Quantifying tree mortality at global scale remains a major unsolved challenge. Earth observation satellites, such as the Copernicus Sentinel fleet have sufficient resolution to map even scattered tree mortality patterns. Coupled with supervised machine learning, they present a promising avenue to map standing dead trees at global scale. To acquire data for training globally transferable machine learning models training data, we need a globally comprehensive reference dataset. 

Here, we present the launch of the community-based deadtrees.earth database, compromising more than 1500 globally distributed centimeter-scale drone orthophotos of forests with delineated standing deadwood. This database acts as a foundation to a machine learning model ecosystem that leverages these drone orthophotos and labels as reference for satellite-based tree mortality mapping. Finally, we present a model that provides a yearly map, fractional cover of standing deadwood at 10 m resolution, across global ecosystems. These products will facilitate to reveal and understand global tree mortality dynamics across.

Speakers

Teja Kattenborn, Freiburg University
Clemens Mosig, Leipzig University