On the need for a multi-dimensional framework to measure accessibility to urban green

As the global urban population is projected to reach 68% by 2050, cities increasingly rely on urban greening and nature-based solutions to improve public health, well-being, and environmental performance. Public Green Areas (PGAs) have been linked to healthier lifestyles, social cohesion, reduced mortality, lower cardiovascular risks, and improved mental health and cognition. PGAs also support biodiversity, carbon storage, soil protection, and heat mitigation. Policy frameworks such as the UN New Urban Agenda and SDG 11.7 emphasize universal access to safe, inclusive, and accessible green public spaces, while recent paradigms (e.g., the 3–30–300 rule) articulate multi-level targets for green visibility, canopy cover, and proximity. Despite proliferating targets and growing emphasis on spatially resolved, data-driven indicators, there is no universally adopted framework to measure accessibility to urban green. Existing efforts vary from minimum distance to nearest parks to exposure metrics from satellite-derived green intensity, with limited consideration of walkable catchments at large scale and challenges integrating behavioral usage data due to availability and representativeness concerns. The research question addressed is whether accessibility patterns derived from different indicators are interchangeable and how parameter choices affect assessments. The purpose is to build and apply a multi-dimensional framework for green accessibility, evaluate indicator stability under parameter perturbations, and assess the overlap and interchangeability among institutional targets to guide policy-making.

Prior work has operationalized green accessibility using diverse data and metrics. Distance-to-park indicators have been built from administrative datasets, OpenStreetMap (OSM), and satellite-derived land use; exposure metrics have used land cover and vegetation indices (e.g., NDVI). Large-scale studies like the GHS-UCDB include immediate exposure measures but often neglect walkable catchments due to computational complexity, which are more common in single-city studies. Emerging research incorporates behavioral data from surveys and user-generated sources (social media, mobility traces, sports tracking, PPGIS) to estimate actual greenspace usage and perceived qualities, but scalability and representativeness constraints limit their broader application. Despite these advances, no systematic investigation has assessed the interchangeability of accessibility patterns across indicators or the policy implications of choosing one indicator over another, revealing a gap this study addresses.

The study develops and deploys a computational framework to measure three families of structural green accessibility indicators across 1040 urban centers (UCs) in 145 countries, using a minimal geographical unit of 9 arc-sec (~196 m at 45° latitude). Distances are computed on the walkable street network assuming a base walking speed of 5 km/h. Indicators are parameterized by greenery type, minimum size thresholds, and (when applicable) walking time budgets; walkable distances can be customized for specific populations (e.g., elderly). Data sources: (1) GHS-POP 2015 for granular population estimates; (2) OSM (May 2022) for PGAs and street network used in minimum distance and per-person indicators and OSRM for routing; (3) ESA WorldCover 2020 (WC-ESA) for land cover used in the exposure indicator and for sample validation. Urban centers are defined using GHS-UCDB (2015, R2019A); up to 50 most populated per country with at least 100,000 inhabitants were retained, then filtered for sufficient OSM green feature quality by comparing OSM-derived green intensity with WC-ESA to derive acceptance intervals, yielding 1040 UCs. The city space is gridded at 9 arc-sec (WGS-84). PGAs are extracted from OSM (parks, grass, forests) and reclassified; urban green coverage for exposure is extracted from WC-ESA (codes 10, 20, 30). Pre-processing includes boundary extraction, buffered clipping (3 km) to mitigate edge effects, OSM local dumps and feature extraction, raster remapping to the analysis grid, and OSRM-based walking distance matrices. Indicator definitions: (1) Minimum distance (MD) measures minutes to the nearest PGA meeting size/type criteria, via the minimum of distances masked by the presence of qualifying green. (2) Exposure (EXP) measures cumulative hectares of green features within a time budget t using WC-ESA land cover within the isochrone. (3) Per-person (PP) measures square meters per person of PGAs within time budget t by apportioning population to PGAs within t proportionally to PGA size and computing green per-person for each cell. Stability analysis evaluates robustness of indicators to parameter perturbations along three dimensions: (i) Kendall rank correlation of area rankings between parameterizations; (ii) stability of targeted populations under two strategies: naive targeting (bottom y% by indicator performance; for MD lower is better; for EXP/PP higher is better) using a weighted Jaccard overlap, and most-disadvantaged targeting (for EXP/PP: zero-access group; for MD: performing y-times worse than mean); and (iii) inequality measured by population-weighted Gini of indicator distributions. The framework is exposed via an interactive web application with functionalities to explore, measure, compare, create, and draw scenarios for policymakers and the public.

- Across cities, small parameter changes can substantially alter area and population rankings for green accessibility, indicating instability when relying on a single indicator or fixed threshold. For the minimum distance indicator, increasing the PGA minimum size reduces stability under both targeting strategies. Median stability under 2% naive targeting drops from 0.77 [IQR: 0.48–1] for a shift from 0.5 ha to 1 ha to 0.53 [IQR: 0.28–0.94] for a shift to 2 ha, corresponding to median conflicting target populations of 12% [IQR: 35%–0%] and 30% [IQR: 56%–3%], respectively. City examples for a 0.5→1 ha change: Sydney 0.32 (51% conflicting), Rio de Janeiro 0.36 (47%), London 0.47 (36%). - Higher area-level rank correlation can mask substantial reshuffling at the bottom of the distribution; conflicting targeted areas tend to have higher population density than stable targeted areas (density ratio > 1 across cities). - Targeted areas cluster more as the minimum PGA size increases, reflecting the spatial scarcity of larger PGAs. - Inequality patterns (weighted Gini) under parameter changes vary by city with no universal trend: some cities show decreasing inequality with larger size thresholds (e.g., Singapore), others increasing (e.g., New York), or U-shaped (e.g., Sydney, London). - For the exposure indicator, a time budget increase from 5 to 10 minutes yields median Kendall rank correlation of 0.69 [IQR: 0.63–0.74] and median 2% naive targeting stability of 0.38 [IQR: 0.23–0.63], indicating notable instability for targeted populations. - Performance against institutional targets shows geographical patterns: European and Australia–Oceania cities generally outperform Global South and North American cities on MD and per-person metrics. Exposure targets are met by larger population shares, especially in Global South and North American cities where green is less organized in public areas or dominated by private gardens/suburbs. - Cross-target stability analysis: Indicators within the same family show higher overlap in targeted populations than across families. The lowest stability occurs between WHO (MD-based) and ESA (exposure-based) indicators due to different green feature definitions (ESA includes elements beyond parks/grass/forests). Among short-distance MD indicators, the overlap between WHO and B1 is smaller than between WHO and N1, suggesting the time threshold exerts stronger influence than the minimum size in short-distance MD evaluations.

Findings demonstrate that accessibility assessments are sensitive to indicator choice and parameterization, undermining the reliability of single-metric, fixed-threshold approaches for prioritizing areas or populations. Instability in bottom-ranked groups means small parameter shifts can reassign target populations, and consistently under-performing areas tend to be less densely populated, complicating policy prioritization. While the study’s targeting strategies are simplified and do not capture real-world constraints or priorities (e.g., feasibility, demographic focus, gentrification risks, co-creation), they reveal the challenge of robustly identifying disadvantaged groups with single indicators. Cross-target analyses further show limited interchangeability between metrics designed to capture different forms of accessibility (e.g., proximity vs. exposure) and even notable discrepancies among ostensibly similar targets, reinforcing the necessity of multi-indicator assessments aligned with multi-level policy goals. A multi-dimensional framework enables two critical functions: evaluating the dependence of outcomes on fixed thresholds and concurrently capturing complementary forms of green accessibility to provide a fuller picture for equitable, effective interventions.

The study introduces a scalable, multi-indicator computational framework for measuring green accessibility—minimum distance, exposure, and per-person—applied to 1040 cities worldwide using harmonized spatial data and walkable catchments. Analyses show that rankings and targeted populations are often unstable to modest parameter changes and that institutional targets from different families are not interchangeable, with discrepancies even among similar MD targets. Consequently, single-indicator assessments can inadequately differentiate disadvantaged areas or subgroups. The work advocates for multi-dimensional evaluation of green accessibility in policy design, supported by the released interactive platform that allows exploration, measurement, comparison, and custom indicator creation. Future research should integrate richer, policy-relevant attributes of greenspaces (e.g., services, biodiversity, safety/quality), improve data completeness and harmonization across sources, and refine approaches for equitable, context-sensitive prioritization while considering risks such as green gentrification.

Primary limitations arise from data completeness and standardization. OSM green feature mapping may be incomplete or heterogeneous across regions, potentially biasing MD and PP indicators that depend on OSM tags and definitions of PGAs; quality checks and filtering reduced the city sample from ~2500 to 1040, but residual biases may remain. The reliance on OSM for PGA classification can mask variation in park characteristics (green level, services, vegetation) across countries or climate zones. Exposure measures depend on WC-ESA land cover categories and their accuracy. Behavioral usage data are not integrated at scale due to availability and representativeness constraints. Although instability metrics compare parameterizations within the same city (limiting cross-city bias), broader generalizability is still conditioned by data quality. Future work should add new data layers (services/facilities, biodiversity, environmental quality, safety), systematically assess multiple green data sources for completeness/suitability, and collaborate with local authorities to enrich attributes while addressing scalability challenges.