Evolvement patterns of usage in a medium-sized bike-sharing system during the COVID-19 pandemic

Abstract

The global outbreak of COVID-19 has fundamentally reshaped human mobility. Compared to other modes of transportation, how spatiotemporal patterns of urban bike-sharing have evolved since the outbreak is yet to be fully understood, especially for bike-sharing systems operating on a smaller scale. Taking Pittsburgh as a case study, we examined the changes in spatiotemporal dynamics of shared bike usage from 2019 to 2021. By distinguishing between weekday and weekend usage, we found different temporal patterns between trip volume and duration, and distinct spatial patterns of within- and between-region rides with respect to naturally separated regions. Overall, the results illustrate the resilience and the vital role of bike-sharing during the pandemic, consistent with previous observations on bike-sharing systems of a larger scale. Our study contributes to a comprehensive understanding of bike-sharing that calls for more research on smaller-scale systems under disruptive events such as the pandemic, which can greatly inform decision-makers from smaller sized cities and enable future studies to compare across different urban regions or modes of transportation.

1. Introduction

In recent years, bike-sharing systems have experienced rapid growth in major cities around the globe. Nowadays, 1590 cities in 92 countries have bike-sharing systems in operation, most of which are located in China, North America, and Europe (Meddin et al., 2020). In the United States, many cities have deployed and subsequently expanded bike-sharing systems, reporting a steady increase in ridership that reached 50 million trips in 2019 (National Association of City Transportation Officials, 2019). Such systems primarily operate in the form of docked bike-sharing, in which users are required to pick up and return at designated docking stations, dockless bike-sharing, in which users are free to pick up and return at any location within the area of operation (Li et al., 2020; Zhang et al., 2021), or a hybrid of both forms. Compared to traditional modes of motorized transportation such as driving and public transit, cycling offers a great number of benefits ranging from promoting physical health, minimizing carbon footprint, alleviating traffic congestion, to increasing urban connectivity and accessibility (Bullock et al., 2017; Fishman, 2016; Zhu et al., 2020). Moreover, like other modes of shared mobility, shared bicycling provides convenient access to means of travel based on individual needs without the requirement of owning or carrying a bike. Bike-sharing has quickly become one of the fastest-growing options of shared micromobility, playing an essential role in urban transport systems in most cities (Erin & Uz, 2020).

However, the global outbreak of the COVID-19 pandemic which was declared by WHO in 2020 has fundamentally interrupted human mobility. To contain the spread of the virus and “flatten the curve”, most governments issued lockdowns and stay-at-home orders to minimize movements and in-person contacts (Hu et al., 2021; Teixeira & Lopes, 2020). This has led to an abrupt reduction in the demand for physical travel, followed by a sharp decline in the volume of urban traffic involving almost every modes of transportation. During lockdowns, the usage of public transit declined as much as 90% compared to the previous year in major cities worldwide (Sahraei et al., 2021), and micromobility options including shared bikes and e-bikes also experienced a significant drop in ridership (e.g., Li et al., 2021; Padmanabhan et al., 2021). In the meantime, a shift of travel modalities has been observed in short- to medium-range trips, where more urban residents switched from public transit and ride-sharing, which usually involve large crowds inside an enclosed space, to walking and cycling that mostly take place in open areas (Bustamante et al., 2022; Pase et al., 2020; Teixeira & Lopes, 2020). To some extent, this modality shift reflects how the pandemic has changed the travel behaviors of urban residents by raising the awareness of personal health and safety. Shared bike-riding has been demonstrated to offer an optimal option to address these concerns related to COVID-19. Many cities have observed a less significant decrease in bike-sharing ridership compared to other transport modalities during the pandemic (e.g., Bustamante et al. 2022, Teixeira & Lopes 2020). Furthermore, bike-sharing often exhibited a fast and strong rebound in ridership, sometimes even exceeding the demands of pre-pandemic times. All together, these suggest that shared bike-riding serves as a highly resilient means of urban transportation and is quite robust to disruptive events as influential as the outbreak of COVID-19 (Hu et al., 2021; Pase et al., 2020; Teixeira & Lopes, 2020).

There is a growing body of literature on bike-sharing that has critically informed researchers and practitioners of various domains including transportation, urban planning, and public policy. However, compared to other traditional modes of travel, shared bicycling is still a relatively new research topic and has been less well-studied. The outbreak of COVID-19 poses additional challenges to understanding how the usage of emerging modalities like bike-sharing has evolved as the pandemic unfolds. With the vital role bike-sharing plays in urban transportation systems and its resilience to the COVID-19 pandemic demonstrated by increased research, there is an urgent need to understand the characteristics and dynamics of bike-sharing from different perspectives in order to build resilient urban transportation systems and promote economic, environmental sustainability, and reachability in urban regions (Bustamante et al., 2022). An important perspective is the different operating scales of bike-sharing systems. Metropolitan regions like New York and London, usually with hundreds of docking stations and thousands or more shared bikes in service, have attracted most of the attention from the research community, while cities with bike-sharing systems of a smaller scale have rarely been studied in the literature (Caulfield et al., 2017; Li et al., 2022). To date, it remains largely unknown how the characteristics and dynamics of smaller-sized systems differ from those operating on a greater scale, including whether the usage has evolved differently since the pandemic. These insights are particularly important for medium- or small-sized urban regions to expand existing bike-sharing systems or introduce new operations, which constitute the primary driving force in promoting mass adoption of shared bike-riding (Li et al., 2022). To address these gaps, we select the city of Pittsburgh, the second largest urban city in Pennsylvania in the US, as a case study and examine its bike-sharing usage over the span of the years before, during, and following the outbreak.

The main research question of this study is how shared bike usage of a medium-sized system (Healthy Ride in Pittsburgh) has evolved over the years since the outbreak of COVID-19. We investigated bike-sharing usage aggregated by year, and further distinguished between work-related and recreational usage by examining weekdays and weekends separately. Temporal and spatial patterns were subsequently explored. For the temporal analysis, we used both trip volume and duration to demonstrate how hourly usage differed across days in a week, how the difference has changed across the years, and whether the changes were comparable between volume and duration. For the spatial analysis, we applied complex network theories to model bike-sharing flows as weighted directed networks to explore the dynamics of the network characteristics. Our results show that although some usage patterns found in other larger cities can also be observed in Pittsburgh, the changes in usage over the years show different directions and magnitudes when characterized by different variables (volume versus duration, weekend versus weekday, between- versus within-region trips). To the best of our knowledge, none of the published studies has focused on the evolvement patterns of bike-sharing usage in a smaller-sized system during the pandemic, covering the periods prior to, during, and following the outbreak. By focusing on the bike-sharing systems that have been much less studied in the literature, the paper's contributions include informing decision-makers from medium- to small-sized cities about bike-sharing usage and enabling future comparisons across different modes of transportation or with other urban regions.

The rest of the paper is structured as follows. The next section summarizes related work on bike-sharing and the impact of the pandemic on urban mobility. Then, we describe the study area and the dataset, followed by the methodology of analyzing spatiotemporal patterns. Finally, we present the results and discussion of the findings, and conclude with limitations and future research directions.

2. Related work

As bike-sharing becomes more prevalent in urban regions, it has also attracted considerable interest from researchers of various domains. Studies on bike-sharing either focus on the supply of bikes and stations, from the perspectives of system design and planning (e.g., He et al. 2022, Soriguera & Jiménez-Meroño 2020), or analyze the demand of bike-sharing, from the perspectives of analyzing users or usage patterns (e.g., Bullock et al. 2017, Campbell & Brakewood 2017, Chibwe et al. 2021, McKenzie 2018, 2019, Zhu et al. 2020). In particular, much research can be found to understand the spatiotemporal patterns of using shared bikes by identifying factors that may contribute to the demand (e.g., Bongiorno et al. 2019, Sohrabi & Ermagun 2021; Younes et al. 2020; Eren & Uz 2020). These factors include variables describing spatial (such as land use and proximity to public transit), temporal (such as time of day and important events), and context information (such as sociodemographic and safety concerns), which can facilitate the emerging interest in predicting urban mobility patterns. For instance, a study by Younes et al. (2020) examined how weather, gas prices, and temporal variables have affected the number of trips and trip duration of bike-sharing in Washington, D.C. Comparing the result with shared scooter usage, the authors showed that bike-sharing users were highly sensitive to gas prices and special events, and demonstrated higher sensitivity to weather conditions than scooter users. Along the similar line, Morton et al. (2021) investigated features of the built environment, demographics, and system characteristics of bike-sharing in London, and found that higher usage of shared bikes can be contributed by better accessibility to rail stations, higher proportion of male and Caucasian residents, and greater capacity of docking stations. In particular, their findings suggest that for certain time periods, the location of stations relative to River Thames appeared to affect trip generation. It provides one of the very few pieces of evidence on the possible association between natural barriers and bike-sharing usage. Different from their study, we considered natural barriers in examining a system of a smaller-scale during time periods that include the COVID-19 pandemic. Our findings can enrich the existing literature by providing the first analysis on how natural barriers may affect bike-sharing usage during disruptive events in a smaller-sized bike-sharing system.

The most recent event that significantly interrupted urban mobility, including bike-sharing, is the global outbreak of the COVID-19 pandemic. As more data from the pandemic becomes available, a rapidly growing body of research has explored how COVID-19 impacted the usage of bike-sharing (Teixeira et al. 2023; Rotaris et al. 2022). A wide range of topics has been addressed, from spatiotemporal changes in usage patterns (Tokey, 2020; Xin et al., 2022), influencing factors (Bergantino et al., 2021; Bustamante et al., 2022), to relationships with other modes of transportation (Teixeira & Lopes, 2020). According to the scale of the bike-sharing system, we distinguish between literature that examined large-scale systems and those examined smaller-scale systems, and provide examples of studies in Table 1 . To show the difference in their periods of study, we further specify whether the data used in the analysis covers the different years prior to, during, and following the outbreak. For instance, a study by Chen et al. (2022) employed trip data from 2019 to 2021 in Washington D.C. and explored the behavioral changes in shared bike users. They showed that the demand recovered quickly from the sharp decline at the beginning of the outbreak, and bike-sharing has become ever more important as a means of travel among city residents. Similar findings on the robust resilience of bike-sharing systems have been suggested by many other studies of urban regions worldwide (e.g., Hu et al. 2021, Teixeira & Lopes 2020). For example, Song et al. (2022) analyzed spatiotemporal patterns of bike-sharing usage in Singapore and illustrated high resilience of bike-sharing by showing an increase in both the intensity and connectivity of ridership amidst the outbreak.

Table 1.

Examples of studies on bike-sharing during the COVID-19 pandemic.

Scale of bike-sharing system	Location	System description	Research study	Period of bike-sharing data used in the study	Includes 2019, the year prior to the COVID-19 outbreak?	Includes 2020, the year of the COVID-19 outbreak?	Includes 2021, the year following the COVID-19 outbreak?
Large-scale	Washington D.C., USA	627 stations, 5,400 bikes	Chen et al. (2022)	Jan 2019 to Dec 2021	Yes	Yes	Yes
	New York city, USA	890 stations, 14,500 bikes	Teixeira and Lopes (2020)	Feb and March 2019 and 2020	Yes	Yes	No
			Pase et al. (2020)	March 2020	No	Yes	No
			Xin et al. (2022)	Jan to April 2019 and 2020	Yes	Yes	No
			Bi et al. (2022)	Jan 2019 to June 2020	Yes	Yes	No
	Barcelona, Spain	500 stations, 7,000 bikes	Bustamante et al. (2022)	Jan 2019 to Dec 2020	Yes	Yes	No
	London, UK	780 stations, 11,500 bikes	Chibwe et al. (2021)	Jan 2012 to June 2020	Yes	Yes	No
	Seoul, South Korea	1,500 stations	Kim (2021)	Jan 2019 to Dec 2020	Yes	Yes	No
	Chicago, USA	608 stations, 5,800 bikes	Hu et al. (2021)	March to July 2019 and 2020	Yes	Yes	No
	Singapore	14,000 bikes	Song et al. (2022)	Dec 2019 to July 2020	Yes	Yes	No
Medium- to small-scale	Lisbon, Portugal	81 stations	Albuquerque et al. (2021)	Jan 2018 to Dec 2020	Yes	Yes	No
	Bangkok, Thailand	50 stations	Sangveraphunsiri et al. (2022)	Jan 2018 to Dec 2020	Yes	Yes	No
	Zurich, Switzerland	153 stations	Li et al. (2021)	Feb to April 2020	No	Yes	No
	Pittsburgh, USA	104 stations, 506 bikes	Tokey (2020)	Mar to Aug, 2020	No	Yes	No
			Present study	Jan 2019 to Dec 2021	Yes	Yes	Yes

An important observation from Table 1 is that there is an unbalanced number of studies investigating large-scale bike-sharing systems and those operating on a smaller-scale. Indeed, bike-sharing systems like the one in Pittsburgh have attracted little attention from the research community. Prior to the pandemic, they were rarely studied in the literature. A few exceptions include Caulfield et al. (2017) who examined bike-sharing usage in the city of Cork in the Republic of Ireland, and Kim and Pelechrinis (2020) who employed data from Pittsburgh to investigate the impact of elevation. Research on smaller-scale systems listed in Table 1 leveraged data during the COVID-19 pandemic. For instance, bike-sharing in Pittsburgh was included in a study by Tokey (2020) to compare with usage patterns of bike-sharing in Washington D.C., Boston, Columbus, and Portland. The findings indicate positive correlations between weekly usage in Pittsburgh and COVID-19 exposure as well as non-work related trips, similar to other cities in the analysis. However, the study did not conduct in-depth investigation into the evolvement patterns of bike-sharing in Pittsburgh, nor did it cover the year following the outbreak. Literature on other smaller-scale bike-sharing systems also provides insights into the impact of the pandemic, but the usage in 2021 has yet to be examined. For instance, Li et al. (2021) leveraged trip data from Zurich, Switzerland before and during the lockdown period. They showed that both trip duration and distance increased during the lockdown, while working from home appeared a main contributor to the decrease in trip volume. Albuquerque et al. (2021) used data from the bike-sharing system in Lisbon, Portugal, to analyze the relationships between mobility patterns and time of day, weather, and the pandemic. Consistent with some observations from larger cities, the results demonstrate higher demand for weekday travel, sensitivity to precipitation and temperature, and the decline in volume following the COVID-19 outbreak. In summary, the impact of the pandemic on bike-sharing systems of a smaller scale remains largely understudied, and more research is needed to fully understand how the usage has evolved during as well as following the outbreak to show any enduring impact on usage patterns. Our study listed in the last row of Table 1 aims to address these gaps and provide a starting point for future research.

3. Study area and dataset

Pittsburgh is the largest city in Western Pennsylvania with a population of 302,971 according to the 2020 Census. It serves as the cultural and economic center of the region with several important industries of education, technology, and health care. Despite being historically dependent on cars as the primary means of transportation, more and more residents in the city choose to bike, take public transit, or walk to work (31% reported in a survey by U.S. Census Bureau1 ). Policymakers of the city have been continuously seeking to incorporate more environmental friendly and sustainable means of transportation to improve urban mobility. One significant effort is the launch of Healthy Ride on May 31, 2015, the only operating bike-sharing system in Pittsburgh until it was rebranded in April 2022 (Meddin et al., 2020). Over the years, Healthy Ride went through several iterations of expansion, including the introduction of dockless bike-sharing and the addition of new stations and bikes. As of 2020, the system was operateing on 113 stations with over 500 bikes, serving more than 66,000 active users (City of Pittsburgh, 2020). Healthy Ride provided three pricing options during its operation. Users can pay the $2 rate for every 30 min ride, purchase standard plans that cost $12 per month for unlimited 30 min rides, or purchase deluxe plans that cost $20 per month for unlimited 60 min rides (Healthy Ride, 2022). Starting 2018, the system has partnered with the operating agency of public transit in Pittsburgh, Port Authority, to provide unlimited 15 min rides to Connect Card pass holders. Fig. 1 depicts the spatial distribution of 113 Healthy Ride stations. As the figure illustrates, two rivers flow through the city and separate it into regions in the north, center, and south. With most of the urban areas (such as universities, parks, commercial neighborhoods) concentrated in the center region, it also has the largest proportion of docking stations, taking about 86% of all the stations. The rest are distributed between the north and south regions in close proximity to the region in the center.

Fig. 1.

Healthy Ride bike-sharing stations in Pittsburgh, with three regions separated by rivers flowing through the city.

Anonymized bike-sharing dataset is publicly available from 2016 to 2021 on the Healthy Ride website (Healthy Ride, 2022). We excluded data from 2015 during which Healthy Ride only operated for seven months since it was launched in May. Quarterly data on Rentals and Stations provides Trip ID, Bike ID, Trip start day and time, Trip end day and time, Trip duration, Trip start station name with station ID, Trip end station name with station ID, and Usertype for each rental record, and Station ID, Station name, Latitude/Longitude coordinates, and Number of individual docking racks for individual rental stations. Dockless trips were reported since 2018 by recording bike IDs as the origin or destination without information to identify the actual locations.

Before the analysis, we identified trips with abnormal durations and removed them from the Rentals dataset, excluding those that were shorter than 60 s (0.36% of total) or longer than 6 h (3.84% of total) (Teixeira & Lopes, 2020). Preliminary analysis on annual trip volume was performed to determine the period of study. The result in Fig. 2 shows a clear “increase-decrease-rebound” pattern from 2019 to 2021 (2019 to 2020: −6.80%, 2020 to 2021: 19.52%), consistent with prior findings on bike-sharing usage in larger cities (e.g., Hu et al., 2021). Therefore, we focus the rest of the analysis on the years of 2019, 2020, and 2021, considering 2019 as the baseline (Tokey, 2020); a total of 321,544 bike-sharing trips were examined in the following analysis.

Fig. 2.

Annual volume of bike-sharing usage in Pittsburgh from 2016 to 2021.

4. Methodology

4.1. Temporal patterns

All analyses were performed in Jupyter Notebooks with Python libraries. To examine temporal patterns, we performed exploratory comparative analysis on trip volume and duration and aggregated them by month and day in a week, respectively. Then, they were grouped by hour to further compare temporal patterns across different days and years. To quantify the similarities of hourly usage, we selected cosine similarity as the measure. It calculates the cosine of the angle between vectors in a multidimensional space (Han et al., 2011) and has been applied to spatiotemporal data in several studies (e.g., McGuire & Tang 2013, McKenzie 2019). We performed year-wise comparisons of aggregated trip volume and median trip duration by hour in a week using cosine similarity. The median of travel duration was chosen to better characterize skewed distributions (Younes et al., 2020). In the context of our analysis, a cosine similarity value of 1 implies the exact same temporal distribution of weekly usage, whereas a value of −1 implies completely different patterns.

4.2. Spatial patterns with network theory approach

We performed geovisualization and network analysis on aggregated trip flows between bike-sharing stations, which marked the origin and destination of individual trips, to examine spatial patterns of bike-sharing usage. Dockless trips were excluded from the analysis due to the lack of information to locate their origins or destinations from the data available. Loop journeys, which describe trips that departed and arrived at the same location, were also removed from the spatial analysis as the trajectories were not available. The remaining 201,166 station-to-station, non-looping trips were included in the analysis.

As the prior temporal analysis suggested heterogeneous usage patterns between workdays and weekends, we grouped trip volume and visualized the flows separately for weekdays and weekends. This can further facilitate the comparisons between work-related and leisure rides. As a result, two separate weighted directed networks were built using data from each year, yielding 6 networks in total. For each network, rental stations were modeled as nodes. An edge was established between nodes that were connected by at least a single trip. Directions of edges were determined by the origin and destination from the rental records, with edge weights corresponding to the aggregated trip volume.

To characterize each flow network, we calculated a set of descriptive statistics. For network connectivity, we analyzed both transitivity, also known as the global clustering coefficient, and weighted clustering coefficient. The former simply computes the proportion of triangles connecting any triad of nodes (Wasserman & Faust, 1994). To further account for the edge directions and weights in our networks, we calculated the geometric average of edge weights (Onnela et al., 2005) to yield the fraction of directed triangles (Fagiolo, 2007) as weighted clustering coefficient. A combination of both metrics provides a better indication of how densely connected the networks are with respect to network topology as well as the number of bike-sharing flows.

For each network, we are interested to learn how the importance of stations regarding bike-sharing flows has changed since the pandemic. In network science, node importance is typically assessed through centrality measures, with betweenness centrality being the most commonly used in transportation networks (Borgatti, 2005). However, such measure calculates the shortest paths between network nodes, which may not describe the behavior of bike-sharing riders. Empirical evidence suggests that users of bike-sharing prefer routes that can significantly deviate from the shortest paths (Lu et al., 2018). As such, PageRank is more suitable for our networks since it evaluates the importance based on links connecting between nodes (Bongiorno et al., 2019). In weighted networks, the probability of visiting a subsequent node from a single node is proportional to the weight of the connecting edge (Gleich, 2015). To interpret the results with respect to our networks, bike stations with higher PageRank scores are of greater importance in controlling the flow of bike-sharing trips.

Another metric of interest is network assortativity. It describes the tendency of network nodes connecting to other nodes that are similar in certain ways (Newman, 2003). In our case, we explored whether bike stations with similar attributes were more likely to attract higher ridership in between, and how the trend has evolved since the pandemic. Real-world networks either show assortative mixing, where nodes alike tend to connect to each other, or disassortative mixing, where links are more likely to form between dissimilar nodes. We focus our analysis on two forms of assortativity: degree and attribute. In weighted directed networks, degree assortativity can be computed according to the direction of source node degrees (in-degree or out-degree) and target node degrees (in-degree or out-degree) connected by an edge from the source to the target (Foster et al., 2010). With respect to our networks, assortativity of in-degrees demonstrates whether stations with higher arriving bike-sharing traffic are more likely to connect. Similarly, assortativity of out-degrees describes the likelihood regarding departing traffic. In transportation networks, nodes with higher volume of arriving/departing traffic are usually considered as hubs (Fleurquin et al., 2014). These measures of assortativity can better reflect any changes in the tendency of bike-sharing hubs connecting to other hubs over the years since the pandemic. Mathematically, assortativity of in-degrees can be measured by the following Pearson correlation (Foster et al., 2010):

$$r\left(in,in\right)=\frac{E^{-1}{\sum }_i\left[\left(j_i^{in}-\overline{j^{in}}\right)\left(k_i^{in}-\overline{k^{in}}\right)\right]}{{\sigma }^{in}{\sigma }^{in}}$$

where E is the total number of edges. A similar equation can be derived for calcuating the assortativity of $r\left(out,out\right)$. A positive correlation indicates a trend of assortative mixing, while a negative value can be observed in disassortative networks. To test for significance, we compared the result with a null model comprising an ensemble of 1,000 randomly generated networks given the fixed degree sequence (Foster et al., 2010). We used z-scores to indicate if the observed assortativity was significantly higher or lower than the expected trend.

Furthermore, network assortativity was analyzed with respect to node attributes, including the location of bike-sharing stations in relation to the rivers and their land use. In other words, we are interested to know whether stations located in the same region as shown in Fig. 1 were more likely to connect by shared-bike trips, or if it can be observed among stations with the same type of land use. We first classified rental stations into those that were located in the northern, center, and southern regions of the city. To derive land use information, we mapped the stations into different zoning districts using the data published by the city of Pittsburgh (Pittsburgh Zoning Code, 2022), and further grouped them into residential, industrial/commercial, educational, and others (such as mixed land use) based on similar functions. Then, for each type of attribute, we calculated assortativity coefficient using the equation (Newman, 2003):

$$r=\frac{{\sum }_i{e}_{ii}-{\sum }_i{a}_i{b}_i}{1-{\sum }_i{a}_i{b}_i}$$

where e_ii represents the fraction of edges connecting nodes with the same attribute, while a_i and b_i are the fractions of edges that start or end at nodes of that attribute, respectively. For significance testing, for each type of node attribute, we built a null model comprising an ensemble of 1,000 randomly generated networks following the bootstrapping technique described in Christakis and Fowler (2013). More specifically, the random networks preserved the structure of the original network, but node attributes were randomly assigned to the nodes in the null model while preserving the prevalence of such attributes. Similar to degree assortativity, we also calculated z-scores to demonstrate the significance of the results.

Lastly, we explored how the commuting behavior of Healthy Ride users has shifted since the outbreak of COVID-19. To this aim, we calculated network reciprocity by comparing the ratio of bidirectional edges to the total number of edges in each network. Taking edge weights into account, we computed the percentage of fully reciprocated edges according to the equation (Squartini et al., 2013):

$$r=\frac{{\sum }_i{\sum }_{i\ne j}\min \left[w_{ij},w_{ji}\right]}{{\sum }_i{\sum }_{i\ne j}{w}_{ij}}$$

A higher value indicates greater resemblance of commuting patterns in a transportation network, assuming that a majority of bidirectional trips comprise regular commutes between home and work.

5. Results

5.1. Temporal patterns

Fig. 3 presents the results of bike-sharing usage aggregated by month and day in a week. It is evident that across all three years, shared-bike usage in Pittsburgh followed a seasonal pattern, in which higher volume and longer trips can be observed in warmer months. The observation is consistent with other larger bike-sharing systems (Talavera-Garcia et al., 2021) and the positive effect of warm, nice weather found in medium-sized systems (Caulfield et al., 2017). Compared to weekdays, weekend usage shows greater volume and longer trip duration. Moreover, we can observe the changes in bike-sharing usage following the critical events during the pandemic. For instance, trip volume in April 2020 dropped by 48.73% compared to the same month in 2019, right after the announcement of the stay-at-home mandate in Pennsylvania. The following months showed a fast rebound in bike-sharing usage, with only a less significant drop in ridership compared to the previous year (−8.21%, −12.22%, −11.53% in May, June, and July 2020 compared to 2019). Trip duration increased in the months following the stay-at-home order, suggesting the trend of using shared bikes for longer trips during the pandemic.

Fig. 3.

Trip volume (top row) and trip duration (bottom row) aggregated by month in a year and day in a week.

To examine temporal patterns by hour, we visualized trip volume and median trip duration in Fig. 4, Fig. 5 . Comparisons of daily distributions in Fig. 4 suggest a “trimodal distribution” of weekday usage in 2019. Hourly volume reached its peaks in the morning, around lunch, and arrived at daily maximum during rush hours in the evening. Such a pattern is consistent with bike-sharing systems of different scales as suggested in previous studies (Bean et al., 2021; Mateo-Babiano et al., 2016). Comparisons across years suggest that during 2020, bike-sharing usage on weekdays demonstrates a single peak pattern, which typically occurs over the weekends.

Fig. 4.

Hourly trip volume of bike-sharing from 2019 to 2021. Individual days are separated by vertical solid lines, noons are marked by vertical dotted lines.

Fig. 5.

Hourly trip duration of bike-sharing from 2019 to 2021. Individual days are separated by vertical solid lines, noons are marked by vertical dotted lines.

Results on cosine similarities in Fig. 6, Fig. 7 further confirm our observations on the hourly usage patterns. Since 2019, there is a persistent increase in similarities between weekdays and weekends. In other words, patterns of trip volume became less distinguishable among days in a week. Year-wise comparisons in Fig. 7 suggest that hourly bike-sharing usage during 2019 and 2020 were least similar. All values of the cosine similarity are greater than 0.9, suggesting similar hourly usage of shared bikes.

Fig. 6.

Similarities of hourly volume distribution between days in a week in 2019 (left), 2020 (middle), and 2021 (right).

Fig. 7.

Similarities of weekly distribution of hourly volume (left) and median trip duration (right) across the years.

For trip duration, one can observe from Fig. 5 that in general, trips during weekdays were shorter than those over the weekends. The figure clearly shows that trip durations in 2020 were much higher than 2019 and 2021 for all days in a week. Results from cosine similarities in Figs. 7 and 8 further showed how 2020 stood out from the other years. For instance, duration of bike-sharing trips was the most distinct between weekdays and weekends during 2020 (Fig. 8). Moreover, 2019 and 2020 showed the most distinct patterns of trip volume as well as trip duration (Fig. 7). Comparing the graphs from left to right in both Figs. 6 and 8, we can find that for the year of 2021 following the outbreak, trip duration returned to a pattern similar to the pre-pandemic times whereas trip volume became more deviated from 2019.

Fig. 8.

Similarities of the distribution for hourly median duration between days in a week in 2019 (left), 2020 (middle), and 2021 (right).

5.2. Spatial patterns

5.2.1. Flow visualization

We performed descriptive analysis on loop journeys before excluding them from the network analysis, since they may inform leisure usage of bike-sharing (Mateo-Babiano et al., 2016). They constitute 27.27% of station-to-station trips between 2019 and 2021. Fig. 9 depicts the volume and duration of loop trips aggregated by day in a week. It can be observed that most loop journeys took place on weekends, consistent with their characterization of being mostly recreational rides. The significant increase in volume and duration in 2020 indicates a potential rise in using bike-sharing for leisure trips during the pandemic.

Fig. 9.

Bike-sharing trip volume (top) and duration (bottom) of loop journeys.

Fig. 10 shows the geovisualization of daily bike-sharing flows. Dots represent Healthy Ride stations and are connected by lines whose widths and opacities are proportional to the average daily trip volume. Row-wise comparisons yield a substantial decrease in usage from 2019 to 2020, and a clear rebound in the following year. We further distinguish between within-region (e.g., trips started and ended at the stations in the north region) and between-region trips (e.g., trips started from the station in the north region and finished at the station in the south region) by whether they started and finished in the same region illustrated in Fig. 1. Upon close inspection, we found a sharp decline in between-region trips from 2019 to 2020, which is more profound during the weekdays. The following year shows a clear rise in trip volume that connects different regions over the weekends, but for workdays between-region trips remained sparse. In terms of rides within the same region, Fig. 10 highlights two areas with interesting patterns. For the subregion located in the center, the bike-sharing rides experienced a steady increase in volume especially over the workdays. This area is home to the campuses of Carnegie Mellon University and University of Pittsburgh, with most of its residents being college students and university employees. The observed spatial patterns indeed align with the universities’ responses to the pandemic, such as the return of in-person instructions in 2021 (University of Pittsburgh., 2021). Another area with a clear pattern of evolvement is the region in the south. Bike-sharing trips inside the region increased over the years, even during the COVID-19 pandemic. The stations in the south side are in close proximity to riverside parks, restaurants, shopping, and bike trails.

Fig. 10.

Flow of daily bike-sharing trips between rental stations in Pittsburgh on weekdays (left) and weekends (right) in 2019 (top), 2020 (middle), and 2021 (bottom).

5.2.2. Network analysis

Table 2 summarizes the descriptive statistics of bike-sharing networks representing workdays and weekends from 2019 to 2021. The values of transitivity suggest that topologically, weekday networks are more densely connected than weekend networks, but the difference became less obvious since 2019 as the transitivity of weekend networks increased. Taken together, they indicate the emerging usage of shared bikes for leisure between previously disconnected stations. Considering the volume and direction of trip flows, the weekend network of 2021 is the most densely connected among the rest. Also notable is the pattern of decrease-and-rebound in the weighted clustering coefficients that characterize weekend usage, which is different from the decrease in the connectivity of work-related rides.

Table 2.

Descriptive statistics of bike-sharing flow networks.

Network		Number of nodes	Number of edges	Weighted clustering coefficient	Unweighted clustering coefficient (transitivity)
2019	Weekdays	113	5421	0.02	0.61
	Weekends	113	4025	0.02	0.53
2020	Weekdays	100	4322	0.02	0.60
	Weekends	100	3731	0.01	0.56
2021	Weekdays	101	4715	0.01	0.62
	Weekends	100	4153	0.03	0.60

Fig. 11 and Table 3 present the results of PageRank scores on the importance of bike-sharing stations of each network. On the one hand, as the figure indicates, the scores are positively correlated over the years, implying that important stations of bike-sharing flows are quite similar. Moreover, the correlation between 2019 and 2021 is the least evident, suggesting more significant shifts in important stations before and after the outbreak in 2020. On the other hand, the figure shows that the PageRank scores for weekend networks are concentrated around lower values, whereas a more uniform distribution can be found for the weekdays. In addition, we examined stations located in the different regions separately and compared their PageRank scores across different years in Table 3. It shows how stations in separate regions differ in their importance over the years of the analysis. While the stations in the north and in the center have highly correlated node importance, no significant linear correlations were found in the southern stations during workdays except for the comparison between 2020 and 2021 (2019 and 2020: p = 0.34, 2019 and 2021: p = 0.16). Again, for the weekends, none of the comparisons show linearly correlated PageRank scores for those stations in the south (2019 and 2020: p = 0.14, 2020 and 2021: p = 0.06, 2019 and 2021: p = 0.29).

Fig. 11.

PageRank scores of the networks on weekdays (top) and weekends (bottom) from 2019 to 2021. Colors represent the regions in which each station is located.

Table 3.

Pearson correlation coefficients of PageRank scores of bike-sharing stations between flow networks.

		r(pr19, pr20)	r(pr20, pr21)	r(pr19, pr21)
Weekdays	All stations	0.85	0.87	0.76
	Stations in the north	0.88	0.96	0.82
	Stations in the center	0.87	0.85	0.75
	Stations in the south	0.39	0.90	0.55
Weekends	All stations	0.90	0.94	0.91
	Stations in the north	0.95	0.95	0.95
	Stations in the center	0.93	0.95	0.93
	Stations in the south	0.56	0.69	0.43

The results of in-degree and out-degree assortativity of each network are shown in Table 4 . Trip flows in the weekdays of 2021 show assortative mixing regarding both arriving and departing bike-sharing traffic, in which hub stations were more likely to attract rides between them, while the rest of the networks demonstrate disassortative mixing patterns. Compared to the null model, the mixing patterns are significant at the alpha level of 0.001.

Table 4.

Results of assortativity and reciprocity of the bike-sharing networks.

Network	Degree assortativity		Attribute assortativity		Reciprocity
	In-degree assortativity coefficient (z-score)	Out-degree assortativity coefficient (z-score)	Station location (z-score)	Land use (z-score)
2019 weekdays	−0.01 (18.36)	−0.01 (18.99)	0.08 (9.18)	0.05 (9.85)	0.32
2019 weekends	−0.03 (17.93)	−0.03 (17.79)	0.09 (9.02)	0.06 (8.32)	0.29
2020 weekdays	−0.05 (7.52)	−0.04 (8.67)	0.09 (10.83)	0.05 (8.32)	0.34
2020 weekends	−0.03 (11.95)	−0.03 (11.37)	0.09 (8.97)	0.05 (8.20)	0.30
2021 weekdays	0.01 (20.51)	0.04 (23.94)	0.07 (8.09)	0.05 (8.38)	0.34
2021 weekends	−0.03 (13.73)	−0.03 (12.55)	0.07 (7.87)	0.05 (7.93)	0.32

Table 4 also summarizes the results on assortativity measured by node attributes and network reciprocity. For stations in different regions, a significant assortative mixing can be observed for all the networks, indicating a strong tendency of using shared bikes for within-region trips. The pattern is particularly evident in 2020 compared to the other years, while the coefficient became smaller in the following year. The finding is consistent with the earlier observations on the geovisualization which demonstrate a significant rise in between-region trip volume. Similarly, for land use, the networks again illustrate assortative mixing patterns, where stations with the same type were more likely to connect by bike-sharing rides. However, the variability across the networks appears less evident than other measures of assortativity, which may yield different results if the time window of aggregation was selected differently or the zoning districts were categorized using a different approach. The last column of Table 4 presents the results of network reciprocity. It shows that weekends had lower reciprocity than their weekday counterparts, which is consistent with the commuting behavior of bike-sharing users over the workdays. Both the proportions of reciprocal flows during weekdays in 2020 and 2021 were higher than the value prior to the pandemic, while the trips over the weekends also experienced a slight increase in bidirectional bike-sharing flows over the years.

6. Discussion

6.1. General discussion

Taking Healthy Ride in the city of Pittsburgh as a case study, we examined how spatiotemporal patterns of a medium-sized bike-sharing system have changed over the years prior to, during, and following the outbreak of COVID-19. Our method combines comparative analysis, geovisualization, and network theory approaches, providing an example of how the usage of shared micromobility evolves during highly disruptive events in a system that operates on a smaller scale. Currently, it remains largely unknown how such systems have been affected by the pandemic, especially over the course of multiple years before and after the outbreak. Our research contributes to existing literature by addressing the lack of such understanding and further allows comparisons across different geographical regions or modes of transportation. A better understanding of urban micromobility requires an extensive examination of systems with varying scales, which can further promote the adoption of bike-sharing and inform practitioners in making critical decisions. Our research also draws attention to the uneven number of published studies, in which bike-sharing systems of a smaller scale have been less investigated.

Overall, the annual volume of bike-sharing trips followed a “slight decrease and strong rebound” pattern since the outbreak, with a 6.80% decrease in 2020 and a 19.52% increase in 2021. Similar trends have been observed in a number of bike-sharing systems operating on a greater scale (e.g., Bi et al. 2022, Hu et al. 2021, Jurdak 2013, Song et al. 2022, Teixeira & Lopes 2020). The drop in ridership demonstrates the substantial impact the pandemic had on bike-sharing mobility across systems of varying scales. Most cities in 2020 adopted similar measures to prevent the spread of the virus, including lockdowns and stay-at-home orders, leading to the prevalent decrease in travel demands. In the same year, we found a higher percentage of trips made by non-subscribers compared to the periods before and after the pandemic (49.14% in 2019, 64.20% in 2020, 53.26% in 2021). Although we did not contrast bike-sharing with other modes of travel in the study, the observation hints at a possible modal transfer among city residents who started using shared bikes as a substitute for other means of transportation during the outbreak. In the year of 2021, trip volume in Pittsburgh reached the maximum since the adoption of bike-sharing. Together with our observation on the nearly instant rebound in monthly ridership following April 2020, the strong recovery illustrates that bike-sharing can be extremely robust and resilient to disruptive events even for a smaller-scale system without a long history of deployment, and can continue to grow its ridership. Examining weekends and weekdays separately provides additional insights into different purposes of travel. The rise in weekend trips suggests that the growing demand for leisure rides was the primary contributor to maintaining ridership during the outbreak. Combining the results from loop journeys, we showed an emerging preference of riding shared bikes for recreational purposes since the pandemic. Similar interests in cycling for exercise and recreation have been observed in other cities during COVID-19 (Albuquerque et al., 2021; Buehler & Pucher, 2021). Taken together, they suggest that the demands for work-related and leisure travel could be differently affected by disruptive events. With lockdowns and stay-at-home orders, city residents may show an increasing need for recreational trips to compensate for the reduction in work-related travel.

Further comparisons showed that as the pandemic continued, temporal patterns of trip volume between weekdays and weekends became increasingly similar. In other words, daily usage of bike-sharing became less distinguishable, where weekday patterns shifted from “trimodal distribution” to single-peak usage, a typical pattern for weekends before the pandemic. Such trend has been illustrated in other larger cities (Chai et al., 2020; Pase et al., 2020) as well as in systems of a similar scale (Li et al., 2021), but over a shorter time period. The most important contributor could be the promotion of work-from-home in most urban regions, leading to the reduction in commuting behavior that has been similarly observed across many transportation modalities (Buehler & Pucher, 2021). Moreover, our results illustrate that the trend continued to develop in the following year, with ever more homogeneous patterns between weekdays and weekends (all cosine similarities greater than 0.95). However, we did not find other works to compare our findings with. It remains to be answered by future research whether the observation is specific to systems with similar scales or to the city in the study, and what the potential explanations could be. Another important finding is how annual trip volume has been redistributed among hours in a week. It illustrates the importance of aggregating data using different time intervals, for instance, by hour, by day in a week, and by month, to discover changing patterns in ridership at different temporal granularities.

Beside trip volume, we also examined temporal patterns of trip duration and how they have evolved across the years since the pandemic. Trip duration increased especially around the beginning of the COVID-19 outbreak, a trend that has been observed in many other cities (Albuquerque et al., 2021; Hu et al., 2021; Pase et al., 2020; Xin et al., 2022). It shows that bike-sharing can play a significant role in supporting urban movements of various purposes during disruptive events like the pandemic, across systems of different operating scales, and has likely served as a substitute for other travel modalities for longer-distance trips. Factors contributing to the longer usage could include reduced operation of public transportation, fear of disease transmission in crowded spaces, and increased awareness of physical health (Shokouhyar et al., 2021). Comparing changes in trip volume and trip duration, we found distinct patterns of development over the years. While trip volume became more different from pre-pandemic times, trip duration showed a trend of returning to the prior patterns. It demonstrates how the two variables describe distinct changes in bike-sharing usage, characterizing different aspects of user behavior. As a result, future studies may find it necessary to analyze travel demand using variables that describe volume and duration separately.

Observations on the geovisualization showed the different magnitudes with which between- and within-region trips have evolved. We found a general sharp decline in trip volume between different regions, except for the unprecedented demand over the weekends in 2021. Results on the mixing patterns using station locations confirm our observations on between-region trips, showing that the pandemic has affected between- and within-region travel differently over the years. Ridership across the rivers is more sensitive to disruptive events during both the weekdays and the weekends, exhibiting lower resiliency compared to within-region trips. Many reasons may explain the observed decline in between-region usage of shared bikes. Beside the decreased demand for commuting in response to work-from-home, we believe there exist other spatial cognitive factors that could have been in play in the observed aversion of between-region travel. More specifically, empirical evidence from psychological studies suggests that barriers could distort the distance judgement between locations (Egenhofer & Mark, 1995; Tansan et al., 2022). It is likely that the residents perceived destinations that were located across the rivers being further than the actual distance, thus were less willing to travel across different regions using shared bikes. Although not being the focus of this study, we believe it is an interesting research direction to systematically examine the possible effect of natural barriers on urban mobility and disassociate psychological factors from other potential contributors.

In addition, the geovisualization shows how the proximity to universities and recreational activities could have impacted bike-sharing usage since the COVID-19 outbreak. To rule out the effect of possible station expansion, we checked the number of racks of the stations of interest and found little to no changes across the years. To be more specific, bike-sharing is an ideal choice for travelling across the campus, primarily by university students and employees, which has led the increase in ridership the year following the outbreak. Such an increase is more evident than other regions with similarly concentrated point-of-interests in the city. The observation may be specific to Pittsburgh, as it involves two populated universities adjacent to each other, with a total of more than 45,000 students in enrollment each year. It will be interesting to compare the results with bike-sharing usage around adjacent university campuses from other cities, to further explore the underlying association between student population and usage of shared micromobility during the pandemic. Moreover, we also found an emerging ridership connecting between stations in the south side even during the outbreak. These stations are not only close to parks and shopping, but they are also located along a trail with dedicated cycleways. One possible explanation, consistent with our previous finding on the increased leisure usage, is that more residents chose shared bikes for leisure trips and in particular preferred these stations for the convenience of getting on and off the bike trail. Proximity to cycling infrastructure has been shown to positively associate with bike-sharing usage in many cities (Bustamante et al., 2022; Chibwe et al., 2021; Pase et al., 2020). Our findings indicate that it can also promote bike-sharing usage in smaller-scale systems especially during disruptive events like the COVID-19 pandemic.

At the network level, we analyzed and compared connectivity, degree assortativity, and reciprocity across the years regarding weekday and weekend usage. The separate measures of connectivity (global and weighted clustering coefficients) indicate differing trends with respect to network topology of bike-sharing flows. Topologically, the findings suggest that while weekend networks became increasingly connected, work-related networks experienced a slight decrease in connectivity when the pandemic first hit followed by a rebound in the next year. Previous studies have also shown that bike-sharing networks in other cities became topologically less clustered in the first year of the pandemic (Hu et al., 2021; Xin et al., 2022), but weekday and weekend usage were not examined with separate networks. Together with the results on weighted clustering coefficients, we showed that bike-sharing stations became more densely connected following the pandemic, even for a system operating on a smaller scale like Healthy Ride.

Considering both arriving and departing traffic, all networks show disassortative mixing patterns except for the weekdays during 2021. In a previous study, Bongiorno et al. (2019) found that the network of bike-sharing flows in Boston showed assortative mixing patterns. There are several possible explanations for the discrepancy in the findings, which requires investigations in future research. First and perhaps the most important, the bike-sharing system in Boston operates on a much greater scale than the one in Pittsburgh, with more rental stations densely distributed in the urban region. For a smaller-scale operation like Healthy Ride, hub stations may play a more important role in connecting non-hub locations, as a result of the stations that are sparsely and unevenly distributed. Second, the period of study in Bongiorno et al. (2019) is from 2014 to 2015, different from ours that covers 2019 to 2021. Interestingly, in comparing the assortativity coefficients across the years, we found little to no change in the mixing patterns over the weekends. For weekdays, the 2020 network was more disassortative than the previous year, but became assortative in the following year. Taken together, they indicate that during the outbreak, users were more likely to use bike-sharing to travel between stations with higher traffic (hubs) and less busy locations, especially over the weekdays. The demand may be attributable to the lack of alternative modes of transportation during lockdowns. Shared bikes thus became an important mode for connecting all locations including those with lower traffic volume.

Another network characteristic that has dynamically changed since the pandemic is reciprocity. The result supports our previous observations on the possible shift of travel modes to bike-sharing and the growing demand for leisure rides. More specifically, the uptick in weekday reciprocity reveals that users were more likely to use shared bikes for both directions of travel. Weekend reciprocity also increased as the network flow of recreational rides surged. Previous studies found that bike-sharing networks had higher reciprocity than pedestrian flow networks (Bongiorno et al., 2019). However, it remains to be fully understood how the difference between bike-sharing and walking has manifested during the COVID-19 pandemic. Our case study can be used by future research to compare with the commuting behavior of other urban regions, to discover the similarities and differences in how bike-sharing is used by commuters in smaller-sized systems.

At the node level, we investigated important stations using PageRank and mixing patterns of node attributes. The outbreak of COVID-19 did not lead to significant changes in the stations that were important for bike-sharing flows, as their importance was quite similar across the years in our comparison; except for stations in the south, whose importance was found to change more substantially from year to year. Previous observations on the geovisualization also demonstrated the unique patterns of trip flows among these stations, which may contribute to the shifts in their importance. Moreover, from the distribution of PageRank scores, we found that weekend trips were more concentrated on hubs when compared to weekdays, with a few stations having higher scores of importance. In other words, weekend networks showed a more prominent hub-based structure than weekday networks, with only a small number of stations more popular than the rest. The difference between weekdays and weekends further emphasizes the importance of distinguishing between work-related and leisure rides in understanding bike-sharing flows.

Finally, our findings on node attribute assortativity showed an increased preference for within-region rides during the year of the outbreak, with the assortativity coefficients changed from 0.8 to 0.9, followed by a rise in between-region rides in the next year, with the coefficients dropped to 0.7, consistent with the prior observations on the changes in these two types of trips. These results demonstrate how shared bikes can greatly support urban mobility and accessibility by connecting regions separated by natural barriers, especially during the times of recovering from the outbreak. Also, the demand for between-region travel calls for more attention to the bike-related infrastructure that supports riding across the barriers, including bridges and ferries. It also demonstrates the need for proper planning of rental stations for docked bike-sharing systems in order to promote reachability and equitable access to more locations in the city (Shokouhyar et al., 2021), which is particularly important in the times of disruptive events.

6.2. Practical implications

Our results have several important implications for decision-makers looking to introduce bike-sharing services or improve current systems, in particular for cities smaller than metropolitan regions. First and foremost, our findings demonstrate the resiliency of bike-sharing systems during disruptive events as influential as the COVID-19 pandemic, even those operating on a smaller scale. Policymakers from smaller sized cities can take our research as a case study of how bike-sharing systems can greatly boost resiliency of urban transportation systems and complement other means of urban travel during challenging events. Together with other economic and environmental benefits offered by shared bikes, decision-makers should seriously consider deploying bike-sharing services or prioritizing bike-sharing in preparation for future disruptions to improve urban mobility. Also, the boost in ridership after incorporating unlimited rides for public transit card users in Pittsburgh (Healthy Ride, 2022) provides a great example of how policymakers can promote bike-sharing when introducing it to city residents by encouraging collaboration with other means of transportation such as public transit.

Moreover, the changes in usage patterns over the period of the pandemic highlight some important aspects service operators and policymakers should consider in optimizing supplies, improving infrastructure, and promoting bike-sharing. One key distinction of the bike-sharing trips analyzed in this study is the rides between versus within urban regions separated by natural barriers. Waterways that segment urban areas into naturally disconnected regions are commonly found in cities worldwide. For system operators, categorizing bike-sharing trips with respect to natural barriers offers another lens of understanding the demand, which may facilitate making decisions on distributing bikes among stations. For city planners and policymakers, our findings stress the importance of integrating and improving bike-related infrastructure that serves to connect those separated regions, including bridges and ferries. Not only can such measures enhance the safety and efficiency of cycling, but they can also improve accessibility and reachability of urban regions. Other findings from our study can also inform demand modeling during normal periods as well as during the pandemic. For instance, the observations from our geovisualization suggest that areas around universities and bike trails experienced rising demands for shared bikes during and following the pandemic. Besides, we also demonstrate different usage patterns of work-related and recreational rides since the outbreak of COVID-19. Bike-sharing operators and policy makers should therefore consider the shift in purposes of using shared bikes (Li et al., 2021) to better understand previous demands and make future predictions. Important stations identified by network modeling should also be taken into account in prioritizing maintenance and monitoring, expanding station capacity, and allocating the number of bikes available to the users.

7. Conclusion

Taking Pittsburgh as a case study, we examined the changes in spatiotemporal patterns of bike-sharing usage of a medium-sized system over the years before, during, and following the outbreak of COVID-19. Specifically, we focused on distinguishing between weekday and weekend usage to investigate whether commuting and leisure rides showed different patterns of evolvement. The results showed that the usage patterns of weekdays and weekends indeed changed in different directions with distinct magnitudes. Moreover, we found evolvement in temporal patterns that were distinct between bike-sharing volume and duration, and over various aggregation scales. The analysis of spatial patterns also demonstrates the distinction between trips within and between regions separated by natural barriers. Overall, our results suggest that bike-sharing is a resilient means of transportation during interruptive and uncertain events, contributing to current literature by providing a detailed case study of a smaller-scale bike-sharing system. To fully understand how shared bikes, and more broadly, shared micromobility will develop during events that challenge urban mobility, it is of great importance to investigate bike-sharing systems of varying scales and identify generalizable and unique usage patterns. Such an understanding will help system operators and policymakers take a better position in preparing for future challenges and building resilient and sustainable urban transportation systems.

One major limitation of our study is that the observed changes in bike-sharing usage may be contributed by factors other than the pandemic. For instance, Pittsburgh introduced shared e-scooters in July 2021 (Downtown Pittsburgh, 2021), which could have affected how shared bikes have been used since then. Moreover, like many urban regions, the city of Pittsburgh has been constantly seeking to promote bike-sharing and other sustainable travel modalities. An effort is the Bike(+) master plan published in 2020 that aims for “building a safe, comfortable, and convenient bike network for all types of riders and all types of trips” (City of Pittsburgh, 2020). In response to the plan, the city has been actively installing on-street bike facilities, adding new trails, and incorporating bike lanes into the existing road network. Although the primary aim of this study was to discover the changes in usage during the pandemic, it will be meaningful for future research to model bike-sharing usage of smaller-scale systems with explanatory variables, incorporate comparisons before and after COVID-19, and contrast bike-sharing with other means of transportation. Such studies will not only enrich existing literature but also provide valuable insights for system operators and policymakers in improving urban mobility.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

• The data and code for replicating the result have been uploaded to OSF: https://osf.io/kuzjc/?view_only=3c5640990d734aed8e6360df90762705.