Smoothing spline assessment of the accuracy of enteric hydrogen and methane production measurements from dairy cattle using various sampling schemes

Estimating daily enteric hydrogen (H2) and methane (CH4) emitted from dairy cattle using spot sampling techniques requires accurate sampling schemes. These sampling schemes determine the number of daily samplings and their intervals. This simulation study assessed the accuracy of daily H2 and CH4 emissions from dairy cattle using various sampling schemes for gas collection. Gas emission data were available from a crossover experiment with 28 cows fed twice daily at 80% to 95% of the ad libitum intake, and an experiment that used a repeated randomized block design with 16 cows twice daily fed ad libitum. Gases were sampled every 12 to 15 min for 3 consecutive days in climate respiration chambers. Feed was fed in 2 equal portions per day in both experiments. Per individual cow-period combination, generalized additive models were fitted to all diurnal H2 and CH4 emission profiles. Per profile, the models were fitted using the generalized cross-validation, REML, REML while assuming correlated residuals, and REML while assuming heteroscedastic residuals. The areas under the curve (AUC) of these 4 fits were numerically integrated over 24 h to compute the daily production and compared with the mean of all data points, which was considered the reference. Next, the best of the 4 fits was used to evaluate 9 different sampling schemes. This evaluation determined the average predicted values sampled at 0.5, 1, and 2 h intervals starting at 0 h from morning feeding, at 1 and 2 h intervals starting at 0.5 h from morning feeding, at 6 and 8 h intervals starting at 2 h from morning feeding, and at 2 unequally spaced intervals with 2 or 3 samples per day. Sampling every 0.5 h was needed to obtain daily H2 productions not different from the selected AUC for the restricted feeding experiment, whereas less frequent sampling had predictions varying from 47% to 233% of the AUC. For the ad libitum feeding experiment, sampling schemes had H2 productions from 85% to 155% of the corresponding AUC. For the restricted feeding experiment, daily CH4 production needed samplings every 2 h or shorter, or 1 h or shorter, depending on sampling time after feeding, whereas sampling scheme did not affect CH4 production for the twice daily ad libitum feeding experiment. In conclusion, sampling scheme had a major impact on predicted daily H2 production, particularly with restricted feeding, whereas daily CH4 production was less severely affected by sampling scheme.