High-resolution aerial imagery plays a critical role in urban planning, energy mapping, and land-use classification. However, many datasets remain limited to lower resolutions due to acquisition costs or legacy data sources. Super-resolution (SR) techniques offer a means to enhance 25 cm aerial imagery to 8 cm, making it more suitable for object-level analysis. This thesis investigates the capability of a modified SRGAN architecture to enhance the visual and structural fidelity of aerial images, thereby improving the representation of urban features such as rooftops, dormers, and solar panels. The architecture incorporates an EdgeMaskBlock to improve edge awareness and preserve sharp contours in reconstructed imagery. To address the challenges of spatial complexity and temporal misalignment, a two-phase training strategy is implemented. First, the model is trained on synthetically downsampled HR-LR pairs to establish a robust initialization. This is followed by fine-tuning on real-world 25 cm inputs mis- aligned with their 8 cm HR counterparts, enabling the model to generalize under realistic and variable acquisition conditions. Evaluation is conducted across both training iterations using standard image quality metrics (PSNR, SSIM, LPIPS), along with downstream segmentation benchmarks. For Iteration 2, the gen- eralization capability of the model is assessed across new cities and seasonal conditions. Two segmen- tation pipelines are used: the Segment Anything Model (SAM) and the operational semantic segmen- tation system developed by Readar B.V., which detects buildings, dormers, and PV panels using both RGB and DSM data. Metrics such as precision, recall, and F1-score demonstrate that super-resolved outputs significantly outperform bicubic upsampling, particularly for fine-scale rooftop objects. The results show that the proposed SRGAN model improves perceptual quality while enabling effective domain transfer across seasons. These enhancements contribute to more reliable segmentation outputs, reinforcing the potential of GAN-based super-resolution as a practical tool in geospatial workflows that require fine-grained object recognition.

Urban wind simulations are essential for assessing pedestrian comfort, pollutant dispersion, and microclimate design, yet high-fidelity CFD remains computationally expensive. This thesis investigates the potential of a Swin-Transformer-based surrogate model to approximate steady-state 2D results for 100 m × 100 m tiles under a single inflow direction. A dataset of 690 urban tiles was extracted from the 3DBAG and simulated in Ansys Fluent at 0.5 m resolution across five inflow speeds (5–15 m/s) and five types of building layouts. Velocity fields were rasterized to 1 m grids. The surrogate architecture preserves the original Swin backbone, evaluated with three loss variants: original RMSE, buffer weighting, and a divergence penalty. Encoding wind as xy-component consistently outperformed magnitude-only training. Errors scaled linearly with inflow speed and were highest in Mixed and Attached urban forms. Architectural resolution was the dominant factor influencing accuracy since it introduced visible artifacts, while the buffer and divergence losses offered only marginal improvements. Limitations remain near building facades, where sharp gradients are smoothed by the patch-based architecture. Nonetheless, the surrogate offers potential for rapid wind flow estimation suitable for early-stage design or preliminary analysis.

Building heights are important information for a variety of subjects, such as wind analysis, energy demand simulations and solar potential assessment, yet large-scale LiDAR scanning is costly. This thesis introduces PHYSHADE: a set of physics-guided U-Net-based models. It employs shadow projections derived from building footprints and solar geometry into an aerial image shadow segmentation pipeline, for the purposes of building shadow extraction and consequently the large-scale estimation of building heights. Thirty-five aerial images in the Netherlands were manually annotated for buildings and their associated buildings. By employing transfer learning, based on a general purpose shadow-segmentation model, a total of 130 models were trained, which can be categorized into three different implementations of PHYSHADE. Through these different configurations, the performance impact of the various methods of addition of pseudo-shadows to the models was ablated. Afterwards, the best-performing PHYSHADE configurations were used with a raycasting algorithm to convert shadow lengths and solar altitudes back into building heights. The inclusion of pseudo-shadows lifted the mean Dice scores from 0.53 to 0.85, with an average gain of 0.32 and statistical significance across different folds. Physics-guided loss, based on the pseudo-shadows, was not found to be significantly different in most cases, whilst hurting model performance in some cases compared to the pseudo-shadow enabled models. On six out-of-fold test tiles the best PHYSHADE variants retained Dice scores of 0.72 – 0.95, although recall declined in one winter scene. Finally, height estimation on these tiles using the inference from the best PHYSHADE variants resulted in mean RMSE of ≅ 1.9m and MAE of ≅ 1.5m. While its application needs to be tested in broader contexts, PHYSHADE offers a viable low-cost complement to LiDAR for building height estimation.

Urban freight transport is a critical yet complex component of city logistics, shaped not only by transport networks but also by the morphological structure of urban areas. Traditional forecasting models often neglect this spatial heterogeneity, relying primarily on traffic counts or infrastructure topology. This thesis proposes and evaluates ST-SimNet, a Spatio-Temporal Simulation Network designed to enhance freight flow prediction by integrating static urban morphology descriptors with dynamic freight data in a graph neural network framework. Focusing on the city of Amsterdam, the study explores the extent to which detailed urban morphology, including building features, land use, and spatial layout, can improve short-term freight flow forecasts at the road network level. Results demonstrate that incorporating static features significantly reduces error variance, improves peak hour prediction, and enhances node-level stability compared to dynamic-only baselines. Furthermore, analysis reveals that nodes with richer morphology information benefit most, while areas with sparse or noisy static features experience challenges that highlight opportunities for future refinement. The findings offer practical insights for integrating machine learning into digital twin platforms for urban mobility, providing a data-driven, spatially aware layer for freight forecasting in operational city planning systems. Limitations and future directions, including adaptive fusion mechanisms and cross-city generalisation, are discussed. Overall, ST-SimNet advances the integration of urban morphology into spatio-temporal predictive models and demonstrates its practical relevance for modern freight planning in complex urban environments.

High-quality 3D mesh models are increasingly used in urban applications ranging from planning and simulation to environmental monitoring. While aerial imagery provides a practical balance between resolution and coverage, conventional Structure from Motion (SfM) and Multi-View Stereo (MVS) pipelines apply uniform processing to all image regions, often overlooking their geometric relevance. This research investigates whether reconstruction efficiency can be improved by selectively focusing dense matching efforts on image regions most critical to mesh quality. We explore the use of the Segment Anything Model (SAM) to guide dense matching in oblique aerial imagery, applying it for building-level segmentation and importance estimation via entropy and edge-based scoring. Initially, we tested SAM for sparse reconstruction guidance but encountered poor performance. We then shifted to evaluating whether SAMderived importance maps could enable region-aware thresholding to improve mesh reconstruction. These methods were benchmarked against a simpler Canny edge-based distance approach across varying thresholds and scenes. Results show that SAM-based methods can reduce memory usage while maintaining mesh quality at low to moderate thresholds (up to 0.4). However, contrary to expectations, Canny edge detection consistently outperformed SAM across most quality and efficiency metrics, offering better spatial coverage, lower computational overhead, and more stable performance. While SAM-based thresholding led to file size reductions of up to 16% and marginal runtime gains during dense matching ( 2.3%), these were offset by the costly preprocessing pipeline required to generate segmentation masks and importance maps. Overall, this thesis contributes an evaluation pipeline for image region-specific reconstruction strategies and highlights that while SAM shows potential for memory-efficient modeling, simpler methods like Canny edge detection may offer better trade-offs for scalable, time-sensitive workflows. Future research should focus on faster importance estimation techniques and pipeline adaptation for more complex, city-scale datasets.

Urban climate affects how people move through cities, but its influence is difficult to capture with models based on generalized comfort indices that ignore individual experiences of climate. This thesis instead explores a bottom-up approach that uses daily Global Navigation Satellite System (GNSS) traces of people traversing an urban environment, which inherently contain each individual's personal influences on their mobility. A machine learning model was developed and trained using this dataset with the purpose of predicting future mobility values, while assessing the role that climate played in such predictions. The model employed a Spatio-Temporal Graph Neural Network (STGNN) architecture to capture both potential spatial dependencies between visited locations and temporal patterns in their activity.

The work draws on the Seoul Cozie dataset, which recorded six weeks of GNSS location data from wearable devices of 22 participants in Seoul during autumn 2023. Positions were aggregated into a graph structure with road intersections as nodes and transitions between them as edges. Climate features (temperature, humidity and PM10) were interpolated from over 1,000 weather stations using a Triangulated Irregular Network method and added as dynamic node features. STGNN variants were trained and compared based on whether they included climate node features.

Results show forecasts of node visits with low Mean Square Error of around 0.12 . However, precision and recall values for visited/unvisited node detection are low, peaking at 56.51%, reflecting strong class imbalance in the input. Adding climate attributes produced only minor and inconclusive improvements, in part due to the dataset’s short time span. The thesis proposes a reproducible framework linking climate and mobility, while underlining the need for richer datasets and for more flexible model architectures, capable of addressing class imbalances and representing personal mobility datasets.

This study investigates the relationships between urban form elements (UFEs), Urban Heat Island (UHI) effects, and building energy consumption in Heukseok-dong, Seoul, South Korea. Using 10 years of satellite imagery, air temperature measurements, and energy consumption data, the research examines how six UFEs (NDVI, building ratio, building height, building volume, FSI, and GSI) influence air temperatures and building energy consumption across multiple spatial scales and seasons. A Multi-Layer Perceptron (MLP) neural network was developed to convert satellite-derived Land Surface Temperature (LST) to air temperature, achieving an R2 of 0.9684, and tested with independent S-DoT sensors. The independent testing was conducted in two phases: for May-August 2020, the mean temperature difference was 1.84°C, remarkably close to the known systematic temperature difference of 1.8°C between S-DoT sensors and AWS. When extended to 2020-2024, the mean
difference was 0.98°C with an R2 of 0.807, confirming the model successfully predicts actual air temperatures rather than sensor-specific values. The Genizi method and partial correlation analysis were combined to address multicollinearity while revealing both relative importance and directional effects of UFEs. This complementary approach provides more comprehensive insights than traditional regression methods alone. Key findings reveal that NDVI dominates temperature variance in spring (79.3%), fall (64.7%), and winter (71.6%), but building characteristics become more important in summer, with building ratio contributing 71.8% at pixel scale. Scale-dependent patterns emerged, with energy consumption best captured at 100m scale (R2 up to 0.378) while temperature variations appeared more clearly at 300m scale (R2 up to 0.328). The cascade relationship from UFE through air temperature to energy consumption showed air temperature driving 54.3% of electricity variance in summer, while building volume consistently influenced both electricity and gas consumption despite EUI normalization. A decade-long analysis of District 3’s transformation from 478 low-rise buildings to 28 high-rise apartments confirmed the statistical findings. Despite a 2,112.6% increase in building volume and 168% improvement in NDVI, temperature trends showed 0.10-0.16°C/year increases, which are approximately half of Heukseok-dong’s 0.17-0.32°C/year rates, demonstrating that urban design can partially mitigate but not eliminate warming effects. The moderate R2 values (0.067-0.378) indicate that urban form elements explain only a portion of variance, partly reflecting the temporal limitation of correlating single hourly satellite observations with monthly energy totals. The research provides evidence-based recommendations for urban planning policies, including maintaining GSI below 0.55, achieving NDVI above 0.15, and implementing seasonal strategies for temperature mitigation and energy management.

This research explores how deep learning models can improve the detection of efflorescence in masonry buildings in the Netherlands. Efflorescence, caused by moisture-driven salt transport, poses detection challenges due to its variable appearance, similarity to other surface features, and co-occurrence with other forms of masonry damage.
To address this, the study benchmarked two state-of-the-art object detection models: Mask R-CNN and YOLOv8. On a single-class dataset, Mask R-CNN achieved a peak mAP@0.5 of 0.35, outperforming YOLOv8 (0.30–0.33) in segmentation quality and spatial precision. However, both models suffered from false positives, often misclassifying encrustation, lichens, and graffiti as efflorescence due to visual similarity.
To mitigate this, a multi-class training setup was introduced. Graffiti achieved the highest mAP (0.60) and near-perfect precision due to strong visual contrast, while lichens were classified with high stability. In contrast, efflorescence and encrustation remained difficult to separate, resulting in unstable mAP and precision fluctuations over time. This confirmed that misclassification significantly limits model accuracy when damage types share visual characteristics.
Model performance was further evaluated by incorporating thermal imaging (RGBT), combining aligned RGB and infrared data to detect moisture-driven efflorescence. The RGBT model reduced false positives (as few as 3 per evaluation set) and improved detection confidence, reaching a precision of 0.94 and an average confidence score of 0.96, although it required more epochs to converge and showed increased false negatives in ambiguous scenes. Still, RGBT improved the confidence of the visual detection in real-world, poorly lit, or heritage conditions, where over-segmentation is costly.
A spatial co-occurrence analysis of annotated masks indicated a statistically significant correlation between efflorescence and adjacent damage, supporting the potential of dual-class detection
In conclusion, while deep learning models can support efflorescence detection, especially when enhanced with thermal input and multi-class refinement, their performance depends heavily on dataset quality, annotation strategy, and class separability. These findings offer a foundation for scalable, automated inspection in conservation and diagnostics of masonry.

Daylight is a significant factor in the decision-making process of urban planners and architects when intervening in an urban environment. A considerable element in such a complex environment, assessed in daylight simulations, is greenery, particularly trees, due to their interaction with sunlight and their shading effects.
This thesis investigates the impact of diverse tree 3D representations produced by Actueel Hoogtebestand Nederland (AHN) point cloud data when performing daylight simulation on scenes integrating them. The assessed tree 3D representations are point cloud based representation using small cubes instead of the points from the original point cloud, voxels, alpha shapes of individual trees, and convex hull of branch points. The tool for the daylight simulation was the Daylight Availability Workflow of Climate Studio plugin in Rhino
software which produces time series of simulated illuminance values based on the location and the local solar data, the scene, the material properties of the objects of the scene and surface on which simulation is performed. Specifically, regarding the assigned materials for the voxel presentation, two approaches were followed; one related to the predefined opaque material properties in Rhino software and the other to transparent materials defined by the density point inside a voxel. The reference of the simulations was a dataset of actual illuminance values recorded by the sensor located on the west facade of the CCC building, where is the
position that the simulation was conducted. Also, to examine how the seasonal changes influence the simulation results, simulations for two months, February and June, were performed. For the June simulations, a synthesized point cloud was generated by combining the AHN points with additional points representing the tree canopy.
Next, results from all days in both months the results showed that the point cloud based representation caused a significant overestimation of simulated illuminance values, whereas the Alpha Shape and Convex Hull representations resulted in underestimation. In contrast, the simulation outcomes for voxel representations depended on their material properties (opaque or transparent), spatial allocation, and size. It was proven that for February via the voxel representations of sizes close to 0.10 m there was the best fit between simulations and sensor data, yet it is not clear which voxel material is the most optimal for all the different
sky conditions. However, in June the simulations were not as accurate as in February, as the synthesized point cloud representation probably did not include a sufficient number of points and as the simulation system performed less accurately for days under clear sky conditions.
Consequently, the study demonstrates that voxel representations with sizes around 0.10 m provide the most reliable results for tree modeling in daylight simulations in February, while clarifying the limitations that gave rise to to the poorer performance of alternative representations and of June simulations.

Fields like geology, mining, and construction have long relied on manual rock and soil classification methods, which remain time-consuming and labour-intensive. Accurate rock and soil classification is necessary for hazard assessment, urban management or resource exploration, which the rapid advances in artificial intelligence and remote sensing might facilitate. Though optical imagery has dominated remote sensing in the last decades, thermal data offers unique advantages such as sensitivity to material properties, sunlight independence or seasonal and diurnal changes responsiveness. This study explores the use of thermal imagery, along with SAR and NDVI data as complementary data, to leverage and automate the classification process with deep learning. Therefore, the aim is to develop two deep learning models capable of rock and soil segmentation with multi-modal datasets, and evaluate the performance of these models, Convolutional Neural Networks (CNNs) and Convolutional
Long Short-Term Memory networks (ConvLSTMs) across diurnal, seasonal and multi-source
scenarios as well as the effect of vegetation on the prediction results.
Results show CNN tends to have a strong overall performance, especially with thermal and SAR data. Instead, ConvLSTM excels at capturing temporal dependencies, improving the classification of most datasets and variations. The findings demonstrate the potential of thermal imagery to be used as a powerful classification tool when combined with deep learning methods for rocks and soils. They also demonstrate the slightly negative influence of vegetation on the predictability of the outputs. By combining spatial and temporal dynamics, the models offer an automated, scalable, fast and cost-effective approach to more traditional workflows. This work intends to contribute towards modernizing geological practices for more informed decision-making in geology, mining, urban planning, hazard and resource management.

The building sector represents the highest share of operational energy consumption across all sectors, with a significant portion attributed to the inefficiency of the existing building stock. In this context, building retrofit plays a crucial role in enhancing energy efficiency and reducing environmental impact. However, conventional models for assessing retrofit scenarios are highly computationally expensive, thereby slowing down the retrofit process. This research addresses this challenge by developing an AI-based surrogate model using Multi-Task Learning (MTL). The proposed MTL model significantly reduces computational costs while simultaneously predicting energy consumption, costs, embodied carbon, and thermal comfort. Additionally, Multi-Objective Optimization (MOO) and Multi-Criteria Decision Making (MCDM) techniques are employed to select optimal retrofit solutions Results demonstrate that the MTL model accelerates the retrofit simulation process from 90 minutes to just 2 seconds, highlighting its potential to streamline and enhance retrofit decision-making processes.

As urban architectural environments become increasingly complex and densely populated, the demand for precise registration of legal statuses, encompassing both private and public interests, has become more urgent. Traditional 2D cadastral registration systems are increasingly inadequate for addressing the multifaceted and vertical nature of modern urban landscapes. These systems are limited in scope and unable to fully capture the intricacies of multi-level property rights, overlapping parcels, and underground constructions. This study uses a BIM/IFC model for the building’s physical representation. The party and Rights, Restrictions, and Responsibilities (RRRs) data are stored in a DBMS, with all data and the building location being fictitious. Visualization is achieved in 3D over the web using Cesium JS, an extensible globe viewer. Unlike earlier 3D cadastral systems, this research has developed a new 3D Land Administration prototype based on ISO 19152-2 (Land Administration Domain Model (LADM)). The objective is to explore improved methods for analyzing and visualizing RRRs in complex buildings. Novel techniques include presenting UML instance-level LADM diagrams for selected parties and/or apartments, showing RRRs and BAUnits linking them. The study introduces a new method for displaying surrounding buildings at varying Level of Detail (LoD), with closer buildings rendered in higher detail and more distant buildings shown in less detail. This selective detailing enhances both performance and clarity in visualizations. A key feature of this digital twin system is its real-time update capability. This study supports the updating of party and rights information in the backend database, accurately reflecting these updates in the public front-end version. This ensures the maintenance and visualization of the most current property rights data. The system also integrates sunlight simulation, which is crucial for urban planning, architectural design, and aiding buyer decision-making. For more details, visit the project’s website 3D web-visualization platform.

Balancing the requirements of urban planning with the need for rapid and efficient iterative design, while accurately assessing its detailed impact on existing surrounding buildings at a per-room level, presents a significant challenge in architecture design, the conversion and integration between multiple platforms introduce additional complexity, establishing high standards for designers to achieve. Integrating Building Information Modeling (BIM), Geographic Information Systems (GIS), and per-room daylight simulation into a unified workflow offers a promising solution. This thesis develops a web application that leverages Grasshopper(GH) scripts for backend processes, Rhino Compute as the intermediary API, andMapboxGL as the GIS-based front-end platform, enabling seamless user interaction with BIM data for customized daylight simulations.

Since the advent of digital tools, architects have sought ways to enhance and streamline the design process. Integrating these technologies bridges the gap between urban planning and indoor daylight simulation, facilitating real-time interaction and data manipulation. Grasshopper scripts are developed to compute Aperture-BasedDaylight Modeling (ABDM) metrics, Sunlight Beam Index (SBI), and sky lumen, providing a foundation for evaluating their relationships with traditional Daylight Modeling (CBDM) metrics like Daylight Autonomy (DA), Spatial Daylight Autonomy (sDA), and Useful Daylight Illuminance (UDI).

This research explores the potential of combining these advanced tools to create an interactive web application that allows users to manipulate building typology, location, scale,façade materials, and levels of detail (LOD) for surrounding city models. The study demonstrates a robust linear relationship between SBI and the daylight metrics DA, sDA, and UDI, with room geometry playing a crucial role. Skylumens, however, show limited effectiveness as a CBDM alternative.

The thesis concludes that MapboxGL is a promising platform for integrating BIM models into room-based daylight simulations, offering an intuitive user interface for urban planners and designers. Rhino Compute technology ensures smooth data transfer and interaction, significantly enhancing the workflow. The study results show that SBI has a linear relationship with traditional CBDM metrics. However, multiple moderating variables affect this relationship, and it does not apply to north-facing rooms. Thus, SBI is not a current substitute for traditional CBDM metrics. Overall, this thesis provides a comprehensive framework for integrating BIM, GIS, and daylight simulation into a cohesive system, paving the way for more efficient urban planning and design practices.

The transition to a fully electrified energy grid in the Netherlands is essential for the country's energy transition strategy. However, the existing grid infrastructure struggles to support the increased demand due to electrification, leading to congestion and potential outages anticipated before 2030. Microgrids, decentralized networks where generation, heat, storage, and consumption are coordinated, offer a promising solution by providing the required flexibility for the integration of Distributed Renewable Energy Resources (DERs) and potentially reducing energy losses associated with long-distance energy exchange.

This study explores the implementation of Deep Reinforcement Learning (DRL) in a district-level microgrid in the Netherlands. The primary objective is to minimize grid connection dependency and reduce energy exchange distances within the microgrid cost-effectively, thereby enhancing grid reliability and sustainability, answering the research question: \textit{Main Research Question}: \emph{How can Distributed Energy Resources be deployed and managed in a cost-effective manner within an electrified microgrid to achieve a balance between energy consumption and production in order to minimize the burden on the central grid given fluctuating demand?}

A comprehensive framework was developed offering a structured approach for simulating and optimizing microgrid scenarios, providing practical solutions and frameworks for the efficient management and deployment of microgrids. The study demonstrates that DERs can be effectively deployed and managed within an electrified microgrid through a combination of simulation techniques with geographical data input, a DRL model with a Deep Q-Network (DQN) that controls the system, and dynamic mappings and visualization. While the DQN demonstrated its potential in optimizing microgrid configurations, challenges were noted due to the large action spaces required for routing optimization. To overcome these limitations, future work is proposed to combine DRL with Graph Neural Networks (GNNs). This novel method shows promise in enhancing the scalability and efficiency of the optimization process, enabling distance and routing optimization without an aggregated model.

Scenarios were created to test the influence of different components, demonstrating how various factors affect overall system performance. The results of the model were as expected, and undertakings to optimize the model further led to significant improvements in the variability of the learning curve and of the rewards.

The study concludes that the developed pipeline contributes to existing works by offering a structured approach for simulating and optimizing microgrid scenarios and testing different setups. While this research approach has proven effective, future work should enhance the realism of components and address some computational limitations encountered. Refining these methodologies by incorporating GNNs, integrating real-world data, and extending the model with additional algorithms for performance comparison will further advance the field. This effort ultimately supports the vision of a reimagined, sustainable, resilient, and efficient energy grid.

Local Climate Zone (LCZ) classification plays a crucial role in understanding and managing urban environments, particularly through the lens of Land Surface Temperature (LST) behavior. This study investigates the effectiveness of a Convolutional Neural Network (CNN) with U-net architecture for classifying urban LCZs using spatio-temporal thermal imagery.

The research addresses several key sub-questions, including the creation of a representable training dataset, optimizing hyperparameters for the U-net model, and assessing the impact of temporal factors on classification performance. An unsupervised clustering approach was adopted to label the training data, utilizing the Iterative Self-Organizing Data Analysis Technique (ISODATA) algorithm on stacks of thermal images to generate clusters based on thermal behavior, which were then manually refined and validated. Through experimentation, optimal hyperparameters were identified: a learning rate of 0.001, a patch size of 64, and the SparseCategoricalCrossentropy loss function. The study also highlights the significant influence of temporal factors, when using daytime and Spring/Summer thermal images for training and testing the model better classification outcomes were obtained compared to nighttime and Autumn/Winter images.

The research contributes to the LCZ classification by incorporating both spatial and temporal dimensions of LST patterns, providing valuable insights for urban planning. The findings demonstrate that a CNN with U-net architecture is highly suitable for classifying urban LCZs, particularly when the dataset captures diverse seasonal and extreme conditions. This approach offers a robust and adaptable framework for urban environmental monitoring and planning. This thesis has explored the utilization of a new source for LCZ classification, providing a useful starting position for further enhancement of standardized LCZ classification. Future work is therefore recommended to focus on integrating additional geospatial data sources, refining classification categories, and integrating the standardized LCZ classification system by Stewart and Oke [2012].

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.

Computational Fluid Dynamics (CFD) simulations for residential areas are becoming increasingly important as the urban population keeps growing and extreme weather conditions due to climate change are becoming more prevalent. Since CFD simulations simulate and visualise fluid flows, users can analyse and predict air flows in urban environments, giving insight in pollution, heat dissipation, and weather conditions. Performing these simulations demands expertise and time: 3D urban models must be prepared, and pre-run setups must be defined to obtain accurate results. Partial automation can streamline this process. Therefore, we developed a method that identifies geometric errors and defines mesh parameters for the open source CFD software OpenFOAM. This method follows the ISO19107 standard and recent CFD guidelines for urban areas, ensuring accurate simulation results. We validated algorithms with a variety of urban models and parameters, and analysed the workflow developed for the mesh parameters definition. Additionally, we created a prototype, in the form of a web application, in which these algorithms are implemented. Our prototype will simplify the use of CFD simulations for urban areas, making them more accessible to everyone.

This thesis presents the development of a real-time daylight analysis tool for architectural and urban development using the Unreal Game Engine. The tool offers architects and urban planners a fast and precise way of analysing outdoor daylight conditions in their designs. The Unreal Game Engine provides real-time visualisation and analysis of daylight conditions, which makes it an effective tool for real-time decision-making during the design process. The study compares the light values and the process of extracting these values in Unreal with the validated light model Radiance used in Honeybee & Ladybug in the visual scripting program Grasshopper for Rhino. The thesis compares the values of Unreal & Honeybee/ladybug based on outdoor illuminance values, calculation time and ease of use. The comparison demonstrates the potential of Unreal as a valuable daylight analysis tool, with measurements showing a Mean Absolute Error of 9.78% between Honeybee and Unreal. In terms of computational time, the Unreal application requires only 0.6ms to execute and recalculate the daylight analysis, whereas the complete Honeybee script took an average of 93 minutes to calculate daylight values, and a new angle calculation took approximately 645 seconds. Furthermore, unlike Honeybee, which is sensitive to the complexity of urban or complex geometries and requires challenging adjustments, the Unreal application effortlessly accommodates complex geometry without the need for extensive modification.
The thesis concludes that the tool provides a robust and efficient method for analysing daylight conditions in architectural and urban design. The tool's ease of use and real-time visualisation capabilities make it an essential addition to the design workflow. Finally, the thesis presents the tool as a proof of concept for a geospatial urban development platform with built-in geospatial analysis. The research has demonstrated the potential of real-time simulation and analysis using the Unreal Game Engine as a powerful tool for architects and urban planners.

The production of cocoa beans contributes to 7.5% of European Union (EU) driven deforestation. For this reason, the recent European Union Deforestation-free Regulation (EUDR) requires producers to perform comprehensive tracking of cocoa farm extents. However, cocoa crops present unique detection challenges due to their complex canopy structure, spectral similarity to forest, variable farming methods, and location in frequently cloudy regions. Previous work employs Multispectral Imagery (MSI) and/or Synthetic Aperture Radar (SAR) for pixel-based classification of satellite images. Convolutional Neural Network (CNN)s offer
a promising approach to semantic segmentation of cocoa parcels that considers = both spectral and spatial characteristics. This thesis aims to evaluate the impact of combining SAR and MSI data in the training of a CNN for cocoa detection, in order to demonstrate the importance of texture, moisture and canopy characteristics in identifying cocoa canopies. A U-NET is employed to evaluate
how prediction results are impacted by the stacking of MSI datasets with different SAR polarizations, seasons and temporality. The results show that the addition of single-day and temporal SAR to a single-day MSI image can improve the predictions, reaching an F1 score of 86.62%. This research demonstrates the influence of SAR measurement season and polarization, and ground truth classes, on the semantic segmentation of cocoa.

*Public version of internship report*
Urban wildlife plays an invaluable role in cities, including the promotion of overall biodiversity and greenery, pollination and human connection to nature. This internship project involved the development of a prototype that quantifies the potential for local ambassador species to visit a building site based on the amount and connectivity of habitat in the neighbourhood. The workflow integrates vector layers from open datasets and Arup ecologists’ expertise on animal behaviour in order to quantify the cost of moving through an urban environment. Initial results show that the prototype enables the numerical and visual comparison of connectivity for 7 out of the 10 ambassador species, with the Bee demonstrating the highest connectivity, and Toad having the lowest connectivity. This prototype has the potential to support Arup’s work on facilitating more connected urban environments for local wildlife, all the while improving urban ecology overall.