Urban trees play a pivotal role in enhancing environmental quality, human well-being, and climate resilience within urban areas. However, conventional approaches for overseeing Urban Green Infrastructure (UGI) frequently exhibit a lack of accuracy and scalability. This dissertation investigates how machine learning (ML), graph data structures, and 3D modeling can support automated urban tree inventories and optimize tree planting locations.
The research addresses four significant gaps: (1) the need for conceptual frameworks to guide tree management towards targeted leaf locations, (2) the limited automation in tree inventory and species recognition, (3) the underutilization of models for predicting tree crown development and optimizing tree planting location, and (4) the absence of an integrated toolbox for simulation and decision support in urban forestry.
A 3D voxel-based design framework is proposed for target-driven planting, allowing urban trees to grow without interfering with infrastructure while maximizing canopy benefits (Chapter 2). A unique multi-layered dataset (TreeML data) is presented, consisting of over 3,700 high-resolution LiDAR-based tree models, quantitative structure models (QSM), and graph representations (Chapter 3). Utilizing this comprehensive dataset, a graph neural network achieved over 84% classification accuracy in tree species recognition by analyzing branch geometries and structural relationships (Chapter 4).
Furthermore, a machine learning model was developed to predict tree crown geometry based on surrounding urban features. The most effective model, a histogram-based gradient boosting regressor (HGBR), attained R² values of up to 0.83 in height prediction and 0.7 in crown radius prediction across eight directions, thereby providing a robust tool for canopy prediction (Chapter 5).
A novel optimization tool integrates these predictive models to suggest optimal tree planting locations, taking into account future crown expansion, competition, and spatial constraints. The tool demonstrated its applicability in several scenarios, confirming its potential for dynamic, long-term urban planning (Chapter 6). To facilitate the translation of research findings into practical applications, the "GroTree" toolbox was developed as a set of Rhino/Grasshopper plugins. This toolbox integrates all the core models necessary for simulation, planting, pruning, and performance evaluation (Chapter 7).
In sum, this dissertation contributes a comprehensive, data-driven approach to supporting UGI management. It integrates field data, advanced modeling techniques, and design integration, providing practical tools to support planners, landscape architects, and ecologists in creating resilient, functional urban green spaces.
«
Urban trees play a pivotal role in enhancing environmental quality, human well-being, and climate resilience within urban areas. However, conventional approaches for overseeing Urban Green Infrastructure (UGI) frequently exhibit a lack of accuracy and scalability. This dissertation investigates how machine learning (ML), graph data structures, and 3D modeling can support automated urban tree inventories and optimize tree planting locations.
The research addresses four significant gaps: (1) th...
»
Login
BibTeX