Long-Term Monitoring of Greenhouse Gas Emissions and Interpretable Machine Learning Models for Methane Prediction in Dairy Cattle Slurry Storage

Greenhouse gas (GHG) emissions from continuously filled dairy cattle slurry storage remain difficult to estimate under practical conditions. This study monitored methane (CH4), nitrous oxide (N2O) emissions, and slurry physicochemical properties for 15 months in laboratory-scale storage tanks designed to reproduce key features of on-farm storage, including continuous filling, farm-derived temperature dynamics, and periodic emptying events. N2O concentrations remained below 0.5 ppm and contributed negligibly to measured direct GHG emissions. In contrast, CH4 showed pronounced seasonal variability: 94.5% of annual CH4 emissions occurred during summer and autumn, whereas only 5.5% occurred during winter and spring. Summer emptying caused a sharp decline in the CH4 flux followed by partial recovery, whereas emptying during colder periods had limited effects. Machine learning (ML) models achieved robust performance during temporal external validation (R2 = 0.82) and identified pH, temperature, and volatile fatty acids as key nonlinear predictors of CH4 emissions. By combining cross-season monitoring with interpretable ML models, this study highlights the importance of long-term observations for characterizing CH4 emissions from continuously filled dairy cattle slurry storage estimation and targeted mitigation.