Layer 04 — the product

Uncertainty quantification in carbon MRV

A point estimate cannot responsibly issue a carbon credit. A bounded estimate can. Uncertainty quantification in carbon MRV is the propagation of measurement, sampling, classification and model error into plot- and project-level confidence intervals that a registry can deduct against.

Last updated 6 August 2026

A registry buys one thing. The error bar.

Why uncertainty is the binding constraint — not detection accuracy

Demonstrating 95% flooded/drained classification accuracy on a single instrumented demo district is table stakes. Issuance decisions are made across tens or hundreds of thousands of heterogeneous ~1 ha plots that were never instrumented. The question is whether the confidence interval remains defensible after sparse sampling, SAR classification error, and flux-model parameter uncertainty are combined.

After Verra's 2024 rice credit invalidation, markets treat unverifiable precision as a liability. Buyers burned by overstated reductions will prefer a wider, honest interval to a narrow, undefended one.

Error propagation from sparse sensors to plot-level estimates

1. Sensor measurement error2. Sparse-sample representativeness3. SAR classification error4. Flux model parameter uncertainty5. Aggregation to project tonnes
Propagation chain — each stage widens or shifts the interval unless modelled.
  1. Sensor measurement error — water-level instrument bias, drift, and installation effects on the calibrated subset.
  2. Representativeness error — the statistical gap between sparsely instrumented plots and the uninstrumented population across soil, slope, and irrigation strata.
  3. SAR classification error — false wet/dry labels from canopy, speckle, angle, wind, and boundary mixing.
  4. Flux model parameter uncertainty — emission-factor and process-parameter ranges when converting flooding regime to CH₄.
  5. Aggregation — how plot-level intervals combine into project tonnes and what conservativeness factors apply.

How registries treat uncertainty

Methodologies do not treat uncertainty as optional colour. They apply conservativeness: deductions, discount factors, or required quantification approaches that bias toward under-crediting when evidence is weak.

Read normative text on the registry sites before treating any summary here as operational guidance. See Gold Standard MRV requirements.

What is still open at smallholder scale

Plot-level uncertainty quantification under monsoon cloud, across millions of heterogeneous holdings near one hectare, remains an open engineering problem. We say so because it is true. In a market recovering from a verification scandal, naming the unsolved layer is more credible than claiming satellite-plus-AI has closed it.

Our TRL-3 work attacks this layer specifically. Methods, datasets, and validation reports will publish on Research with DOIs as they exist. Until then, treat marketing claims of solved smallholder uncertainty—from anyone—as incomplete.

Sources

  1. Verra — 2024 rice project rejections (integrity context)
  2. ICVCM — Gold Standard rice methodology CCP decision
  3. Isometric — Rice Methane Reduction Protocol v1.0 (uncertainty & monitoring requirements)
  4. Isometric — Protocol announcement (conservative IPCC pathway with uncertainty discount)

Registry-ready verification, independent of origination. Measure a district with us.