Conservation Integrity: independent baselines for forest carbon

An ESA Business Applications Kick-start activity

Conservation Integrity on the ESA Business Applications portal ↗

The counterfactual baseline problem we set out to solve

The hard part of forest carbon has never been counting and measuring trees. Satellites have been good at that for years.

The hard part is the counterfactual baseline. Before you can say a project saved a forest, you have to say what would have happened to that forest if the project had never existed. That number decides how many credits get issued, and today it is often produced by the people being paid based on the answer.

That is the credibility gap the voluntary carbon market has been criticised for, and it is the one we work on.

“The challenge is not measuring trees on a project area. It is establishing a credible baseline, meaning what would have happened without the project.”

What we built: independent reference areas from satellite data

ramet5 is a geospatial AI pipeline that selects the reference areas a project’s impact is measured against. What sets it apart is how it chooses them. It finds the land that most closely resembles the project area, from the satellite record rather than from a circle drawn round the project or from negotiation.

Independence comes with that, and it is no longer rare on its own. The reference areas are chosen by the model, not by the developer, and the uncertainty around the result is quantified rather than assumed. The difference is how closely the comparison land matches the project.

ramet47 is the version of the engine built for Verra’s VM0047, the methodology for afforestation and reforestation projects.

The pipeline draws on:

• Sentinel-1 and Sentinel-2, ESA’s radar and optical imagery
• GEDI, NASA’s canopy structure lidar
• The Landsat archive, for the long historical record that carbon methodologies require
• ESA’s BIOMASS mission, which we are integrating as data becomes available

On top of that sits a cloud embeddings pipeline, the matching and analysis layer, and automated reporting built for audit rather than for presentation.

Who independent baselines are for

The Kick-start work was shaped around three groups, and we spoke to all of them:

• Project developers originating forest conservation and restoration projects under standards such as Verra’s.
• Conservation organisations running reserves and reporting impact to their donors.
• Sustainable timber companies demonstrating low-impact forestry in their concessions.

What they asked for was consistent. Reference areas not chosen by the developer. Results an auditor can reproduce. And an honest number for how uncertain the baseline is.

A fourth group matters as well: carbon credit buyers, who want to know how wide the uncertainty is on the additionality different projects claim. That is not a rating. It is information that helps them decide which projects to buy from.

What the ESA Kick-start work has produced

An operational pipeline producing reference areas at jurisdictional scale, rather than a prototype that works on one site.

We have demonstrated it on forest sites in Sabah, Malaysia, southern Africa, Brazil and Switzerland, and tested it against coastal blue carbon scenarios.

belian.earth Ltd ran the activity, which began in November 2025. We announced the award in January 2026.

What we are doing next: VM0047 and landscape screening

The methodology work continues. We have extended the pipeline to Verra’s VM0047 for afforestation and reforestation, and are extending it to improved forest management under VM0010. We are working on landscape-scale screening, which turns the same machinery around to ask where forests would regenerate on their own and where they would not.

We will keep publishing methods as the work develops.