CLEARFARM Co-designed welfare monitoring platform for pig and dairy cattle

CLEARFARM Co-designed welfare monitoring platform for pig and dairy cattle

The acceptance of producers and consumers is essential for the adoption of technological solutions to improving animal welfare. Therefore, using the methodology of Design Thinking, ClearFarm gathered different profiles involved in the value chains of dairy cattle and pig products so that their needs and sensitivities were fully represented to ensure that the newly developed technological solution fit into their requirements.

In this sense, a sequence of events, both focus groups and co-creation workshops, were organized with consumers, producers, retailers, regulators, academics, animal welfare organisations and policy makers. After understanding the consumer’s needs, the requirements and opportunities for companies and the technological and institutional constraints, ClearFarm designed market-based solutions, based on Precision Livestock Farming, to provide easy to understand information on animal welfare status, as well as other environmental and economic sustainability information, to producers and consumers.

The proposed solutions included measurable indicators and standards, innovative technology integration, value proposition towards consumers, data management and data protection measures, development of ethical and systematic decision-making approaches, connections to animal welfare labelling organisations and should consider the impact of the system itself.

"The ClearFarm platform was designed to collect sensor-based animal welfare information from dairy cattle and pigs on farms located in different countries. This information is then transmitted to our computer servers in the cloud for storage and analysis. The data is then processed for end-users, such as farmers, to help them improve welfare monitoring, disease detection and decision making, and for consumers to help them take well informed decisions about the animal products they want to buy. The platform consists of two components: the back-end and the front-end.
Back-end
The data is gathered in dairy and pig farms using Precision Livestock Farming (PLF) technologies, such as collars, rumen bolus, GPS, Copeeks, and SoundTalks. For dairy cattle, welfare data is collected based on their activities (e.g., time spent on eating, walking, ruminating, etc.), health status (e.g., cleanliness of udders, legs, diseases) and milk production. For pigs, data come from a wide range of sources: 3D cameras or automatic feeders, measurements of ambient temperature, humidity, CO2 and NH3 level in the barn, among others. Additional data was obtained from veterinary records and visits to be used as reference values. Although this data is used to calculate the animal welfare indices, it is not as regular as the data generated by the PLF.
All the information gathered is transferred via internet to the platform’s data centre for animal welfare analysis and welfare index calculations. Depending on the technology used, the downloads from these farms can be made in different routines (e.g., daily, hourly, etc.)
The information is saved in MongoDB, an open-source database that does not follow traditional relational principles. Every data entry is in the form of a document, which can have several fields and is adaptable in size. Unlike other types of databases, the data in MongoDB doesn’t require a specific structure. Due to its ability to scale and adapt, we chose it as our database and installed it on a server dedicated primarily to its operation. This server is also utilised for the routine gathering of data from various sources using both PLF and non-PLF technologies.
To make back-end web development easier, we use the Flask framework. Flask is based on Python, a programming language that reduces development time and is easy to use and implement.
A additional machine learning server is set up to connect to the back-end, which usually requires more processing power for intensive calculations. This server is loaded with computer programs that calculate welfare indices using various algorithms developed in the ClearFarm project.
Front-end
The web server is the back-end that connects to the two servers mentioned above. It provides various interfaces in response to requests from end-users of the platform application. Users can only access pages created specifically for them on the web browser after logging in.
There can be as many front-ends as requested by different users: farmers, consumers, processors, retailers, food service providers, feed manufacturers. Here are two examples of front-ends created for dairy cows:
(1) Farmers: The farmer interface includes information on welfare scores (see Figure 1). It shows the animals’ time spent on feeding, health, housing, and global scores (an average calculation of all indicators), which range from 1 to 10. Users can also select a range of dates to view the scores. A colour scheme (red, yellow, green) is used to indicate the welfare of the cows (at individual level, and also at farm level). In addition, the application allows users to set up alerts if the scores fall below a certain threshold. The application also allows farmers to see the likelihood of individual animals suffering from diseases, such as mastitis and acidosis (see Figure 2).
(2) Consumers: By using the app on their mobile phones, consumers can scan the QR code on the product to find out what the welfare scores are for feeding, health, housing and global (see Figure 3). This will help them know that these products are from farms that prioritize animal welfare and enable them to check the animal welfare status on the farm. The scores range from 0 to 10, with 10 being the highest positive score and 0 being the least.
In summary, the ClearFarm platform has proven to be functional in providing users with information on farm animal welfare. It helps farmers to be aware of diseases and health issues that their animals may be suffering from. Other users, such as consumers, have the opportunity to be better informed about the welfare aspects of the animal products they buy."
 

"Sensor-based monitoring systems form the core of the ClearFarm project. The sensors installed in the ClearFarm project can retrieve information at pen level (e.g., climate sensors, video-based activity sensors), but wherever possible they are used to gather information at the individual animal level, for instance through accelerometers or electronic feeding stations.
Individual-level data enables us to identify animals with diseases, among other things. Additionally, the considerable amount of data collected on individual animals across time has provided us with insight into the large individual differences in behaviour exhibited by different animals.
Our initial understanding of animal individuality emerged when we studied the daily rhythms of feed intake behaviour in growing-finishing pigs. Similar to humans, we anticipated that most pigs would prefer to eat at certain times of the day and rest at other times, resulting in a clear daily rhythm. Although we observed a daily rhythm in pigs on approximately 57% of the days in the barn on average, this percentage ranged from 0% to 100% in individual pigs. These results showed that pigs were highly variable in the extent to which they exhibited daily rhythms.
In a follow-up study, we looked more closely at these individual differences between pigs and examined whether pigs had individual strategies in their feeding behaviour that they adhered throughout their time in the barn. In other words, we wanted to see whether pigs 1) differed from each other and 2) were consistent in these differences as they got older. We found that pigs exhibited feeding strategies in four different ways.
First, pigs showed feeding strategies by varying 1) the frequency and size of their meals. It is common knowledge that some people prefer snacking over eating full meals, while others eat in meals or choose something in between. Likewise, pigs seemed to have such a preference, ranging from nibblers, who frequently visited the feeding area but consumed small amounts, to infrequent meal eaters, who ate in large quantities.
Additionally, pigs exhibited differences in their 2) speed of feeding, ranging from fast to slow, and in 3) the amounts they consumed at night. Although all pigs consumed most of their food during the day, some also fed on a significant quantity of food at night while others refrained from doing so. Lastly, 4) some pigs were more consistent than others in their feeding times, having a strong daily rhythm and eating at the same time each day, while others were more flexible in their feeding schedules.
These results clearly demonstrate individual differences between pigs. This has consequences for ClearFarm’s monitoring systems, which aim to detect deviations from normal behaviour that are indicative of welfare issues. Whether or not a behaviour should be considered as deviating may differ for pigs with different feeding strategies. For example, lameness, a painful welfare issue, causes pigs to visit the feeder less often because they avoid the pain of standing up and walking. A pig might be identified as lame if it visits the feeding station fewer than five times a day. As a consequence, a pig who eats a lot in one meal and visits the feeder infrequently (i.e. a meal eater) may be mistakenly identified as lame. In contrast, a nibbler would need to significantly reduce its number of feeding visits before its lameness would be detected.
This detection issue can be overcome by modelling animal behaviour at the individual level, where each animal’s own normal behaviour is quantified. If there is a welfare problem, we can tell when an animal behaves differently from its usual behaviour. ClearFarm is improving this method to detect welfare issues more accurately in farm animals."
 

ClearFarm interviewed all the relevant actors of the pig and dairy value chains about their needs and preferences about Precision Livestock Farming.

Consumers agreed that Precision Livestock Farming could lead to many advantages for farm animals, farmers, producers and consumers. For them, technology can help to satisfy short-term needs pertaining to consumption, like taste and food safety. It also can help to satisfy their needs pertaining to animal-welfare and the environmental impact and importantly it may respond to an unfulfilled need for greater transparency and trustworthiness of the animal production chains.

Their main concern was that PLF would become a form of robotization of livestock farming at the expense of animals and farmers with even more intensified production systems. Consumers were also worried about how the technology would increase prices and they highlighted the importance of transparency, reliability and trustworthiness.

Farmers associated PLF with the opportunities to take care of every individual animal, to stay competitive on the market and to offer new sales opportunities because of an improved product segmentation. At the same time, it was indicated that integrating PLF technologies requires an increased acceptance of innovation and that the perceived benefits need to outweigh the risks.

For farmer cooperatives, PLF was expected to allow to produce sufficient quantity of meat and dairy products to satisfy the global market and that it allows to directly respond to individual animals’ needs. They also considered privacy as is a key point that needs to be taken into account.

Processors & slaughterhouses indicated the opportunity that PLF has the possibility to evaluate the entire lifetime cycle of animals. In addition, PLF can provide an approach to harmonise welfare standards and labelling across Europe. Their requirement was that PLF technologies should be non-invasive for the animals.

Retailers mentioned that PLF is an opportunity to access new market segments and to trade more transparent products. They also suggested that PLF can increase the transparency of the value chain and help to harmonise labelling approaches across the EU.

For technology providers, PLF has the capability of optimising routine processes whilst reducing the farmers’ workload. However, they warned of the requirement to have a robust and secure data (storage) system.

Consultants and researchers indicated that PLF can optimise the whole value chain, integrating innovative technology, algorithms and data management tools. The requirement that PLF has to cope with the claims of modern livestock farming was mentioned.

Finally, animal interest groups pointed that PLF has the opportunities to assure animal-friendly treatment of animals, that it facilitates the process of setting standards for animal welfare and that PLF might fosters public discussions, which may contribute to a critical evaluation of the own consumption behaviour.

Play behaviour has been proposed to be an indicator of animal welfare. As part of the ClearFarm research, the validity of play behaviour in weaner pigs is currently investigated as an animal welfare measure. To do so, the team explores the relationship between play behaviour and broadly used welfare measures, such as weight gain, stress hormones, presence of disease and injuries, and feeding behaviour. The researchers also assess the relationship of play behaviour with novel behavioural measures related to emotions such as tail posture and tail motion.
More stress, less play
In nature, sows decrease their nursing activity gradually and pigs are naturally weaned when they are 4-to-5 months old. However, pigs raised in modern production systems shift from milk to solid feed much earlier, as they are abruptly weaned from their sow at approximately 3 to 5 weeks of age.
A ClearFarm study demonstrated that pigs weaned at approximately 26 days of age showed a drastic reduction in time spent playing in the first 24 h after weaning compared to before weaning. However, the study also showed that weaning stress was reduced by keeping litters socially intact in their familiar environment after weaning, as indicated by a less pronounced reduction in time spent playing from the day before weaning to the day after weaning and a stepper increase in time spent playing on the second day after weaning.
These findings illustrate a suppression of play behaviour when pigs experience hunger and social instability, and a higher engagement in playing when social conditions are more stable. In fact, early weaning is a management practice known to inflict nutritional, physiological, and psychological stress in pigs, constituting a suitable context for examining the relationship of play behaviour with other animal welfare measures.
With this, ClearFarm will advance the understanding and validation of play behaviour as an animal welfare indicator and potentially promote the use of play behaviour in on-farm welfare assessment protocols.

Farm animal welfare is increasingly emphasised as a quality attribute of food and a growing number of EU citizens would like to have more information about how farmed animals are treated. Product labelling can inform consumers effectively about food quality and sustainability, increase the transparency of farming and provide better protection to EU producers who apply a high animal welfare standard.
Animal welfare labels reviewed by ClearFarm mostly measured animal welfare by using resource- rather than animal-based measures. Only few of them referred to the technical specification ISO/TS 34700:2016 on animal welfare management.
Generic EU product labelling regulations apply also to animal welfare labelling. For instance: 1) product must not be marketed with properties it does not have or that can be assumed to be shared by all other products, such as minimum legal requirements, 2) the marketing must not be misleading, and 3) comparative claims should be verifiable.
All packing labels should also be easily noticeable, readable, and comprehensible. In addition, the European Commission has provided best practice guides for voluntary certification schemes for agricultural products and foodstuffs.
The key factors to successful animal welfare labelling are 1) business operators’ desire for openness, 2) the participation of industry, retailers, and interest groups in designing and implementing the labelling, as well as 3) consumers’ awareness of the label and its benefits. Moreover, 4) the transparency of the labelling scheme and 5) wide involvement of actors in decision-making were identified as additional important factors. Particularly farmers appreciate the possibility to take part in the label’s decision-making. Informing consumers requires adequate, correct, and coherent communication so that they understand the benefits of the label. It is recommended that the verification of conformity is reliable, made by an independent body, and that regular inspections with clear, understandable, and realistic criteria take place.
Financial viability of labelling is important. A commercial label must provide adequate value to all relevant actors, including animals, consumers, farmers, food business operators and other business entities involved.
Studies (Yang & Renwick, 2019; Cicia & Colantuoni, 2010) suggest that consumers are typically willing to pay some 15-30% price premium for products originating from high animal welfare farming systems. However, products are different, and consumers are a heterogeneous group, with dynamical changing shopping habits. While some are interested in animal welfare and willing to pay a price premium for welfare improvements, others may not be.
Digitalisation offers new opportunities to change the way labels are used in a business environment. It allows innovative technologies to be used for the benefit of consumers, which can both change the way of communication and give rise to new business models based on precision livestock farming technologies and data.
Quality labels are recommended to adhere to the principle of continuous improvement. The technical standard of animal welfare label affects how much impact the label can generate. Enhanced quality is not delivered for free. Rather, it requires effort at the different stages of the value chain. As the price rises, some consumers become excluded as the buyers of welfare-labelled products. Thus, the additional cost of delivering high quality cannot be too high.

"Machine learning plays a key role in the ClearFarm project, as it is necessary to process and analyse the large amounts of data ollected by automatic sensors. Machine learning models can be used to predict the likelihood of a particular animal suffering from a disease and notify farmers, allowing them to take the necessary corrective action and improve animal welfare and performance. In this practice abstract, we will focus on the use of machine learning to predict the probability of disease in dairy cows, in particular the probability of a cow having mastitis or acidosis.
Accurate prediction In a general machine learning approach, the goal is to create a model that can accurately predict the relationship between predictor variables and predicted variables. Predictor variables are the inputs or features used to make predictions, while predicted variables are the outputs or targets to be estimated or classified. To train the model, a dataset is divided into a ‘training set’, a ‘validation set’ and a ‘test set’. The first two are used to automatically optimise the model’s parameters, allowing it to learn from the patterns and relationships in the data. During training, by adjusting its parameters, the model tries to minimise the difference between its predicted output and the actual output on the training set, and the model’s performance is first assessed on the validation set until training produces an acceptable prediction score, or accuracy. This is the rate of adequate predictions, which is usually set at 70-80%.
Once the model has been trained on the ‘training set’ and assessed on the ‘validation set’, it is evaluated on the ‘test set’ to assess its predictive ability. The ‘test set’ contains data that the model has not seen during training, and its performance on this set helps to estimate how well it will perform on unseen data.
Detecting mastitis and acidosis Mastitis is a common and significant health issue in dairy cows. It is an inflammation of the mammary gland, usually caused by a bacterial infection. This condition can have a negative impact on milk production and animal welfare. Infected cows may show symptoms such as swelling, pain, redness and changes in milk consistency. Mastitis not only reduces milk yield but also leads to the presence of somatic cells and bacteria in the milk, making it unsuitable for human consumption. These somatic cells in the milk increase its conductivity, and a cow producing milk with a conductivity above 5.5 mS/cm is considered a potential case of mastitis. Therefore, conductivity will be our predicted variable for mastitis prediction.
Acidosis is a common metabolic disorder in dairy cows. It occurs when the pH balance in the rumen becomes excessively acidic. Acidosis can have serious consequences for a cow’s health and productivity. Affected cows may show symptoms such as decreased appetite, diarrhoea, lameness, and reduced milk production. In addition, acidosis disrupts the balance of beneficial rumen micro-organisms, leading to further digestive disturbances. Therefore, a cow with a rumen pH below ~5.8 is considered a potential acidosis case. The target pH used for training and testing will come from sensors in the cow’s rumen bolus.
For both mastitis and acidosis, the predictor variables will be the automatic data collected by the sensors and other information about the animal:
• Behavioural data from today: time spent eating, ruminating, lying down, and walking.
• Accumulated behavioural data from previous 20 days.
• Days in milk (days the cow has been lactating).
• Number of lactations per day.
Both of our predicted variables are continuous, and therefore a regression model is used to predict them. The model used is a simple vanilla encoder-only transformer, with a regression head on top that outputs the predicted variables.
Our model shows excellent performance in predicting both mastitis and acidosis on the test set. For mastitis, the model effectively identifies animals with high conductivity values, indicating potential mastitis cases. While there are some instances where the model fails to detect mastitis in some animals, most of the few errors of the model are on the side of caution, producing false alarms for animals that do not have mastitis.
Similarly, for acidosis, the model successfully identifies most animals with acidosis by considering pH values below the specified threshold. However, there may be occasional false alarms, typically of low probability, where the model identifies cases just above the threshold. Again, it is important to emphasise that these false alarms are preferable to missed cases, as it is ultimately the farmer and veterinarian who will determine the course of action by visiting the potentially affected cow.
In conclusion, we can predict with reasonable confidence the probability of a cow developing mastitis or acidosis. These predictions can be made on a daily basis using the data automatically collected by the sensors, giving the farmers a daily snapshot of the status of the animals on their farm. By incorporating this near real-time feedback, farmers should be able to improve the welfare and performance of their animals."
 

"Machine Learning is a rapidly growing field of artificial intelligence that involves leveraging data to automatically make predictions or decisions without explicit programming. The ability to process large amounts of data has made Machine Learning increasingly useful in recent years, allowing for the efficient analysis and understanding of complex information across a variety of industries, including healthcare, finance, manufacturing, and animal production.
In the context of the livestock production, data is collected using Precision Livestock Farming technologies, which use sensors to gather a wide range of data on individual and group animals in real-time. These data include information on the animal’s behaviour and activity such as the time an animal spends feeding, lying, walking, and ruminating, as well as on the physical state of the animal such as body temperature or rumen pH. The goal of Precision Livestock Farming is to provide farmers with valuable insights and information that can help them improve the welfare and productivity of their animals.
However, the complexity and volume of data that Precision Livestock Farming generates can be overwhelming for a manual processing. This is where automatic approaches based on Machine Learning come into play. Machine Learning models are trained using veterinarian assessments as reference values, to predict the likelihood of animal illnesses such as mastitis in dairy cows or respiratory problems in pigs.
These predictions can be used to send early warnings to farmers, allowing them to perform the necessary corrective measures and increase the welfare and output of the animals.
In addition to these predictive models, unsupervised learning methods such as clustering and statistical techniques can be employed to identify abnormal behaviour, based on the normal ranges defined by veterinary experts. Identification of animals that are out of the normality range may help to identify problems of welfare related to health but also to feeding to housing and to appropriate behaviour welfare domains.
In summary, Precision Livestock Farming is a technology that provides farmers with valuable data on the behaviour and physical state of their animals in real-time.
To make these data useful, Machine Learning techniques are key to process the data into human understandable alerts and scores, which will help farmers raise the welfare and productivity of their animals."
 

Mastitis is one of the most common diseases of high-producing dairy cows with significant economic and animal welfare implications.
Mastitis is a multifactor disease and results from the inflammation of the mammary gland. The severity of the inflammation can be classified into either sub-clinical or clinical mastitis. In the clinical form, milk alterations, mammary gland modifications and changes in animal status are easily visible. but no visible symptoms are observed in subclinical mastitis.
The actual ability to identify early stages of mastitis or predict the likelihood of mastitis in healthy herds is limited, and it can be expensive.
Cows with clinical mastitis may show sick behaviours and/or changes in their behavioural patterns (i.e. changes in feed and water intake, and feeding, ruminating and lying time). Identification of these behavioural changes at the onset of mastitis can be useful as an early detection system.
Precision livestock farming (PLF) (e.g. accelerometers’ technology), permits non-invasive real-time data collection providing continuous information about behaviours related to individual activity (i.e. feeding, ruminating, lying behaviour, standing, walking, etc.). from which the health of the cow can be estimated.
The implementation of accelerometery collars used in the ClearFarm project, recorded the daily hours spent ruminating, eating and lying down in dairy cows. Behaviour was monitored all time. For the study 5 days prior to diagnosis of mastitis, the day of diagnosis and 3 days after were considered. All cows were treated with antibiotics at the day of diagnosis.
Our results seem to demonstrate that cows with mastitis showed a significant reduction in time spent ruminating four days before the diagnosis and that this behaviour recovers after antibiotic treatment. Suggesting that rumination may be a proxy behaviour to early detection of mastitis. An early diagnosis of clinical mastitis can improve the welfare of cows and can reduce the costs, treatments, and the cow’s replacement rate.
ClearFarm will advance the understanding of the possible role of PLF technology in monitoring dairy cows’ behaviours to identify automatically sick cows and to predict the onset of diseases that threaten animal welfare in the dairy industry.

The Five-Domains Model for animal welfare assessment inspires the methodology of welfare assessment used in ClearFarm. This model is an integrative approach for animal welfare assessment based on the assumption that animal welfare is structured into five different domains including (1) nutrition, (2) physical environment, (3) health, (4) behavioural interactions and (5) mental state. Each of these domains can be assessed by a list of indicators that provide a quantitative outcome and that can be monitored continuously, 24 hours a day and 7 days a week.
For instance, the time a cow spends lying is an indicator that can be used to monitor her physical comfort, which pertains to the physical environment domain. ClearFarm will identify the existing Precision Livestock Farming (PLF) tools (sensors) available on the market that can effectively monitor these indicators, in order to know the welfare status of an animal at any time.
Following on with the example of laying time and comfort, activity sensors may help monitoring the time a cow spends lying. With the data from sensors on the farm, ClearFarm will develop a digital platform that will integrate the welfare indicators providing precise and continuous information. Farmers will have continuous access to precise information about the welfare status of their animals with regard to different domains.
ClearFarm will inform farmers when a cow is sick, and whether sickness has implications for other domains such as nutrition, comfort, or else. Real-time monitoring will not only allow to identify the problem (and to put in place the most appropriate remedy) but also to monitor the recovery, so treatments could be refined according to each animal’s needs. This approach will facilitate a comprehensive welfare assessment and help farmers to identify any critical aspect that compromises animal welfare at any time, which should be used to refine the corrective measures to improve the animal’s quality of live.
Animal welfare is crucial to ensure the maximum efficiency of livestock production. So, the maximum profit will only be achieved if animals are in the finest conditions. Having the ability to control the status of animals on farm continuously will help farmers to refine livestock production and achieve the highest efficiency in their production system.

EU Regulation 1169/2011 on the provision of food information to consumers establishes a labelling system with a list of indications that each food label is obliged to include (name, list of ingredients and allergens, etc.).
In addition to these obligatory elements, Article 36 provides for the possibility to include additional information on the label on a voluntary basis: in this case, the information must not mislead the consumer, must not be ambiguous or confusing, and, where appropriate, must be based on "relevant scientific data".
Uneven labelling schemes
The reference to animal welfare only appears on the label through the voluntary channel (as this parameter is not included among the declarations indicated as mandatory).
This has led to the development of different animal welfare labelling schemes in the different Member States, each based on different criteria, mostly unknown to the consumer, which often carry the risk of having a misleading impact on them.
The ClearFarm project is based on the development of a software platform capable of collecting data from various types of sensors, integrating them and extrapolating information on animal welfare to make it available to producers and consumers (by entering the data on the label) to help them in their decision-making.
Thus, the project creates a digital system powered by precision livestock technologies that allows not only to monitor the welfare of each animal but also to constantly verify in real time the supply chain's compliance with current regulations, reinforcing the final consumer's confidence. Therefore, everything is based on objective and scientific analysis methods, i.e., detectors (IoT [Internet of Things] sensors, optical readers, NFC [Near-Field Communication] devices, etc.) based on animal-based measurements (ABM) —such as behavioural, physiological, pathological, and production parameters—which are constantly updated with the latest scientific evidence.
The use of the latest precision livestock farming (PLF) technologies confers absolute reliability, transparency, and immediacy to the system, drastically reducing the possibility of consumer deception and confusion and the risk of providing potentially misleading information.
ClearFarm system is, therefore, not only fully in line with European provisions on voluntary food labelling, but also capable of adding value to European agri-food communication, providing consumers with essential information to guide them towards more informed consumption choices.

"The ClearFarm project is based on the use of technology on the farm to obtain information on animal welfare on pig and dairy farms through a specific algorithm, allowing the monitoring of the physiological or health status of each animal and offering an exceptional tool for tracing animal welfare throughout the supply chain.
This project is not only part of the development process of Precision Livestock Farming (PLF), understood as the management of livestock farms with the support of advanced technologies, but also of the more general evolution, primarily scientific, that marks the passage from an evaluation of animal welfare through the consideration of minimum essential parameters (food, water, space, air, temperature, etc.), to the use of more objective and scientific methods of analysis, capable of measuring behavioural, physiological, pathological and productive parameters of the animals.
In spite of the fact that this vision responds to a multifactorial concept of animal welfare, which includes both the physical state of the animal and the psychological one, it is worth emphasising that, to date, the aim of PLF is -still- exclusively that of controlling and maximising production processes in terms of quantity and quality, guaranteeing the farm a saving of resources (water, energy, fertilisers, etc.), a containment of costs and a reduction in environmental impact. The exploitation of PLF technologies in terms of supporting animal welfare management is therefore only indirect, since it is based on the use of data not specifically produced for this purpose -and it is precisely in this direction that the ClearFarm project is developing.
This scenario, however, must be seen in a context that is constantly changing and evolving, both from a scientific and a juridical point of view.
If it is true that the Clearfarm project is part of a European overview characterised by a regulatory framework on farm animal welfare that is still unsatisfactory, it is also fair to point out, however, that the European Union is currently undergoing a thorough review of all its animal welfare legislation in order to modernise it, taking into account the latest scientific evidence, with a view to a regulatory review scheduled for summer 2023.
This circumstance, which undoubtedly testifies to the EU’s willingness and commitment with respect to updating and raising animal welfare standards, allows us to reflect on the increasingly central role that technological development and PLF tools will play with respect to the ability to monitor and protect the welfare of animals along the production chain, especially assuming the future development of ad hoc PLF technologies, specifically programmed to assess parameters concerning the physiological state and health of the animals.
In these terms, therefore, Clearfarm must be seen not only as a cutting-edge project, which enhances the potential contribution that PLF can bring to the welfare of production animals, but also as a precursor project, which opens the door to future scenarios that are certainly innovative and increasingly specifically aimed at supporting animal well-being."
 

"The pig industry has intensified in recent years, esulting in farms with bigger herds and fewer employees. Sensor technology can be a vital tool under these new conditions to ensure animal welfare, as it permits continuous and simultaneous monitoring of various indicators.
Currently, several sensors are available in the market or under development that can monitor animal welfare on farms. Examples include a 3D camera, accelerometers, microphones, climate sensors, and RFID (radio frequency identification) ear-tags, amongst others. These sensors can collect information from the animals and/or the housing environment based on the technology they use.
Information about the environment can include variables such as room temperature, humidity, or air quality, while information about animals may incorporate body temperature, activity level, feeder visits, spatial distribution in the pen, coughing, or lameness, to name a few.
In the ClearFarm project, we used a sensor device that can measure information from the animals (hereafter referred to as animal-based parameters): activity level, spatial distribution and posture, to be precise.
*Measuring the activity levels*
Research has shown that the activity level of pigs can vary substantially due to incidents like leg injury (lameness), diseases or stress. In addition, the spatial distribution and posture of pigs can indicate their behavioural patterns during the day, including their resting and feeding habits and the level of comfort while resting in the pen.
The aim of the trial was, firstly, to validate the sensor data against manually collected data (the so-called ‘ground truth’ or ‘gold standard’) and, secondly, to investigate whether these animal-based parameters could be used in practice to measure the welfare status of pigs.
Four sensor devices were installed on commercial farms, and three batches of nursery and fattening pigs were monitored. Our study results showed a high level of agreement between computer vision and human observation for spatial distribution and posture.
However, it was not possible to validate the activity level data collected by the sensor device due to significant differences in the way the camera and the human observer recorded the activity. Moreover, it was found that changes in the environment caused alterations in the activity level.
For example, an increase in temperature or a decrease in humidity, CO2, and NH3 led to an increased level of activity. Furthermore, the increase in activity was linked to higher levels of aggression, stress, and inflammation, as determined by skin lesion counts and salivary biomarkers. The activity level also varied between days and production stages.
Despite the inability to confirm the activity level through human observation, our study shows a correlation between events that could adversely affect pig welfare (e.g. changing environment, air quality, aggression level, and physiological status in pigs). To summarise, our findings indicate that sensor-collected activity levels could be used as a tool for pig welfare monitoring."
 

Life Cycle Assessment or LCA is a standardized methodology (ISO 14040 ff) that assesses the environmental impacts of a product, process, or service throughout its entire life cycle. When applied to livestock systems such as the pork and dairy value chains, the LCA informs about the impact of these production systems on global warming, land use, eutrophication, acidification, water use, ecotoxicity, among other indicators. It can also help identify hotspots (phases in the value chain that contribute most to the pollution or resource use) and to define strategies to reduce the environmental impact of animal-derived products.
The amount and the composition of feed, water, the energy consumption on- and off-farm, or manure management, for instance, are examples of relevant input data needed for an LCA.
The required primary data is collected directly from farmers and producers such as integrators or cooperatives. Countries included in ClearFarm LCA are Spain, Italy, Germany, Finland, and The Netherlands, representing different regions and production systems throughout Europe. Based on the completeness and quality of data, some assumptions and secondary data sources of information are required to fill data gaps. The ClearFarm consortium has a good combination of diverse expertise, especially veterinary physicians, which helps to make such assumptions based on their knowledge of the system.
An important aspect when performing an LCA is the definition of the system boundaries (i.e. what processes to include). Another important aspect in animal-derived products is the assessment of animal-derived systems considering co-products as butter, cheese, and meat from the dairy farms. The environmental impact has then to be allocated to different products leaving the farm. Since there are different allocation methods, the outcome may also be different.
Once all data are gathered and the system boundaries are defined, the potential environmental impacts of the pork and milk production can be calculated. The calculation is based on existing data repositories such as Ecoinvent, GaBi, as well as databases with local inventories as the LCADB® developed by ICTA-UAB. To make such calculations, a series of characterization factors have been defined and agreed. For example, one of the most used impact categories in LCIA is the global warming potential (GWP). The reference substance to account for the GWP is carbon dioxide (CO2). All substances contributing to GWP –e.g., methane (CH4) and nitrous oxide (N2O), gases commonly emitted in livestock production through enteric fermentation or manure management–, are accounted for in kg or g of CO2 equivalent.
In ClearFarm, performing an LCA of the pork and dairy value chain from over five different countries helps comparing differences due to day-to-day practices and the implementation of diverse farm technologies across countries, and how they affect the environment.
One of the most important challenges in ClearFarm is understanding the potential contribution of LCA results to the assessment of animal welfare in the various farms assessed. There is still limited literature linking how a change in the farm management can benefit or constraint animal welfare. Identifying potential benefits and trade-offs between animal welfare and environmental performance would be a major outcome of the project. To achieve such goal, the ClearFarm team develop a unique platform and concentrate a high level of expertise to develop scores that reflect both the environmental and the animal welfare domains.

Precision Livestock Farming (PLF) is the use of advanced technologies to reduce labour and to monitor and optimize farming processes. It is based on the use of sensors capable to monitor a vast range of variables with an interest for animal welfare, environmental impact and productivity.

These technologies provide real-time data of individual animals and groups of animals as a whole, that allow to monitor their welfare status and make sure that a fast reaction will occur to solve specific problems.

There are different technologies developed for monitoring pig and dairy cattle production. For instance, sensors installed on animal or in the nearby environment can detect sudden change in animals’ behaviour, such as in feeding, drinking, rumination, moving, vocalization or productivity. Moreover, the physical state of the animal, such as temperature, progesterone level or rumen pH can be monitored by thermal cameras, automatic milking stations or rumen boluses respectively, to give some examples.