Global near-surface methane distribution and emission hotspots from TROPOMI: A geospatial machine-learning approach

Methane (CH₄) is a major greenhouse gas; however, the global information on its near-surface distribution and emission hotspots remains sparse. Although satellite instruments provide column-averaged CH₄ (XCH₄), the link between these and surface-level concentrations remains unclear, especially across diverse source environments. This study combines Tropospheric Monitoring Instrument XCH₄ mixing ratios with meteorological and environmental variables to estimate global near-surface CH₄ mixing ratios from May 2018 to May 2024. Four machine-learning models were trained and evaluated using near-surface CH₄ measurements from 44 World Data Centre for Greenhouse Gases sites. Among these, the RF model performed most consistently (root mean squared error = 0.033 ppmv, mean absolute percentage error = 1.23%). Satellite XCH₄, tropospheric NO₂, season, and surface temperature emerged as key drivers of the estimated CH₄ variability, linking the model to both emission strength and boundary-layer conditions. The resulting fields reproduce large-scale gradients between continents and reveal enhanced CH₄ over industrial, agricultural, and wetland regions, with localized increases of up to approximately 0.06 ppmv above nearby background levels. These hotspot patterns are broadly consistent with previously reported anthropogenic and natural source regions. Furthermore, comparisons with the emission database indicate stronger correlation between our estimates and actual emissions (correlation coefficient (R) = 0.73) than XCH₄ (R = 0.06), especially in fossil fuel sectors. Thus, our model effectively extracts surface signals, extending CH₄ information to areas with little or no in situ monitoring. This study effectively combines satellite observations and machine learning to link column and surface CH₄ information, thereby supporting regional methane emission assessment and future mitigation efforts.