Skip to content
Back to projects

Technical project

Spatial Justice through Urban Green Space Connectivity

Reproducible R pipeline and spatial MCDA that assesses green space accessibility, ecological quality, equity and connectivity in Delft and Yuexiu (Guangzhou) to prioritise nature-based solutions.

  • R
  • sf / terra
  • Quarto
  • OSM
  • Sentinel-2 NDVI
  • GBIF
  • Spatial MCDA

Overview

Group project for Applied Spatial Analytics at TU Delft (2026), with Roxanne Vuijk and Belina Aileen. We built a fully reproducible R pipeline that evaluates urban green space (UGS) quality across Delft (Netherlands) and Yuexiu District, Guangzhou (China), and combines four dimensions — accessibility, ecological quality, spatial justice and connectivity — in a Spatial Multi-Criteria Decision Analysis (S-MCDA) that pinpoints where nature-based solutions are needed most.

Context

Rapid urbanisation puts pressure on green space, and its benefits — cooling, health, biodiversity, social contact — are often unevenly shared. Accessibility, ecological quality and connectivity are usually assessed separately. The two study sites contrast sharply: Yuexiu is the dense historic core of Guangzhou, a city that grew its greenway network through the Pearl River Delta programme, while Delft integrates small-scale blue-green infrastructure into compact Dutch urban planning.

Urban green space in Yuexiu (jiedao, left) and Delft (wijken, right).

Objective

Answer the question: how can urban green space quality assessment inform nature-based interventions in Delft and Yuexiu? Five sub-questions cover accessibility, typology and ecological quality, equity of distribution, connectivity, and prioritisation of interventions.

Method

  1. 01

    Accessibility (SQ1): green space per capita from WorldPop, population share within 300 m and 500 m buffers, distance to nearest green space, and access points where the road network meets green boundaries

  2. 02

    Ecological quality (SQ2): OSM/BGT typology classification, Sentinel-2 NDVI zonal statistics, GBIF observation density per hectare, and a blue-green balance index

  3. 03

    Spatial justice (SQ3): Lorenz curve and Gini coefficient of green space per capita, a 3×3 bivariate choropleth of green vs population density, and correlation with CBS income (Delft) or VIIRS night-time light (Yuexiu)

  4. 04

    Connectivity (SQ4): fragmentation metrics (NP, MPS, ENN) and a patch graph with edges at a 150 m dispersal threshold, analysed for degree, betweenness and connected components

  5. 05

    S-MCDA (SQ5): min-max normalised sub-scores weighted 30% accessibility, 25% ecology, 25% connectivity and 20% equity, classified into priority tiers and overlaid with proposed green corridors for isolated patches

  6. 06

    Engineering: eight sequential R scripts with every path, CRS and parameter in 00_config.R, rendered with Quarto and published on GitHub Pages

Analytical pipeline: five sub-questions integrated through a weighted MCDA synthesis.

Data

  • OpenStreetMap (green space, roads, water) and BGT for Delft
  • WorldPop 100 m population grids
  • Sentinel-2 NDVI and ESA WorldCover
  • GBIF species occurrence records
  • CBS neighbourhood income (Delft) and VIIRS night-time lights (Yuexiu)

Analysis

  • Yuexiu is about 5× denser than Delft (≈25,000 vs 4,856 residents/km²), so its low green space per capita (≤10 m² vs up to 400 m² in Delft) reflects more people sharing green space, not only less green space
  • Almost all of Delft lives within 300 m of green space; Yuexiu's coverage is clearly lower at 300 m and improves considerably at 500 m
  • Green space is distributed more equally in Delft (Gini 0.598) than in Yuexiu (Gini 0.742); income/night-light correlations are near zero in both cities (r = −0.082 and −0.116)
  • At a 150 m threshold, Delft forms a dense network (37,936 edges across 4,039 patches), while Yuexiu has only 42 edges across 114 patches, split into 83 isolated components
  • Over 30% of Yuexiu's patches exceed 1 ha, against less than 10% in Delft: a few large, isolated parks versus many small, close-knit patches
Green space per capita by administrative unit (pseudo-log scale).
Lorenz curves of green space per capita: Delft (Gini 0.598) vs Yuexiu (Gini 0.742).
Green patch network at a 150 m dispersal threshold.
Landscape fragmentation metrics: patch count, mean patch size, nearest-neighbour distance and share of patches over 1 ha.

Results

  • Delft performs better on every dimension: higher accessibility, more consistent vegetation, lower inequality and a denser green network
  • Both cities' historic cores score highest in the MCDA, with Delft's Binnenstad the strongest priority in the dataset (0.95) and Kuangquan leading in Yuexiu (0.83)
  • Connectivity is the most urgent criterion in both cities (mean urgency ≈ 0.81–0.82)
  • Yuexiu needs far more corridor interventions, such as stepping stones, green roofs and street trees, to link its isolated parks, while Delft can reinforce an already connected network with small interventions
  • The pipeline relies on open, global datasets and can be transferred to other cities by editing a single config file
MCDA composite urgency score per subdistrict/wijk. Darker red indicates higher priority for intervention.
City-mean MCDA sub-scores. Connectivity carries the highest relative urgency in both cities.
Proposed green corridors linking isolated patches, overlaid on high-priority areas.

Reflection

  • OSM completeness is lower in Chinese cities than in Western Europe, which can bias typology and connectivity metrics for Yuexiu
  • The socio-economic proxies differ (CBS income vs VIIRS night-light), so equity correlations are exploratory rather than directly comparable
  • GBIF density reflects recording effort as much as biodiversity, and WorldPop placeholder values had to be filtered (pop ≥ 100) to avoid spurious priorities
  • MCDA weights were set a priori and priority tiers are relative rankings, so the next steps are sensitivity analysis, network-based walking access and validation with residents