Precision Agriculture Tools for Livestock Production

Caleb J. Grohmann

FFAR Fellow, University of Missouri

Columbia, MO

Livestock producers make several critical decisions daily that impact the health of animals and the farmer’s bottom line. Yet the supply of quality livestock technicians and managers who can make these decisions is decreasing across all sectors of animal agriculture, especially in the United States swine industry. Precision livestock farming technologies offer an interesting opportunity to provide support in making accurate decisions that positively impact the productivity and sustainability of swine operations. In my Ph.D. research at the University of Missouri–Columbia, we are striving to build tools to help pig farmers proactively and positively impact pig survivability in wean-to-finish pig barns.

Many factors contribute to increases or decreases in mortality rates in commercial pig production, such as genetic selection programs, environmental variation, management protocols and infectious diseases. To reduce mortality rates in a wean-to-market pig barn, intervention in the form of vaccination, injectable or broad-use water medication, ventilation, facility modification or other changes in management protocols are required when one or more factors negatively impact pig health and performance. Historically, deciding when to intervene on the pig, room or farm level was based on a visual appraisal of pig health and performance or barn climatic conditions by herd managers or veterinarians. The increasing development, acceptance and implementation of modern technologies in animal agriculture have allowed “big data” collection and continuous monitoring of individual animals, farm subsections (i.e., rooms or barns) or entire farms. In general, these technologies are non-invasive and fully automated. They’re comprised of platforms based on digital images, sounds or sensors. Robust and efficient systems to collect, store, analyze and communicate data from sensors and computer vision applications have the potential to enhance management practices, product quality and animal health and well-being.

Well-calibrated and precise predictive models remove human error and time considerations from these decisions by incorporating diverse and automated data on variables that influence mortality. We have evaluated the efficacy of a multi-faceted array of sensors (cough, nutrient availability and environmental), manually collected production data (treatment administration and inventory) and developed a robust algorithm to automatically label the start, peak and end of mortality episodes in wean-to-finish pig barns, regardless of location, season and management. Using these labels as a dependent variable, machine learning models can be deployed that predict the onset of a mortality episode in near real-time (i.e., 1 to 5 days before the start). This allows producers to devise optimum intervention plans before severe losses are realized. We expect the utilization of an automated intervention trigger through machine learning models, alongside visual inspection of animals and facilities by technicians, will reduce the overall mortality in farms that adopt the technology compared to farms that rely solely on manual visual appraisal. Implementing these predictive models in the industry will decrease the overall time to intervention, reducing the intensity and duration of mortality episodes.

The Foundation for Food and Agricultural Research Professional Development Fellowship has been an integral part of my PhD training. The skills I have acquired through this program will greatly benefit my future academic and professional career, such as interpersonal relationship building,  communication in the workplace and efficient time management. The work I have completed thus far would not have been possible without funding from the USDA-FFAR program and The Maschhoff’s, LLC, and I am forever grateful for their past and continued support.

ID: 548795