Animal-mounted sensor technology to predict ‘time to calving’ in beef and dairy cows

Application: The availability of early detection and alerts for parturition/dystocia will enable farmers to intervene in a timely manner to prevent the losses associated with dystocia, thus optimising the economic and production efficiency of their business. Introduction: In the UK, the average herd size and animal to stockman ratio is increasing within the beef and dairy sectors, thus the time devoted to monitoring individual animals is reducing. In order to optimise the production efficiency of the UK livestock sector, there is a requirement for the development and use of cost-effective animal monitoring solutions to inform on the health and productive status of individual animals. The non-invasive nature of behavioural observations and the availability of a number of sensors on the market, or near to market, designed to monitor different elements of cattle behaviour provides opportunities for translation of current behavioural and technology validation research into a multisensor platform for the prediction of calving onset and calving difficulties. Lying and standing behaviour, eating and rumination patterns, social behaviour and tail raising events are known to change during the 24 hours prior to calving. This study explored the potential of technologies on the market or near-to-market for related and other uses (e.g. detection of oestrous) for their capabilities in the detection of calving and dystocia. Material and methods: Two trials were conducted at separate locations: i) beef cattle (n = 144) at SRUC’s Beef and Sheep Research Centre, SRUC, UK and (ii) dairy cattle (n = 110) at a commercial dairy farm in Essex, UK, under the control of staff from Writtle University College. Three sensors were deployed on each cow: 1) Afimilk Silent Herdsman (SHM) collars monitoring time spent ruminating, eating and the relative activity level of the cow, 2) Tail mounted sensors to detect tail-raise events and 3) a real time locating system (Omnisense - OMS) to monitor cow location, orientation and activity level. Cows were monitored from between >30 days to one day prior to calving until 12 hours post calving (beef) or immediately post calving (dairy). CCTV cameras were used to determine the exact time of calving. Machine learning (Random Forest (RF)) algorithms were developed to detect the onset of calving at the earliest opportunity using single sensors variables and by integrating multiple sensor data-streams. Results: The RF models showed that SHM and tail sensors can predict time of calving for both beef and dairy cows for up to four hours pre-calving. When presented with a validation dataset, the tail raise RF for beef cows had a balanced accuracy of 90% one hour prior to calving, dropping to 62% six hours prior to calving (Figure 1a). Results were similar for dairy cows. Time spent ruminating was a better predictor of time prior to calving than time spent eating and combining these two variables (Figure 1b) resulted in only a slight improvement in the beef model performance 0-3 hours prior to calving (particularly at two hours prior where accuracy increased from 59 to 68%), but not the dairy. Combining data-streams from SHM and tail sensors did not substantially improve model performance over tail sensors alone for either beef or dairy. Cumulative deviation from mean acceleration of activity from OMS data indicated a decline in activity up to 10 days prior to calving. CUSUM analyses showed drops in activity greater than two standard deviations from the mean for dairy cows who had difficult calvings – this may prove to be a useful indicator of dystocia. Conclusion: Tail sensors were effective at predicting calving up to five hours prior to calving. Afimilk collars were less or as good at predicting time to calving. There was no significant improvement to model performance in combining datastreams from these two sensors. Activity level, as measured by Omnisense, may provide a useful early warning indicator of dystocia.