Meta Announces Canopy Height Maps v2 Using DINOv3 for Forest Mapping

On March 10, 2026, Meta announced the release of Canopy Height Maps v2 (CHMv2), which utilizes the DINOv3 self-supervised vision model to enhance forest mapping capabilities. This new version replaces DINOv2 and is trained on the SAT-493M dataset, a large collection of satellite imagery. The model's accuracy has improved significantly, with an R² value ranging from 0.53 to 0.86, according to John Brandt, Data Science Lead at the World Resources Institute (WRI).

CHMv2 is designed to support various initiatives, including the European Commission's commitment to plant at least 3 billion trees by 2030. The maps will be utilized by the WRI Ross Center for Sustainable Cities in its Cool Cities Lab, which aims to promote urban cooling through the Smart Surfaces Coalition, an initiative endorsed by mayors from 10 cities.

The original high-resolution canopy height maps were released in 2024, and CHMv2 builds on this foundation to provide more consistent and scalable monitoring of forest structures across diverse landscapes. This advancement follows previous efforts to improve global forest cover mapping, which utilized the earlier version, CHMv1.

The introduction of CHMv2 is expected to enhance the capabilities of researchers and urban planners in their efforts to monitor and manage forest resources effectively.