The microbes underfoot that quietly eat methane
On a still morning after rain, a patch of brittle desert crust softens and wakes. Invisible to the eye, microbes stir in the top few centimeters of soil, drawing in air that holds only a few parts per million of methane. They live on the trace—a diet so lean it once seemed negligible to global accounting. A new analysis argues it is anything but.
Methane is a greenhouse gas roughly 27–30 times more potent than carbon dioxide over a 100-year span. For decades, scientists have tried to close the books on where it comes from and where it goes. A team led by Youmi Oh now says one of the biggest lines in that ledger—the amount of methane that soil-dwelling, methane-munching microbes remove from the air—has been underestimated.
A long-standing mismatch in the methane budget
There are two major ways to estimate methane’s global comings and goings. Bottom-up, process-based models simulate how ecosystems behave—wetlands bubbling methane up, soils and microbes drawing it down—by representing the underlying biogeochemistry. Top-down, atmospheric inversion models start with observed methane in the air and work backward to the sources and sinks that would best reproduce those concentrations.
These two approaches have not agreed. In particular, bottom-up estimates for biological methane sources from wetlands and inland fresh waters have tended to run higher than top-down, atmosphere-based reconstructions. One way to bring those ledgers closer is to recognize a larger sink—the amount soils take out of the air.
The study that adds a third voice
Published in the Journal of Geophysical Research: Biogeosciences, the new work by Oh and colleagues runs three kinds of models side by side. They keep the familiar pair—process-based ecosystem simulations and atmospheric inversions—but add a third, data-driven machine learning model. The idea is simple: when independent approaches converge, confidence grows and uncertainties can shrink.
They also did some housekeeping inside the process model itself. The team tweaked how microbial dynamics are represented and expanded the map to include places and microbes that earlier studies had left out. Then they compared what all three methods said about the size of the soil methane sink and fed the revised sink into the atmospheric inversions to see if the atmosphere’s patterns and isotopic fingerprints were better reproduced.
How much methane do soils eat? More than the books showed.
Previous estimates suggested soil methanotrophs—organisms that biologically remove methane from air—might take up an average of 28–35 gigatons of methane per year globally. Oh’s three-model analysis points higher: both their process-based and machine learning models produced similarly sized sinks of 40–45 gigatons per year.
That range is larger than in global climate assessments, including those of the Intergovernmental Panel on Climate Change. Crucially, when this larger sink was plugged into the top-down atmospheric inversions, the models better matched observed methane in the air as well as its stable carbon isotope composition.
Why this matters for the rest of the budget
The methane budget is a balance sheet. If the bottom-up side has been counting too little removal by soils, then it has been forced to count too much emission from wetlands and inland waters to explain the air’s methane. A larger soil sink relieves that pressure and narrows the gap with top-down estimates without inventing new sources or sinks.
This is not a story of a free pass on methane. Rather, it is a bookkeeping correction with real consequences. Better alignment between models and observations means more credible projections and clearer attribution of what is driving methane’s recent trends—which policies target, and which they do not.
The overlooked workforce beneath our feet
Soil hosts a riot of microbial life, much of it still being cataloged from Arctic soils to desert sands. Among these communities are methanotrophs that survive on the thinnest trickle of methane in ambient air. Earlier modeling left out some of these microbes and some of the places they live, not out of neglect but because the field lacked enough data to parameterize them.
By revising microbial dynamics and adding overlooked habitats, the new analysis increased the modeled appetite of this living sink. The result suggests that what seemed like marginal activity at the micro scale adds up to a substantial global service when you integrate across the world’s soils.
Testing the answer against the air itself
Any new estimate has to survive contact with the atmosphere. The authors report that incorporating the larger soil sink into the inversions made the models better at reproducing what is actually measured: methane concentrations and the stable carbon isotope composition of methane.
That dual improvement matters. Concentrations alone can be matched by many combinations of sources and sinks; getting the isotope composition right at the same time is a higher bar. In this study, the larger soil sink helped clear it.
The deeper cut
Inside the three-model cross-check
Process-based models track state variables (soil moisture, temperature, substrate availability) and fluxes via parameterized kinetics for microbial uptake, then scale from plot to globe. Their weaknesses are familiar: sparse field constraints, structural choices about which microbial guilds to include, and parameter values that are often tuned regionally. Atmospheric inversions, by contrast, assimilate observed methane and solve an optimization problem to infer the spatial–temporal pattern of sources and sinks that yields the best fit, subject to prior constraints and transport. They are powerful, but they smear signals where transport is uncertain and can accommodate compensating biases (an overestimated source can be “balanced” by an overestimated sink) if the priors allow it. The machine learning layer sidesteps explicit process representation, instead learning empirical mappings from environmental covariates to observed fluxes where data exist, then generalizing. Its Achilles’ heel is extrapolation: performance depends on the training domain covering the predictor space actually encountered. Running all three together exposes where each one flexes or fails. In Oh et al., the convergence of the process-based and ML sink magnitudes, and the improved inversion fit when that sink is imposed, is precisely the triangulation you want when no single method, on its own, can claim ground truth.
What’s changed, and what hasn’t
The big change is methodological: a three-pronged estimate that brings separate lines of evidence into alignment. The immediate upshot is an upward revision of the soil sink into the 40–45 gigaton-per-year range and a better match to atmospheric constraints, including isotope composition.
What has not changed is the basic shape of the methane problem. Methane still traps heat far more strongly than carbon dioxide over a century. Soils may be doing more work than credited, but that role slots into the global budget rather than erasing it. The study’s authors frame their result as an improvement to global carbon cycle modeling, not as a solution to rising methane.
| Approach | What it does | What changed in this study | Key outcome |
|---|---|---|---|
| Process-based modeling | Represents biogeochemical processes in soils and ecosystems | Tweaked microbial dynamics; included previously overlooked places and microbes | Yielded a soil sink in the 40–45 gigaton-per-year range |
| Machine learning | Data-driven mapping from observations and covariates | Added as a third, parallel line of evidence | Produced a similarly sized soil sink as the process-based model |
| Atmospheric inversion | Starts with atmospheric methane and infers sources/sinks | Incorporated the larger soil sink from the other models | Improved reproduction of observed methane and its stable carbon isotope composition |
The next questions researchers will ask
If soils are taking more methane out of the air than thought, where exactly is that happening, and how stable is it? The study broadened the roster of places and microbes represented, but it does not enumerate regional winners and losers. That gap points to field campaigns and monitoring aimed at pinning down spatial patterns and year-to-year variability.
Another question is how this larger sink interacts with changes in land use and climate. The present work does not resolve those sensitivities. But by tightening the overall budget and improving agreement between methods, it sharpens the baseline from which such effects will be measured.
A quiet correction with outsized consequences for models
Oh and colleagues do not claim a silver bullet for methane. They claim something humbler and more consequential for the science: that a living, overlooked sink is larger than the models used to assume, and that acknowledging it brings our reconstructions of the atmosphere into better focus.
In climate work, alignment across methods is hard won. Here it required adding a data-driven voice, revisiting microbial assumptions, and checking the answer against the air itself. For microbes that live on the thinnest fare imaginable, that is a loud result.
The paper: https://dx.doi.org/10.1029/2025jg009668
Sources: Soil methane sink may be larger than thought, three-model analysis finds (phys.org); Soil methane sink may be larger than thought, three-model analysis finds (phys.org); Farm mice study shows antibiotic resistance is an environmental problem, not just a medical one (phys.org)
Images: Cover: Everyman Science (illustration)
How this article was made: Everyman Science uses AI tools to structure, format and optimise its articles, and occasionally to produce illustrations where no free photograph exists. The reporting these articles are based on is human-produced and cited above. Spotted an error? Write to [email protected] and we will correct it. — The editors How we work.
Science desk team of Everyman Science, curating and reporting on the day’s most significant developments in research, space exploration, and technology.
