Monitoring tree canopy height is essential for understanding forest health, biodiversity, and carbon sequestration, as it reveals key aspects of forest structure and ecosystem dynamics. While LiDAR is the preferred method for accurate measurements, its limited availability and the labor-intensive nature of traditional approaches often restrict coverage to small regions.

This Deep Learning Package (DLPK) features Meta’s High-Resolution Canopy Height model, designed to estimate tree canopy height using high-resolution satellite imagery where LiDAR data is unavailable. The model leverages a vision transformer backbone pretrained through self-supervised learning on millions of satellite images worldwide. A convolutional decoder, trained on LiDAR-derived canopy height data, generates canopy height predictions in meters above ground. This model enables scalable, automated estimation of canopy height across large geographic areas using satellite imagery.