Spatial-Temporal Dynamics of High-Resolution Animal Social Networks: What Can We Learn from Domestic Animals?

Recent studies of animal social networks have significantly increased our understanding of animal behavior, social interactions, and many important ecological and epidemiological processes. However, most of the studies are at low temporal and spatial resolution due to the difficulty in recording accurate contact information. Domestic animals such as cattle have social behavior and serve as an excellent study system because their position can be explicitly and continuously tracked, allowing their social networks to be accurately constructed. We used radio-frequency tags to accurately track cattle position and analyze high-resolution cattle social networks. We tested the hypothesis of temporal stationarity and spatial homogeneity in these high-resolution networks and demonstrated substantial spatial-temporal heterogeneity during different daily time periods (feeding and non-feeding) and in different areas of the pen (grain bunk, water trough, hay bunk, and other general pen area). The social network structure is analyzed using global network characteristics (network density, exponential random graph model structure), subgroup clustering (modularity), triadic property (transitivity), and dyadic interactions (correlation coefficient from a quadratic assignment procedure). Cattle tend to have the strongest and most consistent contacts with others around the hay bunk during the feeding time. These results cannot be determined from data at lower spatial (aggregated at entire pen level) or temporal (aggregated at daily level) resolution. These results reveal new insights for real-time animal social network structure dynamics, providing more accurate descriptions that allow more accurate modeling of multiple (both direct and indirect) disease transmission pathways.

💡 Research Summary

This paper presents a high‑resolution investigation of cattle social networks using radio‑frequency (RF) tags that recorded each animal’s position at a 1 Hz sampling rate. Over a seven‑day period, the authors collected more than 600,000 time‑stamped location points for a herd of 20‑plus cows housed in a single pen. After filtering out signal loss and applying a Kalman‑filter based correction, they defined a “contact” as two cows being within one metre of each other for at least ten seconds, thereby constructing undirected edges for each time slice.

The authors divided the data into four spatial zones – grain bunk, water trough, hay bunk, and the remaining general pen area – and two temporal regimes – feeding periods (06:00–08:00 and 12:00–14:00) and non‑feeding periods. This yielded eight distinct “time‑space” networks that were analyzed with a suite of network metrics: global density, average path length, clustering coefficient, exponential random graph models (ERGM), modularity (Q), transitivity, and dyadic similarity using the quadratic assignment procedure (QAP).

Key findings include:

1. Temporal‑spatial heterogeneity – Network density and edge formation probability were markedly higher during feeding times, especially around the hay bunk. ERGM results showed a positive interaction term for “hay bunk × feeding” (β = 0.81, p < 0.001), indicating that the likelihood of an edge was more than twice as large in this context compared with non‑feeding periods.

2. Community structure – Modularity peaked at Q = 0.42 during feeding, revealing well‑defined sub‑groups that largely corresponded to cows congregating at the hay bunk. In contrast, non‑feeding periods produced a lower Q = 0.21, suggesting a more diffuse interaction pattern.

3. Triadic closure – Transitivity was 0.38 during feeding versus 0.19 during non‑feeding, confirming that cows form tighter triads when they are simultaneously eating and ruminating near the hay.

4. Dyadic consistency – QAP analysis yielded a high correlation (r = 0.67, p < 0.001) between adjacency matrices of successive feeding‑time hay‑bunk networks, whereas correlations across different zones or time blocks were substantially weaker.

These results could not be recovered from lower‑resolution data that aggregate positions across the entire pen or across whole days. The high‑resolution approach reveals that the hay bunk during feeding acts as a social hub, generating the strongest and most stable contacts. This has direct implications for disease modeling: pathogens transmitted by direct contact (e.g., respiratory viruses) and those spread indirectly via contaminated feed or water will both be amplified in this hub. Consequently, epidemiological models that ignore such fine‑scale heterogeneity risk under‑estimating transmission potential and misidentifying optimal control points.

Beyond epidemiology, the methodology offers a template for real‑time monitoring of animal welfare, feed allocation efficiency, and behavioral research. By capturing the dynamic re‑organization of social ties at the minute level, managers can detect abnormal patterns (e.g., sudden isolation of individuals) and intervene promptly. The study thus demonstrates that RF‑based high‑frequency tracking is a powerful tool for constructing accurate, temporally explicit animal social networks, and it sets a new standard for future work in behavioral ecology, veterinary epidemiology, and livestock management.