Why agricultural carbon data fails audit

Carbon farming is one of the largest levers available on climate. Most programmes that struggle do not struggle on the science — they struggle because the field evidence cannot survive verification. Here is where it breaks, and what prevents it.

Agriculture, forestry and other land use account for roughly 24% of global greenhouse gas emissions, according to the FAO. That makes farmland one of the largest available levers on climate — and carbon farming a genuinely capable strategy for producing food more sustainably while paying landholders to do it.

The obstacle is rarely the science. It is whether the evidence survives contact with an auditor.

Why agricultural carbon is harder to evidence than industrial carbon

An industrial carbon project has one site, one operator and a meter. An agricultural carbon programme has thousands of smallholdings, hundreds of surveyors, and a claim that rests on what each of them recorded in a field, often on paper, often with no connectivity.

Carbon farming initiatives also depend on convincing farmers to participate at all. Adoption is shaped by scheme complexity, by whether programme targets align with what a landholder actually wants, and by the skills available on the ground. Project developers run training and awareness programmes, onboard farmers, and collect data on farmers, farms and the practices implemented. That collection phase is the critical one: an error introduced there propagates all the way to the credit.

Where field data goes wrong

The same handful of failures show up across programmes, and each has a specific technical answer.

Farm boundaries that do not match reality

The area recorded on paper, the area in government records, and the area actually under cultivation are frequently three different numbers. Drawing boundaries against live satellite imagery — geo-fencing — anchors the record to the ground rather than to a recollection.

Photographs with no provenance

A field photograph proves nothing on its own. It proves a great deal when it carries date, time and farm ID encoded at capture, tying the image to a specific plot on a specific day.

Duplicate and overlapping records

A farmer with several plots can be enrolled several times. Two surveyors can interview the same household. Adjacent boundaries can overlap so the same hectare is counted twice. Duplicate-record and boundary-overlap prevention has to run at capture, because these are close to impossible to unpick afterwards.

Units that do not reconcile

Local land and yield units vary by district, sometimes by village. Conversion factors built into the collection tool prevent a silent order-of-magnitude error entering the dataset.

Language and connectivity

A surveyor working in a language they do not read will make errors regardless of how good the form design is. The same is true of a tool that requires a signal where there is none. Multi-language support and genuine offline capture are data quality features, not conveniences.

Where verification goes wrong

Even good field data can fail at the verification stage, usually for structural reasons rather than dishonest ones:

• The time gap. Months between collection and verification means the field has changed, and the practice being verified is no longer visible.
• No geographic coordinates. Data that cannot be located cannot be re-checked, which makes it unverifiable regardless of accuracy.
• No practical way to physically check. A verifier cannot revisit every plot; without located, timestamped evidence they are reduced to sampling and inference.
• Incomplete or mismatched records. Photographs that do not correspond to the record they sit against are worse than no photograph, because they undermine confidence in the whole dataset.

The answer is to verify continuously rather than at the end. A first pass checks the quality of what was collected, with edit, approve and reject options. A second pass checks the implementation activities themselves and audits the edits made at the first. Both run while field teams are still deployed and corrections are still cheap.

What good looks like

• Every record located and timestamped — so it can be found again by anyone, including a third-party verifier.
• Errors caught at capture — duplicates, overlaps and unit mistakes blocked at the point of entry rather than found in analysis.
• Verification running in parallel — not as a phase that begins after collection ends.
• Live visibility of survey progress — so a systematic problem in one team's method is caught in week one, not at the end.
• Export in formats verifiers already use — spreadsheet and GIS-compatible, so no one is re-keying data at the last step.

None of this is exotic. It is the difference between a programme that issues on schedule and one that spends a season reconstructing its own evidence.

Related reading

The AWD dMRV platform page sets out how these controls work in a rice methane programme, and Apps covers the wider family of field applications. For the full measurement workflow from field to credit, see digital MRV for carbon programmes, and DMRV methods for how each methodology is measured.