Enhancing the Detection of Potential Anthropogenic Methane Emission Sources in China Using Machine Learning and TROPOMI Observations

Methane emissions are strongly influenced by human activities. The global methane budget framework indicates that direct anthropogenic emissions, including those from agriculture, waste, fossil fuel extraction, and biomass/biofuel combustion, account for 57% (bottom-up) to 65% (top-down) of total global emissions, underscoring the dominant role of human activity. Here, we analyze methane emission sources and classify their types across China using TROPOspheric Monitoring Instrument (TROPOMI) satellite observations from 2019 to 2023. Results show that the number of potential methane emission sources retrieved from satellite data increased initially and then declined over the study period. A multiparameter machine learning classification model successfully distinguished between potential natural and anthropogenic methane sources, achieving accuracy scores of nearly 0.75 on the validation set and approximately 0.7 on the test set. We find that 12–25% of the anthropogenic methane sources detected by TROPOMI over the five-year period were not recorded in EDGAR. These unlisted sources exhibit similar longitudinal patterns to EDGAR but show marked latitudinal discrepancies, particularly in northwestern and northeastern China. Our study provides an effective framework for methane source inversion and identification using TROPOMI, offering valuable support for informing mitigation strategies in China.