Natural England's AI Peat Map Misread the Landscape It Was Built to Protect
What happened
Peatlands cover roughly 12 percent of England's land area and store more carbon than all of the country's forests combined. Protecting and restoring them is a stated policy priority, and in May 2025 Natural England published AI4Peat, a computer-vision model trained to map peatland surface features from aerial and satellite imagery. The map was positioned as an innovative tool to guide restoration decisions at national scale, reaching landscapes where ground surveys are expensive and slow.
The problems were not subtle. Reviewers found the map misidentifying bogs as stone walls, quarrying scars, and granite outcrops. These are not edge cases in difficult terrain. Stone walls are linear, hard, and dry; bogs are diffuse, saturated, and organic. A model that conflates them is not making a close call. It is producing outputs that diverge from physical reality in ways that anyone who has walked the ground would catch immediately. The errors were widespread enough to draw criticism from land managers and policymakers who had expected to use the map for planning decisions.
The stakes are not abstract. Peatland restoration funding in England runs to tens of millions of pounds annually, and allocations increasingly follow spatial data about where restoration is most needed. A map that places bogs where stone walls stand, or marks quarry faces as peat, could direct that funding to land that does not benefit, while genuine peatland goes unrecorded and unrestored. Conservationists warned that the inaccuracies risked misdirecting resources and creating inappropriate land use restrictions in areas the system had misclassified.
The incident is a case study in deployment outpacing validation. AI4Peat was built at a scale and resolution that ground surveys cannot match, which is exactly what made it appealing to policymakers. But scale and coverage do not compensate for a calibration process that fails to capture the variety of the actual landscape. England's upland terrain includes peat over granite, peat adjacent to drystone walls, and intact bog that from above can resemble disturbed ground. A model that has not been tested against these specific configurations cannot be trusted to distinguish them reliably, and the map was released before that gap was closed.
What the incident leaves unresolved is the verification trail. There is no publicly available record of which test sites the model was validated against, what its error rates were by landscape type, or which map cells carried low-confidence outputs before the product was released for policy use. That is the gap accountability infrastructure is built to close: a provable record of what a system produced, which ground conditions it was actually tested against, and where its outputs should not have been trusted without further local review before decisions followed from them.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.