63. Evaluation of enteric methane emissions from dairy cattle for spot sampling schemes using quantitative approaches

Introduction: Measuring daily enteric methane (CH4) emitted from dairy cattle using spot sampling, as applied in GreenFeed systems, requires accurate frequency and timing of gas collection. Measurement accuracy increases with sampling frequency (Lee et al., 2022), and a minimum sampling frequency is particularly required for dairy cattle fed restrictedly (Van Lingen et al., 2023). The aim herein is to assess the accuracy of sampling schemes for dairy cattle enteric CH4 emissions across various studies using simulation approaches. Material and methods: Data were taken from five in vivo experiments in which CH4 emission was measured for 72 h using climate respiration chambers with sampling intervals ≤ 15 min. Four experiments applied feeding at 80–95 % of ad lib twice daily and the fifth twice daily ad lib feeding. Per observation, smoothing splines were fitted to the CH4 emission data using a penalized least squares objective function to control the wiggliness of the curve. The area under the curve (AUC) of the smoothing splines quantifies the daily CH4 production. Next, to estimate daily CH4 production from the fits, 5 different sampling schemes with and without error inclusion were applied, including sampling at 0.5, 1, and 2 h intervals starting at 0, 0 and 0.5 h from morning feeding, respectively (0.5_0.0, 1.0_0.0, 2.0_0.5), at 6 h intervals starting at 2 h from morning feeding (6.0_2.0), and at 1, 7 and 16 h from morning feeding (1.0 + 7.0 + 16.0). Further, raw data points were sampled using the 5 sampling schemes. Results and discussion: Fitted smoothing splines commonly followed the data points showing the two post-prandial peaks per day (Fig. 1a), which indicated the applied penalty on the least squares was appropriate. CH4 productions sampled from smoothing splines without error inclusion for the 0.5_0.0, 1.0_0.0, 2.0_0.5, 6.0_2.0 and 1.0 + 7.0 + 16.0 sampling schemes were not different from the AUC for 3, 3, 2, 0 and 0 out of the 4 restricted feeding datasets, respectively (e.g. Fig. 1b). Error-included smoothing spline sampling for the 5 sampling schemes had CH4 productions not different from the AUC for 3, 3, 3, 2 and 0 out of these 4 datasets, respectively. For raw data sampling, this applied to 5, 5, 4, 4 and 1 of these 4 datasets, respectively. For the ad lib data, no sampling scheme differed from the AUC for all 3 procedures.