Parks and greenspaces in England deliver an estimated £6.6 billion ($8.4 billion USD) in health, climate change, and environmental benefits every year. Currently, with 80% of people in England residing in towns and cities, one-third of the English population still doesn’t have access to high-quality green space. In 2023, the UK government launched a bold Environmental Improvement Plan, which states that every person in England should live no more than a 15-minute walk from a greenspace and be able to see at least three trees from their home. In addition, the government made a commitment to increase tree canopy cover in England by 2% by 2050 by planting trees and woodlands.
Because the national Environmental Improvement Plan envisions an impressive future for UK residents, frequent and accurate monitoring of canopy cover — over time at the individual tree level — is crucial. In April 2025, Forest Research, the research agency of the Forestry Commission and the organization responsible for monitoring Great Britain’s trees, woodlands, and forests, published a new Trees Outside Woodlands (ToW) Map. This was funded by the UK Government’s Natural Capital and Ecosystem Assessment (NCEA) program, which is assessing England’s land, freshwater, and coastal ecosystems to produce a baseline of the country’s natural assets by 2029. The tool mapped England’s trees outside woodlands and — for the first time — made the data freely available. Forest Research is now using DINOv2 to build a model that will enhance the accuracy of their maps. This improvement is also essential for the NCEA program, strengthening its baseline assessment and providing robust evidence to inform policy decisions by the Department for Environment, Food and Rural Affairs (Defra).
Previously, Forest Research used a combination of surveys and LiDAR data, a type of geospatial information acquired via lasers and satellite imaging, to evaluate the presence of tree canopy throughout the country. However, these data sources are prohibitively expensive to procure on an ongoing basis, making accurate forest monitoring an ongoing challenge. Monitoring lone trees, groups of trees and areas of woodland under 0.5 hectares is particularly difficult, and with approximately 30% of England’s canopy falling in these woodland types, DINOv2 has been extremely valuable in mapping and monitoring efforts.
Collaborating with the World Resources Institute (WRI), Meta trained its open source computer vision model, DINOv2, on 18 million satellite images to create an open source global map of tree canopy height at a 1-meter resolution, allowing the detection of single trees at a global scale. Since launching the map in April 2024, many governments around the world have shown interest in leveraging the model and maps to improve their own reforestation efforts.
“The high-resolution canopy height model based on DINOv2 is the most powerful open source AI model to be released in recent years and is a game changer for national-scale individual tree detection and monitoring,” says Freddie Hunter, Head of Remote Sensing for Forest Research.
Hunter and the team are currently working to apply Canopy Height Maps to national aerial photography, with the intention of producing more up-to-date canopy cover and height estimates for lone trees, groups of trees, and small woodlands on a national scale, as well as estimates of timber volume loss from trees being cut down. Applying the model to aerial photography may result in a synthetic Canopy Height Model (CHM) that is consistently higher-quality — at least in terms of individual tree identification — than the Environment Agency’s national LIDAR survey. This could allow Forest Research to estimate tree density and canopy area on a rolling three-year time frame.
Forest Research is also interested in leveraging structural information from Canopy Height Maps to enhance tree type distribution mapping efforts at national scales, by developing features for a model and canopy estimation in complex environments like towns and cities.
The use of Meta’s and WRI’s open source model may potentially remove the need for Forest Research to create and train its own models to predict canopy height, which would reduce their reliance on LiDAR and survey data to monitor progress toward national greenspace targets. Additionally, using the canopy model to delineate canopy and estimate tree density may result in major cost savings on data procurement, increase data update frequency, and enhance data quality. Ultimately, this will likely improve monitoring of government tree canopy targets, which could directly impact environmental policy.
Building on the success of DINOv2, Meta recently introduced DINOv3 to help improve visual intelligence. Our hope is that by further enhancing the accuracy of the Canopy Height Maps, governments around the world will have the option to leverage open source models — including DINOv3 — to help monitor their investments in reforestation.
Want to learn more about how partners are using Meta’s AI for social impact applications? Visit our AI For Good website.
Originally published at: Meta
