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 CH emission prediction models based on a variable CH conversion factor (Y)-based approach that quantitatively relates Y 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-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 CH 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; %), and milk protein (M; %). 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 value of 0.065 and adjusted Y values according to production levels, as provided in the Intergovernmental Panel on Climate Change guidelines, were also assessed as existing representative Y-based prediction models. Among the Y-based models developed in this study, the model with the highest predictive performance was as follows: CH emissions (MJ/d) = exp(-2.74 + 0.000325 × BW - 0.00883 × MY + 0.116 × M - 0.142 × M) × GEI. This model captured the variation in Y, with an R of 0.30. Notably, although a substantial bias related to production levels was observed in the existing Y-based models, no such bias was observed in the variable Y-based models developed in this study. Although the proportion of Y variance accounted for by BW, MY, and milk composition was relatively low, our results highlight the advantages of using the variable Y-based models to predict CH 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-based models in situations where feed characteristics data are not available.