Ask most people how to measure carbon in a forest and you get the same answer: count the trees. Measure all of them and there is nothing left to estimate. It sounds like the most rigorous thing a project could do.
It is also the approach that fails first, and it fails for reasons that are arithmetic rather than technological. A project area holds hundreds of thousands of stems. Nobody measures them all, because they measure the ones they can reach and then extrapolate. At that moment the census has quietly become a sample, but one with no design behind it and no confidence interval anyone can calculate.
Why the Census Idea Is So Appealing
The intuition is sound. Estimation introduces uncertainty; census eliminates it. If you know every tree, you know the carbon.
The arithmetic is less kind. A moderately sized project area holds hundreds of thousands to millions of stems. Measuring each one to the standard a verifier accepts, including species identification, diameter at a defined height, height or a validated height model, and condition, takes minutes per tree under good conditions and considerably longer in dense understorey or on steep ground.
Multiply that out and the cost of a single complete inventory exceeds the credit revenue of most projects. And a carbon project does not need one inventory. It needs repeated inventories across a crediting period that may run decades.
So the census is attempted once, or partially, and then extrapolated, at which point it has quietly become a sample, but an undesigned one.
The Problem Nobody Mentions: An Undesigned Sample
This is the part that matters for verification. A designed sample has known statistical properties: you can state the confidence interval, defend the sample size, and show that the sampled units represent the population.
A partial census cannot do any of those things. The trees that got measured are typically the accessible ones, being near roads, on flat ground, in open stands. The trees that did not are the ones in difficult terrain, which frequently differ systematically in size, species and density.
| Designed sample | Attempted census, partially completed | |
|---|---|---|
| Selection | Probabilistic, defined frame | Accessibility-driven |
| Confidence interval | Calculable | Not calculable |
| Bias | Quantifiable and correctable | Unknown direction and magnitude |
| Verifier treatment | Assessed on design | Discounted conservatively |
| Repeatability | Same frame, same plots | Rarely reproducible |
A verifier facing an estimate with no calculable uncertainty does not reject it outright. They apply the most conservative assumption available, and the deduction comes out of issuable tonnes.
So What Actually Drives the Error?
Forest carbon varies enormously across short distances. Species composition, stand age, soil depth, hydrology and disturbance history all drive biomass, and they vary at scales of tens of metres.
Counting more trees does not address heterogeneity. Sampling proportionate to heterogeneity does. This is what stratification means in practice: divide the area into strata that are internally more uniform than the whole, then sample within each. Variance within strata is lower, so fewer plots achieve the same precision.
The result is counterintuitive to anyone attached to the census idea: a well-stratified sample of a few hundred plots can produce a tighter confidence interval than an unstratified measurement of vastly more trees.
What Individual Tree Data Is Actually Good For
None of this makes tree-level measurement useless. It makes it the wrong unit for estimation.
Individual tree records are essential for building and validating allometric equations, for tracking permanent sample plots through time, for species-specific growth modelling, and for demonstrating measurement quality to an auditor. Those are calibration and assurance functions.
The estimate itself, which is the number that becomes tonnes, comes from the sample design, and its credibility comes from the statistical properties of that design.
What Replaces It — and Why It Costs Less
Multi-layer statistical sampling: strata defined by the variables that actually drive biomass variance, nested plots sized to each carbon pool, sample sizes derived from measured variance rather than convention, uncertainty quantified and propagated through every conversion, and remote sensing used to extend and monitor rather than to replace ground observation.
That is a harder system to build than a tree counter. It is also the one that survives a verification body's technical reviewer, because every number in it traces back to a defensible decision.
Image credit
Hero image: A dense tropical forest by Ibrahim Achiri, CC BY-SA 4.0, via Wikimedia Commons.
Sources
1. IPCC 2006 Guidelines Vol. 4 (AFOLU), sampling and uncertainty guidance — https://www.ipcc-nggip.iges.or.jp/public/2006gl/vol4.html (accessed 16 Sep 2026)
2. Kauffman & Donato (2012), CIFOR Working Paper 86 — nested plot protocols — https://www.cifor-icraf.org/knowledge/publication/3749/ (accessed 16 Sep 2026)
3. Verra VM0048 and methodology requirements — https://verra.org/methodologies-main/vm0048-vmd0055-faqs/ (accessed 16 Sep 2026)
4. ICVCM Core Carbon Principles — https://icvcm.org/core-carbon-principles/ (accessed 16 Sep 2026)
Move beyond estimates
Verifiers test sampling design, uncertainty and whether a number traces back to the field. TREEO dMRV captures that evidence in real time, in one auditable chain.
