Method: Comparing averaging methods for gas flux data generated by automated head chamber systems

Researchers are increasingly using automated head chamber systems (GreenFeed; C-Lock inc., Rapid City, SD) for estimating gaseous emissions, such as carbon dioxide and methane, and consumption, such as oxygen. Our objective was to explore different data preprocessing methods. For this investigation, we collated data from 5 previously published manuscripts – 3 from grazing studies and 2 from studies utilizing finishing beef steers. We compared simple arithmetic or time-bin (8, 3-h intervals) averaging and least-squares means (LSMEANS) methodologies to arrive at a single estimate for each animal from gas estimates for each visit. For the LSMEANS approach, a mixed effects model was fit for each gas as the dependent variable, animal ID as fixed effects, visit duration and average airflow as covariates, and date and hour of day by animal ID as random effects. If duration and average airflow were not significant, they were removed from the model. After fitting the model, LSMEANS were generated for each animal with a standard error of the mean for each animal estimate. We then analyzed the data for each experiment according to the model presented in its respective manuscript, to obtain residual standard deviation and to calculate the coefficient of variation. Time-bin averaging increased unexplained error relative to arithmetic averaging and the LSMEANS approach. The increased unexplained error resulted in time-bin averaging having a greater coefficient of variation by 11.2% for pasture and 6.1% for finishing trials compared with arithmetic averaging and by 13.5% for pasture and 6.1% for finishing trials compared with the LSMEANS approach. We conclude that the proposed LSMEANS approach controls for any potential diurnal variation in gas flux, without increasing unexplained error as seen by time-bin averaging.