Identifying the main drivers of methane flux in wetlands using machine learning and FLUXNET data across climate zones

Global atmospheric methane (CH₄) emissions have risen significantly, tripling in atmospheric concentrations since preindustrial times. Wetlands, as the largest natural source of CH₄ emissions, contribute significantly to the global CH₄ budget. However, quantifying wetland CH₄ emissions remains highly uncertain due to the complex interplay of hydrological and biogeochemical processes. In this study, we develop a random forest (RF) and SHapley Additive exPlanations (SHAP) framework to identify the main predictors of CH₄ emissions across different climate zones and on a global scale. We used monthly global environmental variables and CH₄ flux emissions from FLUXNET-CH4 dataset, incorporating 39 wetland sites over the globe. These sites are classified into tropical, temperate, and boreal regions by latitude. Key variables considered in the analysis included mineral-associated organic carbon, soil organic carbon, soil moisture, and canopy height. Our findings reveal that air temperature and latent heat are the most important predictors of CH₄ at both global and regional scale. Regionally, tropical wetlands are primarily influenced by canopy height, water table level and soil organic carbon while soil temperature emerges as the dominant driver in temperate and boreal wetlands. Furthermore, we analyze the similarities and differences in CH₄ predictors across climate zones to improve our understanding of regional and global wetlands CH₄ dynamics. Understanding the main predictors of CH₄ emissions across wetland regions is essential for improving CH₄ budget accuracy on both regional and global scales.