How do forests grow in different regions of the world? With ECHOSAT, this question can now be systematically answered on a global scale and over multiple years.

While previous tree height maps only represented individual years, ECHOSAT (Estimating Canopy Height Over Space and Time) tracks forest growth, disturbances, and regrowth over multiple years, providing important data for CO₂ monitoring and climate protection efforts.

Jan Pauls, Dr. Karsten Schrödter, Sven Ligensa, and Prof. Fabian Gieseke from the Chair of Machine Learning and Data Engineering at the Department of Information Systems develop ECHOSAT together with international partners: Martin Schwartz and Philippe Ciais (Laboratoire des Sciences du Climat et de l’Environnement (LSCE), France), Berkant Turan and Max Zimmer (Zuse Institute Berlin (ZIB)), Sassan Saatchi (Jet Propulsion Laboratory (JPL), California Institute of Technology, USA), and Sebastian Pokutta (Technical University of Berlin).

The research team combines multi-sensor satellite data with a specialized vision transformer model. A self-supervised growth regularization ensures that the estimated tree heights follow biologically plausible growth curves, including gradual increases as well as abrupt declines due to fire or deforestation.

For example, in Les Landes (France), ECHOSAT clearly reveals the cycles of logging and regrowth, while large parts of the Amazon rainforest remain stable over the observed period. Both patterns are detected automatically, without additional post-processing.

ECHOSAT is part of the AI4Forest project, a German-French research cooperation. Recognizing the critical importance of forest ecosystems for climate and biodiversity, the initiative investigates how state-of-the-art artificial intelligence methods can be used to better understand, monitor, and respond to climate-driven environmental changes.

Links related to this news

Access the full paper here.

The maps are publicly available here.

More information about the AI4Forest project can be found here.