Quality Assessment of Cycling Environments Around Metro Stations: An Analysis Based on Access Routes

Abstract

Cycling significantly contributes to improving metro accessibility; however, the quality of bicycle environments surrounding metro stations remains insufficiently studied. This study develops a criteria–indicators assessment framework that incorporates both objective characteristics of bicycle infrastructure and subjective perceptions of bicycle access to metro stations. The framework consists of four primary criteria—accessibility, convenience, safety, and comfort—along with eighteen sub-level indicators. Taking central Tianjin as the study area, the study evaluated the cycling environment quality around eight representative metro stations by employing information entropy and the analytic hierarchy process, with cosine similarity used to compare the outcomes against human–machine adversarial scoring result to ensure analytical robustness. The findings reveal substantial disparities in cycling infrastructure, with safety and accessibility exhibiting higher scores than convenience and comfort. Additionally, cycling environment quality is higher around comprehensive and public-service stations compared to residential stations, while commercial stations exhibit the lowest quality. The study underscores the necessity of expanding protected bike lanes, enhancing route directness, and improving parking and wayfinding facilities to promote cycling as an effective first- and last-mile metro access mode.

1. Introduction

Over the past decades, numerous cities worldwide have swiftly expanded their metro systems to ease traffic congestion, reduce parking demand, and mitigate the environmental externalities of excessive car use [1,2,3]. Like other forms of public transit, the metro system involves access and egress trips to and from stations, which significantly influence the metro’s overall attractiveness. These trips can be made using various modes, with cycling emerging as one of the key options due to the rise of bike-sharing systems [4]. Nevertheless, although metro networks have expanded rapidly, the bicycle infrastructure around stations has received insufficient attention and cannot meet the requirements of bicycling as an effective link to or from metro stations [5,6]. Assessing the quality of bicycle environments in these areas is crucial for improving cycling access, maximizing metro ridership, and supporting long-term urban development.

Previous studies on cycling environments within the metro station catchment areas primarily concern criteria that align with cyclist needs and relative indicators associated with the built environments [7,8]. A key aspect of these studies is cycling accessibility, which plays a crucial role in metro ridership [9,10]. Scholars have examined both spatial and psychological factors, such as suitable cycling distances and perceived satisfaction [11,12]. Attributes like designated bike lanes, road segment length and density, intersection density, land use diversity, as well as the number of bus lines and stops, have been regarded as critical factors impacting bicycle accessibility [13,14]. Moreover, the directness of cycling routes has also been recognized as an essential indicator influencing metro users’ choice to access stations by bicycle [15].

Beyond accessibility, scholarly explorations have extensively examined the safety and desirability of cycling within the metro’s service areas. The presence of unsafe road conditions serves as a deterrent to cyclists [16], while infrastructural enhancements, such as street lamps, bicycle parking facilities, and landscape aesthetics, increase the overall appeal of cycling [17]. Prior research suggested that bike-sharing users experience anxiety regarding the potential unavailability of bicycles when needed or the inadequacy of parking spaces upon trip completion. In contrast, private bicycle users prioritize the security of parking [18,19,20]. Additionally, the convenience and comfort of cycling significantly shape individuals’ preferences to utilize bicycles for first- and last-mile connectivity to metro stations. A study investigating the determinants of bicycle–railway station integration in the Netherlands demonstrated that perceived network connectivity and the qualitative characteristics of cycling infrastructure play a pivotal role in choosing cycling as an access mode [18]. Similarly, a study conducted in Nanjing, China, found that the perceived adequacy and comfort of cycling networks, along with the availability of safe parking slots, exhibit strong and statistically significant correlations with cycling to metro stations [21].

Despite the progress in prior research, three critical gaps persist. First, many studies treat accessibility and safety as fundamental prerequisites for making cycling feasible. As a result, criteria such as enjoyment and comfort are often reviewed as secondary concerns, considered only after basic needs are met [16,22,23]. However, limited research has systematically examined these fundamental and secondary criteria thoroughly. Second, some commonly used indicators fail to properly account for cycling infrastructure. For example, safety is often measured in terms of street crime, rather than road conditions, thereby capturing general security rather than the specific safety of cycling environments [18,24]. Third, the quality of bicycle networks depends not only on physical infrastructure but also on how cyclists perceive them [17,25], yet few studies have comprehensively integrated both objective and subjective perspectives into their analyses.

To bridge these research gaps, our research proposes that four primary criteria—accessibility, safety, convenience, and comfort—along with their corresponding sub-level indicators, influence the bicycle network quality surrounding metro stations from both objective and subjective perspectives. To validate this hypothesis, the study employed a comprehensive assessment model that combined information entropy and the analytic hierarchy process to facilitate a systematic evaluation of bicycle facility quality, with human–machine adversarial scoring and cosine similarity applied to validate the results. Eight metro stations, categorized into four typologies, within the central area of Tianjin, China, are selected as the study cases. By providing a quality assessment of bicycle environments near the eight chosen stations, this research seeks to support urban policymakers and planners in comprehending and enhancing bicycle–metro integration.

The paper is organized into the following sections. Section 2 reviews the literature related to this study. Section 3 presents the study area and data sources and introduces the research methods employed. Section 4 interprets the assessment results, with conclusions and implications for planning policies and future perspectives discussed in Section 5.

2. Literature Reviews

2.1. Bicycle Environment and Cycling Access

Numerous studies have established the correlations between the built environment for cycling and patterns of bicycle-based access to metro stations [26]. The cycling accessibility of metro stations, which quantifies the extent to which individuals can traverse via bicycle between their origin and a station, or vice versa, is regarded as one of the fundamental determinants of metro ridership [23]. Investigations into metro accessibility have analyzed the optimal spatial and temporal thresholds for station reachability, demonstrating that suboptimal cycling accessibility is often associated with diminished metro usage [9,10]. To enhance the spatial coverage of metro services, scholars have examined both physical and cognitive dimensions of bicycle-based access and egress distances [11,12]. For instance, Gan et al. (2020) employed cycling distance metrics alongside subjective satisfaction indices to assess bicycle accessibility within metro catchment areas in Nanjing, China [25]. Furthermore, micro-scale urban design elements and built environment attributes exert significant influence on bicycle accessibility around metro stations [26,27]. Key infrastructural parameters, including the width of cycling lanes, the spatial density of road networks and intersections, as well as route directness, have been identified as critical determinants that significantly contribute to enhancing the bicycle-friendliness of metro station environments [13,14,15].

Previous analyses have also underscored the crucial importance of cycling safety considerations [28,29]. The safety dimension encompasses the interplay between the structural and operational characteristics of cycling infrastructure. The availability and quality of dedicated bike lanes are instrumental in ensuring a safer cycling experience. Among various infrastructure designs, protected bike lanes—designed to provide physical separation between cyclists and motorized traffic—have been empirically validated as the most effective intervention for augmenting both perceived and actual cyclist safety [16]. Additionally, the implementation of secure and exclusive bicycle parking facilities within transit-oriented environments has been demonstrated to facilitate increased bicycle-to-metro integration [18]. A study conducted in Xi’an, China, revealed that users of bike-sharing systems experience psychological distress concerning the potential unavailability of bicycles on demand and the insufficiency of designated parking spaces at metro station termini [20].

Cycling convenience and comfort further modulate the preferences toward cycling access or egress [29]. A GPS trajectory-based study in Seattle, USA, revealed that convenience is a significant determinant of cyclist route selection, with most cyclists exhibiting a preference for routes characterized by minimal elevation variation, reduced travel distance, and well-integrated cycling infrastructure with low vehicular traffic intensity. Additionally, some cyclists demonstrate a preference for routes featuring adequate street illumination and diverse land-use patterns [30]. Research on the determinants of bicycle–rail integration in the Netherlands has highlighted that perceived network connectivity, along with qualitative aspects of cycling infrastructure, exerts a significant influence on the adoption of cycling as an access mode to railway stations [18]. Similarly, an empirical investigation in Nanjing, China, identified strong and statistically significant associations between perceived network adequacy, cycling comfort, and the availability of secure bicycle parking, all of which contribute to individuals’ propensity to cycle to metro stations [21]. Despite the recognition by researchers of the necessity for multi-dimensional criteria that encapsulate cyclists’ demands in assessing cycling environments surrounding metro stations, fundamental criteria such as accessibility and safety serve as essential prerequisites for cycling feasibility, whereas those such as enjoyment and comfort function as supplementary elements that enrich the overall cycling experience [16,22,23]. However, limited scholarly investigations have comprehensively analyzed these criteria in conjunction. Consequently, this study addresses the existing research gap by systematically evaluating the quality of cycling environments around metro stations through the lenses of cycling accessibility, safety, convenience, and comfort, with the objective of advancing the understanding of cyclist behavior in relation to metro accessibility.

2.2. Quality Assessment of Bicycle Environment

Bikeability can be evaluated using various established methods, each applying specific criteria to assess cycling conditions. Several models and indices have been developed to classify bikeability. For instance, the Dutch Design Manual for Bicycle Traffic outlines five key criteria for a high-quality cycling network: coherence, directness, attractiveness, safety, and comfort [31]. The Copenhagenize Index, developed by the Copenhagenize Design Company, uses 14 categories to score cities from zero to four points, with additional bonus points awarded for exceptional efforts. These categories include factors such as NGO lobbying, cycling culture, infrastructure, bike-sharing systems, gender and modal splits, perceived safety, political support, social acceptance, urban planning, traffic reduction, and bicycle logistics [32]. Another commonly used method is the Bicycle Level of Service, which focuses on assessing the accessibility and comfort of urban cycling environments [33]. However, because the sub-level indicators vary in the criteria they include and how they weigh them, their results may differ even when applied to the same city.

Traditionally, data on cycling environments have been gathered through field surveys, questionnaires, and video recordings [34,35,36,37], methodologies that are both time- and labor-intensive, thereby limiting their efficiency in capturing the user-friendliness of cycling infrastructure. The availability of online multi-source open data has facilitated large-scale quantitative analyses for evaluating the quality of bicycle networks, enabling a more comprehensive and systematic understanding [38]. For instance, researchers have investigated the effects of urban design characteristics and transportation network metrics on bicycle usage through trajectory data analysis [39,40]. Gu et al. (2018) employed OpenStreetMap road networks, street view images, and points of interest data to evaluate the bicycle-friendly degree of road networks in four Chinese cities, namely Tianjin, Chongqing, Kunming, and Shijiazhuang [41]. While these studies have integrated multi-source data to enhance assessments of bicycle-friendliness, their analyses remain constrained to individual road segments and are predominantly conducted from an objective perspective.

Regarding evaluation methodologies, statistical approaches are commonly employed to analyze the influence of cycling environment-related indicators on bicycle-friendliness through regression models and other analytical frameworks. For instance, previous studies have utilized linear regression [27,35], principal component analysis (PCA) [31,42], and generalized additive mixed models [43] to examine the associations between cycling frequency, trip duration, safety, comfort, and built environment characteristics. While these methodologies effectively identify key influencing factors, their applicability in practical implementation remains challenging. An alternative approach involves assigning weights to criteria and indicators to assess their relative importance. The analytic hierarchy process and analytic network process are widely used to determine subjective weightings for criteria and indicators. For example, Li et al. (2012) developed an assessment framework for cycling comfort perception using an analytic network process incorporating 16 indicators related to road geometric design, and environmental and traffic conditions [44]. Additionally, the information entropy weighting method is frequently adopted as it mitigates the subjectivity and arbitrariness associated with manually assigned weights while preventing information redundancy among variables. Gu et al. (2018) applied the information entropy method to derive street scores for bikeability across four Chinese cities—Tianjin, Chongqing, Kunming, and Shijiazhuang [41]. Furthermore, a discrete choice model was employed to assess and rank the significance of various cycling infrastructure-related factors influencing bikeability in Barranquilla, Colombia.

Although prior studies have offered high-resolution quantitative assessments and hold significant potential for facilitating the development of targeted and effective strategies to improve cycling infrastructure and promote cycling, they are inherently limited by their emphasis on either objective or subjective assessment methodologies, rather than integrating both perspectives comprehensively.

3. Research Design and Method

3.1. Method

3.1.1. Modelling Design

This study evaluates the bicycle network quality around metro stations by integrating the information entropy and analytic hierarchy process, with cosine similarity used to compare the outcomes against human–machine adversarial scoring results. The modeling process consists of five main parts: (1) Filtrating evaluation indicators. Based on the literature review and three rounds of filtrations, 18 sub-level indicators for the four criteria to measure cycling environments were selected. (2) Calculating objective weights. The Information Entropy was used to assess the indicators’ objective weights. (3) Calculating subjective weights. The Analytic Hierarchy Process was used to assess the indicators’ subjective weights. (4) Calculating composite weights. The composite weights were calculated based on objective and subjective weights. (5) Validating evaluation results. The cosine similarity was used to compare the outcomes against human–machine adversarial scoring results to ensure analytical robustness.

3.1.2. Filtration of Evaluation Indicators

The analysis started by reviewing 40 relevant studies [3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,22,23,24,25,26,27,28,29,34,35,36,37,38,39,40,41,42,43,44,45,46] on bicycle network quality published since 2015, sourced from the China National Knowledge Infrastructure and Google Scholar database, which led to the identification of 29 potential indicators. In the first round of refinement, 8 indicators with low frequency were removed. In the second round, the expert’s scoring data collected by the Likert scale method were used to calculate each indicator’s Kendall’s coefficients of harmony, coefficients of variation, and mean values. Two indicators, traffic marking completeness and monitoring facility density, were eliminated because of low mean values and high variations. In the third round, Pearson’s correlation coefficients were calculated, leading to removal of the commercial facility density indicator, which had strong associations with the daily service facilities’ density and diversity. A total of 18 indicators were finally selected for further analysis. The above steps were performed in SPSS 22, with 18 indicators selected for further analysis.

3.1.3. Objective Weights of Indicators

The Information Entropy was used to determine the objective weights of indicators by analyzing information entropy, utilizing multi-source big data along with field investigation data. Firstly, to ensure comparability, the data were standardized and normalized, and an evaluation weight matrix was constructed as follows:

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

(1)

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

(2)

In the equations, m and n are separately the number of cycling paths and indicators; and A_ij denotes the indicator j (from 1 to n) of the cycling path i (from 1 to m).

Secondly, the indicators’ information entropy was computed by Equation (3) to measure its degree of dispersion. Thirdly, the entropy weights of both indicators and criteria were calculated through Equation (4) based on the proportion of their corresponding entropy values. The above steps were performed using Pycharm.

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

(3)

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

(4)

In the equations, H_j is the indicator j’s entropy; m is the number of cycling paths; and EW_j is the indicator j’s weight.

3.1.4. Subjective Weights of Indicators

The Analytic Hierarchy Process was applied to assign subjective weights for indicators based on expert opinions gathered through surveys. Firstly, a hierarchical framework was structured, with the evaluation of bicycle network quality as the top level, the four criteria as the middle level, and the 18 indicators as the bottom level. Then, pairwise comparisons were conducted within the same criterion to assess the relative importance of indicators using the expert’s scoring data collected by the 1–9 scale method; see Equation (5):

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

(5)

where λ_max and W_i are the matrix C’s maximum eigenvalue and the index’s normalized eigenvector, separately; n represents the matrix C’s layers; and CI, CR, and RI are the consistency index, coefficient’s verification, and random consistency index, respectively.

Next, the consistency of pairwise comparison matrices was performed by Equation (6) to ensure logical consistency. Finally, the weights of both indicators and criteria were derived from eigenvectors of the judgment matrices via Equation (7). The above steps were performed through the YAAHP 12.11 software platform.

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

(6)

In the equation, X_ij and X_ijn are the indicators i’s and j’s final and n-th relative importance.

$${AW}_i={AW}_{c, i}\times {AW}_{ i, c}$$

(7)

where AW_i represents the indicator i’s final subjective weight, and AW_c,i and AW_i,c are the criterion’s weight and the indicator i’s weight, separately.

3.1.5. Composite Weights of Indicators

Indicators’ composite weights were computed by combining the objective and subjective weights through Equation (8). Considering the number of cyclists on each route, the quality of the bicycle network was assessed by Equation (9).

$${CW}_i={\displaystyle \frac{\sqrt{{EW}_i{AW}_i}}{{\sum }_{i=1}^{18}\sqrt{{EW}_i{AW}_i}}}$$

(8)

where CW_i, EW_i, and AW_i are the i-th indicator’s composite, objective, and subjective weights, separately.

$$\left\{\begin{array}{c}{CWF}_{1,m}={\sum }_{j=1}^n{\displaystyle \frac{r_{mj}}{R_m}}{CW}_{1,mj}\\ {CWF}_{2,m}={\sum }_{j=1}^n{\displaystyle \frac{r_{mj}}{R_m}}{CW}_{2,mj}\\ ⋮\\ {CWF}_{i,m}={\sum }_{j=1}^n{\displaystyle \frac{r_{mj}}{R_m}}{CW}_{i,mj}\end{array}\right.$$

(9)

where CWFi,m represents the value of the CWi indicator to metro station m; n denotes the number of cycling paths leading to metro station m; CWi,mj indicates the value of the CWi indicator for path j to metro station m; rmj refers to the cyclist number on the path j to metro station m; and Rm is the total cyclist number traveling to metro station m.

3.1.6. Composite Weights of Criteria

On the basis of indicators’ composite weights, we computed the weights of each criterion through Equation (10):

$${CrW}_i={\sum }_{i=1}^r{CWF}_i$$

(10)

In the equation, CrWi is the criterion i’s weight, where i ranges from 1 to 4 (representing accessibility, safety, convenience, and comfort); and CWFi is the sub-indicator i’s final composite weight; r is the number of indicators.

3.1.7. Validating Evaluation Results

To validate the evaluation results, 85 experts evaluated street view images taken along the cycling paths, and their evaluations were used to train the human–machine adversarial model developed by China University of Geosciences [47] for training. The mean score of all images for each station was calculated and compared with the evaluation outcomes using cosine similarity.

3.2. Study Area

Tianjin is a directly administered municipality and part of the Jing–Jin–Ji megapolis, serving as a key national central city along the shore of the Bohai Sea. Its metropolitan area consists of sixteen districts, with six forming the central urban area. These six districts are our study area, with a 3.91 million population over 60.20 square kilometers in 2023. According to the traffic survey conducted by the Transportation Commission of Tianjin, from 2000 to 2022, the proportion of walking travel remained basically unchanged, rising slightly from 34.70% to 35.00%. In contrast, bicycle travel saw a continuous decline from 53.40% to 19.40%. During the same period, car travel experienced a sharp rise from 3.20% to 28.20%, while public transportation usage increased from 8.70% to 14.8%, with the metro accounting for 5.70% [48]. The rapid growth of private motorized transport in the central area has resulted in increased congestion and parking shortages. In response, the government has prioritized the expansion of the metro network and operated six metro lines with 87 available stations by the end of 2023.

Distance plays a crucial role in shaping individuals’ preferences and behavior when choosing cycling as an access mode. Previous studies have identified the cycling access radius around metro stations in Chinese cities to typically range from 800 to 1600 m [49,50]. Given the high concentration of metro stations in central Tianjin, a smaller cycling radius of 800 m is sufficient to encompass the primary service areas of the stations; see Figure 1. Therefore, an 800 m catchment area was used for subsequent analysis.

Shaped by historical evolution and rapid urbanization, the central area exhibits diverse spatial layouts and land use patterns, creating distinct urban environments around metro stations. In order to explore the diverse typological characteristics of cycling environments near these stations and make the evaluation results more representative, it is necessary to classify the stations to select typical ones for evaluation. According to the “Urban Land Classification and Planning Construction Standards”, K-Means clustering was employed to calculatethe areas of different types of urban construction lands, including lands for residence, public facilities, commerce, industry, transportation, municipal utilities, green space, and other purposes, in an 800 m radius around each station. Through this process, the 87 stations were categorized into six types, including residential, public service, commercial, transportation, comprehensive functions, and others.

Table 1. Results of land clustering around the metro stations.

Type of Station	Ratio of Total Constructed Area (%)					Number of Station
	Ratio of Residential Land	Ratio of Land for Public Facilities	Ratio of Commercial Land	Ratio of Land for Transportation	Ratio of Land for Other Purposes
Residential Station	43.15	11.24	20.78	15.21	9.62	35
Public-service Station	20.65	37.97	9.85	15.24	16.29	9
Commercial Station	25.12	11.28	35.17	17.97	10.46	19
Transport Station	12.84	10.24	6.16	47.15	23.61	4
Comprehensive Station	22.76	16.34	17.45	17.22	26.23	13
Other Station	9.45	3.58	4.46	13.24	69.27	7

presents these categories and the corresponding proportions of lands for different usage. Notably, transport stations function as hubs in train and high-speed railway stations and the airport, while the “others” category includes stations that are either under construction or just planned.

To account for station diversity and data availability, we selected eight stations from the four primary categories—residential, public service, commercial, and comprehensive functions—as study sites. Each station was examined within an 800 m radius to provide a comprehensive understanding of cycling conditions in Tianjin’s central metro areas.

3.3. Data Sources

3.3.1. Open Data

This study employed five types of online big data: bicycle network data, AMap’s walking routes data, building height data, Points of Interest (POI) data, and street view images.

A bicycle road network offers cycling routes to metro stations. To acquire this infrastructure, we extracted bicycle-related road facilities from OpenStreetMap using the Overpass API and imported the data into ArcGIS (version: 10.8). By removing unrelated roads and connecting inconsistencies, we obtained a network with detailed information such as road names, classifications, lengths, and widths.

AMap’s route planning Application Program Interface (API) provides cycling route queries along with detailed cycling distance and estimated travel time for each route. Using Dijkstra’s greedy algorithm, the API suggests the most optimal cycling routes by analyzing and prioritizing road conditions, such as the most commonly used paths, routes with fewer red lights, shortest cycling time, and road congestion. To enhance route planning efficiency, we divided the 800 m catchment areas of metro stations into traffic zones based on land use and population activity intensity. For each traffic zone, the center point was used as the origin, while the entrance of the corresponding metro station served as the destination. We then input the coordinates of these locations into the route planning API through PyCharm (version: 2024.3.2) and generated 61 cycling routes, see Figure 2. These routes were loaded into the bicycle network map in ArcGIS and visualized with detailed cycling data such as total distance, estimated time, names of the street segments, and turning instructions.

The building height data, which reflect the intensity of area development around the metro stations, was adopted to evaluate cycling volumes on the access paths. These data were sourced from BuildingHeightModel (https://github.com/lauraset/BuildingHeightModel, accessed on 27 January 2021), published by Wuhan University, and were estimated by a multi-task deep learning network that integrates multiple spectra from multiple perspectives [51]. For instances of missing data, we supplemented the information with values crawled from Baidu 3D Map API.

POI data, a reliable source reflecting land use and function characteristics of the station’s catchment area, were collected from Baidu Map API and mapped on the bicycle network in ArcGIS, with attributes including name, address, longitude, latitude, and type. According to travel demands and attractions, a total of nine categories were selected, namely catering services, corporations and enterprises, shopping and consumption, financial institutions, hotels and accommodations, science and educational services, life services, leisure and entertainment, and medical care.

Street view images provide visual features of cycling environments around metro stations. Panoramic static images captured in four directions (0°, 90°, 180°, and 270°) of the 61 cycling routes at intervals of 30 m were downloaded from Baidu Map API. A semantic segmentation model based on a fully convolutional network was utilized to detect elements such as sky, greenery, roads, and motor vehicles, and calculate their corresponding proportions. The proportions were projected onto the cycling routes in ArcGIS. Table 1 lists the descriptions of the eighteen indicators.

3.3.2. Survey Data

This study applied two types of survey data, including expert opinion survey data and field investigation data.

The expert opinion survey, conducted in October 2023, comprised three sub-surveys. The first part collected experts’ perceived importance of indicators. To validate these indicators for modeling, experts evaluated their significance using the Likert scale method, assigning scores from 1 (not important at all) to 5 (extremely important). A total of 135 professionals in urban planning responded, yielding a response rate of 98.54%. The second component involved pairwise comparisons of indicator importance within the same criterion. In this phase, a total of 51 urban planning experts rated each indicator’s relative importance using the 1–9 scale method developed by Thomas Saaty. The third component evaluated bikeability based on street-view images captured on cycling paths to metro stations. A total of 85 urban planning experts participated in assessing these images.

The field investigation aimed to gather additional data on cycling environments, including route guide signs, quality and slope of bike lanes, and cycling congestion. Conducted in November 2023, the investigation was carried out with the assistance of four trained postgraduate researchers.

Based on the open data and field investigation results, we obtained the values of evaluation indicators (

Table 2. Objective descriptions of the indicators.

Criteria	Indicators	Formulas	Descriptions
Accessibility	Width of bike lanes	${WBL}_i={\displaystyle \frac{{\sum }_{j=1}^n{WBL}_j}{n}}$	WBLi is the width of bike lanes along path i; WBLj is the width of section j along the route; and n is the total number of sections.
	Bicycle route directness	${BRD}_i={\displaystyle \frac{L_i}{D_i}}$	BRDi is the bicycle route directness of route i; Li is the actual length of the cycling route; and Di is the Euclidean distance between the origin and station of route i.
	Degree of interruptions	${DI}_i={\displaystyle \frac{{SUM}_{i,I}}{L_i}}$	DIi is the interruption degree of route i; SUMi,I is the number of interruptions along route i; and Li is route i’s actual length.
Safety	Percentage of bike lanes	${PBL}_i={\displaystyle \frac{{BL}_{i,I}}{L_i}}\times 100\%$	PBLi is the proportion of bike lane along route i; and BLi,I and Li are the length of bike lane and the actual length along route i, separately.
	Percentage of protected bike lanes	${PPB}_i={\displaystyle \frac{{PBL}_{i,p}}{L_i}}\times 100\%$	PPBi is the rate of protected bike lanes along route i; PBLi,P is the length of protected bike lanes along route i; and Li is route i’s actual length.
	Encroachment of bike lanes	field investigation	Whether the bike lanes along the cycling routes to stations are occupied by motor vehicles or bicycles, scored from 1 to 5, with 1 representing unappropriated bike lanes and 5 indicating encroached ones.
	Percentage of bicycle routes disturbed by motor vehicles	${PBM}_i={\displaystyle \frac{1}{n}}{\sum }_{j=1}^n{\displaystyle \frac{S_{ij,I}}{S_{ij}}}\times 100\%$	PBMi represents the proportion of bicycle routes disturbed by motor vehicles for path i; Sij,I denotes the road area occupied by motor vehicles in street view image j of path i; Sij refers to the road area of street view image j for path i; and n is the total number of street view images for path i.
	Density of entrances and exits	${DE}_i={\displaystyle \frac{{SUM}_{i,E}}{L_i}}$	DEi is the density of entrances and exits along the i-th route; SUMi,E is the number of entrances and exits along route i; and Li is route i’s length.
Convenience;	Density of bicycle parking spots	${DBP}_i={\displaystyle \frac{{SUM}_{i,I}}{L_i}}$	BBPi is the density of bicycle parking points for route i; SUMi,I is the number of parking points along the way for route i; and Li is route i’s actual length.
	Route guide signs	field investigation	Whether the layout and number of guide signs meet the needs of cycling to stations, scored from 1 to 5, with 1 representing a fuzzy guiding system and 5 indicating a clear one, respectively.
	Density of daily service facilities	${DSF}_i={\displaystyle \frac{F_i}{L_i}}$	DSFi is the density of daily service facilities along route i; Fi is the number of facilities along route i; and Li is route i’s actual length.
	Diversity of daily service facilities	${MSF}_i=-{\sum }_{j=1}^n\left({\displaystyle \frac{P_{i,j}}{{\sum }_{j=1}^n{P}_{i,j}}}*\mathit{ln}{\displaystyle \frac{P_{i,j}}{{\sum }_{j=1}^n{P}_{i,j}}}\right)$	MSFi is the diversity of daily service facilities along route i; Pi,j is the number of POI points of type j along route i; and n is route i’s total types of POI.
Comfort	Quality of bike lanes	field investigation	Whether the quality of bike lanes meets cyclists’ demands, scored from 1 to 5, with 1 representing a low quality of bike lanes and 5 indicating high quality, respectively.
	Slope of bike lanes	field investigation	Whether the bike lanes are steep or flat, scored from 1 to 5, with 1 representing flat and 5 indicating steep, separately.
	Cycling congestion	field investigation	Feelings of overcrowding when cycling to metro stations because of narrow bike lanes or an excessive number of bikes, scored from 1 to 5, with 1 representing an uncrowded environment and 5 indicating a crowding one, respectively.
	Percentage of routes with street lamps	${PRS}_i={\displaystyle \frac{{LST}_i}{L_i}}$	PRSi is the street lamps’ proportion along route i; LSTi,L is the length of route i with street lamps; and Li is route i’s actual length.
	Greenery	$G_i={\displaystyle \frac{1}{n}}{\sum }_{j=1}^n{\displaystyle \frac{S_{ij,G}}{S_{ij,T}}}\times 100\%$	Gi is the greenery of route i; Sij,G is the green area in street view photo j of route i; Sij,T is the total area of street view photo j of route i; and n is the total number of street view photos in route i.
	Openness of vision during cycling	${SV}_i={\displaystyle \frac{1}{n}}{\sum }_{j=1}^n{\displaystyle \frac{S_{ij,S}}{S_{ij,T}}}\times 100\%$	SVi is the openness of vision when bicycling along route i; Sij,S is the sky area in street view photo j of route i; Sij,T is the total area of street view photo j of route i; and n is route i’s total number of street view photos in route i.

) to compute their objective weights. Simultaneously, expert opinion survey data were used for calculating the indicators’ subjective weights.

4. Results

4.1. Quality Evaluations Based on Cycling Routes

4.1.1. Accessibility

Table 3. Results of quality evaluations of bicycle routes.

Criteria	Indicators	Residential Station		Public-Service Station		Commercial Station		Comprehensive Station
		South Hongqi Road	Jinshi Bridge	Zhoudeng Memorial Hall	Culture Centre	Yingkoudao	Heping Road	Changhong Park	Tianta
Accessibility	Width of bike lanes	2.24	2.37	2.31	2.29	1.77	1.79	2.19	2.25
	Bicycle route directness	1.51	1.64	1.67	1.50	1.44	1.50	1.58	2.05
	Degree of interruptions	0.29	0.27	0.33	0.32	0.37	0.39	0.30	0.34
Safety	Percentage of bike lanes	71.88	45.57	83.27	46.63	53.99	57.09	75.26	69.23
	Percentage of protected bike lanes	59.67	7.44	52.93	14.55	12.30	13.65	50.65	20.27
	Encroachment of bike lanes	2.20	2.88	2.18	2.55	3.62	3.28	2.96	2.85
	Percentage of bicycle routes disturbed by motor vehicles	18.96	17.56	16.96	14.84	15.07	14.34	18.68	22.82
	Density of entrances and exits	0.61	0.67	0.50	0.55	0.87	0.77	0.97	0.87
Convenience	Density of bicycle parking spots	0.44	0.45	0.46	0.43	0.37	0.33	0.38	0.36
	Route guide signs	1.82	2.50	2.78	3.21	2.88	2.73	1.87	1.80
	Density of daily service facilities	7.25	7.36	6.49	5.41	9.82	11.04	8.18	8.77
	Diversity of daily service facilities	0.28	0.34	0.15	0.19	0.39	0.41	0.21	0.30
Comfort	Quality of bike lanes	3.06	2.93	2.96	3.26	2.61	2.49	3.14	3.14
	Slope of bike lanes	1.46	2.03	1.14	1.67	1.51	1.32	1.50	1.52
	Cycling congestion	2.58	2.74	1.97	2.54	2.89	3.47	2.24	2.65
	Percentage of routes with street lamps	85.75	81.04	91.80	85.33	95.25	94.14	84.22	89.02
	Greenery	13.74	15.19	18.95	15.98	20.49	19.30	10.82	11.36
	Openness of vision during cycling	28.29	23.56	27.66	25.72	16.19	15.85	27.34	24.26

displays the outcomes of quality assessments for bicycle paths located within an 800 m radius of eight stations analyzed. Bike lane widths vary across the study areas, with the narrowest lanes found near Heping Road station, measuring approximately 1.78 m. This may be attributed to commercial stations being located in Tianjin’s central district, where available space for cyclists is scarce. Furthermore, the bicycle route directness values are the lowest around commercial stations, averaging 1.47. This is probably a result of the detailed land use structure and the presence of small-sized blocks in the central district [52], resulting in more direct cycling routes. In contrast, the value of bicycle route directness around Tianta station is the highest at 2.05, suggesting a less straightforward cycling network in that area.

The degree of interruptions ranges from 0.27 at Jinshi Bridge station to 0.39 at Heping Road station, reflecting variations in route continuity. On average, bicycle routes around residential stations have the lowest frequency of interruptions, approximately 0.28 times every 100 m. In contrast, paths within commercial station catchment areas encounter the highest frequency of interruptions, averaging 0.38 times per 100 m. Public-service and comprehensive station areas exhibit a fairly consistent level of interruption, with values around 0.32 occurrences per 100 m. This aligns with previous research indicating that residential areas generally tend to have larger blocks and wider roads, while commercial areas often characterized by smaller blocks and narrower roads in major Chinese cities [29,52].

4.1.2. Safety

The percentage of bike lanes varies significantly, ranging from 45.57% near Jinshi Bridge station to 83.27% around Zhoudeng Memorial Hall station, indicating substantial differences in cycling infrastructure coverage. However, the proportion of protected bike lanes is much lower. Only the catchment areas of Zhoudeng Memorial Hall (52.93%) and Changhong Park (50.65%) have more than half of their bike lanes protected, while in other areas, the presence of protected bike lanes is considerably lower. In particular, the catchment areas of Jinshi Bridge, Culture Centre, Yingkoudao, and Heping Road stations have less than 15.00% of the total bike lanes protected, with Jinshi Bridge having the lowest proportion at just 7.44%. Furthermore, bike lane encroachment is a common issue across all station types. In commercial station areas, bike lanes are often occupied by stalls and parked shared bicycles, while those around other types of stations are mainly obstructed by shared bicycles. These patterns are consistent with findings from previous studies [11,13,22], which highlight the impact of informal commercial activities and bicycle-sharing programs on cycling space.

The percentage of bicycle routes disturbed by motor vehicles varies across station types, reflecting differences in vehicular interference. Residential and comprehensive station areas tend to experience more disturbance due to the presence of urban arterial and secondary roads, which facilitate motor vehicle traffic. In contrast, commercial station areas have lower disturbance values, as commercial streets and small-scale roads often impose traffic restrictions or speed limitations on motor vehicles.

Entrance–exit density along cycling routes is highest around comprehensive and commercial stations, with Changhong Park station (0.97) and Yingkoudao station (0.87) having the highest scores. In contrast, public-service and residential station areas have lower entrance–exit densities, with Zhoudeng Memorial Hall station (0.50) and Culture Centre station (0.55) scoring the lowest. This pattern suggests that the fine-grained and diverse urban structures for land use in commercial and comprehensive station areas lead to more entrances and exits, a characteristic commonly observed in Chinese cities [52].

4.1.3. Convenience

The density of bicycle parking spots is relatively high near public-service and residential station areas, averaging around 0.45 per 100 m, while it is lower around commercial and comprehensive stations, at approximately 0.36 per 100 m. This difference may be attributed to the need to balance space usage in high-density areas. In commercial and comprehensive stations, where pedestrian and vehicular traffic is more intense, bicycle parking density is strategically reduced to prevent congestion and maintain smooth traffic flow. Consequently, measures such as reducing density are implemented to enhance overall accessibility. Additionally, the placement of parking spots across all station areas remains relatively fixed, primarily concentrated near the entrances of metro stations, residential communities, and business buildings.

The availability of route guide signs varies across different station types. Public-service and commercial station areas have higher route guidance scores, with Culture Centre station (3.21) and Yingkoudao station (2.88) standing out, whereas residential and comprehensive station areas generally have lower scores. This trend suggests that commercial and public-service stations, which attract a higher volume of urban activities due to their diverse land use, are equipped with more comprehensive signage to facilitate bicycle navigation.

Cycling routes accessing commercial station areas offer the greatest concentration and diversity of daily service amenities, with Heping Road station having the greatest concentration (11.04 in density and 0.41 in diversity). This is followed by routes accessing comprehensive and residential stations, which also have relatively high densities and diversities of service facilities, reflecting large populations around these stations. In contrast, public-service station areas exhibit the lowest concentration and diversity of daily service facilities, indicating fewer amenities available along cycling routes in these areas.

4.1.4. Comfort

The quality of bike lanes reflects their smoothness, cleanliness, and level of damage. Among the evaluated stations, the bike lanes in the catchment areas of the Culture Centre station demonstrate the highest quality, followed by the ones near Changhong and Park Tianta stations. The bike lane quality around Yingkoudao and Heping Road stations is relatively poor, likely due to aging infrastructure and insufficient maintenance, leading to issues such as broken pavement and potholes. The slope of bike lanes, which affects cycling comfort and effort, remains relatively stable across most stations, averaging around 1.50. However, Jinshi Bridge station has the highest slope score, likely due to the presence of landscape paths with undulating terrain, making cycling more challenging. Cycling congestion is a subjective experience for cyclists accessing metro stations. Congestion levels remain manageable across most station areas, except for Heping Road station, where heavy commercial and business activities contribute to significant crowding on bike lanes.

Street lighting coverage is highest in commercial station catchment areas, with nearly 95.00% of routes equipped with street lamps, a proportion higher than that of other station types. This trend is expected; on one hand, the high nighttime population density in commercial areas increases the demand for adequate lighting, while on the other hand, well-lit commercial areas enhance cycling activity and support local businesses [53].

Cycling routes accessing commercial stations also have the highest greenery scores, averaging around 20.00%. This could be attributed to the narrower streets surrounding the stations, which feature more roadside vegetation [54]. In contrast, cycling routes near residential and comprehensive stations receive lower greenery ratings, primarily because of the wider street layouts, reducing the concentration of green spaces. Interestingly, the vision openness while cycling is lowest in commercial station areas, averaging around 16.00%. This may be due to the higher proportion of trees and vegetation, which, while enhancing greenery, can also obstruct visibility—a phenomenon observed in other studies [22].

4.1.5. Overall Evaluations

Table 4. Results of overall evaluations of the bicycle routes.

Criteria	Residential Station		Public-Service Station		Commercial Station		Comprehensive Station
	South Hongqi Road	Jinshi Bridge	Zhoudeng Memorial Hall	Culture Centre	Yingkoudao	Heping Road	Changhong Park	Tianta
Accessibility	0.206	0.175	0.204	0.176	0.164	0.169	0.210	0.186
Safety	0.225	0.193	0.225	0.194	0.181	0.186	0.232	0.205
Convenience	0.066	0.058	0.068	0.059	0.055	0.056	0.070	0.062
Comfort	0.080	0.068	0.080	0.069	0.064	0.066	0.082	0.072
Overall	0.577	0.494	0.577	0.498	0.465	0.477	0.595	0.525
Rankings	2	6	2	5	8	7	1	4

presents the overall evaluations of bicycle routes in the 800 m areas of studied stations. Among them, Changhong Park station has the highest-ranked bicycle network, followed by South Hongqi Road, Zhoudeng Memorial Hall, and Tianta stations. Those around Culture Centre and Jinshi Bridge stations fall into the mid-range, while those at Heping Road and Yingkoudao stations receive the lowest rankings. Notably, the overall quality of bicycle networks near commercial stations exhibits smaller variations compared to those around residential, public-service, and comprehensive stations, suggesting more uniform conditions in commercial areas.

Regarding the criteria, safety receives the highest scores across all station areas, followed by accessibility, with comfort and convenience receiving the lowest scores. This distribution makes sense, as safety and accessibility are fundamental factors that promote cycling to metro stations, whereas comfort and convenience play a secondary role by enhancing the overall cycling experience. These findings align with earlier studies, which highlight the significance of safety and accessibility in encouraging bicycle use, while comfort and convenience influence cyclists’ willingness to continue cycling [16,55].

4.2. Verification of Assessments

The outcomes of the human–machine adversarial scoring analysis are shown in

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

	Residential Station		Public-Service Station		Commercial Station		Comprehensive Station
	South Hongqi Road	Jinshi Bridge	Zhoudeng Memorial Hall	Culture Center	Yingkoudao	Heping Road	Changhong Park	Tianta
Scores	39.445	39.032	43.244	35.423	34.667	34.098	42.010	35.004
Rankings	3	4	1	5	7	8	2	6
Ranking changes	↓1	↑2	↑1	0	↑1	↓1	↓1	↓2

Note: ↑ and ↓ indicate that the ranking is rising and falling down, respectively.

. In general, the ranking remains fairly consistent, exhibiting only slight fluctuations. The ranking of the bicycle road network quality at the Cultural Center station remains unchanged. In contrast, the rankings for ZhouDeng Memorial Hall, Yingkoudao, and Jinshi Bridge stations have increased, with the first two stations advancing by one position and the latter by two positions. Conversely, the rankings for South Hongqi Road, Heping Road, Changhong Park, and Tianta stations have declined, with the first three stations decreasing by one position and the last by two positions.

The computed cosine similarity between the quality evaluation results and human–machine adversarial scoring outcomes is 0.981. This high similarity coefficient indicates a strong alignment between the two assessments, suggesting that the assessments are highly reliable and the evaluation framework is strong.

5. Discussions and Conclusions

This research provides a comprehensive assessment of cycling environments surrounding metro stations in Tianjin, China. By integrating objective data and subjective perceptions, the research highlights key factors influencing bicycle network quality and their implications for urban planning.

The analysis of accessibility, safety, convenience, and comfort across eight metro stations reveals significant disparities in cycling conditions. Accessibility is influenced by bike lane width and route directness, with commercial station areas exhibiting the lowest values due to space constraints. Safety assessments indicate a notable deficiency in protected bike lanes, particularly in commercial and high-traffic areas, resulting in frequent lane encroachment and vehicular disturbances. While safety receives the highest overall scores among the four key criteria, comfort and convenience show lower ratings, suggesting that basic cycling infrastructure is available but lacks enhancements that improve the overall user experience.

The evaluation results further align with previous studies emphasizing the importance of accessibility and safety in fostering bicycle–metro integration. However, the findings also highlight overlooked aspects such as route interruptions, signage clarity, and cycling congestion, which significantly impact cyclists’ overall experience. The human–machine adversarial scoring verification confirms the robustness of the assessment framework, with high cosine similarity between computed evaluations and expert-reviewed results.

To improve cycling environments around metro stations, urban planners and policymakers should consider a multi-faceted approach. Increasing the proportion of protected bike lanes, especially in commercial and high-traffic areas, will enhance safety and encourage cycling. Redesigning cycling pathways to provide more direct and uninterrupted routes can improve accessibility. Expanding bicycle parking facilities and enhancing wayfinding systems can increase convenience for cyclists. Implementing better road maintenance, greenery, and street lighting can improve the overall cycling experience.

While Tianjin’s metro stations exhibit varying levels of cycling-friendliness, significant opportunities exist to enhance cycling environments through targeted infrastructure improvements. Future research could extend this study by incorporating real-time cycling data (e.g., use shared-bike trajectory data to enable a more accurate assessment of route frequency) and investigating user preferences through behavioral modeling. By adopting a holistic approach, cities can foster more sustainable and integrated urban mobility systems, ensuring that cycling remains an attractive and viable transportation mode for metro access.