Soil carbon MRV

Measurement, reporting and verification of soil carbon is the hardest subject in this cluster, and the difficulty is entirely metrological. The stock of organic carbon in the top 30 cm of a typical arable field is tens of tonnes of carbon per hectare. The annual change under carbon-building practices is a fraction of a tonne. The task, then, is to detect a change of order one percent in a quantity that varies routinely by tens of percent from point to point within the same field.

Why it takes either many samples or many years

Everything else follows from that ratio. The minimum detectable difference scales with spatial variability and falls as the square root of the number of samples. At the coefficient of variation typical of within-field carbon, detecting a change of a few percent at a defensible confidence level requires either a very large number of cores or a re-sampling interval of five to ten years, over which the signal accumulates above the noise. Sampling the same georeferenced points rather than a fresh random set removes much of the spatial component, but none of the temporal one: differences in moisture and sampling date propagate into bulk density.

Where the gain actually comes from

A stock is not a concentration. It is carbon content multiplied by bulk density and by depth. Bulk density is itself changed by the practices: no-till compacts the surface horizon, loosening decompacts it. Sample to a fixed depth and the two cores contain different masses of soil, so part of the apparent “gain” is simply more soil in the same volume. The equivalent-soil-mass correction, introduced by Ellert and Bettany in 1995, compares equal masses rather than equal depths, and applying it shrinks or removes a substantial share of published gains. This is not a protocol footnote; it is where a non-existent result is most often born — together with the 30-centimetre depth limit which, as the carbon farming page notes, mistakes redistribution of carbon towards the surface for accumulation of it.

Spectroscopy and models are predictions

Hence the attempts to replace costly laboratory analysis with in-field sensing. Infrared spectroscopy — near-infrared and especially mid-infrared — gives a fast estimate of carbon content, but it does not measure it: it predicts the laboratory result through a chemometric calibration built on real cores. Accuracy depends on whether that soil type is represented in the training set, and field conditions add moisture and surface-structure interference on top.

Remote sensing is bounded harder still: an optical sensor sees the top few millimetres, and only on bare soil, outside the growing season and clear of residue. Process models of the soil carbon cycle are calibrated on long-term experimental sites and then extrapolated to fields where no such experiment was ever run.

Neither is a measurement, and a sound protocol says so: it combines sparse coring with a model or a spectral map, states the residual uncertainty and deducts it from the credited amount. It is the size of that deduction, rather than the central estimate, that decides whether there is anything to transact.