A novel random effect based simulation system for the replication of dairy cattle methane emission experiments

Methane (CH₄) emissions produced by dairy cattle (DC) represent production inefficiency within the animal, as well as a key source of agricultural greenhouse gas emissions. Predictive models based on animal information can be used to estimate CH₄ emissions, enabling low CH₄ emitting DC to be selectively bred, yet accumulating datasets of sufficient size to produce robust models requires considerable investment. Therefore, in this study, we attempted to develop a simulation system which could accurately replicate authentic DC CH₄ emission datasets. To assess the accuracy of the system, we compared the performance of 23 extant DC CH₄ emission prediction models between the original dataset the simulation system was based on, and the simulated versions developed. Assessed via the Root Mean Square Prediction Error (RMSPE) and Concordance Correlation Coefficient (CCC), the respective model metrics ranged from 45.92 (g/d) to 70.44 (g/d) and 0.41 to 0.84 on the original data and 36.01 (g/d) to 66.29 (g/d) and 0.44 to 0.89 on the simulated data. The improved performance of the models on the simulated data compared to the original data suggest a slight bias within the current system. However, once ordered by RMSPE and CCC, the model hierarchy between the original and simulated data remained consistent, with a model on the simulated data, never being more than 2 places from its order when evaluated on the original data. Therefore, the current simulation system still provides a solid foundation from which to produce DC CH₄ emission experiment datasets, and could assist in the development of DC CH₄ emission prediction models once its accuracy has been refined.