Results and Discussion

1) Cattle navigate their environment and the virtual fence system by learning the relationship between audio warnings and shocks.

In order for virtual fencing to be successful in Alberta, we need to know whether cattle learn to use the warning and shock system to navigate their environment. A birds eye view of exactly how the cattle interacted with the virtual fence system allows us to see a clear divergence in numbers of warnings and shocks (Figure 7), which is further illustrated by the change in the ratio of shocks to warnings over time (Figure 8). At the beginning of the training period we see that the ratio is about 0.34, meaning that an audio warning resulted in a shock 34% of the time. We see that on July 7 there was something causing a spike in electrical shocks, but over all the shock to warning ratio quickly decreases, hovering between 0 and 0.05 for the rest of the experiment. This suggests that cattle learn to avoid electrical shocks as the trial progresses.

Figure 7. A comparison of the numbers of audio warnings and shocks as they change over time. The black dotted lines represent the grazing period intervals, the blue line represents the number of audio warnings that occurred each day, and the pink line represents the number of shocks that were administered each day. The Y axis was square root transformed in order to improve the visual interpretation of data.

Figure 8. The ratio of electrical shocks to audio warnings over time. The black dotted lines represent the grazing period intervals and the red line represents the number of electrical shocks divided by the number of audio warnings, for each day. This shock:warning ratio represents the frequency of an audio warning resulting in a shock.

After a birds eye view of the data, it is prudent to take a closer look at the numbers. Using a linear regression model in R I determined that audio warnings per day had a moderate increase over time (r=0.267) but a 95% confidence interval of r=0.0326 to 0.474 (p=0.0265) - this is not very precise (Figure 9)! But the weakness of the correlation between audio warnings and time does not matter as much as the relationship between electrical shocks and time. Electrical shocks per day had a strong decrease over time (r=-0.498) and a 95% confidence interval of r=-0.657 to -0.296 (p=1.344e-5), which is more precise and statistically significant (Figure 11).

From these numbers I conclude that time may have a weak effect on increasing the number of audio warnings that cattle experience, but that it strongly correlates to a decrease in electric shocks. In other words, cattle do not seem to be avoiding the audio warnings but quickly learn to avoid the electric shocks. I propose that they are using the audio warnings, and the learned expectation and a shock will follow a warning, to "see" where the virtual boundaries are and navigate their environment. This is good news for cattle producers who may want to implement virtual fencing on their farm - the customization and real-time grazing management is a key component of the promise that virtual fencing technology represents. It is crucial to rotational grazing that farmers be able to change pastures and trust that their cattle will be able to adjust their behaviour.

Figure 9. Linear regression illustrating a weak positive relationship between audio warnings per day and time.

Figure 11. Linear regression illustrating a strong negative relationship between electrical shocks per day and time.

Figure 10. Histogram of the residuals of the regression in Figure 9, showing an approximately normal distribution.

Figure 12. Histogram of the residuals of the regression in Figure 11, showing an abnormal distribution.

2) Grazing phase directly impacts the number of audio warnings animals receive, but does not effect electrical shocks.

Again we start with a bird's eye view of the data. We can see how the amount of forage supply available in the pasture typically starts out higher at the beginning of the rotation, and it falls back down towards the forage demand line by the end of the grazing period (Figure 13). I used the ratio of forage supply:forage demand to calculate the grazing pressure; as forage in the pasture is eaten supply goes down while demand remains the same, which pushes the supply:demand ratio closer to zero and means that grazing pressure increases.

If we compare the visual trend in audio warnings (Figure 7, placed below for convenience) with the visual trend in forage supply (Figure 13) we can see that as forage supply decreases across a grazing period, the number of audio warnings tends to increase.

Figure 13. A comparison of forage supply and forage demand over time. The black dotted lines represent the grazing period intervals, or the period of time for which the forage supply available in a specific pasture must sustain the forage demand of 51 cattle for that interval of time. The blue line represents the amount of forage available in the pasture (error bars indicate standard error), and the red line represents the forage demand of the cattle herd for that grazing period. Error bars indicate standard error.

To further investigate the relationship between warnings, shocks, and grazing pressure I used another linear regression model in R to characterize the relationship. We see that there is a moderate negative relationship between audio warnings per day and the forage supply:demand ratio (r=-0.343, p=3.967e-3) meaning that when there is more forage supply present in the pastures, the ratio of supply:demand is higher and we see fewer audio warnings than days when supply (and therefore the supply:demand ratio) is decreased (Figure 14). Conversely, when a linear regression was used to assess the relationship between shocks and grazing pressure, no statistically significant relationship was seen (Figure 16).

To investigate this relationship further I conducted two ANOVA analyses in R, comparing the number of audio warnings and electric shocks to three "treatments": the beginning, middle, and end of the grazing periods (which represent low, medium, and high grazing pressures respectively). Based on the ANOVA results I conclude that the variation of the audio warnings per day between the three treatments were statistically different (F = 26.6), with a high degree of confidence (p=7.31e-7), and that the variation of electrical shocks per day between the three treatments was far less significant (F=4.28, p=0.040).

These results are very interesting, and provide strong evidence that cattle truly learn to avoid electric shocks by responding to the audio warnings. It makes logical sense that as forage supply diminishes the cattle interact with the virtual boundary more often, as they seek out fresh food. However according to the linear regression models and supported by the ANOVA analyses, the increase in warnings is not accompanied by a significant increase in shocks, meaning that cattle still respond appropriately to the audio warnings. Additionally, it is important to understand the way in which external stressors (like grazing pressure) affect the virtual fencing system. In this case we can say that for animals experiencing a supply:demand ratio from approximately 1 to 4, there is a moderate affect on audio warning frequency but not effect on electrical shocks. More research is needed to investigate the effect of more intense, and a wider variety, of stressors.

Figure 14. Linear regression comparing audio warnings per day to grazing pressure using the forage supply:forage demand ratio. When forage supply decreases and the ratio of supply to demand is low (closer to 1) this means the grazing pressure is high; when the ratio of supply to demand is higher (closer to 4) this means the grazing pressure is low.

Figure 16. Linear regression comparing electric shocks per day to grazing pressure using the forage supply:forage demand ratio. When forage supply decreases and the ratio of supply to demand is low (closer to 1) this means the grazing pressure is high; when the ratio of supply to demand is higher (closer to 4) this means the grazing pressure is low.

Figure 15. Histogram of the residuals of the regression in Figure 14, showing an approximately normal distribution.

Figure 17. Histogram of the residuals of the regression in Figure 16, showing an approximately normal distribution.

3) Virtual fencing can be used to manage a rotational grazing system in Alberta.

There are multiple ways to define a successful management system for rotational grazing and for the purposes of this trial I will give success three parameters: 1) do the cattle learn to use the virtual fence system, 2) is there an animal welfare cost, and 3) is the technology practically feasible in the context of Alberta.

I believe I have answered the first question and sufficiently proven that cattle do learn to use the virtual fencing system. During this trial we monitored animal welfare in two ways. The first was through weighing the cattle every two weeks. We saw no significant changes in weight that would indicate illness or injury, and a desirable gradual weight gain throughout the summer (Figure 18). I also monitored the condition of the animals' necks where the collars were placed, and saw no evidence of rubbing or chafing throughout the course of the trial, therefore I conclude that my trial had no significant effect on animal welfare.

Finally, we must consider whether this technology is practical to use in Alberta. One of the main issues that occurred during this trial was unreliable cell phone service. While the collars themselves are fully functional once the virtual pasture is downloaded and do not need cell service, network connection is necessary for the collars to communicate with the mobile app, through which the collars are managed.

Ultimately this trial has revealed that virtual fencing has the potential to be applied to cattle farms in Alberta, but there is more work that needs to be done regarding cell network connection.

Figure 18. The average weights (kg) and standard errors of the herd of cattle before the experiment started, every two weeks, and then at the conclusion of the trial.

Conclusion

The overall goal of this trial was to investigate whether the NoFence virtual fencing system works in Alberta, whether it may be a viable tool that cattle producers may choose to adopt, and to discover factors that might affect the success of the system. To meet this goal I asked three questions: do cattle learn to avoid electrical shocks by responding correctly to audio warnings; does grazing pressure affect the way cattle interact with the virtual fence; and is it practical or possible to use a virtual fencing system in Alberta. I discovered that cattle do learn to avoid electric shocks, and I proposed that they may even use the audio warnings as a form of echolocation to "see" and navigate the virtual boundaries. I also concluded that increased grazing pressure can cause an increase in audio warnings, but has no significant affect on electrical shocks. Finally, it is my opinion that virtual fencing technology is at a point where it can be effectively used in Alberta but that there are many improvements that will increase the efficiency of the system.

The adoption of this technology has the potential to revolutionize the way that farmers raise cattle. In conventional farming farmers must invest time and money into building fences, must spend hours on maintaining and fixing those fences, and put a lot of labour into manually checking cattle and moving them to different areas, if needed. This cost is passed onto the consumer through beef prices, and it also inhibits farmers from making management choices (like implementing rotational grazing) that would ultimately benefit the customer and the environment. Virtual fencing technology is an exciting possible solution to many cattle farmers' woes and the technology is, it seems, at your fingertips.