1 Introduction
Urban vitality represents the comprehensive manifestation of regional social attractiveness and sustainable development potential (
Mouratidis and Poortinga, 2020). High-vitality metro station areas can enhance public transit service efficiency, minimize transportation energy consumption, and improve public life (
Chen et al., 2022). Among the various elements that constitute metro station area vitality, pedestrian activity is one of the most critical components (
Delclos-Alio et al., 2019), with persistent high-density pedestrian flows serving as its essential foundation (
Ouyang et al., 2022). In pedestrian-oriented environments, individuals engage in diverse social and recreational interactions, which further catalyze commercial economic activities, enhance regional employment opportunities and quality of life indicators, thereby forming a virtuous cycle that strengthens urban vitality (
Liu and Shi, 2022).
Based on the close association between urban vitality and pedestrian activity, walkability can be viewed as a key dimension affecting the vitality of metro station areas (
Sun et al., 2016). Since its inception, the Transit-Oriented Development (TOD) model has maintained an intrinsic connection with walkability, with its fundamental characteristics including high-density development, mixed land use, and pedestrian-centered street design (
Calthorpe, 1993). These characteristics collectively create an environment conducive to pedestrian activity, thereby promoting vitality in metro station areas. However, in practical implementation, many metro station areas have failed to meet pedestrian mobility demands due to economic incentive mechanisms and car-centric planning approaches, resulting in insufficient vitality. As rail transit plays an increasingly important role in urban transportation systems, the need to enhance urban vitality in metro station areas has become particularly urgent (
Pan et al., 2017;
Sung et al., 2014). Consequently, measuring walkability and examining its influence on metro station area vitality has become a crucial component in formulating effective urban planning and development strategies.
Despite extensive research on walkability and vitality, limitations remain in the current literature. First, numerous studies employ the 5D frameworks to measure walkability in metro station areas (
Ewing and Cervero, 2010). However, these frameworks mainly focus on functional provision and service accessibility while neglecting the influential impact of public spaces and building form on pedestrian experience (
He and He, 2023). Although scholars have attempted to address this limitation by incorporating additional metrics, inconsistent classification standards have resulted in limited comparability across studies (
Kang, 2018). Second, recent research often relies only on big data to measure urban vitality (
Wu et al., 2023;
Xiao et al., 2021), which inadequately reflects vitality from the pedestrian activity perspective (
Chen and Yuan, 2024). While big data sources such as GPS and social media check-ins can provide patterns of human presence, they fail to capture the nuanced behavior of pedestrians. This methodological constraint leads to a misunderstanding of the associations between vitality and walkability. Third, while existing TOD research provides valuable insights (
Bivina et al., 2020;
Niu et al., 2021), it has yet to fully address the differential modification potential within context-specific built station areas. This requires examining nonlinear threshold effects and synergistic interactions, where strategic combinations of interventions may amplify the positive impacts on station area vitality.
To bridge these gaps, this study investigates walkability characteristics and their influence on urban vitality across 274 metro station areas in the Shanghai central area. We address the first limitation by employing a comprehensive walkability evaluation framework that integrates urban morphology research hierarchies across four dimensions: land use, street networks, public spaces configuration, and building form. To overcome the second limitation regarding vitality measurement, the research combines field-collected pedestrian density data with multi-source big data (metro ridership, smartphone positioning, and social media check-ins) to measure metro station area vitality from a pedestrian perspective. Finally, to address this gap, we employ GBDT with SHAP interpretation to capture complex nonlinear relationships and interaction effects between walkability and station vitality. This machine learning approach identifies critical intervention thresholds and amplification effects from strategic element combinations, enabling us to generate station-specific optimization pathways based on each station’s unique environmental context.
The contributions of this research are as follows: First, it enhances cross-study comparability by focusing directly on physical spatial elements through the integration of walkability measurements with urban morphology research hierarchies, while incorporating public space and building form dimensions related to pedestrian experience. Second, it explores urban vitality measurement methods from the pedestrian’s perspective. By integrating field survey walking activity data with multi-source big data, this study achieves a pedestrian-oriented vitality assessment at the station area scale. This method addresses the scalability constraints of field surveys and the insufficient capture of pedestrian activities by big data. Third, it establishes differentiated optimization strategies for TOD urban construction through interpretable machine learning. The nonlinear analysis identifies threshold ranges and reveals interaction effects that offer renewal pathways for existing station areas, combined with local effect interpretation to achieve context-specific interventions.
2 Literature review
2.1 Measurement of walkability
In urban planning and public health research, the measurement of walkability has often relied on the “3D” framework―Density, Diversity, and Design―first conceptualized by
Cervero and Kockelman (1997). This framework was subsequently expanded into the “5D” framework with the addition of Destination accessibility and Distance to transit dimensions (
Ewing and Cervero, 2010). Contemporary walkability studies typically operationalize these dimensions by examining building density, land use mix, street connectivity, destination density, and distance to station (
Cheng et al., 2019;
Jiang et al., 2021).
However, many empirical studies suggest limitations in the 5D framework for comprehensive walkability assessment. Walkability encompasses not merely the functional ability to reach diverse destinations within a reasonable time but also the experiential quality of the walking itself (
Doaa and Ahmed, 2021). Recent literature demonstrates that pedestrian behavior is significantly influenced by public space and building form, which shape the perception of safety and environmental attractiveness, beyond mere travel efficiency considerations (
Basu et al., 2022;
Distefano et al., 2023;
Mateo-Babiano, 2016). More specifically, the limitation of the 5D framework lies in its conflation of walking experiential factors into the over-simplified “design” dimension, while neglecting crucial spatial elements that shape the pedestrian experience―such as public space and building form.
Though some research endeavors aim to expand upon the 5D framework’s indicators, the integration of experience-related factors into this framework has been inconsistent across studies, thereby reducing cross-study comparability. For instance, the GSI, commonly classified as a “density” indicator to quantify development intensity, simultaneously functions as a “design” parameter by influencing spatial enclosure and visual interest (
Gehl, 2010;
Jacobs, 1961).
He and He (2023) aptly note that aggregating disparate elements, such as public space and building under the “design” dimension, obscures their distinct contributions to walkability. Even when proposing the 5D framework,
Ewing and Cervero (2010) pointed out that there was an ambiguous and unsettled definition, and believed that these boundaries might change in the future to cope with the overlaps among dimensions.
To address these limitations, we refine the walkability assessment framework to prioritize pedestrian experience-related factors through a more nuanced treatment of public space and building elements. Drawing from previous walkability research (
Doaa and Ahmed, 2021;
Dovey and Pafka, 2020;
He and He, 2023;
Lucchesi et al., 2023) and urban morphology research framework (
Elzeni et al., 2022;
Oliveira, 2016;
Zhang et al., 2023), this research measures walkability in four fundamental components: land use, street network, public space, and building form. This reconfiguration has several advantages: First, it provides analytical granularity by decomposing the formerly monolithic “design” dimension into discrete morphological elements (street network, public space, and building form), enabling more precise examination of design-related walkability factors. Second, it facilitates direct translation between research findings and practical design interventions by linking walkability metrics to specific physical environmental elements that can be manipulated through design and policy instruments. Third, establishing classification criteria based on objective physical entities rather than abstract dimensions enhances methodological consistency and improves comparability across diverse research contexts.
2.2 Measurement of urban vitality
The concept of urban vitality, initially proposed by
Jacobs (1961), encompasses rich and diverse street life and social activities.
Gehl (1971) subsequently identified that public spaces providing a distinct “sense of place” constitute essential elements for urban vitality. Further,
Maas (1984) attributed urban vitality to three key factors: dense pedestrian flows, high facility utilization rates, and diverse street activities. Classical urban studies have consistently defined urban vitality as the public and social activities emerging from pedestrian concentrations. With the evolution of urban research, the conceptualization of urban vitality has expanded to incorporate economic prosperity and population density indices (
Chen et al., 2022;
Jin, 2007). However, pedestrian activities remain the fundamental component, with pedestrian activity significantly contributing to urban economic development, environmental quality enhancement, and social cohesion (
Giles-Corti et al., 2014;
Lucchesi et al., 2023;
van den Berg et al., 2022).
Academic interest in urban vitality assessment methods has increased significantly in recent years. However, current approaches still demonstrate limitations in accurately capturing urban vitality from the pedestrian perspective. Traditional methods have relied heavily on governmental geographic data and quantitative approaches. Researchers have implemented GPS positioning data (
Delclos-Alio et al., 2019;
Feng et al., 2019) and transportation ridership (
Jeong and Woon, 2020) to measure urban vitality. However, Baidu Heatmaps aggregate multiple transportation modes, including pedestrians, drivers, and delivery personnel, without distinction, while metro ridership data excludes local residents and individuals using alternative transportation modes such as bicycles or private vehicles. The emergence of Volunteered Geographic Information (VGI) has expanded the data sources available for urban vitality measurement.
Goodchild (2007) conceptualized citizens as sensors. This paradigm has enabled new forms of data collection, with
Liu et al. (2015) demonstrating how social media data, as a form of VGI, enables “social sensing” of urban environments through platforms like Weibo check-ins. At the same time,
Haklay (2010) found that VGI quality can match authoritative datasets in certain contexts. Specifically, VGI data is primarily uploaded by specific groups of internet-active users, which may introduce bias in urban vitality measurement (
Pan et al., 2021). From the pedestrian perspective, field survey data through direct pedestrian counting provides the most accurate representation of urban vitality (
Gehl, 2010). This methodology presents considerable challenges regarding time and resource requirements when implemented across extensive urban areas.
To address these constraints, this study establishes urban vitality indicators by integrating multi-source big data with pedestrian density collected through field surveys. Previous studies have demonstrated the validity of correlating emerging urban big data sources with traditional survey methodologies (
Singleton and Runa, 2021;
Yang et al., 2022). By synergizing field-collected data with multi-source big data, this approach effectively resolves the constraints of inaccuracies in big data and the scalability limitations of conventional field survey methodologies (
Ye et al., 2023). This integrative methodological framework has been successfully implemented across diverse research domains (
Azcarate et al., 2021;
Jestico et al., 2016).
2.3 Relation between walkability and vitality
Extant literature has examined linear relationships between walkability and urban vitality, with primary emphasis on land use and street network. Those studies demonstrate that high-density mixed land use and interconnected street networks serve as key determinants of urban vitality, while giving insufficient attention to public spaces and building form, which are closely related to walking experience. The subsequent discussion elucidates the relationships between urban vitality and walkability from four aspects: land use, street network, public space, and building form.
Land use diversity and density emerge as important aspects to promote urban vitality across multiple studies. Several studies demonstrate that mixed-use areas generate more walkable trips and potentially more people on the streets, thus fostering higher urban vitality (
Garau and Alfonso, 2022;
Lima et al., 2021).
Wu et al. (2023) found that the Floor Area Ratio (FAR) significantly enhances urban vitality, underscoring the importance of high building capacity in creating dynamic and walkable environments. Similarly,
Mouratidis and Poortinga (2020) suggests that functional diversity in land use effectively strengthens social connections and enhances walkability, aligning with
Jacobs’ (1961) perspectives. The presence of essential services is a key component of livability; suitable land use patterns usually mean greater vitality and walkability, linked to better quality of life and health (
Carmona, 2019;
Frank et al., 2010). Additionally,
Xia et al. (2020) observed that increased diversity in land use and Points of Interest (POIs) directly contributes to vibrant urban areas.
Street network close related to pedestrian route choice and walking distance, dense street network, and shorter street lengths can create a high vitality area (
Pakoz et al., 2022). A well-designed road network facilitates pedestrian activity, thus improving urban vibrancy―a viewpoint originally proposed by
Jacobs (1961). Furthermore,
Ye et al. (2018) identified a significant correlation between street centrality and urban vitality by analyzing network accessibility. However, the role of motor vehicles in walkability and vitality remains debated.
Doan et al. (2024) found that motor vehicle-induced air pollution negatively affects urban vitality, yet they also observed that increased motor vehicles can enhance vitality in some contexts. Conversely,
Buehler and Hamre (2015) argue that more motor vehicles tend to detrimentally impact walkability and urban vitality by reducing pedestrian accessibility and mobility.
The quantity and quality of public spaces serve as critical catalysts for urban vitality, particularly in metro station areas.
Zapata-Diomedi et al. (2019) compared amenities (e. g., daily living destinations, transit) and greenfield urban areas in terms of activity intensity, finding that green spaces have a greater attraction for walking activities.
Ottoni et al. (2016) similarly support this view, emphasizing the role of public spaces in promoting vitality. Research has primarily focused on parks and squares, often assessing their impact through POI density or green coverage rates (
Xiao et al., 2021;
Xia et al., 2020). These studies highlight that abundant public spaces promote urban vitality by providing walkable destinations and fostering pedestrian activity.
Building forms significantly shape the pedestrian experience. A higher GSI is considered a key factor in determining urban vitality. For example, using building form analysis with GSI values, studies have shown that different street block types with varying building densities influence urban vitality, with general trends of increasing vitality with higher GSI values (
Berghauser Pont and Haupt, 2010).
Jiang et al. (2021) observed that increased building density supports walkability by attracting pedestrians. However,
Ye et al. (2018) and other researchers have found that even when accounting for the GSI, the positive impact of building form types on vitality remains significant and multifaceted.
Despite offering valuable insights, most studies have relied on linear models to examine the relationship between walkability and urban vitality. This approach overlooks the potential for non-linear and complex interactions among walkability indicators. For instance, commuters may tolerate suboptimal walking environments out of necessity, suggesting that the relationship between walkability and vitality in some situations isn’t a simple linear correlation. Furthermore, once basic walking needs are satisfied, additional improvements to the walking environment may yield diminishing returns regarding metro station area vitality. Additionally, failing to consider the synergistic relationships among walkability indicators can obscure critical differences between station environments, leading to oversimplified comparisons. For example, high-density single-function areas may appear equivalent to low-density mixed-use areas (
Dovey and Pafka, 2020), despite their differing mechanism of influence on vitality.
To bridge these gaps, machine learning models were employed to analyze the effect of walkability on metro station area vitality. Compared to linear models, machine learning models can effectively capture complex nonlinear relationships and identify the synergistic interactions, providing a more comprehensive understanding of the walkability effect on metro station area vitality.
3 Materials and methodology
3.1 Research framework
This research examines the relationship between walkability and vitality in metro station areas through three main steps: data preprocessing, model selection, and data analysis (Fig. 1). First, walkability and vitality data from 274 metro stations in Shanghai’s central area were collected and preprocessed. Then, machine learning algorithms were employed to establish a relationship model between walkability and metro station area vitality. Finally, SHAP and PDP methods were used to interpret the machine learning model results, revealing how walkability elements influence metro station area vitality.
3.2 Study area
Shanghai is one of China’s iconic international metropolises, characterized by high population density that has necessitated extensive rail transit development. The Shanghai Metro system ranks among the world’s largest transit networks in terms of operational rail length and ridership volume. Approximately half of Shanghai’s residents rely on rail transit as their primary mode of daily transportation, with peak daily passenger volumes exceeding 13 million. Despite this heavy reliance on public transit, the built environment surrounding Shanghai’s metro stations often prioritizes commercial interests and vehicular accessibility while neglecting pedestrian-friendly design. This imbalance represents a widespread challenge across numerous high-density global cities, such as Delhi and Seoul (
Bivina et al., 2020;
Kang, 2019).
Consequently, Shanghai is an ideal case study for examining pedestrian environment issues in metro station areas within high-density urban contexts. The research encompassed 274 metro stations across Shanghai’s central area. For each station, 500-m buffers were generated around all entrances and merged to form a single station area (Figs. 2 and 3). These buffer zones for each station were subsequently merged to create metro station area research units. The 500-m threshold was selected as it represents a 5 min walking distance for pedestrians, aligning with Shanghai’s central area metro station spacing of approximately 1 km.
3.3 Data collection and preprocessing
3.3.1 Walkability data collection
Pedestrian activities can be categorized into two fundamental requirements: mobility needs and stationary activity needs. Mobility needs are influenced by destination attractiveness and path accessibility, while stationary activity needs relate to public space provision and environmental experience quality. Based on these requirements, this research measures the walkability through four dimensions: Land Use, Street Network, Public Space Configuration, and Building Form.
The Land Use dimension examines the density and diversity of functional destinations that generate pedestrian trips through service facility provision. It evaluates how service function capacity and density potentially stimulate pedestrian activity(
Frank and Pivo, 1994). Metro passengers require high-capacity buildings and sufficient service facilities to support daily activities within station areas. Land Use data includes building area, block functions, and POI. To measure construction and service facility capacity, we employ FAR and POI Density indicators. The spatial distribution of POI across different categories is presented in Fig. 4. Additionally, functional diversity ensures rail passengers can access various services within walking distance. We assess this diversity through Mixed Land Use and Mixed POI Diversity indicators, which quantify the diversity of urban functions available to pedestrians.
The Street Network dimension evaluates spatial connectivity and route directness that facilitate pedestrian access and movement convenience within station areas. Dense, well-connected networks minimize pedestrian detours and facilitate direct access to destinations. Street network data comprises road centerlines and broad outlines. We measure network accessibility through three key indicators: road density, average street length, and street centrality. Street centrality is calculated using Spatial Design Network Analysis (sDNA) tools, which evaluate how each street segment connects to surrounding segments in the network (
Ye et al., 2018). Additionally, appropriate motorway width must balance basic vehicular needs while protecting pedestrians from pollution and noise exposure, which we measure using the motorway width indicator.
The Public Space Configuration dimension assesses how public spaces’ supply and spatial arrangement fulfill pedestrians’ needs for social and leisure activities. Research confirms that public space design affects pedestrian lingering, interaction, observation, and sense of place (
Gehl, 1971). Public space data involves the location and scale of squares, parks, and pedestrian malls, which derive from the Area of Interest (AOI). “Scale” and “quantity” are critical measures of public space provision. Appropriately sized public spaces foster urban life―spaces too small cannot accommodate social gatherings, while oversized spaces may feel impersonal. Spatial distribution also matters: multiple, dispersed spaces improve accessibility for diverse groups, while fewer, concentrated spaces may create beneficial clustering effects. We measure public space provision using area and quantity density, and evaluate placement by measuring walking distances from stations to public spaces. We analyze squares, parks, and pedestrian malls separately due to their different functions and relationships with travel demands.
The Building Form dimension analyzes the visual qualities of building volumes, which shape pedestrians’ spatial experience and environmental perception. Research indicates that building form directly influences street enclosure and landscape continuity, affecting pedestrians’ comfort and sense of place (
Gehl, 2010). Building Form data includes building footprints and heights. Compact building arrangements create well-defined street spaces with continuous facades, measured by GSI. Appropriately scaled buildings avoid overwhelming pedestrians, measured by average building ground floor area and height. Variation in building heights creates visual interest, but excessive differences can disrupt street coherence. We measure building harmony using standard deviations of building ground area and height.
3.3.2 Vitality data collection
This study integrates field survey data with multi-source big data to measure metro station area vitality. Field surveys provide accurate pedestrian activity measurements but are limited in spatial coverage, while big data enables large-scale assessment but may lack granularity for pedestrian-level analysis. We therefore use field data to calibrate big data sources for vitality indicator development. Four types of data were acquired to measure metro station area vitality.
Pedestrian Density: Field surveys were conducted at 28 metro station areas selected from 274 stations in central Shanghai. Station selection followed a three-step process: K-means clustering of stations based on walkability indicators with four optimal clusters identified through elbow method and silhouette coefficient analysis; candidate station screening within each cluster by selecting stations within 1.5 standard deviations of cluster centroids using Mahalanobis distance, followed by additional filtering criteria including removal of stations with substantial undeveloped land areas, exclusion of areas under urban renewal, and prioritization of stations with high ridership levels; dynamic selection of 7 stations per cluster with representativeness validated through Kolmogorov-Smirnov tests and variance coverage analysis. After selection, the pedestrian counts were collected at street midpoints within 28 station areas during April‒June 2023 and April‒June 2024, covering one weekday and one weekend day from 7:00 to 21:00.
Metro Ridership: Shanghai Shentong Metro Group provided ridership data for one weekday and one weekend in 2023, comprising 21.5 million transaction records.
Smartphone Positioning: Location data were obtained from Baidu heat maps for one week in 2023, collected daily from 7:00 to 21:00, yielding 3.26 million spatiotemporal data points.
Weibo check-in data were collected for one week in 2023. Raw records contained user IDs, timestamps, coordinates, content, and device information. Corporate and government accounts were excluded based on comment content and device signatures, retaining only individual user check-ins within study areas, yielding 8862 valid records.
The distribution of vitality-related data is presented in Table 1.
3.3.3 Data preprocessing
The data preprocessing phase involves several steps to ensure analytical accuracy. First, walkability indicators normalization to standardize values, facilitating cross-indicator comparability. In the vitality data integration phase, a weighted multiple linear regression model is used to relate pedestrian density to multi-source data, based on field survey coefficients from 28 representative stations. Spatial unit matching is then conducted to ensure consistency across walkability and vitality data within 500-m buffer zones. Outlier treatment is applied using the 3σ principle to identify and handle extreme values, ensuring data quality. Finally, a multicollinearity test is performed using VIF<10 to exclude highly correlated variables, reducing bias in the model.
The calculation methods and descriptive statistics of the above data are detailed in Appendix. A1.
3.4 Methodology
3.4.1 Machine learning model selection
In analyzing the effect of walkability on metro station area vitality, this study employed three machine-learning models: Gradient Boosting Decision Trees (GBDT), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). Before applying these models, independent variables were selected using the Variance Inflation Factor (VIF<10) to mitigate multicollinearity. The dataset was randomly divided into a training set (70% of observations) and a testing set (30%). The performance of the GBDT, XGBoost, and RF models was compared using 5-fold cross-validation alongside the Optuna tool. The GBDT model demonstrated the best predictive performance with parameters: n_estimators = 105, learning_rate = 0.029, max_depth = 8, and min_samples_leaf = 10, achieving a test set R2 of 0.797 and RMSE of 1421.83 (Table 2). The model stability was validated through 5-fold cross-validation with consistent performance across folds (Appendix. A2). Consequently, the GBDT model was utilized to determine the walkability effects on metro station area vitality.
3.4.2 SHAP and PDP-based interpretation
To enhance the interpretability of our machine learning models, we employed both SHAP (Shapley Additive exPlanations) and Partial Dependence Plots (PDP). SHAP analysis quantified individual walkability indicators’ contributions to model predictions through Shapley values, providing feature importance rankings and revealing non-linear relationships between indicators and metro station area vitality. Complementing this, PDP visualizations captured the marginal effects of indicators and their interactive influences, offering insights into synergistic effects that individual feature analysis might overlook. Beyond overall interpretations, we specifically aimed to identify critical thresholds in the relationships between walkability and vitality.
This research employed segmented linear regression combined with curvature analysis to identify nonlinear thresholds in SHAP dependency plots. Zero-crossing thresholds represent points where feature effects shift from inhibitory to promotional, identified through linear interpolation at the intersection of fitted functions and the zero axis. Saturation thresholds represent points where positive effects begin to diminish, identified through numerical curvature analysis to detect inflection points. Statistical stability was validated through 500 bootstrap resamples, retaining only thresholds with detection rates >80% and coefficients of variation <0.2.
Bootstrap resampling was employed to evaluate the robustness of nonlinear relationships and synergistic effects, with nonlinearity quantified through partial dependence plot (PDP) analysis across 50 grid points per feature, requiring mean CV<0.2 across grid points and inter-curve Spearman correlation >0.8 between bootstrap samples, with synergistic effects quantified through Friedman’s H-statistic measuring the proportion of variance in joint effects. Statistical reliability was established through H-statistic(>0.05) and coefficient of variation (<0.2) across bootstrap iterations.
4 Result
4.1 Shanghai metro station areas’ vitality
Figure 5 illustrates the calculation of metro station area vitality. Our analysis focused on 274 metro stations across Shanghai’s central area. To address the relationship between pedestrian density and proxy indicators, we examined correlation coefficients between pedestrian density and three data sources: metro ridership, smartphone positioning, and Weibo check-in data. These correlations ranged from 0.53 to 0.72 (Appendix. A3), with all VIF values below 10, confirming acceptable multicollinearity among variables. We developed a weighted Multiple Linear Regression (MLR) model that achieved an R2 of 66.37% (Appendix. A4), resulting in the following formula: metro station area vitality index = 0.009 × Metro Ridership +0.496 × Smart Phone Position +1021.107 × Social Media Check-in.
The spatial distribution of urban vitality indicators reveals a distinct pattern characterized by high vitality in central and western areas, contrasting with lower vitality in peripheral and eastern regions. This distribution pattern aligns with Shanghai’s historical development trajectory and corroborates findings from previous research (
Shi et al., 2021, 2023;
Yue et al., 2019).
4.2 Relative importance of walkability indicators
This study ranked variable importance using mean SHAP values to identify key walkability indicators influencing metro station area vitality. As illustrated in Fig. 6 and Table 3, land use indicators emerged as the most significant contributor, accounting for 61.09% of the influence effect. Public space configuration ranked second with a 19.72% contribution. Street network indicators followed in third position, contributing 11.46%, while building form indicators accounted for 7.73% of the influence effect.
In the land use dimension, catering services make the most significant contribution to vitality at 27.16%. Four of the top 5 most important indicators relate to land use, confirming that high capacity and mixed functionality are primary determinants of metro station area vitality. Corporate services demonstrate notably lower importance in stimulating vitality compared to other facility types. While high-density corporate services indicate office area aggregation, increased street activity also requires diverse services and public spaces that attract office workers for dining, rest, and socialization during breaks. Mixed POI demonstrates far greater importance than mixed land use, suggesting that distributing diverse service functions across various blocks within metro station areas proves more effective than concentrating different functional buildings in specific blocks.
In the street network dimension, average street length and street centrality contribute 4.86% and 3.53% respectively, emphasizing accessible and convenient road networks for street activities. Longer streets diminish urban function connectivity, forcing pedestrians to travel extended distances to access different services. The significantly higher importance of average street length compared to street density likely stems from terrain configurations, water systems, and planning influences resulting in non-linear street networks. This creates environments with high street density but longer street segments, making average street length a more accurate reflection of the pedestrian walking distance.
In the public space configuration dimension, squares and pedestrian malls contribute significantly to metro station area vitality, while parks demonstrate relatively lower impact. The number of square spaces proves considerably more important than their size, as dispersed multiple squares provide entrance areas and connecting corridors for surrounding buildings, enhancing spatial richness and visual interest along walking paths. Pedestrian mall characteristics strongly influence station area vitality, with size, distance, and number contributing 5.46%, 2.39%, and 1.45% respectively, underscoring the importance of coordinated development between public transport facilities and pedestrian-oriented commercial spaces. Park-related indicators show minimal impact on urban vitality (contributions below 1%), potentially due to varying traveler needs. For commuters, densely vegetated and winding parks may impede efficient pedestrian movement, while large urban parks near stations might force commuters to detour. Conversely, for residents and tourists, parks offer attractive venues for leisure and social activities.
Regarding building form, the ground floor area standard deviation significantly impacts station area vitality, contributing 3.19% and ranking tenth in importance. GSI, average ground floor area, and weight building height show moderate influence, each contributing between 1.2% and 1.5%. Ground area standard deviation measures building volume harmony on the horizontal dimension, with higher values indicating greater urban texture heterogeneity. In this study, heterogeneous urban texture primarily results from large-scale commercial complex development. High urban texture heterogeneity increases the likelihood that pedestrians will encounter imbalanced street scales and fragmented urban landscapes.
4.3 Non-linear effects of walkability indicators on metro station area vitality
Examining whether walkability indicators maintain consistent influence on metro station area vitality or demonstrate varying effects across different value ranges is essential for optimizing walkability. This analysis employs dependence scatter plots to reveal the nonlinear impacts of key walkability features on vitality. Due to space constraints, we present the three indicators from each dimension that demonstrate pronounced nonlinear effects.
4.3.1 Land use and metro station areas vitality
Figures 7(a)—7(c) show that in the land use aspect, FAR, catering services, and Mixed POI all have positive impacts on the vitality. Figure 7(a) indicates that when FAR is below 1.3, urban vitality is suppressed (SHAP value is negative). In the range of 1.3—2.0, there is a slight positive effect, and between 2.0 and 2.5, the impact on urban vitality significantly increases. However, after exceeding 2.5, its influence remains at a stable level and does not continue to enhance. This suggests that higher building capacity can promote walking activities, but once the necessary capacity is met, further capacity increases have limited promotional effects. Figure 7(b) shows that catering services have a negative impact on urban vitality when below 100. Between 100 and 200, the impact fluctuates between positive and negative effects, and when exceeding 200, the positive impact is significantly enhanced. This result effectively supplements the research by
Wu et al. (2023). Figure 7(c) illustrates that Mixed POI has a significant negative impact when increasing from 0 to 1.6, stabilizes between 1.6 and 1.8, and the positive impact effect significantly enhances after exceeding 1.8. Additionally, the synergistic interactions between financial services and these indicators also exhibit positive impacts on urban vitality.
4.3.2 Street network and metro station areas vitality
Figures 7(d)—7(f) illustrate the relationship between the street network and metro station area vitality. Figure 7(d) shows that when the average street length is below 0.2, there is a significant positive effect on pedestrian activities. In the range of 0.2—0.4, the negative impact significantly increases, and when exceeding 0.4, the inhibitory effect on pedestrian activities gradually weakens. Shorter street segments imply more intersections and path choices, while longer street segments lead to pedestrian detours. However, as the street segment length further increases, the deterrent effect on pedestrians no longer continues to increase, possibly because, in this value range, street segment length is no longer the main factor influencing pedestrian travel decisions. Figure 7(e) shows that when street centrality is within the range of 0—200, it has a suppressive effect on pedestrian activities. When it exceeds 200, it significantly promotes vitality. This result is consistent with existing studies, indicating that high street centrality promotes an increase in pedestrian activities (
Stavroulaki et al., 2019). Additionally, catering services and street centrality can synergistically promote pedestrian activities. Figure 7(f) shows that motor width promotes pedestrian activities when it is within the range of 15—20. This indicates that motorways that are too narrow cannot meet the necessary traffic demands, while motorways that are too wide increase high-speed non-destination motorized traffic, thereby disrupting pedestrian activities. When motorways are too wide, high-street centrality can effectively mitigate their negative impact on vitality.
4.3.3 Public space configuration and metro station areas vitality
Figures 7(g)—7(i) illustrate the relationship between public space configuration and metro station area vitality. Figure 7(g) shows that when the square number is below 2, the impact on vitality is relatively low. Between 2 and 4 squares, the positive impact becomes significant, and above 4 squares, the promotion of vitality exhibits a leap-like change. Additionally, when the size of squares is large and the number of squares is small, it indicates that having only concentrated large-scale squares in a station area has a negative impact on vitality. Overly expansive squares exceed the range of human visual and psychological comfort and may disrupt the existing street network, reducing the connectivity of urban spaces. Figure 7(h) shows a weak association between the supply and layout of parks and vitality. Figure 7(i) shows that when the distance between pedestrian malls and metro stations is between 0 and 150 m, there is a promoting effect on pedestrian activities, and this impact linearly weakens as the distance increases. When the distance exceeds 150 m, the effect becomes slightly negative, but the negative impact does not increase further with distance. Additionally, when pedestrian malls are closer to metro stations, and there are more of them, their promoting effect on the vitality of station areas is most significant.
4.3.4 Building form and metro station area vitality
Figures 7(g)—7(i) illustrate the relationship between building form and metro station area vitality. Figure 7(j) shows that when the GSI is between 0 and 0.25, there is an inhibitory effect on pedestrian activities. In the range of 0.25—0.35, the positive impact sharply increases. An increase in GSI results in more street-facing building spaces, enhancing the enclosure of pedestrian areas and creating opportunities for social and recreational activities. However, when GSI exceeds 0.35, its impact on pedestrian activities shifts back to an inhibitory effect. This indicates that overly dense building coverage reduces public open spaces, thereby decreasing the likelihood of pedestrian activities. Figure 7(k) shows a promoting effect on urban vitality when the standard deviation of the ground floor area is below 500, which indicates that pleasantly scaled and harmonious building volumes effectively promote urban vitality. Figure 7(l) demonstrates that when the average building height exceeds 20 m, the impact on urban vitality changes from negative to positive. One explanation is that it necessitates vertical growth of buildings to meet the essential demands of urban life. However, when the average building height exceeds 25 m, this positive impact begins to weaken. This may be due to excessively tall, point-like buildings failing to provide the necessary enclosure for streets and potentially causing a sense of oppression for pedestrians.
4.4 Synergistic effects of walkability indicators
PDP analysis was conducted to further investigate interaction effects between walkability indicators. While theoretically 25 × 25 plots could be generated for all indicator pairs, due to space limitations, this study focused on key indicators frequently referenced in urban planning and design literature, including: FAR (land use), Average Street Length (street network), Square Number (public space configuration), and GSI (building form).
4.4.1 Synergistic effects of FAR with other walkability indicators
Figures 8(a)—8(d) demonstrate that metro station areas achieve higher vitality when characterized by high FAR, abundant catering services, proximity of pedestrian malls to stations, and smaller average ground floor areas.
When FAR exceeds 2.0 and catering service density surpasses 300, their combined effect significantly enhances vitality, indicating that high-capacity development paired with diverse dining options fosters social interaction. In areas with FAR below 1.5, increased street density produces minimal vitality improvement, suggesting that development density takes precedence over street network connectivity. This occurs because even well-connected street networks in low-density areas maintain substantial distances between destinations. In high-capacity areas (FAR exceeding 2.5), pedestrian malls maintain their vitality-enhancing effects even when located farther from subway stations, demonstrating that density supports the efficacy of distant pedestrian infrastructure. The combination of low FAR and expansive ground floor footprints significantly diminishes urban vitality, as disproportionately large structures situated in underdeveloped areas create excessive spatial intervals that inhibit street-level social interactions and pedestrian engagement.
4.4.2 Synergistic effects of average street lengths with other walkability indicators
Figures 8(e)—8(h) show that urban vitality in station areas is enhanced by a combination of shorter street lengths, higher FAR, greater street centrality, more squares, and increased GSI. The analysis identifies 0.28 as a critical threshold for average street length. When streets are shorter than this threshold, increasing FAR substantially improves urban vitality.
Conversely, when average street length exceeds 0.28, the positive impact of higher FAR diminishes significantly. This occurs because longer street segments reduce permeability, constraining pedestrian movement. Despite increased construction capacity providing more services, pedestrians struggle to navigate efficiently to their destinations. Similarly, when average street length remains below the threshold value, other factors―street centrality, number of squares, and GSI―demonstrate more pronounced positive effects on vitality. This relationship highlights how an efficient street network can effectively amplify the benefits of other urban design elements, collectively enhancing the vitality of the metro station area.
4.4.3 Synergistic effects of square number with other walkability indicators
Figures 8 reveals that station area vitality positively correlates with higher square number, increased FAR, dense street network, and smaller ground floor areas. When square number exceeds four, FAR’s positive impact becomes particularly significant, demonstrating that strategic placement of squares around high-FAR buildings effectively balances dense development with quality open spaces―optimizing both land use efficiency and environmental quality.
A synergistic effect emerges when combining more than two squares with street density exceeding eight, as the enhanced street network improves square accessibility. While larger average ground floor areas generally inhibit vitality, this negative effect diminishes once square count exceeds 5, likely because the resulting public space network mitigates the monotony of large building facades. Notably, square count and square distance show no significant interaction, reflecting the homogeneous demand for squares in station areas. Optimal square placement requires both proximity to station entrances/exits for transit flow management and even distribution throughout the area to provide diverse social and leisure spaces for various user groups.
4.4.4 Synergistic effects of GSI with other walkability indicators
Figures 8(m)—8(p) reveal that within an optimal GSI range of 0.25—0.35, station area vitality significantly increases with higher FAR, greater street density, and reduced pedestrian mall distance. This specific GSI threshold represents an ideal balance between street enclosure and open space availability.
When GSI falls within this optimal range, the positive effects of both FAR and street density are substantially amplified. This balanced building coverage ratio creates conditions where high-capacity development provides abundant service facilities, while the dense street network effectively connects open spaces and functional amenities to form an integrated urban living network, thereby enhancing overall urban vitality. Furthermore, the data indicate that building height variation, as measured by standard deviation, plays a particularly important role within this optimal GSI range, significantly contributing to increased pedestrian activity and movement. This suggests that building volume diversity within a controlled building footprint creates a more engaging urban environment.
4.5 Local effects and metro station areas vitality promoting strategies
Assessing the local effect of walkability indicators on metro station area vitality provides significant implications for urban renewal strategies. By analyzing the similar patterns of local effects among metro station areas and proposing targeted intervention strategies, we can optimize station areas from a pedestrian perspective. Through hierarchical clustering analysis, we constructed a similarity spectrum of local effects for metro station areas. Figure 9 reveals that based on vitality impact characteristics, four cluster types exhibit significant differences―Type 1 shows the highest predicted vitality, while type 4 displays the lowest predicted vitality level. Figure 9 further reveals the spatial distribution pattern of these types: type 1 is primarily concentrated in the urban core area, type 2 is distributed in both the urban core and multiple peripheral sub-centers, while types 3 and 4 are mainly distributed in peripheral urban areas. Notably, type clusters formed based on local effect similarities demonstrate clear spatial agglomeration, aligning with Shanghai’s “one main center, multiple subcenters” urban spatial development pattern, reflecting the intrinsic connection between urban development logic and pedestrian vitality distribution. To gain a deeper understanding of each type’s characteristics, we selected typical cases from each type for micro-analysis, with Fig. 9 visually demonstrating the local explanation mechanisms of these cases.
Lujiazui Station, located in the new urban center, exhibits the highest vitality level among the four cases. Its positive influence factors are comprehensive, including diverse POIs (such as high-density catering, financial, and corporate services) and open plaza spaces. However, this station area still faces constraints from negative factors such as low street density. Vitality could be further enhanced by adding pedestrian alleys to divide large blocks into smaller ones, increasing pedestrian street density, and strengthening road network permeability.
Changshulu Station represents a medium-high vitality station area type, with multiple positive influence factors, but still has room for improvement in pedestrian mall numbers and corporation services. Therefore, urban renewal for this station area should focus on creating pedestrian commercial streets and incorporating corporate service functions.
Dongan Road Station, as a representative of medium vitality station areas, has its vitality primarily supported by factors such as Mixed POI, FAR, street centrality, and financial services, while catering services, tourist services, street density, and corporation services have negative impacts on vitality. Based on this, urban renewal for this station area could adopt an “acupuncture-style” intervention strategy, precisely increasing the density of catering service facilities and adding capillary-like pedestrian alleys and walkways within existing blocks to form a multi-level, comprehensive pedestrian network, effectively activating the station area’s vitality potential.
Fudandaxue Station exhibits low vitality characteristics, with street centrality being the core factor supporting the station area’s limited vitality. This station area faces multiple overlapping negative factors, including insufficient density of catering services, financial services, and corporation services, a lack of POI diversity, and a shortage of pedestrian malls. It should fully utilize the area’s high road network accessibility advantage by introducing a diverse mix of functional facilities through policy incentives while strengthening the creation of commercial street frontages.
Given the varied local effect patterns across metro station clusters, policies should be customized for each station’s specific needs, prioritizing intervention in lowvitality types 3 and 4. For Donganlu Station (type 3), increasing catering facility density from 107 to 200 would boost vitality. Fudandaxue Station (type 4) requires both higher catering density and greater POI diversity (from 1.5 to 1.8+) to create a vitality area. This approach enhances urban renewal precision and provides evidence-based guidance for creating vibrant, human-centered metro station areas.
5 Discussion
5.1 Non-linear effects of walkability indicators
The research results partially support and complement existing literature, revealing non-linear relationships between transit station area walkability and urban vitality (
Doan et al., 2024;
Yang et al., 2021). These findings reveal priority levels among different walkability features, effectively assisting policymakers, urban planners, and designers in developing strategic guidelines.
From the perspective of land use, this research suggests that the relationship between development intensity and urban vitality follows a non-linear pattern, challenging previous linear assumptions. While development intensity is necessary for promoting urban vitality, excessive density provides no additional benefits. Specifically, our research observed distinct FAR thresholds: values below 1.3 negatively impact urban vitality; between 1.3 and 2.0, positive effects begin to emerge; between 2.0 and 2.5, enhancement accelerates rapidly; beyond 2.5, additional capacity yields diminishing returns. This demonstrates that optimal station area vitality requires careful calibration within specific density ranges rather than simply maximizing development intensity. Additionally, we found a critical Mixed POI threshold at 1.8: values below this point correlate with reduced vitality, while those exceeding it show significant positive effects. This finding aligns with studies by
Yang et al. (2021) and complements research by
Doan et al. (2024).
Regarding street networks, this research uncovered differential impacts and mechanisms among various network features on vitality. For instance, shorter street segments (below 0.2 km) substantially enhance pedestrian activity, while moderately longer segments (0.2—0.4 km) show dramatically reduced pedestrian attraction. Interestingly, when segment length exceeds 0.4 km, the inhibitory effect slightly diminishes―a nuanced finding that contrasts with previous research (
Xiao et al., 2021), which did not detect this diminishing effect with continuously increasing street lengths. Longer street segments may offer other attractive qualities (such as better landscape views, fewer intersection disruptions), partially offsetting the negative effects of length. Street Centrality exhibits a linear positive correlation with vitality in the 0—400 range, with promotional effects weakening beyond this threshold, suggesting that fundamental accessibility requirements of the road network have been satisfied at this point. Notably, our analysis identified an optimal range for motorway width (15—20 m) that positively influences urban vitality, indicating that excessively narrow lanes fail to accommodate vehicle access needs, while overly wide lanes generate increased through-traffic that suppresses station area vitality―a parameter rarely addressed in existing literature.
Concerning public space configuration, the study confirmed that squares and pedestrian malls are closely related to station area vitality, while the relationship between parks and station area vitality remains unclear. These results emphasize the importance of rationally organizing different types of public spaces within the metro station area to promote walking activities―an aspect often overlooked in station area vitality measurement and impact factor research. The study found that when the square number exceeds 4, public space networks form more easily, with promotional effects on walking significantly higher than when values are in the 2—4 range. When the path distance between pedestrian malls and metro stations exceeds 150 m, their impact on vitality shifts from positive to slightly negative. These conclusions partially fill gaps in existing research. Existing studies often only consider the quantity and density of public spaces (
He and He, 2023), without examining their spatial relationship to metro stations.
Regarding building form, our research confirms a significant correlation between building morphological harmony and urban vitality. When the ground floor area standard deviation remains below 500, a linear promoting effect on station area vitality is observed; however, exceeding this threshold produces an inhibitory effect as the linear relationship diminishes. This demonstrates how harmonious building configurations enhance pedestrian experiences and stimulate station area vitality. Additionally, while increasing GSI generally promotes station area vitality, values exceeding 0.35 counterintuitively suppress walking activity―likely due to reduced public open spaces that inhibit street life development. This finding contradicts previous linear analysis-based research, which predominantly suggested only positive correlations between GSI and urban vitality (
Yang et al., 2021).
5.2 Synergistic effects between walkability indicators
This research examines the synergistic relationships between multiple walkability indicators and urban vitality. Our analysis reveals that the influence of individual walkability indicators on urban vitality changes in response to variations in other indicators. Leveraging these synergistic effects can inform more effective urban design practices.
For instance, pedestrian malls located beyond 150 m from metro stations typically have minimal impact on urban vitality; however, when the FAR exceeds 2.5, these pedestrian malls enhance station area vitality despite greater distances―indicating that high-capacity developments support commercial streets at more distant locations through population concentration. When average street length falls below 0.28 km, the positive effects of increased building capacity, improved network accessibility, more squares, and higher building coverage ratios are amplified. Additionally, when the number of squares exceeds 5, the inhibitory effects of oversized building volumes on urban vitality are effectively mitigated.
The research confirms significant synergistic effects between different walkability features, highlighting the necessity of considering multiple environmental dimensions simultaneously when designing metro station areas. These findings provide quantitative explanations for the cooperative relationships between different elements in transit station areas and contribute to the coordinated allocation of urban spatial resources. Urban planners should therefore adopt a holistic approach that considers how various elements interact rather than optimizing individual features in isolation.
5.3 Policy implications
Using metro stations as catalysts for urban renewal through enhancing spatial quality and optimizing pedestrian environments promotes efficient station-city synergy and constructs high-vitality metro station areas―a crucial approach to achieving sustainable urban development and improving residents’ quality of life. This research provides quantitative support for MSA urban renewal design by explaining local effects, thereby enhancing the application of machine learning methods in specific urban renewal project practices.
Specifically, the research offers explanations of the local effects of specific metro station cases. For example, in selected low and medium-vitality cases, insufficient station area vitality often results from fewer service facilities and low function diversity. Urban renewal for these station areas should primarily focus on increasing POI density and diversifying service facilities. However, in selected medium-high and high vitality cases, although street activities are already relatively prosperous, targeted enhancement opportunities persist. Two critical interventions warrant implementation: optimizing street network density to mitigate low connectivity constraints and strategically developing public spaces that activate urban social life through enhanced community interaction.
This portion of the research demonstrates through four examples how machine learning methods can identify problems in specific metro station area pedestrian environments and urban vitality, providing quantitative support for metro station area urban renewal practices. By integrating computational analysis with practical design considerations, planners can develop targeted interventions that address the unique challenges and opportunities of each metro station area, ultimately creating more vibrant, accessible, and sustainable urban environments.
6 Conclusion
This research investigates the non-linear relationship between walkability and urban vitality in Shanghai’s central area. The findings demonstrate that a high-quality pedestrian environment is crucial for promoting urban vitality. Among the four dimensions analyzed, land use explains 61.09% of the variance in vitality, public space configuration contributes 19.71%, street network accounts for 11.46%, and building form adds 7.73%. The study identifies several key factors that enhance urban vitality: dense catering services, ample square quantity, highly permeable street networks, and harmonious building forms. Importantly, the research reveals significant threshold effects in how walkability influences vitality. All variables exhibit specific thresholds where their impact on vitality shifts from negative to positive, with most variables showing significantly different effects across various value ranges. Notably, excessively high GSI negatively impacts station area vitality, while motorway width only promotes vitality within an optimal range. Furthermore, the research identifies synergistic effects between certain indicators. For instance, when the average street length is relatively small (below 0.28 km), increased FAR and enhanced network accessibility generate more significant positive effects on urban vitality. Through the interpretation of local effects, this study quantitatively assesses the shortcomings in pedestrian environments around specific metro stations. This methodology effectively supports decision-making for urban planning professionals engaged in metro station area renewal projects. By validating theories of station area vitality and exploring the synergistic and non-linear threshold effects among walkability indicators, this research provides valuable references for sustainable development and vitality enhancement in transit station areas.
Despite this study’s contributions to sustainable development and vitality creation in station areas, several limitations remain. First, our aggregation of built environment characteristics as walkability indicators overlooks the heterogeneity across different zones within station areas and fails to examine micro-scale environmental quality impacts. Future research should investigate metro station area vitality from multi-zone and multi-scale perspectives. Second, our methodology, combining field surveys with big data to characterize pedestrian-oriented vitality, entails substantial labor costs. More cost-effective measurement approaches should be developed (e.g., webcams or Wi-Fi devices). Finally, our findings may be context-specific, as cities at different development stages or with different cultural backgrounds likely exhibit variations in the importance and thresholds of walkability indicators. Extensive comparative studies are needed to establish more universally applicable conclusions.
2095-2635/2025 The Authors. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd.
PDF
(11571KB)
