Assessment of Dairy Cow Welfare through Predictive Modelling of Individual and Social Behaviour

Dairy cow welfare is increasingly a subject of public concern. A recent European report of leading scientists concluded that lameness and mastitis of cows were the most important factors in reducing the welfare of dairy cows due to the pain associated with these conditions. Unfortunately, the Farm Animal Welfare Council in the UK also reports that the dairy industry has made little progress in addressing these problems, mainly due to a reduction in profitability affecting investment and the lack of welfare surveillance systems available. A major challenge in improving the welfare of food production animals is in developing methods of automating the detection of such welfare problems. Such detection systems should be able to operate as early warning systems and detect the early signs of disease or illness within dairy herds and individual cows.

Thanks to new technological developments there are potential solutions. Until recently, it has not been logistically possible to monitor the complex behaviour associated with animals kept in large social groups, such as sheep, pigs or cows. However, novel local positioning wireless sensors such as those designed by our project partner, Omnisense, can be deployed over large networks of animals and give accurate positioning information for individuals over long periods of time. For the first time we will be able to record large quantities of data regarding the behaviour and social interactions in a whole herd of dairy cows.

Research studies have shown that diseases such as lameness in dairy cattle can affect general behaviour, such as how long cows spend lying down. Similarly, social interactions between individual animals, such as how much time they spend close to each other or how closely they synchronise their behaviour, have been suggested as possible measures of animal welfare. However, it is a non-trivial problem to determine and quantify changes in individual and social behaviour and subsequently to use such changes to predict the onset of disease.

In this project we will be using automated data collection techniques to record patterns of space use, movement, and social interactions within commercial dairy herds. In the first year, the behaviour of animals with lameness, mastitis or metabolic disease will be compared with healthy animals to determine differences in behaviour. In year two, a full dairy herd will be monitored for an extended period from calving to measure changes in their behaviour with the natural onset of disease in order to identify early changes that might be used to subsequently predict disease occurrence. In year three, the study will be repeated on three other farms to test whether such predictions are still relevant on different intensive dairy units.

The behavioural data will be analysed using cutting-edge mathematical and statistical techniques. Using information about the observed changes in both individual cow behaviour and herd social structure we will develop a predictive model for the onset of disease and other welfare changes within individual cows. This will lead to the development of an on-farm automated 'early warning' system for disease detection. Such a system would be invaluable for improving the welfare and productivity of dairy cows.

Using the techniques we develop for predicting the onset of disease we will also determine if it is possible to use behavioural changes to identify other important welfare changes in dairy cattle, in particular the onset of oestrus and the time of calving.

The aim of the project is to develop an automated system that uses behavioural observations to predict the onset of diseases such as lameness and mastitis within a commercial dairy herd. The project is highly timely: the latest automated wireless sensors provided by our project partner will allow us to collect an unprecedented amount of data on the behaviour of large herds of cows (140+ individuals) over months at a time.

With an automated system the true behavioural states of each individual cow are not known but instead an observed output state is determined from the positional and activity data collected automatically using the wireless sensors. This is an ideal scenario to use Hidden Markov Models (HMM). HMMs are a flexible statistical tool that allow one to model and analyse sequences of behaviour and have not previously been used in this context. In particular we will use HMMs to identify atypical cow behaviour and we hypothesise that this will be a useful predictor of disease state.

Social Network Analysis (SNA) is a powerful framework which provides metrics that quantify social structure at different levels of organisation. These metrics can be used to test hypotheses regarding the relationship between an individual's social network position and its attributes such as disease status. A highly novel aspect of this project will be to use SNA of the behavioural data to make predictions about changes in welfare of individual cows within the herd.

Using computer learning techniques such as artificial neural networks we will combine the results of the HMM and SNA analysis together with other indicators of welfare to predict disease onset in individual cows. This will lead to the development of an on-farm automated 'early warning' system for disease detection. We will also apply the methodology to try to predict two other important aspects of cow welfare: the onset of oestrus and the time of calving.

Veterinary and animal science research has had a longstanding partnership between academia and industry. The main beneficiaries for this research are the 15,000 dairy farms in the UK with their 1.8 million cows. Disease is a common occurrence in dairy cows with 15-39 per cent of cows suffering from lameness (with up to 79 per cent of cows on a single farm) with an individual case costing an estimated £240. A direct end-user for the research findings will be our industry collaborator DairyCo who represent the UK dairy industry and will disseminate findings through their 'Cow Signals' programme, a national knowledge transfer initiative that directly engages and informs farmers about the importance of animal behaviour in welfare monitoring. We will also engage directly with producers through our collaboration with Milk Link who are owned by over 1500 UK dairy farmers. Milk Link employs over 1200 people at eight processing and packaging facilities in the UK and has an annual turnover of £586 million (2010/11).

This research will also have significant international impact, as exemplified by the fact that lameness and mastitis have been highlighted as the most important factors in a survey on the consequences of poor welfare in cows by the European Food Safety Authority. The results of our research will be relevant to all high producing nations maintaining intensive dairy units.

Better systems for understanding the link between behaviour and welfare have the potential to have significant impact on other sectors of the food production industry that rear animals in social groups (e.g. beef cattle, sheep and pigs). Our research will also be of significant interest to those seeking to have more informed welfare standards including the leading supermarket chains and animal welfare charities. The general public are increasingly interested in the welfare of farm animals as evidenced by the popularity of food produced through Farm Assurance schemes. The research will have a significant impact on our project partner, Omnisense Limited, enabling them to develop and test novel technology that could be used for a commercial 'early warning system' for on-farm disease detection.