Assessing Bike-Transit Accessibility

Abstract

Bicycles are a potential first-last mile mode that can augment the service area of public transit, yet it is difficult to fully account for bike-to-transit trips in planning and travel demand modeling processes. This paper presents a methodology for assessing bicycle first-last mile trips from one area to many possible areas using three visualizations on accessibility, travel times, and transit mode(s) utilized. Two configurations of bicycle first-last mile travel are considered: bringing the bicycle aboard transit to have the bicycle for biking at both ends of the trip (bike-transit-bike) and leaving the bike at the first stop (bike-transit-walk). Three locations in and near Atlanta, GA, U.S., are selected for analysis, and the optimal routes to all possible destinations in the transit service area are calculated for walk-transit-walk, bike-transit-walk, and bike-transit-bike. The walking and biking portions of trips are modeled using Dijkstra’s algorithm, and the transit portion is modeled using the round-based public transit optimized routing (RAPTOR) algorithm. Results indicate that bike-transit-bike and bike-transit-walk decrease travel and wait times for transit, and in many cases reduce the number of transfers required compared with walk-transit-walk. Transit services with higher travel speeds or frequencies, such as heavy rail, greatly increased the number of accessible destinations and reduced travel times. Thus, an origin’s distance to rail service had a major impact on the number of accessible TAZs. Planners and engineers can use this research to examine how public transit service changes and new cycling infrastructure can affect the accessibility of bike-transit trips.

Keywords

bicycles human factors modeling and forecasting planning and policy public transportation accessibility multimodal

Low-density land use, sprawl, and Euclidean zoning (i.e., separation of commercial and residential land uses) in U.S. cities reduce the effectiveness of public transit by reducing the number of homes, amenities, services, and jobs near transit stops ( 1 , 2 ). This gives rise to the first-last mile problem, where transit riders must travel long distances to access transit from their origin and from transit to their final destination ( 3 ). This problem exacerbates the travel time difference between private motor vehicles and public transportation, negatively affecting those that rely on public transportation. One study in the Metro Atlanta area found that the average difference in travel time between driving and using public transit was 72 min for a selection of 2,082 GPS traces of automobile trips, which corresponded to a Transit Capacity and Quality of Service Manual (TCQSM) Level of Service of F ( 4 , 5 ). Another study of 14,000 trips in Atlanta and Seattle, routed through Google Maps’ routing API, found that the average travel time by public transit was more than twice the travel time by automobile ( 6 ).

Bicycles can extend the service coverage area of a transit stop or station by allowing transit users to cover a greater distance in the same amount of time. While the TCQSM sets walking distances for transit service coverage at 0.25 and 0.5 mi for bus and rail stops, it sets biking distances at 1.25 and 2.5 mi for bus and rail stops ( 4 ). Not only can people reach transit stops faster on a bicycle than they could by walking; people using bicycles can also reach more transit stops within the same time frame. Lastly, people may be able to avoid bus feeder routes and cycle directly to higher service quality transit routes (such as rail).

Using a bicycle as a first-last mile mode (henceforth referred to as “bike-transit”) is often more flexible than park-and-ride (i.e., drive, and park car at a transit station) journeys, in that a bike can be parked at the transit station, brought aboard the transit vehicle for use at the destination transit stop, or stored at the destination stop to be ridden to the final destination. Bikeshare or other shared micromobility services could also be used in place of a personal bicycle. In the U.S., most current bike-transit trips involve bringing the bike aboard transit, yet studies have shown that secure bike parking can result in more people parking their bike at the station ( 7 , 8 ).

Bike-transit trips are not common in the U.S. as the vast majority of transit trips start and end on foot ( 9 ). Only about 1% of Metropolitan Atlanta Rapid Transportation Authority (MARTA) riders used a micromobility mode (e.g., bicycles, scooters, skateboards) to access transit compared with 85% by walking and 14% by shared or personal vehicle ( 10 ). This statistic stands in stark contrast with countries with improved bike-transit infrastructure, such as the Netherlands, where the bicycle was used for 43% of first-mile trips accessing the train and 14% of last-mile trips from the train to destination ( 11 ). The factors that influence people to choose bike-transit depend on the quality and speed of public transit, the provision of cycling infrastructure, the availability of secure bike parking, and land use ( 12 , 13 ). In public transportation systems with park-and-ride oriented stations, such as MARTA, stations can be surrounded by large, high-speed arterial roadways where the majority of people would not feel comfortable or safe cycling. A transit station’s theoretical catchment area for cyclists can shrink drastically when only considering low-stress streets (i.e., streets that the majority of people would feel comfortable or safe cycling on) as viable routes to transit ( 14 ). Despite bike-transit’s potential for shortening travel times, bike-transit is not typically incorporated into the traditional travel demand modeling process ( 15 ).

The purpose of this paper is to demonstrate, using bicycle and transit shortest path algorithms, how bike-transit improves transit’s accessibility to destinations by reducing overall travel times, transit waiting times, and the number of transit transfers needed. Bike-transit’s impact on accessibility likely varies with the built environment and available transit service as well; therefore, three different locations in Atlanta, GA, are studied. Currently, this research does not incorporate cyclists’ preferences for cycling infrastructure and low-stress streets. However, these preferences can be modeled using impedance functions in the bicycle routing algorithm, and these impedance functions are currently being developed through the authors’ ongoing research.

In the next section of this paper, previous work in modeling bike-transit access is described. Then, in the Methodology section, the study area and selected locations, the street network and transit data available for these analyses, and the routing procedure are presented. The results are then presented and discussed. The Results section is followed by the Limitations, Future work, and Conclusions section.

Literature Review

Unlike traditional shortest path routing with bicycles or automobiles, public transit operates on a fixed schedule. One of the past methods for transit shortest path routing is a time-expanded Dijkstra algorithm. This method is inefficient, because every trip stop in the transit network is modeled as a node; therefore, a route consisting of 5 stops and 10 trips would require 50 nodes, 40 links, and the links needed for transfers to and from other routes (16 –18). Another past method involved converting the transit network to a static network graph with average wait times ( 19 ). This second method loses the ability to properly consider how the transit schedule affects travel time, which is often important in the U.S. where high-frequency transit is rare.

Today, there are multiple transit routing algorithms that are used in popular services, such as Google Maps and OpenTripPlanner (OTP). OTP uses the round-based public transit optimized routing (RAPTOR) algorithm ( 16 ). RAPTOR is not a traditional graph-based routing algorithm like Dijkstra’s algorithm. RAPTOR utilizes a transit system’s property of fixed routes and schedules to focus on finding the earliest arrival time for each stop that can be reached from a source transit stop, given a specified number of transfers. Each round of RAPTOR is split into two phases: the first phase searches for stops along a route, while the second phase searches for footpaths to the final stop. Every stop along a route is visited exactly once per round, and the earliest arrival time at a stop is updated if an earlier arrival time is found.

While existing trip planning services are capable of outputting the shortest paths for bike-transit trips, there have been only a few instances where these services have been used for assessing bike-transit at a regional scale. This is because bike-transit trips are computationally intensive to calculate, given the number of possible transit stop pairs and departure times. However, there have been a few instances of these assessments in the literature.

First, using a combination of Google Maps API for the bike portion and OTP for the transit portion to solve the bike-transit travel time, Ling assessed the feasibility of replacing single-occupancy-vehicle trips from and to a large employer in Atlanta, GA, with alternative modes ( 20 ). Bike-transit was one of several modes assessed, and the ratio of each alternative modes’ travel time was compared with the drive-alone travel time, a TCQSM quality of service metric. However, only rail transit was considered, and cyclists only brought their bikes aboard transit. Employees within biking distance of rail stations had a median travel time for bike-transit of 40 min and driving of 24 min. Bike-transit was considered viable if the bike-transit time was less than or equal to twice the driving time.

In addition, Conveyal, a company that provides a web-based analysis tool for assessing transit accessibility to jobs, has extended their tool to incorporate level of traffic stress for bike-transit trips ( 21 , 22 ). Conveyal also incorporates uncertainty in passenger departure times, transfers, and transit vehicle arrivals ( 23 ). This tool allows for all possible configurations of bike-transit, and considers both bus and rail for routing. In addition to the online tool, Conveyal has developed several open-source libraries that can be used for solving bike-transit trips online or offline. However, Conveyal does not currently store the underlying routes that are being taken by transit to reduce storage requirements, so it is not as clear what transit routes are being used the most to access jobs.

This paper extends previous work done in assessing how bike-transit improves the effectiveness of public transit by visualizing and comparing average travel times, accessibility, the minimum number of transfers, and transit modes/routes utilized for bike-transit and walk-transit. To communicate the increase in accessibility, three visualizations are generated for three locations of varying distances from bus and heavy rail service in Atlanta, GA.

Methodology

The methodology of this research compares the walk-transit mode with the bike-transit mode for three selected locations using bicycle and transit shortest path routing. For this paper, the shortest path routing algorithm is based on the shortest travel time; it does not include cycling-specific impedances that account for cyclists’ routing preferences for low-stress streets and cycling infrastructure. These cycling-specific impedances are being developed in the authors’ ongoing research using revealed preference data from cyclists. Walk-transit (referred to as walk-transit-walk) and two types of bike-transit are examined. The first is where the person brings the bike aboard transit (referred to as bike-transit-bike), and the second is where a person parks their bike near the first transit stop and walks to their final destination after the transit trip (referred to as bike-transit-walk).

Analysis Context and Data Employed in Analyzing the Bike-Transit Mode

This application of the bike access to transit methodology uses the public transit system in Atlanta. MARTA’s rail stations have elevators and wide ADA fare gates to accommodate users with bicycles. MARTA’s buses are equipped with a bike rack that holds two bikes; if the bus bike rack is full, additional cyclists must wait for the next bus. While there are no restrictions for bringing bicycles aboard MARTA’s heavy rail vehicles, there are no bike racks for storing a bike during travel.

For transit schedule information, the general transit specification feed (GTFS) data for service between December 25, 2022, and April 21, 2023, were used; at that point, service had largely been restored to pre-pandemic service levels. Only MARTA’s heavy rail and bus fixed route transit services were considered, and only the static GTFS schedule was used.

The study area was delineated using a Euclidean buffer distance of 2.0 mi around all of MARTA’s bus and rail stops. This value is between the TCQSM’s recommended bicycle access thresholds of 1.25 mi for bus stops and 2.5 mi for rapid transit stations. Traffic analysis zones (TAZ) from the Atlanta Regional Commission’s (ARC) 2020 activity-based model run that fell within this buffered area served as potential origins and destinations. There were a total of 2,221 TAZs in the study area and 5,922 TAZs in the entire metro area.

There are 113 unique bus routes and four heavy rail routes in the study area, as shown in Figure 1. The bus routes are represented with grey lines and the rail routes are shown with white lines with black borders. Two of the rail routes are north-south oriented, and the other two rail routes are east-west oriented. The rail station at the intersection of these routes, Five Points (purple), serves as the transfer point between the rail routes. TAZs are shown as light grey dots. The City of Atlanta’s borders are crosshatched in red, but MARTA’s potential bike-transit service area extends far beyond these borders. The street network for the study area was retrieved from OpenStreetMap (OSM) through Osmnx and Overpass API (24 –26). The OSM data were filtered to only include public roads and multi-use paths that allowed bicycle travel. Supplementary data from ARC’s 2022 Regional Bicycle Facility Inventory were used to add cycling infrastructure not present in OSM ( 27 ). The final network was composed of 366,926 links and 323,275 nodes. Three TAZs were selected to represent the variability in transit service and land use throughout the study area. These TAZs’ locations are shown in Figure 1 as white dots and include:

Midtown—This commercial center is composed of dense mixed-use development and features several MARTA heavy rail stations that serve as terminus points for many of MARTA’s bus routes. There are several bike lanes and two cycletracks that enhance the bikeability in this area. The street network forms a grid in this area, which allows for shorter walking and biking distances to transit services.

Brookhaven—This northeast suburb features one MARTA rail station and four bus routes. There is a large park-and-ride lot next to the station, and many of the surrounding streets in this area are large multi-lane arterials with high vehicle speeds and volumes. There are single-family homes near the station, but most of the development directly surrounding the station is commercial. The street network is circuitous and disconnected, and the rail line and adjacent major arterial act as obstacles to street connectivity.

Campbellton—This location was selected because of its proximity to the Campbellton bus route, which has high ridership and frequent service. However, this location is the furthest away from rail, and this location is largely composed of low-density single-family homes and circuitous, disconnected streets which increase the required walking and biking distances to transit.

Figure 1.

Study area using a 2 mi buffer around all of Metropolitan Atlanta Rapid Transportation Authority (MARTA)’s bus and rail stops.

Routing Procedure

The bike and walk portions of trips were routed using Dijkstra’s algorithm, and the transit portion was routed using the RAPTOR algorithm implemented in the Transit Routing Python module ( 18 , 28 ). This module was selected, in part, because the open source algorithm developer offered to work directly with the research team on streamlining the integration. The general constraints for routing are listed below. These constraints were based on values suggested in the TCSQM, engineering judgment, and lived experience with both walk-transit and bike-transit trips.

Assumed an average travel speed for all street network links:

Walking: 2.5 mph

Biking: 8.0 mph

Assumed a 15 min transit access threshold:

Walking: 0.625 mi

Biking: 2.0 mi

Origin-destination pairs with a shortest path distance less than the transit access threshold distance were not considered for transit routing.

The maximum distance for walking or biking was the first access threshold + second access threshold:

Walk-transit-walk: 1.25 mi (30 min)

Bike-transit-bike: 4.0 mi (30 min)

Bike-transit-walk: 2.625 mi (30 min)

Transfer restrictions:

No more than one transfer is considered for this research.

Transfers are performed on foot at 2.5 mph and are limited to 440 ft (2 min).

The boarding time for transit vehicles is 30 s.

The optimal route between an origin and destination is the route with the least travel time given all the above constraints. The access thresholds were modified from the recommended TCQSM values to account for the lower assumed walking and cycling speeds. These lower speeds were chosen to reflect the hilly terrain of the study area and delay from signalized intersections. Trips requiring more than one transfer were restricted on the grounds that these trips would be behaviorally unlikely for bike-transit-bike. Each additional transfer requires the user to wait for an additional transit vehicle, which decreases the feasibility of that trip. In addition, cyclists may have similar average speeds to many bus routes, meaning cyclists would avoid unnecessary bus transfers unless a bus transfer significantly improved travel time. In the end, one transfer was allowed between any transit mode to see if there were certain areas where these trips were feasible.

Given these constraints, the routing procedure followed several steps and is visualized in Figure 2. First, all TAZs and transit stops were matched to the nearest OSM node via the shortest Euclidean (straight-line) distance for shortest path routing purposes. Then, Dijkstra’s algorithm was used to find the shortest network distance between all possible TAZ pairs to eliminate origin-destination pairs that were too close together according to set access thresholds (0.625 mi for walking, and 2.0 mi for biking). It would likely be faster to walk or bike between these TAZs than to incorporate transit. Next, from each TAZ, all transit stops within a 2.0 mi Euclidean (straight-line) distance were found. On average, a TAZ had around 22 stops within this buffer. If there were multiple stops servicing the same route within this buffer, then only the closest stop was considered a candidate.

Figure 2.

Routing procedure and process flow for analyzing bike/walk-transit trips.

Dijkstra’s algorithm was then used to find the shortest distance from each TAZ to each candidate transit stop. Then, another run of Dijkstra’s algorithm was used to find the shortest distance from each candidate transit stop to each TAZ. These steps were run separately to account for street directionality because cyclists are not allowed to travel the wrong way on one-way streets. TAZ to candidate transit stop pairs and candidate stop to TAZ pairs that were higher than the set access thresholds for the mode taken were removed from consideration. These remaining pairs were then exported to feed into the RAPTOR algorithm.

The required RAPTOR algorithm inputs are the starting station, arrival station, departure time, and maximum number of transfers. All possible starting and ending transit stop combinations for each TAZ pair were considered. Departure times between 8:00 a.m. and 10:00 a.m. in 15 min intervals on a weekday schedule (when MARTA offers its most frequent service) were considered, to capture the effect of the transit schedule and service frequencies on travel time. Arrival times at the first transit stations were calculated using the shortest path results from all TAZs to all candidate transit stops. The final arrival time from the last transit stop to the destination TAZ was calculated using the shortest path results from all candidate transit stops to all TAZs. All arrival times were rounded to the nearest minute.

Once a solution was calculated, the route taken, including the first-leg, transit-leg, and last-leg, were saved as Geopackage files for validation. Additionally, the total travel time, transit travel time, transit waiting time, number of transfers, and the transit modes utilized were recorded to generate figures visualizing accessible TAZs within 60 min, including the minimum number of transfers required, travel times to TAZs within 60 min, and the transit mode(s) and route(s) utilized.

Results

Using a desktop with a 10-core 3.70 GHz CPU, 32 GB RAM, and a 1.2 TB NVMe, the total run time for solving and processing the three TAZs locations was 1 h for walk-transit-walk, 5 h for bike-transit-walk, and 8 h for bike-transit-bike. The significantly longer computation time for bike-transit-walk and bike-transit-bike was expected because the higher travel speed and access threshold for cycling resulted in more trip combinations to examine. Similarly, origin TAZs that were near more transit stops and other TAZs took longer to route. In the next sections, example routes and three types of visualization (accessibility, travel time, and transit modes/routes utilized) illustrating the results of these analyses are presented.

Example Routes

Figure 3 shows an example bike-transit-bike and walk-transit-walk route starting at 8:15 a.m. from the Campbellton TAZ and ending at a TAZ adjacent to the east-west rail line and just west of the Five Points transfer station. The bike-transit-bike trip begins with a 6 min bike ride to a bus stop. Once the bus arrives, the bus then travels 17 min to a rail station along the north-south rail line. Once at the rail station, a northbound train is taken for 4 min to a stop just south of the Five Points transfer station. The rest of the journey is accomplished via a 10 min bike ride. The total travel time was 47 min, the total transit time was 29 min, and the total wait time was 9 min. Given the constraints specified, this appears to be a reasonable route. Depending on the departure time, the suggested route will change for bike-transit-bike. For an 8:45 a.m. departure time, an express bus along an uninterrupted parkway connects to a rail station further south on the north-south rail lines, and the train is taken to the same ending station as the 8:15 a.m. departure time.

Figure 3.

An example route of bike-transit-bike (top) and walk-transit-walk (bottom) from the Campbellton traffic analysis zone (TAZ) to a TAZ near the east-west rail line.

The walk-transit-walk trip begins with a 15 min walk to a bus stop. Once the bus arrives, the bus travels 21 min to a rail station at the western terminus of the east-west line. Once at the rail station, an eastbound train is taken for 7 min to a stop near the ending TAZ. The rest of the journey is accomplished via a 9 min walk. The total travel time was 59 min, the total transit time was 35 min, and the total wait time was 7 min. Had the number of transfers not been limited, a comparable route for walk-transit-walk could have been taking the bus line used in the bike-transit-bike route to the north-south rail station, transferring to the east-west line at the Five Points transfer station, and walking to the TAZ. However, given the constraints, the route seems reasonable. Out of all the departure times examined, this destination TAZ was only within 60 min for walk-transit-walk at 8:15 a.m., whereas this destination TAZ was accessible for all departure times via bike-transit-bike. This suggests that bike-transit-bike is a more consistent mode for reaching this TAZ within a certain time limit.

Midtown

Figure 4 shows a comparison of walk-transit-walk, bike-transit-walk, and bike-transit-bike accessibility within an average 60 min of total travel time and the minimum number of transfers required from the Midtown TAZ. TAZs that are accessible via walk-transit-walk, bike-transit-walk, and bike-transit-bike are displayed in green, blue, and red, respectively. TAZs that are accessible via walk-transit-walk are also accessible via bike-transit-walk and bike-transit-bike, and areas that are accessible via bike-transit-walk are also accessible via bike-transit-bike. This figure also includes TAZs that can be walked or biked to directly.

Figure 4.

The accessible traffic analysis zones (TAZs) and minimum number of transfers required from the Midtown TAZ.

Starting from the source TAZ (white) in Figure 4 and moving outwards, 705 TAZs are accessible via walk-transit-walk or walking from the source TAZ. The geographic extent of the walk-transit-walk accessible TAZs is large, but this is expected because MARTA’s network layout prioritizes transporting people to and from this business center. The Midtown TAZ is a short walking distance to both rail and bus routes that go in every direction. However, there is higher walk-transit-walk accessibility south and directly east and west of the Midtown TAZ than there is north of the Midtown TAZ. Bike-transit-walk and bike-transit-bike increased the number of accessible TAZs to 872 (24% increase) and 1,149 (63% increase), respectively. Most of the added accessible TAZs were toward the extremes of the study area where transit service coverage is sparse or a second transfer would be required for walk-transit-walk. The average wait time was 11 min for walk-transit-walk, 7 min for bike-transit-walk, and 6 min for bike-transit-bike.

TAZs that require a transfer to access are crosshatched and the orientation of the cross-hatching indicates if a transfer was required for walk-transit-walk or bike-transit-bike. If both cross-hatching patterns are present, then a transfer was required for both walk-transit-walk and bike-transit-bike. For visual clarity, the minimum number of transfers required for bike-transit-walk is not displayed. The minimum number of transfers was identified by finding the trip with the least number of transfers and a travel time of no more than 2 min greater than the average travel time to that TAZ. While the first transfer on MARTA is free, adding a transfer can add additional wait time, especially if the transfer route operates on a lower frequency or is not timed for transfers. Delays can also make timing transfers more uncertain.

For walk-transit-walk, the TAZs around the train stations along the north-south route and the areas directly west and east of the origin TAZ can be reached with no transfers. Bike-transit-walk and bike-transit-bike eliminated approximately 21% and 63% of these transfers, respectively, particularly along the north-south route. In these cases, low-frequency bus routes are being replaced with bike trips, which removes the burden on the traveler having to time a transfer. Although, in certain cases, bike-transit-bike had more transfers, as seen in the western portion of the study area just north of the east-west rail line. In these cases, two rail trips combined with biking at both ends improved travel time over the most direct option, a single bus trip.

In addition to reducing the number of trips requiring a transfer, bike-transit-walk and bike-transit-bike also reduced the average wait time from walk-transit-walk by 6 and 8 min, respectively. Reducing wait time is important, as wait time is often perceived as more costly than travel time. When considering TAZs accessible by bike-transit-walk and bike-transit-bike, bike-transit-bike reduces the wait time by 2 min, and the number of transfers by 45%. Since most transit routes would be a short bike ride from the Midtown TAZ, it is logical that having a bike for the last leg of a trip would make it possible to reach more TAZs, especially when these TAZs are more sparsely spread out than near the Midtown TAZ.

In Figure 4, the accessible TAZs do not form a contiguous surface. There are several isolated “accessibility islands” such as the walk-transit-walk accessible TAZs in the far east section of the study area. These TAZs were typically reached with long-distance bus routes that traversed restricted access roads such as interstates. There are fewer stops along these routes, which reduces travel time. This also occurs on the rail routes.

Figure 5 illustrates the average travel time (across all the departure times studied) to accessible TAZs for bike-transit-bike and walk-transit-walk. For visual clarity, bike-transit-walk is not displayed, as its travel times were comparable with bike-transit-bike. The source TAZ (white) is bounded by a black dashed border on both maps. The small white dots within this area are TAZ centroids within the transit access threshold (0.625 mi for walking and 2.5 mi for cycling) that were not considered for transit routing. Outside of this border, accessible TAZs are color-coded according to average travel time rounded to the nearest minute; non-accessible TAZs are shown as light grey dots.

Figure 5.

Average travel times from the Midtown traffic analysis zone (TAZ) to all other TAZs for bike-transit-bike (left) and walk-transit-walk (right).

Whereas Figure 4 displayed the disparity in accessibility, Figure 5 presents the disparity in travel time between the bike-transit modes and walk-transit-walk. For TAZs that were accessible by all three modes, bike-transit-walk reduced the average travel time from walk-transit-walk by 7 min, and bike-transit-bike reduced the average travel time from walk-transit-walk by 12 min. These results are not surprising, given the increased speed provided by the bike, but these results do highlight areas where there is potential for bike-transit-bike. When considering TAZs accessible by bike-transit-walk and bike-transit-bike, bike-transit-bike reduces the travel time by 5 min.

For all three TAZs locations studied, there were relatively few TAZs that were accessible within 15 min regardless of mode taken. This is because the transit option had to already be faster than a 15 min walk or bike ride (TAZs within 15 min were not considered for transit routing). Still, a few TAZs near rail stations were within 15 min for walk-transit-walk, and several more were within 15 min for bike-transit-bike in Figure 5. In general, travel time appears to increase with distance, but the travel time increases slower along the north-south rail routes than along bus routes.

Lastly, Figure 6 shows the combination of transit modes used to access a TAZ from the source TAZ and the transit routes utilized for bike-transit-bike and walk-transit-walk. Again, bike-transit-walk is not pictured. Once the fastest trip per departure time was calculated, the most frequent transit mode utilized (rail, bus, “bus and rail”) was found. While “bus and rail” indicates that there was a transit transfer, TAZs marked as bus or rail could have been reached with one or two transit trips. Transit mode can be important in the mode decision process because certain modes, such as rail service, often feature higher travel speeds and frequency than bus service.

Figure 6.

Fastest transit mode(s) to accessible traffic analysis zones (TAZs) from the Midtown TAZ for bike-transit-bike (left) and walk-transit-walk (right).

Starting again from the source TAZ, TAZs that were bikeable or walkable are green, TAZs that were reachable with one or two bus trips are dark yellow, TAZs that were accessible via one or two rail trips are blue, and TAZs that were accessible via a combination of bus and rail (in either order) are grey. The lines represent the utilized transit routes. Light blue is for rail and light tan is for bus. As depicted in Figure 6, the number of TAZs accessible via rail is limited for walk-transit-walk. Most other TAZs are reached by bus service (west and east of the source TAZ) or connecting by bus after a rail trip. With bike-transit-bike, the number of TAZs reachable via rail service increases substantially. In both cases, the full rail network was utilized, but bike-transit-bike utilized more bus routes.

Campbellton

The next location assessed was the Campbellton TAZ. Unlike the other two study locations, the Campbellton location is far from rail but has access to a high frequency bus route with 15 min headways. The number of accessible TAZs in Figure 7 for walk-transit-walk, bike-transit-walk, and bike-transit-bike were 277, 347 (25% increase), and 592 (113% increase), respectively. Despite the distance from rail, the proximity to the high-frequency Campbellton bus route resulted in relatively low average wait times. The average wait time was 10 min for walk-transit-walk, 9 min for bike-transit-walk, and 8 min for bike-transit-bike.

Figure 7.

The accessible traffic analysis zones (TAZs) and minimum number of transfers required from the Campbellton TAZ.

A transfer was required to access many of the TAZs in Figure 7. For TAZs accessible by all three modes, bike-transit-walk and bike-transit-bike only decreased the number of transfers by 3% and 13%, respectively, from walk-transit-walk. For TAZs accessible by bike-transit-walk and bike-transit-bike, bike-transit-bike reduced the number of transfers required from bike-transit-walk by 10%. Additionally, many of the TAZs along the eastern section of the east-west line are inaccessible to all three modes because only one transfer was permitted. Had bus-to-bus transfers been removed, TAZs west of the north-south rail line would not have been accessible because many MARTA bus routes terminate at rail stations rather than continuing through the rail stations.

Figure 8 shows the travel time to the accessible TAZs for the bike-transit-bike mode and bike-transit-walk mode for the Campbellton TAZ. Walk-transit-walk was not displayed because there were too few accessible TAZs. For TAZs that were accessible via all three modes, bike-transit-walk and bike-transit-bike only provided a 1 and 2 min reduction in travel time and a 2 and 3 min reduction in wait time, respectively. For TAZs that were accessible via bike-transit-bike and bike-transit-walk, bike-transit-bike reduced the travel time by 2 min and the wait time by 1 min. The impact of having a bicycle for the last leg is most evident for the TAZs near the Midtown location as there are more TAZs within the 16–30 min range for bike-transit-bike than bike-transit-walk.

Figure 8.

Average travel times from the Campbellton traffic analysis zone (TAZ) to all other TAZs for bike-transit-bike (left) and walk-transit-walk (right).

Figure 9 shows the fastest transit mode(s) and transit routes utilized from the Campbellton TAZ for both bike-transit-bike and bike-transit-walk. As with Figure 8, walk-transit-walk trips were not displayed, because of the similarity of those trips with bike-transit-walk trips. For both bike-transit-bike and bike-transit-walk, there were no TAZs that could be reached by only rail. This is expected, given the distance between the Campbellton TAZ and the nearest rail station. TAZs in the northeastern section were primarily accessed via “bus and rail”. Unlike the Midtown location, not all of the rail network was utilized. In fact, none of the east-west line was used for bike-transit-walk. The effect of MARTA’s network layout on accessibility can be seen here as well. Many MARTA bus routes serve as feeder routes to rail stations. As such, many of the transit routes utilized for both bike-transit-bike and bike-transit-walk head toward the rail lines.

Figure 9.

Fastest transit mode(s) to accessible traffic analysis zones (TAZs) from the Campbellton TAZ for bike-transit-bike (left) and bike-transit-walk (right).

Brookhaven

The final location assessed was the Brookhaven TAZ. In contrast with the two other TAZs studied, very few TAZs (15) are accessible from the Brookhaven station via walk-transit-walk, as shown in Figure 10. As such, comparisons with walk-transit-walk are not made for this TAZ. The low accessibility of walk-transit-walk is because of the long walking times to the nearest transit stops and the low frequency of the available service. The bike-transit-walk mode greatly increases the number of accessible TAZs to 571 (370% increase), and bike-transit-bike increases the number of accessible TAZs even further to 909 (600% increase). The Brookhaven location’s proximal access to a rail station elongates its accessible area for both the bike-transit-walk and bike-transit-bike modes. This indicates that just having a bicycle at the first leg of this TAZ can transform the effectiveness of public transit. Locations similar to the Brookhaven TAZ are likely more suitable for bike-transit-walk and bike-transit-bike trips than walk-transit-walk trips. The average wait time for walk-transit-walk, bike-transit-walk, and bike-transit-bike was 29, 8, and 7 min, respectively. The large wait time for walk-transit-walk demonstrates just how low-frequency the transit service was near the source TAZ. On the subject of transfers, TAZs that were not directly adjacent to the north-south rail line, excluding the northwestern spur, could typically be reached without needing to transfer. For TAZs accessible by bike-transit-walk and bike-transit-bike, bike-transit-bike removed 70% of the transfers required for bike-transit-walk.

Figure 10.

The accessible traffic analysis zones (TAZs) and minimum number of transfers required from the Brookhaven TAZ.

Figure 11 illustrates the travel time to the accessible TAZs for bike-transit-bike and bike-transit-walk for the Brookhaven TAZ. Since there are so few accessible TAZs via walk-transit-walk, bike-transit-walk is displayed instead. For TAZs accessible by walk-transit-walk, bike-transit-walk reduced travel times by 29 min, and bike-transit-bike reduced travel times by 31 min. For TAZs accessible by bike-transit-bike and bike-transit-walk, bike-transit-bike reduced travel time by 5 min. There were no travel times below 15 min for either mode, but TAZs that were accessible via both modes could, on average, be reached 5 min faster via bike-transit-bike. The decrease in travel time is especially evident for the TAZs near Midtown TAZ. Bike-transit-bike would also have two fewer minutes of wait time.

Figure 11.

Average travel times from the Brookhaven traffic analysis zone (TAZ) to all other TAZs for bike-transit-bike (left) and bike-transit-walk (right).

The Brookhaven location had a similar pattern in modes utilized to the Midtown location. Figure 12 shows that TAZs near rail stations were traveled to primarily via rail, and areas adjacent to the Brookhaven location but not along rail lines were reached via bus. Two modes were required for trips that were further away from rail service but not adjacent to the Brookhaven location. Additionally, access to a bike for the last leg of the trip greatly expanded the geographic reach of rail stations, as the bike-transit-walk side of Figure 12 shows only a few TAZs being accessed via rail along the rail lines.

Figure 12.

Fastest transit mode(s) to accessible traffic analysis zones (TAZs) from the Brookhaven TAZ for bike-transit-bike (left) and bike-transit-walk (right).

Similar to the Midtown scenario, the full rail network was utilized with both bike-transit-bike and bike-transit-walk. Unlike the Campbellton TAZ, rail was close enough to bike to, yet, unlike the Midtown TAZ, rail was too far to walk to. Bike-transit-bike used fewer bus routes than bike-transit-walk near the center of the rail network, but bike-transit-bike utilized more routes in the northeast section of the study area.

Discussion, Limitations, and Future Work

These analyses demonstrate the potential increase in transit accessibility from the bike-transit-bike mode over the walk-transit mode, even when the bike is only used for the first leg of the trip. In each location, bike-transit-walk and bike-transit-bike reduced total travel times, transit wait times, and the number of transfers needed, as shown in Table 1. However the extent of the reduction varied by the location and the set constraints.

Table 1.

Accessibility of Modes and Comparisons across Modes

Accessible by:	Total TAZs accessible (% of all TAZs) N = 2,221	Average wait time (min)	Reduction using:	Travel time reduction (min)	Wait time reduction (min)	Transfer reduction (%)
Midtown
Walk-transit-walk	705 (32%)	11	Bike-transit-walk	7	6	21%
Bike-transit-walk	872 (39%)	7	Bike-transit-bike	5	2	45%
Bike-transit-bike	1,149 (52%)	6
Campbellton
Walk-transit-walk	277 (12%)	10	Bike-transit-walk	1	2	3%
Bike-transit-walk	347 (16%)	9	Bike-transit-bike	2	1	10%
Bike-transit-bike	592 (27%)	8
Brookhaven*
Walk-transit-walk	15 (1%)	29
Bike-transit-walk	571 (26%)	8	Bike-transit-bike	5	2	70%
Bike-transit-bike	909 (41%)	7

Note: TAZ = traffic analysis zone.

There were too few accessible TAZs via walk-transit-walk from Brookhaven to draw meaningful comparisons.

Bike-transit’s greatest impact on improving transit accessibility was in areas that were within biking distance but not within walking distance of heavy-rail stations, such as the Brookhaven TAZ. Wait time reductions were consistent across the Midtown and Campbellton locations but were large for the Brookhaven location because of the few accessible TAZs via walk-transit-walk. Reductions in the number of transfers were highest for the Midtown location, which indicated that feeder bus routes were replaced with faster bike trips.

In these analyses, the impedance used for bike routing only considered travel time, leaving out several potential attributes that cyclists consider in choosing a route, such as elevation, the number of turns, the presence of cycling infrastructure, and other roadway characteristics known to influence cycling routing ( 29 , 30 ). Locations further away from the urban core, such as Brookhaven and Campbellton, lack cycling infrastructure and a connected network of slow residential streets. As such, there would likely be fewer accessible areas here than presented. Future studies should focus on incorporating cycling-specific impedances into assessments of bike-transit accessibility. Once these impedances are added, new cycling infrastructure projects can be evaluated based on how the infrastructure improves the accessibility of the bike-transit mode.

These cycling-specific impedances will need to be balanced against a set of impedances for public transit (which is also currently set as a time-only impedance). For example, even with two transit routes that provide an equivalent total transit link travel time, the route with the shorter wait time will likely be scored by users as having lower impedance (most users penalize wait time more than travel time in their mode choice decision-making) ( 4 ). Additionally, if the bike route to access the transit route with the shorter waiting time had a higher impedance than the other transit route, some users may endure the higher-impedance bike route to access the route with a shorter wait time. Cyclists may also have public transit-specific preferences that can also be accounted for in estimating impedance, such as a preference for rail service over bus service and/or avoiding transfers. The constraints used in this report were designed to model reasonable walk-transit and bike-transit trips, but constraints can cause threshold effects that influence results. For instance, individuals may be willing to bike or walk further than the set access thresholds, and higher access thresholds can change the number of accessible TAZs. Similarly, not allowing more than one transfer also affected the number of accessible TAZs, particularly for walk-transit-walk.

Because accurate pathways to and through rail stations were not available, the estimated travel times in this paper may not be representative of real-world conditions. Each transit stop was snapped to the nearest network node, but rail stations often have multiple entrances, including bridges that span across otherwise impassible features such as rail yards and interstates. MARTA does not currently provide a pathways file in their GTFS data. Incorporating these data in the future would greatly help in making modeled travel times more representative of actual travel times, especially for walk-transit where smaller detours can cause larger increases in travel time because of the slower travel speeds. In this paper, the transit schedule and routes were not modified. However, one future application of this research is in evaluating transit network re-designs. New transit networks or schedules can be fed into RAPTOR in GTFS format to see how bike-transit accessibility changes. Additionally, methods that incorporate the uncertain parts of transit routing (i.e., delays, leave time, cancellations) can increase the confidence in results.

Lastly, this paper focused on evaluating bike-transit accessibility using a one-to-all approach using TAZs in the interest of generating visuals that explained the difference in accessibility, travel time, and transit mode(s) utilized. In addition to these types of analyses, future work should examine origin-destination pairs from travel diaries to gain insight into the feasibility of bike-transit for real trips. In addition, point of interest data on schools, jobs, grocery stores, and hospitals, or census data on population or employment could be used in lieu of the current TAZ approach to generate the desired accessibility metrics.

Conclusion

In this paper, RAPTOR and Dijkstra’s algorithm were used to model walk-transit-walk, bike-transit-walk, and bike-transit-bike trips originating from three locations in Atlanta, GA. The results were compiled into several visualizations that compared accessibility, travel times, and transit mode(s) taken. In general, both bike-transit-walk and bike-transit-bike expanded accessibility, reduced travel time and wait time, and reduced the number of transfers required relative to walk-transit-walk. Travel times increased with distance from the origin TAZ but would increase less rapidly for destination TAZs near rail service because of the higher frequencies and travel speeds of rail. Similarly, access to rail greatly increased the geographic extent of accessible destination TAZs. These initial results suggest the service area of stops serving rail and high-frequency bus routes can be expanded with bike-transit-walk and bike-transit-bike provided that there are safe cycling routes to and from these stops.

These results suggest that bicycles as a first-last mile mode could provide a large expansion to transit accessibility with the current level of transit service. Thus, planners and engineers should consider bike-transit-walk and bike-transit-bike trips in their planning and travel demand modeling processes to maximize the utilization of their existing transit network. Planners and engineers can repeat the analyses presented in this research to identify where bike-transit would be most effective. Additionally, this methodology can be used to assess how public transit service changes and new cycling infrastructure can affect the accessibility of bike-transit trips. Other first-last mile modes, such as bike-share and scooter-share, can also be assessed.

Once planners and engineers have located the areas with the greatest potential for bike-transit, secure bicycle storage (both at stations and on board vehicles) and bicycle infrastructure can be introduced to encourage these trips. In many cases, constructing high-quality cycling infrastructure, such as cycletracks and multi-use paths, that feeds into the existing transit network will be less costly and more timely than transit-oriented development or expanded transit service ( 31 ). In addition, encouraging bike-transit trips could replace automobile trips and reduce transportation greenhouse gas emissions and vehicle miles traveled ( 32 ). Lastly, since bike-transit is a complicated mode because of the large number of potential transit stop and route combinations, transit operators should also work to provide mobile trip planners that offer step-by-step bike-transit routing. Widely used trip-planning apps do not include this option for most areas, which makes it difficult for travelers to plan these trips.

Footnotes

Acknowledgements

The research team would like to acknowledge Fizzy Fan for her contributions on the transit routing algorithm and the GTFS processing, and Prateek Agarwal at the Indian Institute of Science, Bangalore, India, for his input on implementing the RAPTOR algorithm. The authors would like to thank the NCST and the USDOT for their support of university-based research in transportation, and especially for the funding provided in support of this project.

Author Contributions

The authors confirm contributions to the paper as follows: study conception and design: R. Passmore, R. Guensler, K. Watkins; data collection: R. Passmore, R. Guensler; analysis and interpretation of results: R. Passmore, R. Guensler, K. Watkins; draft manuscript preparation: R. Passmore, R. Guensler, K. Watkins. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded, partially, by a grant (reference number GT-DOT-717) from the National Center for Sustainable Transportation (NCST), supported by the U.S. Department of Transportation (U.S. DOT) through the University Transportation Centers program.

ORCID iDs

Reid Passmore

Kari Watkins

Data Accessibility Statement

All of the data used in this project are publicly available through their respective sources. The road network data used in this project was obtained from OpenStreetMap. The bicycle facility inventory and traffic analysis zone were obtained from the Atlanta Regional Commission’s open data hub. The GTFS data were obtained from MARTA. The code used to perform the analyses in the paper can be found at .

References

Walker

Human Transit. Island Press/Center for Resource Economics, Washington, D.C., 2012.

Hall

Divide and Sprawl, Decline and Fall: A Comparative Critique of Euclidean Zoning. University of Pittsburgh Law Review, Vol. 68, No. 4, 2007, pp. 915–952.

Zuo

Wei

Chen

Zhang

First-and-Last Mile Solution Via Bicycling to Improving Transit Accessibility and Advancing Transportation Equity. Cities, Vol. 99, 2020, p. 102614. https://doi.org/10.1016/j.cities.2020.102614

Transportation Research Board. Transit Capacity and Quality of Service Manual, Third Edition. The National Academies Press, Washington, D.C., 2013. https://doi.org/doi:10.17226/24766

Zuehlke

Impossibility of Transit in Atlanta: GPS Enabled Revealed Drive Preferences and Modeled Transit Alternatives for Commute Atlanta Participants. Master’s thesis. Georgia Institute of Technology, Atlanta, GA, 2007. https://hdl-handle-net.web.bisu.edu.cn/1853/19788. Accessed March 4, 2024.

Watkins

K. E.

Accessibility Disparity Between Transit and Automobile: A Study of Atlanta and Seattle. Presented at 97th Annual Meeting of the Transportation Research Board, Washington, D.C., 2018.

Pucher

Buehler

Integrating Bicycling and Public Transport in North America. Journal of Public Transportation, Vol. 12, No. 3, 2009, pp. 79–104. https://doi.org/10.5038/2375-0901.12.3.5

Rose

Weliwitya

Tablet

Johnson

Subasinghe

Bicycle Access to Melbourne Metropolitian Rail Stations. Proc., 38th Australasian Transport Research Forum 2016, Melbourne, 2016.

Wang

Liu

Bicycle-Transit Integration in the United States, 2001–2009. Journal of Public Transportation, Vol. 16, No. 3, 2013, pp. 95–119. https://doi.org/10.5038/2375-0901.16.3.6

10.

Atlanta Regional Commission. 2019Regional Transit On-Board Survey. n.d. atlregional.github.io/ActivityViz/index.html. Accessed July 13, 2023.

11.

Jonkeren

Kager

Harms

te Brömmelstroet

The Bicycle-Train Travellers in the Netherlands: Personal Profiles and Travel Choices. Transportation, Vol. 48, No. 1, 2019, pp. 455–476. https://doi.org/10.1007/s11116-019-10061-3

12.

Martens

The Bicycle as a Feedering Mode: Experiences From Three European Countries. Transportation Research Part D: Transport and Environment, Vol. 9, No. 4, 2004, pp. 281–294. https://doi.org/10.1016/j.trd.2004.02.005

13.

Kosmidis

Müller-Eie

The Synergy of Bicycles and Public Transport: A Systematic Literature Review. Transport Reviews, Vol. 44, 2024, pp. 34–68. https://doi.org/10.1080/01441647.2023.2222911

14.

Bearn

C. B.

Measuring Low Stress Bike Access to MARTA. Master’s thesis. Georgia Institute of Technology, Atlanta, GA, 2015. https://hdl-handle-net.web.bisu.edu.cn/1853/54474. Accessed March 4, 2023.

15.

RSG and The Rand Corporation. Evaluation of Walk and Bicycle Demand Modeling Practice (NCHRP 08-36/Task 141). AASHTO, Washington, D.C., 2019. https://onlinepubs.trb.org/onlinepubs/nchrp/docs/NCHRP8-36CTask141FinalReportRSGRAND.pdf. Accessed March 4, 2024.

16.

Delling

Pajor

Werneck

R. F.

Round-Based Public Transit Routing. Transportation Science, Vol. 49, No. 3, 2015, pp. 591–604. https://doi.org/10.1287/trsc.2014.0534

17.

A Framework for Optimizing Public Transit Bus Fleet Conversion to Alternative Fuels. PhD thesis. Georgia Institute of Technology, Atlanta, GA, 2019. https://hdl-handle-net.web.bisu.edu.cn/1853/64018. Accessed March 4, 2024.

18.

Dijkstra

E. W.

A Note on Two Problems in Connexion With Graphs. Numerische Mathematik, Vol. 1, No. 1, 1959, pp. 269–271. https://doi.org/10.1007/bf01386390

19.

de Freitas

L. M.

Becker

Zimmermann

Axhausen

K. W.

Modelling Intermodal Travel in Switzerland: A Recursive Logit Approach. Transportation Research Part A: Policy and Practice, Vol. 119, 2019, pp. 200–213. https://doi.org/10.1016/j.tra.2018.11.009

20.

Ling

S. W. H.

An Atlanta-Based Analysis on the Feasibility of Employee Commute Options Programs and Switching From Driving Alone to Alternative Commute Modes. Master’s thesis. Georgia Institute of Technology, Atlanta, GA, 2021. https://hdl-handle-net.web.bisu.edu.cn/1853/67180. Accessed March 4, 2024.

21.

Conway

M. W.

Byrd

van der Linden

Evidence-Based Transit and Land Use Sketch Planning Using Interactive Accessibility Methods on Combined Schedule and Headway-Based Networks. Transportation Research Record: Journal of the Transportation Research Board, 2017. 2653: 45–53. https://doi.org/10.3141/2653-06

22.

Eldred

Modeling Bicycle Comfort With Conveyal Analysis (Part 1). 2020. medium.com/conveyal-blog/bike-lts-with-single-point-analysis-inconveyal-55eecff8c0c7. Accessed March 4, 2024.

23.

Conway

M. W.

Byrd

Eggermond

M. V.

Accounting for Uncertainty and Variation in Accessibility Metrics for Public Transport Sketch Planning. Journal of Transport and Land Use, 2018, Vol. 11, No. 1. https://doi.org/10.5198/jtlu.2018.1074

24.

OpenStreetMap Contributors. OpenStreetMap. OpenStreetMap Foundation, Cambridge, UK, 2024.

25.

Boeing

OSMnx: New Methods for Acquiring, Constructing, Analyzing, and Visualizing Complex Street Networks. Computers, Environment and Urban Systems, Vol. 65, 2017, pp. 126–139. https://doi.org/10.1016/j.compenvurbsys.2017.05.004

26.

OpenStreetMap Wiki Contributors. Overpass API. OpenStreetMap Wiki, n.d. wiki.openstreetmap.org/wiki/OverpassAPI. Accessed July 13, 2023.

27.

Atlanta Regional Commission. 2022 Regional Bikeway Inventory. n.d. opendata.atlantaregional.com/datasets/regional-bikeway-inventory-2022. Accessed July 13, 2023.

28.

Agarwal

Rambha

Transit Routing. n.d. github.com/transnetlab/transit-routing. Accessed July 13, 2023.

29.

Meister

Felder

Schmid

Axhausen

K. W.

Route Choice Modelling for Cyclists on Dense Urban Networks. Policy and Practice, Vol. 173, 2023. https://doi.org/10.1016/j.tra.2023.103723

30.

Fitch

D. T.

Handy

S. L.

Road Environments and Bicyclist Route Choice: The Cases of Davis and San Francisco, CA. Journal of Transport Geography, Vol. 85, 2020, p. 102705. https://doi.org/10.1016/j.jtrangeo.2020.102705

31.

NACTO. Urban Bikeway Design Guide. Island Press, New York, NY, 2014.

32.

European Environment Agency. The First and Last Mile: the Key to Sustainable Urban Transport: Transport and Environment Report 2019. Publications Office, 2020. https://data.europa.eu/doi/10.2800/200903