Origin-Destination Spatial-Temporal Patterns of Dockless Shared Bikes Based on Shopping Activities and Its Application in Urban Planning: The Case of Nanjing

Abstract

The utilization of dockless shared bikes for shopping purposes has become increasingly prevalent. This research seeks to optimize the configuration of facilities and transportation policies for shared bike travel by analyzing the spatiotemporal patterns of shopping trips from the perspectives of destination (D), origin (O), and O-D correlation in Nanjing’s main city area. As the second-largest commercial center in East China, Nanjing offers a significant context for this research. First, we introduce the “cycling intensity” indicator to analyze the patterns of shared bicycle trips with shopping facilities as destinations at both the subdistrict and road section scales. Second, we utilize spatial autocorrelation analysis and k-means clustering to explore the outflow patterns of shared bicycle trips originating from shopping facilities. Finally, we employ grey correlation analysis to investigate the dynamic flow correlations of shared bicycle O-D trips around various grades of shopping facilities at both subdistrict and road section levels. Concurrently, we endeavored to delineate the practical transformation and application of the research findings. Our results indicate the following: (1) There is a high concentration of cycling intensity around shopping facilities on east–west and north–south roads, with community shopping facilities primarily associated with north–south roads. (2) The outflow of shared bikes from shopping areas can be categorized into four distinct modes. (3) The inflow and outflow of shopping trips exhibit significant synchronicity, particularly on the branch routes. These findings can provide valuable insights for zoning planning, construction of bicycle infrastructure, and formulation of sustainable urban transportation policies.

1. Introduction

Since the late 20th century, there has been a growing interest in sustainable development and climate change. One of the most effective ways to achieve sustainable transportation is through bicycle travel. In recent years, dockless shared bikes have gained widespread promotion and implementation. These bikes, unlike the traditional dock-based system, utilize mobile applications for rentals instead of IC cards [1,2]. They also do not require fixed docking facilities, but rather, parking within zones defined by virtual geofencing [3]. The absence of dock restrictions, convenient operation via mobile apps, abundant docking facilities, and straightforward payment methods significantly enhance the travel experience of residents and facilitate the integration of public transport [4,5]. Shopping trips, often overlooked, constitute a substantial portion of shared bicycle journeys. According to Aurora Big Data, commuting accounts for 36.9% of shared bicycle trips, followed by shopping (25.4%), leisure activities such as sports and sightseeing (24.6%), and other purposes (2.9%). Shopping trips are more prevalent on weekends and holidays [6]. A report by Hello bike and Eleme app on the May Day holiday in 2020 revealed that 30.3% of the trips were related to shopping malls, appliance stores, and commercial streets [7]. Furthermore, shopping trips exhibit a mix of randomness and discernible patterns, such as routine grocery shopping and the purchase of essential items. While they lack the consistency of commuting travel and the unpredictability of leisure trips, identifying shared bike usage specifically for shopping purposes holds significant value, potential for innovation, and practical applicability. Therefore, this study aims to explore the spatiotemporal patterns of shared bike shopping trips, and to investigate their transformation and practical application.

Previous research has demonstrated that the usage patterns of shared bikes are closely related to user behavior, with travel chain components exhibiting spatial-temporal heterogeneity. This heterogeneity is classified by destination, origin, and origin-destination (O-D) correlation. Studies focusing on destinations generally involve identifying travel purposes and analyzing their spatiotemporal activity characteristics [8], leading to the creation of model methods to improve the accuracy of travel purpose identification. In contrast, studies centered on origins primarily focus on subway stations, including investigations into the mobility characteristics of shared bikes in bus interchange areas and the factors influencing their use [9]. However, current research on O-D travel chains remains relatively limited, primarily encompassing the extraction of spatiotemporal features and comparative studies of travel chains based on purpose or type identification [10].

Although destination-focused research has primarily improved the accuracy of destination identification methods, there is a significant gap in investigating the characteristics of shared bike travel post-identification. This includes a lack of comprehensive comparison and analysis from various spatial analysis scales. Furthermore, while a considerable amount of research focuses on identifying and categorizing commuting purposes, studies on alternative travel activities, such as shopping and leisure, are relatively underdeveloped. Additionally, the complexity of analysis methods and modeling techniques related to O-D travel chains has led to a nascent field full of research opportunities and identified gaps.

Therefore, it is imperative to integrate spatiotemporal usage patterns, which mirror supply and demand dynamics, into the strategic assessment of dockless bike-sharing systems. This approach enhances service efficiency and capacity, aligns with community needs, fosters scientific road network planning, and boosts the economic efficiency of urban environmental management. As a result, it aids in promoting sustainable transport and improves the convenience and comfort of residents’ travel experiences.

This study examines the spatial-temporal patterns of dockless bike-sharing for shopping purposes, employing a robust analytical framework to analyze usage patterns in terms of origins, destinations, and O-D correlations. A “cycling intensity” index is introduced, categorized using the natural breaks method, to delineate destination-based patterns at both subdistrict and road section scales. This study investigates origins through spatial autocorrelation and k-means clustering techniques, while grey correlation analysis is used to elucidate the characteristics of O-D interactions. The specific objectives of this paper are as follows:

• This study focuses on the under-researched area of shopping trips in the context of dockless bike usage, with the objective of better aligning the system to meet both shopping and transportation needs.
• Conduct a thorough analysis of spatial-temporal usage patterns and trends utilizing the adaptation method. Based on these findings, systematic enhancements in various domains are proposed, such as improving urban public transportation systems, planning bicycle transportation infrastructure, and developing urban public service facility center systems.
• This study serves as a bridge between theoretical and applied research, providing a significant resource for future optimization of dockless bike-sharing systems.

2. Background Studies

2.1. Research on Dockless Shared Bikes Based on Destination, Origin, and O-D Travel Chains

2.1.1. Thematic Research Based on Origin or Destination

Current research on shared bike destinations can be categorized into two groups based on spatial scope and content. The first group includes studies that examine the spatial and temporal distribution patterns of cycling destinations on a city-wide scale, with a specific emphasis on the relationship between cycling destinations and their associated purposes. Both Sun, DH and Wang, B [11] and Xing, YY et al. [12] employed POI (Point of Interest) data and cluster analysis to explore the travel purposes of shared bikes in Shanghai. Xing, YY et al. [12] further classified destinations into five categories (dining, transportation, shopping, work, and residence) and analyzed the cycling patterns related to these typical activities. Li, SY et al. [13] enhanced previous studies by using the gravity model, identifying four additional types of travel activities in Shenzhen: leisure, education, life service, and medical. In terms of travel purpose recognition, models such as decision tree, neural network, linear discrimination, support vector machine, nearest neighbor analysis, gravity model, empirical distribution fitting, and Bayesian probability rules have been refined and applied [14,15,16]. Simultaneously, new models like P3M and eigendecomposition have been introduced to improve the accuracy of travel purpose identification [17,18].

The second category of research investigates the usage patterns of shared bikes, with a particular focus on the type of POI or land use at the destination [19,20]. This category mainly includes three aspects: the dynamics of shared bike inflow and outflow around public transportation hubs [21], the spatiotemporal characteristics of cycling origins, and the factors influencing trip attraction and resistance at different times of the day [22]. It also examines the impact of built environment factors on the generation of shared bike travel [23]. Common analytical methods include regression analysis techniques such as the Somber regression model and multiple regression model, geographical detection methods, spatial autocorrelation analysis, and the logit model [24,25,26]. These methods are equally applicable to the study of usage patterns in destination-based bike-sharing. Typically, most studies employ individual models to analyze cycling origins and destinations, with few considering both simultaneously, primarily at the station level [27].

2.1.2. Compound Research Based on O-D Travel Chains

In contrast to studies that solely focus on the origins or destinations of cycling, research examining the O-D travel chain places significant emphasis on revealing the complex correlations between origins and destinations, as well as the spatial-temporal rules governing shared bike flow throughout the entire cycling process [28]. These studies typically adopt a holistic approach, observing the entire period in terms of origin and destination, such as location and scale. Broadly, these investigations can be divided into two categories. The first category focuses on the overall regulation of the spatial-temporal patterns of O-D flow changes for shared bikes, without distinguishing between the specific travel purposes of shared bikes. For example, Dong, J et al. [10] investigated the spatial-temporal aggregation and flow patterns of shared bikes in Beijing, introducing the DestiFlow model to delineate such patterns. Xu, Y et al. [18] collected four months of data on Singapore’s bike-sharing system and applied eigendecomposition to uncover underlying rules. The second category uses models to identify types of O-D travel chains, followed by detailed analyses and comparisons of the similarities and differences in the spatial-temporal patterns of various O-D travel chains. The most common method in purpose-based travel chain analysis is the commuting–non-commuting dichotomy. Xie, XF and Wang, ZJ [29] categorized the bike-sharing travel chain in the Washington area into non-leisure (e.g., commuting) and leisure travel. Sun, QP et al. [30] used a Back Propagation (BP) neural network prediction model to identify a total of five sub-types of travel patterns under the two categories of commuting and non-commuting.

2.1.3. Research Gap

On the one hand, in thematic studies of shared bike travel purposes, spatial and probability dimensions are often used to improve identification accuracy [31]. However, temporal factors are not fully considered. Due to limited space, most models cannot further explore the spatiotemporal patterns of shared bike trips for different purposes. Therefore, there is a lack of systematic comparative analysis of the spatiotemporal characteristics of shared bikes for different travel activities from multiple perspectives and spatial analysis units. On the other hand, although many studies on shared bicycle usage rules based on travel purpose identification focus on commuting [32,33], only a few studies analyze travel activities for other purposes [34]. This may be because most of the relevant research has been conducted in China, where shopping and leisure are also important reasons for using shared bicycles. Therefore, it is important to understand the spatiotemporal patterns of shared bicycle use for shopping and leisure, but there is still little understanding in this area. In addition, there are still deficiencies in the content, perspectives, and methods of origin-based thematic research at non-station levels. Research on the O-D travel chain should start with an analysis of the origin/destination. However, due to the complexity of analytical methods and modeling techniques, such research is still in its infancy and mainly focuses on the commuting travel chain. To address these gaps, this study provides a comprehensive analysis of the spatial-temporal patterns of dockless bike-sharing shopping activities. This is achieved through origin and destination-based topic research at both the subdistrict and road section levels, as well as relevant O-D correlation investigations at the same levels. Furthermore, this study also examines the practical application of transformative analysis results.

2.2. Application of Radiation Range of Business District

The maximum walking radius users can cover from the drop-off point to various shopping facilities is influenced by several factors, including the scale and types of shopping facilities, their location, the traffic environment, and the layout of the electronic fence [35]. For instance, to ensure an orderly and smooth traffic flow around a shopping center, users are generally required to park their shared bikes in designated parking areas. The “electronic fence” of a community convenience store is often located at the store’s entrance, which may result in a larger walking radius for users traveling to a large shopping center within a complex traffic environment compared to the community convenience store itself. The radiation range of the business district can help explain and quantify the maximum acceptable walking distance for shared bike users from their drop-off location to shopping facilities.

The term “radiation range” in the context of a business district refers to the geographical area within which a specific retail establishment can attract customers. This concept is similar to that used for shared bike users, signifying the spatial appeal of shopping facilities that prompt users to park their bikes nearby and visit the store. The service capability coefficient of retail commercial facilities is a crucial metric for assessing their attractiveness to customers. However, this study limits its scope to customers who use dockless shared bikes and park them at retail commercial facilities for shopping. Therefore, the calculation of this coefficient must take into account both the appeal of the retail commercial facilities and the surrounding environmental factors that could impact shared bike parking.

Reilly’s Law and the Huff Model serve as foundational frameworks for assessing the radiation range of commercial zones. Reilly’s Law suggests that a city’s attractiveness is directly proportional to its size and inversely proportional to the distance from it, thereby elucidating the relationship between city size and the establishment of retail zones [36]. On the other hand, the Huff Model outlines how a commercial facility’s business area, scale, and time influence consumer visitation probabilities [37]. While Reilly’s Law emphasizes commercial attractiveness based on city size, the Huff Model underscores the scale and accessibility of individual facilities. Subsequent research has built upon these core theories, identifying various key factors that influence the radiation range of commercial zones, such as geographical location, transportation accessibility, commercial scales, business diversity, and product pricing [38,39]. This comprehensive approach allows for an assessment of the overall service capability coefficient of commercial facilities. Building on this foundation, we integrated national standards and selected representative influencing factors related to user parking of shared bikes to participate in the calculation of the radiation range of retail commercial facilities (i.e., involving building area, format brands, and transportation location).

3. Materials and Methods

3.1. Study Area

Nanjing, located in the middle and lower reaches of China’s Yangtze River Plain, is the political, cultural, and economic center of Jiangsu Province [40]. With a high population density, an advanced public transportation system, and abundant shopping facilities, Nanjing has become the second-largest commercial city in East China after Shanghai. The dockless shared bike program was introduced in January 2017. As of June 2019, there were 500,000 bicycles and 6 million users [41]. In December 2021, the average daily number of rides reached 320,000, with a maximum of 680,000 [42]. Therefore, Nanjing is one of the most developed cities in China in terms of urban development, shared bike supply, user scale, and market maturity. It is suitable for studying dockless shared bike usage.

This study examines the Jiangnan main city area of Nanjing, where shared bike usage is most prevalent and consistent. The research area comprises 41 subdistricts, with defined boundaries extending roughly 25 km from north to south and 18 km from east to west. The region is largely flat, with an average altitude of 20–30 m, except for Zhongshan in Xiaolingwei (No. 20), which exceeds 400 m in altitude. This flat terrain is conducive to shared bike usage. Residential areas are scattered throughout the main city, with Hexi in the west focusing on modern business and residential development, while Chengnan in the south is primarily cultural and commercial. Shopping facilities are mainly located in the old town, with large-scale commercial spaces emerging near Hexi, Chengbei, and Yuhua Subdistrict (particularly around Nanjing South High-speed Railway Station). In terms of commercial circle structure, a spatial layout has emerged, with Xinjiekou as the primary commercial hub, Nanjing South Railway Station as the core of the second-level commercial circle, and Fuzimiao, Jiangdong, and Hunanlu Subdistricts as the tertiary commercial circle. This framework guides the delineation and division of the spatial analysis area based on shopping facilities, with statistical units divided into subdistricts and road sections, thereby facilitating a more comprehensive depiction of usage patterns from various analytical perspectives (Figure 1). The figure provides the correspondence between subdistricts and numerical codes, which applies to all subdistrict related figures throughout the paper.

3.2. Data Acquisition and Processing

3.2.1. Data Sources

The data were collected for two weeks from 13 May to 26 May 2019. The weather was sunny and there were no holidays or special events (legal holidays, summer vacations, etc.) during this period. Schools and companies in Nanjing were operating as usual, which can reflect the usage of shared bikes in the city. Nanjing is located in a sub-tropical humid climate zone with short spring and autumn, long winter and summer, and large annual temperature difference. The average winter temperature is about 2.7 °C, mainly affected by northeast wind. When the temperature drops and the wind speed increases, it is more difficult to ride shared bicycles. Therefore, the usage rate of shared bicycles decreases in autumn and winter. Some studies collect data for one week [43,44], while others collect data for one month or longer. A one-week dataset may not be enough to identify the influence of external factors on the fluctuation of usage. If the data are collected for two weeks or longer, they can be cross-checked, i.e., the first- and second-week’s data can be compared to find and remove outliers, which makes the data more stable. In this way, we can analyze and compare the regularity of usage and the overall trend over the whole period and specific time periods. The latest data are not used because China gradually lifted COVID-19 prevention policies in 2023. In the past three years, people have changed their habits of using shared bicycles due to COVID-19 prevention policies, so the usage patterns are different from those before the epidemic. This limits the generalizability and applicability of the research results after the epidemic.

The shared bike usage data for Nanjing was sourced from Tecdat, a company specializing in domestic data collection and analysis using R language and Python. As the data are not publicly available, Tecdat used a Python script to collect the data by decompiling the application’s source code and employing Packet Capture tools to identify APIs (Application Programming Interface) from the server side within the Hellobike, Mobike, and DiDi Bike applications and mini programs. The data were captured at 30 s intervals to ensure real-time accuracy of new order origins and to minimize discrepancies with previous destinations, thereby reducing manual dispatch errors and GPS inconsistencies. This method improves the precision of user-locked pre-ride destination positioning. The dataset includes over one million records daily, containing order number, user ID, vehicle ID, start and end times, and geographical coordinates for both origin and destination. Using Amap API’s path planning based on the shortest duration criterion, a closely simulated cycling path for shared bikes with consistent origin-destination (O-D) pair data was generated. This reduces errors associated with previous O-D straight connection methods and enhances the reliability and accuracy of the analysis.

POI and AOI (Area of Interest) data for Nanjing in 2019 were downloaded from the Amap API. Within the study area, 69,234 POI data points from shopping facilities were extracted, along with their corresponding AOI data. Additionally, in compliance with national regulations, shopping facilities were classified into three types: municipal, district, and community (

Table 1. Features of different types of shopping facilities.

Level	Service	Location	Radiation Radius	Frequency of Daily Use	Features of Adjacent Cycling Lanes
Municipal	Cater to sophisticated and fashionable shopping preferences and offer catering, leisure, and entertainment services, along with business social amenities.	In the city’s core business circle or central business district.	Cover or even exceed the city’s boundaries.	Generally, the frequency tends to decrease from municipal to community level. It is not uncommon for residents to travel to higher-level shopping facilities located at a greater distance (particularly on weekends) to fulfill diverse consumption needs. Those facilities situated near workplaces and along commuting routes are typically utilized more frequently.	Connect adjacent functional zones within the city, typically experiences high volumes and complex traffic conditions.
District	Offer a wide array of products and services.	Typically situated close to main traffic roads or urban nodes.	Approximately 5 km or more.		Establish internal connections within urban functional groups, catering to large public facilities or residential areas.
Community	Primarily focus on meeting the daily needs of nearby residents.	Situated on the ground floor of dwellings and along the streets that link the community to external areas.	Usually between 300 and 1000 m.		Facilitate transit and enhance connectivity between plots and major roads, with low bicycle traffic flow.

and Figure 2).

3.2.2. Data Processing Based on Destination

We first pre-processed the two-week shared bike data collected from Tecdat to eliminate invalid records caused by anomalies such as overly short rides due to GPS drift or overly long rides due to failure to lock the bike. First, we filtered the dataset to include only records with at least one starting point or ending point within the study area. Then, we calculated the duration of each ride using the record start and end times. Using the Amap API (JS API 1.4) and Python 3.9, we simulated the riding routes and geographically calibrated them, which allowed us to calculate the average riding speed for each order. We found that 97.06% of the rides lasted less than one hour. Considering that the average rental cost of a one-hour shared bike (CNY 6) is sufficient to cover the cost of other public transportation for longer distances, we decided to focus on rides that were less than one hour in duration and had an average speed between 1 km/h and 30 km/h, considering these as valid data [45]. All subsequent data processing, whether based on destination or origin, was conducted exclusively on this valid dataset.

The primary objective of processing shared bike data based on destination is to identify trips associated with shopping activities. This study enhances the methodologies for identifying shopping-related trips using shared bike data, moving beyond traditional POI-based approaches. It takes into account variations in shopping facility operating hours and service grades, which significantly influence cycling times and destinations (Figure 3):

First, the two-week data (13 May to 26 May) covering 10 working days and 4 weekends were filtered separately based on valid data. The working hours on weekdays were determined by the peak commuting time in the morning and evening, while the working hours on weekends were determined by the operating hours of commercial facilities. According to previous studies and experience, the period from 10:00 to 16:00 on weekdays is between the peak commuting times in the morning and evening. Therefore, this period includes lunch breaks and is defined as working hours [46]. Within these time ranges, shared bike data with order start times of 10:00–16:00 and 19:00–21:00 on weekdays and 9:00–21:00 on weekends were retained. This filtering was performed to eliminate the impact of commuting peaks and retain only shopping and leisure trips. In addition, according to the analysis of sample data, shopping trip during 10:00–16:00 on weekdays accounted for 40% of valid data. This indicates that shared bike users who may have flexible work schedules or are not employed still have a certain demand for shopping trips. Therefore, it is necessary to retain this period of data. The above processing step can remove the concentrated commuting data, and the following steps will further extract the shopping trip data to ensure that the non-peak shopping behavior is captured and conforms to the business hours (9:00–21:00).

Secondly, we determined the maximum walking radius of users to retail commercial facilities. By integrating the results from prior studies [47] with our current data, we set a reference value of 200 m. This value signifies that at least one POI corresponding to 90.02% of shared bikes was situated within this radius. Furthermore, the growth rate of the proportion curve for the maximum walking radius data volume was observed to be zero (Figure 4).

Third, as mentioned in Section 2.2, the acceptable walking distance from the drop-off point to the shopping facility exceeded the assumed threshold of 200 m. To cope with this change, we propose a set of formulas for calculating the radii of retail commercial facilities.

Coefficient of service capacity of retail commercial facilities,

$$I_C=S_C+{\displaystyle \frac{N_{poi}}{S_C}}+(N_r+{\displaystyle \frac{1000}{d_{nb}}}+{\displaystyle \frac{1000}{d_{nm}}})$$

(1)

where $I_C$ is the service capacity coefficient of retail commercial facilities; $S_C$ is the building area of facilities; $N_{poi}$ is the amount of POI data covered by facilities; and $N_r$, ${\displaystyle \frac{1000}{d_{nb}}}$, and ${\displaystyle \frac{1000}{d_{nm}}}$ are the number of roads connected with retail commercial facilities and the distance (in km) between retail commercial facilities and the nearest bus station and subway station entrances.

It should be noted that before the index superposition, dimensionless processing should be carried out by the method of linear function normalization. The formula is as follows:

$$X_{nom}={\displaystyle \frac{X-X_{min}}{X_{max}-X_{min}}}$$

(2)

where $X_{nom}$ and $X$ are the normalized and original values of a certain type of indicator, respectively; and $X_{min}$ and $X_{max}$ are the maximum and minimum values of a certain type of indicator, respectively.

Radiation ranges of retail commercial facilities,

$$R_C=200\ast (1+I_C)$$

(3)

where $R_C$ is the radiation range of retail commercial facilities.

Finally, we used the method proposed by Xing et al. [12] to extract the dockless shared bike usage data within the radiation range of retail commercial facilities. To address the local optimization problem associated with the random initialization of K-means cluster centers, we employed the K-Means++ algorithm proposed by Arthur and Vassilvitskii [48] (pp. 1027–1035). This algorithm optimally initializes cluster centers to maximize dispersion and minimize the number of iterations required for convergence. The method is based on the tendency of dockless bikes to park near user destinations, and infers travel destinations based on the proportion and clustering of POI types. This method has several advantages, including its ability to use POI data to reveal potential traveler destinations, its optimization of initial cluster center selection within the K-Means++ framework, and its emphasis on the spatial analysis of shopping trip regions, which are defined by the overlapping influence radii of retail commercial facilities.

The method consists of three key steps:

• Gaussian function-based spatial attraction attenuation for POIs.
• Weighted spatial attractiveness assessment of POIs to user travel destinations by POI type distribution.
• K-Means++ algorithm-driven determination of shared bike travel purposes (Formula (4)).

$$\underset{C_k}{\min } {\sum }_{j=1}^k{\sum }_{x\in C_k}{‖S_j-{\mu }_k‖}^2$$

(4)

where $S_j$ is the vector of the largest weighted spatial attractiveness values of each POI category associated with trip j; $k$ is the cluster index; ${\mu }_k$ is the centroid vector for cluster $k$; ${\mu }_k$ has the same dimension as $S_j$; and $C_k$ is a list of bike-sharing trips in cluster $k$.

3.2.3. Data Processing Based on Origin

O-based bike data processing extracted valid data of dockless shared bikes where the cycling origin is at shopping facilities. Spatial analysis range was determined by the “change inflection point” of cycling data contained in the multi-loop buffer as follows:

• Establish a multi-layer buffer ring with 200 m as the unit.
• Calculate the number and rate of change of the origin data of shared bikes contained in the buffer ring of each layer. For data of shared bikes falling into two or more spatial buffers simultaneously, repeated counting was allowed to obtain a more accurate spatial range under sufficient statistical samples.
• Determine the spatial analysis range of the origin of cycling for shopping facilities as “the range formed after the fusion of a circular area with a radius of 400 m” (Figure 5).

3.3. Analysis Methods

3.3.1. An Explanation of Different Analysis Methods

Generally, the methods used to analyze the spatiotemporal patterns of shared bikes based on origins can also be applied to those based on destinations. This includes kernel density estimation, origin-destination (OD) network analysis, and other methods described in Section 2.1.1. These different methods have two main advantages: they are often the best choice for specific purposes and scenarios, making full use of technical advantages; and they provide a multi-angle, multidimensional perspective that makes the results more accurate and innovative. Specifically, the destination-based analysis used cycling path data, calculated cycling intensity using Formulas (5) and (6), classified it using the natural break point method, and analyzed spatial-temporal patterns at the subdistrict and road section levels. The origin-based analysis used starting and ending points, spatial autocorrelation analysis, and k-means clustering to explore outflow patterns and distribution at the origins of shopping facilities. Considering the spatial characteristics of both origin and destination in the travel chain, these methods represent different analysis scenarios, with the former focusing on internal patterns and the latter on external diffusion. For O-D correlation analysis, the results of both destination and origin analyses were integrated into a unified coordinate system.

3.3.2. Methods of Spatial-Temporal Patterns of Shared Bikes Based on Destination

The spatial-temporal patterns of shared bikes, based on destination, were primarily analyzed through the quantitative calculation of specific indicators. Utilizing the statistical unit of subdistrict and road section, we introduced an index termed “cycling intensity”. This index encapsulates the sum of the road segment per unit area and the average daily cycling path length of shared bikes per unit road section. The formulas for these calculations are as follows:

Subdistrict unit,

$$I_{subdistrict}={\displaystyle \frac{{\sum }_{i=0}^n{d}_{bn}/t}{{\sum }_{i=0}^n{d}_{rn}/S_n}}$$

(5)

where $I_{subdistrict}$ is the cycling intensity of shared bikes under the subdistrict unit, $d_{bn}$ is the riding path length of a shared bike carried by subdistrict unit, $d_{rn}$ is the length of a road section, $S_n$ is the subdistrict area, and $t$ is the number of days.

Road section unit,

$$I_{road}={\displaystyle \frac{{\sum }_{i=0}^n{d}_{bi}/t}{d_{ri}}}$$

(6)

where $I_{road}$ is the cycling intensity of shared bikes under the road section unit, $d_{bi}$ is the riding path length of a shared bike carried by the road section, $d_{ri}$ is the length of the road section, and $t$ is the number of days.

In light of this, the natural break point method was employed on the ArcGIS platform, utilizing the Jenks automatic classification algorithm to evaluate “cycling intensity”. This method’s core principle is similar to clustering. However, it holds a significant advantage over clustering due to its capacity to consider the range and quantity of elements in each category, thereby yielding more interpretable classification results.

3.3.3. Methods of Spatial-Temporal Patterns of Shared Bikes Based on Origin

This study aimed to analyze the spatial-temporal patterns of shared bikes based on their origin, specifically focusing on the outflow patterns and modes of strong source points. A source point is defined as an origin with a net outflow value greater than zero [21]. Consequently, the strong source points under investigation in this study were required to meet three criteria: a net outflow quantity greater than zero, placement in the first quartile of all data, and the demonstration of spatial clustering with surrounding points.

The procedure is outlined as follows:

Firstly, we conducted a screening for high outflow points. We calculated the “net outflow” (the difference between the outflow and inflow of bikes) and the rate of net outflow at the cycling origin. If both indicators exceeded the mean value, we identified it as a high outflow point. In this process, we utilized the definitions of the net outflow rate proposed by Gao et al. [21] and Guo, D et al. [49]. The formula is presented below:

$$P_i={\displaystyle \frac{O_i-D_i}{D_i+O_i}}$$

(7)

where $P_i$ is the net outflow rate, and $D_i$ and $O_i$ are the number of destinations and origins of shared bikes within the spatial analysis range of different POI types, respectively.

Secondly, we identified the strong source points. Utilizing “net outflow density” (defined as the ratio of net outflow to unit area under the spatial analysis unit) as our parameter, we conducted a global autocorrelation analysis to discern the spatial aggregation pattern of high outflow points. Subsequently, we employed local autocorrelation analysis to pinpoint the strong source points (hot spots with a 99% confidence interval) within the distribution of high outflow points.

Third, we analyzed the outflow pattern and distribution of shared bikes emanating from strong source points. Given the non-overlapping nature of morning and evening peak source points, an initial integration was performed. We selected the “ratio of net outflow at strong source points” and the “cycling intensity of road sections” as parameters. After standardization, these parameters were amalgamated to formulate the “outflow intensity” index. This index not only mirrors fluctuations in shared bike flow from prominent source points, but also encapsulates usage patterns in neighboring sections influenced by these points within a specified radius. Following this, k-means clustering was employed to categorize outflow intensities and distill shared bike outflow patterns. To ascertain the optimal number of clusters (K), one-way ANOVA was utilized to guarantee high intra-cluster similarity alongside marked inter-cluster differences.

3.3.4. Correlation Analysis of Spatial-Temporal Patterns of Shared Bikes Based on O-D

The practical challenges in associating spatial-temporal patterns based on cycling O-D arise from the differences in data sources and analysis results between destination and origin. These differences prevent a direct comparison of the two in a unified coordinate system. To address this issue, we employed grey correlation analysis. This method offers several advantages: it primarily measures the degree of correlation among factors based on their developmental trend similarities or dissimilarities, referred to as the “grey correlation degree”. This approach is adaptable to data interface inconsistencies and is suitable for large sample calculations. The procedure can be summarized as follows:

Firstly, we standardized the scope of our analysis. In terms of spatial considerations, we identified the intersection of destination-based and origin-based spatial analysis areas as the area for O-D correlation study. Regarding temporal aspects, peak periods for shopping trips were determined to be between 9:00 and 10:00 and 17:00 and 21:00. The remaining time intervals were deemed to be non-peak or flat periods.

Secondly, we calculated the indicators. The cycling intensity based on origin and destination was computed using Formulas (5) and (6). We used “Cycling intensity under the statistical unit of road section” as an index because it comprehensively represents both the frequency and intensity of cycling. This approach provides a more accurate reflection of spatial-temporal patterns based on origin and destination.

Third, we conducted an origin-destination (O-D) correlation analysis. We employed grey correlation analysis to compute and contrast the “correlation degree” of shopping trips during peak and flat periods. This was done under the statistical unit of subdistrict and road section, specifically examining the degree of synchronous change between the origin and destination. The formula for this analysis is derived as follows [50]:

Suppose $x_{0=}\left\{x_o\left(k\right)|k=\mathrm{1,2},\dots ,N\right\}$ is the reference sequence, $x_{i=}\left\{x_i\left(k\right)|k=\mathrm{1,2},\dots ,N\right\},i=\mathrm{1,2},3,\dots ,m$ is $m$ comparative series,

Correlation coefficient,

$${\zeta }_i\left(k\right)={\displaystyle \frac{\begin{array}{c}min\\ i\in m\end{array} \begin{array}{c}min\\ k\in N\end{array}\left|x_0\left(k\right)-x_i\left(k\right)\right|+\rho ·\begin{array}{c}max\\ i\in m\end{array} \begin{array}{c}max\\ k\in N\end{array}\left|x_0\left(k\right)-x_i\left(k\right)\right|}{\left|x_0\left(k\right)-x_i\left(k\right)\right|+\rho ·\begin{array}{c}max\\ i\in m\end{array} \begin{array}{c}max\\ k\in N\end{array}|x_0\left(k\right)-x_i\left(k\right)|}}$$

(8)

where ${\zeta }_i\left(k\right)$ is the correlation coefficient of $x_i$ to $x_o$ at time $k$, $\begin{array}{c}\min \\ i\in m\end{array} \begin{array}{c}min\\ k\in N\end{array}\left|x_0\left(k\right)-x_i\left(k\right)\right|$ and $\begin{array}{c}\max \\ i\in m\end{array} \begin{array}{c}max\\ k\in N\end{array}\left|x_0\left(k\right)-x_i\left(k\right)\right|$ are the minimum and maximum absolute values of the difference between all sequences and sample number columns at any corresponding time, respectively, ρ is the resolution coefficient (0 < $\rho$ < 1), and the value of $\rho$ is 0.5.

Correlation degree,

$${\gamma }_i={\displaystyle \frac{1}{N}}\sum _{k=1}^n{\zeta }_i\left(k\right)$$

(9)

Both correlation coefficient and degree represent the O-D correlation strength of a certain type of travel under spatial statistical units. The larger the O-D correlation, the stronger the O-D correlation, and vice versa.

3.4. Questionnaire Survey

A questionnaire survey was conducted from 6 May to 26 May 2023. The primary objectives were twofold: firstly, to determine the discrepancy between the routes suggested by Amap and the actual cycling paths of shared bike users; secondly, to validate the accuracy of the methodology proposed in this study for identifying shopping trips using shared bikes. Due to the high mobility of bike-sharing users, a significant amount of time was dedicated to locating willing participants for our survey. To ensure an adequate collection of questionnaires, we administered the survey on both weekdays and weekends, taking into consideration factors such as the popularity, turnover, and geographical distribution of retail commercial facilities. Specifically, Deji Plaza on Xinjiekou (No. 18) Subdistrict, Wanda Plaza on Xinglong (No. 34) Subdistrict, and Suguo Market on Wulaocun (No. 22) Subdistrict were selected as representative examples of municipal, district, and community shopping facilities, respectively. Through random sampling, over 60 individuals were interviewed at each location, resulting in a total of 196 participants who reported using shared bikes for shopping. During these interviews, respondents were asked to recount their most recent shopping experience at the surveyed location and to indicate the starting point, destination, and general route taken on a map. Subsequently, Amap was used to generate a cycling route based on the respondent’s information, and the distance traveled was confirmed with the respondents and marked on a paper map. The proposed method was then used for identification and verification purposes. We found that 86.67% of users chose to cycle along the planned path on Amap, including those with temporary changes in destination. By calculating the proportion of shared bike shopping trips in the overall interview data, we found that the accuracy of the method proposed in identifying shopping trips reached 81.3%. This empirical evidence underscores the efficacy of the suggested approach in precisely identifying shared bike shopping trips.

4. Results

4.1. Spatial-Temporal Patterns of Shared Bike Shopping Trips Based on Destination

Time can be categorized into working days and weekends, while space is divided based on two statistical units: subdistrict and road section. This study will primarily focus on the spatial-temporal patterns of shared bikes around shopping facilities of varying grades.

During weekdays, areas of high shared bike cycling intensity around municipal shopping facilities in subdistricts were primarily concentrated in Xinjiekou (No. 18), Wulaocun (No. 22), and Hongwulu (No. 25). This formed a “single core” distribution, while district shopping facilities encircled Wulaocun (No. 22) to create a “ring”. Community shopping facilities displayed a “cross structure”, with Meiyuanxincun (No. 19) as the peak. On weekends, the high-intensity areas around municipal shopping facilities decreased, district facilities became more compact, especially in Wulaocun (No. 22), and community shopping facilities significantly expanded, adopting a “dumbbell” distribution (

Table 2. Comparison of spatial-temporal patterns around shopping facilities of different grades on weekdays and weekends.

Spatial Analysis Unit	Time Period	Municipal	District	Community
Subdistrict statistical unit	Weekdays
	Weekends
Road section statistical unit	Weekdays
	Weekends

). Overall, the distribution of high cycling intensity areas was more concentrated with higher levels of shopping facilities, and there was greater weekday similarity between municipal and district shopping facilities in high cycling intensity areas.

The analysis of cycling intensity on road sections showed that weekday commuting was mainly along east–west urban trunks and north–south branches, or residential roads around municipal and district shopping facilities. The high-intensity sections near community shopping facilities were shorter, with more branch and residential roads. On weekends, the distribution around municipal facilities was similar to that on weekdays, while the high-intensity sections around district and community facilities were more dispersed, mainly along north–south roads in residential areas. This may be related to the distribution of shopping facilities along different grades of roads and the road network structure in Nanjing. High-grade shopping facilities were mainly distributed along major east–west roads, with a higher concentration in the north–south direction near Xinjiekou (No. 18) due to large-scale facility clusters. Community-level shopping facilities were more evenly distributed without obvious directionality. There were more primary roads in the east–west direction than in the north–south direction, and more residential roads in the north–south direction outside the old town. Overall, there was a significant difference in cycling intensity between weekdays and weekends around district shopping facilities (Table 2). It can be concluded that community facilities meet daily needs, while municipal facilities are related to high-end consumption. The continuous use of shared bicycles near these facilities indicates stable demand, while there is some randomness around district shopping facilities.

4.2. Spatial-Temporal Patterns of Shared Bike Shopping Trips Based on Origin

We extracted origin types associated with shopping trips (shopping facilities) and analyzed the temporal distribution of shared bike mobility. We identified the morning and evening peaks (7:00–10:00 and 17:00–19:00) as the primary statistical units.

The outflow analysis (Figure 6a,b) revealed two main source point clusters during the morning peak: one located in the old town center and another in the northern section of Maqun (No. 41) Subdistrict in the eastern part of the main city. Conversely, the evening peak displayed a “multi-core scattering” pattern. This included the expansion of the old town center to Xuanwumen (No. 15) and Meiyuanxincun (No. 19) Subdistricts, and the emergence of new hotspots south of Maqun (No. 41) and west of Xuanwuhu (No. 17) Subdistricts. The density cores of strong source points during both peaks were closely clustered and could be divided into two categories: (1) large commercial complexes, such as Xinshijie Department Store and Central Shopping mall; (2) community shopping facilities, exemplified by retail commercial facilities along the western side of Taiping South Road. The changes in the core distribution of source point clusters between morning and evening suggest that travel volumes from shopping facilities are more dispersed in the evening than during the day. This could potentially indicate an increased travel demand across various regions. These increases may be part of a travel chain where residents visit nearby shopping facilities for shopping or dining after work on weekdays, followed by a return home. Alternatively, they may reflect a separate travel chain involving weekend shopping trips, with residents either heading to other destinations or returning home post-purchase. In conclusion, managers and operators should prioritize the evening deployment and supply of dockless shared bikes.

The outflow patterns (Figure 6c) reveal that the strong source points of shopping facilities can be categorized into four types: “morning high-evening high” (Type 1), “morning high-evening low” (Type 2), “morning low-evening high” (Type 3), and “morning low-evening low” (Type 4). Type 1 represents a concentrated source point of shared bike trips to shopping areas during peak hours, with surrounding points of high outflow facilitating centralized departures. Conversely, Type 4 indicates a high net outflow but lacks centralized departure points due to surrounding areas of low outflow. The remaining types suggest centralized departures during either the morning or evening peak only. These four modes differ based on the presence of a centralized departure point and the specific peak period it operates. Type 1 points displayed a linear distribution along roads, with a concentrated aggregation around subway stations. Type 2 points were concentrated in the old town center, with large commercial facilities and commercial blocks exhibiting a concentrated strong source point presence. Type 3 points were scattered across Xuanwumen (No. 15), Meiyuanxincun (No. 19), Chaotiangong (No. 21), and Wulaocun (No. 22) Subdistricts, with a primary aggregation on the east side of Central Shopping Mall (in a “group” distribution), and along Zhujiang Road and Changbai Street (in a “line” distribution). Type 4 was distributed in proximity to Type 3, yet it exhibited a more scattered distribution than Type 3.

4.3. Spatial-Temporal Patterns of Shared Bike Shopping Trips Based on O-D

The parent sequence was represented by the origin-based cycling intensity in the subdistrict (road section) statistical unit, with data averaged for dimensionality reduction. Correlation coefficients were computed between the destination-based cycling intensity and its corresponding parent sequence during peak and flat periods separately. For both subdistrict and road section units, the correlation degrees for peak and flat periods were 0.734 and 0.717 (subdistrict), and 0.982 and 0.984 (road section), respectively. It is noteworthy that the O-D correlation degrees were consistent across units during both periods, although road sections exhibited slightly higher correlation degrees during flat periods compared to peak periods, which was contrary to subdistrict units. This suggests a synchronous variation in shopping trip inflow and outflow, albeit with differences in degree and timing distribution among units.

The correlation coefficient analysis (Figure 7) revealed two areas with a higher O-D correlation during both peak and flat periods. One area was located at the edge of the old town, which acted as a primary hub for shared bike-based shopping trip gathering and distribution (i.e., more inbound and outbound trips). The other area was situated on the periphery of the main city, characterized by lower shopping trip activity (i.e., fewer inbound and outbound trips). During peak hours, subdistricts with a high O-D correlation demonstrated a dispersed and uniform distribution, while during the flat period, there was a significant concentration in the northwest. Subdistricts with low synchronous inflow and outflow may require careful management of shared bike dispatch and supply. For example, shared bikes could be relocated from high-inflow, low-outflow subdistricts to those with low-inflow, high-outflow to optimize resource distribution.

Figure 8 demonstrated a consistent distribution of road sections with high correlation during both peak and flat periods. During the flat period, the number of such sections around the old town, particularly in Xuanwumen (No. 15) and Jianninglu (No. 5) Subdistricts, surpassed those during the peak period. High O-D correlation sections were predominantly located on the eastern, northern, and southwestern edges of the main city during both periods. These sections were primarily branch roads, with trunk roads constituting approximately one quarter of the branch roads. Therefore, the central main city, especially the old town areas with low O-D correlation, necessitated increased attention for timely dispatch of shared bikes on trunk roads.

5. Discussion

This study contains two main innovations:

• It introduces a novel analytical framework for investigating the spatial-temporal usage patterns of shared bikes, focusing on destination, origin, and O-D correlation. In contrast to existing studies that typically analyze bicycle destinations or origins in isolation, often at the station level [27], this research proposes an approach that allows for simultaneous observation of the correlation between destinations and origins, as well as the shopping travel chain. This addresses a gap in the literature regarding non-station level research.
• This study improves the technical methods by adding the “service capability coefficient” of retail facilities to the destination-based data processing. It makes up for the lack of consideration of the difference in the maximum walking distance from parking facilities to different levels of shopping facilities [39]. The addition of cycling intensity and net outflow density indicators can more comprehensively analyze the spatiotemporal pattern. Based on the existing research methods [21,49], it is convenient to draw more detailed conclusions (e.g., identifying the strong source points and their outflow patterns and the spatial distribution of O-D correlation), which expands the application range of multivariate analysis technology of spatiotemporal pattern.

Furthermore, our findings provide valuable theoretical insights for the development and optimization of dockless shared bike transportation systems. This includes enhancements to the bike lane network and refinements to the layout of shared bike parking spaces in proximity to shopping facilities.

This study elucidates the spatial-temporal patterns of shared bike usage around shopping facilities of varying grades, thereby identifying key scheduling areas for shared bikes in different periods based on user needs. It also identifies high-intensity road sections around facilities of different periods and grades. Furthermore, it provides a foundation for optimizing bike lane planning, zoning, and stratification, as well as enhancing the cycling environment. By taking into account the primary traffic functions and the spatial-temporal patterns of shared bike shopping trips in Nanjing’s main city, a rational plan for bike lane layout can be devised (Figure 9). Primary cycle lanes facilitate long-distance travel between and within functional zones, thereby supporting significant bicycle traffic and serving as the core of the bicycle network. Secondary cycle lanes manage internal connections within urban areas, serving adjacent buildings. Tertiary cycle lanes accommodate passing traffic, thereby improving connectivity between plots and main roads with lower bicycle flow, and complementing the bicycle network.

In relation to the morning and evening peak hours, we analyzed the intensity and patterns of shared bike outflow around the origins of four types (as detailed in Section 3.2.2). Our findings suggest that shared bikes typically demonstrate either a linear or clustered distribution during peak times. Moreover, the demand for travel during the evening peak may exceed that of the morning peak, aligning with the results of existing studies [9,45]. This information can be leveraged to optimize the design and configuration of bike parking spaces near shopping facilities and public transportation hubs, thereby integrating with the urban commercial center system.

Considering the linear distribution characteristic of “morning high-evening-high” type, it is recommended to set up parking areas within 100 m of the main entrances and exits of commercial facilities (Figure 10a), which can accommodate a large number of source points with “morning high-evening high” type and strong O-D correlation. At the same time, flexible storage areas are suggested to be arranged along the roadside to alleviate the sharp rise and fall of shared bicycle demand in peak and flat periods, as well as to meet the parking needs of surrounding residential areas, commercial facilities, subway stations, and other places with similar tidal characteristics of shared bicycle use. In peak periods (especially in the evening), these storage areas will provide extra parking spaces to ensure a sufficient supply of shared bicycles to meet travel demand; in flat periods, these flexible spaces will return to the pedestrian system to reduce the impact on walking space.

In accordance with the “morning (evening) high-evening (morning) low” pattern, which is more pronounced in group distributions, we propose the establishment of large-scale centralized parking areas within 300 m of the primary entrances and exits of facilities. These facilities should have a higher prevalence of “morning (evening) high-evening (morning) low” type source points and a moderate O-D correlation (Figure 10b). This strategy is particularly relevant for main roads where residents frequently access certain facilities, thereby accommodating the demand during peak hours in the morning or evening. Simultaneously, it also caters to the surrounding area with single peak travel times. By providing intensive support during peak hours and making adjustments throughout the day, the total number of shared bikes in the area can be maintained at a stable level. This approach not only standardizes the parking of shared bikes, but also alleviates the pressure on usage.

Considering the dispersed occurrence of the “morning low-evening low” pattern, we propose the creation of large-scale centralized parking zones in proximity to the primary entrances and exits of shopping complexes (Figure 10c). We advocate for the reservation of a specific quota of parking spaces and a designated quantity of bicycles to accommodate intermittent surges in usage and parking demands. These parking zones can also cater to other adjacent facilities and properties, thereby addressing the shared bicycle requirements of the local residents.

6. Conclusions

This study employed big data and a range of quantitative analytical methods to conduct an exhaustive examination of the spatial-temporal usage patterns of dockless shared bikes for shopping trips in Nanjing’s main city. The analysis explored these patterns in terms of destination, origin, and O-D correlations. The findings and recommendations generated from this research are as follows:

• The distribution of areas with high cycling intensity demonstrated notable differences between weekdays and weekends. On weekdays, access to shopping centers was primarily via east–west trunk roads and north–south branch roads, while community centers were more commonly associated with north–south branch roads. Conversely, areas near lower-grade facilities were predominantly located on north–south branch roads and residential streets (based on destination).

This suggests that the active zones for dockless shared bike flows around high-level shopping facilities maintain a consistent pattern, whereas those surrounding district-level and community-level shopping facilities tend to concentrate on lower-grade urban roads during weekends. The implications of this finding for policy formulation include:

• Enhancement of Integration with Public Transportation. The development of bicycle road networks should prioritize the incorporation of low-grade roads with connections to public transportation systems. The integration of bike-sharing and public transportation has been demonstrated to amplify the advantages of both modes [51]. Typically, the connectivity between low-grade roads and public transportation is less robust than that of trunk roads. Determining which low-grade roads require enhanced connections to public transportation can be a complex task. However, by referencing the distribution of areas with high cycling intensity, planners can effectively refine the route selection for public transportation and identify sections of low-grade roads that may require improvement.
• Development of Non-Motorized Isolation Facilities. This analysis also provides useful information for optimizing the cross-sections of urban bicycle lanes and improving non-motorized isolation facilities. The operation of turns and cycling at intersections is considered a significant safety risk associated with bicycle riding [52]. By examining the distribution of areas with high cycling intensity, it is possible to assess the current traffic flow and congestion levels in bicycle lanes during peak hours. This data can then be used to adjust the road cross-sections as necessary to ensure smooth bicycle traffic. Furthermore, by considering the volume of motor vehicle traffic and its potential disruption to bicycle travel, various non-motorized isolation facilities (e.g., the establishment of green belts, railings, and marking lines, among other measures.) can be implemented to ensure the safety of both cyclists and drivers.

2.

The outflow patterns from shopping facilities could be categorized into four types: “morning high-evening high”, “morning high-evening low”, “morning low-evening high”, and “morning low-evening low” (based on origin).

This categorization highlights the dynamic nature of spatial distribution for high-outflow areas during peak morning and evening hours in dockless shared bike usage, which fluctuates throughout the day. This fluctuation mirrors the temporal and spatial changes in residents’ travel demands, particularly around shopping facilities. Similar findings have been observed in studies on shared bicycle usage in other Chinese cities [46], suggesting that there are consistent dynamic changes and patterns in shared bicycle usage during peak hours. This implies that universal policies could be implemented to enhance the convenience of shared bicycle usage across a broader area. The implications of these observations for policy development are as follows:

• Enhancement of the Dynamic Modification of Electronic Fence Configurations. This finding not only offers insights into the layout of shared bike parking spaces, as outlined in the three scenarios in Section 5, but also serves as a crucial reference for operational companies when developing management policies for electronic fences. Specifically, these companies can adjust the spatial configuration of electronic fences dynamically during peak periods, based on the outflow patterns around facilities. By aligning these adjustments with the physical environment’s parking spaces, companies can establish electronic buffer zones to alleviate the surge in parking demands during peak times.
• Parking Management and Collaborative Governance. A collaborative governance model should be implemented for the management of shared bike parking, engaging residents, street managers, and enterprises. This strategy would reduce the burden on personnel responsible for maintaining order in shared bike parking zones. Bicycle companies should be assigned specific areas for maintenance. Concurrently, local street managers ought to concentrate on overseeing these companies, ensuring parking discipline, and swiftly addressing any improperly parked bicycles.

3.

Significant synchronicity is observed in the fluctuations of shopping trip inflow and outflow at both subdistrict and road section levels, with notable variations in magnitude and timing. Specifically, branch roads exhibit more pronounced simultaneous changes in bicycle traffic compared to trunk roads (based on O-D).

This implies that the inflow and outflow of shared bikes on branch roads generally achieve equilibrium. However, a significant imbalance is observed on the trunk roads of the old town, where either the inflow surpasses the outflow or vice versa. This imbalance often mirrors the uneven distribution of residents’ demand for dockless shared bikes, resulting in scenarios where supply either falls short of or exceeds demand. Numerous studies have demonstrated that shared bicycle usage is predominantly concentrated in the core areas of cities, where shopping facilities exhibit a significant degree of shared bicycle usage irrespective of their grade [2,43]. Consequently, the dynamic zoning scheduling and route allocation of shared bikes are crucial in adapting to fluctuations in the supply and demand of residents’ usage. The implications of this observation for policy development include:

• Development of a Cloud-Based Smart Transportation Sharing System. The establish establishment of an intelligent transportation cloud-sharing system is imperative to address the needs of users, companies, and government regulators. In situations where there is a mismatch between demand and supply, it is crucial for users to avoid areas with significant discrepancies in supply and to comprehend the dynamic distribution of shared bikes. Companies are obligated to swiftly dispatch shared bikes to regions where demand surpasses supply and retrieve them from areas where supply is surplus for redistribution purposes. Governments play a pivotal role in overseeing the management of these companies. All such dispatch strategies necessitate the analysis of spatiotemporal patterns, which are based on inflow and outflow.
• Optimization of the Public Facility System. The optimization of urban public facility systems can be achieved by considering population mobility patterns. The dynamic circulation of shared bikes often reflects the movement and redistribution of the population to a certain degree. Regions where inflow surpasses outflow suggest that these facilities are more appealing and capable of accommodating larger demographic flows. On the other hand, areas with a higher outflow may indicate a lower level of attractiveness and service capability, necessitating a reevaluation of these facilities’ role within the urban public facility system and an enhancement in people-oriented urban development. The correlation between inflow and outflow fluctuations provides a valuable reference for identifying such facilities.

In conclusion, this study illuminated the spatiotemporal patterns of shared bicycle usage in relation to various shopping facilities. This understanding facilitated targeted scheduling adjustments to accommodate the varying demands of users across different time periods. Such insights were instrumental in refining zoning plans, optimizing bicycle lane designs, enhancing the efficiency of urban public facility systems, and systematically improving cycling infrastructure. Moreover, they contributed significantly to the development of sustainable urban transportation solutions.

There are also limitations to this work. First, the focus on hierarchical planning of bike lanes for shopping trips may cause bias because it does not consider the various usage patterns of shared bikes for other purposes such as commuting and leisure. Future studies should expand the research scope to plan bike lane classification more comprehensively. Second, it is difficult to identify shopping trips with 100% accuracy, especially considering that the current method cannot distinguish between incidental commuting and leisure riding. Technological advancements are needed to accurately track the entire travel paths of dockless shared bikes over short periods. Third, the data used in this study were collected only for two weeks in spring, which does not fully reflect the impact of bad weather on shared bike use in other seasons [53,54]. In addition, due to data acquisition constraints, a two-week sample size may be insufficient. To improve the reliability of analysis results, future studies could collect data over a longer period. Finally, the results of this study are more suitable for East Asian countries with similar bicycle usage patterns; regional applicability requires detailed analysis of spatial-temporal usage patterns.

Furthermore, several issues arising from the findings of this study warrant further investigation in future research. First, the increase in weekend shared bike trips to shopping centers not only indicates an increase in trip volume, but also reflects a preference for longer distances to larger shopping facilities due to more leisure time. Since shared bicycles are often used as the last mile of public transit trips, the increase in shared bicycle usage may be related to the growth in passenger flow on public transit. The correlation between these two needs to be further studied and discussed. Second, chain home furnishing stores (e.g., IKEA) and independent bulky goods retailers have different delivery methods, with the former using a unified system and the latter often provided by the retailer, which can be free or paid. For those who choose to use shared bicycles to these destinations, other transportation modes are generally needed for the return trip. How shared bicycles match with other transportation modes, what role they play in the entire travel chain, and what characteristics these trips have are all worth studying in depth. Finally, the travel patterns of shared bicycles are closely related to the structure of urban road networks, including primary and secondary arterial layouts, connectivity orientations, and road network density. What impact does the morphology of the urban road network have on the distribution of shared bicycle traffic volume? To what extent does it affect it? These issues need to be further explored in future research.