Predicting Enteric Methane Emissions in Dairy Cows Using Deep Learning Models

This study evaluates the potential of deep learning (DL) models to predict enteric methane (CH₄) emissions in dairy cows using data from automated milking and feeding systems, behavioral sensors, and public weather databases. Methane emissions were recorded for 52 cows from October 2022 to December 2023 using the sniffer technology at Gladbacherhof, an organic research farm run by the Justus Liebig University Giessen, Germany. Among the tested models, Long Short-Term Memory (LSTM) networks outperformed Convolutional Neural Networks (CNNs) and hybrid CNN-LSTM models given that data from all sources were available (Scenario S1), achieving an R² of 0.88 and a mean bias error (MBE) of 13.55 ppm CH₄. To assess model applicability under varying data scenarios, features were categorized as "rare," "moderate," or "public" based on their ease of acquisition. Using only public weather data (Scenario S2) resulted in poor predictions, while incorporating moderate-effort farm data (Scenario S3) improved accuracy (R² = 0.45, MBE = 17.60). Adding three rarely available feed-related features, namely feed efficiency, concentrate intake, and total dry matter intake considerably enhanced performance (Scenario S4: R² = 0.74, MBE = 14.36). Random forest analysis highlighted feed-related data as critical for improving prediction performance. These findings demonstrate the capability of DL models to accurately predict CH₄ emissions using readily accessible farm data integrated with a small set of high-impact feed-related features. This approach provides a valuable tool for developing targeted strategies to mitigate methane emissions in dairy farming.