Understanding the Interannual Variability and Trend of Boreal Arctic Wetland Methane Emissions Using an Observation-Driven and Long-Term Upscaling Dataset

Methane ranks as the second-largest contributor of current anthropogenic warming, with a Global Warming Potential (GWP) approximately 28-34 times larger than that of carbon dioxide (CO₂) over a 100-year time horizon. The CH₄ emissions from the Boreal-Arctic (BA) wetlands significantly contribute to global CH₄ emissions, and are sensitive to climate change. However, in the BA area, substantial discrepancies (ranging from ~9 to 53 TgCH₄ yr^-1 regarding estimated annual emissions) exist between the bottom-up (BU) biogeochemistry and top-down (TD) atmospheric inversion models, and the long-term responses of wetland CH4 emissions to climate change remain highly uncertain. To better understand the complex nature of wetland CH₄ emissions in the BA area and to benchmark TD and BU models, we upscaled the Boreal-Arctic wetland CH₄ emissions from 2002 to 2021 using an observation-driven and causality-guided machine learning model. With the upscaled dataset we analyzed the environmental drivers of the interannual variability and trend during the past two decades. We found that our model achieved reasonable accuracy, and the upscaled wetland methane emissions showed reasonable magnitude, and spatial distribution compared to those of 13 BU models and 21 TD models. During 2002-2021, a significant increasing trend and strong interannual variations in wetland CH₄ emissions were found. Temperature and vegetation activities predominantly explained the trend and interannual variations of wetland CH₄ emissions. BU and TD models exhibited poor performance on capturing the long-term temporal dynamics of wetland CH₄ emissions, with possible underestimation of future-warming driven wetland methane emission changes in the BA area.