Quality of Pedestrian Networks Around Metro Stations: An Assessment Based on Approach Routes

Abstract

Walking is the primary mode of reaching metro stations, yet the quality of pedestrian networks around these stations has not been well researched. Considering the objective physical characteristics of pedestrian networks and the subjective assessments of walkers on the routes, this study developed an evaluation model that integrated the Analytic Hierarchy Process and Entropy Weight Method with human–machine adversarial scoring and cosine similarity to validate the reliability. Nineteen indicators concerning four fundamental criteria, including accessibility, convenience, safety, and comfort, were applied with data acquired from eight stations in Tianjin, China. Results reveal that accessibility and safety indicators weigh more than convenience and comfort indicators. The quality of pedestrian networks around the public-service and comprehensive stations scores higher than that around residential stations, while walking environment quality near commercial stations shows significant disparities. These findings highlight the importance of prioritizing accessibility and safety while enhancing convenience and comfort in the renewal of the pedestrian network in Tianjin. The assessment model provides a valuable tool for urban policymakers and planners, enabling the formulation of sound pedestrian-network policies, facilitating higher-quality walking access and egress trips to stations, and encouraging transit-oriented development.

1. Introduction

Metros, widely believed to mitigate traffic congestion, address parking shortages, and improve air quality, have been rapidly constructed in more than 40 cities across China over the past decade [1,2]. A crucial component of a successful metro system is pedestrian access to and from metro stations [3]. However, despite the rapid expansion of metro networks, there has been insufficient emphasis on integrating pedestrian networks in the areas surrounding metro stations [4]. Evaluating the quality of pedestrian networks within these catchment areas is essential for enhancing pedestrian access and ensuring that well-constructed metro systems achieve their full potential in attracting ridership and supporting future development [5,6].

Existing studies on walking environments around metro stations primarily focus on criteria that align with pedestrian demands and indicators that link to built environment characteristics, using quantitative and model-driven analyses [4,7]. Central to these studies is walking accessibility, which significantly affects metro ridership [8,9,10]. Researchers have explored the spatial and psychological aspects of access, such as suitable walking distances and perceived satisfaction levels [11,12,13]. Attributes like sidewalk width, street crossings, traffic signals, and street greenery, have been recognized as essential factors influencing pedestrian accessibility [5,6,14,15,16,17,18,19]. Furthermore, pedestrian route directness has also been identified as a key motivator for metro users to access stations on foot [13,20].

Beyond accessibility, research has also explored walking safety and attractiveness within the metro’s service areas. Unsafe streets deter pedestrians [21], while enhancements like streetlights and well-equipped amenities increase the appeal of walking [22,23]. Arellana et al. (2020) highlighted that pedestrians prioritize attractiveness in large cities, whereas security and traffic safety are more critical in smaller cities [24]. Furthermore, walking convenience and comfort also influence individuals’ preferences towards walking to metro stations [25,26]. Frameworks like Jiao et al. (2017)’s six-criteria model (including convenience, comfort, safety, pleasantness, connection, and conspicuousness) [27] and Manaswinee et al. (2024)’s user satisfaction-based approach reflect the growing recognition of multi-dimensional evaluation criteria [28].

Despite these advancements, significant research gaps remain. Firstly, previous studies have primarily regarded accessibility and safety as prerequisites for walking to be feasible, treating criteria like comfort and enjoyment as additional considerations that are unlikely to be prioritized unless fundamental needs are met [29,30,31,32,33]. However, a limited amount of research has thoroughly explored these criteria. Secondly, some of the walking indicators do not focus on walking facilities. For instance, safety indicators often emphasize street crime instead of the conditions of roads, tending to measure security rather than safety [34,35]. Thirdly, the quality of the pedestrian networks is influenced by both its physical characteristics and the subjective evaluations of walkers [7,36], yet few studies integrate these perspectives comprehensively.

To address these gaps, this study hypothesizes that the quality of pedestrian networks around metro stations is determined by four core criteria, including accessibility, convenience, safety, and comfort, and that a novel methodology integrating subjective and objective evaluations can effectively identify critical criteria and indicators for enhancing pedestrian networks. To test this hypothesis, this paper introduces a model that combines the Analytic Hierarchy Process (AHP) with the Entropy Weight Method (EWM), enabling a comprehensive assessment of pedestrian network quality, and this model is applied to four types of metro stations in Tianjin, one of China’s four municipalities. By offering a robust tool for assessing walking environments, this study aims to assist urban planners and policymakers in understanding and improving pedestrian access to metro stations.

The paper is structured in five parts. Section 2 discusses the quality assessment models, and data collection is introduced in Section 3. Section 4 interprets the results of our analysis. Section 5 concludes and discusses the implications for planning policies and future perspectives.

2. Methodology

2.1. Modelling Framework

This study applies a combination of the AHP and EWM to assess the quality of pedestrian road networks in metro catchment areas. The AHP model is a structured technique that uses experts’ scoring data, collected through a questionnaire survey, to estimate the weights of criteria and indicators. It is a subjective weighting method achieved by comparing the correlations and importance of the indicators, constructing judgement matrices and solving for the eigenvectors [37]. The EWM model is an objective method for assessing the weights of indicators based on the information they contain, which can mitigate the subjectivity associated with the AHP method. The higher the entropy value of an indicator, the greater its degree of dispersion and the amount of information it provides. Consequently, it should have a more significant influence on the quality evaluation and be assigned a larger weight, and vice versa [38]. Online multi-source data reflecting the walking environments around the selected metro stations in Tianjin, integrated with supplementary data collected through field investigation, are used to estimate the EWM model. Additionally, a human–machine adversarial scoring model [39] and cosine similarity are employed to test the evaluation system’s rationality and the evaluation results’ accuracy.

The modelling framework is shown in Figure 1.

2.2. Modelling Process

2.2.1. Filtration of Evaluation Indicators

We conducted a comprehensive search of the CNKI database using keywords such as “pedestrian network”, “road network evaluation”, “travel quality”, “walking facilities”, and “sidewalks”, identifying 51 relevant papers published since 2017 [4,5,6,7,8,10,12,15,19,20,23,24,27,28,33,36,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74]. From this pool, we extracted 39 indicators for evaluating pedestrian network quality and analyzed their frequencies, eliminating 11 with low occurrences. Subsequently, we adjusted and refined 6 indicators by incorporating factors such as street space quality, service facility layouts, and road accessibility to enhance their applicability to daily walking. Ultimately, we grouped the remaining 22 indicators into four categories—accessibility, convenience, safety, and comfort—establishing a preliminary index system for assessing walking route quality in station catchment areas.

To validate the 22 indicators, we developed a Likert-scale questionnaire to measure their importance, with scores ranging from 1 (very unimportant) to 5 (very important). The survey was conducted in October 2023, and 137 professional workers from the areas of urban planning and transportation were invited to participate. A total of 135 responded. Based on the experts’ scoring data, we calculated the mean values, coefficients of variation, and Kendall’s coefficients of harmony of each indicator. The results showed that the Kendall’s coefficients of harmony were all larger than 0.20 and passed the 0.05 significant test, indicating a high degree of consensus among the experts. However, 2 indicators, namely block sizes and social security, were excluded due to mean scores below 3.50 and coefficients of variation exceeding 25.00%. To further ensure the independence of the indicators, we calculated the Pearson’s correlation coefficients. The density of the commercial facilities was excluded due to its strong correlation with the density and diversity of daily service facilities. This refinement process resulted in a final set of 19 indicators. All calculations were conducted using SPSS 26 software.

2.2.2. AHP-Based Subjective Weights

The core idea of the AHP model is to divide complex decision-making problems into multiple layers. In this study, the target layer evaluates the quality of pedestrian networks within the stations’ catchment areas. The criterion layer encompasses the accessibility, convenience, safety, and comfort of walking routes to metro stations, while the factor layer comprises indicators related to walking environments along these routes. The structured hierarchical framework is depicted in Figure 2.

After structuring the hierarchy, we conducted pairwise comparisons of indicators under the same criterion, thereby generating an importance degree for judgement. To improve the credibility and accuracy of the data, we invited 51 experts in urban planning to score the importance of each indicator using the “1–9 scale method” [75]. To avoid logical errors, we performed consistency tests on the judgement matrices. For example, assuming there are 3 indicators under “accessibility”, if the importance rankings indicate C1 > C2 and C2 > C3, then C1 > C3 should be true. If, instead, C1 < C3, it signifies a failure of the consistency test. The equations are as follows:

$$\left\{\begin{array}{c}{\lambda }_{max}={\sum }_{i=1}^n{\displaystyle \frac{{AW}_i}{{nW}_i}}\\ CI={\displaystyle \frac{{\lambda }_{max}-n}{n-1}}\\ CR={\displaystyle \frac{CI}{RI}}\end{array}\right.$$

(1)

where λ_max is the maximum eigenvalue of matrix C; W_i is the normalized eigenvector of the index; n is the number of layers in matrix C; CI is the consistency index; RI is the random consistency index; and CR is the verification coefficient. When CR < 0.1, the judgement matrix passes the test and the indicator weight value is valid. When CR > 0.1, there are contradictions in value assignment and it is necessary to rescore and adjust the judgement matrix to meet the requirements.

Given that there are multiple export scoring matrices, it is necessary to calculate the geometric mean of each judgement matrix that has passed the consistency test; this can be achieved through Equation (2), as follows:

$$X_{ij}={{(X}_{ij1}+X_{ij2}+X_{ij3}+\cdots +X_{ijn})}^{{\displaystyle \frac{1}{n}}}$$

(2)

where X_ij is the final relative importance of indicators i and j and X_ijn is the relative importance of indicators i and j by the n-th expert. Finally, the subjective weights of the indicators are calculated through Equation (3), as follows:

$$W_i=W_{c,i}\times W_{f,i}$$

(3)

where W_i is the final subjective weight of indicator i; W_c,i is the weight of criterion to which indicator i belongs; and W_f,i is the weight of indicator i relative to its corresponding criterion.

The calculations were conducted using YAAHP 12.11 software.

2.2.3. EWM-Based Objective Weights

To address the subjectivity inherent in the AHP, we employed the EWM model to calculate the indicators’ objective weights. The process involved three key steps. First, we constructed a standardized evaluation matrix. Second, we converted the standardized evaluation matrix into evaluation weight matrix P_mn, calculated as follows:

$$P_{mn}=\left\{\begin{array}{cccc}{P}_{11}& P_{12}& \cdots & P_{1j}\\ P_{21}& P_{22}& \cdots & P_{2j}\\ ⋮& ⋮& ⋮& ⋮\\ P_{i1}& P_{i2}& \cdots & P_{ij}\end{array}\right.$$

(4)

$$P_{ij}={\displaystyle \frac{A_{ij}}{{\sum }_{i=1}^m{A}_{ij}}}$$

(5)

where m denotes the number of walking routes; n signifies the number of evaluation indicators; and A_ij represents the j-th evaluation indicator of the i-th walking route, with i ranging from 1 to m and j ranging from 1 to n. The entropy value of each indicator can be used to characterize its degree of dispersion, thereby determining the importance of the information that the indicator reflects within the overall evaluation system. The equation is shown as follows:

$$H_j=-ln{\left(m\right)}^{-1}{\sum }_{i=1}^m{P}_{ij}ln{P}_{ij}$$

(6)

where H_j represents the information entropy value of indicator j; m indicates the number of walking routes; and P_ij denotes the weight of the j-th evaluation indicator for the i-th walking route in the matrix P_mn. Thirdly, we obtained the indicator’s entropy weight by calculating the proportion of its entropy value; this was achieved through Equation (7), as follows:

$$K_j={\displaystyle \frac{1-H_j}{{\sum }_{j=1}^n(1-H_j)}}$$

(7)

where K_j is the weight of indicator j and H_j is the information entropy value of indicator j.

2.2.4. Composite Weights and Evaluation Formula

Based on the subjective and objective weights, we obtained the composite weights of the indicators; this was achieved through Equation (8), as follows:

$$Q_i={\displaystyle \frac{\sqrt{W_i{K}_i}}{{\sum }_{i=1}^n\sqrt{W_i{K}_i}}}$$

(8)

where Q_i, W_i, and K_i are the composite, subjective, and objective weights of the i-th indicator, respectively, and n is the total number of indicators.

Using the composite weights of the indicators and the number of passengers on the approach routes, we developed a formula to assess the quality of the pedestrian road networks within the catchment areas of metro stations (see Equation (9)):

$$\left\{\begin{array}{c}{C}_{1,i}={\sum }_{j=1}^n{\displaystyle \frac{r_{ij}}{R_i}}{C}_{1,ij}\\ C_{2,i}={\sum }_{j=1}^n{\displaystyle \frac{r_{ij}}{R_i}}{C}_{2,ij}\\ ⋮\\ C_{m,i}={\sum }_{j=1}^n{\displaystyle \frac{r_{ij}}{R_i}}{C}_{m,ij}\end{array}\right.$$

(9)

where C_m,i is the final value of the C_m indicator for the i-th metro station’s catchment area; n is the total number of routes within the i-th metro station’s catchment area; C_m,ij is the value of the C_m indicator for the j-th route in the i-th metro station’s catchment area; r_ij is the number of travellers on the j-th route in the i-th metro station’s catchment area; and R_i is the total number of travellers in the i-th metro station’s catchment area.

2.2.5. Validation of Evaluation Results

To test the rationality of the evaluation model and the accuracy of its results, human–machine adversarial scoring was applied. Initially, 85 urban planning experts were invited to assess street view images along the walking routes to minimize the influence of individual evaluation biases on machine learning accuracy. Subsequently, the street view images were imported into a human–machine adversarial model, where the training results were utilized for scoring. For each street view, the average of the 85 scores was taken as its final evaluation score. Finally, the average score for all street view images within the catchment area of each station was calculated.

To validate the model’s performance, a cosine similarity calculated as Equation (10) was employed to compare the human–machine adversarial scoring results with the quality evaluation outcomes. A cosine value closer to 1 indicates a smaller angle between the vectors, signifying a higher similarity between the two vectors.

$$\left\{\begin{array}{c}\overrightarrow{A}=A(a_1,a_2,\cdots ,a_n)\\ \overrightarrow{B}=B(b_1,b_2,\cdots ,b_n)\\ cos(\theta )={\displaystyle \frac{{\sum }_{i=1}^n{a}_i{b}_i}{\sqrt{{\sum }_{i=1}^n{\left(a_i\right)}^2}\sqrt{{\sum }_{i=1}^n{\left(b_i\right)}^2}}}\end{array}\right.$$

(10)

where $\overrightarrow{A}$ is the human–machine adversarial scoring vector; a_i is the pedestrian road network score of the i-th station area obtained by human–machine adversarial scoring; $\overrightarrow{B}$ is the quality evaluation calculation vector; b_i is the pedestrian road network score of the i-th station area obtained by quality evaluation; n is the total number of evaluation stations; and $cos\left(\theta \right)$ is the cosine similarity.

3. Data Collection

3.1. Study Area

Tianjin is our study city. Its core area, comprising six districts, had a population of 3.90 million in 2023. According to Tianjin’s comprehensive traffic survey, individual motorized transportation accounted for 28.2% of daily travel, resulting in increasingly severe traffic congestion and parking shortages in the core area. To address these challenges, the government has actively facilitated the development of the metro system. By the end of 2023, the core area had six operating metro lines with a total of 87 stations.

Distance is a primary factor influencing the preference for walking as an access mode. Urban and transport planners typically use an 800 m walking radius to define the service areas of metro stations [11,12,13]. However, based on the distribution and density of metro stations in the core area of Tianjin, we found that a walking radius of 600 m effectively covers the major service areas of the stations. This radius was adopted as the buffer zone for our analysis (see Figure 3).

Metro stations in Tianjin vary significantly in spatial layouts, land use, population density, and passenger flow characteristics within their catchment areas. To comprehensively understand the walking environments around the stations, we applied the K-Means clustering algorithm to analyze the features of urban construction land within a 600 m radius of the 87 stations. This analysis grouped the stations into six categories, including residential, commercial, public service, transport, comprehensive functions, and others. Among the six categories, transport stations primarily serve as hubs connecting to train stations and the airport, and the “others” encompass stations that are either planned or currently under construction. Considering the diversity of station types and the availability of relevant data, we systematically selected eight stations across the remaining four types, each with a 600 m radius buffer, as our study sites.

3.2. Online Multi-Source Open Data

3.2.1. Pedestrian Networks

The pedestrian network data, which is essential for understanding the walking environments within the stations’ catchment areas, was downloaded from OpenStreetMap. After filtering out unnecessary roads and correcting inconsistencies, the network was saved as a shapefile in ArcGIS for subsequent analysis.

3.2.2. Points of Interest (POIs)

POIs transform specific places where people conduct daily activities into point data with geographical attributes, reflecting the functional characteristics of the station’s catchment area. The POIs data employed in this study were obtained through the Baidu Map API and mapped on the pedestrian network using ArcGIS (version: 10.8).

3.2.3. Building Heights

Building heights depict the development intensity in the area surrounding the stations and can be used in estimating traveller volumes on the routes. This study utilized open data on Tianjin’s building heights published by Wuhan University [76]. To ensure accuracy, we used ArcGIS and the Baidu 3D Map API to verify and correct the downloaded data.

3.2.4. Street Views

We progressed through street view images within the stations’ 600 m radius areas at intervals of 30 m through the panoramic static images API provided by the Baidu Map Open Platform. A fully convolutional network semantic segmentation model was applied to classify and quantify elements in street view images, such as greenery and sky [42]. The results were spatially projected onto the walking routes through ArcGIS.

3.2.5. AMap’s Walking Routes

AMap’s walking-route planning interface offers answers for walking route queries and distance/time calculations. To streamline the process of route acquisition, we divided the stations’ catchment areas into traffic zones and designated the centre point of each traffic zone as the origin. By inputting the coordinates of the origins and metro stations into AMap’s API, we obtained walking route data.

3.3. On-Site Travel Survey Data

The on-site travel survey includes two parts. The first part collected respondents’ social demographic information and trip characteristics. To collect the accurate walking approach routes, participants were required to map their actual access routes to stations either on paper maps or electronically using iPads. The second part involved a field investigation, aiming to collect supplementary data on walking environments. The survey was conducted in November 2023 by four trained postgraduate students. In total, 320 respondents were randomly selected across eight metro stations, with a response rate of about 94%.

Using ArcGIS, we integrated walking routes from the on-site survey with those generated by AMap. Through this process, we ultimately acquired 61 commonly used walking access routes, as visualized in Figure 4. The descriptions of the 19 indicators are listed in

Table 1. Descriptions of evaluation indicators.

Criteria	No.	Indicators	Descriptions
Accessibility	C1	Degree of interruptions	The ratio between the number of interruptions on the access route and the actual length of that route
	C2	Width of sidewalks	The average width of sidewalks along the access route
	C3	Pedestrian route directness	The ratio of the actual walking route length to the Euclidean distance between the route’s origin and the station
Convenience	C4	Barriers to walking	The number of obstacles, like tree pits and trash cans, on the sidewalks along the access route
	C5	Route guide signs	The number of guide signs along the access route
	C6	Density of daily service facilities	The ratio between the number of facilities along the access route and the actual length of that route
	C7	Diversity of daily service facilities	The variety and range of facilities (POIs) available along the access route that cater to the daily needs and services of individuals
Safety	C8	Percentage of sidewalks	The ratio between the length of sidewalks on the access route and the actual length of that route
	C9	Encroachment of sidewalks	The areas of sidewalks along the access route that are occupied by motor vehicles or bicycles
	C10	Percentage of protected routes	The ratio between the length of protected sidewalks on the access route and the actual length of that route
	C11	Completeness of pedestrian crossing facilities	The ratio of the number of street-crossing facilities along the access route to the total number of intersections
	C12	Density of entrances and exits	The ratio between the number of entrances/exists on the access route and the actual length of that route
Comfort	C13	Quality of sidewalk paving	The smoothness, slip resistance, and water permeability of the sidewalk paving along the walking routes to stations
	C14	Provision of barrier-free facilities	The provision of barrier-free facilities that meets the demands of the elderly and the disabled
	C15	Density of recreational facilities	The ratio between the number of recreational facilities along the access route and the actual length of that route
	C16	Percentage of routes with street lamps	The ratio between the length of the access route with street lamps and the actual length of that route
	C17	Walking congestion	The degree of crowding during walking to stations because of narrow sidewalks or too many pedestrians
	C18	Greenery	The ratio between the green area and the total area of the street view photos of the access route
	C19	Openness of vision during walking	The ratio between the sky area and the total area of the street view photos of the access route

.

4. Results

4.1. Weights of Pedestrian Network Indicators

Table 2. The weights of the indicators.

Criteria	Number	Indicators	Subjective Weights	Objective Weights	Composite Weights	Final Ranks
Accessibility	C1	Degree of interruptions	0.1318	0.960	0.0867	2
	C2	Widths of sidewalks	0.0637	0.916	0.0836	3
	C3	Pedestrian route directness	0.1682	0.936	0.1071	1
Convenience	C4	Barriers to walking	0.0391	0.967	0.0481	10
	C5	Route guide signs	0.0169	0.970	0.0358	13
	C6	Density of daily service facilities	0.0379	0.993	0.0238	19
	C7	Diversity of daily service facilities	0.0644	0.956	0.0498	9
Safety	C8	Percentage of sidewalks	0.0821	0.992	0.0392	11
	C9	Encroachment of sidewalks	0.0425	0.969	0.0366	12
	C10	Percentage of protected routes	0.0285	0.879	0.0832	4
	C11	Completeness of pedestrian crossing facilities	0.0998	0.949	0.0686	6
	C12	Density of entrances and exits	0.0509	0.973	0.0564	7
Comfort	C13	Quality of sidewalk paving	0.0208	0.963	0.0271	18
	C14	Provision of barrier-free facilities	0.0189	0.963	0.0332	14
	C15	Density of recreational facilities	0.0293	0.959	0.0765	5
	C16	Percentage of routes with street lamps	0.0488	0.977	0.0309	16
	C17	Walking congestion	0.0132	0.953	0.0299	17
	C18	Greenery	0.0295	0.963	0.0519	8
	C19	Openness of vision during walking	0.0137	0.963	0.0317	15

shows the weights of the indicators. The subjective and objective weights are presented in the middle columns, respecting the subjective assessments of walkers on the routes and the physical characteristics of pedestrian networks, separately. A Pearson correlation test revealed no statistically significant association between the subjective and objective weights, indicating that the two approaches capture distinct dimensions of the indicators.

The composite weights, derived from the subjective and objective weights, reflect the indicators’ impacts more thoroughly. Among the top six weighted indicators, three pertain to accessibility and three to safety, suggesting that metro passengers prioritize these two criteria when walking to stations. This preference is understandable, as passengers value routes with fewer detours, shorter waiting times at intersections, and minimal conflicts with other traffic flows [73,77].

Among the indicators, pedestrian route directness ranks as the top priority, followed by the number of interruptions. This makes sense, as most passengers prefer more direct and uninterrupted walking routes that minimize travel time to their destination stations [27]. Sidewalk widths and the proportion of protected routes rank third and fourth, respectively, highlighting the significant role of the pedestrian networks in enhancing walking access to metro stations. This finding aligns with the reality that sidewalks near metro stations are often occupied by shared bikes and even motor vehicles, forcing pedestrians to walk in bike lanes or even vehicle lanes [78]. Consequently, sufficiently wide sidewalks are essential for providing adequate walking space, while protected sidewalks are equally important for separating pedestrians from other traffic and improving safety. The importance of sidewalk width and protection also suggests a broader need for urban planners to address challenges in shaped spaces.

Walking congestion, quality of sidewalk paving, and the density of daily service facilities rank last, indicating that metro passengers prioritize efficient access to stations over utilizing services along their routes. Furthermore, passengers exhibit a higher tolerance for issues like uneven sidewalks, esthetic concerns, and congestion. This tolerance may stem from the flexibility and adaptability of walking as an access mode, as well as the efficiency of metro travel, which may reduce the perceived importance of sidewalk quality [79].

4.2. Evaluations of Walking Routes Quality

4.2.1. Accessibility

Table 3. Results of quality evaluations of walking routes.

Quality of Walking Routes		Types and Names of Metro Stations
Criteria	Indicators	Residential S.		Commercial S.		Public-Service S.		Comprehensive S.
		Jinshi Bridge	South Hongqi Road	Heping Road	Yingkoudao	Culture Centre	Zhoudeng Memorial Hall	Tianta	Changhong Park
Accessibility	Numbers of interruptions	0.19	0.22	0.44	0.48	0.29	0.36	0.27	0.26
	Widths of sidewalks	2.21	2.78	1.56	1.61	2.25	2.17	2.15	2.17
	Pedestrian route directness	1.45	1.50	1.41	1.31	1.42	1.70	1.63	1.57
Convenience	Barriers to walking	2.39	2.81	3.59	3.36	3.16	3.20	3.31	2.85
	Route guide signs	2.85	2.31	3.28	3.30	2.98	3.02	2.22	2.81
	Density of daily service facilities	8.10	7.21	10.68	9.27	5.64	6.03	9.04	8.31
	Diversity of daily service facilities	0.35	0.30	0.42	0.36	0.16	0.18	0.30	0.20
Safety	Percentage of sidewalks	91.22	90.61	97.05	94.30	90.68	90.44	94.53	92.39
	Encroachment of sidewalks	3.22	2.99	3.58	4.00	2.79	3.37	3.76	3.96
	Percentage of protected routes	8.73	19.97	47.27	25.32	84.82	21.31	33.11	38.40
	Completeness of pedestrian crossing facilities	91.47	92.32	94.26	93.79	100.00	94.31	90.16	91.24
	Density of entrances and exits	0.55	0.80	0.82	0.76	0.54	0.52	0.81	0.93
Comfort	Quality of sidewalk paving	3.02	2.60	2.05	3.32	3.46	2.10	2.09	3.00
	Provision of barrier-free facilities	1.91	2.30	2.36	1.65	2.82	2.25	2.02	2.65
	Density of recreational facilities	0.30	0.17	0.43	0.36	0.41	0.35	0.27	0.31
	Percentage of routes with street lamps	70.20	83.48	96.03	95.63	85.90	93.13	86.02	86.15
	Walking congestion	2.40	2.65	3.63	2.57	2.40	2.27	2.94	2.29
	Greenery	15.75	14.97	18.80	21.08	19.06	16.18	11.71	13.56
	Openness of vision during walking	24.25	27.54	13.86	18.38	24.00	26.39	22.82	28.89

Note: S. is the abbreviation of station.

presents the results of quality evaluations of walking routes within the 600 m catchment areas of the eight metro stations in Tianjin. Overall, walking routes near residential stations experience the fewest interruptions, averaging about 0.20 times per 100 m. In contrast, routes around the commercial stations face the most interruptions (approximately 0.46 times per 100 m). Public-service and comprehensive stations show relatively stable interruption rates, maintaining at around 0.30 times per 100 m. These patterns are consistent with the findings suggesting that commercial areas feature smaller blocks with denser road networks whereas residential areas typically have larger blocks with wider roads [6,27].

Sidewalk widths in commercial station catchment areas average approximately 1.60 m, which is narrower than those near other types of stations. This is likely due to the commercial stations being situated in the old district of Tianjin, where space for pedestrians is limited. Meanwhile, because of the fine-grained land use patterns and small-scale blocks in the old district [6], the values of pedestrian route directness within the commercial station catchment areas are the lowest.

4.2.2. Convenience

Street furniture, such as planting pools and garbage cans, is the most common barrier regarding walking access to stations. Routes near residential stations exhibit the lowest obstacle scores, with Jinshi Bridge scoring 2.39 and South Hongqi Road scoring 2.81. In contrast, the obstacle scores of walking routes around commercial stations are the highest, with Heping Road and Yingkoudao scoring 3.59 and 3.36, respectively. This may be due to the narrower sidewalks around these two stations, which increase the proportion of space occupied by obstacles of the same size [80], making these barriers more significant for metro passengers.

Route guidance scores are higher in commercial and public-service station areas, reflecting how the greater diversity of land use around these stations attracts higher urban populations, leading to more comprehensive signage to guide pedestrian traffic. Walking routes accessing commercial stations have the highest density and diversity of daily service facilities, followed by routes accessing residential and comprehensive stations, while the ones accessing public-service stations rank the lowest. Figure 5 visualizes these patterns.

4.2.3. Safety

Sidewalks in all station catchment areas are well constructed, with over 90.00% of the walking routes being situated on sidewalks. However, encroachment is a common issue: sidewalks near commercial stations are often obstructed by stalls and parked motor vehicles, while shared bicycles frequently occupy those near comprehensive stations [81,82]. The provision of protected sidewalks shows a low correlation with the types of stations, from 84.82% at the Culture Centre station to 8.73% around the Jinshi Bridge.

Pedestrian crossings are prevalent, covering 90.00% of intersections. However, zebra crossings dominate, with pedestrian bridges and underpasses being rare. Entrance-exit density scores along the walking routes to commercial and comprehensive stations are higher than those to residential and public-service stations. The potential reason might be that the fine-grained and varied land use patterns around commercial and comprehensive stations generate more entrances and exits [6].

4.2.4. Comfort

Sidewalk paving quality varies widely. The sidewalks surrounding Culture Centre have the highest paving quality, while those within the service areas of Heping Road, Tianta, and Zhoudeng Memorial Hall exhibit poor paving conditions due to ageing infrastructures and inadequate maintenance. Barrier-free facilities, designed for vulnerable groups such as the elderly and the disabled, achieved high scores around the Culture Centre and Changhong Park stations but low scores in the Yingkoudao and Jinshi Bridge station areas.

Recreational facility density is highest along the routes to commercial and public-service stations (e.g., Heping Road and Culture Centre reaching 0.43 and 0.41 facilities per 100 m, respectively) and lowest near residential and comprehensive stations (e.g., only 0.17 facilities per 100 m in the catchment area of South Hongqin Road). Meanwhile, approximately 96.00% of the routes in the commercial station catchment areas have street lamps, reflecting the high pedestrian activity at night in these areas [83].

Congestion levels are generally manageable, except near Heping Road station, which experiences significant crowding due to commercial and business activities. The greenery scores of the walking routes accessing commercial stations are higher (e.g., Yingkoudao station at over 20.00%), while routes accessing residential and comprehensive stations have lower scores due to wider streets, as shown in Figure 6. The openness of vision during walking access to commercial stations is the lowest (see Figure 7), possibly due to the narrow streets and the higher proportions of greenery.

4.2.5. Overall Evaluations

Table 4. Overall quality evaluations of the pedestrian road networks.

Quality Criteria	Types and Names of Metro Stations
	Residential S.		Commercial S.		Public-Service S.		Comprehensive S.
	Jinshi Bridge	South Hongqi Road	Heping Road	Yingkoudao	Culture Centre	Zhoudeng Memorial Hall	Tianta	Changhong Park
Accessibility	0.29	0.31	0.36	0.33	0.34	0.34	0.32	0.33
Convenience	0.12	0.14	0.16	0.14	0.15	0.15	0.13	0.15
Safety	0.24	0.26	0.30	0.28	0.28	0.28	0.27	0.28
Comfort	0.14	0.15	0.17	0.15	0.16	0.16	0.16	0.16
Overall	0.80	0.85	0.99	0.90	0.92	0.94	0.90	0.91
Ranks	8	7	1	5	3	2	5	4

Note: S. is the abbreviation of station.

reports the quality evaluation of pedestrian networks within the 600 m catchment areas of the studied metro stations. The network around the Heping Road station achieves the highest quality score, followed by those near the ZhouDeng Memorial Hall and Culture Centre stations. The networks around the Changhong Park, Tianta, and Yingkoudao stations are rated as moderate in quality, while those near the South Hongqi Road and Jinshi Bridge stations rank the lowest. Meanwhile, the overall quality scores of pedestrian networks around public-service and comprehensive stations show smaller differences compared to those around commercial and residential stations. Among the four key criteria, accessibility scores the highest, followed by safety, while convenience and comfort score the lowest. This is reasonable, as accessibility and safety encourage walking to the stations while convenience and comfort enhance the willingness to do so.

4.3. Verification of Quality Evaluations

Table 5. Results of the human–machine adversarial scoring.

	Types and Names of Metro Stations
	Residential S.		Commercial S.		Public-Service S.		Comprehensive S.
	Jinshi Bridge	South Hongqi Road	Heping Road	Yingkoudao	Culture Centre	Zhoudeng Memorial Hall	Tianta	Changhong Park
Scores	63.88	57.50	72.60	67.75	65.44	70.45	59.43	64.25
Ranks	6	8	1	3	4	2	7	5
Changes in ranks	↑2	↓1	0	↑2	↓1	0	↑2	↓1

Note: S. is the abbreviation of station.

presents the results of the human–machine adversarial scoring. The overall ranking remains stable, with only minor variations. The rankings for pedestrian network quality at the Heping Road and ZhouDeng Memorial Hall stations remain unchanged. However, the rankings for the South Hongqi Road, Cultural Centre, and Changhong Park stations have each dropped by one place, while the rankings for the Jinshi Bridge, Yingkoudao, and Tianta stations have risen by two places.

The cosine similarity between the human–machine adversarial scoring results and the quality evaluation outcomes is 0.999. This exceptionally high value suggests that the evaluation results are highly accurate and that the quality assessment is reliably robust.

5. Discussions and Conclusions

This study evaluated the quality of pedestrian networks for walking access to metro stations by employing 19 indicators categorized into four core criteria: accessibility, convenience, safety, and comfort. To achieve this, we developed an assessment model that integrates the AHP with the EWM, validated using human–machine adversarial scoring and cosine similarity to ensure reliability. The model was applied to eight representative metro stations in Tianjin, China, aiming to access the quality of walking routes leading to these stations. The research drew on diverse data sources, including export scoring data, detailed information on walking routes approaching the stations, and micro-scale walking environmental data.

This study bridges a critical gap in pedestrian network quality assessment by integrating objective and subjective measures into a unified model. The model not only demonstrates robust reliability but also provides highly practical utility for urban renewal. Additionally, this research establishes a comprehensive criterion-indicator system for evaluating the quality of pedestrian networks. For the studied metro stations, the accessibility indicators and two safety indicators, including the proportion of protected routes and the completeness of pedestrian crossing facilities, are identified as crucial factors in quality evaluation weights. Conversely, indicators related to convenience and comfort are weighted less heavily. Our findings also reveal that pedestrian networks around public service and comprehensive stations score higher in quality compared to those around residential stations, with more minor variations being observed in their quality scores.

The model, including the criterion-indicator system, is an effective tool for urban planners and policymakers, enabling them to evaluate the quality of walking environments around metro stations and allocate regeneration funding more effectively. By identifying the most influential indicators, planners and policymakers can prioritize targeted improvements in assessed station areas to enhance overall walkability. Lower-weighted indicators can serve as secondary considerations during the initial stages of urban regeneration. Additionally, stations with more minor variations in quality scores can act as benchmarks for extending improvements to other areas. These efforts are expected to facilitate metro access and enhance the overall performance of the metro system.

In this study, we adopted 19 indicators across four criteria (accessibility, convenience, safety, and comfort) to quantify the quality of pedestrian networks within the 600 m catchment areas of metro stations. Future research could expand upon our findings by incorporating more refined indicators, like walking speeds [84] and sidewalk slopes [74]. Additionally, we designated the centre points of traffic zones around metro stations as the origins, generating the walking routes between these origins and destination stations. Future studies utlizing revealed route choice data and GPS trajectories [85] between accurate buildings and metro stations are encouraged.