Methane emission prediction models for lactating cows based on feed intake, body weight, and milk yield and composition: Variable methane conversion factor-based approach

The first objective of this study was to develop CH₄ emission prediction models based on a variable CH₄ conversion factor (Y_m)-based approach that quantitatively relates Y_m to BW, milk yield (MY), and milk composition. The second objective was to evaluate the predictive performance of the developed models, particularly focusing on differences between the constant and variable Y_m-based models. A dataset of 266 records, sourced from previous experiments performed using whole-body respiration chambers or headboxes, was used for analysis. The models were developed using linear mixed models and generalized linear mixed models, with the latter performed to constrain the relationship between CH₄ emissions and gross energy intake (GEI) to a straight line passing through the origin. Different combinations of variables were used for the model development, including DMI (kg/d) or GEI (MJ/d), BW (kg), MY (kg/d), milk fat (M_FAT; %), and milk protein (M_PROT; %). The prediction accuracy and precision of the developed models were assessed using k-fold cross-validation. Additionally, prediction bias regarding milk production levels was evaluated. Models based on a default Y_m value of 0.065 and adjusted Y_m values according to production levels, as provided in the Intergovernmental Panel on Climate Change guidelines, were also assessed as existing representative Y_m-based prediction models. Among the Y_m-based models developed in this study, the model with the highest predictive performance was as follows: CH₄ emissions (MJ/d) = exp(-2.74 + 0.000325 × BW - 0.00883 × MY + 0.116 × M_FAT - 0.142 × M_PROT) × GEI. This model captured the variation in Y_m, with an R² of 0.30. Notably, although a substantial bias related to production levels was observed in the existing Y_m-based models, no such bias was observed in the variable Y_m-based models developed in this study. Although the proportion of Y_m variance accounted for by BW, MY, and milk composition was relatively low, our results highlight the advantages of using the variable Y_m-based models to predict CH₄ emissions without bias regarding production levels. Although the models developed in this study may have challenges in terms of universality due to the limited dataset, the modeling methods proposed in this study would represent useful tools for developing country-specific Y_m-based models in situations where feed characteristics data are not available.