Introduction

Urban green spaces, as a crucial component of the urban environment, provide essential ecosystem services, including microclimate regulation, air pollution mitigation, and biodiversity conservation services that are vital to human well-being1. With the deepening of the “human-centered” urban development concept, the multidimensional impact mechanisms of green spaces on residents’ physical and mental health have become a key topic in environmental justice research2. Exposure to green spaces not only directly enhances psychological well-being but also influences mental health through mediating effects such as nature connectedness and social interactions3. Against this backdrop, urban development is undergoing a paradigm shift from “scale expansion” to “quality enhancement”4. It is imperative to acknowledge that environmental concerns transcend the realms of technical and managerial challenges; the distribution of environmental resources is inherently politically charged5. The adoption of “green” urbanism policies and practices carries the potential to amplify existing social inequalities, as the process of green gentrification has been observed to serve to reinforce existing class stratification and power imbalances within society6. Urban greening initiatives have the potential to displace vulnerable groups, thereby preventing them from benefiting from green infrastructure that is ostensibly designed to meet their needs. This dynamic is especially evident in the Global South. In recent years, the concept of the “15-minute city” has introduced a new spatial governance framework for creating livable communities7. It can alleviate the disproportionate burden of seasonal environmental pressures borne by different populations, as well as the long-term inequities in the availability of local green infrastructure. The community-based “living circle” model aligns with the “proximity principle” in urban planning8. In 2023, China’s Ministry of Commerce, along with 12 other departments, issued the Three-Year Action Plan (2023–2025) for the Comprehensive Promotion of 15-Minute Convenience Living Circles, identifying the “doorstep” (within a 5–10-minute walk) and “surrounding area” (within a 15-minute walk) as the core focus for residential planning and infrastructure development.

The rational planning of urban green spaces has thus become an essential component in constructing high-quality 15-minute communities9. However, existing urban planning strategies have predominantly focused on the static coverage of green spaces while overlooking the impacts of spatial accessibility and seasonal dynamics on residents’ actual green exposure10.

Limitations of green exposure assessment methods

The uneven distribution of urban green spaces may result in unequal green exposure among city residents within 15-minute communities10. Previous studies have shown that communities with lower levels of green exposure often experience negative physiological and psychological impacts. These impacts include reduced social interaction frequency, impaired cognitive function, and decreased well-being and life satisfaction11. These adverse effects can further contribute to social isolation within communities, exacerbating feelings of loneliness and anxiety among residents12. However, existing planning practices exhibit significant limitations. First, traditional 2D measurement methods, such as the NDVI derived from Landsat imagery, struggle to accurately represent residents’ actual perception of street greenery13. Second, most studies focus on static green space coverage while neglecting the coupled effects of spatial accessibility and seasonal dynamics, leading to a paradox between “data equality” and “experiential inequality”14. For example, in northern cities dominated by deciduous trees, the visible green volume of street vegetation can differ by up to 65% between summer and winter.

Previous research on green inequality has primarily focused on green areas within and around residential neighborhoods, typically using a planar perspective with neighborhoods as the fundamental unit of analysis14. Satellite remote sensing images are often used to assess green space exposure based on coverage extent, area size, and quantity15. NDVI is a commonly used indicator for measuring green space16. While NDVI-based methods enable straightforward comparisons across different scales and directions, they rely on a top-down aerial perspective rather than a ground-level human perspective17. As a result, 2D remote sensing images are inadequate in capturing the street layout and 3D structure of urban greenery. This oversight results in an inaccurate representation of the green exposure experienced by residents18.

Furthermore, the dynamic nature of human movement suggests that green exposure should be assessed as a dynamic process. However, previous studies have typically centered green exposure calculations on residential locations18. Spatial accessibility assessments are also subject to model biases; for instance, the traditional Two-Step Floating Catchment Area (2SFCA) method does not account for the topological constraints of pedestrian networks, leading to an overestimation of 15-minute coverage by 12.3%19. This discrepancy results in a “green volume illusion” in planning practices—for example, a community in Xi’an has a satellite-detected green coverage of 35%, but pedestrian-visible green volume is only 17.2%, with a significantly low spatial match (Kappa = 0.31)9. Additionally, static analytical frameworks fail to consider vegetation phenological rhythms. In northern cities dominated by deciduous broadleaf forests (such as Xi’an), NDVI values can differ by as much as 0.65 between summer and winter, yet 83.4% of existing studies rely on single-season cross-sectional data20. Seasonal fluctuations in vegetation lead to systematic biases in studies based on single time points. Research has shown that the marginal effect of green exposure on emotional regulation in spring and summer is 2.3 times higher than in autumn and winter21.

Environmental justice challenges of green exposure inequality

Research from the perspective of environmental justice indicates that green exposure inequality exhibits significant spatial differentiation patterns22. Numerous studies on environmental inequality have demonstrated that urban green exposure varies according to socioeconomic status (SES)23. Inhabitants of communities with lower SES tend to reside in areas with less green space, which limits their access to nature and the associated benefits, such as improved mental health, recreational opportunities, and physical well-being24. In contrast, communities with higher SES typically possess greater financial resources, cultural and social capital, and stronger political influence in maintaining and enhancing green spaces25. Astell-Burt, et al.26 utilized negative binomial distribution and logit regression models to examine the relationship between green space accessibility and socioeconomic status. Their findings indicated that in Australia, the higher the proportion of low-income residents in a community, the more limited their access to green spaces. Sit, et al.27 used Sentinel-2 satellite imagery to explore the spatial patterns of green space coverage near residences and along 10-minute walking routes, examining their relationship with housing types and urban forms. Their results showed that in Hong Kong, residents of private housing in low-income, high-density areas within the urban core endure the most severe lack of green space and overcrowded living conditions.

Pham, et al.28 analyzed vegetation distribution in Montreal using data extracted from high-resolution satellite imagery. Their results revealed an unequal distribution of vegetation across the city, disproportionately disadvantaging low-income populations. Li, et al.29 conducted a quantitative study on the spatial distribution of various types of urban green spaces in Hartford, Connecticut, using green indicators derived from multi-source spatial datasets. Their results showed that high-income communities in Hartford generally have more abundant green street resources compared to low-income communities. Rigolon, et al.30 conducted a systematic review and meta-analysis, concluding that green spaces provide greater protective benefits for low-SES individuals and communities compared to wealthier groups. They also found that the protective effects of green spaces vary depending on the distance from home, influencing different SES groups differently. This spatial deprivation effect may be further reinforced by housing price gradients, forming a “green space-housing price” positive feedback loop19.

Multi-source data provides opportunities for green exposure research

Street View Images (SVI) provide rich geographic and environmental information, including buildings, roads, vegetation, pedestrians, vehicles, and other elements31. First, Li, et al.32 have already used street view images to assess variations in greening across different residential streets. Second, the ability to process large-scale data enables the analysis of vast amounts of images within a short period, supporting large-scale urban and environmental studies. This provides a breakthrough in overcoming the limitations of insufficient sample sizes in past research. For example, Zhang, et al.33 used street view images and machine learning to develop a framework for assessing urban green space exposure, addressing the limitations of traditional data collection methods. Meanwhile, SVIs typically offer extensive coverage, spatiotemporal continuity, and greater intuitiveness and interpretability. They provide direct visual information that facilitates public understanding and engagement. Additionally, time-series street view images have also been used for evaluating community spaces34.

In urban studies, street view images have become a key data source for assessing green exposure35. First, SVIs can be used to extract green space indicators (i.e., green view index) as a data source for assessing green exposure. The proportion of green measured by street view images can, to some extent, serve as a substitute for the NDVI36. Second, SVIs can be used to extract the vertical distribution of street-level green exposure. Morakinyo and Lam37 conducted a correlation analysis between green exposure and air pollutants such as traffic emissions and street dust. Additionally, SVIs have been used to assess the impact of urban green exposure on residents’ subjective choices, including potential effects on travel preferences and comfort38. At the same time, SVIs enable large-scale evaluations of green exposure inequality among different population groups, contributing to valuable explorations of higher levels of spatial justice16.

Research objectives and scientific value

To address the limitations of existing research and further explore the seasonal variations in green exposure levels and their relationship with environmental justice, this study takes Xi’an as a case study and constructs a multidimensional “spatiotemporal-social” analytical framework. It aims to address two core questions: (1) examining the seasonal differences and fairness of green exposure across different living circles; (2) identifying the coupling mechanism between housing prices and green exposure, as well as their spatial distribution patterns. The findings of this study can provide decision-making support for climate-adaptive green space planning and promote the transition of environmental justice from quantitative balance to qualitative equity.

Methods and data

Research framework and study area

This study developed to assess seasonal green space inequality in the 15-minute city. This framework integrates the green view index (GVI), deep learning techniques, Moran’s I index, local binary relationship, Gini index, and urban rental price data (Fig. 1). First, we obtained street view images from Baidu Maps. Then, we performed semantic segmentation on these images to extract GVI. We estimated the level of green exposure at the community level by setting different temporal buffers. And, the Gini coefficient was used to express the green exposure inequality at different temporal buffers. Furthermore, using Linear regression analysis to examine the correlation between SES and seasonal differences in green exposure under different walking times. Finally, we used Moran’s I index and Local Binary Patterns (LBP) to examine the correlation between SES and seasonal differences in green exposure. Figure 2 illustrates the framework of this study.

Fig. 1
Fig. 1
Full size image

City housing price distribution map. Basemap from OpenStreetMap (https://www.openstreetmap.org).

Fig. 2
Fig. 2
Full size image

Research framework diagram.

This study uses Xi’an as an example. Xi’an has a warm temperate semi-humid continental monsoon climate, characterized by distinct seasons with variations in temperature and precipitation. As a renowned ancient capital of thirteen dynasties and a major tourist city, Xi’an has prioritized green space development as a core urban strategy since the launch of the Western Development Programme in 1999. The city has established a multi-layered green space system of “parks + green belts + ecological corridors”, including 63 urban green parks such as Ring City Park and Qujiang Pool Park. This study focuses on the urban core of Xi’an (within the Second Ring Road) as the research area (Fig. 3).

Fig. 3
Fig. 3
Full size image

Study area. Basemap from OpenStreetMap (https://www.openstreetmap.org).

Data collection

Street view summer and winter image collection

This study utilizes the Baidu Maps API to obtain the required street view images, collecting a total of 31,965 panoramic images from April 2014 to April 2022, covering all four seasons. To ensure an accurate representation of real-life green space distribution, we retrieved detailed road network information for the study area, including urban arterial roads, auxiliary roads, and dedicated roads39. We used a quantitative method to collect BSVIs samples based on grid units (100 × 100 m of each grid) generating from road network40. Centroids derived from road network-based grid units were designated as sampling points. These points were transmitted to the BSVI application programming interface (API) to download multi-angle images for each location, capturing front, left, rear, and right views31, which were subsequently stitched together (Fig. 4).

Fig. 4
Fig. 4
Full size image

Street view image collection. Basemap from OpenStreetMap (https://www.openstreetmap.org). image taken by the author.

Xi’an, located in northwestern China, experiences extended summer and winter seasons, with summer lasting an average of 105 days and winter lasting approximately 137 days. To deal with differences in images caused by changes in cities between 2014 and 2022, we collect SVIs in the summer (June to August) and winter (November to March) months from same sampling points. This made sure that the information we had was always the same. Using these selection rules, we found 22,573 summer and 47,504 winter images during the specified period. ArcGIS was used to visualize the collected data41 and buffer zones of 5 min (360 m), 10 min (720 m), 15 min (1080 m) and 30 min (2160 m) were created (Figs. 5 and 6). Subsequent to the implementation of seasonal filtration, the final dataset comprised 9,988 valid sampling points.

Fig. 5
Fig. 5
Full size image

5 min and 10 min sampling points and buffer zones.

Fig. 6
Fig. 6
Full size image

15 min and 30 min sampling points and buffer zones.

Housing price data

Property values are often closely related to residents’ wealth levels, making house prices a commonly used indicator to measure socioeconomic status (SES)42. According to a study by the UK Office for National Statistics (ONS), living near a park not only improves the residential experience, but also increases property value43. Therefore, housing prices may reflect the economic burden on community residents to access green spaces44. This study used housing (rental) price data from the secondary property website Anjuke (https://xa.anjuke.com) as of 5 November 2024 to assess the SES of residents’ communities. Anjuke provides listings for new homes, second-hand homes, rental properties, commercial real estate, office buildings, and overseas properties, as well as detailed community information including related prices, areas, and property rights. Python was used to scrape community characteristic data from Anjuke’s property section. This data covers 85 communities in the central area of Xi’an (Fig. 1). We calculated the average housing price for each community.

Data analysis

Greenery extraction

Green View Index (GVI) has been shown to be an effective measure for assessing the visibility of urban forests45. To extract greenery from the obtained street view images-including roadside trees, shrubs, herbaceous plants, and climbing plants-the SegFormer model was applied46. Its application in urban studies has been extensive, due to the highly accurate prediction results it generates. SegFormer is a frequently employed model for semantic segmentation, which is often used for the purpose of identifying urban green spaces and measuring greenness in high-resolution remote sensing images47. The model has also been employed for the purpose of visual quality assessment of Zhujiang Park in Guangzhou, China48. The images were preprocessed by resizing each to 512 × 512 pixels and standardizing pixel values to the \(\:\left[\text{0,1}\right]\) range33. SegFormer combines the advantages of Transformers and convolutional neural networks, offering strong global modeling capabilities along with the efficiency of a multi-layer perceptron (MLP) decoder. For detailed technical details, please refer to Supplementary Material S1.

Using the collected street view panoramic images, a standardized procedure is applied to count the selected green vegetation pixels and divide this by the total number of pixels in the image, yielding the ratio of green vegetation to the total image area49. In this study, multi-seasonal street view images were selected to explore seasonal green exposure at the community level within a 15-minute city framework. Then, the average green exposure specific image values for each community were calculated based on buffer zones of 5 min (360 m), 10 min (720 m), 15 min (1080 m), and 30 min (2160 m). The formula is as follows:

$$\:\begin{array}{c}GVI=\frac{\sum\:_{n=1}^{4}Are{a}_{{g}_{n}}}{\sum\:_{n=1}^{4}Are{a}_{{t}_{n}}}\end{array}$$
(1)

where \(\:Are{a}_{g\_n}\) refers to the number of green pixels in the image captured from the nth direction at the sample point. \(\:Are{a}_{t\_n}\)refers to the total number of pixels in the image captured from the \(\:n\)th direction at the sample point.

Regression analysis and spatial statistical analysis

The present study utilized linear regression to investigate the correlation between green space exposure and SES. Linear regression is a commonly used statistical method for analyzing the linear relationship between a dependent variable and one or more independent variables50. This model is simple and intuitive, commonly used for prediction, trend analysis, and exploring relationships between variables35. Evaluation metrics include R-square and Mean Squared Error (MSE). \(\:Pricet\) represents the SES at a walking time of \(\:t\:\)at is the intercept indicating the price when \(\:t\:=\:0\:\). \(\:bt\) is the slope, representing the rate of change in community SES over time. \(\:ct\) represents the seasonal variation in green exposure. The formula is as follows:

$$\:\begin{array}{c}Pric{e}_{t}={a}_{t}+{b}_{t}\times\:t+{c}_{t}\end{array}$$
(2)

We further examined the correlation between spatial SES and seasonal differences in green exposure using Moran’s analysis. Spatial autocorrelation is often used as a method to study urban green space exposure inequality. In this paper, both global bivariate Moran’s I \(\:{I}_{P,A}\) and local bivariate Moran’s I \(\:{I}_{P,A}^{{\prime\:}}\) are used to analyze the spatial autocorrelation between housing prices and seasonal green exposure. The formulas are as follows:

$$\:\begin{array}{c}{I}_{P,A}=\frac{N\sum\:_{i}^{N}\sum\:_{j\ne\:i}^{N}{W}_{ij}{Z}_{i}^{P}{Z}_{j}^{A}}{\left(N-1\right)\sum\:_{i}^{N}\sum\:_{j\ne\:i}^{N}{W}_{ij}}\end{array}$$
(3)
$$\:\begin{array}{c}{I}_{P,A}^{{\prime\:}}=\sum\:_{j=1}^{N}{W}_{ij}{Z}_{j}^{A}\end{array}$$
(4)

where, \(\:N\) represents the total number of community units within the study area; \(\:{Z}_{i}^{P}\) denotes the standardized value of housing prices in the \(\:i\)-th community; \(\:{Z}_{j}^{A}\) represents the standardized value of seasonal green space exposure in the \(\:j\)-th unit; and \(\:Wij\) is the spatial weight matrix used to measure the correlation between the \(\:i\)-th and \(\:j\)-th units.The Moran’s I statistic ranges from − 1.0 to + 1.0. A value greater than 0 indicates positive spatial correlation, with proximity to + 1.0 reflecting stronger clustering. A value of 0 suggests spatial randomness16.

In addition, based on spatial autocorrelation theory, clustering analysis using the Local Bivariate Moran’s I is considered an effective method for identifying inequality hotspots or concentrated areas of environmental exposure. The clustering map generated by the Local Bivariate Moran’s I reveals four spatial association patterns between community housing prices and the GVI: Low-Low (where low SES areas are surrounded by low GVI areas), Low-High (where low SES areas are surrounded by high GVI areas), High-High (where high SES areas are surrounded by high GVI areas), and High-Low (where high SES areas are surrounded by low GVI areas).

Gini index

In this study, the Gini coefficient is used to analyze seasonal green exposure across different SES51. The formula is as follows:

$$\:\begin{array}{c}Gini=1-\sum\:_{i=1}^{K}{p}_{i}^{2}\end{array}$$
(5)

The Gini index ranges from 0 to 1. A Gini index of 0 indicates that the node is perfectly pure, meaning all samples belong to the same class. The larger the value, the more unequal it represents.

Results

Seasonal variation and inequality of green exposure from the 15-minute City perspective

Based on the multi-temporal interpretation of Street View imagery, the study reveals the seasonal attenuation pattern of green exposure at the community level in Xi’an (Figs. 7 and 8). In the 5-minute walking buffer zone (360 m), the number of communities with green exposure values ≥ 0.3 in summer is 26 (23.0%), which decreases to 16 in winter (a decrease of 38.5%). In contrast, the area with low values (GVI < 0.1) increases to 32 in winter (a 39.1% increase compared to summer). The spatial distribution analysis shows that the high value areas in summer are mainly concentrated in the eastern part of the city (GVI = 0.42 ± 0.05), while in winter they shift northwards to form a new hotspot (GVI = 0.38 ± 0.07). As the walking distance increases to 15 min (1080 m), the seasonal differentiation shows a spatial reversal: the high value area in the northwest in summer transitions to the dominant southeast zone in winter (ΔGVI = 0.21, p < 0.05). These results show that green exposure at the community level shows significant seasonal and spatial disparities, with distinct geographical distribution characteristics.

Fig. 7
Fig. 7
Full size image

Comparison of seasonal changes in 5-minute and 10-minute green exposure. Basemap from OpenStreetMap (https://www.openstreetmap.org).

Fig. 8
Fig. 8
Full size image

Comparison of seasonal changes in 15-minute and 30-minute green exposure. Basemap from OpenStreetMap (https://www.openstreetmap.org).

Additionally, a comparison was made of green exposure at different distances at the community level between summer and winter (Fig. 9). It was found that at the 5-minute (360 m) scale, the southeastern and some northwestern areas of the study region exhibited significant seasonal differences in green exposure. At the 10-minute (720 m) scale, the central and some northern areas showed more pronounced seasonal differences. The areas demonstrating significant seasonal differences at the 5-minute and 10-minute scales were found to be relatively dispersed. At the 15-minute (1080 m) scale, the northwestern and southeastern parts of the study region exhibited greater seasonal variations in green exposure, while at the 30-minute (2160 m) scale, the northwestern and some central areas demonstrated larger seasonal differences. The regions exhibiting significant seasonal variations at the 15-minute and 30-minute scales were also relatively scattered. These findings indicate that, from the perspective of a 15-minute city, seasonal inequality in green exposure exists at the community level. During walking times of 5 and 10 min, green exposure in summer is higher than in winter, whereas at walking times of 15 and 30 min, green exposure in winter surpasses that in summer.

Fig. 9
Fig. 9
Full size image

Comparison of seasonal green exposure inequality. Basemap from OpenStreetMap (https://www.openstreetmap.org).

Furthermore, Fig. 10 further illustrates the findings of our fairness analysis using the Gini coefficient. The results point to a non-linear rise in green exposure inequality as accessibility time increases. Specifically, the Gini coefficient climbs from 0.2686 (with a 95% confidence interval of 0.2512 to 0.2860) at the 360 m scale to 0.4428 (95% CI: 0.4210 to 0.4646) at the 2160 m scale. This overall trend is statistically significant (F = 19.32, p < 0.001). Notably, a local inflection point is observed at the 15-minute travel threshold (corresponding to the 1080 m scale, Gini = 0.4038, 95% CI: 0.3855 to 0.4221). At this point, the rate at which inequality increases slow down substantially; the decay rate is approximately 32.4% lower compared to preceding intervals. This indicates the marginal fairness improvement effect of urban green spaces. The spatial heterogeneity index indicates maximal seasonal disparity along the northwest-southeast axis (Moran’s I = 0.47, p = 0.002), which may be related to the spatial coupling between vegetation phenology and the urban heat island effect.

Fig. 10
Fig. 10
Full size image

Gini Index trend across different walking time.

The coupling relationship between green exposure and urban housing prices

This study utilizes spatial correlation and local spatial autocorrelation analysis to elucidate the spatial association characteristics between housing prices and seasonal differences in green exposure at multiple scales. The relationship between the two exhibits dynamic characteristics that change with spatial scale (Fig. 11). At smaller spatial scales (5-minute, 10-minute, and 15-minute walking buffers), a negative correlation is observed, which weakens as distance increases (5 min = -0.156 > 10 min = -0.039 > 15 min = -0.019). This phenomenon indicates that within closer living circles, residents are more likely to choose residential areas with stable green environments, and regions with smaller seasonal variations tend to have higher housing prices. However, when the spatial scale expands to the 30-minute walking circle, the correlation shifts significantly, showing a weak positive correlation (r = 0.101). This shift unveils an intriguing phenomenon: in larger spatial areas, regions exhibiting significant seasonal variation in green space tend to have higher housing prices, which may be indicative of residents’ preference for a more diverse plant phenology experience within their broader activity range.

Fig. 11
Fig. 11
Full size image

Correlation analysis of housing prices and seasonal differences in green exposure at different walking times.

Next, local spatial autocorrelation analysis further reveals the evolution of spatial clustering at different scales (Fig. 12). Within the 5-minute walking buffer (360 m), the spatial distribution shows significant local clustering characteristics. High-High regions (17 areas) are mainly concentrated in the southern and south-eastern parts of Weiyang District, the north-western part of Lianhu District and Beilin District, forming several stable high-value aggregation zones. Meanwhile, High-Low areas (25 areas) form contiguous distributions in the western parts of Lianhu and Beilin districts, forming a unique transitional zone. This detailed pattern of spatial differentiation clearly reflects the interaction between housing prices and green spaces at the micro scale. As the spatial scale expands to the 10-minute walking buffer (720 m), the spatial clustering pattern is reorganized. The number of High-High areas increases to 21, with more stable spatial aggregation zones forming in the southeastern part of Weiyang District and the western part of Beilin District. At the same time, the spatial distribution of other area types tends to become more balanced, reflecting the restructuring of the mesoscale spatial correlation.

Fig. 12
Fig. 12
Full size image

Local bivariate Moran’s I spatial correlation analysis between walking time and SES. Basemap from OpenStreetMap (https://www.openstreetmap.org).

Within the 15 min walking buffer (1080 m), the number of High-High areas increases to 24, but shows greater spatial dispersion. This transition suggests a weakening of spatial association strength. When extended to the 30-minute walking buffer (2160 m), the spatial pattern shows macro-organizational characteristics. High-High areas form significant aggregation zones in the north-west, north-east and south-east sectors, while other area types show relatively regular patterns of differentiation. This multiscale spatial clustering analysis shows how housing price-green exposure correlations evolve with increasing mobility ranges, revealing systematic spatial-seasonal interdependencies in urban systems.

From the perspective of overall spatial evolutionary trends, as the walking range expands from 360 m to 2160 m, the study area exhibits three notable spatial characteristics. First, the western region shows strong spatial stability, maintaining a relatively consistent spatial association pattern across different scales. Second, the eastern region exhibits significant dynamic characteristics with changing spatial scales, reflecting the complexity of the area’s urban functional structure. Third, the central region shows pronounced spatial heterogeneity, which may be related to the functional mix of the area and the spatial configuration of the green space system. These multi-scale spatial analysis results not only reveal the complex spatial dependence between housing prices and seasonal variations in green exposure, but also provide new research perspectives for evaluating the value of urban green space systems. These findings have important practical implications for optimizing urban green space planning and improving residential environmental quality.

Discussion

This study takes Xi’an City as a research object, combines Green View Index, deep learning, spatial statistical analysis methods, and big data on urban rental prices, establishes an evaluation framework for seasonal green exposure inequality in cities, and analyses seasonal green exposure inequality in Xi’an City at the 15-minute community level. The results show: (1) in the context of the 15-minute city, the level of green exposure shows a declining trend, which is exacerbated by seasonal changes; (2) green exposure inequality increases nonlinearly with accessible time and exhibits spatial heterogeneity; (3) there is a significant spatial synergy effect between housing prices and seasonal differences in green exposure.

Spatial differentiation characteristics of green exposure

In this study, we found that seasonal green space exposure varies across communities and is closely related to SES, suggesting that residents in communities with higher economic levels may enjoy better urban green space across multiple seasons. This finding is consistent with the socio-economic gradient of green space theory proposed by Dadvand, et al.52. However, we need to be careful when using rental prices as a way to measure SES. Several factors may contribute to this finding. First, high-income municipalities tend to invest more in greening projects, with landscape designs emphasizing three-season flowering and year-round greenery, resulting in better green spaces in wealthier areas. In addition, the old city of Xi’an contains a large number of city walls and historical heritage sites. This phenomenon is consistent with the ‘heritage-greening trade-off’ observed by Rochel53 in Paris, Brussels and Lisbon, where strict zoning regulations in heritage areas often limit green expansion, resulting in an uneven distribution of green resources. Due to heritage considerations, road construction in these areas avoids cultural sites, resulting in narrower roads and higher spatial point densities in buffer zones, increasing seasonal variations in green exposure. Narrower streets in historic urban cores physically constrain planting strip width and canopy growth potential, tending to accommodate smaller or fewer tree and shrub specimens relative to wider thoroughfares in newer districts. This intrinsically limits the achievable maximum GVI. Historically evolved urban geometry may generate distinctive microclimates, such as reduced airflow, altered solar radiation, and seasonally amplified effects. Deciduous species prevalence due to historical planting practices or spatial constraints may further drive pronounced seasonal GVI contrasts. In contrast, in newly developed areas, road construction is less constrained by existing buildings. In addition, the need to facilitate rapid transport on ring roads results in less seasonal variation in green exposure. Although most green spaces in Xi’an are publicly accessible and free, there are still inequalities between communities of different socio-economic status (SES), consistent with previous studies. The disparities in streetscape greenery extend beyond mere reflections of contemporary SES, they are profoundly intertwined with historical urban morphologies, planning priorities, and structural infrastructural legacies. Policymakers should prioritize the improvement of streetscapes to ensure that residents of all income levels have access to adequate green spaces throughout the seasons. The formulation of an integrated spatial plan that explicitly links heritage conservation, climate-adaptation objectives, and equity goals is also crucial. Furthermore, the institution of participatory budgeting for community greening pilots is necessary, as is the devolution of greening resources and decision-making authority to sub-district offices or the community level.

Exploration of green exposure from an environmental justice perspective

Based on Moran’s I index, we identified spatial autocorrelation patterns between SES and seasonal green exposure across municipalities, categorized into High-High, High-Low, Low-Low and Low-High clusters54. High-Low clusters were observed in the high-tech zone of Xi’an, while Low-High clusters were observed in Huyi and Chang’an districts in the southern suburbs. These findings are in line with previous studies. The High-Tech Zone serves as a hub for high-tech industries and is characterized by high-density office buildings that limit the availability of green space55. In the city center, both the protection of underground cultural relics and the semi-arid climate of northwest China contribute to the relatively low number of trees56. Furthermore, as a historical and cultural city, Xi’an attracts visitors from all over the world to its heritage-rich city center, such as the ancient city walls. This tourism-driven demand also results in higher rental prices in the city center. Conversely, in Huyi and Chang’an districts, the proximity to the Qinling Mountains - one of China’s major climatic transition zones - provides abundant natural greenery57. From the perspective of rental prices, these districts were transformed from counties to urban districts in 2002 and 2017, respectively. The transition from county-level to district-level rental price averages requires a longer period of economic development, resulting in relatively lower housing prices58. The city center combines heritage preservation constraints with intense tourism pressure. Elevated rental prices here are primarily driven by location premiums from proximity to cultural landmarks and demand for tourist accommodations, creating conditions where high SES coexists with heritage-imposed low GVI. Resident populations may consequently experience environmental inequities. This complexity underscores the limitations of using rental prices alone as an SES proxy in multifunctional urban cores, necessitating multifaceted approaches in future research. This study focuses on the 15-minute community-scale analysis of green space, which differs from previous studies that mainly rely on two-dimensional remote sensing data to assess green space. The use of street view imagery as a dataset for green space assessment may contribute to discrepancies in findings compared to previous studies59. In the context of High-Low areas, a proportion of the economic benefits generated by high-value locations should be allocated to a dedicated green infrastructure fund, with the objective of financing the development of green space. Concurrently, the implementation of incentive programs should be encouraged, with the objective of encouraging residents to adopt small-scale greening solutions that are compatible with heritage contexts that are site-constrained. In the context of Low-High zones, proactive measures must be implemented to prevent the erosion of existing environmental advantages due to uncontrolled urbanization pressures. Future urban development in these areas must ensure that new constructions comply with high greening standards from inception, thereby averting emergent socio-spatial disparities.

Innovations in methods and limitations

This study develops a novel approach to quantifying seasonal green space variation using street view imagery and machine learning techniques. Compared to traditional green space assessment methods, this approach is more time-efficient and effective, and makes a significant contribution to understanding the relationship between green space quality and residents’ well-being. However, several limitations should be addressed in future research. First, the use of rental prices as the sole indicator of SES has certain limitations and other socio-economic factors should be considered. This is particularly true in Chinese cities, where the presence of informal housing and state-subsidized units makes it difficult to use SES proxies. Second, this study uses the Green View Index, which only reflects street-level greenery captured in Street View imagery. Many green spaces, such as those in residential areas or areas without Street View coverage, are not included. As Biljecki and Ito31 point out, even when using crowdsourced data (e.g. Mapillary), 35% of microgreens remain uncovered due to camera placement on vehicle roofs, which limits access to pedestrian-only or restricted areas. Third, while street view imagery provides time-series data, it does not fully cover all historical periods. Given the rapid pace of urban development in China, some areas are experiencing drastic changes in green space coverage due to changing urban planning policies, and these changes may not be captured in the available imagery. Fourth, seasonal data availability varies, and the number of data points within 15-minute buffer zones is uneven across municipalities. Future studies can use more crowdsourced data to improve the accuracy of seasonal green exposure assessments.

Conclusion

The issue of inequality in green space distribution has received increasing attention. This study focuses on the inequality of seasonal green exposure in the main urban area of Xi’an, integrating the Green View Index, deep learning, linear regression, spatial statistical methods and the Gini coefficient to construct an evaluation framework for seasonal green exposure within the “15-minute city” concept. Unlike traditional remote sensing methods, we use multi-seasonal street view data to extract green space distribution patterns, analyze residents’ access to green spaces at different walking distances in different seasons, and assess the impact of seasonal differences in green exposure on housing prices. Linear regression analysis reveals a negative correlation between housing prices and green space stability within a 5–15 min walking radius, which shifts to a weak positive correlation in the 30 min zone. Through bivariate Moran’s I analysis, we find that as the walking radius increases (from 5 to 30 min), high value clusters transition from localized dense formations to broader macro-scale belt-like structures, developing along a northwest-southeast axis from the southern part of Weiyang District. This pattern shows spatial stability in the west, dynamic change in the east, and heterogeneity in the central region, allowing for targeted interventions at different spatial scales and regional levels. In the 5-minute life circle, for High-High clusters (e.g. southern Weiyang District), priority should be given to multi-seasonal uniform greening by planting evergreen species to increase green space stability. For High-Low areas (e.g. Lianhu and Beilin districts), vertical greening can compensate for the seasonal loss of green cover caused by deciduous trees in winter. In the 30-minute living circle, taking into account the northwest-southeast directional change and residents’ demand for seasonally diverse vegetation in distant travel zones, tree species with pronounced seasonal variation should be selected to create a seasonal landscape corridor along the northwest-southeast axis. Meanwhile, for stable western areas, efforts should focus on improving winter greening while maintaining the existing green system. For dynamic eastern areas, priority should be given to resilient plant species with high adaptability, such as cold-resistant, drought-tolerant and water-resistant varieties. For heterogeneous central areas, the introduction of more pocket parks can help increase green space in densely built environments.

In order to achieve green space equality in the “15-minute city” and improve urban comfort and equity, we emphasize three policy recommendations for future planning: (1) When constructing infrastructure in newly developed areas, attention should be paid to multi-seasonal green space planning, ensuring a composite planting structure and low fluctuations in the Green View Index. (2) Establish a plant-level intervention system at the community level, taking into account the inherent characteristics of plants to ensure their survival rate. (3) Build seasonal green exposure walking units for the “15-minute city” to address the spatial and temporal mismatch in seasonal green equity.