Sensor Technologies for Nutritional Management of Ruminants

Abstract

Precision livestock farming is gaining popularity in both the research and production setting. Despite this, current technologies are limited in the ability to explore the rumen environment. The overall goal of this work was to explore sensing technologies that could enable shifts in management to maximize productivity in ruminant production systems. In the first study, we assessed the use of existing sensor technologies to monitor ruminal volatile fatty acid (VFA) concentrations using pH sensing. Four ruminally cannulated Holstein cows at maintenance were included in a Latin Square Design. Treatments consisted of a) chopped grass hay, b) 85% chopped grass hay and 15% cracked corn and soybean meal, c) 70% chopped grass hay and 30% cracked corn and soybean meal, and d) 55% chopped grass hay and 45% cracked corn and soybean meal. Prior to receiving treatment diets, cows were individually housed and underwent a fasting period of up to 24 hours. During each period, cows were allowed access to treatment diets from 0600 to 0800 hours, and rumen fluid samples were collected hourly for twelve hours beginning at feed delivery. A bench pH meter was used to obtain rumen fluid pH levels at sampling times. Concentrations of individual VFA and branch-chain VFA were analyzed statistically using two linear mixed effects models. In one model type, VFA were estimated through fixed effects terms for treatment, time, and the treatment by time interaction. For comparison, the other model estimated VFA concentrations using linear and quadratic effects for the sensed pH data. Both models leveraged random effects for animal and period. Models utilizing diet data and time showed better performance in estimating VFA concentrations compared to models leveraging pH data, indicating minimal predictive capacity was identified for the pH sensing. The second study explored opportunities to track ruminal VFA concentrations based on aqueous sensing of ruminal CO2, temperature, and conductivity across four diets differing in energy and protein supply. Four ruminally cannulated Holstein cows at maintenance were included in a Latin Square Design. Treatments consisted of a) chopped grass hay (9.07 kg), b) chopped grass hay (9.07 kg) plus cracked corn (4.08 kg), c) chopped grass hay (9.07 kg) plus soybean meal (2.13 kg), and d) chopped grass hay (9.07 kg) plus corn (2.38 kg) and soybean meal (0.83 kg). Prior to receiving treatment diets, cows were individually housed and underwent a fasting period of up to 24 hours. During each period, cows were allowed access to treatment diets from 0600 to 0800 hours, and rumen fluid samples were collected hourly for twelve hours beginning at feed delivery. A CO2 electrode and conductivity probe were placed in the rumen of each cow to investigate the relationship between aqueous ruminal CO2, temperature, and conductivity, with sensor measurements recorded every three minutes beginning at 0545h. Concentrations of individual VFA were analyzed statistically using a linear mixed effects model with fixed effect for treatment and sensing data and random effects for animal and period. Single-point-in-time modeling of VFA concentrations from sensor data demonstrated comparable or improved results in terms of error variance and Concordance Correlation Coefficient (CCC) compared to models using diet and time variables. Incorporating time-lagged sensor variables further improved the predictive capacity and reduced residual error variance. Adding diet descriptions to the lagged sensor data did not enhance the ability to explain variability in VFA concentrations. These models indicate VFA concentrations can be well characterized from aqueous, ruminal sensing of CO2, temperature, and conductivity, in a manner apparently independent of and robust across diets.