Can Acoustic Sensors Provide Automated Measurement of the Eructation Patterns of Cattle?

Methane is emitted from cattle primarily via eructation. The aim of this study was to use collar‐attached acoustic sensors and automated deep learning models to predict the frequency and intensity of eructation from cattle in outdoor pasture‐based grazing systems. Trials were conducted with 10 cows for 3 days and a total of 1848 eructation events were manually labelled over the period. Models were calibrated to a 1‐h period and validated to a second 1‐h period and a 3‐day period. The accuracy of the eructation classification model was 87% and 82%–83% for the calibration and validation datasets. The measured average eructation frequency over the 3‐day trial was 0.73 eructations per minute with a between‐cow variation characterised by a standard deviation of 0.21 ± 0.05 eructations per minute. The measured average intensity of eructation (1–3 integer scale, which was a subjective measure of eructation volume) over the 3‐day trial was 2.09 with a between‐cow variation characterised by a standard deviation of 0.30 ± 0.07. There was a strong relationship between the predicted and measured between‐cow variation in (1) the eructation frequency for the calibration ( R 2 = 0.75) and validation ( R 2 = 0.80) datasets and (2) the eructation intensity for the calibration ( R 2 = 0.97) and validation ( R 2 = 0.90) datasets.