Predicting methane production in dairy cows

Abstract: Methane is a potent greenhouse gas, to which enteric fermentation from ruminants contributes significantly. Reliable and accurate predictions of methane (CH₄) production from dairy cows would be of interest to develop mitigation strategies and for national inventories. Thus, the overall aim of this thesis was to predict CH₄ production in dairy cows by modelling approaches. Predicted in vivo CH₄ production decreased with increased sample size in the gas in vitro system. Molar proportion of acetate decreased at the expense of propionate. Digestibility also decreased with increased sample size. Predicted CH₄ production based on stoichiometric equations of volatile fatty acids was in good agreement with observed values of CH₄ production from the gas in vitro system. Dry matter intake per kilogram of body weight, organic matter digestibility and dietary concentrations of neutral detergent fibre, non-fibre carbohydrates and ether extract were the variables of the best fit model predicting CH₄ energy as a proportion of gross energy (prediction error 4.65% of the observed mean). The non-linear models developed proved to be more applicable over a wider range of intake for predicting total CH₄ production than linear models. Adjusting the exponents for dietary concentration of fat, proportion of non-fibre carbohydrates in total carbohydrates and organic matter digestibility improved the model. The sub-model predicting CH₄ production in the Karoline model was revised. Modifications were made to equations predicting digesta passage kinetics, microbial cell synthesis, digestion in the hind-gut and utilisation of hydrogen. The sensitivity analysis suggested that accurate values for digestion kinetic variables are required for accurate and acceptable predictions of CH₄ production with mechanistic models. The Karoline model was evaluated against published data (n=184 diets) reporting CH₄ production from in vivo trials. There was a good relationship between observed and predicted values of CH₄ production, with a small root mean square error of prediction (10.1% and 6.1% of the observed mean for fixed and mixed models, respectively). The mean bias was small (<2%) but statistically significant, and there was no slope bias. Most of the error was due to random bias (96.4%), whereas the contributions of mean and slope bias were small (3.4 and 0.2%, respectively).