New Gradient unveils AI tool for woodland monitoring

New Gradient has developed a machine learning tool for woodland monitoring across the UK, backed by the UK Space Agency.

The Edinburgh-based AI consultancy says the system is designed to improve how woodland creation projects are measured and verified. It targets parts of the market where investors and land managers depend on carbon and biodiversity data to assess project performance.

Woodland creation is central to UK climate policy. Forests currently remove about 17.4 million tonnes of carbon dioxide equivalent each year, roughly 4% of national greenhouse gas emissions, and government targets call for tree and woodland cover to reach 16.5% by 2050.

Plans to plant 30,000 hectares of new woodland each year are expected to save up to 12 million tonnes of carbon dioxide equivalent in total. Delivering that expansion will require significant private funding, with the release estimating a GBP £1.8 billion gap.

One source of that funding is the Woodland Carbon Code, a government-backed framework that allows businesses to support verified woodland projects in return for carbon credits. More than 93,000 hectares of UK woodland are already registered under the framework, but progress has been slowed by the way projects are monitored.

Conventional surveys are often manual and carried out infrequently, leaving gaps in data on carbon capture and biodiversity outcomes. That can slow verification and make it harder for investors to judge the quality and value of credits generated by a project.

New Gradient's tool uses automated digital Monitoring, Reporting and Verification (dMRV) to collect and process woodland data at a greater scale. It applies machine learning to assess individual trees and provide information on tree health, height, biomass, carbon storage and species identification.

According to the company, the model expands measurement coverage from less than 1% of woodland to full coverage and increases the number of trees directly analysed 100-fold. Early results showed crown segmentation accuracy of 72.3%, above published academic benchmarks, it said.

Funding support

The work was backed through the UK Space Agency's third Climate Services Call, a programme intended to support Earth observation-based services for climate-related applications. The funding is part of a wider GBP £380,000 investment in the UK's Earth observation and climate services sectors.

The woodland model builds on earlier work in peatland monitoring. New Gradient is developing the tool within Calterra, its joint-venture platform with peatland-restoration specialist Caledonian Climate.

That earlier work gave the business a base for extending digital measurement tools into another nature-based carbon market. In this case, it seeks to address long reporting cycles that can leave project developers and investors waiting for years for updated verification.

"Our solution represents a breakthrough for Woodland Carbon Code verification. By replacing expensive, time-consuming manual field surveys with automated dMRV across a broader range of performance indicators, we are putting landowners in a far stronger position to attract investment into these critical woodland projects," said Ewan McMillan, founder of New Gradient.

He said the model could also support more frequent reporting. "The solution facilitates continuous monitoring, shifting verification from decade-long cycles to annual reporting. This gives investors, verifiers and land managers near real-time visibility of woodland carbon delivery, bolstering confidence in verified woodland carbon units."

Model details

The project produced a self-supervised foundation model pre-trained on 1.8 million multi-modal UK aerial image pairs. It also developed four specialist models for tree counting, crown segmentation, species classification and canopy height, along with a biomass and carbon estimation pipeline.

These tools are intended to provide a broader picture of woodland condition and environmental outcomes than standard survey methods. For investors in woodland carbon projects, the aim is for a larger, more regular data set to reduce uncertainty around the assets they are backing.

McMillan said the technical work had delivered a validated method for automated woodland measurement using Earth observation data. "The UK Space Agency grant enabled us to extend our existing peatland dMRV capability, currently deployed in our Calterra platform, into the woodland domain.

"The project has delivered a self-supervised foundation model pre-trained on 1.8 million multi-modal UK aerial image pairs, alongside four expert models for tree counting, crown segmentation, species classification and canopy height, as well as a biomass and carbon estimation pipeline. The results have been overwhelmingly positive, demonstrating a validated approach to automated woodland measurement, reporting and verification using Earth observation data," he said.