Public transportation can support sustainability initiatives by reducing the frequency of use of private cars and associated environmental impacts. As public transit transports people collectively it is found to produce 45% less CO₂, 95% less CO, and 48% less NO₂ than private vehicles [9]. A study after a rail system opening in Taiwan indicates that CO was reduced by 5% - 15% and another environmental assessment after rail service expansion in Germany indicates the reduction of pollutants such as NO, NO₂ and CO [10]. It is estimated that if 5% of Americans used public transit instead of private car or if every American used public transit for 5% of their trips during 1970 to 1998, the CO pollution reduction would be more than all the CO emitted from all metal processing plants and chemical manufacturing section combined [11].

According to a report published by American Public Transportation Association (APTA) in 2009, for one billion dollars of annual investment in public transportation, there would be more than $1.7 billion dollars of added annual GDP [12]. Depending on some factors such as mileage reduction, declining vehicle ownership, etc., a shift from automobile to transit provides a variety of cost savings [13] including fuel and oil, insurance costs, parking costs, etc.

Available public transit services can be especially beneficial for people with low income who cannot afford automobile ownership and for elderly and disabled persons by offering convenient and affordable service. Thus, it increases social and economic opportunities for physically, socially and economically disadvantaged people along with achieving equity objectives [1].

According to the US Center for Disease Control (CDC) and prevention, at least 30 minutes of daily physical activity such as bicycling, or walking is necessary to stay healthy [14]. An Atlanta, Georgia survey results show that almost two-thirds of the recommended daily physical activity is achieved by the transit users which is ten times greater than the average walking reported by the non-transit users [15]. It is also worth noting that medical expenses are 32% lower ($1019 per year) for adults who achieve the recommended physical activity than those who do not ($1349 per year) [16]. According to another study, 21 minutes of walking can help to burn 65.1 to 98.7 calories and 100 kilocalories burn per day might save $12,500 dollars per person in obesity-associated medical costs [17]. These findings clearly show the value of transit in the wellbeing of transportation users, an issue that is often overlooked by decision makers when they appropriate funding for transportation services and projects.

The results from a seemingly unrelated regression equations (SURE) model showed that 50 percent increase of city bus routes in highly populated areas of Taiwan reduce car usage by 1.4% which corresponds to 300,000 vehicles [18]. Another study on Bay Area Rapid Transit (BART) system indicates that in absence of BART services during the morning peak, driving times increased more than four times in multiple corridors [19]. Similar studies [20] [21] on investigating the effects of transit absence indicated increased traffic volumes, and longer delays. On the contrary, a study on Salt Lake City’s University TRAX light-rail system in 2014 found that typical vehicle traffic has reduced by nearly 50% with the expansion of the light rail system [22]. Some other studies [23] [24] also provide links between transit availability and transportation mode choices and highlight the potential positive impacts from introduction or expansion of transit services in a region, for individuals, the transportation network operations, and the community. However, given local differences, it is important to conduct local studies in order to gain an understanding of potential impacts of transit ridership increase on local congestion and quantify such impacts.

Based our earlier work on comparisons of available simulation platforms [25] [26], the simulation platform selected for this study was the Multi-Agent Transport Simulation platform (MATSim). The platform is designed to simulate large-scale scenarios by adopting a computationally efficient queue-based approach [27]. It incorporates time choice, mode choice, and/or destination choice into an iterative loop, leading to a stochastic user equilibrium. The MATSim model is publicly hosted and allows for simulation of the impact of changes in transit ridership in medium sized cities such as Birmingham, AL.

An open-source software, MATSim requires its input files to be as XML files. Minimum input files required to run the software are: 1) Configuration file; 2) Network file; 3) Population/plans file. The configuration file builds the connection between MATSim tool and all other Extensible Markup Language (XML) files (e.g. network, population, etc.), and contains a list of settings that influence how the simulation behaves. MATSim’s network file consists of nodes and links and describes the infrastructure that agents can use to move around. Nodes are defined by coordinates while the link requires definition of several attributes including the length of the link, capacity, speed, and the number of available lanes. The population file provides information about travel demand, e.g. a list of agents and their travel diaries. The travel demand is described by the daily plans of each agent. The population file contains a list of transportation users and their daily plans, activities, and legs.

Each simulation job executes in iterations where each iteration executes the selected plans of all agents over an underlying road network. It starts with an initial population demand (a.k.a. plans) in the studied area which is application-dependent. Each agent in the population maintains a pool of up to 5 day-plans. In each iteration, 1) MATSim’s “mobsim” simulation executor runs the selected plans of the agents in the synthetic road network environment; 2) Then, a scoring function assigns a score to each plan based on the corresponding agent’s experiences with the selected day plans (e.g., if congestion happens or not); 3) Afterwards, the replanning step selects a candidate plan based on the plan scores in each agent’s day-plan pool, and may modify this plan for execution in the next iteration.

As far as simulation outputs are concerned, MATSim creates output data that can be used to monitor the current simulation setup progress as well as to analyze results. In each iteration, a linkstats file containing hourly count values and travel times on every network link is generated by the model. The user can specify the output interval of simulation statistics for individual links. MATSim provides overview summaries of counts and other statistics for the whole network, but also analyzes for individual links. Also, a google maps-based visualization is available, showing simulation output results for each station in a pop-up window. Details about MATSim are available in [27] and at https://www.matsim.org/.

MATSim was used to develop a model considering private cars only (base) and three scenarios including public transit. The Population was generated by population synthesis [24] of travel diary from a survey and using open source data [28]. To speed up computational performance, and similarly to MATSIM-based studies reported in the literature [29] [30], 10% of the total population was used for the simulation. Criteria considered for the development of each scenario are listed in Table 1.

For the first scenario, the probability of choosing public transit is set as equal or less than 0.1 because with this probability, 1.1% of total agents (travelers) choose public transit. This percentage represents public transit ridership in Birmingham at the time of study [31] [32]. To determine the impact of increased ridership, the probability is then set as less than or equal to 0.5 and 0.9 respectively for scenario 2 and scenario 3, respectively. Table 1 also shows the expected market share of the various modes considered in the study (namely car, transit, and walk) under the 3 scenarios tested (namely probability of ≤0.1, ≤0.5, ≤0.9).

Simulation model calibration and validation are important steps toward increasing the confidence in use of any simulation model. Comparing traffic counts retrieved from the simulation model with actual traffic counts is an approach commonly used to validate simulation models, including MATSim [29]. In this study, traffic counts from the MATSim model (scenario 1) were compared to traffic volumes obtained from Alabama traffic data collected by the Alabama Department of Transportation (ALDOT) for two time periods (7 AM to 8 AM and 5 PM to 6 PM). Traffic volume data used for this calibration were for selected freeway and state highway sections/links located in the Jefferson and Shelby counties. A total of 90 links along I-459 N, I-459 S, AL 25 S, I-65 N, AL 3 N, AL3 S, AL 5 E, AL 38 E, and the junction of I-65 and I-59 were considered for this validation.

Figure 1 and Figure 2 show model validation results, where the X-axis represents the field volume and Y-axis represents the simulated volumes. The three diagonal lines in the graphs represent the simulated versus real volume ratio of 2:1, 1:1 and 1:2 which are named as 2 counts, 1 count and 0.5 count respectively. Counts falling between 2 and 0.5 count are considered acceptable [33] and most of the data points are within these boundaries, thus the model validation is deemed acceptable.

MATSim generates output according to the link ID which refers to the identity of the roadway sections in the MATSim platform. This study analyzed the performance of road network as a result of increased transit ridership by examining links near the bus stops. The procedure used for selecting 93 links from the 1761 bus stop links within the study network is shown in Figure 3.

Based on the data available at the live traffic website [34], traffic congestion in the Birmingham region peaks from 5 to 6 PM for a typical weekday. During this time slot, an addition 50% of travel time is needed on average to complete a trip, compared to the travel time under free flow conditions. As far as congestion severity time period is concerned, the 5 PM to 6 PM period is followed by 4 PM to 5 PM, 3 PM to 4 PM, 7 AM to 8 AM and 8 AM to 9 AM, during which travel times are 43%, 36%, 33% and 27% higher respectively, as compared to travel time under free flow conditions. The study selected the same five time periods for further analysis.

To accommodate the big range of volumes (120 veh/hr to 2520 veh/hr), links were classified in five different groups based on the procedure described in Figure 4. Resulting link groupings from this procedure are summarized in Table 2.

It should be noted that the simulated traffic volumes are 10% of the total. For example, group 3 includes simulated volumes of 51 - 90 veh/hr which refer to links with actual traffic volumes of 510 - 900 veh/hr as seen in Table 2. Also, the volumes are total volumes per link (directional volumes, all lanes combined).

To determine the performance of the road network in the event of increased transit ridership, traffic volume, speed, and average travel time were analyzed for the selected 5 time slots. The results are summarized next.

Figure 5 shows the mean traffic volume for three of the scenarios, 5 time slots and 5 groups. The purpose of showing the traffic volume data for the study scenarios based on the groups is to allow for understanding how increased transit ridership affects network operations under different volumes levels. The black line, blue line, red line, green line and purple line in Figure 5 stands for the traffic volumes between 7 AM to 8 AM, 8 AM to 9 AM, 3 PM to 4 PM, 4 PM to 5 PM and 5 PM to 6 PM respectively. To facilitate comparisons, the line color scheme used in the analysis represents the same time slots for all the groups.

As the probability of choosing public transit is 0.1 in scenario 1 and then increased to 0.5 and 0.9 in scenarios 2 and 3 respectively, an improvement in traffic performance is expected in terms of speed increases associated with higher transit ridership. It is postulated that increased transit ridership will result in increases in the average vehicular speeds. Figure 6 shows the resulted speed for the three scenarios, 5 time slots and 5 groups studied.

Travel time for a roadway link refers to the time needed to drive from start point to the end point of that link. The expectation is that modal shifts from automobile to transit may result in reduction of link travel times, thus resulting in an improvement of traffic network performance. To verify this expectation, an analysis of travel time data was performed for the study links and for the 5 study time periods. Average travel times for the three scenarios and 5 groups are shown in Figure 7.

To understand whether the traffic flow reduction associated with changes in transit ridership is statistically significant, several significance tests were performed. First, a significance test was performed to test difference within the three scenarios. If the traffic flow reduction was significant within the scenarios, then tests between the scenarios were also performed. Table 7 shows the p-value of the significance tests performed herein. The tests were performed using the 95% confidence level, hence, a p-value less than 0.05 indicates that the difference is statistically significant between/within groups.

The results from the statistical tests between the base model and scenario 1 imply that the current level of public transit use does not have any significant impact on traffic volumes. This is evident from the high p-values documented in Table 7 resulting from the comparison of traffic volumes between the base model and with scenario 1. Furthermore, results show that traffic flow reduction is statistically significant when comparing results from scenario 1 to scenario 2 as well as scenario 1 to scenario 3. The only exception is for group 1 which shows a p-value of 0.224 > 0.05, indicating that there is no evidence to support that there is a statistical difference in traffic volumes from scenario 1 to scenario 2 during free flow conditions (group 1).

Table 8 and Table 9 show the p-values of the significance tests done for speed and travel time changes respectively. The results show that there is not enough evidence to suggest that there is statistically significant differences in speed or travel time associated with the increase in transit ridership as per the study scenarios.