A carbon removal credit is a promise that one tonne of CO₂ has been taken out of the atmosphere and stored durably. A buyer cannot check that promise by looking, so the industry has a three-part discipline for backing it up. MRV stands for measurement (collecting the data), reporting (writing up the result by an agreed set of rules) and verification (an independent party checking both). Carbon removal 101 explains why credits need this; this module is about how it is done for enhanced rock weathering (ERW).

Compare two kinds of removal. In direct air capture (DAC), fans push air over a chemical filter, and the captured CO₂ leaves the plant as a concentrated gas in a pipe. You can put a flow meter on that pipe, measure the gas going into a storage well, and count tonnes almost the way a petrol pump counts litres. ERW has no pipe. The "reactor" is a farm field, and the product is invisible ions dissolved in soil water that trickles away in every direction.

Five things make ERW difficult to measure, and every tool below is a response to one or more of them.

• Water moves through. The dissolved products don't wait to be counted. They drain downwards and sideways, so you have to catch them on the way or infer them afterwards.
• The signal is small against a large background. Soil already contains calcium and magnesium, and already releases and absorbs CO₂. The background (what the soil does with no rock added) is often far larger than the change you are trying to see.
• Soils vary enormously from spot to spot. Two cores taken a few metres apart can differ in chemistry by tens of percent, so any single sample is a poor guide to the field.
• It is slow. Basalt grains dissolve over years, so the signal builds up gradually and has to be tracked across several seasons.
• The carbon changes hands several times. Between rock grain and ocean it passes through soil water, clay surfaces, groundwater and rivers, and some of it can leak back out along the way, as described in The chemistry of enhanced weathering.

Put the first two numbers together and the problem is clear. You are looking for the partial disappearance of a 1% ingredient in a material whose natural composition wobbles by tens of percent. That is closer to a mineral explorer's problem, finding a faint signal against a noisy background (see Prospecting & the exploration pipeline), than to reading a meter.

Because the carbon moves, the measurements are arranged along its path. The diagram below follows it from rock grain to ocean and shows which tool samples each stage. No single tool sees the whole journey, so serious projects combine several and check them against each other.

1 · Soil-based mass balance. The most direct question is: how much of the rock we spread has dissolved? One way to answer it is to go back and measure what is left. A mass balance is simple bookkeeping: what went in, minus what is still there, equals what has left.

The trick is to use two kinds of element. Rock-forming elements split into two groups when rock weathers. Mobile elements, such as the cations calcium (Ca), magnesium (Mg), sodium (Na) and potassium (K), dissolve readily and are carried away in water. Immobile elements, such as titanium (Ti), zirconium (Zr), thorium (Th) and some rare earth elements, end up in minerals that barely dissolve at all, so they stay where the rock was put. (The same chemistry leaves laterites rich in iron and aluminium, as you saw in Weathering, erosion & soil.)

So measure the ratio of a mobile element to an immobile one, say Ca to Ti, in the rock before you spread it. That ratio is the rock's fingerprint. Later, measure the ratio again in what remains. Titanium hasn't gone anywhere, so if the ratio has dropped, the missing calcium must have dissolved. Using a ratio rather than a plain concentration has a valuable side effect: if rain or ploughing physically carries some rock grains off the field, Ca and Ti leave together, the ratio doesn't change, and you don't mistake erosion for weathering.

In a real field there is one complication: you can't pick the rock grains back out of the soil. A sample is a mixture of soil and rock, and the soil has its own calcium and titanium. So you need the composition of three things, the original soil (a baseline, sampled before spreading), the rock, and the mixture later, and you use the titanium to work out how much rock is in the mixture.

That last weakness matters. A mass balance tells you cations left the rock. It does not tell you where they went. They might have been released by a strong acid such as nitric acid from fertiliser, in which case no CO₂ was captured. They might be sitting on clay surfaces, held by the soil's cation exchange capacity (its ability to hold positively charged ions loosely on clay and organic matter), or taken up by the crop. So a mass balance is best read as an upper limit on CO₂ removal, to be paired with measurements of the water.

The next three tools follow the dissolved products as they move.

5 · Models. Measurements are expensive, patchy and slow, and nobody can sample the ocean end of the chain at all. Models fill the gaps. A dissolution-rate model predicts how fast grains of a given mineral, size and surface area dissolve under a given temperature and soil chemistry. A reactive-transport model goes further: it simulates water moving down through the soil, the reactions along the way, cations swapping on and off clay, and what reaches the drains.

Geophysicists face the same kind of problem when they image the ground beneath their feet, and Inversion & uncertainty explores it in depth. Running a model forwards from assumed rock, soil and weather to predicted porewater chemistry is a forward problem. Using measured porewater and soil data to work out the dissolution rate is an inverse problem, and it has the same pitfalls: several combinations of rate, water flow and cation retention can fit the same data equally well. A model is only as trustworthy as the measurements it is calibrated against (tuned so that its predictions match observations), and its output should carry uncertainty bands, not a single line.

Suppose drainage water from a treated field shows rising alkalinity this year. Was it the rock? Maybe. Or it was a wet year that flushed more water through, or the farmer limed, or the crop changed. On its own, a change over time can't tell you. Experimental design is how you rule those explanations out.

Put the time and space comparisons together and you get before-after-control-impact (BACI), a standard design in environmental science. You measure treated (impact) and control plots both before and after. The rock's effect is the change in the treated plots minus the change in the controls. Step through an example below.

Sampling has its own design choices. Soil is usually taken as composite samples: several cores from across a plot mixed into one bag, which averages out small-scale patchiness and saves lab costs. Sampling the same marked spots each time (a paired before-and-after design) removes much of the spot-to-spot difference from the comparison. And every lab batch carries quality-control samples: blanks that should contain nothing, certified standards of known composition, and duplicates. Mineral explorers call this QA/QC (see Prospecting & the exploration pipeline). It matters because a mass balance hunting for a change of a few percent is ruined by a lab that drifts by a few percent.

Every soil sample gives its own estimate of how much rock has dissolved, and because soils vary, those estimates scatter widely around the truth. A few samples might say 60%, a few might say less than zero (noise can do that). The spread is measured by the standard deviation (σ, "sigma"), roughly the typical distance of a single sample from the average.

The average of many samples is much more reliable than any one of them, and statistics says exactly how much more. The uncertainty in the average is the standard error:

where n is the number of samples. The square root is the painful part. Four times the samples only halves the uncertainty; a hundred times the samples cuts it tenfold. Precision gets expensive fast.

From the standard error you build a 95% confidence interval: the mean plus or minus about two standard errors (1.96, for large samples). In plain words, it is a range built by a method that captures the true value 95 times out of 100. It is not a guarantee about any single campaign: about one interval in twenty will miss. With few samples the multiplier is a bit larger than 1.96 (it comes from Student's t distribution), because the standard deviation itself is then only roughly known.

So how much can you claim? The usual answer is conservative crediting: claim a value you are confident is not an overestimate. A common choice is the lower end of the confidence interval rather than the middle. Then apply further discounts for losses you know happen but can't measure well, such as carbon that may escape further downstream. A buyer who receives conservatively counted tonnes is protected even when the uncertainty is large. Try it below.

Measuring is only a third of MRV. The results must be reported in a way others can check, and then checked.

The rules come from registries: organisations that set standards for carbon credits, keep a public record of each credit issued, and make sure no tonne is sold twice. A registry's rulebook for a particular removal method is called a methodology (or protocol). As of 2026, Puro.earth and Isometric are the main registries with published enhanced-weathering methodologies that are issuing credits; Verra has paused its draft, and the EU's certification scheme (CRCF) has yet to adopt one. They differ in detail and are revised as the science improves, so always read the current version. But they tend to cover the same ground:

Put the whole module together and you get a credit funnel: the chain of deductions between what the soil says has dissolved and what can actually be sold. Each step can only take tonnes away. It also links the course's ERW modules: the losses come from the chemistry, the uncertainty discount from the sampling above, and the emissions from From quarry to field.

A few ideas recur across methodologies, described here generically:

• Independent verification. A third-party verifier, with no stake in the result, audits the project's data, lab results, calculations and field practice before credits are issued.
• Measured, not promised. Many approaches credit removal only after measurements show it has happened (ex-post crediting), not on the strength of a forecast. Because weathering takes years, a field may be re-measured and credited in several instalments over a crediting period.
• Buffers and discounts. Some credits are deducted for uncertain losses, or set aside in a buffer pool, a shared reserve that can cover any credits later found to have been over-issued.
• Net, not gross. The emissions from mining, grinding and moving the rock, covered in From quarry to field, are subtracted from the removal.
• Transparency. Raw data, methods and uncertainties are documented so that a sceptic, not just a friendly auditor, could redo the sums.

Mining learned this lesson the hard way. In 1997 the Bre-X scandal, in which a company faked gold results by adding gold to its rock samples, led the industry to tighten its reporting rules. Today a mineral resource estimate is signed off by a qualified, personally accountable expert (a Competent (or Qualified) Person) and audited, because investors can't inspect the rock themselves. A carbon credit is verified for the same reason: buyers can't inspect the field. Mining also grades its estimates by confidence, and those classes are a useful way to think about how much confidence a removal number deserves. This is an analogy for learning, not an official ERW classification.

The underlying mindset is the one geophysicists bring to inversion (see Inversion & uncertainty): data are noisy, many models can fit them, and an honest answer comes with its uncertainty attached. In ERW the uncertainty isn't a footnote. Through conservative crediting it directly sets how many tonnes can be sold.

ERW measurement is young, and the methods are still being standardised. Researchers, companies and registries are actively working on questions such as:

• How much of the cation release in a field turns into durable carbon removal, and how much is held in the soil or lost on the way to the ocean?
• How quickly do cations held on clay surfaces move on, and how should that delay be credited?
• Which combination of soil, water and modelling evidence gives the best accuracy for the cost, and how many samples are enough?
• Can models calibrated on a few intensively measured sites be trusted for many lightly measured ones?
• Can new tracers, such as isotopes, or cheaper sensors cut the noise?

These sit alongside the wider questions in Open problems in Earth science. Some of the answers will come from people who are new to the field today.