Individual milk fatty acids are potential predictors of enteric methane emissions from dairy cows fed a wide range of diets: Approach by meta-analysis.

There is a need to quantify methane (CH) emissions with alternative methods. For the past decade, milk fatty acids (MFA) could be used as proxies to predict CH emissions from dairy cows because of potential common rumen biochemical pathways. However, equations have been developed based on a narrow range of diets and with limited data. The objectives of this study were to (1) construct a set of empirical models based on individual data of CH emissions and MFA from a large number of lactating dairy cows fed a wide range of diets; (2) further increase the models' level of complexity (from farm to research level) with additional independent variables such as dietary chemical composition (organic matter, neutral detergent fiber, crude protein, starch, and ether extract), dairy performance (milk yield and composition), and animal characteristics (days in milk or body weight); and (3) evaluate the performance of the developed models on independent data sets including measurements from individual animals or average measurements of groups of animals. Prediction equations based only on MFA [C10:0, iso C17:0 + trans-9 C16:1,cis-11 C18:1, and trans-11,cis-15 C18:2 for CH production (g/d); iso C16:0, cis-11 C18:1, trans-10 C18:1, and cis-9,cis-12 C18:2 for CH yield (g/kg of dry matter intake, DMI); and iso C16:0, cis-15 C18:1, and trans-10 + trans-11 C18:1 for CH intensity (g/kg of milk)] had a root mean squared error of 65.1 g/d, 2.8 g/kg of DMI, and 2.9 g/kg of milk, respectively, whereas complex equations that additionally used DMI, dietary neutral detergent fiber, ether extract, days in milk, and body weight had a lower root mean squared error of 46.6 g/d, 2.6 g/kg of DMI, and 2.7 g/kg of milk, respectively). External evaluation with individual or mean data not used for equation development led to variable results. When evaluations were performed using individual cow data from an external data set, accurate predictions of CH production (g/d) were obtained using simple equations based on MFA. Better performance was observed on external evaluation with individual data for the simple equation of CH production (g/d, based on MFA), whereas better performance was observed on external evaluation mean data for the simple equation of CH yield (g/kg of DMI). The performance of evaluation of the models is dependent on the domain of validity of the evaluation data sets used (individual or mean).