Examining the Heterogeneity of Geographical and Social Equity of Urban Green Space Exposure at Overhead and Eye Levels

Highlights

	· Develops an urban green space exposure assessment framework, including indicators at both overhead and eye levels
	· Evaluates the social equity of urban green space exposure using four spatial regression models
	· Identifies spatial correlations between housing price and green space exposure indicators at overhead and eye levels
	· Proposes improvements for areas with varying green space exposure levels based on spatial regression results

Abstract

Enhancing green space exposure is a crucial strategy for proactively intervening in public health from an upstream perspective. However, the distribution of green spaces in urban areas is often uneven, leading to issues such as "green inequity." This study aims to systematically assess the level of green space exposure at overhead and eye levels, analyze the geographical and social equity of green space exposure, and propose planning and regulatory strategies. Focusing on Nanjing as the study area, the research team first constructed a green space exposure assessment system based on the composition and configuration of urban green spaces at the overhead level, and the quantity and perceived quality of street green space at the eye level, assessing the geographical equity of green space exposure. Next, by selecting housing price as a socio-economic indicator, the research used various spatial regression models to analyze the spatial correlation between green space exposure and housing price, evaluating the social equity of green space exposure. The research finds 1) significant imbalances in both the geographical and social equity of green space exposure within the study area; 2) the spatial correlation between eye-level green space exposure indicators and housing price ranges from 0.08 to 0.29, generally higher than that at overhead level (ranging from 0.02 to 0.13); 3) significant heterogeneity in the spatial correlation between green space exposure and housing price, with people in higher-priced housing being more likely to benefit from green space services. The results can accurately identify blind spots in green space exposure and imbalance areas between green space supply and socioeconomic status, providing guidance for "scientific greening, " and further promoting empirical studies in Exposure Ecology.

Graphical abstract

Keywords

Exposure Ecology / Green Space Exposure / Urban Green Spaces / Geographical Equity / Social Equity / Spatial Regression Model

1 Introduction

Green spaces, as critical urban ecological features, are essential for constructing resilient cities and improving the quality of life^[1][2]. These spaces not only provide extensive ecosystem services but also offer vital venues for residents to connect with and experience nature^[3][4]. However, in urban contexts, particularly in high-density areas, green spaces are scarce and often unevenly distributed, leading to a series of equity-related challenges^[5][6].

Green space exposure (GSE) refers to direct or indirect access and interaction between residents and urban green spaces^[7]. Ensuring sufficient exposure to urban green spaces is a prerequisite for promoting health benefits^[8]. Over the years, both domestic and international scholars have explored the characteristics of urban GSE, including its spatial patterns and health effect measurements^[9][10]. Current GSE assessments are predominantly conducted at the overhead level, utilizing indicators such as area of green space, normalized difference vegetation index, percentage of tree canopy, and percentage of vegetation^[11]~[13]. Some studies also evaluate the configuration of green spaces, e.g., fragmentation, shape complexity, and patch cohesion^[14][15]. However, such studies primarily rely on satellite imagery to reveal large-scale spatiotemporal GSE patterns, omitting the vertical dimension of exposure or the actual spatiotemporal behaviors of residents in urban green spaces. Consequently, these studies fail to reflect the inherent complexity, diversity, and dynamism of GSE. With the advancement of big data and machine learning technologies, scholars have begun to preliminarily explore eye-level GSE metrics, such as using street view images to measure the green view index and quantify the quality of street spaces^[16][17]. Although there has been a shift from single-dimensional measurements to multi-dimensional GSE evaluations^[9][18][19], comprehensive urban GSE assessment models remain scarce. Bin Chen et al. have emphasized the importance of integrating green space quantity, quality, type, and structure attributes in quantitative studies and advocated for GSE measurements that consider spatial, temporal, and social disparities^[20]. Theoretically, Zhaowu Yu et al. proposed the framework of Exposure Ecology to systematically understand the nexus of (urban) natural ecosystems, ecological exposure, and health^[21]. Geographical and social equity in urban GSE is a critical study component of this framework, focusing on the "object–reality" dimension.

Green space equity research has evolved from quantity equity to spatial distribution equity, and to social equity^[22]. Early studies predominantly employed conventional green space coverage indicators (e.g., ratio of green space, per capita green area) to assess the service capacity of green spaces in a given administrative area. While these metrics reveal the supply levels, they often fail to reflect the demand–supply relationship between populations and green spaces^[20][23]. Later studies incorporating population distribution and socio-economic attributes into GSE equity analyses have gained prominence^[24][25]. Evidence suggests that GSE equity varies across different geographic regions and socio-cultural contexts^[26]. Inequities in GSE potentially exacerbate health disparities, particularly among socio-economically disadvantaged groups^[27]. The United Nations' Sustainable Development Goals proposes to "provide universal access to safe, inclusive and accessible, green and public spaces, in particular for women and children, older persons and persons with disabilities"^[28]. Systematically assessing urban GSE, identifying blind spots, understanding geographical and social disparities, and proposing planning interventions are pivotal for achieving green space growth scientifically and for upstream public health interventions^[29]. Despite growing interest, existing research lacks comprehensive evaluations of both geographical and social equity of GSE in high-density urban areas, limiting the ability to propose targeted optimization strategies for diverse urban contexts.

Building on the framework of Exposure Ecology, this study aims to construct an assessment system for urban GSE at both overhead and eye levels. It seeks to systematically evaluate the geographical and social equity of urban GSE and address the following questions: 1) What are the spatial distribution characteristics of urban GSE indicators at overhead and eye levels? 2) Is there any spatial correlation and heterogeneity between GSE indicators and housing price?

2 Research Methods

2.1 Study Area

This study focuses on the central urban area of Nanjing City, China (Fig.1). By the end of 2022, Nanjing had a total population of approximately 9.5 million, with 8.26 million urban population^[30]. According to the Master plan of Nanjing Territorial Spatial Planning (2021–2035), the central urban area encompasses four centers: Xinjiekou, Hexi, Chengnan, and Jiangbei, covering a total planned area of 804 km². The study area has a prosperous economy and plays a critical role in culture and history as it covers the cultural core of ancient capital. It also features the highest population density and the most intricate road network in the city. This research divided the study area into 500 m × 500 m fishnet grids using ArcGIS 10.3, with each grid as a sample unit.

Fig.1 Location of the study area.

Full size|PPT slide

2.2 Construction of the Urban GSE Assessment Model

This research developed an urban GSE assessment system at both overhead and eye levels. The overhead-level indicators include two categories: green space composition and configuration (Tab.1). The eye-level indicators consist of two categories: quantity and perceived quality (Tab.2).

Tab.1 Overhead-level GSE indicators

Category	Indicator	Description
Composition	Normalized difference	· A reflection of vegetation density and health of the ground vegetation
	vegetation index (NDVI)	· Higher values indicate higher vegetation density and healthier condition
	Tree coverage rate	Proportion of tree area in the grid
	Grassland coverage rate	Proportion of grassland area in the grid
	Cropland coverage rate	Proportion of cropland area in the grid
	Percentage of landscape (PLAND)	The ratio of the total area of green space patches to the total area in the grid
Configuration	Largest patch index (LPI)	· The ratio of the area of the largest green space patch to the area of the grid
		· Values closer to 0 indicate smaller patches
	Number of patches (NP)	· Number of green space patches in the grid
		· Higher values indicate greater fragmentation
	Edge density (ED)	· The ratio of the total edge length of all green space patches to the area of the grid
		· Higher values indicate greater fragmentation
	Shape index (SHAPE)	· Average shape index (ratio of the patch perimeter to the circumference of a circle with the same area) of all green space patches in the grid
		· Values closer to 0 indicate simpler shapes
	Fractal dimension index (FRAC)	· Average edge complexity of all green space patches in the grid
		· Values closer to 0 indicate simpler shapes
	Patch cohesion index (COHESION)	· Degree of aggregation of green space patches
		· Values closer to 0 indicate simpler shapes

Tab.2 Eye-level GSE indicators

Category	Indicator	Description
Quantity	Green view index (GVI)	Percentage of vegetation in a person's field of view
Perceived quality	Vegetation abundance	Diversity of plant species in street green spaces
	Walkability	Degree to which the street environment supports walking activities
	Accessibility	Ease with which people can reach and use street green spaces
	Amenity	Convenience of facilities and services provided in street green spaces
	Openness	Connectivity and openness of the street network
	Neatness	Cleanliness of green spaces
	Safety	Safety conditions of street green spaces, including objective factors (e.g., crime rate, nighttime lighting, and emergency facilities) and subjective perceptions of the street atmosphere

2.2.1 Selection and Data Acquisition of Overhead-Level GSE Indicators

This research first downloaded Sentinel-2 satellite imagery (10 m resolution) of SeptembeR₂021 via the Copernicus Open Access Hub. The NDVI for the study area was calculated using the Sentinel Application Platform (SNAP) and ArcGIS 10.3. Next, the data of green space composition indicators were derived from the European Space Agency's WorldCover v100 land cover dataset (10 m resolution) and calculated with Fragstats 4.2, identifying eight land cover types: tree, shrub, grassland, cropland, built-up land, bare/sparsely vegetated land, water body, and herbaceous wetland. Given the minimal area proportions of shrubland (0.001%) and herbaceous wetland (0.013%), only the coverage rates of tree, grassland, and cropland were included in the GSE composition indicators. The landscape pattern indices of tree, grassland, and cropland as a whole were calculated using Fragstats 4.2 as configuration dimension indicators.

2.2.2 Selection and Data Acquisition of Eye-level GSE Indicators

First, this research employed ArcGIS to establish 79, 777 observation points at 200-meter intervals along the first class, second class, and third class roads. Street view images from Baidu Maps were adopted to capture images in four cardinal directions (0°, 90°, 180°, and 270°) at each observation point, resulting in a total of 319, 108 images (640 × 480 pixels). Next, a fully convolutional network model (FCN-8s) was trained in Python 3.7 based on the ADE20K dataset, which includes 150 object categories (e.g., trees, buildings, cars). The images were semantically segmented to identify their compositional elements. Pixels corresponding to different semantics were filled with specific colors, and the percentage of pixels representing vegetation was calculated to determine the GVI:

(1)$GVI=\frac{\sum _{i=1}^{m}Are{a}_{\mathrm{g}}}{\sum _{i=1}^{m}Are{a}_{\mathrm{t}}}\times 100\mathrm{\%},$

where Area_t represents the total number of pixels in each street view image, Area_g is the number of pixels occupied by vegetation, and m is the number of images captured at each observation point (4 in this research).

To evaluate people's perceived quality of the street view images, a human-machine adversarial scoring method was employed. The research team randomly selected 5, 000 street-view images from the database, covering various landscape features, to construct a training dataset (Fig.2). Forty volunteers (1∶1 male-to-female ratio) were randomly recruited from university researchers specializing in landscape architecture, architecture, and urban planning, as well as from local shopkeepers familiar with the area. Volunteers rated seven perceived quality indicators of the images online, on a scale of 0 (extremely low quality) to 100 (extremely high quality). A Python-based random forest model was used to train the ratings and the proportions of different landscape elements, enabling automated scoring of the entire street view image dataset. The values of the GSE indicators for each grid were calculated as the average score of all corresponding images within the grid. To avoid bias caused by insufficient street view images, grids with fewer than five observation points (i.e., fewer than 20 images) were excluded. Ultimately, 2, 750 grids were included as valid samples for the analysis.

Fig.2 Examples of semantic segmentation results for street view images.

Full size|PPT slide

2.3 Socio-Economic Indicators

Studies have shown that population density, per capita GDP, and housing price are closely linked to socio-economic conditions^[31]~[33]. Housing price, as an important indicator refecting socio-economic status, has shown reliability and validity in empirical research. For instance, Jinguang Zhang employed housing price as a proxy for household income to analyze spatial and social inequalities in GSE accessibility, availability, and attractiveness in central urban area of Nanjing^[34]. Siqi Yu et al. revealed spatial disparities in urban park accessibility among different socio-economic groups in the main urban districts of Nanjing based on housing prices^[35]. Accordingly, this research uses housing price as the primary socioecnomic indicator in spatial analysis, with population density and per capita GDP included as covariates. Population density was obtained from the 2020 WorldPop database (100 m resolution). Per capita GDP was calculated by dividing 2020 district-level GDP by the registered population, using data from the Statistics Bureau of Nanjing Municipality. Housing price data were collected from online property listing and rental platform—Lianjia. First, ArcGIS 10.3 was adopted to segment and calculate the average population density for each grid; second, the per capita GDP of the administrative district in which each grid is located was assigned to that grid; finally, the average housing price of all residential properties within each grid was calculated.

2.4 Data Analysis

In the ArcGIS 10.3 platform, the natural breaks classification method was used to divide the GSE indicator values, observed at both overhead and eye levels, into seven levels of visualization, aiming to assess the geographical equity of GSE in the study area. For social equity analysis, univariate Moran's I was first used to examine the spatial autocorrelation of housing prices, yielding a value of 0.924 (p = 0.001), indicating a strong and positive spatial autocorrelation in housing price across the study area. Subsequently, a bivariate local Moran's I test and spatial correlation analysis were conducted separately for housing price and GSE indicators, accompanied by visualization. Four regression models were then employed to analyze the relationships between housing price and GSE indicators, including ordinary least squares (OLS) model, spatial lag model (SLM), spatial error model (SEM), and geographically weighted regression (GWR). Population density and per capita GDP were used as covariates. The OLS model is expressed as follows:

(2)$H{P}_i={\alpha }_0+{\alpha }_{1}G{S}_i+{\alpha }_{2}P{O}_i+{\alpha }_{3}GD{P}_i+{\epsilon }_0,$

where HP_i, GS_i, PO_i, and GDP_i represent housing price, GSE indicators, population density, and per capita GDP in the i-th grid, respectively; α₀ is the constant term, and ε₀ is the error term.

A spatial autocorrelation test was conducted on the residuals of the OLS model for each GSE indicator. The results showed that all residuals had significantly positive Moran's I values (e.g., the Moran's I for the residuals of the OLS model with NDVI as the independent variable was 0.878, p = 0.001). This indicates that the current OLS model failed to adequately capture the spatial structure in the data. Therefore, based on Eq. (2), spatial regression models (SLM and SEM) were constructed with distance weighting as follows:

(3)$H{P}_i={\alpha }_0+\gamma W_i+{\alpha }_{1}G{S}_i+{\alpha }_{2}P{O}_i+{\alpha }_{3}GD{P}_i+{\epsilon }_0,$

where γW_i represents the spatial lag of neighboring regions in SLM and the spatial error in SEM. The spatial weights were constructed using a minimum threshold distance of 864 m, ensuring that each grid had at least one neighboring grid. To test the robustness of the results, sensitivity analysis was performed with a threshold of 1, 000 m and weights constructed based on Queen's case and Rook's case adjacency criteria.

Although SLM and SEM account for spatial autocorrelation, they capture only global spatial relationships in the study area. In contrast, the GWR model provides local estimates of regression coefficients for each grid, revealing spatial heterogeneity. Using the spgwr package in RStudio, cross-validation determined an optimal bandwidth of approximately 485 m for the GWR analysis. The GWR model is expressed as follows:

(4)$\begin{array}{rl}H{P}_i=& {\alpha }_0\left(u_i,v_i\right)+{\alpha }_1\left(u_i,v_i\right)G{S}_i+{\alpha }_2\left(u_i,v_i\right)P{O}_i\\ & +{\alpha }_3\left(u_i,v_i\right)GD{P}_i+{\epsilon }_0\end{array}$

where HP_i, GS_i, PO_i, GDP_i, and ε₀ have the same definitions as in Eq. (2) and Eq. (3). α₀(μ_i, ν_i) represents the intercept for the i-th grid, and α₁(μ_i, ν_i), α₂(μ_i, ν_i), and α₃(μ_i, ν_i) are the location-specific regression coefficients, which vary with the geographic coordinates (μ_i, ν_i).

The models were then evaluated for goodness of fit using parameters such as the coefficient of determination (R₂) and Akaike information criterion (AIC). The GWR model produced spatially varying regression coefficients, which were visualized to reveal the spatial variation of each GSE indicator across the study area. This analysis is instrumental in identifying areas and GSE indicators significantly correlated with housing price, uncovering spatial social heterogeneity in GSE and providing valuable insights for developing scientific and refined grid-based intervention strategies to address blind spots of GSE.

3 Results and Discussion

3.1 Spatial Distribution Characteristics of GSE Indicators

At the overhead level, GSE composition indicators revealed uneven distributions (Fig.3). NDVI, tree coverage rate, and PLAND showed similar spatial patterns, with lower values in densely built-up central area and higher values near mountains; while peripheral areas generally exhibited higher values than the central area, indicating higher levels of GSE in these areas. Grassland and cropland coverage rate displayed similar spatial patterns, with lower values in the central area and the areas near mountains, and relatively higher values in peripheral areas. For GSE configuration indicators, NP and LPI exhibited contrasting distributions: green spaces in the central area were smaller but more numerous, whereas those in the northeastern and peripheral areas were larger but fewer. ED, SHAPE, FRAC, and COHESION displayed relatively balanced distributions, suggesting that green spaces in the study area had regular shapes and high aggregation.

Fig.3 The spatial distribution of GSE indicators at the overhead level.

Full size|PPT slide

At the eye level, the quantity and perceived quality also exhibited uneven spatial distributions (Fig.4). GVI, vegetation abundance, and openness showed spatial patterns similar to NDVI, with higher values in peripheral areas compared with densely built-up central areas, reflecting better greenery and more open streetscapes. Walkability, accessibility, amenity, neatness, and safety exhibited similar patterns, with generally high values, indicating sound overall quality of street environments in the study area.

Fig.4 The spatial distribution of GSE indicators at the eye level.

Full size|PPT slide

3.2 Spatial Correlation Between GSE Indicators and Housing Price

In the study area, the housing prices were the highest in the cultural core of ancient capital and Hexi Area, gradually decreasing outwards (Fig.5). The bivariate local Moran's I results (Fig.6) showed that the larger the absolute value of the Moran's I, the stronger the spatial correlation between GSE indicators and housing price. Results indicated significant spatial correlations between housing price and all 19 GSE indicators. Notably, grassland coverage rate, cropland coverage rate, NP, and vegetation abundance showed negative spatial correlations with housing price. This suggests that better green space conditions (e.g., dominated by trees with high quantity and quality), larger area, greater shape diversity, and higher cohesion were associated with higher housing price. Furthermore, the absolute value of Moran's I for housing price and overhead GSE indicators were generally lower than those for housing price and eye-level GSE indicators. Among them, tree coverage rate and cropland coverage rate had relatively higher absolute Moran's I of 0.129 and 0.171, respectively. For eye-level indicators, amenity, neatness, and accessibility showed higher absolute Moran's I of 0.292, 0.254, and 0.232, respectively. These findings suggest that housing price is closely associated with both quantity and perceived quality, aligning with the results of previous studies exploring the relationship between urban green space and housing price^[36][37].

Fig.5 The spatial distribution of housing price in study area.

Full size|PPT slide

Fig.6 The spatial correlation between housing price and different GSE indicators.

Full size|PPT slide

Tab.3 presents the regression analysis results of the four models for different GSE indicators at both overhead and eye levels. Overall, the spatial regression models (SLM, SEM, and GWR) greatly outperformed the OLS model in terms of goodness of fit, with SLM and SEM achieving the best results, followed by GWR. Sensitivity analysis using different weighting schemes confirmed the robustness of the findings, as SLM and SEM results were consistent with those reported in Tab.3, with R₂ values exceeding 0.9. It validated the robustness of the findings and reaffirmed the significant spatial correlation between GSE indicators and housing price.

Tab.3 Fitting results of the housing price with GSE indicators at the overhead and eye levels

Variable	OLS				SLM				SEM				GWR
	R2	AIC	Log likelihood		R2	AIC	Log likelihood		R2	AIC	Log likelihood		R2	AIC
Overhead level
NDVI	0.131	−3, 948.410	1, 978.210		0.939	−10, 480.800	5, 245.420		0.939	−10, 477.500	5, 242.766		0.903	−9, 688.662
Tree coverage rate	0.136	−3, 964.150	1, 986.070		0.939	−10, 481.700	5, 245.830		0.939	−10, 477.000	5, 242.495		0.903	−9, 696.385
Grassland coverage rate	0.107	−3, 872.520	1, 940.260		0.938	−10, 457.700	5, 233.830		0.938	−10, 459.700	5, 233.833		0.905	−9, 735.061
Cropland coverage rate	0.111	−3, 885.630	1, 946.820		0.938	−10, 458.200	5, 234.100		0.939	−10, 461.800	5, 234.890		0.904	−9, 705.012
PLAND	0.135	−3, 960.390	1, 984.200		0.939	−10, 492.000	5, 251.000		0.939	−10, 485.200	5, 246.622		0.904	−9, 699.222
NP	0.139	−3, 973.290	1, 990.840		0.940	−10, 524.300	5, 267.600		0.939	−10, 506.100	5, 257.050		0.905	−9, 742.258
LPI	0.136	−3, 965.890	1, 986.950		0.939	−10, 498.000	5, 254.010		0.939	−10490.000	5, 248.996		0.905	−9, 716.672
ED	0.108	−3, 876.800	1, 942.400		0.938	−10, 459.800	5, 234.920		0.939	−10, 462.700	5, 235.339		0.901	−9, 641.370
SHAPE	0.129	−3, 942.570	1, 975.280		0.939	−10, 495.000	5, 252.510		0.939	−10, 476.300	5, 242.135		0.903	−9, 960.407
FRAC	0.111	−3, 886.590	1, 947.290		0.939	−10, 465.600	5, 237.820		0.939	−10, 462.200	5, 235.102		0.898	−9, 528.610
COHESION	0.130	−3, 946.140	1, 977.070		0.939	−10, 512.700	5, 261.350		0.939	−10, 501.800	5, 254.885		0.604	−6, 100.009
Eye level
GVI	0.125	−3, 930.530	1, 969.260		0.939	−10, 471.400	5, 240.720		0.939	−10, 469.900	5, 238.950		0.902	−9, 655.261
Vegetation abundance	0.107	−3, 873.060	1, 940.530		0.939	−10, 465.800	5, 237.910		0.939	−10, 474.500	5, 241.269		0.862	−8, 877.027
Walkability	0.126	−3, 931.390	1, 969.690		0.938	−10, 458.300	5, 234.140		0.938	−10, 459.400	5, 233.694		0.895	−9, 514.691
Accessibility	0.153	−4, 016.13	2, 012.060		0.938	−10, 457.100	5, 233.540		0.938	−10, 454.600	5, 231.292		0.881	−9, 235.790
Amenity	0.173	−4, 083.690	2, 045.850		0.938	−10, 460.200	5, 235.110		0.938	−10, 454.800	5, 231.394		0.889	−9, 090.944
Openness	0.109	−3, 836.550	1, 922.280		0.938	−10, 354.400	5, 182.220		0.938	−10, 357.200	5, 182.623		0.864	−8, 838.506
Neatness	0.162	−4, 049.470	2, 028.740		0.938	−10, 464.700	5, 237.370		0.938	−10, 459.700	5, 233.835		0.846	−8, 614.218
Safety	0.152	−3, 971.400	1, 989.700		0.938	−10, 359.600	5, 184.820		0.938	−10, 357.300	5, 182.643		0.872	−9, 062.314

NOTEThe R2 of the GWR model for all GSE indicators in this table is the global mean value and there is no log likelihood term.

3.3 Spatial Association and Heterogeneity Between GSE Indicators and Housing Price

The coefficient estimation results (Fig.7, Fig.8) showed that the estimated coefficients for all the GSE indicators in the study area simultaneously had both positive and negative values, with a notable proportion of negative values, suggesting significant social equity heterogeneity of GSE. According to the distribution of housing price (Fig.5), high-price areas exhibited positive coefficients for NDVI, tree coverage rate, PLAND, LPI, ED, and SHAPE at the overhead level, indicating that in these areas, as vegetation coverage increases, green space size grows, the diversity of green space shapes increases, and the connectivity with the surrounding environment improves, housing price tends to rise. Similarly, in high-price areas, the estimated positive coefficients for GVI, walkability, accessibility, amenity, neatness, and safety at the eye level indicated that as quantity and the quality of the street green space improve, housing price also increases. Conversely, in low-price areas, the coefficients for these indicators were estimated as negative, implying that GSE and housing price were negatively correlated. This result may be influenced by the geographical locations of these areas, which are typically on the periphery of the central city, where green space conditions are relatively good. Therefore, rather than an increase in GSE, urbanization levels may play a more significant role in raising housing prices in these areas. The relationship between the quantity and quality of street green spaces and housing price has also been confirmed in previous studies. Zhaocheng Bai et al. revealed a positive correlation between quantity of street green spaces and housing price in the central urban area of Hangzhou, China^[38]. Ying Li et al. found that street quality characteristics could influence housing price by as much as 36.74% in Guangzhou, China^[39]. Furthermore, in most high-price areas in this research, the coefficients for grassland coverage rate, cropland coverage rate, NP, vegetation abundance, and openness were negative, indicating that subareas with lower building density and urbanization levels tend to have relatively lower housing prices in high-price areas.

Fig.7 Coefficient estimations of housing price and GSE indicators at the overhead level.

Full size|PPT slide

Fig.8 Coefficient estimations of housing price and GSE indicators at the eye level.

Full size|PPT slide

3.4 Planning Intervention Strategies for GSE

Based on the above analysis, this research found that GSE indicators at the eye level exhibit a stronger spatial correlation with socio-economic status than those at the overhead level. Therefore, planning strategies aimed at promoting "green equity" should prioritize enhancing the quantity and quality of street green spaces. At the micro scale, this research identified low-GSE zones requiring optimization by analyzing the heterogeneity of geographical and social equity of GSE indicators, and proposed targeted optimization strategies for high and low socio-economic areas^① within these zones.

① High/low socio-economic status zones were defined based on the spatial distribution of housing price shown in Fig.4, where color yellow represents the intermediate level, deeper blue indicates lower socio-economic status, and deeper red indicates higher levels.

1) Low-GSE, high socio-economic status zones. In these zones, GSE and socio-economic status are negatively correlated, typically characterized by high housing prices but limited green spaces with small, regular shape, and poor overall vegetation coverage. Such sites are often found in well-developed areas with relatively high living standards. The scarcity of available land makes it challenging to increase the amount of green spaces. Thus, structural adjustments to existing ones, such as improving shapes of green spaces to enhance residents' access, could be considered. The more complex the green patch shapes, the higher the coupling between green space boundaries and the surrounding environment, which can provide more accesses for residents to interact with green spaces. Additionally, green corridors can be created along linear spaces like roads and rivers to connect fragmented green spaces and improve overall GSE levels in urban areas.

2) Low-GSE, low socio-economic status zones. In these zones, GSE and socio-economic status are positively correlated. Such zones are typically located on the periphery of the study area, with lower housing price and urban vitality, and also have higher proportions of grassland and cropland compared with tree coverage. As indicated earlier, grassland and cropland coverage rates exhibit negative spatial correlations with socio-economic status, while tree coverage rate exhibits a positive correlation. Therefore, the focus of increasing green spaces in such areas should be on encouraging tree planting. The addition of trees and shrubs to grassland areas could enhance the visual layering of the landscape while improving the quality of GSE.

4 Conclusions and Prospects

This research developed a comprehensive urban GSE assessment model at both overhead and eye levels. Using housing price as a proxy for residents' socio-economic status, this research adopted OLS, SLM, and SEM models to evaluate the geographical and social equity of GSE in Nanjing's central urban area. To conduct a more detailed spatial heterogeneity analysis of GSE's social equity, GWR model was employed to visualize grid-level spatial correlations between GSE and housing prices.

The results revealed an uneven distribution of GSE across the study area. Green spaces in the central area were fragmented, and overall GSE was lower than peripheral areas. A significant spatial correlation was observed between housing price and GSE indicators, with eye-level GSE indicators exhibiting stronger correlations with housing price than overhead-level indicators. Grassland coverage rate, cropland coverage rate, NP, and vegetation abundance showed negative correlations with housing price, while all other GSE indicators displayed positive correlations. Additionally, SLM and SEM models demonstrated the best predictive performance for housing price based on GSE indicators. The GWR model further showed significant imbalances in both geographical and social equity of GSE within the central urban area of Nanjing.

This research identified urban areas with insufficient GSE and proposed targeted planning interventions separately for low-GSE, high socio-economic status zones, and low-GSE, low socioeconomic status zones. Theoretically, the research extends the empirical application of Exposure Ecology studies. Practically, it provides actionable insights for advancing urban GSE assessments and offers guidance for promoting equity in GSE through urban planning and green space system design.

This study has several limitations. First, differences in the number, profession, and age distribution of volunteers scoring street-view image training sets might introduce bias in perceived quality evaluations. Future studies should control these factors to improve robustness of the results. Second, using community housing price as a proxy for socio-economic status has certain constrains. Future research could incorporate additional indicators, such as average household income and education levels, for a more comprehensive assessment. Third, the analysis results based on cross-sectional data cannot reveal causal relationships between GSE and housing price. Future research can use long time-series data to fill this gap. Lastly, the sample selection from high-density urban areas may limit the generalizability of the planning strategies. Future research should explore the heterogeneity of GSE equity in a multi-scale, multitemporal, and cross-regional context.

References

[1] Yu, Z. , Yang, G. , Zuo, S. , Jørgensen, G. , Koga, M. , & Vejre, H. (2020) Critical review on the cooling effect of urban blue-green space: A threshold-size perspective. Urban Forestry & Urban Greening, ( 49), 126630.

[2] Yao, X. , Yu, Z. , Ma, W. , Xiong, J. , & Yang, G. (2024) Quantifying threshold effects of physiological health benefits in greenspace exposure. Landscape and Urban Planning, ( 241), 104917.

[3] Li, M. , Wen, Y. , & Hu, G. (2024) Research status of urban greenspace exposure and public health relation. Journal of Zhejiang Sci-Tech University (Natural Sciences), 51 ( 4), 492– 506.

[4] Lin, T. , Zeng, Z. , Yao, X. , Geng, H. , Yu, Z. , Wang, L. , Lin, M. , Zhang, J. , & Zheng, Y. (2023) Research of urban green space exposure and its effects on human health. Acta Ecologica Sinica, 43 ( 23), 10013– 10021.

[5] Wu, J. , Si, M. , & Li, W. (2016) Spatial equity analysis of urban green space from the perspective of balance between supply and demand: A case study of Futian District, Shenzhen, China. Chinese Journal of Applied Ecology, 27 ( 9), 2831– 2838.

[6] Heo, S. , & Bell, M. L. (2023) Investigation on urban greenspace in relation to sociodemographic factors and health inequity based on different greenspace metrics in 3 US urban communities. Journal of Exposure Science & Environmental Epidemiology, 33 ( 2), 218– 228.

[7] Markevych, I. , Schoierer, J. , Hartig, T. , Chudnovsky, A. , Hystad, P. , Dzhambov, A. M. , de Vries, S. , Triguero-Mas, M. , Brauer, M. , Nieuwenhuijsen, M. J. , Lupp, G. , Richardson, E. A. , Astell-Burt, T. , Dimitrova, D. , Feng, X. , Sadeh, M. , Standl, M. , Heinrich, J. , & Fuertes, E. (2017) Exploring pathways linking greenspace to health: Theoretical and methodological guidance. Environmental Research, ( 158), 301– 317.

[8] Zhang, J. , Yu, Z. , Zhao, B. , Sun, R. , & Vejre, H. (2020) Links between green space and public health: A bibliometric review of global research trends and future prospects from 1901 to 2019. Environmental Research Letters, 15 ( 6), 063001.

[9] Wheeler, B. W. , Lovell, R. , Higgins, S. L. , White, M. P. , Alcock, I. , Osborne, N. J. , Husk, K. , Sabel, C. E. , & Depledge, M. H. (2015) Beyond greenspace: An ecological study of population general health and indicators of natural environment type and quality. International Journal of Health Geographics, ( 14), 1– 17.

[10] Yang, G. , Yu, Z. , Zhang, J. , Liu, H. , Jin, G. , Ju, Y. , Hong, B. , Zhao, H. , Zhang, L. , Yao, X. , Ma, W. , Xiong, J. , Shao, Y. , & Jiang, B. (2024) Research progress and prospects on the health benefits of greenspace exposure from the perspective of exposure ecology. Acta Ecologica Sinica, 44 ( 14), 5914– 5924.

[11] Astell-Burt, T. , Walsan, R. , Davis, W. , & Feng, X. (2023) What types of green space disrupt a lonelygenic environment? A cohort study. Social Psychiatry and Psychiatric Epidemiology, 58 ( 5), 745– 755.

[12] Zhang, J. , Song, A. , Xia, T. , & Zhao, B. (2023) Evaluating the urban park green space exposure from the perspective of the community life circle. Journal of Nanjing Forestry University (Natural Sciences Edition), 47 ( 3), 191– 198.

[13] Gonzales-Inca, C. , Pentti, J. , Stenholm, S. , Suominen, S. , Vahtera, J. , & Käyhkö, N. (2022) Residential greenness and risks of depression: Longitudinal associations with different greenness indicators and spatial scales in a Finnish population cohort. Health & Place, ( 74), 102760.

[14] Grafius, D. R. , Corstanje, R. , & Harris, J. A. (2018) Linking ecosystem services, urban form and green space configuration using multivariate landscape metric analysis. Landscape Ecology, ( 33), 557– 573.

[15] Sheng, S. , & Wang, Y. (2024) Configuration characteristics of greenblue spaces for efficient cooling in urban environments. Sustainable Cities and Society, ( 100), 105040.

[16] Wang, R. , Feng, Z. , Pearce, J. , Zhou, S. , Zhang, L. , & Liu, Y. (2021) Dynamic greenspace exposure and residents' mental health in Guangzhou, China: From over-head to eye-level perspective, from quantity to quality. Landscape and Urban Planning, ( 215), 104230.

[17] Huang, W. , Wang, R. , Liu, F. , Huang, W. , Dong, G. , & Yu, H. (2022) Association between street view greenness and allergic rhinitis in children. Journal of Environmental and Occupational Medic, 39 ( 1), 17– 22.

[18] Reyes-Riveros, R. , Altamirano, A. , De La Barrera, F. , Rozas-Vásquez, D. , Vieli, L. , & Meli, P. (2021) Linking public urban green spaces and human well-being: A systematic review. Urban Forestry & Urban Greening, ( 61), 127105.

[19] Wolch, J. R. , Byrne, J. , & Newell, J. P. (2014) Urban green space, public health, and environmental justice: The challenge of making cities 'just green enough'. Landscape and Urban Planning, ( 125), 234– 244.

[20] Chen, B. , & Webster, C. (2022) Eight reflections on quantitative studies of urban green space: A mapping-monitoring-modeling-management (4M) perspective. Landscape Architecture Frontiers, 10 ( 3), 66– 76.

[21] Yu, Z. , Yang, G. , Lin, T. , Zhao, B. , Xu, Y. , Yao, X. , Ma, W. , Vejre, H. , & Jiang, B. (2024) Exposure ecology drives a unified understanding of the nexus of (urban) natural ecosystem, ecological exposure, and health. Ecosystem Health and Sustainability, ( 10), 0165.

[22] Zhou, M. , & Wang, L. (2023) Change of park green space social equity and its driving forces—From the perspective of population aging. Chinese Landscape Architecture, 39 ( 2), 57– 63.

[23] Yang, L. , Yang, P. , & Chen, L. (2020) Quantitative evaluation on the equity of park green space provision: A case study of central district of Chongqing. Chinese Landscape Architecture, 36 ( 1), 108– 112.

[24] Kabisch, N. , & Haase, D. (2014) Green justice or just green? Provision of urban green spaces in Berlin, Germany. Landscape and Urban Planning, ( 122), 129– 139.

[25] Mu, H. , Gao, Y. , Wang, Z. , & Zhang, Y. (2019) Equity evaluation of park green space service level from the perspective of supply and demand balance: An empirical analysis based on big data. Urban Development Studies, 26 ( 11), 10– 15.

[26] Yan, L. , Jin, X. , & Zhang, J. (2024) Equity in park green spaces: A bibliometric analysis and systematic literature review from 2014-2023. Frontiers in Environmental Science, ( 12), 1374973.

[27] Mitchell, R. , & Popham, F. (2008) Effect of exposure to natural environment on health inequalities: An observational population study. The Lancet, 372 ( 9650), 1655– 1660.

[28] United Nations. (n. d. ). Sustainable development goals—Goal 11: Make cities inclusive, safe, resilient and sustainable.

[29] Geng, X. , Yu, Z. , Zhang, D. , Li, C. , Yuan, Y. , & Wang, X. (2022) The influence of local background climate on the dominant factors and threshold-size of the cooling effect of urban parks. Science of the Total Environment, ( 823), 153806.

[30] Statistics Bureau of Nanjing Municipality. (2023). Nanjing 2022 national economic and social development statistical bulletin.

[31] Dong, Z. , Hui, E. C. M. , & Jia, S. (2017) How does housing price affect consumption in China: Wealth effect or substitution effect?. Cities, ( 64), 1– 8.

[32] Yegorov, Y. A. (2009) Socio-economic influences of population density. Chinese Business Review, 8 ( 7), 1– 12.

[33] Wang, T. , & Sun, F. (2022) Global gridded GDP data set consistent with the shared socioeconomic pathways. Scientific Data, 9 ( 1), 221.

[34] Zhang, J. (2023) Inequalities in the quality and proximity of green space exposure are more pronounced than in quantity aspect: Evidence from a rapidly urbanizing Chinese city. Urban Forestry & Urban Greening, ( 79), 127811.

[35] Yu, S. , Zhu, X. , & He, Q. (2020) An assessment of urban park access using house-level data in urban China: Through the lens of social equity. International Journal of Environmental Research and Public Health, 17 ( 7), 2349.

[36] Liu, Q. , Gu, D. , Liu, Y. , & Lin, X. (2021) The impact of urban landscapes on housing prices in Yuzhong District, Chongqing. Modern Urban Research, 36 ( 12), 109– 115.

[37] Cui, X. , & Fan, L. (2023) Correlation between park green space and housing price in Chengdu City based on landscape index analysis. Journal of Chinese Urban Forestry, 21 ( 3), 68– 73.

[38] Bai, Z. , Qi, J. , & Tang, X. (2023) Social differentiation of urban street green quantity: A case study through the lens of visual landscape. Journal of Tianjin Normal University (Natural Science Edition), 43 ( 1), 60– 66.

[39] Li, Y. , Wang, N. , Tong, Z. , Liu, Y. , An, R. , & Liu, Y. (2023) The nonlinear influence of street quality on housing prices based on random forest model: A case study of Guangzhou. Tropical Geography, 43 ( 8), 1547– 1562.

Acknowledgements

· Project of "A Multi-Scale Coupling Study on the Threshold Efficiency of Urban Green Spaces in Mitigating Urban Heat, " National Natural Science Foundation of China (No. 42171093) · Project of "A Study on the Spatial Allocation Levels and Mechanisms of Public Service Facilities in Urban Residential Areas Integrating Diverse Needs, " Young Scientists Fund of the National Natural Science Foundation of China (No. 42201200) · Project of "A Study on the Threshold Measurement and Influencing Factors of Physiological Health Effects of Urban Population Exposure to Green Spaces, " Young Scientists Fund of the National Natural Science Foundation of China (No. 42401103) · Project of "A Study on the Mitigation Effects of Urban Green Vegetation on Thermal Environments at Multiple Scales, " Natural Science Foundation of Shanghai (No. 21ZR1408500) · Project of "A Study on the Cooling Efficiency Threshold and Energy Dynamics Mechanism of Urban Green Spaces Under Typical Climatic Conditions, " Shanghai Pujiang Program (No. 21PJ1401600) · Project of "A Systematic Measurement and Optimal Allocation of Green Space Exposure Levels for Healthy Cities, " Natural Science Foundation of the Jiangsu Higher Education Institutions of China (No. 22KJB220006)

RIGHTS & PERMISSIONS

© Higher Education Press 2025