Using animal-mounted sensor technology and machine learning to predict time-to-calving in beef and dairy cows.

Worldwide, there is a trend towards increased herd sizes, and the animal-to-stockman ratio is increasing within the beef and dairy sectors; thus, the time available to monitoring individual animals is reducing. The behaviour of cows is known to change in the hours prior to parturition, for example, less time ruminating and eating and increased activity level and tail-raise events. These behaviours can be monitored non-invasively using animal-mounted sensors. Thus, behavioural traits are ideal variables for the prediction of calving. This study explored the potential of two sensor technologies for their capabilities in predicting when calf expulsion should be expected. Two trials were conducted at separate locations: (i) beef cows (n = 144) and (ii) dairy cows (n = 110). Two sensors were deployed on each cow: (1) Afimilk Silent Herdsman (SHM) collars monitoring time spent ruminating (RUM), eating (EAT) and the relative activity level (ACT) of the cow, and (2) tail-mounted Axivity accelerometers to detect tail-raise events (TAIL). The exact time the calf was expelled from the cow was determined by viewing closed-circuit television camera footage. Machine learning random forest algorithms were developed to predict when calf expulsion should be expected using single-sensor variables and by integrating multiple-sensor data-streams. The performance of the models was tested using the Matthew's correlation coefficient (MCC), the area under the curve, and the sensitivity and specificity of predictions. The TAIL model was slightly better at predicting calving within a 5-h window for beef cows (MCC = 0.31) than for dairy cows (MCC = 0.29). The TAIL + RUM + EAT models were equally as good at predicting calving within a 5-h window for beef and dairy cows (MCC = 0.32 for both models). Combining data-streams from SHM and tail sensors did not substantially improve model performance over tail sensors alone; therefore, hour-by-hour algorithms for the prediction of time of calf expulsion were developed using tail sensor data. Optimal classification occurred at 2 h prior to calving for both beef (MCC = 0.29) and dairy cows (MCC = 0.25). This study showed that tail sensors alone are adequate for the prediction of parturition and that the optimal time for prediction is 2 h before expulsion of the calf.

Use of novel sensors combining local positioning and acceleration to measure feeding behavior differences associated with lameness in dairy cattle.

Time constraints for dairy farmers are an important factor contributing to the under-detection of lameness, resulting in delayed or missed treatment of lame cows within many commercial dairy herds. Hence, a need exists for flexible and affordable cow-based sensor systems capable of monitoring behaviors such as time spent feeding, which may be affected by the onset of lameness. In this study a novel neck-mounted mobile sensor system that combines local positioning and activity (acceleration) was tested and validated on a commercial UK dairy farm. Position and activity data were collected over 5 consecutive days for 19 high-yield dairy cows (10 lame, 9 nonlame) that formed a subset of a larger (120 cow) management group housed in a freestall barn. A decision tree algorithm that included sensor-recorded position and accelerometer data was developed to classify a cow as doing 1 of 3 categories of behavior: (1) feeding, (2) not feeding, and (3) out of pen for milking. For each classified behavior the mean number of bouts, the mean bout duration, and the mean total duration across all bouts was determined on a daily basis, and also separately for the time periods in between milking (morning = 0630-1300 h; afternoon = 1430-2100 h; night = 2230-0500 h). A comparative analysis of the classified cow behaviors was undertaken using a Welch t-test with Benjamini-Hochberg post-hoc correction under the null hypothesis of no differences in the number or duration of behavioral bouts between the 2 test groups of lame and nonlame cows. Analysis showed that mean total daily feeding duration was significantly lower for lame cows compared with non-lame cows. Behavior was also affected by time of day with significantly lower mean total duration of feeding and higher total duration of nonfeeding in the afternoons for lame cows compared with nonlame cows. The results demonstrate how sensors that measure both position and acceleration are capable of detecting differences in feeding behavior that may be associated with lameness. Such behavioral differences could be used in the development of predictive algorithms for the prompt detection of lameness as part of a commercially viable automated behavioral monitoring system.

Group navigation and the "many-wrongs principle" in models of animal movement.

Traditional studies of animal navigation over both long and short distances have usually considered the orientation ability of the individual only, without reference to the implications of group membership. However, recent work has suggested that being in a group can significantly improve the ability of an individual to align toward and reach a target direction or point, even when all group members have limited navigational ability and there are no leaders. This effect is known as the "many-wrongs principle" since the large number of individual navigational errors across the group are suppressed by interactions and group cohesion. In this paper, we simulate the many-wrongs principle using a simple individual-based model of movement based on a biased random walk that includes group interactions. We study the ability of the group as a whole to reach a target given different levels of individual navigation error, group size, interaction radius, and environmental turbulence. In scenarios with low levels of environmental turbulence, simulation results demonstrate a navigational benefit from group membership, particularly for small group sizes. In contrast, when movement takes place in a highly turbulent environment, simulation results suggest that the best strategy is to navigate as individuals rather than as a group.

Calculating spatial statistics for velocity jump processes with experimentally observed reorientation parameters.

Mathematical modelling of the directed movement of animals, microorganisms and cells is of great relevance in the fields of biology and medicine. Simple diffusive models of movement assume a random walk in the position, while more realistic models include the direction of movement by assuming a random walk in the velocity. These velocity jump processes, although more realistic, are much harder to analyse and an equation that describes the underlying spatial distribution only exists in one dimension. In this communication we set up a realistic reorientation model in two dimensions, where the mean turning angle is dependent on the previous direction of movement and bias is implicitly introduced in the probability distribution for the direction of movement. This model, and the associated reorientation parameters, is based on data from experiments on swimming microorganisms. Assuming a transport equation to describe the motion of a population of random walkers using a velocity jump process, together with this realistic reorientation model, we use a moment closure method to derive and solve a system of equations for the spatial statistics. These asymptotic equations are a very good match to simulated random walks for realistic parameter values.

Sampling rate effects on measurements of correlated and biased random walks.

When observing the two-dimensional movement of animals or microorganisms, it is usually necessary to impose a fixed sampling rate, so that observations are made at certain fixed intervals of time and the trajectory is split into a set of discrete steps. A sampling rate that is too small will result in information about the original path and correlation being lost. If random walk models are to be used to predict movement patterns or to estimate parameters to be used in continuum models, then it is essential to be able to quantify and understand the effect of the sampling rate imposed by the observer on real trajectories. We use a velocity jump process with a realistic reorientation model to simulate correlated and biased random walks and investigate the effect of sampling rate on the observed angular deviation, apparent speed and mean turning angle. We discuss a method of estimating the values of the reorientation parameters used in the original random walk from the rediscretized data that assumes a linear relation between sampling time step and the parameter values.