From Data to Decision: A Semantic and Network-Centric Approach to Urban Green Space Planning

Abstract

Urban sustainability poses a deeply interdisciplinary challenge, spanning technical fields like data science and environmental science, design-oriented disciplines like architecture and spatial planning, and domains such as economics, policy, and social studies. While numerous advanced tools are used in these domains, ranging from geospatial systems to AI and network analysis-, they often remain fragmented, domain-specific, and difficult to integrate. This paper introduces a semantic framework that aims not to replace existing analytical methods, but to interlink their outputs and datasets within a unified, queryable knowledge graph. Leveraging semantic web technologies, the framework enables the integration of heterogeneous urban data, including spatial, network, and regulatory information, permitting advanced querying and pattern discovery across formats. Applying the methodology to two urban contexts—Thessaloniki (Greece) as a full implementation and Marine Parade GRC (Singapore) as a secondary test—we demonstrate its flexibility and potential to support more informed decision-making in diverse planning environments. The methodology reveals both opportunities and constraints shaped by accessibility, connectivity, and legal zoning, offering a reusable approach for urban interventions in other contexts. More broadly, the work illustrates how semantic technologies can foster interoperability among tools and disciplines, creating the conditions for truly data-driven, collaborative urban planning.

1. Introduction

As urban environments grow increasingly complex and data-rich, the demand for integrative methodologies that can synthesize diverse types of urban information becomes ever more urgent. Urban sustainability today is not merely a matter of efficient infrastructure or environmental metrics; it is a multifaceted issue involving governance, human behavior, spatial configuration, regulatory frameworks, and the lived experience of place [1,2]. Urban planning, therefore, must move beyond disciplinary silos and adopt approaches that embrace this complexity rather than reduce or simplify it.

Traditional methods for analyzing urban sustainability, including static measures such as green area per capita or population density, are useful for benchmarking but often fall short in revealing the deeper spatial logic of a city, which includes how movement, accessibility, and connectivity shape opportunities for intervention [3,4]. At the same time, the tools available for planners and researchers are often fragmented across domains. Geospatial platforms such as GIS offer spatial querying and visualization, supporting urban analysis, infrastructure management, and spatial planning [5]; network analysis provides insights into urban flows and structural connectivity [6,7,8]; regulatory data are embedded in legal documents or loosely structured databases—formats that are typically not machine-readable and poorly integrated with spatial tools [9]. These formats rarely speak to each other in a coherent or scalable way.

The recent literature on urban networks has emphasized how topological metrics and spatial logics are increasingly used to evaluate urban infrastructure, mobility, and green space provision [6,7,8]. Our approach builds on this growing body of work by situating network analysis not as an isolated technique, but as an integral layer within a semantically governed framework. Comparable efforts, such as Wang et al. [4] or Binopoulos et al. [6], demonstrate the need to align network metrics with planning-specific criteria, a direction that we extend through ontological formalization and regulatory reasoning. Rather than simply mapping network flows, we seek to embed those flows within legal and spatial systems to enable cross-domain insight.

Rather than searching for a single superior method or tool, this paper advocates for semantic technologies as a coordinating layer, not to replace domain-specific approaches (e.g., spatial or network analysis, traditional tools, or AI-driven techniques), but to serve as the semantic glue that binds these diverse tools into an integrated, extensible framework. The semantic web, and particularly ontologies, offers a powerful paradigm for representing urban knowledge in a way that is machine-readable, interoperable, and adaptable [10,11]. This enables richer analysis, cross-domain reasoning, and greater transparency in urban decision-making.

In this study, we develop and apply an ontology-driven methodology that integrates geospatial data, street network centrality metrics, and urban regulatory constraints to assess and enhance urban green spaces (UGSs). By converting spatial and analytical outputs into OWL-based ontologies, we enable semantic querying, classification, and filtering across heterogeneous data sources. While other approaches position ontologies as supporting components to enhance semantic interoperability within GIS, typically functioning as data enrichment layers or modular extensions [12], our approach re-centers semantics as the foundational environment through which geospatial, regulatory, and analytical tools interoperate, building on the perspective introduced by Silvennoinen et al. [13], which emphasizes ontologies as core infrastructures rather than adjuncts. This model not only supports inference and rule-based filtering but also enables modular integration of diverse tools, including GIS, network analysis (sDNA), regulation-based constraints, and others, ensuring adaptability to varied analytical needs.

The methodology is demonstrated through a dual-case study approach, applied to two urban contexts with similar spatial footprints but differing planning structures: the Municipality of Thessaloniki in Greece and Marine Parade GRC in Singapore. Thessaloniki serves as the primary implementation site, where the full semantic-network workflow was applied, including spatial, network, and regulatory layers. Marine Parade functions as a secondary test case to evaluate the method’s adaptability across planning systems and data environments. While Thessaloniki represents a dense, historically evolved urban fabric with fragmented green spaces [2,3], Marine Parade offers a well-planned, infrastructure-integrated neighborhood embedded within Singapore’s highly structured regulatory ecosystem [14]. Through both applications, we assess how network centralities can inform green space prioritization and how regulatory constraints affect spatial opportunities for intervention. This dual lens also enables us to explore how urban form, particularly street network configuration, density distribution, and land-use patterns, mediates the relationship between green space provision and sustainable urban outcomes.

The contributions of this work are threefold. First, it proposes an ontology-based framework for integrating spatial, regulatory, and network data, showing how diverse analytical tools can interoperate without being reduced to the lowest common denominator. Second, it demonstrates how network centrality metrics can be semantically linked to other urban elements (such as parks) to evaluate or guide planning decisions [15,16]. Third, it shows how policy-relevant reasoning (e.g., identifying only legally buildable sites near high-scoring streets) can be operationalized through semantic filtering and classification. For urban planners and practitioners, the value of this approach lies in its ability to connect disconnected data sources like zoning rules, spatial maps, and network connectivity into a coherent and searchable model. Instead of manually comparing different files and formats, professionals can ask structured questions (e.g., “Which sites near well-connected streets are legally buildable as parks?”) and obtain immediate, meaningful answers. This reduces guesswork, increases transparency, and supports more informed and coordinated planning decisions.

This study aims to develop and test a decision-support methodology that combines semantic filtering and spatial network analysis to evaluate and prioritize urban green spaces (UGSs) in a way that is both spatially and legally grounded. By integrating regulatory, topological, and morphological data into a shared ontology, the proposed approach enables context-sensitive reasoning and policy-aligned planning support. For example, a site adjacent to a highly accessible street may be deprioritized if it lies within a restricted zoning area, whereas a smaller but legally developable site near a high-centrality corridor may emerge as a planning priority. These distinctions are made possible through rule-based classification embedded in the semantic model, allowing outputs to reflect both spatial performance and legal feasibility.

Beyond its immediate analytical utility, this work exemplifies the broader evolution in urban data science from siloed toolsets toward interoperable, knowledge-based infrastructures capable of adapting to the complexity of contemporary cities. By embedding diverse analytical outputs within a shared semantic framework, it becomes possible to reason diverse dimensions in a transparent, modular way. This approach not only enhances analytical depth but also improves the scalability and reproducibility of urban interventions. In doing so, the methodology contributes to the development of semantic-rich, data-driven planning tools that support integrated urban decision-making across heterogeneous data environments, an increasingly critical priority in the advancement of intelligent urban systems [13].

2. Conceptual and Methodological Background: Toward an Integrated Planning Framework

Urban systems are inherently complex, shaped by physical infrastructure, socio-political dynamics, environmental processes, and human behavior. Their evolution is influenced not only by planning decisions but also by histories of policy, economic patterns, and cultural change [17,18,19]. This complexity exposes the limits of traditional siloed planning, which often overlooks cross-sectoral interdependencies.

Contemporary urban challenges such as climate change, housing shortages, congestion, biodiversity loss, and unequal access to services like green spaces are tightly interwoven [20]. Addressing one often triggers consequences in another, leading to so-called “wicked problems” [21]. UGSs exemplify this dilemma. Though essential for public health, resilience, and wellbeing [22], parks can fail to deliver their full benefits unless they are accessible, legally viable, and environmentally integrated [23]. Effective planning must therefore account for multiple dimensions at once, including spatial distribution (e.g., park size, proximity to underserved communities), regulatory context (e.g., zoning and ownership), connectivity (e.g., pedestrian access), and ecological concerns (e.g., flood zones, habitat corridors). These demands call for interdisciplinary collaboration and tools capable of synthesizing diverse data types [24].

Our study meets this need by adopting an integrated framework that merges spatial analysis, network theory, and legal ontologies. We treat green space planning not as a standalone policy task but as a complex system in which land use, infrastructure, and regulation co-evolve, echoing principles from urban complexity science [18].

A major barrier to such integration is data fragmentation: legal texts in natural language, mobility data in networks, and spatial layouts in GIS often exist in incompatible formats [4,25]. Semantic web technologies, particularly ontologies, help bridge these gaps by offering shared vocabularies and logical rules that enable machine-readable, interoperable analysis [26]. These tools have already supported domains like environmental planning [27,28], energy systems and urban resilience [29].

Beyond these domains, semantic web tools, including ontologies and knowledge graphs, have been widely adopted across a range of fields, demonstrating their effectiveness in handling complexity. In medicine, for instance, they are used to develop semantically enriched learning environments, support mixed reality educational experiences, and improve knowledge retrieval processes [30,31,32]. These tools are also central to law and forensic science, where they enhance crime scene analysis through forensic ontologies and assist in the semantic representation of complex legal concepts [33,34]. In the sports domain, semantic web technologies are employed to facilitate the multidimensional classification and analysis of sports data [35]. Broadly, they play a critical role in data alignment and ontology matching, which ensures semantic consistency across diverse knowledge systems [36]. Additionally, semantic web technologies have been utilized in internationalized linked data projects, such as the Greek DBpedia, showcasing their scalability for multilingual data interoperability [37,38]. They also extend to specialized areas like mathematics [39] and national strategic frameworks [40]. This wide array of applications underscores the semantic web’s capacity to handle intricate datasets and complex systems across various fields, making it an essential tool for interdisciplinary research and problem-solving.

In urban contexts, ontologies have been applied to city planning [41], regulation compliance [42], and smart city infrastructures [43]. Projects like CityGML [44] and The World Avatar [45] demonstrate how ontologies can enable dynamic, machine-interpretable models of cities.

Building on these precedents, our framework integrates the following three key data types:

•

Spatial properties of parks and streets from GIS datasets;

•

Street network metrics (e.g., centrality from graph theory);

•

Regulatory constraints from zoning and planning rules.

Ontologies do not replace these domain-specific tools but allow them to “speak” to one another, supporting multi-scalar reasoning and complex queries across domains. This avoids the need for cumbersome manual integration or multiple disconnected tools.

One focus of our analysis is the application of street network analysis to UGS planning. Traditional metrics like green space per capita often ignore how people move through cities [22]. By contrast, graph-based metrics, like betweenness and closeness centrality, reflect how integrated a location is within the urban grid, revealing structural imbalances in accessibility [46,47]. Empirical studies support this approach. Barbosa et al. [48] and Kalwar et al. [49] show the need for network analysis over traditional evaluation methods. Kropf [50] emphasizes that the spatial logic of networks shapes livability as much as land use.

In our study, network analysis allows us to assess which parks are well connected and which remain invisible or underused due to poor integration. Parks along streets with high centrality values may attract more users, while those in disconnected areas often go unnoticed, regardless of size or quality, a pattern linked to urban morphology and accessibility [47,51].

Ultimately, street network analysis complements spatial and policy tools by adding a functional perspective. It helps identify locations for new parks, opportunities to connect existing ones, and ways to enhance access via the street grid. Whether applied in the early stages of planning or retroactively to improve existing conditions, it supports broader goals of equity, sustainability, and urban resilience.

3. Data Disclaimer and Methodological Decisions

As urban systems become more data-intensive, the way we define, access, and structure spatial data plays a critical role in shaping planning outcomes. This section outlines the key methodological decisions behind our analysis, particularly regarding how we define UGS within a semantic framework, what data was available or lacking, and how broader governance cultures influenced our data strategy. These decisions are both technical and conceptual, reflecting not just how cities are measured but how they are understood and governed.

A central choice was how to define “parks” within the study. Although the term seems straightforward, its interpretation varies widely across institutional, legal, and mapping contexts [52]. For clarity and consistency, we limited our definition to officially designated public parks, excluding semi-public gardens, green verges, sports fields, and private open spaces [53,54]. This constraint ensured analytical precision and policy relevance, aligning with zoning classifications and urban regulatory frameworks that determine land use and public investment. While this approach excludes informal or incidental green areas, which may be ecologically or socially important, it enabled more consistent semantic modeling. This decision aligns with the broader call in UGS research for standardized definitions to enhance comparability and support effective policymaking.

This definitional rigor was critical given the different semantic interpretations between different contexts. In cities like Singapore, green-looking land may include infrastructure buffers or undeveloped state land that appears public and accessible but is neither designated as a park nor fully open to the public. These areas are often intended for future development rather than for recreation or conservation purposes [55]. In Thessaloniki, the urban landscape is characterized by a mosaic of small, fragmented green spaces that often exist outside formal planning frameworks. These spaces, while not officially designated, serve as vital components of the city’s green infrastructure, reflecting a complex interplay between informal community initiatives and the broader urban fabric [56,57,58]. As previous research has shown, lack of clear operational definitions can lead to conflated classifications that weaken planning insights [59].

Data availability also played a key role in shaping the methodological scope of our work. Despite its limited open-data ecosystem [60], Greece offered a rare opportunity through the e-Poleodomia platform [61], a GIS-based portal with regulatory zoning data that we could integrate into our semantic model [62]. This allowed us to query not just spatial attributes but legal constraints on land use and building coverage. Although this access is not indicative of systemic openness, it provided a valuable test case for demonstrating the full semantic-geospatial workflow. Singapore, in contrast, boasts a mature smart city infrastructure and an extensive public-facing data ecosystem [63], including platforms like URA Space [64] and OneMap [65]. However, these datasets are often geared toward visualization and user guidance rather than analytical integration. Without access to legal zoning files, our Singapore case could not implement regulatory filtering. Still, the case served to demonstrate that even in data-limited settings, network-based semantic modeling can yield actionable valuable insights about spatial accessibility and UGS configuration.

Across both cases, OpenStreetMap (OSM) served as a critical harmonizing layer. As a form of volunteered geographic information [66], OSM offers open, editable data with global reach, which is particularly valuable in cities where official datasets are limited or inconsistent [67,68]. However, its reliability varies by feature type and region [69]. We conducted an initial round of manual cleaning and cross-validation across park geometries, street networks, and regulation lines to align with the ontology’s structural requirements. While this ensured basic consistency with our definitions and enabled the semantic integration process, no in-depth or field-level verification was performed. This underscores the broader challenge of data reliability and the ongoing need for curation even within widely used open platforms [59,70].

These differences in data access do not represent a weakness, but rather an opportunity to test the modularity of our approach. The methodology was fully applied in Thessaloniki, where regulatory, spatial, and network layers could be integrated into the ontology. In Singapore, where access to regulatory data was not available for this study, we implemented a streamlined version focused on spatial and topological dimensions. This flexibility highlights the model’s adaptability to varying data environments and its capacity to produce valuable insights even when certain components are missing.

Importantly, the challenges we encountered were not only technical. Data availability reflect institutional cultures and governance paradigms [71]. In both cities, as in many global contexts, public data platforms are primarily designed for communication or service delivery rather than analytical interoperability [72,73]. Features like parks or corridors are rarely defined in ontological terms that allow logical querying or cross-dataset integration [74]. This reflects a broader issue in data governance: openness and transparency are often framed in terms of visual access, not analytical usability [71,72].

Furthermore, concerns around licensing, data ownership, and institutional boundaries frequently restrict external access, even in technically advanced environments [53,75,76]. These epistemic and political conditions affect not only research but also the kinds of planning futures that can be imagined and implemented [77].

While our study cannot resolve these systemic barriers, it demonstrates a methodological path that is both rigorous and flexible. By situating our workflow at the intersection of semantic modeling, network theory, and spatial analysis, we contribute to an emerging planning paradigm that is both technically integrated and critically aware of the sociopolitical context of urban data [73,74].

4. Methodology: Ontology and Network-Based Spatial Analysis

Urban planning is inherently multidisciplinary, combining spatial data analysis, regulatory frameworks, social dynamics, and, increasingly, semantic technologies. As discussed in the introduction, addressing contemporary urban challenges requires methods that can bridge these domains rather than treat them separately. This section outlines the methodological framework of our study, which integrates semantic modeling with network-based spatial analysis to evaluate, categorize, and support strategic interventions related to UGSs. While spatial analysis often handles one layer of information at a time, our approach places a domain-specific ontology at the center of the process, an ontology that is informed and populated by outputs from spatial and network analysis, enabling interoperable reasoning across multiple dimensions of urban regulation, morphology, and infrastructural connectivity.

We begin with the development of the Sustainable Polis Ontology (SPoOn), which builds upon and extends existing The World Avatar (TWA) ontologies, incorporating additional classes and properties tailored to Greek urban planning and finer-grained urban features. SPoOn formalizes both regulatory and morphological knowledge about parks, streets, buildings, and public spaces, serving as the semantic backbone of the workflow.

Next, we perform network analysis using Spatial Design Network Analysis (sDNA) within GIS to identify patterns of accessibility, connectivity, and critical infrastructural hubs across the urban fabric. These outputs are semantically integrated into SPoOn, allowing them to be queried, reasoned upon, and related to other urban elements.

Finally, we apply regulatory constraints and spatial rules encoded within the ontology and supported by semantic queries to filter, classify, and evaluate urban spaces, demonstrating how semantic reasoning enables complex, context-specific insights into green space distribution, access equity, and strategic planning opportunities.

4.1. Ontology Design: Building Sustainable Polis Ontology (SPoOn)

The first step in our methodological pipeline involved the design and implementation of the Sustainable Polis Ontology (SPoOn), acting as a structured semantic framework to formalize knowledge about green spaces, road networks, urban regulations, and urban features. Ontologies (defined as formal, machine-readable representations of domain-specific concepts and relationships) have been increasingly applied in urban studies to bridge fragmented datasets and enable interoperable reasoning across diverse tools and disciplines [4,78,79].

SPoOn was developed by extending and adapting ontologies from the TWA ontology suite [45], a semantic framework originally designed for integrated urban analytics and digital twins [80]. TWA models the world as a dynamic, interconnected knowledge graph that enables real-time data integration and reasoning across domains. It provides a strong foundation by incorporating connections with important spatial ontologies such as GeoSPARQL, a standard for geospatial semantic queries, and CityGML, a data model for 3D city modeling [81], thus supporting spatial reasoning and representation. However, given the specificity of Greek urban planning, additional modeling was necessary. Our approach involved introducing new classes and properties tailored to the structure and terminology found in Greek urban plans and official regulatory documents. Many of these concepts were drawn from authoritative sources such as the official digital platform of the Hellenic Ministry for Urban Planning [61], its accompanying documentation [82], and from legislative frameworks available through the Ministry of Environment and Energy’s official website [83]. The ontology’s development followed a lightweight, iterative methodology inspired by METHONTOLOGY [84] and NeOn [85], involving reuse of foundational vocabularies, competency question modeling, and domain-specific extension aligned with regulatory structures.

SPoOn, as shown in Figure 1, extends existing semantic models by introducing regulatory precision and urban-scale granularity required for spatial and legal reasoning. Rather than acting as a static taxonomy, it serves as a dynamic semantic layer informed by geospatial, network, and regulatory data, enabling complex urban analysis that evolves alongside the dataset. Key extensions in SPoOn include the following:

•

City Objects: These represent the physical components of the urban fabric, such as streets, blocks, parks, and buildings.

•

Urban Representations: These are used to model regulatory or conceptual entities such as construction lines, zoning boundaries, and setback limits.

•

Properties for Analysis: SPoOn incorporates geometric (e.g., hasArea, hasCentroid), topological (e.g., includes), network-based (e.g., hasBetweenness_max, hasConnectivity), and informational (e.g., hasName, hasAddress) properties to support a wide range of spatial and semantic operations.

•

Populated Instances: Real-world individuals (parks, streets, regulation areas) are added to the ontology through automated scripts parsing GIS outputs, spatial metrics, and regulatory shapefiles.

Figure 1. Sustainable Polis Ontology (SPoOn) with TWA and GeoSPARQL classes.

Figure 1. Sustainable Polis Ontology (SPoOn) with TWA and GeoSPARQL classes.

Where applicable, SPoOn classes and properties were aligned with broader semantic vocabularies through relations such as subClassOf or equivalentClass, ensuring compatibility with external ontologies like TWA. This alignment strategy reflects a pragmatic balance between completeness and feasibility; rather than attempting to model all external links exhaustively, SPoOn focuses on critical semantic connections required for case-specific reasoning (e.g., spatial relations, legal eligibility), while remaining open to incremental expansion.

SPoOn is implemented in OWL and populated with RDF triples generated from heterogeneous sources, including OSM, GIS sDNA outputs, and cleaned regulation geometries. Its design supports real-world spatial-semantic reasoning, prioritizing practical interoperability over more complex integrations like 3D modeling (e.g., CityGML), which remain out of scope but theoretically possible.

Unlike traditional models that remain fixed, SPoOn evolves throughout the analytical process. New insights drawn from querying or rule-based classification (e.g., SWRL-based identification of isolated or high-connectivity parks) may reveal the need for additional classes, properties, or distinctions, prompting iterative refinement. In this way, SPoOn operates not as a finished schema, but as an adaptive, extensible framework that bridges GIS, regulation, and semantic reasoning in a coherent and reusable structure.

Figure 2 visualizes this iterative, ontology-centered workflow that underpins the entire methodology from initial data input, through network analysis and interpretative reasoning, to feedback loops for semantic enrichment.

The diagram in Figure 3 illustrates the end-to-end architecture of our urban sustainability network-driven knowledge graph. The process begins with the collection and curation of spatial and regulatory data from sources such as OpenStreetMap and the Greek Epoleodomia platform. These datasets are processed using GIS and Spatial Design Network analysis (sDNA) to generate metrics like betweenness and connectivity. These outputs, along with legal zoning geometries, are then semantically encoded into the SPoOn ontology as individuals with spatial , topological, and legal properties. The ontology allows integrated querying via SPARQL and reasoning via SWRL rules. For instance, a park polygon that borders a high-scoring street but falls outside of construction zones may be classified as a potential development site. Such reasoning processes modify and enrich the ontology itself, creating a feedback loop where each round of analysis informs further semantic refinement. This adaptive structure ensures that the framework remains responsive to different spatial contexts and planning logics.

A simplified version of the SPoOn ontology is available via GitHub for reference: https://github.com/parielis/SPoOn (accessed on 13 August 2025). A complete version is included in the Supplementary Materials accompanying this paper.

4.2. Network and Spatial Analysis in GIS

The second stage of our methodology involves the extraction and analysis of geospatial data within a GIS environment. This component serves two essential purposes, as follows: (1) to generate the spatial and network indicators that will later be associated with individual parks, and (2) to prepare the data for semantic translation into OWL classes and properties for use in the ontology-based reasoning framework.

4.2.1. Data Preparation and Processing

We used OSM as the primary source for both the street network and park datasets, due to its global coverage, open accessibility, and structured tagging system. While OSM provides a strong foundation for cross-contextual application in both Singapore and Greece, its data quality can vary, especially in urban feature classification. To ensure consistency, we applied preprocessing steps in a GIS environment, including the cleaning and topological correction of the street network, removing disconnected segments, fixing geometry errors, and simplifying classifications to reflect meaningful urban structure. For the park data, we applied filtering and manual verification procedures to exclude non-park features and to ensure that selected parks meet criteria, as discussed in Section 3.

4.2.2. Network Analysis Using sDNA

For the analysis of the urban street network, we used Spatial Design Network Analysis (sDNA), a specialized toolkit for quantifying spatial accessibility and flow potential within street systems. The street networks were converted into graphs and analyzed using a suite of sDNA metrics, including but not limited to the following:

•

Betweenness Centrality: This metric identifies streets with high through-movement potential, which are likely to attract traffic flows.

•

Closeness Centrality: This measures how accessible a street is in terms of average distance to all other nodes in the network.

•

Local Angular Connectivity (LAC): This metric evaluates the degree of local integration within a specified radius, reflecting neighborhood-level walkability.

These metrics were selected for their complementary capacity to capture structural accessibility across different urban scales. Betweenness centrality, widely used to identify urban corridors and flow potential [48,50], reflects how parks align with through-movement. Closeness centrality has been used to model user accessibility and proximity to services [51]. Local angular connectivity (LAC), often linked to neighborhood walkability and permeability [6,49], helps identify parks embedded within locally integrated street patterns. Combining these perspectives provides a quantitative characterization of each street segment’s role in the overall network. While only a subset of these metrics was ultimately used for the semantic queries detailed in Section 4 and Section 5, the full analysis ensures a comprehensive understanding of network performance.

4.2.3. Assigning Street-Based Metrics to Parks

Network analysis plays a central role in this study, offering a structured way to assess the spatial performance of parks in relation to the surrounding urban mobility network. Rather than modeling parks as nodes within a dedicated park-to-park system, which might be more suitable for studies of ecological connectivity or recreational routing, we focus on their integration into the street network. This approach reflects our aim to evaluate parks as urban destinations, where accessibility and visibility are shaped by the configuration of nearby streets. While alternative strategies that construct inter-park networks can yield insights for broader planning of green infrastructure, our method prioritizes local embeddedness and direct user access via the public street grid. This perspective aligns with observed mobility behavior, where parks are approached on foot or by vehicle through adjacent roads [47,86,87].

To implement this approach, we assigned network properties to parks based on the street segments that intersect or border each park polygon. For every park, according to these adjacent segments, we recorded both the minimum and maximum values of relevant sDNA metrics including connectivity, betweenness, and integration. This dual-value method preserves edge-level variation, allowing us to capture potential disparities in how different sides of a park connect to the urban grid. Unlike statistical averaging, which may obscure uneven access conditions, this approach retains important spatial nuance.

For example, a park with both very high and very low neighboring street centrality values may indicate that one side is well-integrated into the network while another remains disconnected. Such insights can be critical when evaluating equitable access or identifying optimization opportunities in green space planning.

4.2.4. Spatial Attributes and Preliminary Analysis

In parallel with network values, GIS environment was used to compute basic spatial metrics for each park, like area and distance from network hubs. These attributes were also recorded and linked to semantic properties in the ontology.

The combination of spatial and network indicators provides a robust multi-scalar dataset for evaluating green space performance and distribution. This integrated dataset, as the next section will illustrate, is essential for executing semantic queries that reveal patterns, gaps, and potential across the UGS system.

Future research could extend this approach by modeling internal park pathways or analyzing edge permeability, but the current method offers a scalable and meaningful way to link semantic planning models with real-world spatial accessibility.

4.3. Translating GIS Data into Ontology

Once spatial and network analyses were completed within GIS, the next phase of our methodology involved converting the enriched dataset into a semantic representation. This step is crucial to enable reasoning, complex querying, and integration across diverse knowledge domains; traditional GIS environments, while powerful in spatial terms, are not optimized for these goals.

The translation process involved mapping each GIS feature (parks, streets, and regulatory lines) into individuals within SPoOn. For each entity, the following procedures were applied:

•

Parks were instantiated as individuals of the class Park, with data properties corresponding to attributes such as area, hub distance, maximum betweenness, and other spatial, network-derived metrics or OSM-derived attributes;

•

Street segments were instantiated as individuals of the class Street, carrying properties including length, sDNA, and OSM values;

•

Regulatory lines and planning boundaries were instantiated as individuals of corresponding classes (e.g., BuildingLine, StreetLine, UncoveredSpaceBoundary), populated with their official attributes sourced from the Epoleodomia GIS portal [61].

All geometries were extracted from GIS as well-known text (WKT) representations to enable spatial querying through GeoSPARQL standards (a geospatial extension of SPARQL for querying location-based data). The entire mapping process was automated via a custom Python script (Python 3.11, executed in Anaconda’s Spyder IDE 5.4.3), which parsed shapefiles and attribute tables into RDF triples, ensuring scalability, consistency, and reproducibility across case study areas.

At this stage, no advanced semantic enrichment was performed. Further classifications, inferred relationships, and semantic reasoning were applied during the semantic analysis phase, as described in the following section.

4.4. Semantic Queries and Analysis

Following the translation of spatial and network datasets into SPoON ontology (Section 4.3), the next and final stage of our methodology involved semantic enrichment through querying and rule-based classification. This phase enabled the extraction of complex insights such as accessibility patterns, network hub identification, and compliance with planning regulations, which might be possible with conventional GIS tools but would typically require multiple disjointed processes, manual interpretation, or the integration of several specialized systems. The semantic approach offers a more unified and scalable alternative. Our semantic approach leveraged SPARQL queries (a structured query language for retrieving data from semantic knowledge graphs) and SWRL rules (Semantic Web Rule Language, used to define logical conditions for automated reasoning) to infer new knowledge and dynamically classify urban elements.

4.4.1. SPARQL Queries for Urban Network Analysis

Using the structured SPoOn ontology, we designed a series of SPARQL queries grouped into thematic categories:

•

Accessibility and Connectivity Patterns: Queries retrieved parks exhibiting low local connectivity (?LConn), low link counts (?Lnkn), and large distances to network hubs (?HubDist). This allowed the identification of potentially underserved green spaces;

•

Network Hubs: Parks characterized by both high betweenness centrality and high local connectivity were queried, revealing critical hubs within the UGS network;

•

Composite Scoring: To evaluate and prioritize parks based on their network performance, we computed a composite score (Score) integrating three normalized centrality metrics: local connectivity (normLConn), betweenness centrality (normBtEWln), and hub distance (normHubDist). The composite score was derived using the following weighted linear combination:

Score = normLConn + normBtEWln − normHubDist,

(1)

where the additive terms (normLConn, normBtEWln) capture positive contributions of network accessibility and throughput, while the subtractive term (normHubDist) penalizes remoteness from major activity hubs. Normalization ensured comparability across metrics by scaling all values to a common range prior to aggregation.

This composite score helps planners rank parks not only by size or location, but by how functionally integrated they are within the street network. It identifies spaces that are both accessible and centrally located, enabling prioritization for maintenance, programming, or connectivity improvements;

•

Spatial Typologies: Park categories were identified by querying spatial characteristics such as area, centralities, and adjacency to street networks, informing classification (later formalized through SWRL rules).

These semantic queries allow both generalized assessments (e.g., distribution of well-connected parks across the city) and specific evaluations within neighborhoods. More broadly, SPARQL queries allow planners to extract targeted insights by asking structured questions. This makes it possible to retrieve, for example, “all parks with low walkability that are far from central hubs,” or “areas zoned for public use with proximity to high-accessibility corridors,” without custom coding. This structured querying enables practitioners to explore multidimensional planning problems in ways traditional tools cannot easily support.

4.4.2. Rule-Based Typological Classification (SWRL)

In addition to direct querying, we employed SWRL rules to automate the classification of parks into typological categories and assign them as data properties of the parks. Unlike SPARQL queries, SWRL enabled the ontology itself to infer new knowledge dynamically, creating new property assertions based on logical thresholds.

To classify parks exhibiting spatial isolation, we formalized the “Isolated Park” condition using a SWRL rule that evaluates both connectivity and hub proximity. The rule is defined as follows:

swrlb:greaterThan(?hub,1500)∧spoon:Park(?p)∧spoon:HubDist(?p,?hub)∧spoon:Lnkn_Cl_ma(?p,?Lnkn)∧swrlb:lessThan(?Lnkn,2.0)→spoon:isIsolated(?p,”yes”)swrlb:greaterThan(?hub,1500)∧spoon:Park(?p)∧spoon:HubDist(?p,?hub)∧spoon:Lnkn_Cl_ma(?p,?Lnkn)∧swrlb:lessThan(?Lnkn,2.0)→spoon:isIsolated(?p,”yes”)

This rule assigns the data property isIsolated to parks meeting two criteria: (1) a hub distance exceeding 1500 m and (2) fewer than 2.0 connecting links. These thresholds were selected to operationalize spatial isolation, capturing parks that are both peripherally located and poorly integrated into the urban network. The SWRL engine dynamically infers this classification during reasoning, enabling automated identification of isolated green spaces.

Similarly, other classifications such as NeighborhoodPark, RegionalPark, or PedestrianCorridor were derived through SWRL rules based on combinations of network metrics and spatial attributes. These rule-based enrichments enhanced the ontology’s semantic depth and supported more refined querying and visualization.

Thresholds used in the rules (e.g., hub distance >1500 m for isolation) were initially selected based on ranges found in accessibility and walkability literature [46,47], but were not exhaustively calibrated or validated. Rather than proposing fixed benchmarks, this study focuses on demonstrating a flexible methodology capable of adapting to specific spatial contexts and expert needs. Rules can be refined, expanded, or adjusted depending on local planning criteria, stakeholder priorities, or future feedback loops. In this sense, the current thresholds serve as illustrative defaults; the emphasis is on testing the integrative potential of semantic reasoning within urban workflows, while leaving the fine-tuning of parameters for future work that involves planners and participatory validation.

4.4.3. Semantic Modeling of Regulation Areas and Identification of Potential Parks

In the Thessaloniki case study, a complementary workflow was developed to model urban regulatory boundaries and systematically identify new potential park areas. This process built upon the same core methodology used for parks and street networks, combining geospatial analysis and semantic technologies into a unified pipeline tailored to the local planning framework.

Data Preparation and Polygon Generation

Urban regulation datasets from Epoleodomia were imported into GIS, including street boundaries and construction lines. To generate candidate ParkArea polygons, the following procedures were carried out:

•

Street boundaries were buffered and merged into closed polygons to create StreetAreas and ConstructionAreas;

•

Construction zones were subtracted from enclosed street areas to create park areas. Candidate ParkAreas were calculated as follows:

ParkArea = StreetArea − ConstructionArea

(2)

This reflects Greek urban planning practice, where closed streetlines define plots that may enclose construction zones, and the public open space (for parks, squares, etc.) corresponds to areas outside construction lines but inside streetlines; that is, unbuildable spaces that are not part of a street.

• Exclusion of Existing Parks

To prevent redundant identification of areas already serving as parks, existing parks from OSM were then excluded from the candidate ParkAreas.

• Semantic Translation

The resulting polygons were semantically integrated into the SPoOn ontology by adding new classes: ParkArea, StreetArea, and ConstructionArea. Each was enriched with attributes such as area and geometry (in WKT format).

Querying and Prioritization

SPARQL queries were designed to identify ParkArea candidates that met the following condiditions:

•

Located near (within 500-m distance) high-scoring streets (based on composite network metrics with score higher than 1);

•

Satisfied size (minimum 50 m² area) and zoning conditions inferred from the regulatory data (is ParkArea).

This process enabled a systematic evaluation of new park opportunities, grounded in both urban morphology and planning constraints, and connected to the network structure of the city.

4.4.4. Iterative Feedback Between Querying, Reasoning, and Ontology Evolution

Throughout the querying and rule-based classification phases, an iterative feedback loop was maintained between data analysis and ontology refinement. As queries and initial classifications were applied to the case studies, new insights often pointed to the following:

•

Missing or imprecise property definitions;

•

The need for new classes or subclasses;

•

The necessity to introduce new object or data properties.

For instance, the creation of specific categories such as “Isolated Park” or “Key Hub Park” emerged from patterns detected during the spatial network analysis.

It is important to note that the queries and rules developed here are indicative and tailored to the specific case studies. In a different urban context, other patterns could emerge, such as a need for finer classification based on park sizes or different connectivity thresholds, thus requiring further ontology evolution.

This cyclical process (data → ontology → query → reasoning → refinement) ensures that SPoOn remains adaptable to emerging knowledge, different regulatory environments, and varying spatial patterns across cities. It is precisely for this reason that the methodology was also tested in the Singapore context, to examine its transferability and refine its structure for broader applicability.

5. Implementation and Testing in Distinct Contexts

This section presents the implementation of our semantic-geospatial framework in two urban settings: a complete deployment in Thessaloniki, Greece, and a contextual test in Singapore. The aim is not to compare the two areas, but to demonstrate how the framework adapts to distinct planning environments. This dual application allows us to assess the methodology’s flexibility, extensibility, and capacity to generate meaningful insights under different data and regulatory conditions. The areas were selected due to their comparable size (~19 km²) but contrasting urban morphologies and planning systems. Thessaloniki represents a historically dense, organically developed European city, while Singapore offers a highly structured, infrastructurally integrated district.

The application of a shared semantic model, originally based on TWA ontology and extended to incorporate Greek urban planning regulations, enables a consistent methodological approach across both contexts. At the same time, local adaptations, such as specific classifications, thresholds, and regulatory filters, demonstrate the contextual flexibility and extensibility of the methodology. This dual case study approach serves both to validate the general framework and to highlight how semantic–geospatial tools can accommodate and reveal context-specific challenges and opportunities in UGS planning.

5.1. Core Implementation: Municipality of Thessaloniki (Greece)

The Municipality of Thessaloniki covers approximately 19 km² (shown in Figure 4 below) and houses a population of over 300,000 residents. As a dense, historically layered urban core within a Mediterranean metropolis, Thessaloniki exhibits characteristics typical of many European cities: irregular street patterns, fragmented open spaces, and constrained land availability due to centuries of uncoordinated growth [88,89]. While this morphological complexity contributes to its urban identity and cultural richness, it poses significant challenges for the planning and equitable distribution of UGSs [90].

Despite various local and European initiatives to promote urban sustainability, Thessaloniki ranks among the lowest in green space per capita within the EU, with estimates around 2.6 m² per person; well below the WHO recommendation of 9 m² [3]. This issue is not only quantitative but also spatial–functional; green spaces are often fragmented, unevenly distributed, and disconnected from movement infrastructures [2]. These conditions highlight the need for tools like semantic network analysis to detect spatial deficits and inform more integrated green infrastructure planning.

Figure 4. Municipality of Thessaloniki boundary (purple area) shown on an OSM basemap.

Figure 4. Municipality of Thessaloniki boundary (purple area) shown on an OSM basemap.

5.1.1. Network-Based Findings

The municipality hosts a dispersed network of 185 parks, covering a total area of 118,190 m², reflecting a pattern of smaller, fragmented green spaces embedded within a dense urban fabric. The ontology-based querying of Thessaloniki’s UGSs produced a number of interesting results. Using indicators like local connectivity, betweenness centrality, and distance from hubs, we identified four key parks (Field of Mars, Park of YMCA square, Melenikou Park, Park of Saint George square) that function as network hubs. These parks had the highest combined scores (?BtEWln > 0.5 and ?LConn > 5), indicating that they serve as core UGSs with multifunctional value, offering both recreational benefits and spatial integration.

Interestingly, only one park (Galerius Arch tree alley) was found to meet the criteria for a significant accessibility gap, namely, low connectivity and a long distance from hub nodes. This suggests that while overall provision may be lacking, the spatial fragmentation in Thessaloniki is less about complete exclusion zones and more about network fragility, a finding reinforced by the identification of 13 parks with fragile accessibility (?LConn ≤ 3 and ?Lnkn ≤ 4). These parks are weakly embedded in the surrounding street network, potentially limiting their utility despite being geographically dispersed.

We created a composite indicator by combining normalized values of local connectivity, centrality, and hub proximity, allowing us to compare parks based on a multi-criteria evaluation of spatial performance. Thessaloniki’s highest-scoring park reached 1.708, while 13 parks exceeded the threshold of 1.0, and 88 parks fell below the average value of 0.690. These numbers reveal a relatively polarized performance profile, a few well-integrated parks amidst many underperforming ones. However, this may partly reflect the city’s morphological limitations and irregular infrastructure more than a planning failure.

5.1.2. Park Classifications and Regulatory Context

Out of 185 total parks, Thessaloniki had 70 Neighborhood, 17 Regional, 2 Isolated, and 1 Pedestrian Corridor park. Compared with Singapore, this classification suggests a more balanced typological diversity, possibly reflecting efforts to retrofit open spaces within an already dense fabric. However, the presence of isolated parks indicates challenges in connectivity, often due to physical barriers or street discontinuities.

Figure 5 reveals a clustered pattern of Neighborhood parks and a limited number of Regional parks concentrated in central zones. Isolated parks tend to lie at morphological edges or areas with discontinuous street networks, suggesting possible targets for connectivity improvement.

What sets Thessaloniki apart is the incorporation of Greek urban planning regulations into the semantic model. Specifically, the analysis included spatial regulation lines and zoning constraints to detect more than 100 potentially suitable areas for new green spaces. These include vacant lots, state-owned grasslands, or infrastructure buffers that, while currently undeveloped or unlandscaped, are legally eligible to be designated as public parks under planning rules [61].

Moreover, by correlating the locations of these potential UGSs with the network performance of nearby streets, the model was able to prioritize candidate sites that would be legally feasible and spatially beneficial. This contributes a proactive planning logic; instead of simply documenting deficits, the methodology suggests specific interventions aligned with legal frameworks and spatial logic. Higher-scoring parks, according to a composite of normalized centrality metrics, can guide planners toward areas where access and integration are strongest. Potential sites identified via SPARQL (yellow polygons in Figure 6) exemplify locations where regulatory eligibility intersects with strong network positioning. These results highlight how semantic reasoning can reveal latent spatial opportunities in morphologically complex cities, shifting the narrative from spatial deficit to potential.

5.2. Testing in a Different Context: Marine Parade GRC (Singapore)

Marine Parade GRC in Singapore was selected to test the adaptability of the methodology in a different urban context. Unlike Thessaloniki, where the full implementation included legal zoning layers, this application focused only on spatial and network dimensions, reflecting both practical data constraints and the goal of evaluating the framework’s flexibility. The district’s scale (~19 km²) (shown in Figure 7 below) and planned urban morphology offered an appropriate secondary testbed for validating the method’s flexibility.

Marine Parade was developed in the 1970s on reclaimed land as part of Singapore’s response to post-independence housing shortages [91]. The area blends residential, commercial, and recreational uses, including a large portion of East Coast Park spanning 185 hectares [92,93]. Singapore’s planning system is internationally recognized for its long-term integration of green infrastructure and sustainability goals into urban design [53,94].

While the city-state performs well in terms of green space per capita and design coherence, these indicators emphasize surface provision rather than functional connectivity. Using the semantic network model, we investigated how Marine Parade’s UGSs function within the street network, analyzing local connectivity, betweenness centrality, and proximity to hubs through semantic queries.

Despite having only 69 parks, the district’s total park area is considerably large (332,318 m²), suggesting a focus on fewer but larger UGS. Among these, two parks (Jalan Daud Interim Park and Jalan Eunos Interim Park) stand out as high-performing hubs with both high local connectivity (LConn > 5) and betweenness (BtEWln > 0.5), derived from the corresponding SPARQL query that isshown in Figure 8. Their position within the street network implies strong functional integration and suggests that they serve as connectors as well as destinations.

Figure 7. Marine Parade GRC boundary (purple area) shown on an OSM basemap.

Figure 7. Marine Parade GRC boundary (purple area) shown on an OSM basemap.

Figure 8. Query to find key hub parks in Marine Parade GRC.

Figure 8. Query to find key hub parks in Marine Parade GRC.

However, the analysis also uncovered two green spaces (Sky Park and Lor Marican Park) that are significantly distant from key hubs and have weak local connections (?LConn ≤ 4) and low numbers of neighboring links (?Lnkn ≤ 4). In a traditionally planned environment like Singapore’s, such instances are noteworthy; they suggest emerging micro-gaps in network accessibility that may affect equitable use or social reach. Similarly, three parks (Sky Park, Chiltern Drive Park, and Boundary Park) were categorized as having fragile accessibility, combining low connectivity and low local integration. These may be structurally marginal or poorly integrated into surrounding movement networks, raising questions about design and context-sensitive placement.

The composite indicator was also calculated. The highest scoring park reached a value of 1.666 out of 69 parks, while 8 parks surpassed the threshold of 1.0, and 48 parks scored below the mean value (0.648). This suggests a skewed distribution in spatial integration, with a few dominant nodes and many under-connected UGSs.

In terms of classification, the majority of parks in Marine Parade fell into the Neighborhood category (36 out of 69 parks). Only one park was classified as Regional (Greenville Garden Park), one as Pedestrian Corridor, and none as Isolated. While the total green space footprint in Marine Parade might suggest the presence of larger individual parks, potentially even at regional scale or nature reserves, the classification used here is based on topological rather than purely morphological characteristics. As a result, some large parks may be functionally peripheral or poorly integrated, and thus not meet the thresholds for higher-level classifications.

The absence of isolated parks may indicate a minimum level of spatial continuity within the green infrastructure network. However, this does not guarantee high functional accessibility across the district. The findings highlight that even in systematically planned urban contexts, variability in network integration can emerge, underscoring the importance of tools that assess performance, not just provision.

Taken together, this secondary application reinforces the utility of the proposed methodology. The framework helped surface spatial inefficiencies, identify functional hubs, and flag underperforming areas, demonstrating its value as a flexible and extensible planning tool. These nuances underscore the importance of complementing traditional planning metrics with topological and semantic evaluations to understand how UGS operates within urban systems, rather than just how much of it exists.

6. Insights and Implications for Urban Planning

The application of a semantic–geospatial framework across two distinct contexts revealed both the strengths and adaptability of the methodology. Rather than serving as a comparative analysis between Thessaloniki and Marine Parade, the dual implementation allowed the framework to be tested under varying data conditions and planning cultures. Thessaloniki served as the primary implementation, where the full model was deployed including regulatory, spatial, and network data, while Marine Parade functioned as a secondary testbed to assess the method’s performance using a streamlined dataset. Together, they illuminate how a shared ontology can produce high-resolution, context-aware insights into the performance and potential of urban green infrastructure.

Perhaps most importantly, the results call into question some often-unexamined assumptions in urban sustainability discourse. The implementation in Marine Parade showed that well-planned environments do not necessarily perform better in terms of connectivity or spatial integration. Despite high green-per-capita statistics and advanced regulatory frameworks, the network analysis revealed parks in fragile or peripheral positions, underlining structural inefficiencies that would not be visible through conventional metrics alone. Conversely, the Thessaloniki case, despite its more chaotic morphology and spatial constraints, included high-performing park nodes and over 100 legally and spatially viable sites for future green spaces. These findings demonstrate how data-informed, ontology-driven methods can challenge generalized narratives of urban success or failure, promoting more nuanced and localized interventions.

Across both cities, the framework enabled the following:

•

Detection of spatial blind spots, where parks were poorly integrated, under-connected, or legally constrained;

•

Recognition of high-performing spatial configurations, where parks aligned with central corridors or supported broader accessibility patterns;

•

Identification of actionable opportunities, particularly in Thessaloniki, where regulatory data enabled semantic reasoning to reveal developable green zones.

These capabilities support a dual planning function, comprising the following:

• Problem detection: The model reveals where urban green systems are disconnected, undersized, or misaligned with street networks, diagnosing inefficiencies that would be difficult to perceive through static spatial layers alone.
• Pattern recognition and opportunity identification: By surfacing recurring spatial logics, such as the co-location of parks with highly integrated street segments, the methodology supports replication of successful configurations. Moreover, it enables the semantic filtering of candidate areas for intervention, based on zoning status, area thresholds, and network proximity.

This ontological approach also demonstrates transferability of principles. While the implementation contexts differ, certain findings appear consistent; spatial integration of parks into urban flows, rather than isolated placement, yields greater functional accessibility. Similarly, rules derived in one setting (e.g., thresholds for centrality, legal constraints) can inform interventions in another, especially if refined semantically.

Unlike conventional GIS workflows that require manual combination of layers and context-specific scripting, the semantic model allows such patterns to be formalized, queried, and reused, supporting cross-domain collaboration and future scalability.

Finally, the results suggest that this framework does not only support present evaluations but could guide future planning scenarios. For instance, it could simulate the impact of zoning changes, detect underserved neighborhoods, or help prioritize green corridor extensions. Future work can build on this by developing green space networks, analyzing park systems as interlinked nodes rather than standalone patches, and connecting them topologically, ecologically, and socially across the urban grid. In this way, the methodology positions itself not only as a diagnostic tool, but also as a prescriptive framework, capable of revealing hidden potential, validating regulatory assumptions, and informing smarter, more inclusive, and sustainability-oriented urban planning.

7. Conclusions and Future Work

This study demonstrates the value of a semantic framework enriched with spatial and network data as a tool for more nuanced, regulation-aware urban planning. By extending a shared ontology (based on TWA) with legal designations, morphological attributes, and network metrics, we developed a flexible, context-sensitive method for evaluating the performance and potential of UGSs. Crucially, this approach captures not only spatial configuration but also the legal and regulatory landscape, enabling the identification of both systemic deficiencies and actionable opportunities.

In Thessaloniki, for example, the framework identified over 100 legally suitable areas for new parks, selected through semantic filters that combined zoning eligibility with high-scoring street connectivity. In Marine Parade, Singapore, despite the presence of large green spaces, the analysis exposed parks with weak integration or fragile access. These results underscore the ability of the methodology to detect spatial and regulatory inefficiencies in both historically fragmented and systematically planned environments, offering planners a tool for uncovering actionable, context-specific interventions.

The implementation of this framework in two distinct urban contexts, Thessaloniki in Greece and Marine Parade in Singapore, demonstrates how insights can be generated even under different data conditions and planning cultures, revealing patterns not readily accessible through traditional tools. Identifying legally viable development sites, poorly connected parks, or multi-criteria accessibility gaps would typically demand custom workflows. In contrast, the semantic approach offers a more transparent and generalizable alternative, with adaptable queries rooted in a common urban vocabulary. These results underscore the value of shifting from static, descriptive datasets to dynamic, constraint-aware methodologies that directly inform planning decisions.

A key strength of this framework lies in its modularity and scalability. It supports ongoing integration of new data, automated reasoning, and reuse across spatial contexts and planning domains. This flexibility enables planning tools that evolve with changing conditions and support participatory, transparent governance. Ontology-based models help bridge disciplinary and institutional silos, making urban knowledge more shareable, interpretable, and sustainable over time. Looking ahead, the integration of this semantic framework with natural language processing or AI tools (such as large language models) could make it usable by planners without technical backgrounds. Through intuitive interfaces, users could pose planning questions in plain language and receive complex, multi-layered answers that draw from spatial, regulatory, and infrastructural data. This development could bridge the gap between sophisticated data analytics and everyday decision-making in planning practice.

At the same time, the study highlights an important limitation regarding the critical role of data availability and quality. While Singapore is known for its open data policies, practical access limitations still hindered the full deployment of the framework, demonstrating that availability in principle is not the same as usability in practice. Effective semantic planning tools depend on structured, granular, and legally meaningful data. Cities must go beyond publishing datasets and commit to making them machine-readable, spatially explicit, and interoperable if such tools are to reach their potential.

Looking ahead, one important avenue for development is the shift from treating parks as isolated objects to modeling them as a park network, capturing how green spaces interact across the city, how people move between them, and how these interactions support ecological, social, and infrastructural resilience. This could lead to the identification of green corridors or multi-scale hierarchies of parks, advancing a systemic understanding of urban green infrastructure. Such a network perspective could also support future comparative studies focused on functional integration and spatial logic, should the necessary datasets and variables be available.

The same framework can also be extended to other urban domains, such as public transit nodes, emergency services, or retail clusters, enabling a unified spatial–semantic model that supports city-wide diagnostics and scenario planning. By semantically encoding proximity rules, zoning logic, or demographic thresholds, planners can simulate policy impacts or detect underserved areas across various infrastructures.

Finally, future work might explore the development of dynamic urban ontologies, capable of incorporating real-time data, responding to evolving policies, and adapting to demographic shifts. As cities become increasingly complex and data-rich, planning tools must evolve accordingly. This research points toward a future where urban planning is not only spatial and legal, but also semantic, interpretable by both humans and machines, and capable of supporting more integrated, adaptive, and intelligent urban governance. Looking forward, the semantic methodology presented here holds promise not only for analytical rigor but also for supporting more transparent and accountable decision-making in urban planning. By structuring complex planning data in machine-readable and policy-aligned ontologies, such systems can make regulatory constraints and infrastructural trade-offs more visible to both professionals and the public. If extended with AI components such as large language models or geospatial analytics, the framework can offer enhanced query and analysis capabilities while maintaining interpretability and control through its ontology-driven logic. This balance between machine intelligence and semantic governance could play an important role in future smart and sustainable cities.