1 Introduction
Urban street vitality is a key component of the quality of urban public space. However, the dominance of car-oriented urban development has significantly undermined street vitality, leading to a decline in pedestrian activities and social interactions in street spaces (
Brownson et al., 2009;
Jacobs, 1961;
Zeng et al., 2018). This, in turn, has weakened local economic development, increased social isolation, and contributed to adverse public health outcomes (
Chan et al., 2021;
Mouratidis and Poortinga, 2020;
Wu et al., 2023). In response, cities worldwide have introduced policies aimed at revitalizing street life. For example, the Netherlands’ “Living Street” initiative, based on the shared space concept, exemplifies efforts to redefine streets as vibrant spaces for social activities (
Ben-Joseph, 1995;
Pharaoh and Russell, 1991). To support the development of sound street policies, there has been an emerging academic interest in exploring the relationship between the built environment and street vitality.
Numerous studies have explored the effects of built environment on street vitality. For instance,
Sung et al. (2013) and
Seong et al. (2023) employed ordinary least squares regressions to support Jacobs’ assertion (
Jacobs, 1961) that small block sizes and mixed land use enhance street vitality.
Xia et al. (2020) demonstrated the critical roles of land use intensity, particularly building density and floor area ratio, in improving street vitality.
Jiang et al. (2022) applied hierarchical multiple regression models to reveal that proximity to public services and functional diversity exhibit a positive effect on street vitality.
X. Li et al. (2021) utilized spatial Durbin models to highlight the importance of street network integration in enhancing street vitality.
Zhao et al. (2023) employed geographically and temporally weighted regression to investigate the spatiotemporal heterogeneity of the effects of built environment on street vitality, identifying functional density, floor area ratio, bus station density, and street aspect ratio as key factors. Despite these advancements, most existing studies rely on linear modeling assumptions, which may oversimplify or misrepresent the relationship between the built environment and street vitality by neglecting potential nonlinear effects. Consequently, more sophisticated approaches are needed to better capture and explain these complex relationships.
Recent research has attempted to use machine learning to investigate the nonlinear relationship between the built environment and street vitality. For example,
Han et al. (2024) applied a gradient boosting decision tree (GBDT) model to street samples from 12 Chinese cities, revealing threshold effects in various built environment features. Their findings indicate that while smaller blocks generally enhance vitality, the negative impact of block size becomes more pronounced when it falls below 4 ha. Similarly, building density and floor area ratio positively influence street vitality but stabilize beyond 0.38 and 2.5, respectively, while the effect of the street aspect ratio shifts from positive to negative once it exceeds 3. Likewise,
Doan et al. (2025) employed an XGBoost model with street-level data from Manhattan, identifying similar nonlinear patterns: intersection density no longer contributes to enhance vitality beyond a certain threshold, excessive FAR negatively affects vitality, and functional diversity―measured by mixed point of interest (POI) values―significantly boosts street vitality when exceeding 0.4. These findings highlight the superiority of nonlinear models over linear ones in capturing complex urban dynamics. However, both studies rely on global models that assume spatial homogeneity, overlooking spatial non-stationarity. As a result, they only reflect global nonlinear relationships, masking local variations and interactions within different urban zones. To effectively foster vibrant streets, it is crucial to account for spatial heterogeneity, clarify the functional roles and challenges of distinct urban zones, and develop targeted strategies to enhance street vitality.
To address these gaps, this research aims to examine the local nonlinear relationship between the built environment and living street vitality. First, population heatmap data are collected to measure living street vitality. Second, built environment features―encompassing four dimensions: daily service facilities, built environment quality, morphological characteristics, and accessibility―are integrated into the spatial unit of the living street. A hybrid GBDT-based modeling framework is then employed to analyze the relationship between the built environment and living street vitality. This research makes three key contributions: (i) uncovering the local nonlinear effects of built environment features on living street vitality; (ii) identifying the global and local relative importance of the built environment features; (iii) offering insights for urban planners and policymakers to develop zone-specific strategies to enhance urban street vitality.
The remainder of this paper is structured as follows: Section 2 reviews existing literature to identify theoretical foundations and research gaps. Section 3 details the case study framework, study area characteristics, data sources, and modelling approach. Section 4 presents the results of the local nonlinear effects considering spatiality. Sections 5 and 6 provide discussion and conclusions, respectively.
2 Literature review
2.1 The measurement of street vitality
The concept of street vitality was introduced by Jane Jacobs in the 1960s, defining it as “street life over a 24-h period” and describing it as the “sidewalk ballet”―a dynamic and continuous interplay of human interactions occurring throughout the day in urban environments (
Jacobs, 1961). Montgomery further highlighted that the most prominent feature of street vitality is the flow of people engaging in diverse activities across a 24-h period, facilitated by various facilities (
Montgomery, 1998). Mehta expanded on this by emphasizing that highly vibrant streets typically host a significant number of people participating in regular or spontaneous social activities (
Mehta, 2007). Collectively, these scholars established street vitality as a phenomenon intrinsically tied to human activity, measurable through two key dimensions: (i) vitality intensity, reflecting the density of active individuals within a street space, and (ii) vitality stability, representing the temporal continuity of these activities over extended periods (
Ouyang et al., 2022;
Tang and Ta, 2022).
The measurement of street vitality requires reliable human activity data. Traditionally, field surveys have been the primary method, capturing pedestrian volumes and details about social activities like sitting, conversing, and shopping (
Istrate, 2025;
Sung and Lee, 2015). However, survey-based approaches are constrained by high labor costs, time-consuming processes, and limited data coverage. With advancements in information and communications technology, real-time big data, such as nighttime light data (
Xia et al., 2020), Location-Based Services (LBS) data (e.g., various mobile phone data) (
Jiang et al., 2022;
Lian et al., 2024;
Yu et al., 2024), social media data (
Q. Li et al., 2022), urban sensors (
M. Li et al., 2021), and street-view images (
Doan et al., 2025), enable more precise assessments of street vitality.
Numerous studies have used LBS data to evaluate street vitality. For instance,
X. Li et al. (2021),
Wei and Wang (2024), and
Yu et al. (2024) used Baidu population heatmaps to analyze the spatiotemporal distribution of population density, serving as proxies for both the intensity and stability of street vitality. Furthermore,
Y. Li et al. (2022) integrated multi-source LBS data, including bicycle-sharing trajectories, taxi orders, and Dianping user reviews, to measure multidimensional urban vitality at the street level.
Lian et al. (2024) applied Wi-Fi trajectory data to extract pedestrian volume and analyze trajectory diversity and complexity, providing an assessment of the vitality of pedestrianized commercial streets. Recent research has also explored the use of computer vision techniques applied to street-view images and video footage to estimate pedestrian volumes, offering an alternative means of measuring street vitality. As technological advancements continue to refine data collection methods, integrating multiple big data sources holds immense potential for enhancing the precision and comprehensiveness of street vitality assessments.
2.2 Relationship between built environment and street vitality
Previous research on the relationship between the built environment and street vitality has been conducted from various perspectives. Some studies have investigated the effects of the built environment on fostering street vitality based on Jacobs’ four conditions―small blocks, mixed land use, aged buildings, and density (
Gómez-Varo et al., 2022;
Seong et al., 2023;
Sung et al., 2013). More recent studies have adopted Ewing and Cervero’s 5Ds framework (Density, Diversity, Design, Distance to Transit, and Destination Accessibility) to explore the relationship between the built environment and street vitality (
Doan et al., 2025;
Jiang et al., 2022;
M. Li et al., 2021;
Wei and Wang, 2024;
Zhao et al., 2023).
As for density, functional density and building density are the most frequently examined variables in street vitality research. Numerous studies have reported a positive linear association between functional density and street vitality. For instance,
Istrate (2025) found that the density of small businesses significantly enhances the vitality of living streets. Similarly,
Lian et al. (2024) demonstrated that the concentration of flagship stores plays a crucial role in boosting pedestrianized commercial street vitality.
Wei and Wang (2024) identified a similar positive effect of functional density on historic and cultural streets. Many studies have also emphasized the positive effects of building density on street vitality.
Jiang et al. (2022) observed a strong linear relationship between building density and street vitality across multiple Chinese cities, a finding corroborated by
Xia et al. (2020). However,
Han et al. (2024) and
Doan et al. (2025) highlighted a threshold effect, noting that beyond a certain point, extremely high building density ceases to contribute to further vitality enhancement. The role of residents’ willingness and behavior in shaping street vitality in high-density urban environments also deserves emphasis. High levels of functional and building density tend to support broader social networks, more frequent social interactions, and stronger support from close relationships (
Mouratidis, 2018;
Mouratidis and Poortinga, 2020). These social dynamics are deeply intertwined with residents’ behavior and participation, which significantly influence how street vitality is experienced and sustained.
Regarding diversity, mixed land use and functional diversity (measured through the variety of points of interest, POIs) are the most frequently analyzed factors. Most studies emphasize a positive correlation between diversity and street vitality, aligning with residents’ behavioral preferences: people are naturally drawn to streets offering diverse amenities and services (
E. Chen et al., 2021;
Wang et al., 2014).
Seong et al. (2023) found that mixed land use significantly enhances street vitality, while
Sung et al. (2013) used a residential-to-non-residential balance index to quantify mixed land use and identified a similar trend, aligning with
Jacobs’ (1961) original claims.
Jiang et al. (2022) reported a positive linear relationship between functional diversity and street vitality, a finding echoed by
Yu et al. (2024) and
Wei and Wang (2024), both of whom focused on historic and cultural streets.
Doan et al. (2025) further explored the nonlinear effects of functional diversity, revealing that mixed POI density has a minimal impact on street vitality until it surpasses a critical threshold of 0.4, after which its influence increases sharply.
In terms of design, street length (or block size), sidewalk width, and green view ratio are the most commonly examined variables in street vitality research. Most studies report a negative linear relationship between street length and vitality. This can be partially explained by the higher frequency of resident interactions in small blocks.
Sung et al. (2013) and
Sung and Lee (2015) observed that smaller blocks encourage more frequent resident contact and stronger social interaction, which helps reinforce local culture and place identity, especially in historic and cultural zones (
Huang et al., 2023). This cultural continuity fosters a stronger sense of place, which in turn contributes to enhanced vitality (
Istrate and Chen, 2022).
Istrate (2025),
Seong et al. (2023), and
Wei and Wang (2024) found that shorter street lengths enhance the vitality of living streets, supporting Jacobs’ assertion (1961) that small blocks foster vibrancy. However,
Han et al. (2024) noted that while smaller blocks generally promote vitality, the negative impact of excessive block size reduction becomes more pronounced when the block area falls below 4 ha. Similarly, numerous studies have documented a positive linear relationship between sidewalk width and street vitality (
Istrate, 2025;
M. Li et al., 2021;
Lian et al., 2024;
Wei and Wang, 2024). As for green view ratio,
Lian et al. (2024) found a positive correlation between greenery and street vitality, whereas
Zhao et al. (2023) and
Jiang et al. (2022) reported contradictory findings.
Yu et al. (2024) further indicated that the influence of green view ratio varies across different urban zones, while
M. Li et al. (2021) demonstrated that its impact fluctuates with seasonal changes.
With respect to distance to transit, proximity to bus stations is the most frequently studied factor. Numerous studies have identified a negative relationship between the distance to bus stations and street vitality, suggesting that closer proximity to transit significantly enhances street vibrancy (
Seong et al., 2023;
Wei and Wang, 2024). However,
Doan et al. (2025) further explored the nonlinear effects of bus station density on street vitality, demonstrating that its positive impact on vitality only becomes significant once a critical threshold is reached.
Regarding destination accessibility, street network connectivity and accessibility to large public service facilities are among the most commonly examined indicators. Street network integration is frequently used to assess pedestrian and cycling accessibility. Many studies have identified a positive correlation between destination accessibility and street vitality (
Jiang et al., 2022;
M. Li et al., 2021;
X. Li et al., 2021), likely because residents in areas with greater destination accessibility are more willing to engage in street activities (
Mouratidis and Poortinga, 2020). Improved destination accessibility and walkability also contribute to stronger social cohesion (
Mazumdar et al., 2018). The theory of natural movement further supports this view, suggesting that residents’ travel behavior is strongly influenced by the directness and connectivity of street networks (
X. Li et al., 2022). Nevertheless,
Yu et al. (2024) suggested that improving destination accessibility does not universally enhance street vitality; its impact may be limited to specific urban zones.
Table 1 provides an overview of some key studies that have significantly advanced our understanding of the relationship between the built environment and street vitality. However, several limitations warrant further investigation. First, the influence of built environment factors may vary across different street types, yet limited research has focused on living streets (
Istrate, 2025). The vitality of living streets is particularly critical for residential well-being and community cohesion, making it essential to recognize how these environments interact with street vitality dynamics. Second, most existing studies rely on traditional linear regression models, overlooking potential nonlinear (threshold) effects. While prior research on human activity and the built environment has widely acknowledged nonlinear relationships, such effects remain underexplored in street vitality studies (
L. Chen et al., 2021;
Liu et al., 2024a;
Yan and Chen, 2024;
Yang et al., 2021). On one hand, built environment characteristics may need to reach a certain threshold before significantly influencing vitality (
Han et al., 2024). On the other hand, assuming a continuous linear relationship is unrealistic, as marginal effects may emerge―excessively high building density, for instance, may lead to overcrowding and constrained street spaces, ultimately reducing pedestrian activity (
Doan et al., 2025). Future research should focus on identifying the effective ranges of built environment variables that maximize street vitality. Finally, while some studies consider the spatial heterogeneity of built environment effects, research incorporating both nonlinear assumptions and spatial heterogeneity remains scarce. Empirical evidence suggests that street vitality exhibits strong spatial clustering and undergoes dynamic shifts influenced by residents’ behaviors (
Chen et al., 2022;
Wu et al., 2024;
Zhang et al., 2021;
Zhao et al., 2023). The factors driving street vitality may vary across locations, and some variables could exert entirely different effects at the local scale. Furthermore, the uneven distribution of public amenities and resources in large cities leads to substantial variations in the importance of different built environment features. This suggests that threshold effects may exhibit spatial heterogeneity, necessitating a deeper examination of their localized nonlinear influences on living street vitality.
3 Material and methodology
We developed a geographically weighted gradient boosting decision tree (GW-GBDT) to examine the nonlinear relationships between the built environment and living street vitality. Figure 1 illustrates the analytical framework. First, multi-source data were collected to measure living street vitality and built environment features. Then, we constructed the GW-GBDT model, incorporating spatial weights to account for spatial heterogeneity, to explore the local nonlinear associations by assigning varying importance to built environment indicators. Finally, the relative importance of indicators was compared, and their local nonlinear effects on living street vitality were analyzed. Data and methods are detailed in the following subsections.
3.1 Study area
Xiamen, located in southeastern Fujian Province, China, is a sub-provincial city and a special economic zone, serving as a key central city along China’s eastern coast (Fig. 2(a)). Shaped by historical evolution and rapid urbanization, Xiamen Island has gradually developed into the city’s core urban area, characterized by high population density, intensive mixed land use, and significant spatial development constraints. Since 2010, there has been no newly available land for sale on the island, according to government reports. To alleviate the growing tension between land demand and supply, the city launched urban renewal projects, targeting underperforming or inefficiently used areas. These efforts have included the demolition of dilapidated buildings to free up space for redevelopment, improving living conditions, and increasing urban density, though often at the expense of green space (
Tang et al., 2013).
By 2022, Xiamen Island’s population had reached 2.06 million, within a land area of 132.20 km
2 (
Zhao et al., 2023). Separated by sea from surrounding districts, the island maintains a relatively independent spatial and functional structure. It has developed a diverse land-use structure, featuring a mix of residential, commercial, industrial, historical, and financial functions (Fig. 2(b)). As economic development and living standards have improved, public demand for high-quality living environments has grown. Enhancing the vitality of living streets has emerged as a key strategy for improving livability and public well-being. The Island’s complex and well-connected street network fosters a dynamic urban environment (
Chen et al., 2022;
Wu et al., 2024;
Zhao et al., 2023), making it a suitable case study for exploring the relationship between the built environment and street vitality.
Revealing the relationship between the built environment and living street vitality on Xiamen Island also offers valuable insights for revitalizing the vibrancy of core urban areas of other Chinese cities, particularly those characterized by high population densities and limited developable land. Cities such as Shanghai, Guangzhou, and Shenzhen exemplify this pattern: their core urban areas concentrate dense populations and intensive economic activity, driving a sustained demand for high-quality urban living environments while contending with spatial constraints. In response, these cities are advancing urban regeneration and micro-renewal initiatives to enhance the livability and vitality of their urban cores―for instance, Shanghai’s “15-min community life circle” (
Wu et al., 2021), the renewal of old community in Guangzhou (
Li et al., 2019), and the redevelopment of urban villages in Shenzhen (
Jiang et al., 2024). In this context, Xiamen Island’s strategies for enhancing living street vitality under spatial constraints present a representative and transferable model, offering both theoretical value and policy relevance.
This study focuses on the living streets. First, highways and arterial roads were removed from the street network, and the remaining streets were segmented at intersections. Then buffer zones were generated along the street centerline, with a radius of 50 m for secondary roads and 35 m for branch ones, effectively encompassing the main facilities distributed along both sides of the streets (Fig. 2(b)). Next, streets segments without daily service facilities, identified based on POI data, were excluded. As a result, 893 segments of living streets with corresponding spatial units were obtained (Fig. 2(a)).
3.2 Data sources
As shown in Fig. 1, this study employed six types of data, including population heatmap data, street view images, building data, road network data, POI data, and resident population data. Population heatmaps were obtained through the Baidu Map API, comprising a total of 224 heatmaps collected for weekdays (Tuesday and Thursday) and weekends (Saturday and Sunday) in August 2022, with hourly snapshots recorded from 7:00 to 21:00. These heatmaps were processed using ArcGIS software to calculate the hourly population density for each street spatial unit. Street view images were obtained through the Baidu Map API, captured panoramic images at 40.00 m intervals, with elements such as sidewalks, roads, sky, and vegetation identified using the Deeplab-V3+ semantic segmentation tool and the Cityscapes dataset to calculate area proportions. Building data, obtained through the Baidu Map API, included data on 49,620 individual buildings, detailing the number of floors, vector outlines, and residential attributes (determined in conjunction with POI data). Road network data was obtained from Google Earth, comprising information on 2,545 streets, including names, classifications, lengths, and speed limits. POI data was obtained from Amap, encompassing various types of functional facilities, with a total of 156,204 entries across eight categories: daily life services, public transit stations, financial services, office facilities, catering services, shopping services, educational and cultural facilities, and healthcare services. Resident population data was obtained from WorldPop, an open-source spatial demographic dataset open spatial demographic data). This study employed WorldPop data with a resolution of 100 m per pixel to estimate the resident population for each street spatial unit.
3.3 Methodology
3.3.1 Vitality variables
Existing studies commonly characterize street vitality through the spatiotemporal distribution of population density in the spatial units of streets (
Han et al., 2024;
Jiang et al., 2022;
C. Wu et al., 2023;
W. Wu et al., 2023;
Zhao et al., 2023). Given the significant influence of resident population distribution on the baseline population density of living streets, this study employs the ratio of the average transient population density to the resident population in the street buffer zone to define the intensity of living street vitality (
ITS) (Eqs. (1) and (2)). This approach aims to mitigate the bias introduced by variations in resident population density. Additionally, the stability of living street vitality (
STB) (Eq. (3)) is incorporated, utilizing the coefficient of variation to analyze fluctuations in living street vitality (
Ouyang et al., 2022;
Tang and Ta, 2022), thereby enabling a comprehensive assessment of the spatiotemporal dimensions of living street vitality.
where V represents the value of living street vitality, X denotes the transient population density in the living street buffer zone calculated from Baidu heatmaps, and S signifies the resident population density within the street buffer zone. Vi represents the vitality value of street i at hourly intervals from 7:00 to 21:00 on four weekdays and four weekends in August 2022, with n being the total number of hourly intervals. A higher ITS value indicates a greater intensity of living street vitality, while a lower STB value reflects a higher stability of living street vitality.
3.3.2 Built environment indicators
Building on the key built environment features mentioned in Section 2.2, as well as the foundational work of
Jacobs (1961) and the well-established 5Ds framework (
Cervero et al., 2009), we systematically selected 16 widely recognized built environment indicators as independent variables, categorizing them into four distinct dimensions: accessibility, built environment quality, daily service facilities, and morphological features. These indicators were not only grounded in theoretical and empirical research but were also chosen for their relevance to the available datasets and their applicability to the unique characteristics of living streets. Table 2 presents the details of the definitions and calculation methods for these indicators.
Regarding accessibility, proximity to shopping malls (
He et al., 2024) and subway stations (
Sung et al., 2013) were employed to quantify the locational advantages of streets and residents’ ease of access to essential facilities. In assessing built environment quality, five key indicators were identified: floor area ratio (
Xia et al., 2020), building density (
Lu et al., 2019), residential balance index (
Sung et al., 2013), green view ratio (
Tang and Ta, 2022), and sky view factor (
Jiang et al., 2022), which collectively characterize the environmental quality of living streets. In terms of daily service facilities, four indicators were selected through dual perspectives of density and diversity: daily service facility density (
Seong et al., 2023), daily service facility diversity (
Zhao et al., 2023), commercial facility diversity (
Y. Li et al., 2022), and bus stop density (
Jiang et al., 2022). These indicators comprehensively measure both the accessibility and variety of neighborhood service provisions. Concerning morphological features, five essential indicators were identified to capture street design and spatial configuration: number of road lanes (
Sung et al., 2013), speed limit (
Jiang et al., 2022), street length (
Han et al., 2024), sidewalk percentage (
Lian et al., 2024), and street integration as measured through space syntax (
He et al., 2024).
3.3.3 The hybrid model GW-GBDT
The geographically weighted regression (GWR) model has been empirically validated as an effective tool for addressing spatial autocorrelation and non-stationarity, with demonstrated applications across environmental governance, urban planning, land use analysis, and public health research (
Wu et al., 2024;
Zhao et al., 2023). By embedding spatial coordinates into regression parameters, GWR enables location-specific coefficient estimation and mapping, while incorporating distance decay effects to calibrate localized regression relationships. Gradient boosting decision tree (GBDT) algorithm employs an ensemble learning approach through sequential tree construction to predict continuous outcomes. By utilizing negative gradients for weight updates, GBDT effectively captures nonlinear relationships among variables (
Han et al., 2024;
Wang et al., 2024). Compared with linear regression, GBDT requires no predefined variable relationships and accommodates multicollinearity. Furthermore, it provides interpretable outputs through variable importance rankings and partial dependence plots (PDPs). However, conventional GBDT implementations neglect spatial heterogeneity when processing large-scale geospatial datasets.
To address the limitations of GBDT in handling spatial non-stationarity and GWR in modeling nonlinear associations, this study proposes a hybrid geographically weighted gradient boosting decision tree (GW-GBDT) framework, with the computational workflow illustrated in Fig. 3. The GW-GBDT synergistically integrates GWR’s spatial weighting mechanism with GBDT’s nonlinear learning capacity by incorporating geographic coordinates (X, Y) of living street spatial units as input features. This integration enables the global model to be refined into localized sub models through adaptive distance-threshold-based weighting (see Eq. (3) in Fig. 3). A kernel-adaptive bandwidth selection ensures sufficient sample sizes for each localized regression while maintaining spatial specificity.
The GW-GBDT model was implemented using the “gbm” package in R. Key parameters were optimized as follows: the maximum number of trees was set to 10,000, the shrinkage parameter to 0.001, and five-way interactions were specified to capture complex relationships. To prevent overfitting, five-fold cross-validation was employed during model training. Through multiple boosting iterations, the model generated reliable outputs for each living street spatial unit at optimal bandwidths, effectively balancing model complexity and predictive performance.
To further interpret the nonlinear relationships captured by the GW-GBDT model, partial dependence plots were utilized to analyze the threshold effects of built environment indicators on both the intensity and stability of living street vitality. Specifically, we identified the two most influential built environment indicators from each dimension―accessibility, built environment quality, daily service facilities, and morphological features―resulting in a total of eight key indicators based on their relative importance scores. To systematically explore their nonlinear trends and thresholds, K-means clustering was applied, with the optimal number of clusters determined using the elbow method. This approach enabled a more precise identification of critical threshold values, providing deeper insights into the complex interplay between built environment features and living street vitality dynamics.
Overall, compared with the standard GBDT framework, the proposed GW-GBDT model is better suited for handling large-scale geospatial datasets. GW-GBDT integrates a geographically weighted mechanism into the GBDT structure by applying adaptive spatial kernel functions. This allows GW-GBDT to capture spatially varying relationships between built environment features and human activities, which are often obscured in the global models used in conventional GBDT. Consequently, unlike conventional GBDT models that typically provide only global variable importance and global partial dependence plots (
Han et al., 2024), GW-GBDT extends these analyses to the local scale. This offers greater transparency for urban planners aiming to understand context-sensitive drivers of street vitality. Additionally, the GW-GBDT framework enhances result reliability by reducing the subjective biases commonly associated with resident-based vitality assessments. Rather than relying on perception-based surveys (
Istrate, 2025), the model derives insights objectively from the high-resolution built environment and population data, supporting more evidence-based decision-making. Together, these methodological enhancements justify the use of GW-GBDT in this study and provide a new perspective for analyzing the complex relationship between the built environment and street vitality.
4 Results
Given that the regression results for intensity and stability of living street vitality show no significant differences between weekdays and weekends, the subsequent analysis will focus on weekdays as representative cases to examine the impact degree and nonlinear effects of the built environment on the living street vitality.
4.1 Global relative importance
Table 3 presents the global relative importance of 16 indicators across four dimensions in shaping the intensity and stability of living street vitality. The results reveal that daily service facility density and floor area ratio emerge as the most critical indicators of the intensity of living street vitality (ITS), each contributing over 11.00% of relative importance. Meanwhile, daily service facility diversity and green view ratio demonstrate predominant influence on the stability of living street vitality (STB), both exceeding 9.00% in relative importance. Five indicators―distance to shopping mall, distance to subway stations, floor area ratio, street length, and street integration―demonstrate significant associations with both ITS and STB, consistently ranking within the top seven influential factors. Conversely, speed limit, number of motorized lanes, and public transit stop density exhibit the least impact, ranking lowest in importance. Among the four dimensions, the built environment quality scores the highest in cumulative importance, while transportation accessibility shows the strongest average importance across indicators.
4.2 Local nonlinear effects considering spatiality
4.2.1 Daily service facilities
The daily service facility density (
Dd) demonstrates a positive correlation with
ITS and
STB, consistent with the Jacobs’ observations that higher functional density enhances street vitality (
Jiang et al., 2022;
Q. Li et al., 2022). Notably, the high-vitality density thresholds exhibit significant spatial heterogeneity across different urban zones, with lower thresholds in residential and business-financial zones (0.0007 units/m
2, Street ①), moderate thresholds in historical feature and former industrial zones (0.0009 units/m
2, Street ②), and higher thresholds in core commercial zones (0.0013 units/m
2, Street ③), as illustrated in Fig. 4(a). This may be attributed to the fact that the core commercial zones require a diverse range of services to meet varied consumer demands, resulting in a higher concentration of street-level daily service facilities to attract residents. Figure 4(b) shows that while the daily service facility density has a limited effect on
STB, it significantly influences
ITS, particularly in residential, financial, and former industrial zones. Therefore, increasing the proportion of retail and service facilities on both sides of the living streets in these functional zones will effectively enhance living street vitality.
The daily service facility diversity (
Hd) exhibits positive correlation with both
ITS and
STB, aligning with the finding that functional diversity fosters street activities (
Sung et al., 2013). Notably, its positive effect on
ITS is most significant in southeastern residential zones (Fig. 5(b)), where
Hd score that promotes
ITS mostly falls between 0.70 and 1.30 (Fig. 5(a), Street ①). In contrast, its influence on
STB is more widespread, reaching its peak in core commercial zones. The critical
Hd score for rapidly increasing
STB is between 1.00 and 1.50 (Fig. 5(a), Streets ③ and ④). This finding further corroborates that a diverse mix of street-facing businesses (with an
Hd score of at least 1.00) effectively sustains living street vitality. However, there is an upper limit score to the benefits of
Hd―when it exceeds 1.50, no further increase in both
ITS and
STB is observed. Therefore, a well-calibrated increase in the diversity of street-facing daily service facilities is crucial for maintaining highly vibrant living streets.
4.2.2 Built environment quality
Floor area ratio (
FAR) positively affects the
ITS and
STB. In core commercial and southwestern historic feature zones (Fig. 6(b)), the value of
ITS and
STB shows an irregular stepwise increase as
FAR rises. When the value of
FAR reaches 5.00 (Fig. 6(a), Street ①), both
ITS and
STB stabilize at a high level. In contrast, the impact is less pronounced in business-financial zones, residential zones, and former industrial zones, where both
ITS and
STB surge once the value of
FAR surpasses 1.20, with the stabilization threshold generally below 3.80 (Fig. 6(a), Street ②). These findings confirm that higher
FAR fosters stable and highly vibrant living streets (
Seong et al., 2023), yet excessively high
FAR does not yield further increases in both
ITS and
STB. Therefore, in the redevelopment of living streets, the value of
FAR of street-facing buildings should not exceed 5.00 in core commercial zones and should remain below 3.80 in other functional zones.
The green view ratio (GVR) predominantly exerts a negative effect on both ITS and STB, with the critical influence range falling between 12.00% and 20.00%, where increased GVR leads to a rapid decline in ITS and STB (Fig. 7(a), Street ①). However, in historic features zones, GVR has a positive effect, with a beneficial range of 7.50%—15.00% (Fig. 7(a), Street ②). This may be attributed to the relatively low GVR in these zones, where high-quality greenery and a lower street aspect ratio enhance shading, thermal comfort, and visual appeal, thereby fostering street activities. In contrast, in negatively impacted zones, although GVR is high, unappealing greenery fails to support street vitality. Thus, green space design should prioritize quality over quantity and incorporate well-planned resting areas that facilitate social interactions, thereby enhancing the positive effects of greenery on street vitality.
4.2.3 Morphological features
The effects of street length (
Len) on the
ITS and
STB show no significant spatial variation (Fig. 8(a)). However, its effects on ITS and
STB are opposing. When Len exceeds 0.12 km, ITS declines rapidly, while
STB increases significantly. Once Len reaches 0.32 km, living streets enter a state of low but stable vitality (Fig. 8(b)). These findings underscore the importance of adopting a small-block, dense road network approach in the development of living streets (
Han et al., 2024).
Street integration (
NAIN) primarily shows positive correlations with both
ITS and
STB (Fig. 9). In historic features zones, central residential zones, and business-financial zones, its influence is nearly linear, with
ITS and
STB rising rapidly within the
NAIN score of 0.60—1.20 (Fig. 9(a), Street ①), suggesting that improved pedestrian accessibility attracts more street activities. In core commercial zones, a U-shaped relationship is observed, where vitality remains high and stable when the
NAIN score is either below 0.70 or above 1.40 (Fig. 9(a), Street ②), indicating that both lower and higher levels of pedestrian accessibility are favorable. Conversely, in former industrial zones and southeastern residential zones,
NAIN negatively affects vitality, with a sharp decline occurring within the score of 0.60—1.30 (Fig. 9(a), Street ③), likely due to a lack of pedestrian-friendly infrastructure. Previous studies have highlighted that leisure-oriented pedestrian activity is highly sensitive to
NAIN, but it also requires supportive conditions such as a sufficient sidewalk-to-road ratio and a continuous street network (
Yuan and Chen, 2021). Therefore, measures such as eliminating dead-end streets, connecting long secondary roads, and opening adjacent residential areas can enhance
NAIN and, consequently, boost living street vitality.
4.2.4 Transport accessibility
Distance to subway stations (Dis_S) negatively affects the ITS and STB. When the distance exceeds 0.50 km, ITS and STB decline sharply and stabilize beyond 0.80 km (Fig. 10(b)), indicating a clear influence radius of subway stations. Additionally, the impact follows a concentric attenuation pattern, weakening from the center toward the periphery (Fig. 10(a)). This finding reflects the high dependence of residents in central residential zones and core commercial zones on rail transit, where ITS and STB are significantly influenced by subway access. Therefore, within the 0.50—0.80 km radius of subway stations in these areas, efforts should be made to refine the pedestrian connection system to enhance accessibility and support living street vitality.
Distance to shopping malls (Dis_C) shows a negative correlation with both ITS and STB. As shown in Fig. 11(b), streets within 0.60 km of a shopping mall attract more pedestrian flow, whereas the influence becomes negligible beyond 1.20 km. In terms of ITS (Fig. 11(a)), the emerging eastern urban districts remain highly dependent on shopping malls, with ITS strongly affected by their proximity. Meanwhile, in the southwestern old town, where shopping malls are concentrated, their influence is more pronounced in maintaining STB. Thus, within the 0.60—1.20 km service radius of a shopping mall, efforts should focus on enhancing the spatial quality of living streets to better accommodate pedestrian activity and sustain street vitality.
5 Discussion
5.1 Understanding local nonlinear relationships between the built environment and living street vitality
Our findings are generally consistent with existing studies that demonstrate a strong correlation between the built environment and street vitality. However, past research has primarily relied on linear regression models, focusing on identifying which built environment indicators (BEIs) have significant positive or negative effects. In contrast, our study reveals that the relationship between the built environment and street vitality is nonlinear, meaning the effects of BEIs are not simply positive or negative, but occur within specific effective ranges. More importantly, this nonlinear relationship exhibits significant spatial heterogeneity, which has not been adequately captured by existing studies using global nonlinear models.
Regarding daily service facilities, the density and diversity of such facilities are key predictors of street vitality. This study supports previous findings that higher daily service facility density and diversity are associated with greater street vitality (
Istrate, 2025;
Jiang et al., 2022;
Li et al., 2022;
Lian et al., 2024;
Sung et al., 2013;
Wei and Wang, 2024). However, we also observe an upper threshold for the benefits of facility diversity. Once this threshold is exceeded, further increases in diversity do not lead to additional gains in street vitality. This finding aligns with human preferences, as people are generally attracted to streetscapes with a diverse range of services and amenities (
Chen et al., 2021), although excessive visual complexity beyond a certain threshold may no longer enhance their attractiveness.
For built environment quality, both the floor area ratio (
FAR) and green view ratio (
GVR) significantly contribute to street vitality, in line with previous research. This study confirms earlier findings that, while increased
FAR is positively associated with vitality, once
FAR reaches a certain point, excessive development intensity no longer contributes to further increases in street vitality (
Doan et al., 2025;
Han et al., 2024). This is reasonable, as overly high development intensity can lead to overcrowded and compressed street spaces, which are less conducive to outdoor activities (
Xia et al., 2020). As previous research has shown,
GVR tends to negatively influence street vitality in most functional areas (
Han et al., 2024;
Jiang et al., 2022;
Li et al., 2022;
Wei and Wang, 2024;
Zhao et al., 2023). However, our study finds that in historical districts, increased
GVR correlates with increased street vitality.
In terms of morphological features, street length and street integration (
NAIN) exert the most significant influence on street vitality. This study supports Jane Jacobs’s observations (
Jacobs, 1961) that smaller blocks (i.e., shorter streets) are associated with higher levels of street vitality. However, this finding contradicts the results revealed by
Zhao et al. (2023) and
Istrate (2025); the former reported a positive relationship between street length and vitality, while the latter found no statistically significant effect. These conflicting results may stem from a failure to account for nonlinear effects. For street integration (
NAIN), existing global nonlinear studies have found that higher
NAIN values, representing better walkability, are generally positively associated with street vitality (
Doan et al., 2025;
Han et al., 2024). However, our study reveals that the nonlinear effects of
NAIN on street vitality exhibit significant spatial heterogeneity, showing opposite effects across different functional zones.
Regarding transport accessibility, in line with previous studies, shorter distances to subway stations and shopping malls are associated with higher street vitality (
Doan et al., 2025;
Seong et al., 2023;
Zhao et al., 2023). Additionally, our study identifies a distinct radius of influence for both subway stations and shopping malls, demonstrating the nonlinear effect of transport accessibility on street vitality.
More importantly, uncovering the nonlinear and spatially heterogeneous effects of the built environment on street vitality can support a transition from generic to precision-optimized enhancement strategies. By identifying critical thresholds and the local relative importance of dominant BEIs, this study proposes targeted strategies for enhancing street vitality across five distinct functional zones in Xiamen Island. These strategies include recommended indicator values and prioritized improvements (see Table 4). For instance, in residential zones, increasing the Hd score to 1.00 and optimizing pedestrian connectivity within a 0.50—0.80 km radius of subway stations are crucial for fostering a walkable and livable environment that supports both the stability and intensity of street vitality.
5.2 Policy implications for vitality-oriented urban design
The findings of this study provide valuable insights for urban planning and renewal practices in Chinese cities and may also inform other cities facing space constraints and a growing need for high-quality urban environments. In recent years, compact city development, characterized by high density and mixed land use, has been promoted throughout China (
Jiang et al., 2025). However, relying solely on compact land use and urban morphology cannot guarantee vibrant urban environments. This study offers several implications for urban renewal practices in this regard.
This study confirms that increasing development intensity can enhance street vitality, while excessively high development intensity does not lead to further improvement. Some studies have found that overly high development intensity may even harm urban vitality by contributing to deteriorating environmental and socioeconomic conditions (
Doan et al., 2025;
Wu et al., 2024). Therefore, enhancing urban vitality should not rely solely on continuously increasing development intensity. Instead, it is important to understand the effects of the built environment on vitality at the human scale, within the limits of current development intensity, with particular attention to street-space transformation and small-scale community renewal. Our findings support that a fundamental strategy for improving vitality is to increase the density and diversity of daily service facilities along living streets. When residents have more opportunities to access a variety of daily services within a reasonable distance, they are more likely to engage in street-level activities, which in turn enhances the vibrancy of urban spaces (
Boessen et al., 2018;
X. Li et al., 2022). This approach aligns with the concept of the “15-min community life circle” (
Song et al., 2024), which emphasizes improving residents’ convenient access to daily services.
The Transit-Oriented Development (TOD) model, which encourages commercial and mixed-use development around transit stations to boost functional diversity, is an effective approach to enhancing walkability and vitality in compact cities (
Guo et al., 2024;
Wan et al., 2025). However, most existing urban layouts and land use plans often limit the TOD’s practical implementation. This study finds that transit stations significantly influence the vitality of nearby living streets, with those closer to stations exhibiting higher vitality. Therefore, improving street network connectivity and pedestrian accessibility within transit station catchment areas, particularly through better “last-mile” connectivity (
Chakraborty et al., 2025), provides a practical and effective way to enhance urban vitality.
The positive impact of small urban blocks on urban vitality is well recognized, as smaller blocks often encourage more interpersonal interactions in street spaces (
Jacobs, 1961). Our study also confirms that shorter street segments are conducive to increased street vitality. However, it is important to note that without supportive conditions such as a pedestrian-friendly environment, small blocks may not be sufficient to cultivate vibrant street life (
Gan et al., 2021). In China, the recent “Opening up of gated communities” policy aims to break down barriers in enclosed residential compounds and improve walkability (
Wu et al., 2018). This study echoes the rationale of that policy, suggesting that improving pedestrian and cycling accessibility, particularly through moderate increases in street network integration, can contribute to enhanced street vitality. These efforts, however, must be supported by well-designed walking and cycling infrastructure, such as higher proportions and better continuity of sidewalks and bike lanes (
X. Li et al., 2021).
5.3 Limitations and prospects
This study develops a hybrid GW-GBDT framework to quantify the nonlinear effects of spatial heterogeneity, offering new perspectives on context-specific strategies for enhancing street vitality. However, several limitations remain.
First, although Xiamen Island was selected as a representative case to derive generalizable conclusions, the findings are inevitably influenced by environmental and contextual differences. The conclusions are more applicable to high-density Chinese cities or other East Asian urban environments with similar characteristics. Future studies should expand the sample size to further validate the model’s applicability across diverse contexts.
Second, interactions between variables may influence threshold effects, particularly when interaction effects are strong. Future research should refine the model to improve its ability to capture these complex mechanisms. Additionally, while the GW-GBDT model, based on a distance threshold method, provides localized insights into the nonlinear relationships between variables and street vitality, future studies could enhance model interpretability by incorporating techniques such as inverse distance weighting and shapley additive explanations (SHAP).
Moreover, street vitality is a broad concept that cannot be fully represented by population spatial distribution alone. Future research could leverage physiological sensor data to examine how the built environment affects different age groups, supporting more precise and targeted vitality planning.
Finally, environmental factors such as pollutant exposure and thermal comfort may moderate the relationship between built environment variables and street vitality. However, these factors were not controlled for in the current model. Future studies should incorporate these variables to provide a more comprehensive understanding of the mechanisms influencing street vitality.
6 Conclusion
Although existing research generally acknowledges that vibrant street spaces are closely associated with residents’ mental health, social interactions, and community cohesion, limited attention has been paid to the spatial heterogeneity and nonlinear variation of the effect of built environment. This study addresses this gap by uncovering threshold effects and spatial heterogeneity in the relationship between the built environment and living street vitality, thereby offering more refined empirical evidence and contributing a new methodological perspective to the discourse on urban livability. Additionally, this study introduces an improved measurement of street vitality by integrating population heatmaps with both transient and residential population density. This approach captures the spatiotemporal fluctuations of street vitality and enhances the accuracy of the vitality assessment.
Specifically, using Xiamen Island as a case study, this research quantifies the intensity and stability of living street vitality based on population heatmap data. A hybrid geographically weighted gradient boosting decision tree (GW-GBDT) framework is proposed and applied to assess the local nonlinear effects of the built environment on living street vitality across four dimensions: transportation accessibility, built environment quality, daily service facilities, and morphological features.
Unlike previous studies that mainly relied on linear assumptions (
Istrate, 2025;
Y. Li et al., 2022;
Seong et al., 2023;
Sung et al., 2013;
Xia et al., 2020), our findings reveal a clearly nonlinear relationship between built environment indicators and both the intensity and stability of street vitality. From a global perspective, a higher density of daily service facilities (
Dd ≥ 0.0013 units/m
2) and strong street-facing development intensity (
FAR ≤ 3.80) are crucial for stimulating street vitality, significantly enhancing the intensity of living street vitality. Meanwhile, a moderate level of daily service facility diversity (
Hd = 1.0—1.50) and an optimal green view ratio (
GVR = 7.50%—15.00%) are essential for maintaining the stability of living street vitality. Additionally, proximity to subway stations (
Dis_S = 0.50—0.80 km) and large commercial zones (
Dis_C = 0.60—1.20 km), a fine-grained street network (
Len = 0.12—0.32 km), and high pedestrian accessibility (
NAIN = 0.60—1.20) are all key contributors to both intensity and stability of living street vitality.
Furthermore, our findings not only support previous evidence of spatial heterogeneity in the effects of built environment indicators on street vitality across different urban functional zones (
Zhao et al., 2023), but also advance this understanding by capturing localized nonlinear relationships through the proposed GW-GBDT hybrid framework. By incorporating an adaptive distance-threshold-based weighting method, this framework effectively identifies the relative importance and nonlinear effects of built environment indicators at the local scale. This approach overcomes key limitations of traditional global nonlinear models (
Doan et al., 2025;
Han et al., 2024), which are often inadequate for capturing regional variation. The results reveal that the spatial heterogeneity of dominant built environment factors is manifested in their nonlinear responses, directional influences, and effective ranges, underscoring the need for context-specific strategies to enhance living street vitality across diverse functional zones.
2095-2635/2025 The Authors. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd.
PDF
(6930KB)
