Is Natural England’s Peat Map a monitoring failure?

The England Peat Map was supposed to be an advanced digital tool to support peatland restoration, wildfire risk reduction, and climate action.  

Instead, it’s become a minor controversy.

Despite all of its many achievements, the England Peat Map is a cautionary tale for farmers, land managers, and soil monitoring professionals about what happens when AI models, poor validation, and questionable governance combine.

Launched in May 2025 by Natural England with great fanfare, the map was built using satellite imagery, aerial photography, LiDAR, and legacy datasets, interpreted via machine learning models through a DEFRA-funded project called AI4Peat.  

It promised to underpin the UK's Peat Action Plan with hard data.

But for people who actually live and work on the land, many of whom rely on detailed instrumentation to monitor moisture levels, organic matter, and vegetation, the map simply does not hold up.

If you need to test soil, take a look at our international directory of companies supplying soil analysis equipment.

Misidentified peat

One of the most outspoken critics of the Map has been Dartmoor farmer Cat Frampton, calling it ‘plainly wrong.’

Speaking to Envirotech Online, Frampton claimed that:

• Limestone pavements, like those at Malham Cove, are mapped as deep peat.
• Reservoirs and rivers show up as peat-rich soils.
• Granite tors, quarries, and tree-covered hedgerows are marked as degraded or calluna bog peat.

Actual known peatlands, including some SSSIs, are missing from the map entirely.

‘Since I’ve been asking the farming community to look at their land, I’ve not had one come back and say, “It’s perfect!” They all find something misidentified as peat, or known peatland missing from the map. For myself, my areas of peat are covered with much larger areas of false positives. Rocks are mapped as peat, hedges are mapped as bare peat and the vegetation layer is very odd.”

The vegetation layers are unreliable, claims Frampton, often showing bog plants in impossible places, such as bare rock.  

One might dismiss these as isolated glitches, but they are consistently widespread and bizarre, even affecting flagship sites showcased in Natural England’s own promotional video.

‘The map seems to not have been cross reference with other Natural England maps,” says Frampton. “For example, the Priority Habitats Inventory is part of the map but doesn’t inform the map – so, known habitats are misidentified.’

Was it AI?

The England Peat Map was created by integrating remote sensing data, topographic data from LiDAR terrain models, legacy soil and peat surveys, geological and land-use records, new field survey data and deep learning algorithms.

The resulting model was trained on over 300,000 data points, producing maps of peat extent, depth, vegetation cover, drainage channels, and erosion features.

The developers claim 94–95% accuracy for peat extent and land cover, and they tout the efficiency of machine learning in achieving national coverage without needing to manually visit every site.  

Responding to comments on Natural England's announcement of the Peat Map, the official account stated:

‘While the models are the most accurate picture of England peat resources to date, some areas of peat will have been missed, and there will be places where the map predicts peat but where it may not actually occur. As we say in the blogpost, our models have an overall accuracy measure of over 95% for the extent of peaty soils, and of 94% for vegetation and land cover.’

But that remaining 5% inaccuracy is not randomly distributed or trivial.  

Instead, it appears to cluster around ecologically sensitive or technically complex terrain with features like quarries, rock outcrops, forested hillsides, and water bodies.

Here lies the fundamental challenge: AI, no matter how advanced, is only as good as its training data and validation process.  

AI models are built to extrapolate patterns from visible proxies, i.e., land cover, terrain shape and historical use.  

However, without ground truthing, such models end up trying to predict where peat ought to be, risking false positives.

According to the Moorland Association, Natural England admits that landowners shouldn’t rely on the map to determine peat depth, a staggering admission for a map meant to guide policy, restoration, and funding.

Why does this matter?

Well, this tool is part of a broader government initiative to restore England’s peat for a variety of purposes.  

If used to assess the progress of these projects, it could falsely declare their failure and success.

For those working with soil moisture sensors, organic content analysers, carbon flux instruments, and related tools, the message is clear: don’t trust national datasets without ground-truthing.  

The peat map debacle shows that:

• AI needs field validation: high-tech input doesn’t equal real-world accuracy
• Policy tools must be transparent and correctable, otherwise they undermine trust
• Monitoring professionals must advocate for data integrity or risk being undermined by flawed top-down tools

The map is already being used to shape policy on heather burning, restoration funding, and wildfire prevention, despite Natural England’s own advice not to rely on it.

This disconnect between what looks good on paper and what’s happening on the ground is not just inconvenient, it’s dangerous.

The uplands, and the professionals tasked with stewarding them, deserve better.  

Whether you're a farmer monitoring bog water tables, a researcher tracking vegetation shifts, or a policymaker trying to target restoration funds, the tools you rely on must be accurate, ground-truthed, and fit for purpose.

The England Peat Map could have been a cornerstone. Instead, it stands as a warning: about the risks of AI without accountability, and about the urgent need for real reform in how environmental data is created, tested, and used.