Measuring Soil Carbon Credibly

Soil organic carbon has become the currency of regenerative agriculture, carbon farming and a growing share of corporate climate claims. Every one of those claims rests on a number, and the number rests on a sampling and analysis chain that is far less precise than the confidence with which it is usually quoted.

The laboratory is rarely the problem. Modern dry combustion analysers measure carbon in a prepared sample with excellent repeatability. The uncertainty comes from what the sample represents: how many cores were taken, where, how deep, whether the soil's bulk density changed, and whether the same locations can be found again in five years. Those decisions are made in the field, and they set the ceiling on what the result can ever mean.

Why measurement is the weak link

Soil carbon is spatially variable at a scale that surprises people used to nutrient testing. Two cores ten metres apart in the same field can differ by a third, because of small changes in texture, drainage, past cropping and the position of old furrows or fence lines. That variability, not the instrument, dominates the error of a field estimate.

The practical consequence is that a change of a few per cent in soil carbon is very hard to detect against that background unless the sampling is designed for it. Programmes that sample a field in year one and year five with a new random layout each time are measuring the layout as much as the soil. Programmes that return to georeferenced points, or that use a paired comparison with a neighbouring control area, do far better.

Concentration, stock and depth

Two numbers are routinely confused. Carbon concentration is a percentage of the soil's mass, usually reported as grams of carbon per kilogram of dry soil. Carbon stock is an amount per unit area, in tonnes of carbon per hectare, and it is what markets and inventories actually need. Converting one to the other requires bulk density and depth.

The arithmetic is worth doing once. A hectare of topsoil thirty centimetres deep, at a bulk density of 1.3 grams per cubic centimetre, weighs about 3,900 tonnes. At a carbon concentration of 1.5 per cent, that hectare holds roughly 58 tonnes of organic carbon. A change of 0.1 percentage points is about 3.9 tonnes per hectare, which is agronomically meaningful and, at the prices commonly discussed for soil projects, a modest sum of money.

Depth matters for the same reason. Reporting a stock for a fixed depth assumes the soil surface has not moved, but ploughing, erosion, compaction and surface residue all change how much soil sits above thirty centimetres. If a management change makes the topsoil lighter and deeper, a fixed-depth sample will report a carbon gain that is partly a change in soil mass.

Sampling design decides the answer

A defensible design starts with stratification. The field is divided into units that are internally more uniform than the field as a whole, using soil type, texture, drainage, slope, cropping history or existing management zones. Samples are then allocated within each unit, either at fixed points on a grid or at points chosen to be representative and permanently marked.

The number of locations is set by the variability and by the size of the change the programme needs to detect. Detecting a small change requires far more samples than estimating a baseline, and this is where many projects quietly fail: the baseline is well characterised, the follow-up is not, and the difference between them cannot be distinguished from zero. Compositing several cores into one sample reduces laboratory cost but also hides the variability the statistics need, so a workable compromise is to composite within small homogeneous units and keep enough units to estimate the variance.

Bulk density and equivalent soil mass

Bulk density is the least glamorous measurement in soil carbon work and the one most often skipped. It changes with compaction, tillage, organic matter and the moisture at the time of sampling, and a core taken wet and compressed can read a third higher than the same soil sampled dry and loose.

The standard remedy is to express stocks on an equivalent soil mass basis rather than a fixed depth. The sampler records bulk density, computes the mass of soil in the layer, and the laboratory result is scaled to a common mass per hectare, often around 3,000 tonnes of fine earth. This adjusts for the fact that a compacted soil contains more mass in the same depth, and it stops a management change that merely alters density from being reported as a carbon change.

Laboratory methods

Dry combustion, the Dumas method, is the reference technique: the sample is burned at high temperature and the carbon dioxide released is measured. It quantifies total carbon, so on calcareous soils it includes inorganic carbon from carbonates, which has to be removed or corrected before the result can be called organic carbon.

Wet oxidation, the Walkley-Black procedure, uses dichromate and titrates what remains. It is cheap, but it recovers only part of the carbon, typically 70 to 85 per cent, so results are multiplied by a correction factor that is a convention rather than a measurement. Infrared and laser-induced breakdown spectroscopy predict carbon from a spectrum calibrated against conventional analysis; used well they cut cost and allow much denser sampling, and used carelessly they apply one region's calibration to another region's soils and produce confident numbers that are simply wrong.

How much change can be detected

Published reviews of soil carbon monitoring converge on an uncomfortable conclusion: across most agricultural systems the change over three to five years is small relative to the spatial variability of the field, and detecting it reliably requires many samples, a paired design, or a longer interval. Where the expected rate of change is a fraction of a tonne per hectare per year, a short programme will often be unable to separate signal from noise.

This does not make measurement pointless. It changes what measurement is for. A baseline with a stated uncertainty is useful; a five-year resample that reports a confident gain usually overstates what the data can support. Programmes that acknowledge uncertainty explicitly, and use models constrained by measured data rather than substituting for it, produce more defensible numbers than programmes that report a single figure to three decimal places.

Models and remote sensing

Process models that simulate carbon turnover are useful for extrapolating between measured points and testing which management changes are likely to matter. They are not a substitute for measurement: a model calibrated on another region's soils, climate and cropping system will produce a plausible number and an implausible level of confidence. Remote sensing contributes indirectly, describing how much biomass is produced and therefore how much residue might return to the soil, but it does not see soil carbon. Combining remote observation of management with sparse, well-designed ground sampling is the most credible structure currently available.

Common questions

How many samples do I need? It depends on the variability of the field and the size of the change you want to detect. A baseline needs fewer samples than a change detection programme. Calculate the number from a pilot set of samples rather than assuming a figure.

Can I use a sensor instead of laboratory analysis? For mapping variation within a field, yes, provided the sensor is calibrated against laboratory results from the same field. For reporting a stock or a change to a third party, the number should rest on conventional analysis of a properly designed sample set.

Does no-till always increase soil carbon? It often redistributes carbon towards the surface rather than increasing the total, and the effect depends on how much residue is returned, how the soil is drained and what was grown before. Studies that sample only the top few centimetres tend to overstate the gain.

Why do two labs give different results? Differences usually come from sample preparation, carbonate removal, the oxidation method and the correction factor, not from instrument error. Use one laboratory for a time series, and ask for the method in writing.

Recommendations depend on local soil, climate and crop conditions. Follow the product label and local regulations, and consult a qualified agronomist for field decisions.