Development of validated cognitive and behavioural indicators of welfare in pigs towards a predictive early warning system for poor welfare.

统计学、物理学、工程学和心理学使用了广泛的方法，这些方法可以被改编为基于动物的福利措施，只要它们能够在动物模型系统中得到验证。然后，可以使用经过验证的指标来开发预测性预警系统，以便能够提前发现和减轻潜在的福利问题。先前的研究表明，猪是此类项目的理想模型系统，当提供丰富的营养物质时，显示出显著的福利改善，皮肤损伤的程度可靠地表明福利状态。在这项研究中，将使用物理和环境因素来创建不同福利状态的猪的类别，以开发和验证基于统计学、物理学、工程学和心理学方法的新福利指标。大白×长白猪将被饲养在两种环境条件之一中，每组20头：(1)深秸秆围栏，(2)部分板条地板。共观察800头猪(10批次，4组，每组20头)，为期6周。在整个实验过程中，闭路电视摄像机将每天记录12个小时，每周一次。每只猪都会被单独标记，这样它们就可以从相机中识别出来。在最初的4周内，每头猪都会被评估，并根据受伤程度(身体和尾巴分别受伤)给出一个“福利”评分。在每个猪圈中，平均身体伤害百分比得分最高的4头猪将被列为“低福利”组，而平均百分比得分最低的4头猪将被列为“高福利”组。这将导致四个群体：(A)在深层稻草上获得高福利；(B)在部分板条地板上获得高福利；(C)在深层稻草上获得低福利；(D)在部分板条地板上获得低福利。在每次试验开始之前，将从每个围栏中随机选择5头猪，并在整个试验过程中使用3种认知行为测试进行训练和测试：认知偏差(确定个体的乐观或悲观程度)、功能记忆(评估个体对模式的跟踪能力)和间隔时间(确定不同个体感知的时间流逝)。认知偏差已被证明受到动物福利状态的影响。功能记忆和间隔计时尚未用于动物福利研究；本项目的目的是验证这些方法作为评估福利对认知加工的影响的新的有用工具。此外，在试验结束时，将使用三种认知行为测试来测试来自每个围栏的损伤百分比最高和最低的4头猪。我们还将记录笔内和所选个人在笔内的声音通信。从记录的视频片段中，将对每一只焦点猪进行追溯跟踪，并记录它们的行为，以确定在试验过程中个别活动预算的差异。此外，还将使用一些工具来调查每支笔内的社会群体动态。运动的分形分析和半隐马尔可夫链分析将识别重复的行为序列，社会网络分析将确定每只猪在其围栏网络中的相对位置和群体内的总体连接水平，将评估聚类水平以确定在给定可用空间的情况下，围栏中的个体是否比预期更拥挤，最后我们将评估每个围栏内同步行为的水平。我们将分析每一项指标，以寻找4种所述群体类型之间的差异。最后，将根据整个数据集开发一个预测性统计工具，作为福利问题的早期预警系统。预测模型的目的将是预测哪些围栏和其中的哪些人可能在稍后阶段出现福利问题，从而允许制定缓解策略。

Statistics, physics, engineering and psychology utilise a wide-range of methods that can be adapted for use as animal-based welfare measures if only they could be validated in a model animal system. Validated indicators could then be used to develop a predictive early warning system so that potential welfare problems can be detected and mitigated in advance. Previous research has shown pigs to be an ideal model system for such a project, showing significant welfare improvements when provided with enrichment, and extent of skin injuries sustained reliably indicating welfare state. In this study, categories of pigs of different welfare status will be created using both physical and environmental factors to develop and validate novel welfare indicators, based methods used in statistics, physics, engineering and psychology. Large White x Landrace pigs will be kept in groups of 20 in one of two environmental conditions: (1) deep straw pen, and (2) part-slatted floor. A total of 800 pigs (10 batches of 4 groups of 20) will be observed for 6 week periods. CCTV video cameras will record the pens for 12 hours per day once a week throughout the experiment. Every pig will be individually marked so that they are identifiable from the camera. At intervals over the first 4 weeks, each pig will be assessed and given a 'welfare' score based on measures of injury (to body and tail separately). The 4 pigs from each pen with the highest average % body injury score will be the 'Low welfare' group and the 4 with the lowest average % score will be the 'High welfare' group. This will give four groups: (a) High welfare on deep straw; (b) High welfare on part-slatted floor; (c) Low welfare on deep straw; (d) Low welfare on part-slatted floor. Before each trial starts, 5 pigs will be selected at random from each pen and will be trained and tested throughout the trial using 3 cognitive behaviour tests: cognitive bias (to determine the individual's level of optimism or pessimism), functional memory (to assess how well individuals can keep track of a pattern) and interval timing (to determine the perceived passage of time by different individuals). Cognitive bias has been shown to be affected by the welfare state of the animal. Functional memory and interval timing have not been utilised in animal welfare studies; this project will aim to validate these methods as new useful tools to assessing the impact of welfare on cognitive processing. In addition, the 4 pigs with the highest and lowest % injury from each pen will be tested using the three cognitive behaviour tests at the end of the trial. We will also record acoustic communication both within a pen and of selected individuals within a pen. From the recorded video footage, each of the focal pigs will be retrospectively tracked and their behaviours recorded for differences in individual activity budgets over the course of the trial. In addition, a number of tools will be used to investigate the social group dynamics within each pen. Fractal analysis of movement and semi-hidden Markov-chain analysis will identify repeating sequences of behaviour, social network analysis will determine the relative positions of each pig within its pen network and the overall level of connectedness within the group, levels of clustering will be assessed to determine whether individuals in a pen are crowding together more than expected given the space available to them, and finally we will assess the levels of synchronous behaviour within each pen. We will analyse each of indicator to look for differences between the 4 stated group types. Finally, a predictive statistical tool will be developed based on the entire data set, which can be used as an early warning system for welfare problems. The aim of the predictive model will be to predict which pens and which individuals within them, are likely to develop welfare problems at a later stage and hence permit mitigation strategies to be put in place.