Utilizing Machine Learning to Understand and Predict Methane Emissions in Cattle Farming with Farm-Scale Environmental and Biological Variables

Tom Partridge, Baihua Li, Bashar Alhnaity, Qinggang Meng

Abstract: Cattle livestock contribute to climate change through enteric methane production, making it essential to identify and validate methods for reducing methane emissions. This research correlates GreenFeed cattle methane measurements with farm environment data using the North Wyke Farm Platform (NWFP), a heavily instrumented research facility in the UK. The disparate datasets are combined into a machine-learning-ready dataset capable of mapping methane emissions in grams per day and grams per kilogram of live weight gain. Predictive models are then developed and evaluated for methane prediction. Experimental results indicate that Gradient Boosting achieved the highest accuracy (g/day: r=0.619, RMSE=51.8; g/kg live weight gain: r=0.562, RMSE=65.9). Explainable AI methods are applied to quantify how a broad selection of farm and animal characteristics contribute to methanogenesis. This research provides valuable insights into methane reduction through machine learning and quantitative analytical methods.

External IDs:dblp:conf/icca3/PartridgeLAM24