Infrastructure · Climate · General science

DINOv3 improves open global canopy-height mapping

CHMv2, an open one-metre-resolution global canopy-height map and accompanying DINOv3-based depth-estimation model for forest monitoring.

Summary

A March 2026 preprint and accompanying open release describe CHMv2, which uses a DINOv3-based depth-estimation model to generate a global one-metre canopy-height map. The authors report a higher canopy-height prediction fit than the preceding map and release the model, map, and supporting data for forest, carbon, and biodiversity research.

AI role

A DINOv3 vision-model backbone trained on satellite imagery and airborne-laser-scanning canopy-height references estimates forest-canopy height from optical satellite imagery.

Narrative role

This adds a concrete climate-research infrastructure result: AI produces a reusable global measurement layer that can reduce the data-collection barrier for forest monitoring and downstream environmental studies.

Caveat

The reported accuracy is from the project team’s preprint and depends on uneven airborne-laser-scanning reference coverage; the entry does not establish a measured downstream research-cycle speedup.

Event

Date: 2026-03-06
Timeline role: Notable
Progress relevance: Process acceleration
Validation: Internally reported

Evidence

Preprint · Company announcement

Observed metrics

canopy-height prediction fit: R² 0.86 versus 0.53 for the prior version; The project reports this held-out comparison for the DINOv3-based CHMv2 model against its preceding canopy-height-map model.
map resolution: one metre; The released map provides global canopy-height estimates at one-metre spatial resolution.

Source: arXiv ↗