Abstract

Methane (CH₄) oxidation by microbes is the largest biological sink of global methane, yet its magnitude and long-term variability remain uncertain. Here, we combined process-based (PB), machine-learning (ML), and atmospheric inversion approaches to evaluate the global methane soil sinks and its implication for the atmospheric CH₄ budget in this study. Both PB and ML approaches estimated annual global methane soil sink to be 40–45 Tg CH₄ yr⁻¹, 30%–50% larger than conventional estimates. Although the two approaches agreed on total magnitude, they differ in their representation of variability. PB models simulate stronger spatial heterogeneity, seasonality, and long-term increases in CH₄ uptake because environmental sensitivities are explicitly represented through mechanistic equations. In contrast, ML models reproduce site-level observations more closely but exhibit muted spatial and temporal variability due to their limited environmental sensitivities from sparse and discrete observations used for model training. Atmospheric inversions further indicate that incorporating the larger soil sink improves agreement with observed atmospheric CH₄ and its stable carbon isotope changes and requires larger microbial CH₄ emissions. Together, these results suggest that the global methane soil sink may have been underestimated and demonstrate the value of integrating PB and ML modeling, and atmospheric constraints to improve understanding of global methane cycling.

Link: https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025JG009668