Quantifying Wetland Methane Emissions from the Southeastern United States: A Data-driven Approach, Key Variables, and Spatiotemporal Distributions

Abstract

As the second most significant greenhouse gas (GHG), methane (CH4) contributes ~20% to the cumulative GHG-induced global warming. Among all methane sources, wetlands are the single largest and climate-controlled natural source. Estimating wetland methane emissions typically involves inversion (“top-down”) or process-based (“bottom-up”) models. Nevertheless, estimates derived from these two model types are not independent and exhibit large disparities. To better understand wetland methane emissions and refine the process-based and inversion models, independent high-resolution and long-term wetland methane flux data are needed. Here, we develop a high-spatial-resolution (0.0083° × 0.0083°, approximately 1 km × 1 km) monthly wetland CH4 flux dataset for the Southeastern (SE) United States (US) from 1982 to 2010 using a data-driven random forest (RF) approach. We utilize CH4 flux measurements from four FLUXNET-CH4 wetland sites to develop the RF regression model along with 11 explanatory variables, including air temperature, precipitation, vapor pressure deficit, incoming shortwave radiation, wind speed, Palmer Drought Severity Index, water table level, leaf area index, the fraction of absorbed photosynthetically active radiation, season, and wetland type (tidal versus non-tidal). Wetland CH4 fluxes estimated using the RF model fit well with the measured CH4 fluxes (R2 = 0.91) from four representative FLUXNET-CH4 wetland sites across the SE US, outperforming previous CH4 flux upscaling studies. Leveraging the developed RF model and wetland distribution data from the National Wetland Inventory, we map the spatial distribution of CH4 emissions in the study region. Our mapping reveals large spatial variability in CH4 emissions, ranging from 0 to 266.0 nmolCH4 m-2 s-1, with the coastal wetland areas, the Mississippi Delta, and the Everglades being the predominant sources of CH4. Our dataset demonstrates good agreement with the remote sensing-derived wetland CH4 fluxes from the Carbon Monitoring System Methane Flux for North America product, confirming the credibility of our wetland CH4 flux estimations. Variable importance analysis highlights that air temperature and the Palmer Drought Severity Index are key environmental predictors, explaining 88% of the variance in measured methane emissions from SE US wetlands. This first-ever high-spatial-resolution (0.0083° × 0.0083°) and long-term (1982-2010) monthly gridded regional wetland CH4 flux product over the SE US provide a benchmark and an added constraint for future wetland CH4 flux modeling and upscaling studies in the study region. The insights gained from this study contribute valuable understanding of the environmental controls on CH4 emissions, offering guidance for quantifying GHG emissions at the regional scale.