Optimisation Of On-farm Technologies To Predict Health And Resilience In Dairy Calves

Currently there are no accurate digital tools with decision support to predict calf health and production. Our approach is highly novel as it uses cutting-edge data techniques to develop and use novel features from various calf behaviours (various activities, social networks, feeding, play) and physiology (temperature both core and eye) captured by technologies (automatic feeders, activity location sensors, bolus, thermal cameras) and on-farm data to predict health and production and welfare indicator as play. We will optimise the use of technologies by identifying which information is of value and by conducting a comparative evaluation of the technologies w.r.t their predictive accuracy.

Our approach is different and extends the use of technologies for the first-time to accurately measure and quantify dynamic indicators of resilience in 3 states (behavioural, physiological and production) in calves. Through implementation of a "Living Lab" (LL; first for dairy), a user-centric research methodology for prototyping, refining and validating IoT solutions, the results will inform decision support for farmers. It's timely as results allow optimal and novel use of current technologies and through our consortium involving multiple stakeholders, including commercial partners, we are best placed to exploit these outcomes.

Translation and applicability: The algorithms we will develop in the project will help farmers by providing early disease detection for calves, measures of positive welfare (play) for the herd and predicting production outcomes - these will be of value to both farmers and vets for calf management decisions. The outcome and knowledge of feature importance from different technologies in prediction and their comparative evaluation is of huge value to farmers, vets (for choice and adoption) and wider industry (for innovation). Routes to translation and impact will be via our consortium and hosting of LL workshops during the project lifetime with various stakeholders and through our extensive existing networks.

Using technologies to measure resilience has the added value in that it could promote their embedment in decision support and drive the uptake of technology on farms. This can help farmers and vets to identify animals that are vulnerable and predict how they are likely to respond to a future stressor and have a measure of herd resilience. Our results have applicability to other livestock sectors with digital tools.

Next steps: Our longer-term aim (5 yr) plan will be to further validate the findings from this study, link to lifetime resilience and improve our understanding of early-life conditions that support the development and expression of these markers of resilience in calves. To understand which management interventions enhance resilience and how these markers could be incorporated in breeding programmes. A comprehensive validated resilience index will support a paradigm shift and move the focus from mere disease management to a more holistic and dynamic view of animal health.

Studies to date utilising these tools have failed to use a wide range of behavioural and/or physiological features from monitored data; for example, social and play behaviour, animals coping style (consistent behaviour patterns1 (all generated from single sensor) have been linked to disease; as shown in our work temperature signal can be utilised to give temperature as well drinking. Furthermore, use of these technologies have so far narrowly focussed on disease detection and mostly utilising static features; dynamic features of continuous time series that technologies provide could predict resilience. Insights from use of complex dynamics systems theory in ecology, has shown that as system becomes less resilient, system variables show increasing delays in their recovery from internal or external perturbations. Thus, dynamic features such as (variance, temporal autocorrelation, cross -correlations) of high-resolution data can predict a system's resilience. It is highly novel application for the use of these tools. These data, combined with advanced machine learning algorithms, can be used to automatically monitor health and resilience of calves and provide decision support tools to farmers to predict prevent calf diseases and improve resilience through adaptation of management strategies.

Using unique dataset detailed time series data on animal behaviours, physiology and production, we will generate algorithms that will predict health, production and resilience.
Our hypotheses for this 12-month research are:
H1) Behavioural and physiological features captured from technologies can be optimised to predict health and production in calves
H2) Calves' varying response to natural perturbations and key commercial stressors, measured via dynamic features in their time series of behaviour, physiology and production, can be used to quantify resilience
H3) Dynamic features of resilience in H2 will be correlated between states and to calves cumulative health outcomes