Featured Articles

Introduction: The Evolution and Application of Information Technology to Assist with Animal Production Analytics

The symposium focused on “The Evolution and Application of Information Technology to Assist with Animal Production Analytics” and was introduced by Luis O. Tedeschi, PhD from Texas A&M University. The Nation Animal Nutrition Program (NANP) was established in 2010 to support research requested by stakeholders. It addresses challenges in research, education, and teaching, and supports agencies in animal agriculture. Information Technology (IT) has evolved greatly over the years particularly in combination with data analytics. In the animal industry, understanding the application of information technology of data analytics can provide a competitive advantage. In order to obtain the most value out of data in a competitive environment, it is important to understand what can be learned from data, the appropriate amount of data required, and different viewpoints regarding the data. Data analytics is essential in establishing the true relationships with the given variables in order to enhance research. Predictive technologies are being improved to share, combine, and analyze data, develop models.

Opportunities and Limitations of Modeling and Data Analytics for Precision Livestock Farming

The introduction was followed by a talk about “Opportunities and Limitations of Modeling and Data Analytics for Precision Livestock Farming” by Aline Remus, PhD. According to her, Precision livestock farming (PLF) involves the use of technology such as sensors, scales and cameras to collect information about a group of animals. Processing of this data can be done through mathematical models and/or artificial intelligence algorithms. Mathematical modeling involves an equation or set of equations to represent the behavior of a system. There are several types of models that can be used such as deterministic models that are not associated with a probability and stochastic models which are probability associated and consider variation within a population. Static models do not make time dependent predictions whereas dynamic models predict how quantities vary as a function of time. Empirical models are based on correlation whereas mechanistic models provide a degree of understanding to what is being modeled. In using mathematical models to assess data, there are several limitations. Models work retrospectively using historical population information which assumes that every animal will have the same response to the nutrient provision and models can also be complex for users. Due to this, models must operate at a small group level to consider variations in animal response to adjust parameters accordingly using technologies that measure in real time which PLF allows for. An important component of PLF is precision nutrition in which models operate in real time with the actual data collected allowing for the daily tailoring of diets to individual animals. Data driven empirical models and mechanistic models can be combined to process data coming through sensors and recalibration for machine learning. Artificial intelligence is beneficial for large volumes of data but can be used in combination with mathematical models to control complex PLF components. However, all types of PLF models must be able to consider changes over time from the herd and animal perspective. Control devices such as automatic feeders can manipulate diets to create appropriate outcomes for individual animals and collect real time individual feed information. Overall, PLF has to be user friendly and simple to be adopted and depends on the new generation of training in the field. If implemented successfully, it can also be environmentally sustainable and cost effective.

Application of Precision Sensor Technologies, Real-time Data Analytics, and Dynamic Models on Extensive Western Rangeland Grazing Systems

Dr. Hector M. Menedez and Dr. Jameson Brennan followed with a talk on “Application of Precision Sensor Technologies, Real-time Data Analytics, and Dynamic Models on Extensive Western Rangeland Grazing Systems”. The use of precision livestock management has advanced the monitoring of animal health through the use of sensors and technology. While research has allowed for improvements in efficiency on the basis of the health, reproductive, and nutritional status of animals, the research has been concentrated in dairy operations and feedlot settings. There have been difficulties in incorporating the technology into rangeland production systems due to environmental conditions and challenges of data access across extensive landscapes. New technologies such as low power wide-area network communication allow for the transmission of data over long distances in these areas. Edge computing allows for the amount of data to be reduced in order to be sent. Many challenges exist despite some advances in technology in remote areas including topography, dynamic conditions and climate, lack of control over feed, and low palatability of highly nutritious feed in certain areas. However, precision livestock management along with mathematical models could be used in rangeland settings to improve ruminant nutrition. To overcome the challenges, technology such as satellites can be used to view areas of crop stress. Rangeland Analysis Platform leverages big data such as weather conditions to allow for management based on trends in the forage available. GPS tracking accelerometers can monitor cattle distribution on the landscape and can predict and model factors such as behavior, energy expenditure, and grazing selection. SmartScales measures the front end weights of animals as they drink which helps to determine genetic and environmental factors to consider in analyzing animal production. Furthermore, GreenFeed is an alternative to respiration chambers that evaluates energetic efficiency as well as enteric methane and carbon dioxide conditions. Continuous grazing is typical of animals in the Western United States but precision grazing allows for the use of real time individual data. Data from climate and forage quality can be used to determine where to move animals next for optimal efficiency. To apply the information, virtual fencing systems can be used to draw virtual fences of where the cows should be so that they graze in high quality forage areas. The animal receives a shock stimulus to keep them in the desired area. Some considerations need to be made with the utilization of this system as batteries might not perform optimally in certain climates, animals can show aversion to devices, and communication could be delayed. In the models ecosystem health and energetic cost of animal traveling also must be included. Overall, these technologies will allow for optimal management of animals in the rangeland grazing systems.

Mapping Resilience Indicators and Measuring Emotions of Farm Animals Using Sensor Data

Dr. Suresh Neethirajan presented “Mapping Resilience Indicators and Measuring Emotions of Farm Animals Using Sensor Data”. According to him, sensor-based data is essential for not only the physiological functioning of animals but also for animal welfare. There is an economic value to understanding moment to moment continuous data streamed on a daily basis. Sensors can be used to measure emotions to measure the resistance of an animal against harmful forces. Resilience involves the capacity of an animal to stay healthy and respond minimally and rapidly to circumstances of stress. It is also the ability to accommodate and adapt to external disturbances. Sensors can measure the stimuli that act on animals, the internal state of animals, and emotions of animals in invasive and noninvasive manners to produce solutions. Moment to moment data can provide meaningful critical insights to understand an animal better. Digital phenotyping is a new way of understanding behavior. PLF allows for real time management, indications of health using sensors, and data to create proactive interventions. Types of data include facial expression data, happiness index, and vocalization which can indicate health and heart and respiration rate. For example, ChickTrack uses cameras to measure chicken activity without sensors or handling to monitor social behavior and flock density, and a deep learning based model can be used for frame by frame analysis. Tracking analysis can give insight to the direction in which chickens flock and visualize the data in a user friendly manner. WUR Wolf is an automated facial recognition system that can detect negative, positives and neural postures in cows and pigs as well as change in mood over time. Some challenges in PLF include data storage, poor internet connectivity in rural areas, and capturing and analyzing data. Overall, many modalities are needed to develop a framework for digitalized animal farming.

The Adoption of AI in the Core Scientific Cycle of Feed Research

Marc Jacobs, PhD, a researcher at Nutreco, followed with “The Adoption of AI in the Core Scientific Cycle of Feed Research”. In his view, the application of artificial intelligence (AI) in the animal science industry is not straightforward because of implementation. Data science does not flourish because of difficulties in translating a vision into tangible processes. Most animal models in the feed industry are mechanistic. Valid models give health and growth predictions, allow for least cost formulation and are sustainable. Models are initially based on trials and literature and then used by customers to make it interactive. After considering nutrient requirements, actual animal response is then used to optimize using mathematical models, not necessarily AI. To achieve these goals, near-infrared spectroscopy, laboratory information management system and mycotoxin databases are being combined. Models can be both dynamic and static. Empirical models are dependent on data and predict animal responses. Mechanistic models are more theory based and animal centric, and allow for optimization. Machine learning models can be combined with animal centric models to create a hybrid for systems optimization. Models can be used to predict the nutrition profile of an ingredient, predict the probability of extending above a threshold, obtain early signals on expected safety hazards in animal feed ingredients. This will allow for estimation of nutrient composition of raw materials, assess risk of mycotoxin contamination, improve silage evaluation, estimate nutrient profiles, and improve net energy estimation. Overall, while AI is a valuable tool for research, sometimes the needs of a customer can also be met with simpler mathematics.

Integrating Mechanistic Models with AI for Precision Feeding of Sows

Dr. Charlotte Gaillard spoke about “Integrating Mechanistic Models with AI for Precision Feeding of Sows”. Sow feeding is generally based on the average nutrient requirement for the herd which can lead to over- or under-feeding. Under-feeding can lead to reproductive issues and over-feeding can entail higher feed costs. Feeders, sensors, scales and AI used in precision feeding account for individual variability in estimating nutritional requirements with real-time measurements. Daily nutrient requirements for individual sows during lactation and gestation were calculated using a mechanistic model applied to a database. The mechanistic and dynamic model was a factorial approach to estimate nutrient requirements by compartment in sows. In lactating sows, nutrient requirement was most influenced by milk production and appetite whereas parity, gestation stage, and activity level were most influential in gestating sows. Algorithms were then developed to predict these parameters of interest found when estimating nutritional requirements. For gestating sows, noninvasive cameras are being used to determine individual activities to create an algorithm. Litter weight at weaning and feed intake were predicted for lactating sows using a supervised learning algorithm. The decision support systems composed of models and algorithms were tested on farms. An interface was used to link the decision support systems and the feeders to allow for entry of observational data. The sows that were fed using this precision feeding were found to have reduced lysine and phosphorus intake which led to reduced excretion of phosphorus and nitrogen. There were also reduced feed costs through real-time individual adjustment of nutrient supply. Precision feeding strategies must meet requirements of all sows, improve animal welfare, and reduce cost to optimize sow feeding management.

EnROADS: Overview of Climate Change Modeling

Dr. Charles Jones presented “EnROADS: Overview of Climate Change Modeling”. EnRoads is an online climate interactive model. In his view, showing people research and expecting change does not work and it is instead important to give people a tool to change ways of thinking. EnRoads is a globally aggregated bar scale economy and energy model that serves as a tool to deliver interactive lessons about climate change. It displays several outputs and fields at a time and has sliders that a user can control to simulate taking action in the climate energy system. Categories of sliders include energy supply, transport, buildings and industry, energy efficiency, growth, land and industry emissions, and removal. It allows the user to put economic, social, and environmental changes that a user would like to accomplish or views as a possibility and then the model shows potential outcomes including greenhouse gas emissions, energy demand and consumption, population, land use, costs, and temperature increase by 2100. EnROADS is free, widely used, and available in several languages. Support and training are available, but it is easy to use at a basic level. To teach using the model, Dr. Jones aims to make it a social and emotional experience grounded in science. Currently, there are limits in the model agricultural system but in the future, food demand based on population and wealth as well as emission and land use for crops and animal products will be added. Eventually, a greater level of detail will be displayed on the model to view how changes in animal diet and waste could impact climate change. The model can be used at en-roads.climateinteractive.org.

Statistical Graphics and Interactive Visualization in Animal Science

Dr. Gota Morota presented a workshop about “Statistical Graphics and Interactive Visualization in Animal Science”. In his view, data visualization is fundamental to understanding and visualizing data analysis in animal science. The workshop focused on predicting the body weight of pigs from images. Image processing was done using GNU Octave. Computer visions systems allow for automated animal phenotyping and activity monitoring. In this case, depth sensor cameras set up in swine facilities were used to take videos of the pigs. Dr. Morota demonstrated the selection of frames from the videos to select images that weights could be obtained from. Images can be obtained from restrained or non-restrained animals. Animal restraint can produce higher quality images but with more labor involvement and less practicality. Length, width, and height estimations of the pig can be taken from the image to calculate the volume. The R.A. Shiny application has the potential to improve interactive visualization of data analysis but has limited use in the field of animal science. The workshop demonstrated the capabilities of R Shiny and how it can be applied in the field of animal science using interactive data exploration tools.

A Brief Overview, Comparison and Practical Applications of Machine Learning Models

The symposium concluded with Dr. Dan Tulpan who presented a workshop entitled “A Brief Overview, Comparison and Practical Applications of Machine Learning Models”. The workshop focused on training researchers in formulating problems that can be solved by machine learning techniques. Machine learning techniques can be used to classify, cluster, and perform linear and nonlinear regressions to animal science datasets. The workshop used the Waikato Environment for Knowledge Analysis (WEKA) workbench for machine learning. The workshop consisted of demonstrations of classification problems, regression problems, and data-related artifacts. Data-related artifacts that could hinder the use of WEKA include formatting errors and hidden correlations.