How Much Soil Carbon Can a Hectare Gain Per Year? A Benchmark for Removal Rates

A number arrives in an email. This project sequesters 1 tonne of CO₂ equivalent per hectare per year. Or 0.2. Or 5. The person sending it wants to know whether that is good, and usually the number itself is perfectly sound. What it lacks is the context that gives it meaning. Soil carbon rates are small changes to a very large stock, so the same absolute figure can be an excellent result on one field and an ordinary one on the next.

What makes a rate interpretable is five pieces of context: the unit, the stock it is a fraction of, the depth it was measured over, the control it was compared against, and how long the clock has been running. Attach those and the number becomes checkable. This article supplies the benchmark, so that a rate can be placed against what the literature reports rather than against intuition.

The short version is encouraging. These practices do build soil carbon, and the ranges below are real gains measured in field experiments, some of which have run for more than a century. They are smaller and slower than the figures in general circulation, and they have to be described precisely to be worth anything. But the effect is well established, and the co-benefits for soil function are larger and more certain than the carbon.

It is the soil companion to our cross-system benchmark for tree carbon. Trees and soils fail differently: a tree rate is misread mostly because of stand age and density, while a soil rate is misread mostly because of the stock, the depth, and the comparison. The discipline is the same in both cases. Find the envelope first, then ask where in it your number sits, and why.

A rate only means something as a fraction of a stock

Soil is the largest terrestrial carbon pool. The top metre holds somewhere between 1,500 and 2,400 gigatonnes of carbon globally, and agriculture has already released on the order of 140 to 150 gigatonnes of it2. A spatial reconstruction of twelve thousand years of land use puts the debt at 116 gigatonnes of carbon over the top two metres, split roughly evenly between cropland and grazing land, with the loss rate rising sharply over the last two centuries21. That is the scale of the hole. It is also the reason the annual numbers are so small: a field is being asked to refill a very large container with a very thin stream.

In practical terms, a cultivated topsoil to 30 cm typically holds somewhere between 30 and 120 tonnes of carbon per hectare. Against a stock of 55 tonnes, a gain of one tonne of CO₂ per hectare per year is 0.27 tonnes of carbon, or about 5 parts per thousand of the stock. Against a stock of 120 tonnes it is 2 parts per thousand. Same field measurement, same effort, and a factor of two difference in how impressive it is, purely because of what was there to begin with.

Numbers like these read differently on a laboratory report than in a spreadsheet. A hectare of topsoil to 30 cm, at a bulk density of 1.3, weighs about 3,900 tonnes. A stock of 55 tonnes of carbon in it is a soil test reading of 1.41 percent carbon. Add a good year of 1 tonne of CO₂ and the reading becomes 1.42 percent. That is the whole of a successful year: the second decimal place of a soil test. Run the practice for a decade and you move from 1.41 to 1.48 percent, which is finally a change you can see against the variability of a field.

Almost everything difficult about this subject follows from that one fact. It is why the measurement is expensive, why honest researchers reach different conclusions from the same trials, and why a rate quoted after two seasons is a projection rather than an observation. It is also why the practices are worth doing anyway: the agronomic effects of that extra organic matter show up long before the carbon accounting can resolve it.

The unit itself is the first place claims go wrong. Carbon and carbon dioxide differ by a factor of 3.67, so a rate quoted as “one tonne” without a unit is ambiguous by nearly four-fold. Most soil science reports tonnes of carbon; most carbon markets report tonnes of CO₂ equivalent. The tool below converts between them, expresses the rate as a fraction of whatever stock you set, and places it on a scale whose midpoint is the “4 per 1000” aspiration.

Four per mille is the yardstick, and it costs about double the carbon input

The most useful single reference point in this field is the “4 per 1000” target launched at COP21: an aspiration to grow soil organic carbon stocks by 0.4 percent a year. As a global accounting device it has been criticised, but as a yardstick for a single field it is excellent, because it is expressed as a fraction rather than an absolute. A survey across twenty regions concluded that best management practice can reach 4 per mille and beyond, with rates up to 10 per mille achievable where the topsoil stock starts below 30 tonnes of carbon per hectare, and mostly during the first twenty years after a practice changes1. Soils already at equilibrium cannot add more.

What that target costs in practice is the part usually left out. Sixteen long-term experiments at Rothamsted, spanning 114 treatment comparisons over 7 to 157 years, found that 4 per mille was met or exceeded in most cases, but chiefly under treatments no working farm would sustain. Farmyard manure at 35 tonnes of fresh material per hectare every year, roughly 3.2 tonnes of organic carbon, produced increases of 18 and 43 per mille a year over the first two decades. Applied at lower or intermittent rates it gave 3 to 8 per mille. The authors concluded that there are severe limitations to achieving the goal across ordinary agriculture, for reasons that are practical rather than biological: there is not enough manure, the easy practices are already adopted, and the effective ones are uneconomic or compete with food production14.

Modelling the same question from the input side gives a consistent answer. Across sixteen European long-term experiments, an ensemble of six soil carbon models found that raising stocks by 4 per mille a year for thirty years required an additional 1.5 tonnes of carbon input per hectare per year, an increase of about 119 percent over the control15. Roughly speaking, hitting the target means doubling what the soil receives. Any claimed rate should be accompanied by a plausible account of where that extra biomass came from.

Most practices land between 0.1 and 0.8 tonnes of carbon per hectare per year

With the yardstick set, the practice-level numbers stop looking arbitrary. The table below gathers the central estimates from the syntheses this article draws on, in both units. Every figure is a topsoil rate relative to a stated control, not a project figure, and the spread within each row is usually wider than the difference between rows.

The clearest single source for European cropland is a review of more than 700 records across twelve carbon-farming practices, which reports topsoil rates of 0.40 for cover crops, 0.52 for organic amendments, 0.14 for residue retention, 0.21 for improved rotations, 0.24 for reduced soil disturbance, 0.21 for silvoarable systems and 0.79 tonnes of carbon per hectare per year for converting cropland to permanent grassland6. Grasslands have their own synthesis: improved grazing, fertilisation, sown legumes, irrigation and conversion from cultivation raise soil carbon at rates from 0.105 to somewhat over 1 tonne of carbon per hectare per year7, and at global scale the achievable grassland potential is put at 2.3 to 7.3 gigatonnes of CO₂ equivalent a year for biodiversity restoration against only 148 to 699 megatonnes for improved grazing management8.

Two entries in the table deserve their scepticism up front. Temperate agroforestry averages 0.21 tonnes of carbon per hectare per year in the topsoil across 61 observations, but with a standard deviation of 0.79, which is to say the between-study variation is nearly four times the mean and silvopastoral systems showed a slight loss9. Organic farming reports 0.45 tonnes of carbon per hectare per year across 74 studies, but when the analysis is restricted to zero-net-input systems with measured bulk densities the difference collapses to 0.07 and stops being statistically significant10. In both cases the headline number is real and the interpretation is not what it appears.

A worked example: the cover-crop disagreement

Cover crops are the most instructive case in the whole field, because the disagreement is public, recent, and entirely about method. The widely cited meta-analysis compiled 139 plots at 37 sites and found soil carbon rising at 0.32 plus or minus 0.08 tonnes of carbon per hectare per year at a mean depth of 22 cm, over periods up to 54 years3. That figure has been carried into policy documents and project designs for a decade.

A 2023 reanalysis screened the same body of field studies and found that all 37 had sampled to 30 cm or less, and none had compared treatments on an equivalent soil mass basis. After excluding trials that reported concentrations rather than stocks, used inappropriate controls, ran three years or less, or used cover crops as cash crops, six trials remained. Of those, four showed non-significant trends, one was negative and one was positive, giving a mean of 0.03 tonnes of carbon per hectare per year4. Ten times smaller than the accepted figure, from the same literature.

The original authors published a rebuttal arguing that the screening was too severe and discarded valid evidence5, and the honest position today is that the true value sits somewhere in that range and depends heavily on context. What matters for anyone reading a project document is the mechanism of the disagreement. Nobody fabricated anything. The gap came entirely from sampling depth, stock accounting method, and choice of control. Those three choices are worth more than the practice itself.

Depth, soil mass and control explain most of the disagreement between papers

Sampling depth is the largest lever. Around 55 percent of the carbon in the top metre sits below 30 cm13, and practices that concentrate residues at the surface move carbon upward as much as they add it. A systematic review of 101 long-term tillage trials found the benefit of reduced disturbance confined to the top 30 cm, and the European review found that assessing reduced disturbance over 0 to 50 cm cancelled the topsoil gain entirely, leaving no net climate benefit166. A shallow sample does not measure a smaller effect. It measures a different quantity.

The second lever is how the stock is computed. Management changes bulk density, so a fixed-depth sample compares different masses of soil before and after. Correcting to an equivalent soil mass removes that artefact, and the correction is not small: ignoring it overstated the gain from reduced tillage by 15 to 47 percent depending on the comparison16. The bias does not always run the same way. In tropical land-use change, failing to correct for soil mass understated the effect by 28 percent17. Fixed-depth accounting is not conservative, it is simply wrong in a direction you cannot predict without doing the correction.

The third lever is the control, and it is the one most often chosen for convenience. A practice compared against bare fallow will look transformative; the same practice compared against ordinary local management will look modest; compared against the grassland that preceded it, it may look like a loss. The European review makes the point starkly: set-aside scores 0.75 tonnes of carbon per hectare per year against a cropland control and −0.39 against a grassland one, and poplar plantations score 0.25 on cropland and −0.85 on grassland6. The sign of the answer is set by the comparison, not the practice.

The tool below holds a single hypothetical field fixed and varies only those three choices. Nothing about the soil changes. The reported rate moves by more than a factor of ten, and it can change sign.

A gain below the detection floor is a design problem, not a small result

There is a fourth measurement question sitting behind the other three, and it is the one most often skipped: could this design have seen the change at all? Every monitoring campaign has a floor, the smallest difference it can separate from sampling noise, and that floor is fixed by how variable the field is, how many samples you take and how long you wait. Nothing about the practice enters into it. Soil carbon moves slowly against a large and noisy background, which is why ten-year resampling intervals are the usual compromise and why anything inside three to five years is considered challenging2.

The numbers are sobering. A field with a within-field standard deviation of 11 tonnes of carbon per hectare, sampled at thirty points, cannot resolve anything smaller than about 8 tonnes per hectare. A practice delivering 1 tonne of CO₂ per hectare per year adds 2.7 tonnes of carbon in a decade, so it stays invisible for roughly thirty years at that intensity. The 0.2 case at the top of this article is worse: it stays under the floor even with three hundred samples over thirty years.

So a low number needs reading carefully. A quoted rate of 0.2 tonnes of CO₂ per hectare per year is not simply a modest result; on most fields it sits inside the noise band of an ordinary design, which means a campaign reporting it as a measured gain has probably not measured anything. The honest description is not “a small increase” but “no detectable change, and here is the smallest change we could have detected.” The first invites a credit. The second invites a better design.

Accrual saturates, so the first year is never the thirty-year mean

Soil carbon does not accumulate linearly. Raise the input to a soil and it moves toward a new steady state, quickly at first and then ever more slowly, because the losses rise as the stock does. The cover-crop meta-analysis modelled its own data and found the new equilibrium reached after 155 years, with a total gain of 16.7 tonnes of carbon per hectare3. Almost all of the annual rate people quote comes from the steep early part of that curve.

There is a hard ceiling behind the curve. Mineral-associated organic carbon, the fraction that persists, totals 899 gigatonnes of carbon to one metre globally, yet sits at only 42 percent of its mineralogical capacity in surface layers and 21 percent deeper. Distance from that capacity predicts how efficiently a soil accrues: across 103 accrual measurements, soils at a tenth of their capacity gained carbon roughly three times faster than soils at half of it18. Later work argues that carbon input, rather than mineral surface, is the binding constraint in most soils, with maximum loadings falling from 160 grams per kilogram in coarse soils to 75 in fine ones19. Either way the direction is the same: the emptier the soil, the faster it fills, and the fuller it gets the slower it goes. That capacity question is large enough to deserve its own treatment, and gets one in soil carbon saturation, which works through what sets a soil's ceiling and how close to it a given field already sits.

This is not a technicality for anyone selling or buying a thirty-year removal profile. Accounting that ignores saturation overstates the contribution of soil carbon to mitigation by somewhere between 53 and 81 percent out to 210020. The tool below runs the standard first-order approach to a new steady state. Set a first-year rate and a response speed, and watch the annual rate diverge from the running mean.

Two readings matter. The front-loading ratio is how many times larger the first year is than the thirty-year mean, and it is the size of the error someone makes by quoting an early measurement as a durable rate. The share of the total gain delivered by year twenty shows how much of the opportunity a normal crediting period actually captures. In most plausible settings the answer is that the early years carry the result and the later years are close to flat, which is precisely why a single figure repeated across a project lifetime is the wrong shape of claim.

The largest per-hectare numbers are avoided losses, not gains

Everything above concerns building new carbon in mineral soil, where the honest envelope tops out somewhere near one tonne of carbon per hectare per year and usually sits well below it. Step across to organic soils and the arithmetic changes completely. Croplands on drained lowland peat emit 23.1 plus or minus 10.4 tonnes of CO₂ per hectare per year23, which is twenty times the best accrual rate available on a mineral soil next door. Water table depth overrides every other control on those fluxes, and each 10 cm of reduction in effective water table depth cuts the net warming effect by at least 3 tonnes of CO₂ per hectare per year until the water table is within 30 cm of the surface22.

This is why the global accounting puts protection alongside accrual. Soil carbon is about a quarter of the mitigation potential of natural climate solutions, and 40 percent of that quarter is protecting stocks that already exist rather than rebuilding depleted ones. Soil is 72 percent of the wetland potential against 9 percent of the forest potential24. The lesson for a portfolio is blunt: if there are organic soils anywhere in the footprint, they dominate, and no amount of cover cropping on the mineral fields will offset a drained peat field left alone.

Land-use change is the other place large numbers live, in both directions. Converting lowland tropical forest to oil palm, rubber or cacao plantations loses up to half the stored soil carbon, enough that the authors proposed revising the IPCC Tier 1 change factor from 1.0 to 0.6 for oil palm and cacao agroforestry33. Conversion of primary forest to cropland costs 25 percent of the stock, and to perennial crops 30 percent17. Reversals are slow. In German agricultural soils, cropland converted to grassland took 83 years to reach its new equilibrium, and grassland converted to cropland took 180 years, with fields today still gaining or losing around 0.26 tonnes of carbon per hectare per year because of decisions taken a century ago32.

Four deductions stand between a gain and a claim

A soil carbon gain becomes a climate benefit once four deductions have been made. Most well-run projects survive them comfortably. The value of making them explicit is that you find out early, rather than at audit, and can design around them.

The first is relocation. Organic amendments produce some of the highest recorded rates, up to 5.3 tonnes of carbon per hectare per year in olive orchards receiving mill pomace, with apparent sequestration efficiencies above 100 percent12. An efficiency above 100 percent is a warning, not an achievement: it means more carbon appeared than was applied, which usually indicates that carbon was moved from somewhere else and counted here. Manure and compost that would otherwise have been applied to a neighbouring field is a transfer, not a removal, and the Rothamsted authors say so plainly14.

The second is nitrous oxide. Most practices that add carbon also add nitrogen, and N₂O is roughly 273 times more potent per tonne. The offset is significant enough that the greenhouse gas benefit of soil carbon storage cannot be assessed on carbon alone28. Mineral fertilisation is the extreme case, with no distinct effect on topsoil carbon and a real emissions penalty11.

The third is reversibility. Soil carbon that took twenty years to accumulate can be released in a season of ploughing. The scientific and policy vocabularies around this diverge badly, with policy treating soil carbon as uniformly vulnerable while the science describes a continuum of residence times29, and a review of private soil carbon certificates concluded that permanence cannot currently be guaranteed, with unresolved problems around additionality, leakage and long-term accountability30. None of that makes soil carbon worthless. It makes a non-permanent removal a different product from a permanent one, and it should be priced and described as such.

The fourth is baseline drift, met earlier and the easiest of the four to forget, because it is invisible in the field. A soil still responding to a land-use change made generations ago is gaining or losing carbon on its own, at rates around 0.26 tonnes per hectare per year in the German inventory data32. Measured against zero, that drift is credited to whoever happens to be farming now. Measured against a control that shares the same history, it disappears, which is the whole reason the control matters.

Set against these deductions, the global estimates settle into a narrow band. Technical potential for soil carbon sequestration is put at 2 to 5 gigatonnes of CO₂ a year26, with 4 to 5 as an upper limit under near-complete adoption of best practice25, and a realistic cropland-only figure of 0.28 to 0.43 gigatonnes of carbon a year once manure availability and adoptable area are taken into account27. Overstating those numbers has a cost of its own: inflated estimates give cover for delaying emissions reductions31, and the case for building soil carbon is strong enough on soil-function grounds that it does not need inflating34.

That last point deserves more weight than it usually gets. Most of the practices in this benchmark improve water holding capacity, aggregate stability and nutrient supply, and those returns arrive whether or not a tonne is ever sold. They are also easier to demonstrate. Yield stability and workability show up within a few seasons, while a stock change takes a decade to resolve against its own noise. A programme built on agronomic value, with carbon as a co-benefit, survives contact with a difficult season better than one built the other way round. It is the version we usually recommend.

So: is one tonne a lot?

One tonne of CO₂ equivalent per hectare per year is 0.27 tonnes of carbon. On a typical arable topsoil holding 55 tonnes of carbon per hectare, that is about 5 per mille: above the 4 per 1000 aspiration, comfortably inside the envelope for cover crops, improved rotations or organic amendments, and a good result if it survives a whole-profile, equivalent-soil-mass measurement against a like-for-like control. It is a plausible number, and it is at the better end of plausible.

0.2 tonnes of CO₂ per hectare per year is 0.055 tonnes of carbon, or about 1 per mille of the same stock. That is a real number in the sense that many practices deliver it, and an almost undetectable one in the sense that you would need decades of careful resampling to distinguish it from zero. It is not wrong. It is below the resolution of most monitoring programmes, which is a different problem and one worth naming.

5 tonnes of CO₂ per hectare per year is 1.36 tonnes of carbon, roughly 25 per mille. That exceeds the mean of every mineral-soil meta-analysis cited here, though it is entirely achievable, and four situations routinely account for it: the soil started severely depleted, the system received heavy organic amendments from off-site, a land-use change occurred, or the measurement is on an organic soil. Any of those makes the number credible. If none of them applies, the usual explanations are a unit slip between carbon and CO₂, a shallow sample, or a first-year rate quoted as a durable one.

The checks below compress the article into something portable. They are not a way of catching people out. Most of the time they confirm a number quickly, and the value is in knowing which of the five pieces of context you are still missing.

The deliverable a client should want is not a figure but a bracketed range with its provenance attached: the practice, the baseline stock, the depth, the method, the control, the duration, and what has been netted out. That is harder to put on a slide than “one tonne a year,” and it is the version that survives due diligence. If you are building that range from a model rather than from measurements, our piece on modelling soil carbon change covers what the models can and cannot carry.

Pontos-chave