Near Real-Time Calving Detection in Grazing Cows Using GNSS-Derived Behavioral Anomalies

Timely calving detection is important but difficult in grazing systems because cows may calve in remote areas. This study developed and evaluated a near real-time calving alert system based exclusively on behavioral anomalies calculated from Global Navigation Satellite System (GNSS) collar data. Data were collected from 149 cows across three grazing farms in Spain and Australia, including 76 calving events. Individual, social, and herd-relative behavioral indicators were calculated every 30 min using 24 h rolling windows. Anomalies were defined as directional deviations from recent behavior using delta-self and Z-score transformations and were evaluated either at each 30 min timestamp or accumulated over rolling 3 h windows. Detection performance was assessed within the 20 days before calving, using a 48 h calving window as the reference period. Increasing the number of anomalies required to trigger a calving alert improved precision but reduced recall, while 3 h anomaly accumulation improved the stability of the alert system. Intermediate anomaly thresholds maximized F1-score, reaching 0.41 with individual behavioral indicators and 0.58 with the complete indicator set, irrespective of whether anomalies were evaluated at 30 min intervals or accumulated over 3 h. Alerts and indicator-specific anomalies were concentrated during the final two days before calving, mainly reflecting increased spatial separation and reduced movement relative to the group. These results support GNSS-based anomaly accumulation as a transparent and flexible approach for calving detection under grazing conditions.