Most projects respond to a wide confidence interval by adding plots. Halving the interval that way requires roughly four times the plots — and four times the field budget. Stratification often achieves the same narrowing for the cost of an afternoon with a map.
Why It Works
Sampling error depends on variance within what you are sampling. A project containing both dense mature forest and sparse young regrowth has enormous total variance, because the two conditions are genuinely different.
Split them into separate strata and sample each independently, and each stratum's internal variance is much smaller. The combined estimate inherits the smaller within-stratum variances rather than the large overall one.
| Approach | What varies | Plots needed for a given precision |
|---|---|---|
| Simple random over whole project | Everything | High |
| Stratified by condition | Within-condition only | Substantially lower |
| Stratified by an irrelevant variable | Everything, still | No improvement |
That last row is the trap. Stratification only helps if the stratifying variable actually predicts carbon density. Splitting by administrative boundary, ownership, or planting block produces strata that differ on paper and not in biomass — the variance stays where it was.
What Actually Predicts Biomass
In practice: stand age or planting year, species composition, site quality, and prior land use. Remote-sensing proxies — canopy height, a vegetation index — work well as a stratifying layer precisely because they correlate with the thing being measured.
The test is simple. After stratifying, look at the within-stratum standard deviations. If they are close to the overall standard deviation, the stratification is decorative.
Allocating Plots Between Strata
Once strata exist, plots must be distributed. Three options, in increasing sophistication:
Equal allocation — the same number per stratum. Simple, and wasteful when strata differ greatly in area.
Proportional allocation — plots in proportion to stratum area. Reasonable default.
Neyman allocation — plots in proportion to area times within-stratum standard deviation. Optimal, because it concentrates effort where the variance is. A small but highly variable stratum gets more plots than its area alone would justify.
Neyman allocation requires a variance estimate before you sample, which usually means a pilot survey or prior data. Where those exist, it is the most efficient use of a fixed plot budget available.
The Mistake That Undoes the Gain
Strata must be fixed before sampling and applied consistently across monitoring events.
Re-stratifying between visits because the vegetation changed sounds sensible and destroys the time series: you are now comparing different populations. If strata must change — and in a growing ARR project they eventually must — the correct move is to recalculate earlier monitoring events under the new stratification, so the comparison remains internally consistent.
Projects that quietly re-stratify and report the difference as growth are reporting a definitional change as a carbon change. That is a finding waiting to happen.
What to Document
Stratum definitions, the evidence that they predict biomass, the allocation rule, and the within-stratum variances actually observed. An assessor who can see those four things rarely questions the design. One who sees only a map with coloured polygons usually does.
Turn uncertainty into a plan
Sampling design and growth assumptions decide how much measured carbon survives conservative deduction. TREEO Carbon Simulator models sequestration scenarios and biomass growth before the field season.



