Quantifying Global Wetland Methane Emissions With In Situ Methane Flux Data and Machine Learning Approaches

Area of Science:

• Earth Science
• Climate Science
• Environmental Science

Background:

• Wetland methane (CH4) emissions significantly influence global climate dynamics.
• Current global estimates of wetland CH4 emissions exhibit substantial uncertainties.
• Accurate quantification is vital for climate modeling and mitigation strategies.

Purpose of the Study:

• To develop and apply machine learning (ML) models for estimating global wetland CH4 emissions.
• To reduce uncertainties in CH4 emission estimations through a multi-model ensemble (MME) approach.
• To project future wetland CH4 emissions under various climate change scenarios.

Main Methods:

• Developed six distinct bottom-up ML models using in situ CH4 flux data from chamber measurements and the Fluxnet-CH4 network.
• Employed a multi-model ensemble (MME) approach to synthesize model outputs and reduce uncertainty.
• Incorporated environmental variables such as precipitation, temperature, soil properties, wetland, and climate types into the models.
• Extrapolated the MME to global scale for estimating CH4 emissions from 1979 to 2099.

Main Results:

• Estimated current annual wetland CH4 emissions at 146.6 ± 12.2 Tg CH4 yr-1 (1979-2022).
• Projected future emissions to increase under different Shared Socioeconomic Pathways (SSPs): 165.8 ± 11.6 (SSP126), 185.6 ± 15.0 (SSP370), and 193.6 ± 17.2 (SSP585) Tg CH4 yr-1 by the end of the 21st century.
• Identified Northern Europe and near-equatorial regions as current CH4 emission hotspots.

Conclusions:

• The ML-based MME approach provides a robust data-driven method for estimating global wetland CH4 emissions.
• Future wetland CH4 emissions are projected to rise, with varying magnitudes depending on the emissions scenario.
• Further research should focus on comprehensive CH4 measurements and improved characterization of wetland spatial dynamics to reduce quantification uncertainties.