Uncertainty and Confidence Deductions in Carbon Accounting

Updated 2026-11-02 · 4 min read · By the akar.earth specialist team

Uncertainty deduction carbon credits guide: how confidence intervals are calculated, when deductions apply, and how to reduce them with better design.

An uncertainty deduction reduces the credits a project can claim when its estimate of carbon change is too imprecise, usually judged by the half-width of a 90% or 95% confidence interval relative to the mean. The rule rewards good measurement: a well-designed inventory can avoid the deduction entirely, while a weak one can lose a meaningful share of its credits.

Why uncertainty is deducted

Carbon credits are claims of real tonnes. Every estimate has error from sampling, measurement, models and remote sensing. Integrity frameworks such as the ICVCM Core Carbon Principles expect conservative quantification, and most methodologies do this by:

1. Calculating the uncertainty of the net estimate.
2. Comparing it to an allowable threshold.
3. Deducting tonnes, often using the lower bound of the confidence interval or a formula linked to the excess uncertainty.

The exact formula varies. CDM ARR tools, Verra methodologies and jurisdictional standards each define their own. Read the uncertainty section of the methodology version you use; do not reuse rules from another project.

Sources of uncertainty

Sampling error is the part most projects calculate formally. Systematic errors are handled through conservative choices or explicit propagation, depending on the methodology.

How the calculation works

For a stratified inventory, the estimate of mean carbon stock is the area-weighted mean of stratum means. Its standard error combines the stratum variances and plot counts. Multiply by the t-value (or z-value) for your confidence level to get the confidence interval half-width, then express it as a percentage of the mean.

For change over time (the thing you credit), uncertainty combines the estimates at two dates. Permanent plots help here: because the same trees are measured twice, the correlation between dates reduces the uncertainty of the difference.

When combining several sources, use error propagation (for independent errors, roughly the square root of the sum of squared absolute errors for sums) or Monte Carlo simulation. Monte Carlo is clearer for complex chains with remote sensing models and is accepted under IPCC guidance.

A practical plan to control uncertainty

1. Pilot inventory. Estimate variance per stratum before the main campaign.
2. Size the sample to the threshold with a margin. See plot sampling design.
3. Stratify by what drives carbon, not by administrative boundaries.
4. Use permanent plots for change estimates.
5. Run QA/QC with independent check crews.
6. Validate models on independent data and keep their errors in the budget.
7. Recalculate before verification and add plots if needed, while there is still time.

Worked logic without the numbers

Suppose a stratified inventory gives a mean net removal and a 90% confidence half-width. If the half-width, as a share of the mean, is within the methodology threshold, no deduction applies. If it exceeds it, the methodology’s formula reduces the credited amount, for example toward the lower confidence bound. Adding plots in the stratum that contributes most variance usually shrinks the half-width fastest. Run this check during planning, not after the field season.

Common mistakes

• Reporting the uncertainty of stock at one date and treating it as the uncertainty of the change.
• Ignoring model error in remote sensing estimates.
• Combining percentages instead of absolute errors.
• Using a 95% interval where the methodology asks for 90%, or vice versa.
• Dropping outlier plots without a documented rule. Big trees are real.
• Discovering at verification that the precision target was missed.

What good evidence looks like

Verifiers want the uncertainty calculation to be reproducible. Provide the plot data, stratum areas, the formulas or code used, the confidence level, each uncertainty component, the propagation method, and the resulting deduction applied in the emission reduction calculation. Show that the numbers in the monitoring report match the workbook.

Indonesia and SEA context

Natural forests in Kalimantan, Sumatra and Papua are highly variable, with a few very large trees holding much of the carbon, so sampling error is often high. Smallholder ARR and agroforestry add parcel-to-parcel variability. Remote, expensive field access tempts teams to cut plots, which usually backfires through deductions. Ratings agencies also look at measurement quality, and BeZero’s ratings of ex-post Indonesian projects have averaged ‘A’, so precision is part of how Indonesian supply competes. See credit rating agencies.

Our carbon modelling service builds uncertainty budgets that match the methodology. For more on measurement systems, see the dMRV hub.