Assessing the Accuracy of UAV-based LiDAR for Urban Tree Inventory Mapping

Thesis Abstract

Urban tree inventory management is critical for sustainable city planning, ecological monitoring, and environmental resilience, yet traditional manual methods are often labor-intensive, time-consuming, and prone to human error. With advancements in remote sensing technology, Unmanned Aerial Vehicle (UAV)-based Light Detection and Ranging (LiDAR) has emerged as a promising tool for rapid, high-resolution urban tree mapping. Despite its potential, the accuracy and reliability of UAV-based LiDAR in capturing detailed urban tree attributes remain underexplored, particularly in densely built environments where data collection and interpretation present unique challenges. This study aims to evaluate the positional and attribute accuracy of UAV-derived LiDAR data in performing urban tree inventories within a metropolitan context. The primary objectives are to (1) assess the spatial accuracy of UAV-based LiDAR point clouds in delineating individual urban trees, (2) evaluate the precision of tree attribute measurements including height, crown diameter, and canopy volume, (3) compare LiDAR-derived metrics with ground-truth measurements obtained through traditional field surveys, and (4) analyze the influence of urban environmental factors, such as building density and canopy cover, on data accuracy. In addition, the study seeks to develop a predictive model using regression analysis to estimate tree attributes based on LiDAR data, and to identify the main sources of measurement errors through error analysis techniques. The research adopts a pragmatic mixed-methods approach, combining quantitative accuracy assessment with qualitative evaluation of data collection challenges. The study area comprises a sample of 150 urban trees randomly selected across different neighborhoods characterized by varying building densities in a metropolitan city. Data collection involves UAV flights equipped with a high-resolution LiDAR sensor at an altitude of 60 meters, generating point clouds with a density of 50 points per square meter. Ground-truth data are collected through manual measurements using diameter tapes and total stations during field visits. Data analysis employs spatial accuracy metrics such as Root Mean Square Error (RMSE) and absolute deviation, along with statistical techniques including linear regression, Bland-Altman analysis, and Analysis of Variance (ANOVA) to evaluate differences between UAV-derived and ground measurements. Data visualization and interpretation utilize Geographic Information Systems (GIS) and specialized LiDAR processing software. Expected findings suggest that UAV-based LiDAR can delineate individual urban trees with a positional accuracy within 0.3 meters and measure tree height with a mean error less than 0.5 meters. Crown dimensions are anticipated to have higher variability, with an average error margin of 0.7 meters, influenced by canopy complexity and urban obstructions. The study predicts a strong correlation (R² > 0.8) between LiDAR-derived and manually measured tree attributes, validating the use of regression models for attribute estimation. It is also hypothesized that high building density zones exhibit reduced data accuracy due to signal occlusion and reflectance issues, suggesting that urban environmental factors significantly influence LiDAR performance. This research contributes novel insights into the applicability and limitations of UAV-based LiDAR for urban forestry management, emphasizing the technology’s potential for scalable, cost-effective, and timely tree inventory processes. The findings will inform best practices for urban LiDAR data acquisition and processing, and propose a standardized protocol for accuracy assessment in complex urban landscapes. The study concludes that UAV-based LiDAR provides a reliable alternative to conventional manual surveys for urban tree mapping when environmental and operational considerations are carefully managed. Recommendations include optimizing flight parameters based on urban density, integrating supplementary sensors such as multispectral cameras for improved classification, and adopting advanced data fusion techniques to enhance measurement accuracy. Future research suggestions involve exploring the use of machine learning algorithms for automated tree detection and attribute extraction, as well as extending the methodology to multi-seasonal data collection for phenological studies.