The study was conducted from 29 September 2023 to 26 November 2023 (58 days) at Boat Field, High Woods Country Park, Colchester, Essex (postcode CO4 5WF; central Global Positioning System (GPS) coordinates of 51.902883, 0.910765). The field is a managed pasture, measuring approximately 7.2 hectares and 350 m x 300 m at its greatest extent (longitude x latitude), and consists of an upper, relatively flat section with short, hay-cut grass (Figures 1B,D), and a lower, sloped region with scrubby vegetation and small fenced areas of trees (Figures 1A,C; Supplementary material S1 provides further details, including plant species present). These two distinct areas were separated by a clear ridgeline (marked as the red contour line in Figure 1A), but livestock could move freely around the full field. Key features at the study site included paths, a stream outside the field, and a water trough in the north-west corner (Figure 1). No public access was permitted to the field during the study period.

Ethical approval (reference number ETH2324-0100) was provided by the University of Essex ethics committee. Using a herd of Red Poll cattle (Bos taurus), the study site was managed for conservation purposes on behalf of Colchester City Council by Legacy Grazing (authors RB and LB). On 29 September, Nofence® tracking collars were fitted to three cows (Figure 1D; cattle IDs 294,322, 294,361, and 294,364). These cows were all pregnant and suckling calves during the study, and formed part of a larger herd that varied in size over time (see Supplementary material S1 for further details).

The Nofence® tracking collar units include a GPS sensor, a dynamometer and an accelerometer to measure activity, and a temperature sensor (39, 42). End-users can set a virtual fence to constrain livestock via the collar units using a smartphone application (38, 39). In this study, the virtual fence was set approximately 60 m outside the boundary of the field, which was enclosed by a robust wooden fence, so was effectively unused given our aim was to use the collars to map space-use and activity, rather than specifically explore the impact of the virtual fence on behavior. Default collar settings were used for recording frequency as in Aaser et al. (39), with GPS locations recorded every 15 min when animals were active. To conserve battery, a lower update rate of approximately one recording every 1–2 h is used when animals are inactive, such as at night. Activity was measured using an in-built dynamometer, that counted movements along the heave axis over a defined (commercially sensitive) threshold, cumulatively totaled every 30 min and termed an ‘activity index’ by Aaser et al. (39). The collar unit also recorded the localized instantaneous temperature every 30 min. This dataset is available in the University of Essex Data Repository (44).

The three collared cows were observed in-person by two observers over 8 days in October and November 2023. Each cow was continuously observed for an average of 936 min (minimum = 886 min, maximum = 1,035 min). Behavior was recorded as grazing when the cow was observed with their head lowered, actively consuming vegetation and not ruminating (Supplementary material S2); minor pauses in grazing (e.g., raising head and chewing for a few seconds) were noted but the classification was only changed to non-grazing (or vice-versa) when a clear and sustained change in behavior was observed. This observational data is available in the University of Essex Data Repository (45). All observation sessions were filmed using a video camera to enable validation of behavioral classification. For subsequent analysis, each minute throughout the study period was classified as belonging to a grazing or non-grazing bout, and the number of minutes (proportion of time) spent grazing (or non-grazing) in each 30-min observation period was determined.

To assess the correlation between the Nofence® activity index and observed grazing intensity, a linear model was used to compare the proportion of time each cow was observed to be grazing over each 30-min period to the sensor-recorded activity index from the same period. Model assumptions (linearity, normality, homoscedasticity, and independence) were checked and met. Supplementary material S2 provides further details of the observations and shows no significant difference between observers in the proportion of time observed grazing and activity index (e = −6.13, p = 0.15).

Two methods were used to generate utility distribution (UD; density) maps of space-use (total time spent in given locations), cumulative activity (total sum of activity index), and average activity (mean activity index). The cell count method is computationally efficient but limited by data frequency, while the Brownian Bridge Movement Model (BBMM) estimates movement between recorded locations but it is more computationally intensive and requires model assumptions. For each spatial distribution generated, core and full range sizes were determined by identifying the highest density cells cumulatively adding to 50% or 95% respectively, following previous animal home range analyses [e.g. (28, 46, 47)]. All analyses were programmed using R Statistical Software v4.2.3 with R Studio Cherry Blossom (48, 49), employing the ‘ggplot2’ and ‘ggmap’ (function ggmap()) packages (50, 51) for plotting distribution matrices over satellite imagery.

In the cell count method, a virtual grid (15 m x 15 m = 225 m² cells) was overlayed over the field layout (28). At each time-step, GPS data were used to assign each individual cow to a given cell and the respective cell count value was increased to reflect time spent in the area (based on time duration between 15-min location recordings), cumulative activity (recorded every 30 min and split equally between consecutive locations), average activity or average sensor temperature. A BBMM (27) was used to estimate movement paths between recorded locations and hence generate space-use and activity maps within the same 15 m x 15 m grid, although this required a novel modification to the standard BBMM algorithm in order to map activity across space. Where animals are relatively stationary, the cell count method and the BBMM yield similar results, but the BBMM may give better estimates when movement is more dynamic. Supplementary material S3 provides full details of these spatial mapping methods and their implementation.

Sensor recorded temperatures, influenced by the close proximity of the collar unit to the animal (see Figure 1D; Supplementary material S2), will not directly measure ambient conditions but may nevertheless reflect underlying trends. To demonstrate the relation between hourly sensor-recorded temperatures and hourly local weather station temperature readings (within a mile of the study site, 52), a Spearman’s Correlation test was conducted (non-normal data distribution, Shapiro-Wilks = 0.99, p < 0.001).

The mean observed time for a grazing bout was 20 min 42 s and for a non-grazing bout was 10 min 42 s. A strong significant positive relationship exists between the relative proportion of time cattle were observed grazing over a 30-min period and the activity index recorded by the Nofence® sensor (R² = 0.81, p < 0.001; Figure 2A). Cattle were primarily observed during daytime grazing, resulting in more exclusively grazing periods (n = 31 periods) than exclusively non-grazing periods (n = 5 periods), with n = 36 mixed periods where both grazing and non-grazing behaviors occurred (Figure 2B). During exclusively grazing periods, the activity index is clearly higher (100% grazing over 30 min, median activity index = 4,270) compared to periods of exclusively non-grazing (0% grazing, median activity index = 661) (Figure 2B). Non-zero activity index values were recorded during exclusively non-grazing behavior (Figure 2B), likely due to other head movements such as drinking, scratching, ruminating, or interactions between cows. Results remained consistent when considering data from each observer independently (see Supplementary material S2). For further analysis, and consistent with Aaser et al. (39), we classify activity as ‘low intensity’ when the activity index <2,500 and ‘high intensity’ when the activity index >2,500. This threshold is a heuristic estimate that approximately aligns with the observed proportion of time spent grazing being 50% (activity index = 2,642; Figure 2A) and the midpoint (activity index = 2,233; Figure 2B) between the maximum value for exclusively non-grazing periods (activity index = 1,848) and the minimum value for exclusively grazing periods (activity index = 2,617).

High intensity activity periods (activity index ≥2,500) account for 45% of the sensor recorded activity index values (Figures 2C,D), with a gradual increase in their daily occurrence from October to December (Figure 2C). Across the three cattle, average activity during high intensity periods increased over time (Spearman’s Correlation Coefficient [hereafter ρ] = 0.18, p < 0.001; values above black line in Figure 2C), but average activity during low intensity periods (activity index <2,500) decreased over time (ρ = −0.08, p < 0.001; values below black line in Figure 2C). The daily count of low intensity periods increased over time (Figure 2D; ρ = 0.37, p < 0.001), despite the increase in average activity during high intensity periods over the same period (Figure 2C).

A consistent diurnal activity pattern is shown across all cows, with the mean hourly activity index increasing from low intensity activity overnight (mean activity index <2,500 from 18:00 to 07:00) to high intensity activity in the late morning and early afternoon, with a slight dip around midday, peaking at 10:00 and 16:00 (mean activity index = 3,963 and 4,152 respectively), followed by a decrease into the evening and night (Figure 2E). Mean hourly activity index values are similar for each individual cow, although some differences are apparent (e.g., cattle ID 294361 consistently showed lower mean activity during the day compared to the other cows; Figure 2E). The mean hourly activity index also increased over the study period (by two-week periods; Figure 2F), aligning with the increase in high intensity grazing in Figure 2C.

The cell count method and the BBMM produced qualitatively similar results for space-use across the study (Figures 3A,B), with the BBMM generating a smoother map due to the model estimation of movement in-between actual observed locations. Both models consistently identified core range areas near the water trough (NW corner of upper field), the SW corner of the upper field, along the ridgeline and in the lower field (Figures 3A,B).

Comparing Figures 3A,C for the cell count method, and Figures 3B,D for BBMM, highlights that space-use is highly correlated with cumulative activity (BC = 0.96 for cell count method; BC = 0.99 for BBMM). Nevertheless, there are some notable differences in spatial distribution such as the upper field where the core range for cumulative activity is larger (Figure 3C, core range in upper field = 75 cells for cell count method; Figure 3B, core range in upper field = 54 cells for BBMM) than for space-use (Figure 3A, core range in upper field = 33 cells for cell count method; Figure 3B, core range in upper field = 47 cells for BBMM). The (mean) average activity distributions show larger core range in the upper field (Figure 3E, core range = 79 cells in upper field and 25 cells in lower field for cell count method; Figure 3F, core range = 99 cells in upper field and 29 cells in lower field for BBMM), indicating that there may be more spatial uniformity in average activity across the field when compared to cumulative activity (Figures 3C,D), although the distribution of average activity intensity does not directly overlap with cumulative activity (BC = 0.93 for the cell count method, BC = 0.93 for BBMM).

Supplementary material S3 includes more detailed analysis of spatiotemporal variations in space-use and activity. In particular, Supplementary Figures S3.3–S3.6, respectively, highlight how the cattle shifted their space-use and activity from the lower field at night to the upper field during the day, with a further consistent shift from the upper to lower field as the study progressed.

As expected for this temperate location, mean hourly temperature exhibits a consistent diurnal pattern, peaking at 18.33°C (13.00–13:59) and reaching a minimum of 9.58°C (03:00–03:59) over the study period, with similar patterns in all cow sensors (maximum difference of 1.79°C for 07:00–07:59; Figure 4A). This diurnal pattern is also clear in the spatial temperature distribution maps, showing higher temperatures in the morning and afternoon (Figures 4C,D) compared to the night and evening (Figures 4B,E). There is a significant strong correlation between the sensor recorded temperatures and local weather station temperature readings (ρ = 0.88, p < 0.001), although the sensor temperature measurements were consistently slightly higher, likely due to the proximity between the animal and the collar (Figure 4F). Although the relative variation in mean temperature across the field is small (coefficient of variation = 20.66%, assuming no spatial dependence between cells), the lowest recorded temperatures occur around the ridgeline (Figures 4B–E), possibly due to increased wind exposure which was generally from the SW during the study period. Over the study period, temperatures gradually decreased (Figure 4F), with two-week spatial temperature maps illustrating the first 2 weeks were notably warmer (mean temperature of 18.23°C) than later periods (mean temperatures of 12.55°C, 10.11°C and 9.57°C for F2 to F4 respectively; Figures 4F,G).