UrbanTree3D: An Open Dataset for Urban Tree Species Classification Using Airborne LiDAR and Field Inventory Data

The increasing availability of airborne LiDAR data supports advanced three-dimensional analysis of urban vegetation. However, the development of deep learning methods for tree species classification remains limited by the lack of annotated datasets at the individual-tree level. This study presents UrbanTree3D, a field-validated dataset comprising segmented individual trees extracted from airborne LiDAR point clouds and enriched with species information from field inventory data. The dataset was generated through a structured workflow, including noise removal, vegetation extraction, height normalization based on a digital elevation model (DEM), and temporal consistency verification. Individual trees were segmented using a hybrid approach integrating DBSCAN and Watershed algorithms, and subsequently matched to field inventory data using a nearest neighbor method. A field validation campaign was conducted to ensure data reliability. The final dataset contains 152 individual urban trees and includes six tree species. It provides high-quality annotations, consistent point clouds, and field validation data, supporting its use for training and evaluating deep learning models. UrbanTree3D addresses the current shortage of annotated LiDAR datasets and supports applications in urban forestry, smart cities and urban digital twins.