Training cattle for virtual fencing: Different approaches to determine learning success

Virtual fencing (VF) offers promising future perspectives for grazing, as it simplifies fencing through the use of GPS-coordinated VF-lines. Each animal is equipped with a VF-collar, which emits an acoustic signal when the animal approaches the VF-lines. The signal stops immediately the animal turns around but if the animal continues to move towards the VF-line, an electric pulse is emitted. The use of VF is based on the animal's ability to learn to associate the acoustic signal with the electric pulse, and thus, avoid the electric pulse by reacting appropriately to the signal. The intention of this study was to identify if heifers are able to learn the VF-system during a 12-day-period and how successful learning can be evaluated using three different approaches: i) a reaction score; ii) collar-stored data; iii) the integrated mode change function. 16 Fleckvieh heifers were enrolled in the study and divided into two groups of eight. They were not familiar with VF prior to this study. On the first day of the study the heifers were equipped with VF-collars and the two groups were assigned to two adjacent pastures. The behaviour of four heifers per group was continuously observed by two observers and behaviours were scored according to an ethogram (2 h a.m., 2 h p.m.). We analysed changes over four phases as an indication of successful learning, by measuring (i) the behavioural reactions to acoustic signals and electric pulses, (ii) acoustic signals, electric pulses, success ratio and confidence ratio. The development and calculation of the confidence ratio is our way of weighing the success ratio against the proportion of acoustic signals. Further, (iii) we assessed the time it took for the device to shift from teach mode to operating mode, which is an internal function of the VF-collars. The device changes modes due to the animals' 20th correct response to acoustic signals without receiving an electric pulse. We validated this function and assessed the time until the mode changes in two successive rounds (Collar restart to start round two on day eight) and found a significant difference (p<0.0001) between rounds with a faster mode change occurring in round two. All three approaches suggested successful learning. From our results, learning can be separated in a) learning to avoid the electric pulse and b) learning to interact with the acoustic signal. Therefore, combining different approaches and measuring changes over time is necessary to ensure sustained learning.