minutes and hour summaries. 

The first experiment determined agreement between visual observation and the halter and pedometer. For this experiment, 144 hours of feeding and locomotive behaviors were evaluated The second experiment evaluated correlation of grazing bites and rumination chews and 1,205 minutes were evaluated between visual observation and the halter system. 

Pearson correlations and concordance correlation coefficient (PROC CORR of SAS), bias correction factors (Cb), location shift (V) and scale shift (µ) (epiR package of R software) evaluated associations between direct visual observations and halter observations (Table 1). Correlations for feeding behavior between visual observations and the halter system were 0.84 (P < 0.01) for ruminating, 0.76 (P < 0.01) for eating, 0.39 (P < 0.01) for drinking, and 0.57 (P < 0.01) for other behaviors. Correlations for locomotive behaviors between visual observations and the pedometer were 0.83 (P < 0.01) for standing, 0.91 (P < 0.01) for lying, and 0.38 (P < 0.01) for walking. The correlation between visual observation and the halter system for grazing bites and rumination chews were 0.46 (P < 0.01) and -0.04 (P = 0.79), respectively. Together grazing bites and rumination chews had a correlation of 0.68 (P < 0.01) compared to visual observation. 


The results suggest the RumiWatch System may accurately monitor rumination and eating, as well as standing and lying behaviors. Behaviors such as drinking and walking were seldom observed and may be difficult to accurately monitor in grazing dairy cattle.