Reducing the Discrepancy in Quantifying the Temperature Dependence of Global Wetland Methane Emission

Understanding the apparent temperature dependence of wetland methane emissions (E_M) is critical for predicting climate-carbon feedbacks, yet current estimates remain constrained by observational limitations and methodological inconsistencies. The inherent biogeographic heterogeneity of wetland ecosystems combined with sparse, unevenly distributed flux measurements introduces substantial uncertainty in characterizing spatial patterns of E_M. This knowledge gap impedes accurate projections of wetland methane contributions under climate warming scenarios. Here, we develop a framework that integrates mixed-effects models with artificial intelligence techniques to resolve scale-dependent patterns in methane emission thermodynamics across global wetlands. Our unified framework demonstrates that only 73.6% (5th-95th quantiles: 71.8%-75.4%) of the global wetland area conforms to classical Arrhenius-type temperature dependence. This framework can predict 69.5% (67.9%-71.1%) of the global wetlands with high confidence using the Mahalanobis distance and area of applicability tests. We quantify the weighted mean E_M across high-confidence predictable areas of 0.694 eV, with latitudinal differentiation: tropical (0.634 eV), temperate (0.678 eV), and boreal (0.745 eV) wetlands exhibit progressively stronger temperature responses. Ignoring these biogeographic variations could result in underestimation of projected end-century methane emissions by 4.2%-13.3% across selected socioeconomic pathway scenarios. Our study refined the temperature sensitivity parameter in coupled climate-carbon cycle models, thereby enhancing predictive accuracy of future global warming trends and informing strategic responses to climate change mitigation.