KEY PUBLICATIONS
Science at the forefront
The first time deep learning was used to map billions of individual trees.
​
​
We summarize our experiences in using deep learning for mapping tree resources.
​
We used PlanetScope images to generate 3-m maps on canopy height for all Europe, including both forest and non-forest trees.
We tracked 0.5 billion trees in Indian farmlands over 10 years and found that millions of them had disappeared.
​
OUR VISION
How we think about our work
01
HIGHLY DETAILED
We detect and map trees as objects.
​
​
02
HOMOGENEOUS
Multi-sensor fusion ensures temporal and spatial consistency.
​
03
LARGE SCALE
Our methods scale seamlessly from a single farm to continental and global extents.
04
LONG TERM
Time series analysis going back 25 years to track change over decades.
​​
HOW WE WORK
From pixels to trees
Our pipeline fuses multi-resolution imagery with deep learning to extract tree related information at scales that were previously impossible.
-
PlanetScope, Sentinel-2, Landsat, Skysat, Maxar, GaoFen imagery​
​
-
Deep learning object detection tools
​
-
Tree location, cover, biomass, canopy height outputs
​
-
Validation against ground truth and lidar reference data
GEOGRAPHIC COVERAGE
Work across all continents
Science Advances · PNAS Nexus
EUROPE
We mapped single trees in Denmark and biomass for trees outside forests across Europe.
​
Nature · Nature Climate Change · Nature Ecology · Nature Geosciences · Nature Communications
AFRICA
We mapped tree level biomass for 15 billion trees in the Sahel, 2 million Baobab trees, etc.
​
​
Nature Sustainability · Nature Communications · Nature Cities
​
​
ASIA
We documented a greening over Southern China and also in major Chinese cities, but also found major tree losses in India.
Nature Food · Nature Plants
​
​
​
GLOBAL
We used PlanetScope to map greenhouses and LVOD to estimate biomass dynamics globally.
​
​