Canopy mortality increased tenfold during summer droughts 2018-2020 in Luxembourg

Drought vulnerability and extreme bark beetle outbreaks threaten spruce trees in Central Europe

Forests cover about 31 % of the world's land area and represent a current net sink of more than 20 % of anthropogenic CO2 emissions. In addition to carbon sequestration, forests provide other important ecosystem services such as regulation of the water cycle, wood production, biodiversity and recreation.

In recent decades, we have seen an increase in forest disturbances such as fires and tree mortality due to pathogens and drought. From 2018 to 2020, Central Europe experienced three consecutive dry summers, resulting in widespread forest damage.

As part of a CEDIM project,we investigated the impact of these drought summers using the country of Luxembourg as an example, by automatically detecting canopy mortality using deep learning image classification from aerial imagery (orthophotos). Training data were only collected for the years 2017 and 2019, in order to test the robustness of the approach for years without reference data. The ability to automatically map canopy mortality using deep learning could open up the possibility of continuous monitoring of canopy conditions in future years, without the need for time-consuming steps such as mapping new reference data for each year.

We found a strong increase in canopy mortality from 0.64 km² in 2017 to 7.49 km² in 2020. Taking into account trees removed between years, total canopy mortality was as high as 11 km² in 2020. Of the dead canopy area, 80 % was classified as coniferous and 20% as broadleaf trees. This is remarkable as only 24.5 % of the forest in Luxembourg is coniferous.

The observed vulnerability of conifers can be attributed to the activity of the bark beetle (Ips typographus), which is a parasite of Norway spruce (Picea abies) and can spread rapidly in spruce monocultures. Whilst we observed canopy dieback in broadleaf trees after the drought years, a subsequent field survey showed that most broadleaf trees in the lower levels of the canopy were still alive while the top had died.

Continuous monitoring of large areas using deep learning could help assess forest damage and adapt management strategies, as climate change is likely to lead to an increase in stand altering disturbances in the future.

Associated institute at KIT: Institute of Meteorology and Climate Research Atmospheric Environmental Research (IMK-IFU), Campus Alpin, Garmisch-Partenkirchen
Author: Selina Schwarz (Sep. 2023)

A group of dead Norway spruce (Picea abies) in Luxembourg. Image was taken in July 2023 at a site were widespread canopy mortality was previously mapped by the deep learning algorithm.
Canopy of a birch tree (Fagus sylvatica) which was mapped as dead in 2020 by the deep learning algorithm. Even though the canopy seems dead when observed from above, it has not died completely. Instead the lower canopy levels are still alive.