DATASETS

Canopy Height Maps

Accurate forest mapping can lead to more accountable forest-based carbon offsets and facilitate the development of carbon projects. To support this work, Meta partnered with the World Resources Institute to create Canopy Height Maps. Canopy Height Maps were developed using AI models on high-resolution worldwide Maxar satellite imagery for over half of the globe.

Download Canopy Height Maps Data

Learn More About Canopy Height Maps

ABOUT THE DATASET

Overview

Download Canopy Height Maps Data

Methodology

Step 1: Describe the satellite and aerial light detection and ranging data

Red, Green, and Blue (RGB) imagery is used as the primary input for generating canopy height maps.

Step 2: Feature extraction

A self-supervised vision transformer called DINOv2, trained on satellite RGB images only, is employed to extract features from the imagery.

Step 3: Decoder Training Process

A convolutional decoder is trained from the previously extracted features using aerial light detection and ranging (lidar) data, which provide information about the height of the canopy.

Step 4: Canopy Height Map Generation

The convolutional decoder works on the features extracted by the vision transformer to generate high resolution canopy height maps.

Step 5: Output

The final outputs are high-resolution canopy height maps that provide a detailed representation of the canopy structure.

RESOURCES