Economic segregation has been studied using various social simulation models, including agent-based models. These models, however, typically operate on abstract or schematic spatial representations that strip away cities’ morphological character- istics, limiting the extent to which simulation results can be interpreted as relevant to actual urban contexts. This thesis aims to bridge that gap by developing a six- stage hierarchical regression-based pipeline that takes a stochastically generated synthetic urban structure as input and populates it with household compositions and income distributions calibrated on five Dutch cities: Groningen, Den Haag, Haarlem, Amersfoort, and Maastricht. Using the Ordinary Least Squares method, building type and distance to the city center are identified as the primary spatial predictors of household type, building capacity, and a residential unit’s floor area, with income dis- tribution following household composition. Applied to CityStackGen’s synthetic urban structures, the pipeline produces outputs that are internally consistent, stable across repeated runs, and demographically plausible at the city level. Cross-city scenario generation further demonstrates the pipeline’s utility as a controlled analytical tool for socio-spatial experimentation. The findings of this research aim to contribute to a method for generating spatially grounded synthetic environments that can serve as realistic starting conditions for social simulations of economic segregation. ... Read more

In recent years, several methods have been successfully developed to create 3D building models in LoD2.2 from ALS point clouds. These methods however rely on 2D horizontal uilding outlines whose definitions and quality vary significantly. There are mainly two types of outlines: roofprints and footprints, and having access to both of them allows to reconstruct the façades and the roof independently, therefore improving from LoD2.2 to LoD2.3 models. Moreover, this distinction also allows more precise analyses depending on the use case. Therefore, we present a method to construct both a roofprint and a footprint from ALS data and existing inaccurate outlines. By first precisely deforming and aligning the outline on the roof to turn it into the roofprint, our method can then accurately identify the few ground and façade points available, if there are any, in order to produce a footprint coherent with the roofprint. Our method preserves the validity of the polygons and the topological relations between them. It has been tested successfully on the French national datasets BD TOPO and LiDAR HD. ... Read more

CGAL’s 3D Alpha Wrap robustly produces valid enclosing meshes for defective input using two parameters: α, which controls the carving-ball size, and δ, an offset that controls the distance between the input and the output surface vertices. However, preserving sharp concave features often requires a small global α-value, which also increases refinement in regions where additional vertices do not substantially improve the approximation. This paper proposes a 𝜏-guided extension that adds a local face-to-input distance criterion to the original algorithm. The new parameter 𝜏 triggers additional refinement only where the current wrap remains locally too far from the input.

To make this refinement effective near concave features, we introduce R3 as an additional vertex-insertion strategy alongside the two insertion strategies of the original algorithm. R3 aims to place new vertices more directly near sharp concave features, whereas the original two insertion strategies may reach comparable vertex positions only after repeated refinement. The extension preserves the carving-based structure of the original algorithm, and we prove that the guarantees of termination, strict enclosure, watertightness, and manifoldness still hold. Experiments on 3D BAG building data show that the proposed method improves both average and local approximation quality, achieving lower directed Chamfer distances with fewer output faces than the original algorithm. Although R3 is less time efficient per insertion, it places vertices more effectively near concave features, so fewer insertions are needed to reach lower approximation errors. As a result, the proposed method can reduce runtime in the tested use case. ... Read more

Sidewalks are fundamental urban infrastructure, yet large-scale, openly available geospatial sidewalk datasets remain scarce worldwide, hindering pedestrian routing, accessibility analysis, and urban planning applications. This thesis investigates the feasibility of transforming crowdsourced street view imagery from Mapillary into structured georeferenced sidewalk data using only openly licensed data and pre-trained models. Two complementary automated pipelines are developed and evaluated across three study areas in Amsterdam (Netherlands) and Boston (United States).

The first pipeline, Sidewalk Inventory Mapping, constructs a binary presence inventory (yes/no per roadside) by aggregating pre-computed semantic segmentation outputs and SfM-corrected camera metadata from the Mapillary API onto the OpenStreetMap road network, without any local model inference. The second pipeline, Sidewalk Geometry Reconstruction, downloads imagery and processes it locally using a vision foundation model (DINOv3) for semantic segmentation and monocular metric depth estimation (Depth Anything V3) to reconstruct sidewalk polygons and centerlines.

Results demonstrate that the inventory pipeline achieves algorithmic precision of 97.2% to 98.6% and algorithmic recall up to 86.5%, providing highly reliable sidewalk detection where imagery exists. The geometry reconstruction pipeline successfully produces sidewalk polygons but with limited spatial accuracy (IoU 0.100–0.329, width MAE 1.18–1.40 m), resulting from cascading errors in segmentation, depth estimation, and GPS positioning. Both pipelines are fundamentally bounded by Mapillary's spatial coverage, with system recall dropping substantially in areas lacking imagery. The inventory approach emerges as the practical, scalable solution for city-wide deployment, while the geometry reconstruction demonstrates technical feasibility but requires further refinement for production use. All outputs are compatible with open data standards and can support OpenStreetMap enrichment workflows. ... Read more

Ever increasing global heat has necessitated more robust urban shade analysis in order to keep, among other applications, pedestrians safe and comfortable. Specifically, the impacts of shade-trees on the urban shade landscape is a topic of growing interest and research, and new tools and workflows are being developed in order to allow urban planners to have the clearest understanding of how to best provide shade to citizens. Geometric shade calculation techniques have been used for some time, but they are slow to run at scale and can involve convoluted workflows that slow down analysis. Thus, this thesis investigates the applicability of deep learning to generate shade rasters. The goal is to see if deep learning can allow for faster shade map derivations and a smoother workflow, without sacrificing accuracy. Furthermore, it is the first research of its kind to consider a purely deep learning based approach in deriving tree-aware urban shade maps which capture the contributions of shade trees.

This thesis presents a Conditional Generative Adversarial Network based off of the pix2pix framework, and explores four different combinations of input data and how they affect the final model performance. The model can produce shade rasters ~25 - 45.5% faster than the geometric approach, depending on the parameters used when generating outputs. It also does this using less data and less preprocessing than the geometric alternative. However, while it produces very convincing rasters in some cases, the outputs have insufficient temporal coherence to serve as a reliable alternative to the geometric approach. Thus, while it serves as an interesting foundation for future research, the model as presented in this report is not sufficient to replace geometric shade derivation techniques. ... Read more

Accurate data on inland water surface elevations are becoming increasingly important as climate change intensifies extreme weather events worldwide. Classic data collection methods on water are often expensive and both spatially and temporally limited, making water data largely concentrated in the global north. Ice, Cloud and Land Elevation Satellite 2 (ICESat-2) and its advanced Light Detection And Ranging (LiDAR) instrument are, however, well-suited for collecting data as it is capable of measuring elevation with centimeter-level accuracy across the planet. Its existing inland water data products, ATL13 and ATL22 are, however, constrained by their reliance on water masks that exclude lakes and reservoirs smaller than 0.1 km2 and rivers narrower than 50 m.

The goal of this thesis is therefore to create a Random Forest (RF) model using ICESat-2 data that can predict the presence of inland water, specifically for bodies smaller than 25 m. A set of window-based features that encode the interaction between photons and water was derived at multiple window radii. These features try to quantify and characterize the presence of afterpulses, bottom reflectances, a low slope, a low distribution of photon elevation, and a high photon density. The features were identified by first de-correlating all features using Ward’s linkage clustering and then selecting the best using a Mean Decrease in Impurity (MDI) and permutation importance score. Photon density proved to be the best predicting feature, contributing 41.2% of the total mean decrease in impurity in the model.

The final random forest model used a 2.5 m window and was trained on 7 million points from water segments smaller than ≤ 25 m and an equal number of land points in the Netherlands. The model was evaluated on approximately 520 million ICESat-2 photons across the country and achieved a recall above 80.0% for water segments longer than 6 m and up to 87.1% for water bodies between 10 m and 25 m. Potential improvements of using the features from multiple windows or selecting only windows with a minimum number of photons present proved to be ineffective in gaining better results.

The main sources of misclassifications are currently the presence of snow, uncertainty near water edges, and incorrect ground truth data. Manually validating results in the Swiss Alps, Greenland, and a Mexican mangrove forest showed promising results. The RF performs reasonably well outside its training environment, suggesting that the model identified general photon-water interactions rather than region-specific characteristics. Future work should focus on expanding the training dataset to include geographically diverse data to improve performance and combining the classified photons using clustering to create line segments of water surface elevation. ... Read more

Accurate urban Computational Fluid Dynamics (CFD) simulations require volumetric meshes whose boundary conforms to complex city geometry while maintaining cell quality and enabling adaptive mesh sizing. This thesis develops a Voronoi-based polyhedral meshing methodology tailored to the output of City4CFD, where the simulation domain is provided as a triangulated Piecewise Linear Complex (PLC) whose faces are grouped into semantic boundary patch types. The central goal is to generate a boundary-conforming polyhedral mesh without clipping Voronoi cells against the boundary, and to preserve patch information so that CFD boundary conditions can be transferred consistently to the mesh.

The method constructs a set of boundary (surface) Voronoi sites by intersecting triplets of spheres centered at the vertices of a refined surface triangulation. Building on the sphere-based sampling conditions of VoroCrust, sphere radii are initialized and iteratively adjusted to satisfy smooth-coverage, smooth-overlap, and Lipschitz-type size-transition constraints, while a shrinking step resolves "half-covered" seed configurations. The surface triangulation is refined by splitting triangles and protecting edges, then regularized with centroidal smoothing until the triangles associated with each facet intersect in two points, yielding paired sites on opposite sides of the boundary. These sites induce Voronoi facets that coincide with the input surface, producing triangular boundary faces and avoiding cell clipping. Patch-boundary preservation is enforced by identifying protected edges not only via dihedral-angle sharpness but also via changes in patch groups, ensuring that 1D interfaces between patch types are explicitly represented in the resulting Voronoi boundary.

To populate the mesh interior with Voronoi sites, the thesis evaluates several strategies (uniform random scattering, adaptive distance-based refinement, and structured lattices) to regulate cell density and promote larger cells away from geometric features. A prototype implementation in C++ using CGAL demonstrates feasibility for 2-manifold inputs and produces boundary-conforming Voronoi meshes compatible with OpenFOAM-style polyhedral representations. The approach assumes a valid 2-manifold boundary and does not repair non-manifold or overlapping input geometries. ... Read more

This thesis addresses the challenge of extracting wireframe models which consist of 3d line segments from point clouds of man-made urban linear objects, with a specific focus on power lines and pylons. Wireframe models are essential for various applications including 3D city modeling, infrastructure monitoring, and urban planning. However, the automatic extraction of accurate wireframes from sparse airborne LiDAR point clouds remains challenging due to the complexity of these structures. Current wireframe extraction research either require high-quality data for model fitting or depend on complex pre-processing steps, lacking generality. To address these challenges, this thesis proposes a comprehensive evaluation of multiple wireframe extraction approaches and introduces an energy minimization framework which aims to address the limitations of the existing algorithms.

This thesis investigates four different algorithms for wireframe extraction: 3D RANSAC, 3D-2D RANSAC, Region Growing, and Hough Transform to address their limitations. Additionally, an approach of energy minimization for Markov Random Field is proposed to explore the potential of energy minimization methods in wireframe model extraction. Each algorithm is evaluated using a dataset of power lines and pylons from the Netherlands, with manually extracted wireframes serving as ground truth.

Experimental results demonstrate that each algorithm exhibits distinct advantages and limitations. The 3D RANSAC algorithm struggles with cylinder radius estimation and overlooks significant portions of input data. The 3D-2D RANSAC approach reduces dependency on normal estimation but still faces challenges with fitting accuracy. Region Growing achieves lower overlooking rates but suffers from scattered distribution of extracted elements. Hough Transform performs well on simple structures without requiring normal information but becomes computationally expensive for complex cases. The proposed energy minimization method shows promising results in preserving structural integrity by processing dense input graphs, particularly for complex structures with internal components.

Common limitations across all approaches include difficulties in normal estimation from sparse point clouds, misalignment between extracted primitives and ground truth, and challenges in balancing completeness and accuracy. The research emphasizes the complexity of wireframe extraction from point clouds and provides insights for developing more robust methods that combine the strengths of different approaches while addressing their mutual limitations. ... Read more

Urban digital twins rely on accurate rooftop geometry, yet airborne lidar point clouds are frequently sparse and incomplete, leading to substantial information loss in building reconstruction. This thesis investigates diffusion--based learning as a remedy for high-fidelity roof recovery under severe data corruption.

This thesis proposes a two-stage framework that operates on 2.5D height-map representations. Stage~I introduces a dual-task diffusion model that jointly performs roof height-map completion and roof-line prediction. A novel Bidirectional Control Module enables reciprocal conditioning between the two tasks, enforcing geometric consistency during the denoising process. Stage~II employs a patch-based diffusion upsampler equipped with positional embeddings and a domain-specific global context encoder to synthesise high-resolution height maps while remaining computationally tractable for large and variably-sized buildings. A rigorous preprocessing pipeline further yields two challenging benchmarks, \textsc{S80\_i30} and \textsc{S80\_i80}, derived from 160k real-world building samples.

Extensive experiments conducted on these datasets demonstrate the effectiveness of the proposed approach. Under moderate corruption (\textsc{S80\_i30}), the completion model attains an \textit{RMSE} of \textbf{0.89}~m and a Chamfer distance of \textbf{0.06}, improving upon the state-of-the-art RoofDiffusion baseline by 13.2\% and 17.3\%, respectively. In the severe setting (\textsc{S80\_i80}), the method sustains a 13.5\% \textit{RMSE} reduction. The upsampling stage delivers an additional 10\% \textit{RMSE} gain over the best classical interpolator, and the end-to-end pipeline achieves \textit{RMSE} values of 0.91~m (moderate) and 1.42~m (severe).

The thesis contributes: (i) a structurally-aware diffusion framework for roof completion, (ii) a scalable patch-based upsampler, and (iii) public benchmarks that reflect real lidar degradation. Collectively, these advances close a critical gap between theoretical research and practical generation of LOD2.2 building models, facilitating more reliable urban analytics and planning applications.
... Read more

As urbanization and climate change intensify, managing the urban microclimate becomes increasingly challenging, affecting outdoor thermal comfort. The practical integration of urban microclimate research into urban design remains limited, in part due to the complexity and inaccessibility of existing tools. To support early-stage, climate-sensitive urban design at the neighbourhood scale, I present SOLWEIG For Design (SOLFD): a computation framework that builds on the existing SOLWEIG tool. SOLFD enables urban designers to visualize current microclimatic conditions and assess the impact of design interventions on outdoor thermal comfort. In particular, it focuses specifically on (in)direct solar radiation and its effect as quantified by the mean radiant temperature. Key contributions include: (1) extending SOLWEIG’s 2.5D model to a layered 3D representation for improved accuracy in complex urban geometries; (2) automating the data pipeline using open Dutch geospatial datasets; (3) enabling the modification of the existing urban scene; (4) enhancing output usability through temporally grouped mean radiant temperature maps, derived physiological equivalent temperature maps, and comparison statistics; and (5) significantly reducing simulation time with GPU acceleration. The accuracy of SOLFD was validated using sensor data, achieving an RMSE of 5.39°C. Underneath structures, the RMSE increases to 5.83 °C. The potential of SOLFD is further demonstrated with a case study across various Dutch urban typologies. By laying the foundation for an accessible decision-support tool for outdoor thermal comfort, SOLFD takes a step toward integrating climate-responsive strategies into the urban design process. ... Read more

Standardising data formats for 3D city models is crucial for semantically storing real-world information as permanent records.
CityJSON is a widely adopted OGC standard format for this purpose, and its variant, CityJSON Text Sequences, decomposes large city objects into line-separated objects to enable streaming processing of 3D city model data.
However, the shift towards cloud-native environments and the increasing demand for handling massive datasets necessitate more efficient data processing methods across different platforms and on the web.
While cloud-optimised data formats such as PMTiles, FlatGeoBuf, Mapbox Vector Tiles have been proposed for vector and raster data, options for 3D city models remain limited.
This research aims to explore optimised data formats for CityJSON tailored for cloud-native processing and evaluate their performance and use cases.
Specifically, the study implements FlatBuffers for CityJSON, incorporating features like spatial indexing, spatial sorting, indexing with attribute values, and partial fetching via HTTP Range requests.
The methodology includes designing a complete binary representation of the CityJSON standard using FlatBuffers, conducting a comprehensive review of existing performance-optimised formats, and benchmarking their performance.
Successful implementation of this research will enable end-users to download arbitrary extents of 3D city models efficiently.
The research demonstrates that FlatCityBuf achieves superior read performance compared to CityJSONSeq while generally producing smaller file sizes.
The approach successfully encoded the entire Netherlands dataset into a single 70GB file containing both spatial and attribute indices, demonstrating scalability for national-scale applications.
For developers, the optimised format enables single-file containment of entire areas of interest, simplification of serverless cloud architecture, and accelerated processing by software applications.
Ultimately, this work improves the scalability and usability of 3D city models in cloud environments, supporting advanced urban planning and smart city initiatives. ... Read more

Urban climate simulations increasingly rely on digital twins of cities, yet vegetation remains largely absent or oversimplified despite its strong influence on wind flow and heat exchange. Existing lidar-based tree studies mainly target forests or small plots, and no scalable method currently exists to reconstruct detailed urban trees for computational fluid dynamics (CFD) analysis.

This thesis presents a scalable, automated pipeline for reconstructing CFD-ready urban tree models from open-access airborne lidar. The workflow operates directly on unstructured point clouds and comprises three main components. The first introduces the High-Order Multi-Echo Density (HOMED) vegetation filter, a new approach for distinguishing vegetation from non-vegetation in airborne lidar point clouds. Combined with TreeSeparation, a cuboid-based tree instance segmentation algorithm, it produces clean per-tree clusters for subsequent analysis. The second explores taxonomy-based classification to assess structural separability between species. The third abstracts each segmented point cloud into CFD-suitable crown and trunk geometries and derives per-tree volumetric porosity values, ensuring watertightness, manifoldness, and controlled mesh complexity. Designed to be dataset-agnostic, the pipeline generalises to any comparable airborne laser scanning data.

The workflow was applied to several major Dutch cities—Amsterdam, Rotterdam, Utrecht, and Delft—processing hundreds of thousands of trees from raw point clouds to 3D models within practical runtimes (≈ 13 hours for Amsterdam on 16 CPU cores). Segmentation and reconstruction proved robust and consistent, yielding α-wrapped crowns, cylindrical trunk approximations, and physically meaningful porosity estimates. Taxonomic classification, however, was unreliable due to label noise, class imbalance, and limited structural separability in lidar-only data; supplementary optical features from RGB and infrared imagery were evaluated but proved unreliable for consistent integration.

The results demonstrate that the pipeline enables scalable reconstruction of CFD-ready tree models that preserve canopy structure and aerodynamic properties, allowing their explicit representation in urban digital twins and supporting more realistic urban climate simulations.
... Read more

This report documents the development of Delft3D, a web-based 3D application, designed to visualise Delft’s urban features—such as buildings, vegetation, and water bodies—using open datasets. Developed for the Municipality of Delft, Delft3D customises the Netherlands3D platform to provide accessible, interactive visualisation for residents and employees of the municipality. This report primarily focuses on the process of development and only explains the implemented features conceptually. After the explanation, we provide a concise reflection on the hurdles we faced during the project and we give an overview of the limiting factors. ... Read more

By providing shade for residents in urban areas, cool spaces have been shown to be essential for mitigating the effects of heat stress. In response, the Municipality of Amsterdam developed a map showing walking distances to these spaces. However, the map lacks key information on capacity, accessibility, and precise distance measurements. This project addresses these gaps by identifying quality indicators for cool places and mapping their locations and quality scores across Amsterdam. Additionally, it establishes methods for computing the shortest and shadiest pedestrian routes to these spaces, enabling efficient routing to and from any given location.

To address the research questions, the following procedures were conducted. First, shade maps of Amsterdam were created for each warm month using the Daily Shadow Pattern tool of the Urban Multi-scale Environmental Predictor (UMEP). Second, cool spaces were identified and evaluated based on accessibility, shading, usability, capacity, heat risk, and Physiological Equivalent Temperature (PET) indicators. Lastly, after obtaining and processing the pedestrian network from the Open Street Map database, shade weight was calculated for each street segment, and cool spaces were incorporated into the network, allowing users to generate datasets of the shortest and shadiest distances to cool spaces, and an algorithm that performs four different routing options: the shortest, the shadiest, and two combinations of the shortest and shadiest paths with different weighting ratios either between two locations or from a starting point to its nearest cool space.

The project produced datasets which provide insights into Amsterdam’s cool spaces, their quality, and the shadiest and shortest routes to these locations. Additionally, the code to make these datasets has been made available on GitHub. ... Read more

The growing complexity of urban environments has intensified the need for accurate 3D city models to support simulations and analyses in various fields such as urban planning, energy demand, and computational fluid dynamics (CFD). However, these models often contain geometric and topological errors, such as non-watertight solids, intersecting volumes, and missing surfaces, limiting their usability. This thesis presents AUTOr3pair, an automatic framework designed to repair semantic 3D city models and address these issues.

The process begins with validating 3D city models using val3dity, which ensures geometric validity based on ISO 19107 standards. Errors are addressed hierarchically, starting with ring-level fixes and progressing to solid interaction-level corrections, focusing on localized repairs to minimize alterations while preserving geometric and topological integrity. Existing repair methods are integrated alongside new algorithms designed to meet the specific needs of different use cases, including visualization, energy demand estimation, solar potential analysis, and CFD simulations. Written in C++ for optimal performance, the framework supports CityJSON and OBJ. The repairs prioritize maintaining semantic consistency and minimizing data loss, but textures are excluded. This approach results in a robust validation and repair pipeline that generates detailed error reports and post-processing outputs, significantly improving the overall quality of 3D city models.

Extensive tests on real-world datasets, including 3DBAG tiles from Leiden and a dataset from Brussels, demonstrated that AUTOr3pair successfully repaired most geometric errors, raising validity rates to nearly 100%.
The framework achieved additional requirements, such as watertight geometry and proper surface orientation, for the specific use cases, CFD, energy demand, visualization, and solar power estimation. While some floating-point errors and geometric differences, due to global repairs, persist in complex cases, AUTOr3pair significantly reduces manual pre-processing and improves model suitability for various applications.

This thesis demonstrates that automatic geometry repair is feasible and essential for improving the quality and usability of 3D city models. It provides a foundation for further research and development, particularly in extending the framework to support more file types and refining its capabilities for additional use cases. ... Read more

This thesis introduces the “city stack” framework and the concept of the “typology grid” to analyze and procedurally generate cities using publicly available geospatial data. It builds upon existing urban morphology research and combines existing metrics with new metrics to examine road networks and building footprints across 43 cities worldwide. The resulting unsupervised clustering of road patterns produced mixed outcomes: some global road typologies were consistently identified, while others lacked clarity or validity. Supervised classification of building typologies showed potential for city generation applications but faced challenges related to data quality and validation methods.
To encode the captured urban form, the typology grid was introduced, which allows for straightforward comparison between cities and links the analysis phase with the generation phase. While this approach simplifies complex urban patterns for better understanding, it may oversimplify details needed for effective city regeneration, needing future research into parameterizing the grid.
The simulated annealing optimization technique was applied to generate new city models to produce typology grids resembling those of actual cities. It proved a promising method, with relatively plausible results from just a few shape-based rules in the objective function. Computation time and grid size were a limiting factor, ruling out real-time use. The road and building generation methods based on the typology grid and city stack framework demonstrated the approach’s feasibility but indicated that further refinement is necessary.
Further contributions by this thesis are an open source command line tool for analyzing the urban form real-life cities based on publicly available geospatial data, a proof of concept tool for generating cities, and a publicly available dataset of the analyzed cities.
In conclusion, the city stack framework and typology grids offer a viable method for captur- ing and generating urban form and can be used as a starting point for future research. ... Read more

While three-dimensional coral reef models are valuable for various applications, existing approaches like photogrammetric scanning and manual modeling require substantial time and expertise, limiting their scalability. Previous algorithmic approaches, particularly Agent-Based Models (ABM), have relied heavily on complex ecological simulations and deep domain knowledge. This thesis explores an alternative data-driven approach to automated coral reef modeling, investigating whether empirical data sources can provide a scalable method for generating ecologically plausible 3D models. Rather than simulating ecological processes from first principles, we develop a pipeline that leverages observational data to inform and constrain procedural generation techniques. Through systematic evaluation of available data sources, including the Global Biodiversity Information Facility (GBIF), CoralNet, the Allen Coral Atlas, the Coral Traits Database and the Smithsonian Institution's 3D coral collection, we identified both opportunities and significant limitations in current data availability. The research developed a modular pipeline implemented in Blender that combines procedural terrain generation with the placement of 3D coral models, integrating species occurrence data aggregated over geomorphic zones. To ensure robust data integration across sources and maintain compatibility with evolving taxonomic standards, the pipeline implements automated species name verification through the World Register of Marine Species (WoRMS) (WoRMS - World Register Of Marine Species, n.d.) API. While the resulting pipeline successfully establishes a foundation for automated coral reef modeling, limitations in available structural data necessitated the use of manually configured parameters for critical aspects such as terrain characteristics and population density. The pipeline's modular structure, standardized taxonomy handling, and integration with standardized classification systems position it well for future iterations as improved data sources become available. This research demonstrates the potential of data-driven approaches to coral reef modeling while highlighting the need for more comprehensive, fine-scale structural data to enable fully automated, ecologically plausible modeling of coral reef environments. ... Read more

Roofing material classification is becoming pivotal in urban decision-making, supporting processes like asbestos mapping, disaster preparation, and urban heat island detection. However, research has mainly focused on a restricted set of materials. Additionally, many studies rely on expensive multi- or hyper-spectral imagery and often use outdated or ineffective classification methods, overlooking the transformative capabilities of advanced deep learning and data fusion. This thesis explores using a convolutional neural network for pixel-level classification of Dutch buildings by merging standard aerial images with LiDAR data, enabling detailed mapping and addressing concerns about classification efficacy relative to image- and object-level methods.

A framework was devised to generate a new semantic segmentation dataset with over 15.5 million pixels from 200 randomly selected images nationwide, covering eight distinct materials. To facilitate material identification in unfavourable lighting conditions, true-colour aerial imagery from the BM5 dataset was combined with rasterised features extracted from the national point cloud (AHN4), specifically reflectance, slope, and planar point density. Additionally, a quasi-normalised elevation model (nDRM) was employed, based on the corresponding digital surface model and median roof elevation of buildings in each scene, as provided by the 3DBAG dataset. The research was further investigated using the DeepLabv3+ semantic segmentation architecture with a ResNet-18 backbone, and the model was trained end-to-end on the generated dataset. In this context, a novel stratified splitting algorithm and weighting scheme to combat class imbalance in the training subset were introduced.

After thorough hyperparameter tuning, we achieved a 64.68% mean intersection over union on the test subset. Membranes and gravel outperformed almost every other study. However, there were notable confusion and omission errors with light-permitting surfaces and metal. Further testing of pixel-wise material maps' generalization to different LoDs of the 3DBAG considerably decreased gross errors. However, it might overlook some minor original predictions, thus not improving overall performance notably. Generally, LoD1.2 was inadequate for modelling multi-material roofs of different heights. While LoD1.3 improved this, it still missed small roof sections, unlike LoD2.2, which also had more outliers. Additionally, an ablation study on the LiDAR-derived component of the new dataset showed that removing slope and nDRM reduced performance by 10.31% and 8.61%, respectively, while density had the least impact. All ablated features were semantically linked, suggesting they should be combined into a single dataset.

The thesis showcases the relevance of pixel-based classification with DL and data fusion, providing resources for future research and indicating areas for dataset expansion and improved annotation.
... Read more

The Earth's surface is a complex landscape that is essential for a wide range of applications, from urban planning to environmental monitoring. Digital models of the Earth's surface are generated through mathematical calculations using elevation data collected from various sources and the Digital Terrain Model (DTM) which captures the bare earth's surface topography in 2.5D. The creation of DTM is an approximation of terrain in unsample locations, by using x-y coordinates and one z value. Traditionally, terrain interpolation uses deterministic or geo-statistical methods to calculate elevation. This research would use random forest regression as an alternative method and to compare the results against traditional interpolation. Comparing different locations against traditional interpolation yields similar results overall. Feature importance, within the the points that are closest to the sampled ICESat-2 data point are more significant than other features used in Random Forest model. The correlation between these datasets and the spatial relationship established would impact on the results of the elevation. The improvement overall of using traditional interpolation compared to random forest regression is limited depending on the location and using model trained with local datasets. For model trained on other geographical locations, which shows similar differences. ... Read more

The thesis explores the semantic understanding of urban textured meshes derived from photogrammetric methods. It primarily addresses three aspects with regard to urban textured meshes: 1) semantic annotation and the creation of benchmark datasets, 2) semantic segmentation, and 3) the automation of lightweight 3D city modeling using semantic information.

The first focus of the thesis is the development of a benchmark dataset to evaluate the performance of advanced 3D semantic segmentation methods in urban settings. An interactive 3D annotation framework has been proposed to assign ground truth labels to the urban meshes' triangle faces and texture pixels. This framework achieves efficient and accurate semi-automatic annotation through segment classification and structure-aware interactive selection. In the center of Helsinki, Finland, object-level annotations were made over approximately 4 km\(^2\) (including buildings, vegetation, and vehicles, etc.), and part-level annotations over about 2.5 km\(^2\) (including building parts like doors, windows, and road markings, etc.). The design of the annotation tools improves user operation and enables quick annotation of large scenes, while the resulting datasets allow researchers to refine their deep learning models for urban analysis.

Another research focus is on mesh segmentation algorithms. A novel semantic mesh segmentation algorithm has been introduced for large-scale urban environments, employing plane-sensitive over-segmentation combined with graph-based methods for contextual data integration. This approach, which utilizes graph convolutional networks for classification, significantly improves performance over traditional techniques based on our proposed benchmark datasets.

Finally, leveraging this semantic information, a pipeline for reconstructing lightweight 3D city models has been designed. This facilitates the automated reconstruction of CityGML-based LoD2 and LoD3 city models, ensuring high fidelity in geometric detail and semantic accuracy. The reconstructed large-scale, lightweight, and semantic city models significantly broaden applications in urban spatial intelligence, including automatic geometric measurements, interactive spatial computations, spatial analysis based on external data, and environment simulation using physical engines.

This thesis enhances the practicality of 3D data in real-world applications by utilizing semantic parsing of urban textured meshes to generate lightweight 3D urban semantic models, greatly enriching their usability. It also lays a solid foundation for future progress in understanding, modeling, and analyzing 3D urban scenes. ... Read more