Efficacy of virtual fencing to manage two groups of cattle with adjacent virtual boundaries

Abstract

Virtual fencing (VF) allows cattle producers to manage grazing without additional physical fencing. Although VF is effective for managing one cattle group within a virtual paddock, its efficacy for managing multiple groups within the same pasture, separated only by VF boundaries, remains unclear. This capability could support flexible forage allocation among cattle groups with different nutritional requirements. The objective was to evaluate whether VF could maintain separation between adjacent cattle groups without physical cross-fencing. In this 8-week crossover study, 39 VF-trained yearling steers were assigned to three 13-steer herds. Two experimental herds alternated between a co-fenced treatment (CO), in which another cattle group grazed in an adjacent virtual paddock, and an isolated treatment (ISO), in which no neighboring cattle occupied an adjacent paddock. A third VF-managed group established the adjacent-cattle exposure for CO and was not an experimental herd. A 20-m buffer was implemented between adjacent virtual paddocks. Daily audio cues and electrical pulses per steer were recorded, and GPS locations were used to calculate containment. Containment was analyzed using a beta mixed-effects model with treatment, week, and pasture as fixed effects and herd as a random effect. Cue outcomes were analyzed using repeated-measures mixed models with treatment, day within week, and their interaction as fixed effects, week as a random effect, and an AR(1) covariance structure to account for repeated observations within herd. Fence treatment did not affect the number of audio (P = 0.92) or electric (P = 0.34) cues, nor the audio ratio (P = 0.51) received by each steer. The VF was effective at keeping steers contained, with a containment of 99.96 ± 0.1% and 99.68 ± 0.5% for CO and ISO (P = 0.10), respectively. Thus, VF effectively separated trained cattle groups under the tested 20-m-buffer configuration and could support flexible management of separate cattle groups within the same pasture without physical cross-fencing.

Managed by virtual fencing, trained steers stayed within their assigned grazing areas for more than 99% of monitored time, even with another group of cattle was nearby. This reliable separation could give producers greater flexibility to allocate forage among groups within a pasture.

Introduction

In cattle production systems, grazing management is often constrained by the labor and cost of infrastructure required to construct and maintain physical fences (Whitt and Wallander 2022; Parker et al. 1992; Gillespie et al. 2008). Virtual fencing (VF) has emerged as a management tool that allows producers to establish and modify grazing boundaries without building additional physical infrastructure (Goliński et al. 2022). This technology has the potential to reduce labor demands associated with fencing (Umstatter 2011) while enabling more flexible and responsive grazing management, including applications focused on conservation and adaptive grazing strategies (Melo-Velasco et al. 2024).

In many beef operations in the U.S., multiple classes of cattle with differing nutritional requirements are managed simultaneously. Young cows, replacement heifers, and stocker calves typically benefit from access to higher-quality forage (Heinrichs 1996), whereas mature cows can often utilize lower-quality forage without compromising performance. Grazing strategies that separate animals by nutritional demand, such as leader–follower systems, can improve forage utilization and animal performance (Mayne et al. 1988; French et al. 2001); however, these approaches often require extensive fencing and labor inputs, which can limit adoption compared to cattle managed in a single traditionally fenced management systems (King 2016).

Virtual fencing may provide an opportunity to manage multiple groups of cattle separately within the same pasture without the need for additional physical cross-fencing. While VF has been shown to effectively contain cattle across a range of grazing contexts (Goliński et al. 2022), questions remain regarding its ability to maintain effective separation between cattle groups housed in close proximity. Cattle are highly social animals, and attraction to adjacent groups may increase pressure on virtual boundaries, potentially reducing containment efficacy. For example, Verdon et al. (2021) reported variable containment between two groups of Angus heifers managed with virtual front and back boundaries in adjacent paddocks. Containment declined from 99.9% to approximately 92% across grazing allocations in one virtually fenced group, whereas a second virtually fenced group maintained consistently high containment. Reduced access to fence-line physical contact with other cattle may have contributed to increased boundary crossing by one group. Based on these observations, the authors recommended that VF-managed groups be housed outside visual contact of other cattle. However, these recommendations were based merely on observational findings rather than a controlled study specifically designed to evaluate whether adjacent cattle groups can be effectively managed using VF alone.

Therefore, the objective of the present study was to evaluate the effectiveness of virtual fencing for managing two groups of cattle separated solely by virtual boundaries. Specifically, containment, animal–virtual fencing interactions (stimuli measures), and spatial distribution were compared between cattle managed in adjacent VF paddocks and cattle managed in isolation.

Materials and methods

Animals and treatments

All animal-related research in this study was approved by the University of Nebraska- Lincoln IACUC under protocols 2206 and 2512, and Experiment # 2318. This experiment was conducted at the University of Nebraska-Lincoln Eastern Nebraska Research, Extension, and Education Center (ENREEC) near Mead, Nebraska.

Mixed-breed Bos taurus yearling beef steers (n = 39) were used in this 56-d grazing study. Steers were stratified by VF training responsiveness and randomly assigned to one of three herds (n = 13 steers/herd). Two herds served as experimental units (EU), whereas the third herd was used solely to create conditions for potential social attraction to adjacent cattle and was not included in statistical analysis.

Two virtual fencing treatments were evaluated: co-fenced (CO) and isolated (ISO). In the CO treatment, an experimental herd grazed within a virtual paddock (VP, pastures virtually designated by the virtual fencing system) adjacent to the third herd, with the paddocks separated solely by virtual boundaries and no physical fence present. In the ISO treatment, an experimental herd grazed within a VP without cattle present in adjacent paddocks. The two experimental herds alternated between CO and ISO treatments in 1-week periods throughout the study.

Two pastures with similar sizes (approximately 12.1 ha each) were used in the study (Fig. 1). The forage base consisted primarily of smooth bromegrass (Bromus inermis), and grazing was conducted during the summer months (June to August) of 2024. Pasture 1 consisted of rolling terrain with several hills, whereas Pasture 2 contained several patches of trees. Each pasture was subdivided into three equal-area paddocks of approximately 4 ha each (Fig. 1). Two paddocks within each pasture were assigned to the CO treatment, where an EU herd grazed adjacent to the nonexperimental VF-managed herd. These co-fenced paddocks were separated solely by virtual boundaries, with an approximately 20-m buffer zone between VP as a pragmatic risk-management measure to maintain close visual proximity between groups while preventing overlap of virtual boundaries. The third paddock was assigned to the ISO treatment, where an EU herd of steers grazed without cattle present in adjacent paddocks. For example, when Herd A and the nonexperimental herd (X) grazed in VP1 and VP2 of Pasture 1, Herd B grazed in VP3 of Pasture 2 (Fig. 1). Throughout the study, water sources were positioned on the opposite side of the shared virtual boundaries to allow unrestricted access to water while minimizing unintended interactions with the VF system.

Figure 1.

Experimental pasture layout illustrating the virtual paddock design. Each pasture was subdivided into three paddocks. Paddocks 1 and 2 were assigned to the co-fenced treatment, in which an experimental herd grazed adjacent to another herd separated solely by virtual boundaries without a physical fence. Paddock 3 was assigned to the isolated treatment, in which an experimental herd grazed without adjacent cattle present. Water locations are shown and were positioned away from the adjacent virtual boundaries to minimize unintended boundary interactions. Co-fenced and isolated treatments were not applied within the same pasture during the same week. Dimensions are not drawn to scale.

Virtual fencing system

Steers in this experiment were fitted with eShepherd virtual fencing neckbands (version ES-1, Gallagher Animal Management, Kansas City, MO). These VF neckbands were preliminarily tested to have a positioning accuracy of 2.5 m CEP50 (CEP, Circular Error Probable) and 5.3 m CEP95, which is consistent with the typical global positioning system (GPS) accuracy of 3 to 5 m considered acceptable for animal research and applications (Zhao et al. 2024). Virtual paddocks were created, and steers were individually associated with neckbands on the eShepherd webtool interface. The VP coordinates are transmitted to corresponding neckbands and cattle herds via Internet of Things communication protocols. As cattle approach the VP boundary (within 5 m for ES-1 neckbands), the neckband plays a 5 s audible warning (audio cue) with a rising pitch, followed by a 1 s electrical pulse (pulse) that is similar to the electric energy emitted by electric-wire fences (Goliński et al. 2022). If at any time during this sequence the animal reacted and turned away from the virtual boundary, the 5-s audio warning stops the warning and prevents a pulse; continued approach triggers the 1-s pulse; and three consecutive unsuccessful sequences lead to escaped status, with both stimuli suspended until the steer returns. Escaped animals do not receive any stimuli when returning to the designated VP area. After the animal returned to the VP, the VF system would resume normal sequence of stimuli if the animal approached the virtual boundary again.

Training procedure

Training allows VF-naive cattle to learn that the audio cues predicted a subsequent electrical pulse if they continue toward the virtual boundary and that stopping or turning away prevent pulse delivery (Lee et al. 2009; Lee and Campbell 2021). In this study, the training period lasted 12 days (Fig. 2). On the first day of training, neckbands were fitted on 45 growing steers (average body weight 341.8 ± 17.3 kg) while being restrained in a chute. Immediately following collaring of the neckbands, the steers were turned out into a 5.66 ha training pasture (measured approximately 300 m by 210 m), which was separate from the two experimental pastures. Cattle were allowed to acclimate to the training pasture and neckbands for the first 2 days before the neckbands were activated with the corresponding VP (acclimation period).

Figure 2.

Training phases, duration, and corresponding virtual fencing boundaries used during the steer training procedure. Virtual fencing boundaries were adjusted every 2 to 4 d using the eShepherd user interface. The overall training protocol was conducted over a 12-d period prior to initiation of the trial.

On training day 3, VP boundaries were placed directly over the existing physical perimeter fence of the pasture. Thus, the steers received audio and pulse cues if they came within 5 m of the physical fence (Phase 1). This phase lasted approximately 3 days. Beginning on training day 6, a virtual boundary was moved into the pasture on one side, while the other three VF boundaries remained over the physical fence (Phase 2) to encourage cattle to interact with a virtual boundary. This phase lasted approximately 4 days. Similarly, for Phases 3 and 4, VF boundaries were moved into the pasture away from any physical fence while some portions of the virtual boundary remained over the physical fence. Each of these phases lasted approximately 2 days. As shown in Fig. 2, the complex design of the final two phases (3 and 4) was designed to encourage additional interactions between steers and the VF boundaries.

In order to estimate an individual’s response to VF stimuli, an audio to pulse ratio (audio ratio) was calculated using Equation (1) for each animal (Campbell et al. 2017, 2019; Confessore et al. 2022). Audio ratio is an estimate of animal responsiveness to the cues delivered by the neckbands. As the audio ratio increases, it is assumed that the animals effectively respond to the audio cues with less or no need to receive a pulse. Therefore, a higher audio ratio indicates a more successful training outcome. As suggested by the manufacturer, an over 80% audio ratio for above 80% of the herd indicates that the majority of the steers within the herd were effectively trained with the VF system.

$$\mathit{Audio} \mathit{Ratio}=\frac{\mathit{Number} of \mathit{Audio} \mathit{cues}}{\mathit{Number} of \mathit{Audio} \mathit{cues}+\mathit{Number} of \mathit{Pulse} \mathit{cues}}\times 100$$ (1)

Upon completion of the training period, all 45 steers were stratified by audio ratio (78 ± 15%). During training, one steer continued through the virtual boundary, breached the physical perimeter fence, and left the study site; consequently, it was unavailable for subsequent data collection and was excluded from the study. Two other steers were excluded due to unrelated health reasons. Then three steers with the lowest audio ratio, those who received the greatest number of pulses in relation to audio cues (ie least responsive to audio cues), were removed from the trial, resulting in a total of 39 steers being used for the study, of which 25 had an audio ratio above 80% (64% of the total steers).

For end-of-training stratification, the audio ratio was calculated separately for each steer on each cue-active training day, and the daily ratios were averaged across cue-active days to obtain one training-response value for each steer. These 39 steers were then stratified by audio ratio (81 ± 6%) and assigned randomly into one of 3 herds, with 13 steers per herd. Two herds of steers (A or B) were randomly assigned as the EU, whereas herd (X) was assigned to graze in the adjoining virtual paddock for the CO treatment. Location and boundary interaction data from herd X was collected via VF neckbands but did not undergo statistical analysis.

Data collection and analysis

Training responses

Training responses were summarized descriptively for the 39 steers retained for the grazing trial across the 10 cue-active training days. For each steer-day, audio and pulses were counted. Audio ratio was considered undefined when no cues were delivered; however, zero counts were retained in summaries of audio- and pulse-cue delivery. Daily distributions were summarized using medians, interquartile ranges, and box-and-whisker plots. Detailed results are provided in Figure S1 and Table S1.

Main trial data collection

Immediately following the training period, steers were sorted into their respective herds and turned out to the experimental pastures (day 1). The data collection period was divided into eight 7-d periods (weeks 1–8). During each week, herds were randomly assigned to pastures and housed in either the CO or ISO paddocks within each pasture (Table 1). Based on the location of the pastures, steers were loaded onto trailers and hauled to their respective paddocks. At the conclusion of each data collection week, VP were deactivated, steers were gathered on horseback and then loaded and hauled to their next VP. Once arrived, steers were held in a physical holding area until the VP were activated for all animals (approximately 20 min) before moving into their VP. This entire sequence occurred on the morning of day 8 of each period, which also marked the beginning of the subsequent data collection period. On day 15 (end of period 2), due to personnel shortage, virtual fencing was deactivated, and steers were removed from the trial pastures. All steers were co-mingled in the training pasture and VF remained deactivated for 6 days. On day 21, cattle were re-sorted into their respective herds and data collection period 3 began. Periods 3 through 8 were completed in sequence without interruption. On the morning of day 64 (conclusion of period 8), neckbands were deactivated for all cattle before being removed by animal care technicians.

Table 1.

Weekly assignment of herds^a to pasture, paddock, and virtual fencing treatment during the 8-wk trial.

			Week
Pasture	Treatmentb	Paddock	1	2	3	4	5	6	7	8
1	CO	1	–	X	–	–	B	A	–	X
		2	–	B	–	–	X	X	–	A
	ISO	3	B	–	A	B	–	–	A	–
2	CO	1	A	–	X	X	–	–	B	–
		2	X	–	B	A	–	–	X	–
	ISO	3	–	A	–	–	A	B	–	B

^aHerds A and B were the experimental units. Herd X was not analyzed and was used solely to establish an adjacent group for the co-fenced (CO) treatment. A dash (–) indicates that the herd was not present in that pasture or paddock during the specified week.

^bCo-fenced (CO) indicates that an experimental herd grazed in a virtual paddock adjacent to another herd, with paddocks separated only by virtual boundaries and no physical fence. Isolated (ISO) indicates that an experimental herd grazed in a virtual paddock with no other cattle present in adjacent paddocks.

Neckbands recorded GPS coordinates at approximately 10-min intervals, along with the number of audio and electrical cues delivered within each interval. Records lacking valid latitude or longitude values were excluded from analysis. A data retrieval rate was calculated to assess dataset completeness and was defined as the proportion of successfully retrieved records relative to the total number of expected records.

Containment

Based on the GPS data collected for each EU during each period, containment percentage was calculated using Eq. 2. Because GPS locations were recorded at approximately 10-min intervals, this GPS-derived measure estimates the proportion of monitored steers time spent within the assigned virtual paddock. This metric was used to evaluate the efficacy of the VF system in maintaining cattle within the designated virtual boundaries.

$$\mathit{Containment}=\frac{\mathit{GPS} \mathit{points} \mathit{within} \mathit{the} VP}{\mathit{Total} \mathit{GPS} \mathit{points}}\times 100$$ (2)

Animal—virtual fencing interaction

For each data collection period, data were summarized at the EU level by calculating the total number of audio and electrical cues per EU for each day and period. These values were then divided by the number of cattle in each EU (n = 13) to determine the average number of audio and electrical cues received by each animal per day.

Similarly, GPS coordinates for each EU were compiled for each data collection period. To evaluate cattle spatial distribution relative to the virtual boundaries and adjacent groups, each VP was divided into 5 equal-area zones (Fig. 3). In the CO treatment, Zone 1 represented the area closest to the adjacent cattle group, whereas Zone 5 represented the area furthest from the shared virtual boundary. In the ISO treatment, Zone 1 represented the virtual boundary separating the experimental VP from the remainder of the pasture, and Zone 5 represented the area furthest from that boundary. Water sources were located in Zone 5 in all paddocks to minimize potential confounding effects associated with cattle loafing near water. Spatial distribution was expressed as the proportion of observed GPS points within each zone during each data collection period. These proportions were interpreted as estimates of the proportion of monitored animal-time that each herd allocated to each zone. This measure represents group-level estimated time allocation rather than continuously measured duration. To complement the zone-based analysis, valid GPS observations from the experimental herds were visualized descriptively using two-dimensional kernel-density heatmaps pooled within each experimental pasture across the 8-week trial (Figure S2). These heatmaps were generated only to illustrate localized spatial concentrations of GPS observations and were not used for statistical inference.

Figure 3.

Schematic representation of pasture zones used to quantify spatial distribution of GPS locations relative to the primary internal virtual fencing boundary within each virtual paddock. Each virtual paddock was divided into five parallel zones (Zones 1–5) oriented perpendicular to the boundary. Zone 1 represented the area closest to the primary internal virtual fencing boundary, whereas Zone 5 represented the area furthest from the boundary. Zones were defined consistently across fence treatments and pasture layouts regardless of paddock orientation. GPS locations within each zone were used to estimate cattle distribution and relative time allocation across paddock zones.

The positional dataset of steers was processed in Python 3.7.13 using libraries of pandas and geopandas. The pandas library reads raw data files as tabular dataframes, and the geopandas library is used for geospatial analysis. For instance, geopandas spatially separated GPS coordinates within and outside the designated VP, from which the containment was calculated for each EU during each period using Equation (2).

Statistical analysis

All statistical analyses were performed using the GLIMMIX procedure of SAS (version 9.4; Cary, NC). Normality assumptions were evaluated for all variables using the Shapiro-Wilk test. Variables that did not meet normality assumptions were transformed prior to analysis. Significance was declared at P ≤ 0.05 and tendencies at P < 0.10.

Containment was expressed as a proportion (0 to 1) and analyzed using a beta mixed-effects model to accommodate proportional outcomes. Values of 0 and 1 were adjusted slightly toward the interior of the distribution to satisfy beta distribution assumptions [e.g., a value of 1 (100% containment) was adjusted to 0.999]. The statistical model included the fence treatment, week and pasture as fixed effects and a random intercept was specified for herd to account for clustering of repeated observations within herd.

Animal-fence interactions (audio cues, pulse cues, and audio ratio) were not normally distributed and therefore required transformation prior to analysis. Several transformation methods, including logarithmic, square root, Box-Cox, arcsine and Yeo-Johnson transformations (Weisberg 2001), were evaluated for each variable, with the optimal transformation selected based on the highest Shapiro-Wilk W statistic. The audio cue data was transformed with a square root transformation (W = 0.99), the pulse cue data was transformed using the Yeo-Johnson transformation with lambda value set at -1.2 (W = 0.97), and the audio ratio data was transformed using the arcsine transformation (W = 0.95). Following transformation, the animal-fence interactions were analyzed using mixed-effects models. Each model included fence treatment, day within week, and the treatment × day interaction as fixed effects, with week included as random effect. Herd was specified as the subject, and an autoregressive covariance structure of order 1 [AR (1)] was used to account for correlation and repeated daily observations within herd.

Spatial distribution was expressed as the proportion (0 to 1) of GPS observations recorded within each zone. These proportions were analyzed using a beta mixed-effects model. Fence treatment, pasture zone (1–5), week, pasture, and the fence treatment × zone interaction were included as fixed effects and a random intercept was specified for herd to account for clustering of repeated observations within herd. Since the interaction of fence treatment and zone was significant, pairwise comparisons of CO vs. ISO within each zone were obtained using the DIFF option to determine whether fence treatment affected spatial use of that zone.

Results

Main trial data

The overall data retrieval rate was 99.3% for both herd A and herd B. These retrieval rates indicate that more than 99% of the neckband-recorded data were retained for analysis, with less than 1% excluded due to missing GPS coordinates. Although some variables were transformed prior to statistical analysis, all values presented in this manuscript are reported as raw means.

Containment

The overall containment did not differ between fence treatments over the course of this study (P = 0.10), averaging 99.96 ± 0.1% for CO and 99.68 ± 0.5% for ISO (Table 2). Within the experimental herds, steers that briefly exited their designated VP returned voluntarily without human intervention. In herd X, one steer escaped from the VP and joined the adjacent grazing herd during week 2 and did not voluntarily return to the assigned paddock. This steer was sorted back into the appropriate VP at the conclusion of week 2.

Table 2.

Mean containment, daily number of audio and pulse cues received per steer, and audio ratio^a during the trial period by virtual fencing treatment^b.

	Fence treatment
	CO	ISO	SEM	P-value
Containment (%)	99.96	99.68	0.105	0.10
Audio Cues/steer/day	15.0	15.2	2.70	0.92
Pulse Cues/steer/day	0.57	0.73	0.135	0.34
Audio Ratio (%)	95.68	95.74	0.804	0.51

^aAudio and pulse cues are reported as the raw number of cues received per steer per day. Audio ratio was calculated as the proportion of total cues that were audio cues [audio cues ÷ (audio cues + pulse cues) × 100].

^bCO, co-fenced treatment, in which an experimental herd grazed adjacent to another herd separated only by virtual boundaries; ISO, isolated treatment, in which an experimental herd grazed without adjacent cattle.

Animal-virtual fencing interactions

All animal-VF interactions (audio cue, pulse cue, and audio ratio) are reported as a raw number of cues per steer per day (Table 2). No fence treatment differences were observed for either audio cues (P = 0.92) or pulse cues (P = 0.34) received per day per steer, or the audio ratio (P = 0.51) received during the trial period. Across both treatments, steers received an average of approximately 15 audio cues and less than 1 pulse cue per day.

Spatial distribution

A fence treatment × zone interaction was observed (P = 0.03) for the proportion of time cattle spent within each paddock zone (Table 3). No fence treatment differences were detected for Zones 1, 2 or 4 (P ≥ 0.11). However, cattle in the ISO treatment spent a greater proportion of time in the center of the paddock (Zone 3) compared with steers grazing in CO (25.21 ± 2.41% vs. 15.21 ± 3.48%; P = 0.03). In contrast, cattle in the CO treatment allocated a greater proportion of monitored animal-time to Zone 5, the area furthest from the adjoining virtual boundary, compared to cattle grazing in the ISO treatment (21.75 ± 5.07% vs. 13.10 ± 2.49%; P = 0.05). Overall, cattle distribution was more evenly dispersed across paddock zones in the ISO treatment, while cattle in the CO treatment allocated proportionally more time near the paddock edges (Zones 1 and 5). This general pattern was also apparent in the descriptive kernel-density heatmaps (Figure S2).

Table 3.

Percentage of steers spatial distribution in each zone^a by fence treatment^b over the course of the 8-week grazing trial.

	Fence treatment		SEMc	P-value Treatment × Zone
	CO	ISO
Zone 1	36.40d	27.71d	3.80	0.03
Zone 2	15.03d	18.81d	2.23
Zone 3	15.21e	25.21d	2.41
Zone 4	11.64d	15.16d	1.85
Zone 5	21.75d	13.10e	2.49

^aEach zone represented 1/5 of the total virtual paddock area (4 ha). Zone 1 was the area closest to the primary virtual boundary (closest to adjacent cattle in the CO treatment) whereas zone 5 was furthest from the primary virtual boundary. The water tank for each pasture was located in zone 5.

^bCO = co-fenced treatment, in which an experimental herd grazed adjacent to another herd separated only by virtual boundaries; ISO = isolated treatment, in which an experimental herd grazed without adjacent cattle.

^cRepresents the pooled standard error of the mean.

^d,eMeans within the same row lacking common superscripts differ (P < 0.05).

Discussion

Containment observed in the present study were high (>99%) and were within the upper range reported in the literature for cattle managed using virtual fencing. Many studies reported a containment of above 90%, suggesting VF to be effective at keeping cattle within VP areas (Colusso et al. 2021; Verdon et al. 2021; Perea et al. 2026). Previous studies evaluating the eShepherd system under relatively intensive grazing conditions have similarly reported containment exceeding 99% for dairy cows, even during early grazing periods. For example, Lomax et al. (2019) reported >99% containment in mature dairy cows managed in a strip grazing system using eShepherd, whereas Langworthy et al. (2021) observed similar containment (>99%) in mature dairy cows grazing paddocks approximately 70 × 100 m in size. Harland et al. (2025) utilized the Nofence system to manage beef heifers and cow–calf pairs under rotational grazing and reported overall containment exceeding 99%, although variation in learning patterns was observed among animal cohorts and individuals. These studies suggest that VF can establish effective boundaries for containing cattle within relatively intensive grazing systems.

High containment has also been reported in some extensive grazing environments, although results have been more variable among studies. A study evaluating Rarámuri Criollo cattle managed with the Nofence virtual fencing system in rangeland pastures also reported containment exceeding 99%, despite variation in behavioral and learning responses among individual animals (Campa Madrid et al. 2025). In contrast, lower and more variable containment has been reported under extensive grazing conditions with beef cattle using the Vence system (Merk Animal Health, Rahway, NJ; Boyd et al. 2023; Vandermark 2023). Vandermark (2023) evaluated Vence VF system (version 2 and 3 collars) in large, extensive Northern Great Plains rangeland pastures managed under both continuous and rotational grazing systems. Across years and treatments, mean containment ranged from approximately 54% to 78%, with substantial temporal variation. Similarly, Boyd et al. (2023) reported declining containment over time (30-d) in a fuel-break grazing study using the Vence system with containment decreasing from >94% early in the grazing period to approximately 75% by the end of the trial. Importantly, the decline in containment observed by Boyd et al. (2023) differed among animal classes. Containment of dry cows remained high and relatively stable throughout the grazing period (∼97% to 99%), whereas containment of lactating cows declined substantially over time. The authors suggested that increasing forage limitation within the fuel break, combined with attraction (“pull”) toward cow’s non-collared calves, may have increased pressure on virtual boundaries as grazing progressed.

The role of feed-driven motivation in influencing cattle movement and interactions with virtual boundaries has previously been examined under controlled conditions. Colusso et al. (2021) reported increased boundary challenges early in the experimental period when dairy cows fed an energy-restricted ration were motivated to access supplemental alfalfa cubes. However, cattle subsequently learned to associate audio cues with the virtual boundary and demonstrated improved boundary compliance later in the trial. These findings indicate that elevated feed motivation may influence early interactions with VF but does not necessarily prevent learning or effective containment over time. In the present experiment, each herd of 13 steers grazed an approximately 4-ha VP for a 7-d period before reassignment. Grazing concluded before forage availability declined to limiting levels (based on visual assessment), which likely reduced the motivational pressure to cross virtual boundaries despite cattle having visual access to ungrazed forage beyond the virtual boundary.

In addition to feed-related influences, social interactions both within and among cattle groups may affect VF efficacy. In a study investigating the social hierarchy of cattle using VF, Keshavarzi et al. (2020) determined that out of the 64 angus steers used, 3 “leader” animals were responsible for initiating a majority (75%) of the group’s interactions with the VF. This suggests that within the social hierarchy, animals may learn to respond to other’s interactions with the VF (ie move away from the VF boundary after hearing another animal’s audio cue). In a study conducted by Colusso et al. (2020), cows were trained to a VF system either individually or in groups of 5–6 and were subsequently tested in the alternate social context. During this crossover phase, cows trained in groups and then tested individually had a greater predicted probability of receiving virtual-fence stimuli than cows trained individually and then tested in groups (88% vs. 36%). The authors suggested that, during group training, some cows may have responded to the behavior or VF cues of their conspecifics without directly experiencing the paired stimuli themselves (Colusso et al. 2020). This finding indicates that social facilitation may influence individual learning within a group and that group-training protocols should provide sufficient opportunity for each animal to interact with and learn the VF cues.

Verdon et al. (2021) reported variability in containment when evaluating Angus heifers managed in an intensive cell-grazing system using moving front and back virtual boundaries with the eShepherd system. While one group of heifers maintained near-complete containment throughout the study, a second group exhibited increasing time spent in the exclusion zone, with effective containment declining from 99.9% to 92.0%. Notably, these differences between groups coincide with changes in the nature of contact between adjacent cattle groups. The group exhibiting reduced containment experienced a transition from physical fence-line contact with adjacent cattle to visual contact only using virtual boundaries, without a physical barrier restricting approach toward neighboring animals. In contrast, the second herd maintained fence-line contact with adjacent cattle throughout the study and demonstrated consistently high containment of the VF. Based on these observations, the authors suggested that VF-managed groups should be maintained outside of visual contact of neighboring cattle (Verdon et al. 2021).

In contrast to the case study reported by Verdon et al. (2021), the present experiment was explicitly designed to evaluate containment and spatial distribution when adjacent cattle herds were separated solely by virtual boundaries. Despite experimental conditions that allowed for potential social attraction (“social pull”) between herds, containment in the CO treatment remained consistently above 99%, indicating that the presence of adjacent cattle herd alone was insufficient to reduce containment under the conditions of this study. It should be noted, however, that a 20-m buffer zone was created between adjacent CO herds to increase separation between herds and minimize potential breakouts. The 20-m buffer was selected pragmatically and was not evaluated against alternative separation distances. Consequently, the present findings apply specifically to the configuration tested and do not establish whether a buffer is necessary, whether 20 m is the minimum effective distance, or whether comparable containment could be achieved using a shared virtual boundary.

In the present study, the average numbers of audio and pulse cues received per steer per day were not affected by fencing treatment, indicating that the presence of an adjacent cattle herd did not increase steer interactions with the virtual boundary. Previous studies suggest that daily interaction frequency becomes relatively stable after cattle become familiar with the VF system. Verdon et al. (2021), for example, reported averages of 15.7 audio cues and 0.67 pulse cues per animal per day after training, which were similar to the 15.1 audio cues and 0.65 pulse cues per steer per day observed during the main trial in the present study.

Appropriate training is fundamental to successful VF implementation, although training approaches and duration vary considerably among VF systems and studies (Hamidi et al. 2024; Wilms et al. 2024). In the present study, naive steers first underwent a 2-d acclimation period during which VF cues were not activated, followed by 10 d of cue-active training, during which pulse delivery declined as cattle gained experience with the VF system (Figure S1; Table S1, available as supplementary data at Translational Animal Science online). During the main trial, the audio ratio remained >95%, indicating that most VF interactions were resolved following the audio cue without requiring an electrical pulse. Similar increases in responsiveness to audio cues over time have been reported by Lomax et al. (2019) and others, consistent with continued associative learning. Although cue algorithms and reporting conventions vary among commercial VF systems, these results indicate that steers in the present study were well accustomed to the VF system before and during experimental evaluation.

In the present study, spatial distribution across the five paddock zones differed between fence treatments. However, no treatment difference was detected in Zone 1, providing no statistical evidence that the presence of an adjacent cattle group increased use of the zone closest to the neighboring group. Interpretation of differences for Zones 3 and 5 maybe due to tree patches and/or terrain. In Pasture 2, several tree patches appeared to attract cattle congregation, particularly within the isolated paddock (Fig. 1; Paddock 3). These localized concentrations are illustrated in the descriptive heatmaps presented in Figure S2. Additionally, rolling terrain and hills present in Pasture 1 may have influenced loafing and resting locations of cattle, resulted in certain high animal-time allocation in these zones.

Although still scarce in the literature, a few recent studies have evaluated spatial distribution patterns in cattle managed with VF systems. Lund et al. (2024) examined spatial positioning among Angus cattle managed using the Nofence system and reported no consistent spatial-distribution pattern. Grinnell et al. (2026) reported that heifers spent less time within a 10-m perimeter zone than in the paddock center under both virtual and physical electric fencing. Heifers also moved more slowly near boundaries and concentrated lying behavior in the paddock interior, indicating that spatial patterns near a boundary may reflect boundary recognition. Direct comparison with the present study is limited because Grinnell et al. defined a perimeter around the entire paddock and calculated time from approximately 1-min GNSS fixes, whereas the present study evaluated the proportion of approximately 10-min GPS observations within five equal-area zones oriented relative to a focal virtual boundary. Nevertheless, these studies indicate that virtual boundaries can influence within-paddock spatial distribution even when cattle remain effectively contained.

Conclusion

The results in this study indicate that VF created effective boundaries when cattle were co-fenced, allowing animals to be managed in close proximity to other cattle without the need for a physical barrier. Containment exceeded 99%, and there was no difference in the number of animal–virtual fencing interactions between co-fenced and isolated treatments, indicating that the presence of adjacent cattle did not compromise fencing effectiveness. Consistent with previous literature, steers in this study learned to associate audio cues with the virtual boundary and responded primarily to audio cues with less need for pulse cues. This learning resulted in effective containment across treatments, supporting the use of VF as a reliable management tool under the conditions evaluated. The results obtained in this study highlight the potential for VF to facilitate more flexible grazing strategies, such as leader–follower systems, by enabling separate management of multiple cattle groups without constructing additional cross-fencing infrastructure.

Glossary

List of abbreviations

CEP

circular error probable

CO

co-fenced

EU

experimental units

GPS

global positioning system

ISO

isolated

VF

virtual fencing

VP

virtual paddock

Supplementary data

Supplementary data is available at Translational Animal Science online.

Funding

This research was supported by United States of Department of Agriculture grant 58-3040-1-013: Precision Livestock Management Systems for Western Rangelands.

Conflicts of interest

None declared.