Defining a distance threshold for utilising positioning data from a real-time location system to estimate affiliative dam-calf interactions in a freestall cow-calf contact system

Behavioural data is often collected using direct observations, requiring dedicated labour commitments from observers. Technology provides opportunities to more efficiently monitor behaviour and, consequently, welfare, but external validation of technology-assisted solutions is essential. This project investigated whether filtering positioning data from a real-time location system (RTLS) by distance could identify active affiliative behaviours (i.e., allogrooming and near-udder events) within dam-calf pairs with the purpose of more efficiently collecting behavioural data. Pairs (n = 10) were fitted with ultrawide-band RTLS ear tags in a cow-calf contact system with full, unrestricted contact. Active affiliative behaviours were observed from video footage on 4 consecutive days from 06:00–12:00 h. Positioning data for these observation periods were interpolated to calculate Euclidean distances (sensor-to-sensor) within dam-calf pairs on a per-second basis. The data was filtered for distance thresholds of 0.5–3 m, with 0.25 m increments, and aggregated into close contact bouts when the given distance was exceeded for ≥60 s. For each distance, behavioural observations and RTLS close contacts were compared per second, and a 2×2 contingency table was used to calculate performance metrics. The results of the study demonstrated that positioning data alone could not isolate active affiliative behaviours. However, in this study, 2 m was the optimal distance to capture intervals where active affiliative behaviours were occurring, with an average sensitivity and specificity of 99.5 ± 0.85% (SD) and 86.6 ± 7.90%, respectively, across pairs. The positive predictive value (PPV) was 24.8 ± 13.47%. Although greater distances (2.25–3 m) were able to identify instances of active affiliative behaviours with average sensitivities of at least 99.9%, the specificity and PPV decreased as distance increased. Meanwhile, the average PPV at distances between 0.5–1.75 m only improved marginally (33.6–26.4%) while sensitivity was compromised (23.2–94.5%). Post-hoc video observations revealed that outside of active affiliative behaviours, time in close contact (i.e., a distance threshold of 2 m) was primarily spent with the calf resting and dam standing or resting nearby. These results indicate that RTLS positioning data may be useful in exploring socio-spatial relationships between cows and calves, as this proximity detection approach holds value as a screening tool for capturing dyadic associations encompassing affiliative behaviours. However, without refinement of the method, behaviour specification will require direct observation or complimentary use of other technologies.