Equity Evaluation of Street-Level Greenery Based on Green View Index from Street View Images: A Case Study of Hangzhou, China

Abstract

Equity in urban greenery is essential to improving residents’ well-being and contributing to environmental justice. Research on equity in street-scale urban greenery remains limited, but this study addresses it by employing the green view index (GVI), a widely recognized indicator for assessing green space quality from a pedestrian perspective, using semantic segmentation methods and Baidu Street View (BSV) images to quantify street-level greenery. Through spatial clustering and hot spot analysis, the visibility and spatial distribution of street greenery in Hangzhou’s central urban area were examined. Furthermore, the Lorenz curve, Gini coefficient, and location entropy were applied to evaluate disparities in green visibility across urban spaces. The results show that the average GVI at the sample point level, road level, and district level in the study area are 0.167, 0.142, and 0.177, respectively. Meanwhile, the spatial heterogeneity of the GVI is highly pronounced, with distinct clustering characteristics. The Gini coefficient of street greenery visibility is 0.384, indicating a moderate level of inequality in the distribution of greenery resources. Notably, a higher GVI does not necessarily correspond to better internal greenery equity, highlighting disparities in the distribution of urban greenery. This study offers a more precise and refined quantification of urban greenery equity, providing critical insights for addressing spatial disparities and informing urban planning strategies aimed at promoting equitable green infrastructure.

1. Introduction

Urban greenery, which refers to all visible vegetation elements in cities, is an essential element of urban environments and offers important ecosystem services that contribute to sustainable and livable cities [1,2]. It plays a crucial role in climate regulation [3], noise reduction [4], air pollution mitigation [5], and the enhancement of urban aesthetics [6]. However, research indicates that many cities have developed with significant disparities in access to urban green spaces (UGSs)—areas of urban land partly or completely covered with grass, trees, shrubs, or other vegetation—leading to environmental injustices [7,8,9]. This unfair and uneven distribution of urban greenery can easily lead to considerable disparities in residents’ living conditions and social well-being, ultimately impacting their physical and psychological health [10]. Therefore, assessing and quantifying the equity of urban greenery is essential for addressing environmental disparities and fostering more inclusive and sustainable urban development [11].

Although urban green space has long been recognized as vital to public health and environmental quality, street-level greenery often remains overlooked [12]. In fact, greenery along streets is closely tied to residents’ daily activities such as commuting, recreation, and social interaction, offering direct and frequent contact with nature [13,14]. This raises a critical question: have the streets we travel every day achieved green equity? While streets may appear more evenly distributed than centralized parks or woodlands, their green visual accessibility can vary widely, resulting in potential inequities [15]. For example, Heynen et al. found disparities in tree canopy cover across racial groups in Milwaukee, where Hispanic communities had significantly less canopy coverage than non-Hispanic whites [16]. Zhou and Kim analyzed six cities in Illinois and found that although park access was relatively even, tree canopy cover was disproportionately lower in minority neighborhoods [17]. Similarly, Liu et al. showed that street greenery exposure was positively correlated with household income, underscoring significant socio-economic inequities in urban vegetation visibility [18]. Therefore, assessing residents’ visual perceptions and subjective experiences of urban greenery landscapes at the street level constitutes an important dimension for measuring the level of urban greening.

A diverse range of data sources is available for assessing urban street-level greenery equity. Traditional datasets include the Normalized Difference Vegetation Index (NDVI) obtained from remote sensing imagery [19], tree inventories [20], and various greenery indicators [21]. However, a key limitation of remote sensing imagery is its top-down perspective, which does not accurately capture pedestrian perceptions of greenery [22]. In contrast, street view images have become a more effective tool for evaluating human-scale greenery, offering a perspective consistent with the experiences of pedestrians, cyclists, and drivers [23]. Platforms such as Google Street View, Baidu Maps, and Tencent Maps are providing an increasing number of street view images, providing researchers with large datasets to analyze urban greenery at a finer spatial resolution [24]. Among these, BSV has been archiving images since 2013 and frequently updates its database, making it a valuable resource for tracking urban street-level greenery trends in China [25]. Consequently, the GVI, derived from street view images, has become one of the most widely utilized metrics for quantifying the visibility and accessibility of urban greenery [26,27].

Given the unique role of the GVI in assessing urban greenery equity, improving its measurement accuracy is crucial. Early studies on GVI primarily relied on manual field surveys and photographic analysis, which were not only time-consuming but also susceptible to human error. Li et al. introduced a semi-automated method utilizing Google Street View images and pixel-based color recognition in Adobe Photoshop to estimate GVI [28]. However, this approach still required substantial manual effort, limiting its scalability. In addition, recent progress in computer vision and deep learning have dramatically improved the accuracy and efficiency of GVI extraction through image semantic segmentation techniques [29]. The development of deep learning models such as Fully Convolutional Networks (FCN), SegNet, Pyramid Scene Parsing Network (PSPNet), and DeepLabV3+ has transformed street greenery analysis by enabling high-precision segmentation of vegetation in urban imagery [30,31,32,33]. Among these models, DeepLabV3+ has demonstrated superior segmentation performance and is now widely adopted in GVI studies due to its enhanced accuracy and robustness [34].

Despite the growing number of studies examining equity in the distribution of green spaces, substantial variation exists in the methodologies used to quantify equity. One widely adopted approach is accessibility, usually measured as the distance to the nearest green space or the proportion of the population living within a specific distance of a green space [35,36]. However, accessibility alone does not guarantee that greenery is visible or perceptible in everyday life. By contrast, the GVI captures the green visibility experienced along streets and public spaces [23]. Another commonly used metric is green space provision or green space coverage, which is among the simplest and most straightforward indicators to quantify, making it particularly prevalent in large-scale studies [37,38]. A less frequently employed concept is population pressure, which estimates the likelihood of overcrowding within a green space, based on the assumption that all individuals visit their nearest green space simultaneously [39]. The traditional methods often overlook factors like green space quality, actual usage patterns, and spatial distribution imbalances, leading to incomplete assessments of equity [18,40]. More recently, due to their ability to provide a more nuanced quantitative analysis, measures such as the Gini coefficient and location entropy have been employed to assess the spatial equity of green space distribution. Initially proposed by Gini in 1912 to measure income inequality across regions or countries, the Gini coefficient has been extensively utilized in urban research to evaluate the equitable allocation of equity of street-level greenery [41]. Similarly, location entropy, which was originally applied to measure the concentration of economic development within a specific area, is increasingly being used to assess the spatial fairness of urban green space resource distribution [42,43]. These evolving methodologies provide a more nuanced and comprehensive framework for assessing urban greenery equity, facilitating more informed and equitable urban planning strategies.

While the Gini coefficient and location entropy have been widely adopted to assess the distributional equity of urban green resources, these metrics are often based on planimetric data (e.g., green space area, NDVI), which fail to capture how greenery is actually experienced by pedestrians at eye level [12,28]. As a result, such approaches may overlook the nuances of street-level visual exposure to greenery [13]. This study addresses this limitation by combining the GVI—a visibility-based measure extracted from street-level imagery—with equity indicators. This integration allows for a more comprehensive understanding of the spatial equity of urban greenery from a human-centered perspective. Leveraging BSV images and the DeepLabV3+ semantic segmentation model, this research aims to (1) quantify the GVI across various road networks and urban areas using machine learning-based image segmentation; (2) analyze the spatial clustering and distribution patterns of GVI at different scales to identify areas with high and low levels of street greenery; (3) assess the equity of street greenery distribution by employing the Gini coefficient, Lorenz curve, and location entropy to measure disparities in green visibility; and (4) propose urban planning strategies aimed at optimizing the distribution of street greenery and promoting environmental justice in Hangzhou. This study provides a comprehensive and data-driven approach to evaluating urban greenery equity, offering meaningful guidance for promoting sustainable urban growth and enhancing fairness in green space planning.

2. Materials and Methods

2.1. Study Area

Located on the eastern coast of China, Hangzhou serves as the capital of Zhejiang Province and is one of the 19 national ecological garden cities, which means that extensive urban greenery is needed. The city has made significant progress in urban development and ecological conservation, featuring a forest coverage rate of over 66%, offering 16.2 m² of park green space per capita, and maintaining 45% urban greenery coverage. This study focuses on the Xihu, Binjiang, Gongshu, and Shangcheng districts, which are important components of the main urban area of Hangzhou, characterized by dense infrastructure and a high population density, covering an area of approximately 600 km² (Figure 1). Given the ongoing urban regeneration efforts, research in these areas can offer valuable insights into strategies for enhancing the quality of visible greenery and improving urban environmental conditions in densely populated settings.

2.2. Data Sources

This paper mainly relied on administrative road network data, population data, and BSV images as the core datasets. The urban road network was derived from the OpenStreetMap (OSM) platform (https://www.openstreetmap.org/, accessed on 21 August 2024). Street view images were collected from Baidu Map, using an automated batch extraction process implemented with Python 3.12. Population data at a 100 m spatial resolution were sourced from the WorldPop platform.

2.3. Methods

The Research flow is shown in Figure 2. In this study, the GVI was selected as the primary index to evaluate the extent of urban street greenery, which was calculated using BSV images and semantic segmentation techniques. To examine the spatial distribution patterns of GVI across the study area, hot spot analysis was conducted. Moreover, based on the GVI, we employed the Lorenz curve, the Gini coefficient, and location entropy to assess the equity of street greenery.

2.3.1. GVI Calculation

(1)

Selecting sample points

Given the diverse road classifications in OSM, certain types such as paths, tracks, and steps were excluded due to their limited relevance. The street network was then clipped to administrative boundaries using ArcGIS 10.8. To establish sampling points, a 200 m interval was applied to extract points from the street shapefile, with midpoints selected for roads shorter than 200 m [25,41]. This study selected a total of 8605 sample points, and each sampling point retained the original OSM street attributes and their latitude and longitude coordinates.

(2)

Panoramic street view image acquisition and semantic segmentation

Following the determination of sampling points, 7085 BSV images were collected from Baidu Map, as some points lacked corresponding data. Images were captured at four angles—0°, 90°, 180°, and 270°—forming a continuous 360-degree horizontal panorama of the streetscape (Figure 3) [44]. To maintain consistency, the vertical line of sight was fixed at 0°. Due to limitations in BSV image coverage, a portion of road segments from OpenStreetMap lacked valid panoramic imagery. Sampling points without available street view images were excluded from the analysis to ensure data integrity.

The Cityscapes dataset (https://www.cityscapes-dataset.com/news/, accessed on 25 December 2024) is a large-scale urban landscape dataset specifically designed to benchmark models for image segmentation in urban environments [26]. By providing high-quality labeled data, it facilitates the reliable classification of street view images, thereby improving the precision of GVI assessment. Leveraging this dataset, semantic segmentation was performed using a DeepLabv3+ model pre-trained on Cityscapes (Figure 4). This model, incorporating machine learning techniques, enabled the accurate segmentation of urban landscape features, ensuring a robust and detailed evaluation of street greenery [45].

(3)

Calculation of GVI

The GVI was used to assess the green visibility of urban streets, which specifically represents the percentage of green area in the panoramic image of the streetscape, and is calculated using the following formula [46,47].

$$GVI={\displaystyle \frac{{Area}_g}{{Area}_t}}={\displaystyle \frac{{\sum }_{i=4}^4{area}_{gi}}{{\sum }_{i=4}^4{area}_{ti}}},$$

(1)

where GVI represents the green view index. Area_g is the area occupied by green vegetation in the BSV panorama. Area_t refers to the entire area in the BSV panorama. The area_gi refers to the green-covered area in each directional image (0°, 90°, 180°, and 270°) surrounding the sampling point, while area_ti corresponds to the total area of each individual image (0, 90°, 180°, and 270°). According to the relevant literature [48,49], the GVI is categorized into five classes: extremely high (>35.0%), high (25.0–35.0%), medium (15.0–25.0%), low (5.0–15.0%), and very low (<5.0%).

2.3.2. Spatial Distribution and Agglomeration Characteristics

The distribution of the GVI is analyzed across three distinct spatial scales: point, road, and district. At the point scale, GVI corresponds to the values measured at individual sampling points. At the road scale, the greenery level of a street is represented by the average GVI of all sampling points along the road network. Finally, at the district scale, GVI is aggregated to offer a comprehensive overview of the average greenery level across the four districts, offering a broader spatial perspective.

To further explore the spatial clustering patterns of the GVI, the Getis–Ord Gi* statistical method was employed. As a hot spot analysis method, it calculates the statistics of each feature in the dataset and is a powerful method for identifying localized spatial clustering patterns [50]. This approach effectively differentiates between hot spots (areas where high GVI values are highly clustered) and cold spots (areas where low GVI values are clustered), thereby offering deeper insights into the spatial distribution of urban greenery. Accordingly, this study applies the Getis–Ord Gi* statistic to analyze the spatial clustering characteristics of the GVI, facilitating a comprehensive understanding of greenery distribution patterns.

2.3.3. Equity Assessment of Street Greenery

Street greenery is associated with both tangible and intangible economic benefits [51]. Planting and maintaining urban greenery can yield quantifiable returns, such as improved public health, reduced heat island effects, increased property values, and potential savings in stormwater management costs [52]. Intangible benefits—such as mental health, social cohesion, and equity, which is the focus of this study—are harder to quantify but are an important part of assessing the value of urban greenery. This study uses the Gini coefficient and location entropy to measure equity within the intangible benefits of urban street greenery.

(1)

Lorenz curve and Gini coefficient

The Lorenz curve was proposed by Max Lorenz in 1905 and it graphically represents inequality [53]. Then, Corrado Gini introduced the Gini coefficient in 1912 as a measure to quantify the area beneath the Lorenz curve. In the field of environmental equity, it is commonly used to evaluate the overall fairness level of the research object, and after more than a decade of development, the Gini coefficient and the Lorenz curve have become more mature in the field of environmental equity, and this study utilizes both of them to evaluate the fairness of the overall greenness of Hangzhou from the macro level [54].

The Gini coefficient was calculated to quantify the inequality of street-level greenery across districts. To assess statistical uncertainty, we applied a bootstrap resampling procedure with 1000 iterations to estimate the empirical distribution of each Gini coefficient and derive 95% confidence intervals (CI) [55].

(2)

Location entropy

Location entropy is a critical metric for analyzing the spatial distribution of area-based elements and has gained increasing prominence in the evaluation of green space allocation fairness. In the context of this study, location entropy is defined as the ratio between the supply–demand level of green visibility available to the residential population and the overall supply–demand level of green visibility across the entire study area. Location entropy was calculated based on street view sampling points. The number of valid points with both GVI values and population data is as follows: Xihu District—1466 points, Shangcheng District—1990 points, Gongshu District—1984 points, and Binjiang District—927 points. This metric effectively captures local variations in the distribution of green visibility, thereby providing valuable insights into spatial equity, and is calculated using the following formula [10,43].

$$Q_i={\displaystyle \frac{g_i/p_i}{g/p}}$$

(2)

where Q_i represents the location entropy of GVI at point i, g_i refers to the GVI value at point i, p_i represents the population at that point, and the symbols q and p indicate the total GVI and total population within the study area, respectively. A location entropy value less than 1 suggests that the supply of green visibility relative to demand at a given location is lower than the study area’s average, with smaller values representing an even lower level of supply. In contrast, when the location entropy exceeds 1, it suggests that the green visibility supply–demand level in that area exceeds the overall average, with larger values reflecting a higher concentration of green visibility resources.

3. Results

3.1. Spatial Clustering and Distribution Patterns of GVI

Table 1. GVI statistics at different levels.

Area	Min	Max	Mean	Std
Points	0	0.910	0.167	0.141
Roads	0	0.901	0.142	0.144
Districts	0.152	0.203	0.177	0.018

provides a descriptive summary of the GVI at three spatial scales. In terms of GVI values, the GVI across all sampling points in the study area ranged from 0 to 0.910, with an average of 0.167. At the road level, GVI values ranged from 0 to 0.901, with a mean of 0.142. At the district level, the GVI ranged from 0.152 to 0.203, with an average of 0.177. In terms of spatial distribution, as shown in Figure 5, the GVI in Hangzhou exhibits distinct patterns across different scales. At the point level (Figure 5a), GVI values follow a clear trend, with higher levels concentrated in the urban core and lower levels observed in outlying areas. Higher GVI values are mainly concentrated in the central regions, such as parts of Xihu District and Binjiang District, reflecting effective urban greening planning. Medium GVI values form a transitional zone between urban and suburban areas, while lower levels of GVI are mainly concentrated in the outer regions of both Xihu District and Gongshu District. This distribution can likely be attributed to factors such as higher urban density and industrial land use in these areas. At the road level (Figure 5c), the spatial distribution characteristics of road GVI align with those observed at the sampling point level. At the district scale (Figure 5e), Xihu District exhibits the highest average GVI (0.203), benefiting from its natural landscape. Binjiang District follows with a GVI of 0.181, attributed to well-planned green spaces and waterfront greenery. Gongshu District (0.173) maintains a balance between green infrastructure and urbanization, while Shangcheng District, with the lowest GVI of 0.152, highlights the need for further vegetation development.

In terms of spatial clustering characteristics, high-value clusters of GVI (marked in red) are primarily concentrated in the northeastern part of Xihu District, the northwestern part of Gongshu District, the southern part of Shangcheng District, and much of the central area of Binjiang District (Figure 5b,d). These regions exhibit higher GVI values, likely due to the presence of abundant green spaces, well-planned parks, and riverfront landscapes. Urban planning in these areas appears to prioritize the integration of greenery within residential and commercial spaces. In contrast, low-value clusters (marked in blue) are concentrated in the outskirts of Xihu District, parts of Gongshu District, and some industrialized zones in Binjiang District. The low GVI values in these areas can likely be attributed to insufficient green coverage, high building density, and limited public green spaces. Notably, Gongshu District and Binjiang District display the highest concentration of hot spots, indicating relatively well-developed green infrastructure. Conversely, parts of the Shangcheng District and Xihu District exhibit significant cold spots, suggesting that the planning of green spaces in these regions requires further improvement.

3.2. Equity Evaluation of Street-Level Greenery

Table 2. Gini coefficients and 95% CIs in Hangzhou and its four districts.

Area	Gini Coefficient	95% CI
Hangzhou (Overall)	0.384	0.352–0.415
Xihu District	0.467	0.409–0.518
Shangcheng District	0.317	0.247–0.386
Gongshu District	0.373	0.331–0.418
Binjiang District	0.268	0.218–0.313

presents the Gini coefficient for street-level greenery in Hangzhou overall and for each district, along with their 95% confidence intervals. For the entire city, the Gini coefficient was 0.384 (95% CI: 0.352–0.415). At the district level, the Gini coefficient was 0.467 (95% CI: 0.409–0.518) for Xihu District, 0.317 (95% CI: 0.247–0.386) for Shangcheng District, 0.373 (95% CI: 0.331–0.418) for Gongshu District, and 0.268 (95% CI: 0.218–0.313) for Binjiang District.

As shown in Figure 6, in Hangzhou’s central urban area, the Gini coefficient for street green visibility reaches 0.384, reflecting a moderate level of inequality in the allocation of greenery resources among residents. As observed in the district-level data, there are notable discrepancies in the allocation of green visibility. Xihu District exhibits the highest level of inequality, with a Gini coefficient of 0.467, indicating a more concentrated distribution of greenery resources and lower greenery coverage in certain areas. In contrast, Binjiang District shows the lowest Gini coefficient of 0.268, reflecting a more equitable distribution of greenery resources and less disparity in residents’ access to green spaces. Shangcheng District (0.317) and Gongshu District (0.373) fall between these extremes. Generally, the closer a district’s Gini coefficient is to the diagonal line, the more balanced the distribution of greenery resources.

In light of these findings, to promote fairness in the urban ecological environment, it is essential to prioritize regions with unequal green visibility, such as Xihu District. This could be achieved through the expansion of greening infrastructure and the optimization of green space layouts, ultimately improving the quality of life for local residents.

As shown in Figure 7, The spatial variation of location entropy for street-level greenery in Hangzhou reveals clear heterogeneity. The location entropy of all valid sampled points ranges from 0.02 to 8.97, with a mean value of 1.90 and a standard deviation of 2.11. To interpret these values, location entropy values greater than 1 indicate areas where the availability of street-level greenery is higher than the citywide average, suggesting that residents in these areas enjoy better-than-average access to green visual resources. Conversely, values less than 1 signify areas where green visibility is below the average, meaning residents are likely to experience reduced exposure to greenery in their daily environments.

Based on the interpolated values from all sampled points, the spatial distribution of location entropy for GVI in Hangzhou exhibits substantial variation across districts. Central urban areas within the study region are predominantly characterized by location entropy values lower than 1, revealing that spatial inequities in greenery distribution are more concentrated in the city center. In Xihu District, location entropy values are relatively high, indicating a concentrated distribution of green visibility in certain areas. In contrast, Gongshu District displays generally low location entropy values, suggesting a more even spatial distribution of greenery across the district. Binjiang District and Shangcheng District demonstrate notable internal variability, with areas of both high and low entropy values. These patterns reflect a complex spatial structure of greenery, with clusters of high entropy corresponding to major green spaces such as riverside corridors, parks, or plazas, while other parts of these districts show more limited green visibility.

In conclusion, the distribution of location entropy in Hangzhou reflects the diverse spatial organization of green resources across different districts. Areas with high location entropy values, such as the scenic zones in Xihu District, indicate that green resources are concentrated in limited areas, creating visually prominent but spatially clustered greenery. Conversely, lower location entropy values, as observed in Binjiang District, suggest a more even and balanced spread of greenery throughout the urban fabric. These differences highlight the uneven allocation of greenery benefits across the city’s neighborhoods. Building on this insight, future urban green space planning could incorporate location entropy analysis as a strategic tool to guide the placement and design of greenery interventions.

4. Discussion

4.1. Equity in the Distribution of Urban Street Greenery

This study emphasizes the importance of assessing the equity of greenery, particularly at the street level, where daily interactions between citizens and green infrastructure take place. From the results, the study area demonstrates a generally favorable level of greenery, with an average GVI of 18%, which is slightly higher than previous cities studied, such as the main urban area of Xuzhou, China (15.7%), and within the Sixth Ring Road in Beijing, China (17.1%) [56,57]. Nevertheless, it remains marginally below the levels observed in certain developed regions, including Hartford (22.8%) and Singapore (21.0%), suggesting potential for further improvement [23,29]. Notably, there is considerable spatial variability in the GVI. Areas with high GVI values tend to cluster near extensive green areas, including parks, while low-value areas are mainly distributed within central and commercial areas, where green space availability is reduced and spatially scattered. In terms of equity comparison, we compared Hangzhou’s Gini coefficient results with research findings from other cities (including other megacities in China or overseas). In Guangzhou—a comparable megacity in the Pearl River Delta—Gini values related to urban green space distribution were observed to be as high as 0.523 [58]. Hwang et al. used the Gini coefficient to measure the degree of urban greenery inequality between cities (London, New York, Paris, Tokyo, Seoul, and Beijing) [59]. The research results showed that the Gini value was the highest in Tokyo (0.37) and the lowest in London (0.22). Although these studies differ in data sources and methodologies, the comparative results indicate that Hangzhou performs better than some cities but still lags behind cities with the most equitable urban greenery environments. These comparisons aim to serve as a background reference, positioning Hangzhou within the broader discussion of urban green equity.

The noteworthy phenomenon is that the location entropy value is highest in Xihu District, where the concentration of green resources is also the most uneven, as indicated by the largest Gini coefficient. This spatial inequality arises from an imbalance between green space supply and demand, which plays a critical role in driving inequitable green space distribution [60]. On the one hand, Xihu District benefits from abundant park and scenic green spaces, which together provide a high overall level of street-level greenery. On the other hand, while high-density residential areas exhibit higher local green visibility, this benefit is diminished by the high population concentration and the mismatch between the supply of and demand for green services, negatively impacting residents’ greenery experience. Taking Wulin and Hubin as examples, these are high-density residential areas in Xihu District where population concentration is extremely high. Although they are located near the scenic West Lake area and surrounded by extensive greenways, the benefit of nearby greenery is diluted by the sheer number of residents. Therefore, high population density leads to diminishing marginal returns on services, which is a challenge currently faced by urban greening management. This spatial contradiction is deeply rooted in Hangzhou’s historical land-use zoning and planning trajectory. As one of the oldest and most developed districts in Hangzhou, Xihu District has long prioritized the preservation of ecological landscapes and scenic tourism zones. However, this legacy of land-use protection has led to strict zoning regulations around natural reserves and heritage sites, which limit the integration of green infrastructure into densely built residential zones. Consequently, green benefits are often spatially clustered but socially unevenly distributed. This phenomenon reflects a broader urban planning challenge: in high-density areas, green infrastructure must not only exist in large quantities but also be evenly accessible and effectively embedded within the residential fabric [61,62]. Addressing these issues requires rethinking the spatial logic of green planning beyond aggregate provision, toward demand-responsive and population-sensitive zoning strategies.

4.2. The Difference in Vertical and Horizontal Green Attributes

The Regulations on the Management of Urban Greening in Hangzhou mandate that newly built residential areas allocate no less than 35% of the total land area for green space, while piecemeal renovations in older urban areas require at least 25%. For industrial, mining, and large-scale construction projects—including public buildings, municipalities, and public facilities—the green space requirement is set at no less than 30%. However, several studies have indicated that these land use types do not always correlate positively with high GVI values [63]. In other words, a higher percentage of vegetative cover in land use does not necessarily result in a higher GVI as perceived from the street level.

There is a difference between people’s perception of street greenery and the broader sense of urban green coverage, which in the abstract is a difference between vertical and horizontal perspectives. Specifically, the former includes lawns, trees, shrubs, or green walls that residents see on a daily basis, while the latter is often assessed using metrics such as canopy cover or the NDVI [64,65]. While some cities may exhibit a high NDVI, reflecting abundant green space, their street-level greenery may still fall short, yielding a low GVI. This discrepancy highlights a critical issue: top-down assessments of urban vegetation, such as those based on NDVI, do not always align with the street-level greenery that residents experience. There is a clear mismatch between the spatial characteristics of the GVI observed from the street and the overall coverage of urban green spaces [14,66]. Consequently, while urban green spaces may appear abundant from an aerial perspective, they may not effectively contribute to the visual greenery that directly impacts residents’ daily experiences and well-being at the street level. Although the NDVI and land-use green space ratios provide valuable macro-level information on urban vegetation, they may mask local-scale disparities in green visibility. This study highlights the gap between perceived greenery at eye level and ecological indicators observed from above. Hence, the GVI is not intended to substitute for the NDVI, but to fill a critical perceptual gap in urban green equity research.

4.3. Optimization Strategies for Street Greenery Planning

In addition to improving spatial equity in greenery distribution, urban green infrastructure also plays a critical role in promoting social equity and environmental justice. Areas with low green visibility often overlap with socioeconomically disadvantaged communities, where residents may lack access to well-maintained public green spaces. For example, cities like New York have implemented initiatives such as the MillionTreesNYC program, which strategically increased tree canopy coverage in underserved neighborhoods [67]. Similarly, cities like Rotterdam and Seoul have introduced adaptive green planning to ensure that vulnerable groups—especially the elderly and children—can benefit from accessible greenery [59,68].

Firstly, with limited urban public forest resources, policymakers and urban planners should give priority to tree planting in neighborhoods with low tree cover over those with already high tree cover in order to rationally allocate resources [69]. Especially for areas with low greenery quality, such as the Science and Technology Park in Binjiang District and some old industrial plots in Gongshu District, the level of greenery planting should be optimized by selecting street trees with larger leaf areas and crown sizes to improve the greenery visibility per unit area.

Secondly, in view of the status quo of Hangzhou’s rich water system, some areas are disconnected from the greenery, and we should make full use of the water resources to create an urban greenway network with spatial connectivity and ecological permeability. For example, in the Binjiang District and other newly developed areas, street green belts need to be planned in advance to ensure a balanced distribution of greenery rates.

Finally, in old urban areas with high population density such as Shangcheng District and Gongshu District, we must make full use of marginal land and abandoned idle land to build pocket parks and increase small-scale green space [70]. Introducing green vegetation into vertical structures or walls (e.g., green walls, green façades, and vertical gardens) is widely recognized as an effective strategy to enhance the visibility and landscape appeal of street green infrastructure [71,72,73]. This type of vertical greenery not only helps to improve urban aesthetics, but also effectively increases green coverage in high-density urban environments where space is limited.

In summary, planners could reallocate street trees or pocket parks to enhance localized greenery. For instance, the northwestern segment of Gongshu District, which exhibits high residential density and low greenery visibility, could benefit from targeted tree planting along major corridors such as Xianghong Road. Similarly, underutilized plots near Xinyuan Lane in Shangcheng District may serve as ideal locations for pocket park development. This has been successfully demonstrated in other high-density cities like Seoul and Singapore, where micro-greening initiatives have effectively improved street-level ecological equity [74]. Integrating GVI-based evaluations into green space planning also allows for more responsive, data-driven decisions. By routinely monitoring GVI and correlating it with demographic density and land-use functions, municipal governments can adjust street tree planting strategies or revise zoning codes to ensure equitable access to green visibility across all districts.

4.4. Study Limitations and Future Works

This study is subject to several limitations. First, the analysis is based on BSV images, which may not uniformly capture all urban streets. Certain areas might contain outdated or incomplete image data, potentially compromising the accuracy of GVI estimations. Second, the research offers only a static snapshot of green visibility, without accounting for seasonal variation or long-term urban dynamics. Third, the road network density is notably lower compared to other urban districts in Xihu District, largely due to the presence of extensive natural and scenic areas such as the West Lake Scenic Area, Wu Chao Shan National Forest Park, and Xixi Wetland. These areas contain large water bodies and hilly terrains, which inherently limit road development and the availability of street view imagery. The GVI and spatial equity analysis in this district was therefore conducted using only the accessible roads with valid image data. While this approach preserves analytical consistency, it may affect the comparability of results across districts. Future research should incorporate multi-temporal imagery to monitor temporal changes in urban greenery. Although the GVI serves as a valuable metric for assessing perceived greenery, subsequent studies should investigate how residents interact with urban green spaces by integrating behavioral data and qualitative perspectives. Addressing these limitations will help establish a more holistic framework for evaluating the spatial distribution of urban greenery and its implications for environmental justice.

Regarding the study area, although this research is centered on Hangzhou, the proposed methodology is not geographically constrained and can be generalized to other urban contexts globally. With respect to the research data, the methodology utilizes publicly accessible and regularly updated street view images, which can be customized in future studies to capture images from various angles and at different times of day, thereby enhancing temporal and spatial flexibility. Furthermore, from a methodological perspective, the use of advanced image processing methods—like DeepLabV3+ for semantic segmentation—enhances the scalability and adaptability of the approach across diverse urban environments, irrespective of local terrain or infrastructure. Future research may benefit from employing more sophisticated computational methods and state-of-the-art segmentation models, both to improve the precision of object recognition and to tailor the analytical framework to identify a wider array of urban features based on specific research objectives.

5. Conclusions

This study sought to answer a key question: how equitably is street-level greenery distributed across Hangzhou? By addressing this, the research offers actionable insights for city planners and policymakers. Through utilizing BSV images and advanced semantic segmentation techniques, we assessed the GVI across various districts, revealing significant spatial variation in the visibility and distribution of greenery. The equity evaluation, using tools such as the Gini coefficient, Lorenz curve, and location entropy, highlighted disparities in access to green spaces, with certain districts, like Xihu District, showing higher inequality due to uneven green space distribution. Conversely, Binjiang District displayed a more balanced allocation of greenery resources. These results emphasize the need for targeted urban greenery interventions, particularly in areas with lower GVI values, to improve green infrastructure and promote environmental justice. By optimizing greening strategies, such as enhancing roadside tree planting and incorporating vertical green elements, urban planners can work toward a more equitable distribution of green spaces, thereby fostering a healthier and more sustainable urban environment.