From vision to formula: integrating numerical urban planning heuristics into a spatial accessibility framework for sustainable mobility

1 Introduction

Cities are increasingly expected to reconcile accessibility, environmental quality, mobility, public health, social inclusion, and climate objectives within a single planning agenda. Yet these goals are commonly operationalized through separate policies, institutional responsibilities, indicator systems, and professional vocabularies. Land-use planning may focus on proximity and functional mix, green-infrastructure strategies on environmental quality and ecosystem services, and mobility planning on transport accessibility, modal integration, and decarbonization. Although these domains are strongly interdependent in everyday urban life, they are frequently measured and governed independently. This fragmentation can make sustainability objectives difficult to compare, monitor, communicate, and translate into spatially targeted interventions.

Against this background, a distinctive family of numerically expressed urban-planning heuristics has gained prominence. These rules attempt to convert broad sustainability ambitions into compact and memorable spatial or behavioral propositions. The 15-min city frames urban accessibility around the possibility of reaching essential daily functions within a short walk or bicycle journey and emphasizes proximity, diversity, density, and ubiquity (Moreno et al., 2021). The 3–30–300 rule translates urban-nature objectives into a sequence linking tree visibility, neighborhood canopy coverage, and proximity to public green space (Konijnendijk, 2023). In the mobility domain, the 90–50–70 principle draws attention to walking as an integral part of public-transport journeys and to the first- and last-mile interface between pedestrians and collective transport (EIT Urban Mobility, 2025). Despite their different origins, scales, and metrics, these formulations share an ability to translate complex planning objectives into measurable and communicable spatial conditions.

Their apparent simplicity, however, does not remove methodological complexity. Operationalizing numerical urban rules requires decisions concerning spatial scale, destination definitions, network representation, thresholds, population distribution, data quality, and aggregation. Accessibility outcomes may differ according to whether Euclidean or network distance is used, which travel mode is assumed, how opportunities are classified, and how populations are spatially represented (Geurs and van Wee, 2004). More recent accessibility theory further emphasizes the interaction among transport, land use, digital infrastructure and technologies, individual and social characteristics, and temporal conditions (Geurs and van Wee, 2026). NUSIF-Core should therefore be interpreted as a deliberately bounded spatial operationalization rather than a complete representation of calculated or perceived accessibility. Empirical studies of x-minute and 15-min-city concepts similarly show that results vary with operational definitions and exhibit substantial intra- and interurban heterogeneity (Logan et al., 2022; Bruno et al., 2024). Numerical thresholds are consequently treated here as transparent analytical assumptions and planning benchmarks rather than universal boundaries separating sustainable from unsustainable urban form. Within the broader accessibility-based planning literature, proximity represents one operational dimension rather than a complete representation of accessibility. Reviews of active accessibility show that walking and cycling accessibility can be operationalized through different families of measures and that terminology, parameter selection, network representation, and computational choices can materially influence the resulting assessment (Vale et al., 2016). Population-sensitive approaches are also increasingly used in comparative urban accessibility studies because they account for the uneven spatial distribution of residents rather than treating all locations as equally consequential (Bartzokas-Tsiompras and Bakogiannis, 2023). From a transport-equity perspective, accessibility should additionally be examined in distributional terms, since favorable aggregate accessibility may coexist with substantial inequalities among places and population groups (Pereira et al., 2017; Vale and Lopes, 2023). NUSIF is positioned within this literature as a deliberately parsimonious spatial integration framework: it does not attempt to reproduce all dimensions of accessibility, walkability, or transport equity, but combines selected network-based accessibility conditions while retaining separate territorial and population-weighted interpretations.

An additional challenge is that prominent numerical planning heuristics address different dimensions of the urban system. The 15-min city primarily concerns functional proximity and diversity of everyday opportunities; the 3–30–300 rule addresses exposure and access to urban nature; and the 90–50–70 principle emphasizes the walking component of public-transport journeys. Their original indicators differ in units, spatial scales, data requirements, and conceptual meaning. Integrating them therefore requires the selection of spatially compatible dimensions, normalization of their measures, and explicit treatment of the assumptions introduced when heterogeneous indicators are combined.

This paper addresses that problem through the Numerical Urban Sustainability Integration Framework (NUSIF), conceived as a translation layer between numerical planning visions and spatially measurable accessibility conditions. Rather than reproducing every component of the three conceptual frameworks, the empirical implementation focuses on three dimensions that can be operationalized consistently through network-based spatial analysis: accessibility to essential urban services, accessibility to public green space, and pedestrian accessibility to public transport. The service component represents the proximity logic associated with the 15-min city; the green-space component operationalizes the 300 m accessibility dimension of the 3–30–300 rule; and the public-transport component represents the pedestrian first−/last-mile interface. NUSIF therefore does not constitute a complete compliance assessment of any of the three underlying planning concepts but tests whether selected spatially compatible dimensions can form a coherent and reproducible integrated accessibility framework.

A second contribution concerns the distinction between territorial accessibility and population exposure. Spatial accessibility deficits may extend across many cells containing relatively few residents, while a smaller number of densely populated cells may strongly influence population-level outcomes. The importance of population density and intra-urban heterogeneity in accessibility analysis is increasingly recognized (Bruno et al., 2024). The present study therefore retains territorial and population-weighted measures separately and examines whether intervention performance changes when accessibility is evaluated across space or according to the distribution of resident population.

A third contribution is the transition from baseline diagnosis to intervention analysis. NUSIF is extended through three independent counterfactual scenarios that modify essential-service, green-space, or public-transport accessibility separately while holding the remaining components constant. This design allows comparison of the spatial reach, magnitude, and population-weighted consequences of different accessibility transformations without interpreting them as forecasts of specific infrastructure projects.

Niš, Serbia, is used as the empirical case study. The analysis combines modeled population data with OpenStreetMap-derived services, public green spaces, transport stops, and a pedestrian network. Accessibility is assessed across 873 analytical cells using pedestrian-network distance, after which the three normalized components are integrated into a common NUSIF score. Both territorial and population-weighted outcomes are calculated for the baseline and each scenario. The study addresses four research questions:

RQ1: What is the spatial distribution of essential-service, green-space and public-transport accessibility across the retained analytical territory of Niš?

RQ2: How does integrated NUSIF accessibility differ when evaluated from territorial and population-weighted perspectives?

RQ3: How do independent counterfactual changes in service, green-space and public-transport accessibility alter the spatial distribution of NUSIF?

RQ4: How do the scenarios differ in terms of spatial reach, population exposure, NUSIF change and class redistribution?

A complementary weighting-sensitivity analysis examines whether the principal territorial and population-weighted interpretations depend on the equal-weight reference specification. The study contributes conceptually by integrating selected spatially compatible dimensions of otherwise separate numerical planning heuristics; methodologically by combining network-based multidomain accessibility, component normalization, territorial and population-weighted evaluation, and controlled counterfactual scenarios; and empirically by demonstrating how spatial reach, magnitude of accessibility change, and population exposure can lead to different interpretations of intervention performance.

2 Conceptual foundations and integration logic of numerical urban planning heuristics

2.1 Numerical urban-planning heuristics as instruments of sustainable urban management

Quantitative rules and thresholds have long been embedded in urban planning through regulations concerning development density, land-use intensity, infrastructure provision, environmental performance, and access to public facilities. More recently, a distinctive class of numerically expressed planning heuristics has gained prominence. Rather than functioning primarily as statutory limits, these heuristics condense multidimensional planning objectives into memorable spatial or behavioral propositions that can be communicated among planners, decision makers, urban managers, and residents.

This development is closely connected to the increasing importance of accessibility-oriented planning. Whereas conventional transport planning has frequently emphasized movement, travel speed, and network performance, accessibility-oriented approaches focus on the opportunities residents can reach and the spatial relationships among people, activities, land use, and transport systems. The sustainable-mobility paradigm similarly emphasizes urban structures that reduce unnecessary travel, strengthen active modes, improve public transport, and integrate land-use and transport planning (Banister, 2008). Proximity-based urban models represent one expression of this transition. The 15-min-city concept, for example, combines contemporary proximity planning with established principles of neighborhood-scale development, compactness, mixed land use, and local service provision (Pozoukidou and Chatziyiannaki, 2021).

Despite their apparent simplicity, numerical planning concepts can conceal considerable methodological heterogeneity. Existing applications vary in their treatment of density, proximity, diversity, walkability, connectivity, and service accessibility (Khavarian-Garmsir et al., 2023; Papadopoulos et al., 2023). A numerical threshold may therefore improve the legibility of a planning objective without eliminating decisions concerning spatial units, network representation, eligible destinations, population distribution, service characteristics, or indicator aggregation. Numerical heuristics should consequently be interpreted as decision-support benchmarks rather than universally valid standards, with their assumptions made explicit and open to contextual adaptation.

Similar challenges arise in multidimensional sustainability assessment more broadly, where integration must balance comprehensiveness with interpretability and practical usability (Bigolin et al., 2026). NUSIF adopts this principle by integrating complementary accessibility dimensions while preserving their individual interpretation. Parallel developments in urban analytics have sought to translate complex characteristics of urban form and urban space into reproducible numerical representations. Fleischmann et al. (2022), for example, develop a scalable numerical taxonomy of urban form based on street networks, building footprints, and morphometric spatial units, illustrating both the analytical potential of numerical urban characterization and the importance of how spatial units are defined. Bi and Liu (2022) similarly demonstrate the use of numerical evaluation to structure the assessment of urban streetscape spatial patterns. Although these approaches address different analytical objects from NUSIF, urban morphology and streetscape evaluation rather than integrated accessibility, they reinforce a common methodological requirement: numerical urban indicators should be explicitly defined, reproducible, and interpreted in relation to the spatial support on which they are calculated.

2.2 The 15-min city: from proximity concept to service accessibility

The 15-min city represents a proximity-oriented planning model in which residents should be able to satisfy a substantial proportion of everyday needs within a short walking or cycling journey from their place of residence. Its empirical operationalization, however, is methodologically heterogeneous. Existing studies differ in their treatment of accessibility, spatial units, destination types, transport modes, density, diversity, connectivity, and performance thresholds (Khavarian-Garmsir et al., 2023; Papadopoulos et al., 2023). NUSIF therefore does not interpret its service-accessibility component as a complete test of 15-min-city compliance, but as a spatially compatible operationalization of its proximity logic.

The choice of accessibility measure is particularly important. Walking accessibility varies according to individual and temporal characteristics, including walking speed, age, road conditions, service opening hours, and season (Willberg et al., 2023), while network structure can substantially affect the spatial opportunities that are reachable from a given location. Network-based applications such as Gaglione et al. (2022) demonstrate the value of representing pedestrian-network geometry rather than relying solely on circular distance buffers. Accessibility also has a distributional dimension: proximity-oriented urban structures may coexist with substantial socioeconomic and spatial inequalities in access to everyday opportunities (Calafiore et al., 2022; Vale and Lopes, 2023). Comparative European evidence further shows that substantial intra-urban accessibility inequality may persist even where average pedestrian accessibility is comparatively favorable, reinforcing the need to distinguish aggregate accessibility performance from its spatial distribution.

For the Niš application, NUSIF operationalizes functional proximity through access to six categories of everyday urban services: food, health, education, daily services, culture, and recreation. These categories constitute the empirical taxonomy used in this study and should not be interpreted as a definitive or universally transferable list of 15-min-city functions. The component measures functional diversity rather than facility abundance: access to several functionally distinct categories is treated differently from the presence of multiple facilities belonging to the same category. This distinction is consistent with evidence that accessibility to everyday services depends not only on geographic proximity but also on the distribution and characteristics of available opportunities (Bittencourt and Giannotti, 2023).

Accessibility to the six service categories is evaluated within a 1,200 m pedestrian-network threshold. This distance is a study-specific modeling parameter rather than a universal conversion of 15 min into meters, since walking accessibility varies among population groups and environmental conditions (Willberg et al., 2023). The NUSIF service component should therefore be interpreted as an accessibility-based proxy inspired by the proximity logic of the 15-min city rather than a complete measure of 15-min-city compliance.

2.3 The 3–30–300 rule: from multidimensional greening to proximity

The 3–30–300 rule combines three distinct dimensions of urban-nature exposure: visibility of three trees, approximately 30% neighborhood canopy cover, and access to public green space within approximately 300 m (Konijnendijk, 2023). Its significance lies in shifting attention from aggregate city-wide greenery toward localized exposure and access. Empirical applications confirm that the three dimensions capture different aspects of urban greening and should not be treated as substitutes for one another (Croeser et al., 2024; Bertassello et al., 2026).

The components also require different measurement approaches. Browning et al. (2024) distinguish between methods appropriate for tree visibility, canopy coverage, and green-space proximity, supporting the need to operationalize them separately. The present Niš application therefore incorporates only the 300 m green-space proximity component. It does not assess tree visibility or neighborhood canopy coverage and consequently should not be interpreted as a complete assessment of 3–30–300 compliance.

Green-space proximity should also be distinguished from green-space quality and broader environmental performance. Accessible green spaces may differ substantially in size, vegetation, ecological condition, facilities, safety, maintenance, and thermal performance. The binary NUSIF green-space indicator therefore represents minimum network-based spatial access to an eligible public green destination rather than equivalent environmental quality across all accessible spaces.

The 300 m threshold is retained as the baseline because it corresponds to the proximity dimension of the original 3–30–300 framework. Scenario 2 extends effective network accessibility to 500 m as a counterfactual analytical condition; this should not be interpreted as a revision of the original rule. Rather, the scenario examines how broader effective green-space accessibility would alter the integrated NUSIF outcome while the remaining components are held constant.

2.4 Walking-public transport interface and the 90–50–70 heuristic

The third conceptual component requires particular caution because the 90–50–70 formulation differs in evidential maturity from the preceding concepts. As clarified in the Introduction, the EIT Urban Mobility formulation emphasizes the role of walking in accessing and experiencing public transport; it does not prescribe population-coverage, active-mode, or emissions-reduction targets. NUSIF therefore uses 90–50–70 as a motivating practitioner heuristic for examining the walking–public transport interface, while the empirical specification is grounded primarily in the peer-reviewed literature on pedestrian access to public transport.

Walking access to transit cannot be represented by a universally applicable distance threshold. Empirical studies show that effective catchments vary with transport mode, network structure, pedestrian connectivity, and local conditions, and that fixed service areas may either underestimate or overstate actual accessibility (Daniels and Mulley, 2013; García-Palomares et al., 2013; El-Geneidy et al., 2014; Zielstra and Hochmair, 2011). First−/last-mile accessibility also depends on the quality of the walking environment rather than distance alone (Ha et al., 2023). These findings support the use of network-based accessibility while requiring the resulting thresholds to be interpreted as analytical parameters rather than universal behavioral boundaries. This distinction is also consistent with the broader transit-oriented development (TOD) literature. TOD integrates transport planning, land-use planning, and urban design and therefore treats pedestrian access to transit as an enabling condition within a wider configuration of density, functional mix, urban form, and transport-service characteristics (Ibraeva et al., 2020). Stop proximity alone should consequently not be interpreted as evidence of TOD performance. Within NUSIF, the Tcomponent represents only first−/last-mile spatial availability and is therefore compatible with, but substantially narrower than, a comprehensive TOD assessment.

Accordingly, NUSIF adopts 500 m as the baseline pedestrian-network threshold for access to eligible public-transport stops, while Scenario 3 evaluates an extended effective catchment of 800 m. Neither threshold forms part of the formal 90–50–70 formulation; both are study-specific parameters informed by the wider transit-accessibility literature.

Public-transport accessibility is retained as a distinct NUSIF dimension because access to collective transport constitutes an important enabling condition within broader sustainable-mobility transitions (Rađenović et al., 2026). However, proximity to a stop represents only one element of transport accessibility. The present indicator does not capture service frequency, reliability, timetable coverage, route connectivity, fares, vehicle capacity, accessibility for persons with disabilities, or walking-environment quality. It should therefore be interpreted as an indicator of first−/last-mile spatial availability, while these additional service-quality dimensions belong to an extended formulation of NUSIF.

2.5 Comparative spatial logic: complementarity without equivalence

The three conceptual traditions are complementary but not equivalent. The 15-min-city tradition focuses primarily on functional opportunity and local proximity, the 3–30–300 rule introduces localized environmental exposure and green-space access, and the walking–public transport dimension connects the neighborhood with the wider urban system. Together, they therefore address different but complementary dimensions and spatial scales of everyday urban accessibility.

Their original measurement structures, however, differ substantially. Proximity-based urbanism may be represented through time, network distance, opportunity counts, or functional diversity; the 3–30–300 rule combines counts, percentages, and distance; and public-transport accessibility may incorporate stop proximity, service characteristics, travel time, or network connectivity. These original metrics cannot therefore be directly averaged without prior conceptual selection and normalization. Composite-indicator methodology similarly requires an explicit theoretical framework, transparent variable selection, normalization, weighting, and sensitivity assessment (OECD/European Union/European Commission–Joint Research Centre, 2008).

The NUSIF consequently does not aggregate the original planning concepts themselves. Instead, it extracts three spatially compatible operational dimensions—essential-service accessibility, green-space accessibility, and pedestrian accessibility to public transport—while preserving their component-specific interpretation. Their conceptual and operational relationships are summarized in Table 1.

Table 1. Conceptual comparison and empirical operationalization of numerical urban planning heuristics within NUSIF.

The 1,200 m service threshold and the 500 m public-transport threshold are study-specific analytical parameters rather than formal thresholds of the source planning concepts. The 300 m green-space threshold operationalizes only the proximity component of the 3–30–300 rule. The 90–50–70 formulation is used as a motivating practitioner heuristic; the empirical public-transport accessibility measure is grounded primarily in the peer-reviewed walking-to-transit literature.

2.6 Integration logic of NUSIF

The integration logic of NUSIF rests on the proposition that service proximity, green-space accessibility, and public-transport connectivity represent distinct but complementary conditions of sustainable urban accessibility. Service proximity can reduce the spatial separation between residents and everyday destinations, but it does not by itself guarantee sustainable travel behavior (Banister, 2008). Green-space accessibility introduces an environmental and quality-of-place dimension that may be overlooked by purely functional accessibility measures (Croeser et al., 2024; Bertassello et al., 2026). Public transport, in turn, connects locally accessible neighborhoods with activities and opportunities distributed across the wider urban system, while pedestrian access forms an important part of this connection (García-Palomares et al., 2013; Ha et al., 2023).

These dimensions should therefore be evaluated jointly, since strong performance in one domain may coexist with deficits in another. At the same time, NUSIF retains the individual A, G, and T indicators alongside the composite score because additive aggregation permits compensation among components. Following established composite-indicator guidance, the composite result is therefore interpreted together with its constituent indicators, while normalization and weighting assumptions remain explicit (OECD/European Union/European Commission–Joint Research Centre, 2008).

Equal weighting is used as a transparent reference specification rather than as an empirically optimized or normatively superior weighting structure. The purpose of NUSIF is thus not to erase differences among the three accessibility domains, but to integrate them at a common spatial scale while preserving the ability to identify which component drives a given composite outcome.

Composite accessibility indicators improve communicability and allow heterogeneous dimensions to be considered within a common analytical structure, but they also introduce methodological choices that can affect interpretation. Normalization or scaling determines how heterogeneous components are made comparable; weighting assigns their relative influence; and additive aggregation permits compensation between stronger and weaker component values. Threshold selection additionally determines which spatial relationships are classified as accessible, while spatial aggregation can affect observed patterns according to the scale and configuration of the analytical units. These issues are well established in composite-indicator methodology, where weighting, aggregation, and robustness analysis are treated as substantive rather than purely technical decisions (OECD/European Union/European Commission–Joint Research Centre, 2008; Greco et al., 2019). Spatial analysis introduces the related modifiable areal unit problem (MAUP), whereby results can vary according to the scale and zoning of the spatial units used for analysis; numerical urban-form research similarly emphasizes the importance of explicitly defining spatial support (Fleischmann et al., 2022). NUSIF addresses part of this uncertainty by retaining the individual A, G, and Tcomponents alongside the composite score, reporting fixed operational thresholds and classification rules, and testing alternative component weights. The present application does not, however, test alternative accessibility thresholds, component-scaling formulations, or analytical-grid resolutions; these sources of uncertainty are therefore retained explicitly as limitations and priorities for further robustness analysis.

2.7 NUSIF-Core and NUSIF-Extended

NUSIF-Core comprises the three spatial-accessibility dimensions that can be implemented consistently in the present Niš case study:

NUSIF − Core = { A , G , T }

The A component represents the diversity of accessible essential-service categories; G represents minimum network-based access to eligible public green space; and T represents minimum network-based pedestrian access to eligible public-transport stops. The three indicators are represented on a common theoretical range from 0 to 1 so that they can be integrated transparently.

NUSIF-Extended represents the broader conceptual architecture for more comprehensive future applications. Potential extensions include service characteristics such as travel time, capacity, affordability, and destination quality; environmental dimensions such as tree visibility, canopy coverage, green-space quality, and thermal performance; and public-transport dimensions such as frequency, reliability, connectivity, travel time, and first−/last-mile quality. Environmental accessibility and environmental performance are deliberately distinguished, since proximity to green infrastructure does not necessarily imply equivalent thermal or climate-adaptation benefits (Mančić et al., 2026).

The Core–Extended distinction makes partial operationalization explicit rather than presenting data availability as complete conceptual coverage. This is consistent with evidence that different dimensions of numerical planning concepts may require distinct measurement approaches (Browning et al., 2024; Papadopoulos et al., 2023) (Figure 1).

3 Materials and methods

3.1 Research design and empirical data

The study employed a spatially explicit case-study design to operationalize the Numerical Urban Sustainability Integration Framework (NUSIF) in Niš, Serbia. Three accessibility dimensions were translated into a common GIS-based analytical structure: essential-service accessibility (A), green-space accessibility (G), and public-transport accessibility (T). The components were evaluated independently, normalized to a common 0–1 scale, integrated into an equal-weight composite NUSIF score, and assessed from both territorial and population-weighted perspectives.

The empirical implementation represents selected spatial-accessibility dimensions of the three planning traditions rather than complete compliance assessments of the 15-min city, 3–30–300 rule, or 90–50–70 principle. Accessibility was evaluated through the pedestrian street network rather than Euclidean distance. The workflow proceeded from spatial-data preparation and network modeling to component construction, composite-index calculation, population weighting, and counterfactual scenario analysis. Each scenario modified one NUSIF component while retaining the others at baseline values, thereby isolating its modeled contribution without implying sequential implementation.

For terminological consistency, accessibility refers in this study to the modeled network-based condition by which an analytical cell satisfies a specified destination-access criterion, whereas proximity refers more narrowly to the distance relationship underlying green-space and public-transport criteria. Territorial coverage or territorial reach refers to the distribution of accessibility conditions or scenario-induced changes across analytical cells, while population exposure refers to the modeled resident population located in cells characterized by those conditions. Population-weighted accessibility gives greater influence to accessibility conditions occurring in cells containing more residents. Neither population exposure nor population weighting should be interpreted as observed facility use, perceived accessibility, or realized welfare benefit (Figure 2).

3.2 Study area and spatial unit of analysis

The empirical application was conducted in Niš, Serbia. The analytical study extent was defined using the Niš study-area boundary prepared within the GIS workflow and represented as a single polygon for spatial analysis. The boundary was transformed to WGS 84/UTM zone 34N (EPSG:32634), which was used for all metric accessibility calculations. A nominal 500 × 500 m grid served as the common spatial support for integrating population and accessibility indicators. Boundary-intersected cells were clipped to the study-area extent and therefore could have areas smaller than the nominal 0.25 km²; actual cell area was consequently retained in the analytical dataset. Following geometry and record-level quality control, the final dataset comprised 873 unique analytical cells. Each cell retained a unique cell_id, actual cell area (cell_km²), modeled 2025 resident population (pop_2025) and population density. The same set of 873 cells and identical population values was retained under the baseline and all three counterfactual scenarios. Across the corrected analytical grid, the modeled 2025 population totaled approximately 287,673 residents.

Population density for cell i was calculated as:

D i = P i S i

Where D i is resident population density in cell i, P i is the modeled 2025 population assigned to the cell, and S i is its actual area in km².

Fractional population values were retained because the source population product provides modeled population estimates as continuous values and these values were aggregated spatially to the analytical grid. The exact population product and population-to-grid aggregation procedure are documented in Section 3.3.1 (Figure 3).

3.3 Data sources and preparation

The empirical analysis combined gridded population data with OpenStreetMap-derived pedestrian-network, point-of-interest, public-green-space, and public-transport data. Because the datasets differed in their original geometry, spatial resolution, and coordinate reference systems, they were preprocessed and harmonized to a common analytical extent and spatial reference before the accessibility calculations were performed. Population data were first aggregated to the common analytical grid, whereas the vector datasets were subsequently prepared for pedestrian-network accessibility modeling.

3.3.1 Population data

Population information for the reference year 2025 was obtained from the Global Human Settlement Layer GHS-POP R2023A, produced by the European Commission Joint Research Centre (JRC) (Schiavina et al., 2023). The 2025 product used in this study, GHS_POP_E2025_GLOBE_R2023A_54009_100_V1_0, represents the spatial distribution of residential population as modeled absolute numbers of inhabitants per raster cell. The dataset was obtained at 100 m spatial resolution in the World Mollweide coordinate reference system, with population values represented as floating-point estimates. Population was transferred to the nominal 500 × 500 m analytical grid while both the source population raster and the aggregation grid were represented in the World Mollweide coordinate system. Using QGIS 3.44.12-Solothurn, zonal statistics were calculated for each grid polygon from the first band of the GHS-POP raster. The sum and count statistics were calculated using the prefix p25_, producing the fields p25_sum and p25_count, respectively. The summed population estimate was subsequently retained as the cell-level 2025 population variable (pop_2025). The population-enriched analytical grid was then transformed to WGS 84/UTM zone 34N (EPSG:32634) for integration with the pedestrian-network accessibility calculations. Cell-level population was used to calculate population density, population-weighted NUSIF values, and population distributions across the five NUSIF classes.

3.3.2 OpenStreetMap and Geofabrik data

OpenStreetMap vector data for Serbia were obtained through the Geofabrik Serbia extract. The datasets used in the empirical workflow included the road network, point and polygon points of interest, public-green-space features, and public-transport features. The original Geofabrik attribute structure was retained where relevant, particularly identifiers and the fclass variable used for feature selection and classification. Because the analytical outcome depends not only on the source database but also on the feature classes retained from it, the exact final inclusion and exclusion expressions used for pedestrian routing, essential services, public green spaces, and public transport are reported in Supplementary Table S1. Table 2 provides a concise overview of the principal spatial datasets and their roles in the empirical NUSIF application.

Table 2. Principal spatial datasets used in the empirical NUSIF application.

Detailed OpenStreetMap feature-class selection criteria are reported in Supplementary Table S1. Network-analysis and post-processing parameters are reported in Supplementary Table S2.

Essential urban destinations were organized into six functional categories:

J = { food , health , education , daily services , culture , recreaion }

Employment accessibility was not modeled as a separate destination category, while qualitative characteristics such as service capacity, affordability, opening hours and perceived quality were outside the scope of the present operationalization. Eligible OpenStreetMap point and polygon features were assigned to the six categories using study-specific fclass selection rules. Point features could be used directly as routing destinations, whereas eligible polygon-derived destinations were transformed into representative routing locations where required by the network workflow. For each analytical cell, accessibility to the six service categories was subsequently represented through separate binary indicators. These category-specific variables were aggregated into the number of accessible service categories, n i, and the normalized essential-service component, A i.

3.3.3 Public green-space and public-transport destinations

Public-green-space destinations were derived from eligible OpenStreetMap polygon features selected using the study-specific fclass criteria. The selection was intended to represent potentially accessible public green destinations rather than all mapped vegetation or land with ecological characteristics. Following geometry validation and transformation to WGS 84/UTM zone 34 N (EPSG:32634), a point-on-surface representation was generated for each eligible green-space polygon and used as the routing destination in the network-accessibility procedure. This ensured that the representative destination remained spatially associated with its source polygon while providing a point geometry compatible with the network-service-area workflow. The procedure represents a modeling abstraction of access to the green-space feature and should not be interpreted as identification of an observed or formally mapped pedestrian entrance. The empirical G component operationalizes only the 300 m proximity dimension of the 3–30–300 framework. It does not operationalize visibility of three trees or approximately 30% neighborhood canopy coverage, which require substantially different spatial datasets and measurement procedures (Konijnendijk, 2023; Browning et al., 2024). These aspects are retained conceptually for future NUSIF-Extended applications.

Public-transport destinations were derived from the OpenStreetMap transport-point data distributed in the Geofabrik Serbia extract, including the project layer gis_osm_transport_free_1.shp. Eligible transport features were selected using the final study-specific fclass criteria, transformed to EPSG:32634 and used as destinations for pedestrian-network accessibility modeling. The resulting public-transport component represents pedestrian proximity to eligible public-transport stops.

3.4 Spatial harmonization and pedestrian-network modeling

All spatial layers used for distance-dependent analysis were represented in WGS 84/UTM zone 34 N (EPSG:32634) so that network costs and post-processing distances could be expressed in meters. The population-enriched analytical grid, pedestrian network and accessibility destinations were harmonized to the same coordinate reference system and common study extent before component-level accessibility calculations were performed. QGIS 3.44.12-Solothurn was used for the GIS processing workflow. The pedestrian network was derived from the OpenStreetMap road layer by retaining road classes considered applicable to pedestrian routing. Network accessibility was calculated using the QGIS Service area (from layer) procedure under the Shortest strategy. Travel cost was therefore expressed as distance along the pedestrian network rather than Euclidean distance or modeled walking time. For the green-space and public-transport components, reachable network-line outputs were buffered by 50 m and dissolved to generate continuous accessibility footprints. These footprints were spatially related to the common analytical grid to identify cells satisfying the corresponding network-accessibility condition. Cell-level accessibility was assigned using the spatial-intersection criterion: an analytical cell was classified as accessible when the corresponding accessibility footprint intersected the cell polygon. Any spatial intersection was sufficient for assignment; no minimum overlap proportion or centroid-within-footprint criterion was imposed. The resulting binary value therefore represents accessibility status at the analytical-cell level and should not be interpreted as indicating that every location within a 500 × 500 m cell satisfies the corresponding network-distance threshold. Component-specific routing parameters, including final distance thresholds, destination-to-network settings and post-processing options, are documented in Supplementary Table S2.

Spatial joins were structured to retain the complete set of analytical cells, and cells receiving no accessibility match were explicitly assigned a value of zero rather than being removed. Record-level quality control was subsequently used to ensure uniqueness of cell_id and consistency among baseline and scenario datasets.

3.5 Operationalization of NUSIF accessibility components

3.5.1 Essential-service accessibility

For analytical cell i and essential-service category j, binary service accessibility was defined as:

a ij = { 1 , if service category j is accessible within 1 , 200 m along the pedestrian network 0 , otherwise

The 1,200 m threshold is used as a study-specific network-distance proxy for neighborhood-scale service accessibility. Because no fixed walking speed or time-dependent routing model was applied, the threshold should not be interpreted as a direct conversion of 15 min into distance or as an empirically observed walking-time boundary. The number of accessible service categories in cell i was calculated as:

n i = ∑ j = 1 6 a ij

Where

n i ∈ { 0 , 1 , 2 , 3 , 4 , 5 , 6 }

The normalized essential-service accessibility component was then calculated as:

A i = n i 6

so that

0 ≤ A i ≤ 1

An A i value of 0 indicates that none of the six essential-service categories satisfies the network-accessibility criterion, whereas A i = 1 indicates accessibility to all six categories. Intermediate values represent increasing functional diversity of accessible service categories.

Facility quality, capacity, affordability, opening hours, congestion and individual preferences are likewise outside the present NUSIF-Core specification.

3.5.2 Green-space accessibility

Baseline green-space accessibility was represented as a binary indicator of pedestrian-network access to an eligible public green space within 300 m:

G i = { 1 , if the 300 m green − space accessibility footprint reaches cell i 0 , otherwise

The 300 m criterion corresponds to the green-space-proximity component of the 3–30–300 framework (Konijnendijk, 2023). Its implementation through pedestrian-network distance provides a route-based interpretation of spatial access rather than simple straight-line separation and is consistent with the need to distinguish different measurement dimensions within the broader 3–30–300 framework (Browning et al., 2024).

3.5.3 Public-transport accessibility

Baseline public-transport accessibility was represented using a 500 m pedestrian-network threshold:

T i = { 1 , if an eligible public − transport destination is accessible within 500 m along the pedestrian network 0 , otherwise

Observed walking distances to public transport vary according to transport mode, urban form, service provision and user characteristics (Daniels and Mulley, 2013; El-Geneidy et al., 2014). Accordingly, the 500 m distance is treated as the baseline analytical catchment adopted for the Niš application, rather than as a universal behavioral threshold. Nor should the 500 m threshold be interpreted as a numerical component of the 90–50–70 formulation.

3.6 Construction and classification of NUSIF-Core

The three accessibility components were combined using an equal-weight arithmetic specification:

NUSI F i = A i + G i + T i 3

Because each component is bounded between 0 and 1,

0 ≤ NUSI F i ≤ 1

Equal weighting was adopted as a transparent reference specification in the absence of empirically derived local preference weights. Weighting and aggregation are substantive methodological choices in composite-indicator construction and can influence the resulting interpretation and ranking of spatial units (OECD/European Union/European Commission–Joint Research Centre, 2008). For descriptive interpretation and spatial comparison, the continuous NUSIF score was divided into five fixed classes:

C i = { 1 ( Very low ) , 0 ≤ NUSI F i < 0.20 , 2 ( Low ) , 0.20 ≤ NUSI F i < 0.40 , 3 ( Moderate ) , 0.40 ≤ NUSI F i < 0.60 , 4 ( High ) , 0.60 ≤ NUSI F i < 0.80 , 5 ( Very high ) , 0.80 ≤ NUSI F i ≤ 1.

These class boundaries are analytical interpretation intervals rather than externally validated sustainability standards. The same fixed thresholds were applied to the baseline and all three scenarios, ensuring that observed class changes represent changes in NUSIF rather than changes produced by scenario-specific reclassification (Table 3).

Table 3. Operational definition of the NUSIF-Core indicators.

The 1,200 m service and 500 m public-transport thresholds are study-specific analytical specifications. The 300 m green-space criterion operationalizes only the proximity component of the 3–30–300 rule.

3.6.1 Weighting sensitivity analysis

Because equal weighting represents a transparent reference specification rather than an empirically or normatively derived weighting structure, a sensitivity analysis was conducted to evaluate whether the principal NUSIF findings depended strongly on the assumed relative importance of the three accessibility components. The underlying GIS accessibility calculations, network thresholds, analytical grid, and scenario definitions were not modified. Instead, the already derived cell-level essential-service accessibilityA i, green-space accessibility G i, and public-transport accessibility T i components were recombined using alternative weighting specifications:

NUSI F i ( ω ) = ω A A i + ω G G i + ω T T i

subject to: ω A + ω G + ω T = 1, where ω A, ω G, and ω T denote the weights assigned to essential-service accessibility, green-space accessibility, and public-transport accessibility, respectively.

Three moderate-emphasis specifications assigned a weight of 0.50 to one component and 0.25 to each of the remaining components, while three stronger-emphasis specifications assigned a weight of 0.60 to one component and 0.20 to each of the remaining components. The complete weighting configurations are reported in Supplementary Table S4. These specifications were used as sensitivity tests rather than as estimates of stakeholder preferences or normative policy priorities. Sensitivity was evaluated by comparing the unweighted cell mean NUSIF, population-weighted NUSIF, scenario-specific changes relative to baseline, and cell-level correspondence between the equal-weight and alternative-weight specifications. Cell-level correspondence was quantified using Spearman’s rank correlation coefficient (ρ) between each alternative-weight NUSIF configuration and the corresponding equal-weight configuration.

3.7 Unweighted cell-based and population-weighted assessment

Two complementary aggregation perspectives were used to distinguish territorial coverage from population exposure. First, the unweighted cell mean was calculated by assigning equal weight to each analytical cell:

NUSIF ¯ = 1 N ∑ i = 1 N NUSI F i

Where

N = 873

This measure describes the mean accessibility condition across the final analytical spatial units without accounting for differences in the number of residents living in each cell. Second, population-weighted accessibility was calculated as:

NUSI F PW = ∑ i = 1 N P i NUSI F i ∑ i = 1 N P i

where P i is the modeled 2025 resident population assigned to cell i. The population denominator was held constant across the baseline and all three scenarios:

∑ i = 1 N P i = 287 , 673

The unweighted cell mean and population-weighted index were treated as complementary rather than competing indicators. The unweighted measure describes accessibility across equally weighted analytical cells, whereas the population-weighted measure gives greater influence to conditions occurring in cells containing more residents. This distinction enables a scenario affecting a relatively limited spatial area but a large population to be distinguished from one affecting a larger number of sparsely populated cells.

3.8 Counterfactual scenario design

Three independent counterfactual scenarios were developed to evaluate how changes in each NUSIF accessibility component affect integrated accessibility outcomes. The scenarios are therefore parallel and independent interventions rather than sequential or cumulative transformations. Population values, analytical-grid geometry and the other two NUSIF components were held constant when each individual component was modified. Importantly, these scenarios modify analytical accessibility conditions rather than directly modeling the construction of new infrastructure. In particular, S2 and S3 expand the network-distance thresholds applied to the existing mapped green-space and public-transport destinations; they do not introduce new green spaces, transport stops, routes, or services.

3.8.1 Scenario 1: marginal improvement in essential-service accessibility

For every analytical cell not already attaining the maximum of six accessible service categories, the number of accessible categories was increased by one:

n i S 1 = min ( n i + 1 , 6 )

The scenario-specific service component was:

A i S 1 = n i S 1 6

Green-space and public-transport accessibility remained at their baseline values:

G i S 1 = G i, T i S 1 = T i .

The scenario-specific composite score was therefore:

NUSI F i S 1 = A i S 1 + G i + T i 3

Cell-level change relative to the baseline was calculated as:

Δ i S 1 = NUSI F i S 1 − NUSI F i B

For a cell that had not already reached six accessible categories, an increase of one category changes A i by 1 / 6, resulting in a maximum direct composite increase of:

Δ i S 1 = 1 18 = 0.0556

Cells already attaining six accessible categories retain their baseline score.

3.8.2 Scenario 2: expanded effective green-space accessibility

Scenario 2 evaluates an increase in effective green-space accessibility by expanding the network-distance condition from the 300 m baseline to 500 m, while retaining essential-service and public-transport accessibility at their baseline values. The scenario-specific composite score was:

NUSI F i S 2 = A i + G i S 2 + T i 3

Cell-level change was:

Δ i S 2 = NUSI F i S 2 − NUSI F i B

Because G i is binary, a cell changing from G i = 0 under the baseline to G i S 2 = 1 receives a direct NUSIF increase of:

Δ i S 2 = 1 3

3.8.3 Scenario 3: expanded effective public-transport accessibility

Scenario 3 evaluates increased effective first−/last-mile access to public transport by extending the pedestrian-network threshold from the 500 m baseline to 800 m, while retaining baseline essential-service and green-space accessibility (Table 4). The corresponding composite score was:

NUSI F i S 3 = A i + G i + T i S 3 3

Table 4. Baseline configuration and independent counterfactual scenarios used in the NUSIF analysis.

S1, S2 and S3 are calculated independently from the same baseline and are not cumulative.

The cell-level difference from baseline was:

Δ i S 3 = NUSI F i S 3 − NUSI F i B

A cell newly satisfying the public-transport accessibility condition therefore receives:

Δ i S 3 = 1 3

4 Results

4.1 Baseline distribution of NUSIF

The baseline NUSIF values reveal a strongly differentiated accessibility structure across the 873 analytical cells. The unweighted cell mean was 0.2365, while the median was 0.1111, with a standard deviation of 0.2493. The first and third quartiles were 0.0556 and 0.3889, respectively, yielding an interquartile range of 0.3333. This pattern is also evident in the fixed five-class classification. Under baseline conditions, 492 cells (56.36%) were classified as very low NUSIF and a further 183 cells (20.96%) as low. Thus, 675 of the 873 analytical cells (77.32%) belonged to the two lowest classes. By contrast, only 66 cells (7.56%) were classified as high and 30 cells (3.44%) as very high.

The population distribution across these classes was markedly different from the spatial distribution. The very-low and low classes contained only 10.93 and 11.49% of the modeled population, respectively. Conversely, the high and very-high classes, although comprising only 11.00% of all analytical cells, contained 61.79% of the modeled population. Specifically, 103,920.5 residents (36.12%) were located in high-NUSIF cells and 73,845.7 (25.67%) in very-high-NUSIF cells. The baseline therefore exhibits a clear contrast between territorial coverage and population exposure: low accessibility characterizes a large share of the analytical cells, whereas a substantial majority of residents are concentrated in comparatively high-accessibility cells. The spatial structure underlying these baseline statistics is illustrated in Figure 4, which decomposes the baseline NUSIF into its three constituent accessibility components: essential-service accessibility (A), green-space accessibility (G), and public-transport accessibility (T), together with the resulting integrated baseline NUSIF.

4.2 Effects of the service-accessibility, green-space and public-transport scenarios

Scenario S1 increased the accessibility contribution of essential services by adding one accessible service category where the baseline value had not already reached the maximum of six categories. As a result, the unweighted cell mean NUSIF increased from 0.2365 to 0.2880, corresponding to an absolute increase of 0.0515 or 21.80% relative to the baseline unweighted cell mean. The score increased in 810 cells (92.78%), demonstrating that the service intervention had by far the widest spatial reach of the three scenarios. These cells contained approximately 159,211.6 residents, equivalent to 55.34% of the total modeled population. Nevertheless, because the maximum cell-level NUSIF increment generated by S1 was comparatively small, at 0.0556, most improved cells remained within their original NUSIF class. Only 128 cells (14.66%) crossed a classification threshold.

The resulting class transitions were systematic rather than diffuse: 43 cells moved from very low to low, 53 from low to moderate, and 32 from moderate to high. No cells moved into the very-high class. Consequently, the number of very-low cells declined from 492 to 449, while the number of high-NUSIF cells increased from 66 to 98. The proportion of the modeled population residing in the combined high and very-high classes increased from 61.79% under the baseline to 68.96% under S1.

Scenario S2 produced a substantially different response. Expanding effective green-space accessibility from the baseline threshold to the scenario threshold changed the NUSIF score in only 22 cells (2.52%). The unweighted cell mean therefore increased only modestly, from 0.2365 to 0.2449, an absolute change of 0.0084, equivalent to 3.55% of the baseline unweighted cell mean. Despite its limited spatial extent, every affected S2 cell crossed at least one NUSIF class boundary. The transitions comprised one cell from low to moderate, three from low to high, one from moderate to very high, and 17 from high to very high. Consequently, the number of very-high-NUSIF cells increased from 30 to 48. More importantly, the population represented by the very-high class increased from 73,845.7 residents (25.67%) to 119,041.6 residents (41.38%). The disproportionate population effect arises because the 22 affected cells contain approximately 47,974.8 residents, or 16.68% of the modeled population.

Scenario S3, which expanded effective public-transport accessibility, produced a broader territorial response than S2. A total of 110 cells (12.60%) experienced a score increase, corresponding to approximately 22,039.4 residents (7.66% of the total population). The unweighted cell mean increased to 0.2785, representing an absolute increase of 0.0420 or 17.76% relative to the baseline mean. All 110 affected cells crossed a NUSIF class threshold. Most of these changes occurred in the lower part of the baseline distribution: 61 cells moved from very low to low, 29 from very low to moderate, six from low to moderate and 12 from low to high. Two additional cells moved from high to very high. The number of very-low cells consequently declined from 492 to 402, the largest reduction in the lowest class among the three scenarios. The population share represented by this class decreased from 10.93 to 7.67%.

The contrasting spatial response profiles of the three counterfactual scenarios are summarized in Figure 5. By applying identical NUSIF class thresholds, colors, spatial extent, and analytical units to the baseline and all three scenarios, the figure enables direct cell-by-cell comparison of the resulting accessibility patterns.

4.3 Territorial and population-weighted comparison of scenarios

The unweighted cell-based and population-weighted assessment produce different scenario rankings, demonstrating why both perspectives are required when evaluating accessibility interventions. Under the baseline, the population-weighted NUSIF was 0.6106, substantially above the unweighted unweighted cell mean of 0.2365. This difference reflects the concentration of the modeled population in cells with comparatively high baseline accessibility.

S1 increased the population-weighted NUSIF to 0.6414, corresponding to an absolute improvement of 0.0307 and a relative increase of 5.04%. S2 produced the highest population-weighted value, 0.6662, representing an absolute increase of 0.0556 and a relative increase of 9.10%. S3 increased the population-weighted index to 0.6361, an absolute improvement of 0.0255, or 4.18% relative to the baseline. The ranking therefore depends on the evaluation perspective. Based on the unweighted cell-mean change, the scenarios rank:

S 1 ( + 0.0515 ) > S 3 ( + 0.0420 ) > S 2 ( + 0.0084 )

In contrast, when the distribution of the resident population is taken into account, the ranking becomes:

S 2 ( + 0.0556 ) > S 1 ( + 0.0307 ) > S 3 ( + 0.0255 )

S1 distributes relatively small improvements across 810 cells, whereas S2 generates a substantially larger cell-level NUSIF increment in only 22 cells that collectively contain 16.68% of the study population. S3 produces the same maximum component-induced cell-level increment as S2 across a larger number of cells, but these contain only 7.66% of residents (Table 5).

Table 5. Comparative unweighted cell-based and population-weighted effects of the NUSIF scenarios.

Δ values are calculated relative to the baseline. “Cells with changed NUSIF” denotes cells for which the scenario produced a non-zero change in the continuous NUSIF score, irrespective of whether the change crossed a five-class classification threshold. Population values are derived from the modeled 2025 population assigned to the fixed 873-cell analytical grid. Population and grid structure are identical across the baseline and all counterfactual scenarios.

4.4 Sensitivity of NUSIF to alternative component weights

Across the seven weighting specifications, the baseline unweighted cell mean ranged from 0.1565 to 0.2803, compared with 0.2365 under the equal-weight reference specification. The corresponding population-weighted NUSIF ranged from 0.4752 to 0.6854, compared with the equal-weight value of 0.6106. Under every weighting specification, the population-weighted baseline value remained substantially higher than the corresponding unweighted cell mean, indicating that the concentration of residents in relatively accessible cells was not an artifact of the equal-weight specification. Cell-level results also showed a high degree of rank-order stability. For the baseline configuration, Spearman rank correlations between the alternative-weight and equal-weight NUSIF values ranged from ρ = 0.9749 to ρ = 1.0000. When the baseline and all three scenario configurations were considered together, correlations ranged from ρ = 0.9655 to ρ = 1.0000. The magnitude and, in some cases, the ordering of scenario effects were more sensitive to the weighting specification. Under equal weighting, the unweighted cell-based ranking was S1 > S3 > S2. This ordering was retained under both service-priority and green-space-priority specifications, including the stronger 0.60 weighting variants, whereas assigning greater weight to public-transport accessibility changed the unweighted cell-based ranking to S3 > S1 > S2. Population-weighted rankings were more responsive to the component receiving additional weight. S2 ranked first under equal weighting, both green-space-priority specifications, and the moderate public-transport-priority specification; S1 ranked first under both service-priority specifications; and S3 ranked first under the strong public-transport-priority specification.

The central distinction between spatial reach and population exposure therefore persisted across the sensitivity analysis, although the relative ranking of individual scenarios was dependent on the component-weight structure. A concise summary of baseline robustness and scenario-response stability under the equal-weight and moderate component-priority specifications is provided in Supplementary Table S3. The complete seven weighting specifications are reported in Supplementary Table S4, while the full territorial and population-weighted sensitivity results for all seven specifications are reported in Supplementary Table S5.

5 Discussion

5.1 Main findings and contribution of the NUSIF application

The empirical application of NUSIF shows that integrated urban accessibility can produce substantially different interpretations depending on whether performance is evaluated across spatial units or according to the residential population contained within them. Three findings are particularly important. First, baseline accessibility in Niš was spatially uneven. The result is consistent with the broader accessibility literature showing that citywide or average indicators can conceal substantial intra-urban heterogeneity and that accessibility outcomes depend strongly on the spatial distribution of both opportunities and residents (Calafiore et al., 2022; Bittencourt and Giannotti, 2023). The literature reviewed for the framework similarly establishes that spatial coverage and population exposure should not automatically be interpreted as equivalent outcomes.

Second, the reversal between the unweighted and population-weighted scenario rankings is one of the central empirical findings of the study. The result shows why intervention appraisal based on a single citywide indicator can be misleading. Third, the weighting sensitivity analysis showed that alternative component weights affected the magnitude and, in some cases, the ordering of scenario responses, while the distinction between territorial and population-weighted evaluation persisted across all seven weighting specifications. Taken together, these findings support the principal purpose of NUSIF: not to identify a universally superior urban-planning rule or intervention, but to create a transparent analytical structure through which heterogeneous accessibility dimensions can be examined jointly while retaining their separate spatial and distributional meanings.

5.2 Territorial accessibility and population concentration

The pronounced divergence between the unweighted and population-weighted baseline results deserves particular attention. The baseline population-weighted NUSIF of 0.6106 was substantially higher than the unweighted spatial-cell mean of 0.2365. This indicates that residents are disproportionately concentrated in cells with relatively favorable combinations of service, green-space and public-transport accessibility. This pattern should not be interpreted as evidence that accessibility inequality is unimportant in Niš. Instead, it demonstrates that spatial disadvantage and population disadvantage are different analytical questions. Low-accessibility cells may encompass fewer residents but can still represent substantial portions of the urban or peri-urban spatial structure. Conversely, high accessibility in densely populated locations can increase the population-weighted citywide score while leaving numerous peripheral or less densely populated cells poorly served.

Calafiore et al. (2022) show that proximity-oriented urban structures can coexist with marked distributional differences, while Willberg et al. (2023) demonstrate that accessibility estimates themselves vary according to characteristics such as walking speed, age, season, road conditions and opening hours. These studies reinforce the need to avoid interpreting aggregate accessibility as equivalent to universally experienced accessibility. The distinction is also consistent with the broader accessibility framework developed by Geurs and van Wee, in which accessibility is understood as emerging from interacting land-use, transport, individual and temporal conditions rather than geographic proximity alone. Its population weighting adds a distributional dimension to the spatial indicators, but it does not transform the framework into an individual-level or perceived-accessibility model.

Accordingly, the present results support reporting both the unweighted spatial-cell measure and the population-weighted measure, rather than choosing one as the definitive representation of urban accessibility. The former is useful for identifying spatial coverage and gaps across the analytical grid; the latter indicates how strongly those conditions coincide with the residential distribution.

5.3 Interpretation of the three scenario response profiles

The scenario analysis further shows why the three NUSIF components should not be interpreted as interchangeable policy levers. This result aligns with the logic of proximity-oriented planning, in which the diversity of everyday opportunities is a central characteristic of local accessibility (Pozoukidou and Chatziyiannaki, 2021; Khavarian-Garmsir et al., 2023). However, the present S1 transformation should not be read literally as evidence that one additional physical facility should be provided in every affected cell. Service accessibility is influenced by the spatial organization, availability, capacity and quality of opportunities, not simply facility counts (Bittencourt and Giannotti, 2023). The S1 result is therefore best understood as demonstrating the potential spatial consequences of a marginal increase in functional accessibility.

The markedly different response to S2-Green illustrates the importance of distinguishing spatial reach from population exposure. This finding should not be generalized into a claim that green-space interventions are intrinsically more effective than service or transport interventions. Rather, it reflects the specific spatial coincidence between the cells newly satisfying the green-access condition and the residential population distribution in Niš. Therefore, the strong population-weighted S2 result concerns effective network proximity to eligible public green space, not overall urban greening performance.

S3-Transport produced a different form of spatial improvement. The result suggests that improved first−/last-mile spatial availability can be particularly relevant for analytical cells where baseline integrated accessibility is weak. This finding is consistent with research demonstrating that walking access to public transport varies substantially according to urban and service context (Daniels and Mulley, 2013; García-Palomares et al., 2013; El-Geneidy et al., 2014). The transport-accessibility literature also emphasizes that first−/last-mile conditions depend on characteristics of the pedestrian environment rather than distance alone (Ha et al., 2023). Consequently, the S3 result should be interpreted as an improvement in spatial availability of public-transport access, not as evidence of improved public-transport service quality, ridership or modal shift.

5.4 Implications of integrating heterogeneous planning heuristics

NUSIF addresses the incompatibility among heterogeneous planning heuristics by integrating only three explicitly selected spatial-accessibility dimensions—essential-service accessibility (A), green-space accessibility (G), and public-transport accessibility (T)—after normalization to a common range. This approach enables joint assessment while preserving the visibility and interpretation of the individual components, thereby reducing the risk that aggregation conceals the specific source of accessibility deficits. Composite-indicator methodology nevertheless emphasizes that weighting and aggregation can materially affect results and should remain explicit (OECD/European Union/European Commission–Joint Research Centre, 2008).

The same caution applies to the comparison of the three counterfactual scenarios because the numerical perturbations are structurally asymmetric. Adding one accessible service category produces a smaller maximum cell-level NUSIF increment than changing a binary green-space or public-transport indicator from 0 to 1. The scenarios should therefore not be interpreted as cost-equivalent or intervention-equivalent policy alternatives. Rather, their purpose is diagnostic: they demonstrate how the integrated index responds when different accessibility dimensions are modified independently under controlled counterfactual conditions.

The weighting sensitivity analysis provides an additional test of this interpretation. The results indicate that the principal distinction between territorial accessibility and population exposure remains robust under the alternative weighting assumptions examined here, while the identity and ordering of the highest-performing scenarios remain sensitive to the component-weight structure. NUSIF should therefore not be interpreted as producing a weighting-independent ranking of policy alternatives. Equal weighting is retained as a transparent reference specification rather than as a claim that essential-service, green-space, and public-transport accessibility possess objectively identical policy importance.

5.5 Planning and decision-support implications

From a planning perspective, NUSIF is better suited to multi-criteria diagnosis than to producing a single policy ranking. The three scenarios illustrate how different planning objectives can lead to different priorities: S1 is most relevant when the objective is widespread incremental improvement across the analytical territory, S2 when greater emphasis is placed on accessibility gains affecting larger resident populations, and S3 when the priority is to reduce the spatial prevalence of very-low integrated accessibility.

These differences reflect legitimate planning and distributional choices rather than a single hierarchy of interventions. NUSIF can therefore be used as a screening framework to identify spatial accessibility deficits and to compare how alternative intervention logics affect territorial and population-weighted outcomes. It should not, however, substitute for project appraisal, cost–benefit analysis, transport modeling, service-capacity assessment, or participatory priority setting. Candidate intervention areas identified through NUSIF should subsequently be evaluated in relation to costs, feasibility, land ownership, service capacity, environmental quality, transport operations, and community priorities. In this way, the framework can support an initial transparent diagnosis while recognizing that implementation decisions require more detailed evidence.

An additional area of potential application concerns sustainable urban tourism. Visitor mobility depends on accessibility among arrival gateways, accommodation, attractions, public spaces, and public transport, while accessibility alone does not determine the attractiveness or competitiveness of a destination. Tóth and Dávid (2010), for example, demonstrate that the relationship between transport accessibility and tourism outcomes must be interpreted together with the local characteristics of the destination. For NUSIF, tourism is therefore better considered as a potential extension rather than as an additional NUSIF-Core component. Future visitor-oriented applications could introduce tourist attractions and accommodation as destination categories, transport terminals as urban gateways, and seasonal or daytime visitor populations as alternative weighting surfaces. Such an extension could enable comparison between resident-oriented and visitor-oriented accessibility and help identify whether tourism-related mobility pressures are concentrated in already well-served central areas. The present Niš application remains explicitly resident-centered and does not make claims regarding tourist accessibility.

5.6 Limitations

Several limitations should be considered when interpreting the results. First, NUSIF-Core represents a partial operationalization of the three planning traditions rather than a complete assessment of any of them. The essential-service component captures functional diversity across six everyday service categories rather than facility abundance: accessibility to one eligible facility and to multiple facilities within the same category contributes equally to the category-level indicator. The formulation also treats the six categories equally and excludes employment accessibility and qualitative or capacity-related characteristics such as opening hours, affordability, service quality, congestion, and individual preferences. The green-space component measures network proximity but not tree visibility, canopy coverage, environmental quality, capacity, safety, or thermal performance, while the public-transport component captures pedestrian proximity to eligible stops without incorporating frequency, reliability, timetable coverage, connectivity, fares, capacity, accessibility for persons with disabilities, or observed ridership. These omitted dimensions are conceptually retained within NUSIF-Extended.

Second, the analysis relies substantially on OpenStreetMap-derived network and destination data. OSM provides broad, open, and reproducible spatial coverage, but feature completeness and attribute consistency can vary across locations and feature classes, and the analysis necessarily reflects the mapped database at the time of extraction. Missing or incorrectly classified roads, services, green spaces, or public-transport features could consequently affect network reachability and destination classification. Reporting the exact inclusion and exclusion criteria in Supplementary Table S1 strengthens methodological reproducibility but cannot eliminate uncertainty in the underlying source data. Future applications should therefore validate or compare OSM-derived features with authoritative local inventories where such data are available.

Third, accessibility is represented through network distance rather than observed travel behavior, perceived accessibility, or individual travel time. The 1,200 m service, 300 m green-space, and 500 m public-transport thresholds provide transparent operational parameters, but they do not capture variation in walking speed, terrain, crossing conditions, physical ability, weather, service opening hours, or individual preferences. The 300 m green-space threshold derives from the proximity component of the 3–30–300 rule, whereas the service and public-transport thresholds are study-specific parameters. Importantly, the present sensitivity analysis tests alternative component weights but not alternative network-distance thresholds. Robustness of the results to alternative catchments therefore remains an empirical question for future applications.

Fourth, the nominal 500 × 500 m analytical grid introduces spatial aggregation effects and the modifiable areal unit problem (MAUP). Results may vary under alternative cell sizes, grid alignments, boundary configurations, or origin representations because spatial patterns depend partly on the support on which they are measured (Fleischmann et al., 2022). Boundary-intersected cells may also be smaller than complete cells; consequently, the unweighted cell mean should be interpreted specifically as an equal-cell rather than an area-weighted assessment. Future robustness analysis should examine alternative grid resolutions and alignments and, where appropriate, area-weighted territorial measures. Because accessibility is assigned at the cell level, a cell intersecting an accessibility footprint may contain locations that individually fall outside the corresponding network-distance threshold; this represents an additional spatial-aggregation simplification of the present grid-based operationalization.

Fifth, the population layer represents a modeled 2025 residential population surface rather than individual-level observations. Aggregation to the analytical grid provides a consistent basis for population weighting but does not remove uncertainty associated with the source population model or with the spatial allocation of residents. Moreover, residential population does not capture daytime population, commuting flows, visitors, seasonal populations, or individual socioeconomic and mobility characteristics such as age, disability, income, household structure, or car availability. Population-weighted NUSIF should therefore be interpreted as a model-based measure of residential exposure to accessibility conditions, not as evidence of observed use, individual experience, or distributional equity among social groups.

5.7 Transferability and directions for future research

The NUSIF structure is transferable because it distinguishes between the conceptual architecture of the framework and the empirical parameters adopted in a particular urban context. The three components, destination taxonomy, network thresholds, weighting structure, and classification rules can therefore be adapted to local planning objectives and data availability rather than treated as universal constants.

Future research should pursue four main extensions. First, robustness analysis should move beyond component-weight sensitivity by testing alternative accessibility thresholds, analytical-grid resolutions and alignments, spatial aggregation procedures, alternative component-scaling formulations, and locally elicited or participatory weighting schemes. Future applications should also assess concordance between OpenStreetMap-derived features and authoritative local datasets where available and examine the sensitivity of population-weighted outcomes to alternative population surfaces. Second, NUSIF-Extended should incorporate richer representations of environmental and mobility performance, including tree visibility, canopy coverage, green-space quality, cycling, public-transport frequency and reliability, and pedestrian-environment quality. As Browning et al. (2024) demonstrate, different dimensions of green accessibility require distinct data and measurement approaches.

Third, future applications should incorporate social and distributional vulnerability variables. Population weighting identifies where residents are concentrated but cannot establish whether accessibility gains are distributed equitably among groups facing greater socioeconomic or mobility disadvantage. Combining NUSIF with indicators such as age, income, disability, car availability, or deprivation would enable a more explicit equity assessment. Alternative weighting surfaces representing daytime workers, commuters, or visitors could also complement residential population weighting where temporal population dynamics are important.

Fourth, the framework should be tested across multiple cities and urban forms. Comparative applications would help determine whether the divergence observed in Niš between spatial reach and population-weighted accessibility gain is context-specific or represents a more general characteristic of integrated accessibility interventions. Such comparisons should retain transparent local parameterization rather than impose identical thresholds regardless of urban morphology, pedestrian conditions, and transport-system characteristics.

6 Conclusion

The Niš application shows that integrated accessibility assessment can lead to materially different planning interpretations depending on whether emphasis is placed on spatial coverage or on the resident population located in cells receiving accessibility improvements. Most importantly, the scenario producing the broadest spatial improvement was not the scenario producing the greatest population-weighted gain: the unweighted assessment ranked the interventions S1 > S3 > S2, whereas population weighting produced S2 > S1 > S3. This divergence shows that intervention performance cannot be adequately characterized by a single aggregate score and that spatial reach, magnitude of change, and population exposure should be considered jointly. The findings therefore support the use of NUSIF primarily as a transparent diagnostic and comparative decision-support framework rather than as a mechanism for identifying a universally optimal intervention. The Niš case provides an empirical proof of concept while also indicating the limits of the current implementation, particularly its simplified representations of service diversity, green-space proximity and public-transport access and its dependence on explicit assumptions concerning thresholds, weighting and spatial resolution. Future applications should extend robustness testing to alternative accessibility thresholds, analytical-grid resolutions, spatial aggregation procedures, locally elicited weighting structures, and contrasting urban contexts, while incorporating richer environmental, mobility and socioeconomic dimensions. The weighting sensitivity analysis further showed that the central distinction between territorial and population-weighted evaluation persisted across all seven weighting specifications. Cell-level spatial ordering remained highly stable across the alternative weighting configurations, with Spearman correlations ranging from 0.9655 to 1.0000, while population-weighted baseline accessibility remained substantially higher than the corresponding unweighted cell mean under every weighting structure. Unweighted cell-based and population-weighted scenario rankings also remained non-identical across all seven weighting specifications. However, the precise ordering of the scenarios changed when greater importance was assigned to the component directly affected by a given counterfactual intervention. The analysis therefore supports the robustness of the central distinction between spatial reach and population-weighted accessibility gain while confirming that scenario ranking should remain explicitly conditional on the adopted weighting structure. The principal contribution of the study is not the introduction of new individual accessibility measures, but the development and empirical demonstration of a transparent integration architecture through which established but conceptually distinct accessibility dimensions can be examined jointly. By preserving component-level information, distinguishing territorial from population-weighted outcomes, testing the sensitivity of the composite index to alternative component weights, and comparing component-specific counterfactual response profiles, NUSIF provides a structured means of identifying where accessibility improvements occur, how substantial they are, and how strongly they coincide with the spatial distribution of residents.

Generative AI statement

The author(s) declared that Generative AI was used in the creation of this manuscript. During the preparation of this manuscript, the authors used generative AI solely for language editing, including spelling, grammar, and minor improvements in textual clarity, and as a graphical rendering aid for Figures 1, 2. The manuscript was authored by the authors, and no generative AI was used to develop the study design, methodology, data analysis, calculations, scientific interpretation, or conclusions. The concepts, structure, scientific content, equations, labels, and information presented in Figures 1, 2 were developed entirely by the authors; AI assistance was limited to their visual rendering. All AI-assisted edits and graphical outputs were reviewed and approved by the authors, who take full responsibility for the final work.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frsus.2026.1971777/full#supplementary-material

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Footnotes

Edited by: Jinghu Pan, Northwest Normal University, China

Reviewed by: Lóránt Dénes Dávid, John von Neumann University, Hungary

Rodrigo Henryque Reginato Quevedo Melo, Federal University of Rio Grande do Sul, Brazil