Related Experiment Video
Updated: Aug 2, 2025
08:47
Author Spotlight: UAV Remote Sensing for Efficient Invasive Plant Biomass Estimation
Published on: February 9, 2024
1.5K
Deep learning enables image-based tree counting, crown segmentation, and height prediction at national scale.
Sizhuo Li1,2, Martin Brandt1, Rasmus Fensholt1
1Department of Geosciences and Natural Resource Management, University of Copenhagen, Copenhagen 1350, Denmark.
PNAS Nexus
|April 17, 2023
Summary
A new deep learning framework accurately maps individual trees from aerial images, revealing that trees outside forests significantly contribute to total tree cover. This technology enables better sustainable forest management and digital tree databases.
Area of Science:
- Environmental Science
- Remote Sensing
- Computer Science
Background:
- Sustainable tree resource management is crucial for climate change mitigation, green economies, and habitat protection.
- Accurate tree inventory data is essential but often limited by plot-scale methods that overlook trees outside forests.
- Existing national inventories may underestimate total tree cover by not accounting for non-forest trees.
Purpose of the Study:
- To develop and validate a deep learning framework for individual tree detection and characterization using aerial imagery at a national scale.
- To quantify the contribution of trees outside forests to the total tree cover.
- To assess the transferability of the framework to different geographical areas and data sources.
Main Methods:
- A deep learning-based framework was developed to process aerial images for identifying individual overstory trees.
- The framework extracts tree location, crown area, and height information.
- The model was applied to aerial data covering Denmark and subsequently tested on data from Finland.
Main Results:
- The framework achieved a low bias (12.5%) in identifying large trees (stem diameter >10 cm) in Denmark.
- Trees outside forests were found to constitute 30% of the total tree cover, a figure often unrecognized in national inventories.
- A high bias (46.6%) was observed when evaluating against all trees taller than 1.3 m, indicating limitations in detecting smaller or understory trees.
- The framework demonstrated high transferability to Finnish data with minimal adaptation effort.
Conclusions:
- The developed deep learning framework provides a scalable solution for detailed tree resource assessment using aerial imagery.
- The findings highlight the significant, often underestimated, role of trees outside forests in national tree cover.
- This work supports the creation of digitalized national tree databases for improved spatial traceability and management of large trees.
