A Breakthrough for Forest Science: 10,000+ Detailed 3D Tree Scans Released
Researchers have unveiled TreeScanPL10K, a massive new dataset containing over 10,000 annotated 3D point clouds of Central European trees, captured using terrestrial laser scanning (TLS). The dataset, described in a study published in Scientific Data (DOI: 10.1038/s41597-026-07269-1), promises to accelerate the development of deep learning methods for forest ecology.
How It Was Built
The team used FARO Focus 3D X130 and Trimble TX5 scanners to capture the data. These devices emit near-infrared laser pulses to create highly detailed 3D models, including geometry and intensity values. Scanning was performed at a quarter resolution, achieving approximately 4 mm accuracy at a distance of 10 meters.
To build the dataset, forest plots of 500 m² were scanned from four different positions. Reference spheres were used to align the scans into a single, cohesive point cloud. Individual trees were then detected using algorithms that analyze spatial coordinates and laser intensity. Species labels were assigned by cross-referencing the scans with detailed field inventory data.
"Quality control revealed ground-point contamination and segmentation errors due to crown overlaps. These were corrected manually. Notably, conifers required fewer corrections than broadleaves."
What's Inside TreeScanPL10K
- Scale: 10,417 segmented trees.
- Identification: Species labels are provided for approximately 72% of the trees.
- Species: Includes key conifers (Scots pine, Norway spruce) and broadleaves (European beech).
- Key Feature: The intensity data from the laser pulses helps differentiate species that have similar shapes but different laser reflectance.
- Attributes: Each tree includes a "completelyInside" flag, indicating if the full crown was captured. Precomputed morphological metrics like height and crown projection area are also provided.
Why This Matters
Modern forestry relies on accurate data for biodiversity assessment, carbon accounting, and sustainable management. While satellite and airborne sensing provide a top-down view, TLS offers a detailed bottom-up perspective of stem and crown architecture.
The primary bottleneck for developing AI models in this field has been the lack of large, annotated datasets. TreeScanPL10K directly addresses this gap.
"The dataset supports species classification, biomass estimation, and monitoring of structural changes in forest stands. "
It is now available for research, offering a powerful tool to improve ecological understanding and forest management practices.
Note: This report is based on a pre-publication version of the paper; the study should not be regarded as conclusive.