ML for Earth Observation
Temporal canopy height mapping at global scale using multi-sensor satellite data and vision transformers.
Forest monitoring is critical for climate change mitigation, yet most existing global tree height maps provide only static snapshots. This project develops machine learning methods to track forest dynamics continuously and at scale.
ECHOSAT is a global, temporally consistent tree height map at 10 m resolution spanning multiple years. We train a specialized vision transformer on multi-sensor satellite data (Sentinel-1, Sentinel-2, GEDI LiDAR) that performs pixel-level temporal regression. A self-supervised growth loss regularizes predictions to follow natural tree growth curves, including gradual height increases and abrupt disturbances from fires or logging.
AI4Forest (ICML 2025) presents the first 10 m resolution temporal canopy height map of the European continent for 2019–2022, publicly available on Google Earth Engine.
Papers:
- ECHOSAT: Estimating Canopy Height Over Space And Time — Preprint 2026
- Capturing Temporal Dynamics in Large-Scale Canopy Tree Height Estimation — ICML 2025
Code: github.com/ai4forest/echosat
Demo: Europe Tree Height Map