Mapping individual tree mortality using sub-meter Earth observation data: Advances toward a large-scale global database

The increasing frequency and intensity of droughts and heat waves driven by climate change have led to a significant increase in tree mortality worldwide. However, the lack of accurate and consistent data on the location, timing, species, and structure of dead trees across vast geographical areas limits our understanding of climate-induced tree mortality. Furthermore, standing dead and dying trees are crucial indicators of forest health and biodiversity but are often overlooked in existing forest resource mapping systems.

To address this, we present novel advancements in mapping individual tree mortality events using high-resolution (≤ 0.5 m) multi-temporal Earth Observation data, including both satellite and aerial imagery, combined with deep learning techniques. Our approach represents the first steps towards building an open large-scale database of individual tree mortality events across time. We have trained several U-Net-based deep learning models for detecting individual dead and dying trees from a wide array of imagery, including high-resolution aerial and satellite imagery from boreal, temperate, and Mediterranean forest biomes, enabling the creation of wall-to-wall datasets on tree mortality at national scales. We show results from the first nationwide individual tree mortality mapping, demonstrating the accuracy of sub-meter resolution satellite imagery in providing annual tree mortality data. We also discuss the challenges and limitations associated with detecting and characterizing detected dead trees across entire countries. 

Currently, our database already includes tree mortality data for 10 years in boreal, temperate, and Mediterranean forest biomes for several countries. We welcome scientists from around the globe to contribute to creating a database on individual tree mortality events to support a wide range of tree mortality data needs in different scientific disciplines.