Direct and indirect predictions of enteric methane daily production, yield, and intensity per unit of milk and cheese, from fatty acids and milk Fourier-transform infrared spectra.

Mitigating the dairy chain's contribution to climate change requires cheap, rapid methods of predicting enteric CH emissions (EME) of dairy cows in the field. Such methods may also be useful for genetically improving cows to reduce EME. Our objective was to evaluate different procedures for predicting EME traits from infrared spectra of milk samples taken at routine milk recording of cows. As a reference method, we used EME traits estimated from published equations developed from a meta-analysis of data from respiration chambers through analysis of various fatty acids in milk fat by gas chromatography (FA). We analyzed individual milk samples of 1,150 Brown Swiss cows from 85 farms operating different dairy systems (from very traditional to modern), and obtained the cheese yields of individual model cheeses from these samples. We also obtained Fourier-transform infrared absorbance spectra on 1,060 wavelengths (5,000 to 930 waves/cm) from the same samples. Five reference enteric CH traits were calculated: CH yield (CH/DMI, g/kg) per unit of dry matter intake (DMI), and CH intensity (CH/CM, g/kg) per unit of corrected milk (CM) from the FA profiles; CH intensity per unit of fresh cheese (CH/CY, g/kg) and cheese solids (CH/CY, g/kg) from individual cheese yields (CY); and daily CH production (dCH, g/d). Direct infrared (IR) calibrations were obtained by BayesB modeling; the determination coefficients of cross-validation varied from 0.36 for dCH to 0.57 for CH/CM, and were similar to the coefficient of determination values of the equations based on FA used as the reference method (0.47 for CH/DMI and 0.54 for CH/CM). The models allowed us to select the most informative wavelengths for each EME trait and to infer the milk chemical features underlying the predictions. Aside from the 5 direct infrared prediction calibrations, we tested another 8 indirect prediction models. Using IR-predicted informative fatty acids (FA) instead of FA, we were able to obtain indirect predictions with about the same precision (correlation with reference values) as direct IR predictions of CH/DMI (0.78 vs. 0.76, respectively) and CH/CM (0.82 vs. 0.83). The indirect EME predictions based on IR-predicted CY were less precise than the direct IR predictions of both CH/CY (0.67 vs. 0.81) and CH/CY (0.62 vs. 0.78). Four indirect dCH predictions were obtained by multiplying the measured or IR-predicted daily CM production by the direct or indirect CH/CM. Combining recorded daily CM and predicted CH/CM greatly increased precision over direct dCH predictions (0.96-0.96 vs. 0.68). The estimates obtained from the majority of direct and indirect IR-based prediction models exhibited herd and individual cow variability and effects of the main sources of variation (dairy system, parity, days in milk) similar to the reference data. Some rapid, cheap, direct and indirect IR prediction models appear to be useful for monitoring EME in the field and possibly for genetic/genomic selection, but future studies directly measuring CH with different breeds and dairy systems are needed to validate our findings.