The School of Earth and Atmospheric Sciences Presents Dr. Thomas Weber, University of Rochester

A Machine Learning Approach to Estimate Methane Emissions from the Global Ocean and Laurentian Great Lakes

Oceanic emissions represent a highly uncertain source in the natural atmospheric methane budget, and are thought to be highly sensitive to environmental change. The primary limitation in constraining this term is the very sparse sampling of dissolved methane distributions in the surface ocean mixed layer, which is needed to compute the air-sea flux. 

Here, we overcome this limitation using statistical mapping methods. We compiled a large dataset of methane supersaturation in surface waters, and trained machine learning models to map its climatological distribution as a function of other well sampled biogeochemical variables. 

Our approach yields a global diffusive methane flux of 4.1±2 Tg/yr from the ocean to the atmosphere, or a total flux of 4-15Tg/yr once ebullition is accounted for. These fluxes are towards the lower end of the range adopted by recent IPCC reports (5-25 Tg/yr), but exceed the upper end of estimates extrapolated from individual cruise data (0.3-3 Tg/yr). Our statistical method also provides important insights into methane production mechanisms in the ocean, revealing a significant relationship to net primary production that is consistent with hypothesized aerobic methanogenesis during organic matter cycling. 

Finally, I present ongoing work to apply these methods to the Laurentian Great Lakes, and place new constraints on global freshwater methane emissions.