Pedestrian access to transit: Identifying redundancies and gaps using a variable service area analysis

Transcription

1 Pedestrian access to transit: Identifying redundancies and gaps using a variable service area analysis Ahmed M. El-Geneidy (corresponding author) Assistant Professor School of Urban Planning McGill University Suite 400, 815 Sherbrooke St. W. Montréal, Québec, H3A 2K6 Canada Tel.: Fax: ahmed.elgeneidy@mcgill.ca Paul R. Tétreault Transportation planning and traffic GENIVAR 1600 René-Lévesque Blvd. W., 16th floor Montréal, Québec, H3H 1P9 Canada Tel.: Fax: paul.tetreault@mail.mcgill.ca Julien Surprenant-Legault Master s student School of Urban Planning McGill University Suite 400, 815 Sherbrooke St. W. Montréal, Québec, H3A 2K6 Canada Tel.: Fax: julien.s.legault@mail.mcgill.ca Word Count: 5540 words + 4 Figures + 4 Tables Paper submitted for presentation at the transportation research board 89 th Annual meeting July

2 ABSTRACT Identifying the percentage of the population being served by a transit system in a metropolitan region is a key performance measure. This performance measure depends mainly on the definition of service area. Observing existing service areas can help in identifying existing gaps and redundancies in transit system. In the public transit industry, 400 meter (0.25 miles) buffers around bus stops and 800 meters (0.5 miles) around rail stations are commonly used to identify the area from which most transit users will access the system by foot. This research paper uses detailed origin-destination survey information to generate variable service areas that defines walking catchment areas around transit services in the Montréal region. The 400 and 800 meters service areas are greatly underestimating current coverage around transit stations. The 85 th percentile walking distance to bus transit service is around 550 meters from the origin and 660 meters to the destination. A statistical model is generated to estimate walking distances to transit stations. Walking distances vary based on household and personal characteristics, trip characteristics (especially the headway), type of transit (metro, commuter rail, and bus services), and route characteristics (stop spacing). Accordingly, service areas around transit stations should vary based on the type of service being offered. The generated service areas derived from the statistical model are then used to identify gaps and redundancies in the existing transit network in Montréal region. Finally, detailed analysis examining overlapping service areas along two specific routes shows the usefulness of variable service areas in identifying areas where potential stop spacing revisions can be possible without losing coverage. Keywords: Service area, Pedestrian access, Accessibility to transit, Transit planning, GIS, Redundancy in transit service, and Gaps in transit service 2

3 INTRODUCTION Identifying the percentage of the population served by a transit system in a metropolitan region is a key performance measure [1]. This performance measure depends mainly on the definition of service areas. A service area around a transit station or stop is defined as the area where potential riders are drawn from. Defining the service area around public transit stations is a complex and important issue, and can be used to define optimal stop spacing, identify redundancy and gaps in the system, and understand and predict demand for transit. Many transit planners and engineers depend on simplified methods when determining service areas around transit stations especially in regards to walking. A 400 meter buffer (0.25 miles) is defined around bus stops [2, 3] and an 800 meter buffer (0.5 miles) is used for rail stations [4, 5] as the areas from which most users accessing the system by foot originate. On the other hand, some researchers feel that this definition is not comprehensive enough and accordingly use a more conservative service area based on a 482 meter (0.3 mile) buffer [6]. These simplified service areas assume that all transit stations or stops are ubiquitous for a given mode, which is only true to a certain extent. This method of defining service areas imposes an error when trying to understand the demand for transit and/or when identifying gaps and redundancies in the existing transit service. Redundancy in the transit service provided and can clearly lead to poor and unreliable service. The definition of service areas should be related to the type of service being offered, its frequency [1] and its reliability. In this research paper, we offer a new method for understanding and defining service areas around transit stations for users accessing transit by foot in the Montréal metropolitan region as a case study. This is done through analyzing a detailed origin-destination survey conducted in 2003 [7] and combining it with service characteristics to generate service areas around existing transit stops. This is done with the goal of identifying areas with high levels of redundancy in transit service and identifying gaps where new or improved services are needed. The paper starts with a literature review of transit service area for bus and rail service followed by a description of the study region. The next section pertains to the methodology used to prepare and analyze the data for developing service areas. These sections are followed by a discussion of those results and a conclusion. LITERATURE REVIEW The most common standard measure of walking distance to transit stops and stations has been 400 meters (0.25 miles) [2, 3, 6, 8-11] since However, a large portion of research on this topic attempts to refine the analysis of access to transit facilities. According to Murray and Wu [10], access to transit service is an important factor in transit service planning. The more people residing and/or employed around transit stations, the greater the probability that the service to be used. This definition of the service area involves the use of distance decay to estimate walking distances to transit facilities [3, 6, 9, 12, 13]. Authors using distance decay express distances in terms of proportions of riders who will walk no more than a certain threshold. Zhao and her collaborators [3] noted that in southeast Florida, the number of riders walking over half a mile (800 m) was negligible. In Toronto, Canada, Alshalalfah and Shalaby [14] showed that among transit users, 60% live within 300 metres from their stop and 80% within 500 metres. In Calgary, Canada, Lam and Morrall [13] observed a median walking distance to bus stops of 292 metres, while the average was 327 metres and the 75 th percentile, 450 metres. Also in Calgary, O Sullivan and Morrall [12] distinguished between walking to light-rail transit stations in the suburbs and in the central business district. They found an average distance of 649 metres and a 3

4 75 th percentile equal to 840 metres in the former, while the average distance was 326 metres and the 75 th percentile, 419 metres in the latter. Studying walking distances to rail transit stations in Portland, WA, and San Francisco, CA, Schlossberg and his collaborators found a median distance of 0.47 mile (756 m) [4]. It is clear that a variation exists in the distance users are walking to transit between studies. These variations are according to variances in the study regions and in the sections of the regions where these data are collected. Accordingly, service areas around transit stations should vary according to the service being offered and the location in the region. The first element to consider when analyzing walking distance to transit stops is that pedestrians first and foremost seek to minimize both the distance and time of their the walking portion of their trip [4, 12]. After that, individual characteristics, station and area characteristics, bus route features, and temperature can have an effect on walking distances. According to Loutzenheiser, individual characteristics are the most important factors influencing on walking trips [15]. Population and dwelling density [3, 9, 15] and education [15] have positive effects on walking distances, while higher household incomes [5, 9, 15], a greater number of vehicles available [9, 14], and blue collar neighborhoods [15] have a negative effect. Pedestrian access to a transit service, which is the opportunity for using a system [16], is strongly associated with bus ridership [9] meaning that if a reliable transit system exists within a walking distance from a population the probability of this system to be used by the residents increase. Area characteristics favouring access are the absence of barriers [2, 3], a grid street pattern providing for more pedestrian linkages [3, 9, 15], higher densities, land use mix [3, 9, 15, 17], a small number of parking spaces at the station [15], safety [4, 12, 17], and an attractive and reliable transit service [4]. In terms of transit stops, the number of transit lines at a stop or station [3, 5] increase the willingness to walk, while waiting time [12, 13] and the number of transfers during a trip [14] decrease access walking distances. Finally, the effect of temperature is unclear because temperatures away from 18 C seem to discourage walking in the United States [5], while winter walking distances are slightly longer than summer ones in Calgary [13], a difference that the authors do not explain using temperature. Walking distances yield to buffers or service areas around transit facilities. There is a consensus in the transit literature that Euclidean buffers overestimate the service area of a stop and that network buffers are preferable [2, 3, 6, 9, 11]. This overestimation leads to several errors especially when estimating the demand for transit around stations or stops [11]. Thus, network buffers are better approximations of actual service areas, although they do not account for offstreet shortcuts. In addition, the size of service areas has a direct effect on bus stop spacing strategies. Most studies looking at bus stop spacing use 400 meter (0.25 miles) service areas around bus stops when revising stop spacing [18] or when removing redundancy imposed by poor spacing [10]. Bus stop spacing has a direct effect on the running time and the reliability of service [19]. Service areas are used in several studies to help understand the existing demand and determining the proportion of the population using the service at the station or stop. The transit industry widely applied 400 meter (0.25 miles) and 800 meter (0.5 miles) rules of thumb when estimating service areas around bus and rail stations. The application of these rules of thumb can lead to several measurement errors that need to be addressed and highlighted. Previous research has concentrated on the errors generated based on using Euclidean distance, 4

5 yet to our knowledge there has not been any research looking at the effect of using these rules of thumb at a regional level. Accordingly, more research is needed in order to understand and properly define service areas around transit stations and stops to address redundancy in the system and generate better stop spacing strategies, which are directly related to the quality of service being offered. CASE STUDY Montréal, Québec is the second most populous metropolitan region in Canada with 3.7 million residents. The Agence metropolitaine de transport (AMT) is an agency dependant from the Québec Ministry of Transport that is responsible for regional transit in Montréal. In this study, the region served by the AMT will be used as the study region. The AMT operates 5 commuter rail lines, 16 intermodal terminals, 60 park-and-ride facilities, 2 express bus routes, and 85 kilometers of bus, taxi, and/or high-occupancy vehicle lanes. In addition, the AMT plans future transit and collaborates with the 14 local transit agencies in the Montréal region, the largest ones being the Société de transport de Montréal (STM), the Réseau de transport de Longueuil (RTL), and the Société de transport de Laval (STL). In figure 1, a map of the Montréal metropolitan region and existing major transit corridors is shown. Figure 1: Transit services in the Montréal metropolitan region According to the 2003 Montréal origin-destination (OD) survey [7], 69.3% of trips are done by car, 13.7% by public transit, 10.2% by foot, 4.8% by school bus, and 1.1% by bicycle during a 24 hour period. In terms of trip purpose in the Montréal Metropolitan Region, 18.3% are work trips, 10.2% are school trips, 7.6% are shopping trips, 5.0% are leisure trips, and 44.6% are 5

6 back-to-home trips. The proportion of those trips made by public transit is 15.4% for work trips, 21.7% for school trips, 7.6% for shopping trips, 9.2% for leisure trips, and 14.8% for back-tohome trips. METHODOLOGY AND DATA PREPARATION The objective of this paper is to generate variable service areas based on existing service and neighborhood characteristics to help in understanding redundancies and gaps in the existing transit system. This method of generating service area will be compared to traditional methods and rules of thumb.. The first step towards generating accurate service area is to understand and document how far people are walking to use transit in the studied region. Service areas can be modeled around stations or stops using walking distance information. This is only done by using detailed travel behavior data. In this case we use detailed transit usage data obtained from the 2003 Montréal OD survey [7]. The OD survey is conducted every five years in the Montréal region. The survey records disaggregate trips that were made by each person residing in a household. Five percent of households in the Montréal Region were surveyed. The OD survey includes questions asking each transit user the routes used to reach his/her desired destination and if other transportation modes were employed. Any trip that involved the use of another mode (i.e. car, cycling, taxi, etc.) is excluded from this analysis. Trips using night bus service were also excluded as well as dedicated high school services. Since the OD survey does not record the actual transit station or stop used but only the routes, walking distances to the closest stop served by the first route used were measured. These calculations were done in a geographic information system environment using network analyst in ESRI sarcgis 9.3. Also walking distance from destination to the last transit route used is calculated using the same method. Measuring walking distances to and from transit stations or stops is the first step in preparing the data for the first statistical model. The first model looks at walking distances to access transit using the individual as the unit of analysis. This model is generated to assess the reliability of the data in hand and compare factors affecting walking to transit in the Montréal region to other regions. Several factors need to be controlled for in this model. For example, Controlling for competing routes is an important step in the process of studying the demand for transit [6], therefore controlling for competing transit service is important in a walking distance model. A route is considered as a competing route only if it is accessible within a certain network distance threshold measured from both the origin and the destination of the transit trip. This threshold is defined as the value representing the 75 th percentile of all the walking trips to transit (510.9 meters). These variables include individual, household, trip and route characteristics. These variables are included in the individual model to understand how far people are walking to access transit services. Since we do not know the detailed direction of travel, we summarized the information for both directions on the transit route. The shortest headway of the two directions is assigned to every walking trip. Headway is defined according to the starting time of the trip at the origin. Since some users start walking to transit before service begins in a few cases, we assigned the maximal headway on the route for these observations. Bus stop spacing was calculated using a linear referencing system in ArcGIS 9.3. This was done after snapping the stops to the nearest transit line. Stops were snapped to the transit network and spacing was calculated for both directions using a linear referencing technique. Table 1 lists the variables for the individual and stop models. 6

7 Table 1: Variable definitions Variable Name Description Walking distance Walking distance measured from the trip origin to the nearest transit station or stop along the used transit route Number of vehicles Number of vehicles owned in the household Sex Gender of the individual Age Age of the individual Worker A dummy variable that equals to one if the individual is a worker Student A dummy variable that equals to one if the individual is a student Income above 80K A dummy variable that equals to one if the household income more than 80K per year Headway The headway of the transit route used at the starting time of the trip Headway^2 The headway squared AM peak A dummy variable that equals to one if the trip is started between 6:30 am and 9:30 am PM peak A dummy variable that equals to one if the trip is started between 3:30 pm and 6:30 pm Mid-Day A dummy variable that equals to one if the trip is started between 9:30 am and 3:30 pm Work trip A dummy variable that equals to one for work trips School trip A dummy variable that equals to one for school trips Return home trip A dummy variable that equals to one for trip returning home Number of transfers The number of transfers during the trip Trip distance The total trip distance (in vehicle distance) Walking destination Walking distance measured from the trip destination to the nearest transit station or stop along the transit route last used Metro A dummy variable that equals to one if the Metro is first mode of transport in the trip Train A dummy variable that equals to one if the suburban train is first mode of transport in the trip RTL bus A dummy variable that equals to one if the RTL bus is first mode of transport in the trip STL bus A dummy variable that equals to one if the STL bus is first mode of transport in the trip CIT/CRT bus A dummy variable that equals to one if the CIT/CRT bus (other transit agencies in the region) is first mode of transport in the trip Competing routes The number of competing routes at the origin Stop spacing Average stop spacing between the previous and last stop along the route used Median income Median household income of the trip origin census tract Percentage of transit users to work Percentage of residents usually using transit to go to work in the census tract of the origin Number of intersections Number of street intersections around trip origin within 510 meters 7

8 The following is the suggested specifications for the individual statistical model: (1) Walking distance by individual= f(household and personal characteristics(number of vehicles, Sex, Age, Worker, Student, and Income), Trip characteristics (Headway, Headway^2, AM peak, PM peak, Mid-Day, Work trip, School trip, Return home trip, Number of transfers, Trip distance, and Walking destination), Type of transit (Metro, Train, RTL bus, STL bus, and CIT/CRT bus), and Route characteristics (Competing routes and Stop spacing)) In this model, the walking distance to the métro and commuter rail services are expected to increase relative to STM bus service, with the decrease in the headway, during the am peak relative to early am and late night, and school and work trips relative to other trips. Trips going back home, the number of competing routes, the number of transfers, and the walking distance at the destination are expected to decrease the walking distance from the origin to the nearest transit station along the used routes in a trip. The second step is to generate a more general model that can be used in generating variable service areas for each station or stop in the Montréal region. The findings from this general model will then be used in a comparative analysis, comparing variable service areas with traditional rules in term of identifying redundancies and gaps in the existing services. The following is the suggested model specifications: (2) Walking distance to a stop= f(trip characteristics (Headway, Headway^2, AM peak), Type of transit (Metro, Train, RTL bus, STL bus, and CIT/CRT bus), Route characteristics (Competing routes and Stop spacing), Neighborhood characteristics(median household income, percentage of transit users, Number of street intersections)) Similar to the individual model, walking distance to transit service is expected to increase for accessing metro and commuter rail services relative to STM Bus, with decreases in headway, during the AM peak relative to other times of the day. It is also expected to decrease with stop spacing. DATA A total of 37,411 transit trips are included in the analysis. The median walking distance to a transit station is 270 meters while the 75 th percentile is 510 meters and the 85 th percentile is 666 meters. Separating walking distances by type of service can give a clearer picture in regards to the level of error being imposed by the rules of thumb currently in use. Table 2 shows the summary statistics of walking distances to transit stations (origin) and walking distances from the stations to reach a destination. The 85 th percentile of walking distances to bus service is above 600 meters for all transit operators;for commuter rail, the 85 th percentile is over 1,200 meters and over 749 meters for the metro (subway). 8

9 Table 2: Summary statistics of walking access distances to and from a transit stations or stops in the Montréal region All Transit Service Metro (underground) AMT Train All Bus Service Origin Destination Origin Destination Origin Destination Origin Destination Mean Median th Percentile th Percentile STD STM Bus Service RTL Bus Service STL Bus service Other Bus Service Origin Destination Origin Destination Origin Destination Origin Destination Mean Median th Percentile th Percentile STD It is important to note that the demand around transit stations or stops is not equally distributed and a distance decay affected is observed. Previous research used distance decay curves as a means of understanding service areas [3, 9, 20, 21]. Figure 2 shows a distance decay curve representing walking distances to the metro and STM bus service. Figure 2: Distance decay to the metro and STM bus service The first curve appears to fit the data quite well for the STM bus service (R 2 =0.95). While for the Metro the curve fitting is less accurate (R 2 =0.81) due to the few origins in the near vicinity of 9

10 metro stations. Real estate located next to metro stations e is generally more expensive due to the premium offered by transit accessibility. These impedances reflect the rather limited speeds attainable by pedestrian travel. However, an interesting result is that a surprising number of trips are made at distances up to and even exceeding 1 km (0.6 mile). This result is consistent across trip purposes, suggesting that individuals may be willing to walk considerably greater distances than the 400 meters (quarter mile) threshold, which is considered standard in transit planning. ANALYSIS A multivariate linear regression model for individual walking distances is derived using individual, route and trip characteristics. It is important to note that we tested several model specifications and we found that the one below best explains the desired relationships and follows transit theory. The regression output is reported in Table 3. Statistically significant variables are highlighted in bold. Table 3: Individual walking distance mode Variable description Coefficients t-statistics Significance Number of vehicles Sex Age Worker Student Income above 80K Headway Headway^ AM peak PM peak Mid-Day Work trip School trip Return home trip Number of transfers Trip distance Walking destination Metro Train RTL bus STL bus CIT/CRT bus Competing routes Stop spacing Number of street intersections (Constant) Dependent variable: Walking distance to transit at origin (meters) R-square =

11 As it is clear from the above model walking distances to transit increase with the number of vehicles owned by the household in which the observed individual is related. High levels of car ownership exist in the suburbs of the Montréal region where walking distances to transit are generally higher due to the lower density of transit service. Similarly, this explains the reason we are observing a statistically significant positive effect on walking distances if the individual s annual household income is over $80,000 relative to lower incomes. Males tend to walk 18 meters longer than females, while walking distances to transit decrease by 1 meter for each 1 year increase in age. The most interesting variables in this model are the headway and headway squared. Both show a statistically significant effect on walking distances, yet with different signs. The negative sign assigned to the headway means that walking distances will decrease by 0.61 meters for every 1 minute increase in the transit headways up to a certain threshold. Accordingly, routes with shorter headways (high frequency routes) attract people from further distances compared to routes with longer headways. The square term shows that this phenomenon is true until the headway reaches 30 minutes; when headways exceed 30 minutes, riders are walking further to use this transit service. Such long headways only exist at the fringes of the Montréal region and poorly served areas. It is expected that transit users in these parts of the region are mainly captive and are walking further distances as well as coping with less frequent service. During the AM peak riders walk 16 meters longer relative to the early morning and evening service. Regarding trip characteristics, a work trip adds 40 meters to the individual walking distance when compared to other trip purposes. Meanwhile walking distances for returning home trips are shorter by 40 meters compared to other trip purposes if all other values are kept to their mean. Walking distances tend to decrease by 47.6 meters for every transfer during a trip. On the other hand, for every kilometer traveled along the transit network riders are walking an additional 4.6 meters. These two variables indicate that direct routes are more desirable to riders even if travel distances are longer. It is important to note that for longer trips, people generally prefer to use private vehicles, and most riders undertaking long transit trips are probably captive. Walking distances measured at the destination show statistical significance at the 90 percent confidence level. Although the magnitude of the coefficient is small (0.01 meters), yet interpreting this variable is important. This negative coefficient indicates that riders will tend to walk less between their origin and the bus stop they will be using if the amount of walking between the last stop and their destination is longer. This is a key policy variable for transit agencies when trying to offer or modify services. New routes should be designed taking into account the minimization of walking distances at the origin and destination. Also depending on the phenomenon that service at the home end reaching other routes by transfer is debatable since every transfer will tend to decrease the walking distance at the origin, which is reflected as smaller service areas around the first stop. Service requiring many transfers from the origin corresponds to smaller service areas and accordingly more gaps in the system. It is clear from the model that the type of transit service being offered has an effect on walking distances to use public transit. The underground Metro riders do walk 181 meters more compared to the STM bus users. In addition, users of commuter rail walk an average additional 436 meters compared to STM bus riders. Meanwhile, looking at bus transit users that are using 11

12 other transit networks than the STM s, we find some two interesting observations. Riders using STL and RTL travel 36 meters and 40 meters more than STM respectively. These two agencies serve the first ring of suburbs around off the island of Montréal and lower transit route density than the STM. In addition, CIT/CRT bus transit users walk 148 meters more compared to STM bus users. CIT/CRT transit services mostly operate in the fringes of the Montréal region. These are areas that are mainly dominated by car usage and accordingly users are generally captive. Stop spacing has a negative effect on walking distances. This finding emphasizes the relationship between service area and stop spacing. Finally, a higher number of street intersections in the vicinity of the origin lead to longer average walking distances. This finding emphasizes the effect of the neighborhood characteristics, especially street connectivity, in encouraging walking. The first model follows transit research theory and the expected methodology. Accordingly, a generalized model is generated. This model can be then used to generate variable service areas for each station or stop in the entire Montréal region and to compare it to the traditional methods of generating service areas. The generalized model is presented in Table 4. Table 4: Generalized statistical model Coefficients t Sig. (Constant) Headway AM Metro Commuter Train RTL Bus STL Bus CIT/CRT Bus Competing routes Stop spacing Median income Percentage of transit users to work Number of street intersections Dependent variable: Walking distance to transit at origin R-square All personal characteristic variables are removed from this model. In addition, a couple of variables are added to control for the effects of neighbourhood characteristics. In the generalized model headway is still a statistically significant variable but only at the 90% confidence level. Since it was clear that walking distances vary according to the time of day, 12

13 service areas are also expected to vary. We eliminated all time of day variables except the AM peak variable since we will be using AM peak service to generate service areas as an example in this paper. Service areas can also be generated for other times of the day. Using the specifications obtained from the generalized model, we generated a mean walking distance for every transit stop in the region during the morning peak period. A total of 16,113 transit stops were used. The number of stops excludes the directional effect of the service to avoid double counting. Double counting occurs when two transit stations serving the same route are present across the street from each other, yet each one of them is serving a different direction. Since the headway had a statistically significant negative effect on walking distance, we used the transit route with the shortest headway for generating service areas. Accordingly, this mean walking distance can be used in generating variable service area around each stop. Since service areas are defined as the area including most of potential riders around a transit station, then the mean walking distance to a station or stop needs to be adjusted. Using the ratio between the mean walking distances and the 75 th and the 85 th percentiles for every type of service, STM, RTL, STL, CIT/CRT, Metro, and Commuter rail (see Table 2), a set of distances is generated for every stop used in the generation of variable length service areas. The first step in understanding the existing coverage is to generate different service areas suing Euclidean distance buffers. It is important to note that using Euclidean distance buffers in defining service areas is not an accurate way of representing service areas. Euclidean distance buffers are used in this step to help understand the differences between service areas generated from the mean value obtained directly from the statistical model, modified service areas at the 75 th percentile and 85 th percentile, and service areas generated based on transit theory ( meters service for bus and commuter rail services, respectively). Using the mean value derived from the statistical model, a circular buffer around each transit station is generated. The total area covered by all the buffers is equal 1,174 square kilometers. On the other hand, the area covered by the buffers generated from using the 75 th percentile and the 85 th percentile variable buffers are 1,490 and 1,778 square kilometers respectively. Finally, the area covered by the buffers derived from theory is equal to 1,142 square kilometers. It is important to note that overlapping service areas are measured once in this calculation and no double counting is included. It is clear that the area covered derived by the buffers derived from theory is smaller than the area derived from mean values. Accordingly, it is clear that using 400 and 800 meters service areas around stations underestimates service coverage by approximately 55% when compared to the 85 th percentile estimates. Since measuring distances along road networks is more accurate in generating service areas compared to Euclidean distances, we generated service areas around each transit station or stop using the 85 th percentile estimate measured along the road network. Since we are modeling walking distances around transit stations, freeways were excluded from the network beforehand. Figure 3 shows the overlapping service areas that are generated from the 85 th percentile estimate for the entire region. The figure can serve two purposes. The first is identifying existing gaps in the Montréal region s transit system. Identifying gaps is the first step towards identifying areas where new services or modifications to existing services are needed. The existing gaps in the service are represented as a white area in the figure below. After identifying gaps in the system, transportation engineers and planners can overlay the results with land use information to 13

14 identify if there is a demand for improved services within these gaps. They can also work on modifying the existing service through shortening headways or stop spacing to increase service areas. The second purpose of this map is identifying areas where redundancy in the system exists. The shades from yellow to red are mainly areas with high levels of redundancy in the services being offered. Figure 3: Overlapping service areas using the 85 th percentile estimates network buffers The map above is derived from intersecting 100 by 100 meter grid cells with the network distance service areas. Accordingly, the number of stops displayed represents the count of service areas intersecting with this grid cell. If a bus stop is serving two different transit lines then two variable service areas are derived for this stop based on the route and neighbourhood characteristics. The number of stops in the figure does not represent the number of physical stops since a stop is created for each route operating during the AM peak where multiple routes serve the same stop. Areas with high levels of redundancy need to be explored further to identify whether the redundancy is justified or not. For example in the downtown core, 275 scheduled stops are in service during the AM peak period. The downtown area has the highest number of bus, metro, and train routes in the entire region. Similarly, areas around major transit centers are expected to have high levels of redundancy. Yet, more analysis at the route level is required to understand the reasons for redundancies in other areas. The above figure only generates a general picture of the situation. 14

15 Figure 4 shows the redundancy in the service being offered by two STM bus routes (Bélanger 95 and Beaubien-18). In part A of the figure, we intersected the variable service areas estimated using the 85 th percentile and measured along the network for each stop serving route Beaubien-18 with 30 by 30 meter grid cells. This figure shows the overlapping service area generated from one single route. This method can help in identifying the redundancies imposed by poor stop spacing along a single route. It can be used to evaluate stop spacing as well along a single route. Since transit service does not exist in a vacuum then studying service area requires obtaining information from competing routes as well running in parallel to an existing route. In part B of the figure we intersected the variable service areas generated from the 85 th percentile estimates and measured along the street network for every stop serving both bus routes, Belanger 95 and Beaubien-18, with another set of 30 by 30 meter grid cells. The figure shows the number of service areas generated from scheduled stops intersecting with each grid cell. Figure 4: Bus route sample for overlapping service areas using the 85 th percentile estimates network buffers Looking at part A from the figure above, it is clear that redundancies exist in the middle and the eastern sections of the route where the bus stop spacing is not as consistent. Another important observation is that the area being served by one scheduled stop is equal to 20% of the 15

16 total service area around the transit line, while the area served by two stations represents 15% of the total service area around the entire transit line. This indicates that around 65% of the area served by this route is covered by at least 3 stops or more. Having an overlap in the service area along the same route is acceptable to a certain level. Yet, when the number of overlapping service areas reaches five or six service areas and they represent a big proportion of the service area around a bus route (30% for example) then revisions of stop spacing and route characteristics is a must. Meanwhile, part B of the figure shows the level of redundancy in the service offered by two competing routes. Around 49% of the service area around both lines is being served by at least five scheduled stops from either one of the two studied routes during the AM peak period. Accordingly, the level of service coverage being offered along parallel east-west corridors at this particular location is very high. It is important to note that such method need to be developed carefully to ensure the routes under investigations are competing and not complementing routes. The transit agency has some room to implement bus stop consolidation along several sections of these two routes. It is expected that the additional access time will be offset if not surpassed by the savings in running and waiting time. Savings in waiting and running time translate to saving in operating costs. CONCLUSION This research paper uses detailed origin-destination survey information to generate variable service areas for the Montréal region. It is clear from the summary statistics that service areas generated using rules of thumb greatly underestimate the effective service areas around transit stations. The 85 th percentile walking distance to bus transit service is around 550 meters measured from the origin of the trip and 660 meters measured to the destination of the trip. Meanwhile, for commuter rail walking distances are longer since the 85 th percentile is 1,219 meters and 1,095 meters measured from/to the origin and the destination respectively. The research also highlighted differences between various bus transit operators. It is clear from this research that transit users tend to walk longer distances to use suburban service. The statistical models show that walking distances to transit stations vary based on household, personal and trip characteristics (especially the headway), type of transit (metro, commuter rail, and buses), and route characteristics. Accordingly, service areas around transit stations should vary based on the type of service being offered. The generated service areas derived from the statistical model are used to identify gaps and redundancies in the existing transit network. These gaps and redundancies need to be analyzed carefully and in detail at the route level. Finally, the detailed analysis examining overlapping service areas along two specific routes shows the usefulness of this variable service area in identifying areas where potential stop spacing revisions can be effective without causing much harm to existing riders. It is important to note that revising stop spacing based on such methodology need to be developed for competing routes and not complementing ones. This finding opens a new area of research where stop spacing should be investigated as a variable value depending on the frequency and type service being offered and not just as a service standard given number. It also opens venues for research in the area of transit oriented development and how to identifying the exact service area around transit stations. 16

17 This study concentrated on service areas around transit stations and stops based on measured network walking distances. More research is recommended for deriving service areas around transit stations when other modes of transportation are involved. The generated service areas can be used in operation research studies involving bus stop consolidation. Combining the findings from this research with passenger movement at transit stations can help in generating better estimates of transit demand. In this research paper, we used scheduled rather than actual headways. Actual headways can be generated from archived automatic vehicle location (AVL) data. In addition, using on-time performance measures obtained for the AVL data can be an indication of the reliability of service, another measure of the service attractiveness that could be used to derive service areas more accurately and increase the fitness of the model. Also the generated service areas can be linked back to automatic passenger counter (APC) data, if they were present, to enable a better understanding of the transit demand and the best representation of a service area. Finally, having information related to passenger activities at each transit stop/station could improve the modeling process, through testing several variable service areas beside the 85 th percentile estimate used in here. ACKNOWLEDGMENTS This research funded by the Agence Metropolitaine de Transport (AMT). We would like to thank Ludwig Desjardins of the AMT and his research team for their support and feedback throughout the project. Also would like thank Mr. Daniel Bergeron and Alfred ka Kee Chu of the AMT for providing the detailed Montréal OD survey used in the analysis as well as the transit network. Last and not least, we would like to thank Prof. Raphael Fischler for his feedback at various stages of this project. All opinions in this paper are the responsibility of the authors. REFERENCES 1. Fielding, G., R. Glauthier, and C. Lave, Performance indicators for transit management. Transportation, : p O'Neill, W., D. Ramsey, and J. Chou, Analysis of transit service areas using geographic information systems. Transportation Research Record, : p Zhao, F., et al., Forecasting transit walk accessibility: Regression model alternative to buffer. Transportation Research Record, : p Schlossberg, M., et al., How far, by which route, and why? A spatial analysis of pedestrian preference, in MTI Report , Mineta Transportation Institute & College of Business, San José State University: San José, CA. 5. Kuby, M., A. Barranda, and C. Upchurch., Factors influencing light rail station boardings in the United States. Transportation Research Part A, : p Kimpel, T., K. Dueker, and A. El-Geneidy, Using GIS to measure the effect of overlapping service areas on passenger boardings at bus stops. Urban and Regional Information Systems Association Journal, (1): p Agence metropolitaine de transport, Enquête origine-destination : Montréal, QC. 8. Neilson, G. and W. Fowler, Relation between transit ridership and walking distances in a low-density Florida retirement area. Highway Research Record, 1972(403): p

18 9. Hsiao, S., et al., Use of geographic information system for analysis of transit pedestrian access. Transportation Research Record, : p Murray, A. and X. Wu, Accessibility tradeoffs in public transit planning. Journal of Geographical Systems, (1): p Gutiérrez, J. and J.C. García-Palomares, Distance-measure impacts on the calculation of transport service areas using GIS. Environment and Planning B: Planning and Design, : p O'Sullivan, S. and J. Morrall, Walking distance to and from light-rail transit stations. Transportation Research Record, : p Lam, W. and J. Morrall, Bus passenger walking distances and waiting times: A summerwinter comparison. Transportation Quarterly, (3): p Alshalalfah, B. and A. Shalaby, Case study: Relationship of walk access distance to transit with service, travel, and personal characteristics. Journal of Urban Planning and Development, (2): p Loutzenheiser, D.R., Pedestrian access to transit: Modeling of walk trips and theor design and urban form determination around bay area rapid transit stations. Transportation Research Record, : p Murray, A., et al., Public transportation access. Transportation Research Part D, (5): p Fitzpatrick, K., D. Perkinson, and K. Hall, Findings from a survey on bus stop design. Journal of Public Transportation, (3): p Furth, P. and A. Rahbee, Optimal bus stop spacing through dynamic programming and geographic modeling. Transportation Research Record, : p El-Geneidy, A., et al., The effects of bus stop consolidation on passenger activity and transit operations. Transportation Research Record, 2006(1971): p Levinson, H. and O. Brown-West, Estimating bus ridership. Transportation Research Record, : p Upchurch, C., et al., Using GIS to generate mutually exclusive service areas linking travel on and off a network. Journal of Transport Geography, : p