Tradeoffs between costs and greenhouse gas emissions in the design of urban transit systems

Many studies have attempted to quantify or compare the emissions from buses (Herndon et al 2005, Puchalsky 2005, Ally and Pryor 2007, Chester and Horvath 2009, Cui et al 2010) and rail transit (Puchalsky 2005, Messa 2006, Chester and Horvath 2009, Chester et al 2013), but fewer have attempted to examine the life-cycle beyond the operations phase (Ally and Pryor 2007, Chester and Horvath 2009, Chester et al 2010, Cui et al 2010). This discrepancy may be due to a greater policy focus on tailpipe emissions. As well, estimating the environmental effects of infrastructure is complicated while Puchalsky (2005) compared emissions from bus rapid transit (BRT) and light rail transit (LRT), he only examined emissions from the operation of the vehicles, omitting the significant emissions for infrastructure construction, maintenance, and operation identified by Chester and Horvath (2009). For example, the per-passenger-kilometer-travelled operating GHG emissions are comparable for peak-hour urban diesel and commuter rail, but when the rest of the life-cycle is included, the GHG emissions of commuter rail are nearly double that of bus (Chester and Horvath 2009). Furthermore, life-cycle assessment studies have generally been case studies, making it difficult to generalize results to other locations and various technologies. For example, emissions from electric rail or bus services are dependent on the local electricity mix (Messa 2006, Chester and Horvath 2009, Cooney et al 2013).

A recent Transit Cooperative Research Program report (Gallivan and Grant 2010) identifies several ways in which transit agencies are reducing GHG emissions: expanding transit service, increasing vehicle passenger loads, reducing roadway congestion, promoting compact development, alternative fuels and vehicle types, vehicle operations and maintenance, infrastructure construction and maintenance, and reducing emissions from facilities and nonrevenue vehicles. Some of these approaches, however, are not necessarily cost effective or effective at reducing emissions. The approaches can be generalized into those that reduce the emissions of the transit system (Cook and Straten 2001, Schimek 2001, Stasko and Gao 2010) and those that cause transit to displace other emission sources (Vincent and Jerram 2006, Hensher 2008, McDonnell et al 2008). Schimek (2001) found that it is more economical to retrofit diesel engines rather than buy new vehicles, while Stasko and Gao (2010) developed a model for optimizing vehicle retrofit, replacement, and assignment decisions. Reductions due to displaced emissions are difficult to estimate or forecast as they require an understanding of how mode choice may be affected by improvements in transit service. A case study of several regions in Europe found that improved transit quality attracted non-motorized users, not drivers, causing a net increase in emissions (Poudenx 2008). Emissions per passenger kilometer travelled are also highly dependent on ridership (Chester and Horvath 2009).

While Gallivan and Grant (2010) mention 'transit agency operations' in their report, their focus in that area is on the reduction of tailpipe emissions through engine upgrades and low-carbon fuels, reduction of energy consumption in facilities, and the impacts of construction and maintenance. Other operational improvements that could improve emissions include (1) increasing spacing between stops in order to increase the average vehicle speed and reduce the number of accelerations and decelerations, (2) signal priority, (3) using smaller vehicles, and (4) reducing the number of vehicles required to satisfy user demand. These approaches have not been discussed at length in the literature. In one study, an optimal bus stop spacing of 700–800 m, rather than the US average of 330 m, was found to reduce fuel consumption and carbon dioxide emissions substantially by reducing stops and starts and increasing the average speed, but had little effect on other air emissions (Saka 2003). Others have examined ways of optimizing the assignment and scheduling of 'green vehicles' within a fleet that includes older vehicles to reduce total emissions (Beltran et al 2009, Li and Head 2009). Dessouky et al (2003) jointly optimized costs, service, and emissions for a demand-responsive transit service, but a similar approach has not yet been applied to fixed-route public transportation systems Diana et al (2007) compared the emissions impacts of traditional fixed-route and demand-responsive service at different demand and services levels. Emissions were based solely on distance travelled, and analysis ignored the potential impacts of average speed, acceleration, and deceleration.

There have been numerous studies on transit system design, but very few have included any environmental metrics (Dessouky et al 2003, Saka 2003, Diana et al 2007, described above). Continuum approximation (CA) models have been used to optimize transportation network design to minimize system and user costs. These models can provide general insights into how to structure efficient transit systems by making generalizations that simplify the analysis. Several studies have used CA to optimize stop spacing (Wirasinghe and Ghoneim 1981, Kuah and Perl 1988, Parajuli and Wirasinghe 2001) along with other network attributes such as headway (Chien et al 2010). Others have examined the structure of transit networks, such as grids, radial systems (Byrne 1975, Tirachini et al 2010), and hub-and-spoke systems (Newell 1979). Tirachini et al (2010) compared light rail, heavy rail, and BRT on a radial transit network. Daganzo (2010) went a step further by determining a system design and operating characteristics that could make transit competitive with the automobile. Applying his models to Barcelona, he found that optimal service would reduce the total number of vehicles significantly, thus reducing total transit emissions. Sivakumaran et al (2013) explored the influence of access mode on choice of trunk technology, and the research in this letter builds on the models developed for that research. Continuum approximation models are a promising approach for the joint optimization of costs and emissions, having been used for cost minimization in previous research.

Beginning with pair-wise analysis of the four city scenarios, we can observe the system changes that occur when β is changed. Figure 3 shows the Pareto frontiers for optimal transit system design by mode for a small city with low (scenario 1) and high values of β (scenario 2). For scenario 1, BRT is just barely the lowest cost option for most values of the GHG emissions constraint. At the cost-optimal point at the right end of the curve, however, bus and LRT have lower emissions than BRT. Metro is not competitive in this scenario as its costs are higher than both bus and BRT and its emissions at the cost-optimal point are at least four times that of the other modes. The attributes of the systems at the cost-optimal point (shown in table 5) reveal how the mode parameters impact the relative costs and emissions. The low infrastructure cost of bus allows for small stop (s = 0.8 km) and route (ps = 1.6 km) spacing compared to the other modes, which have route spacing between 1.8 and 3.6 km. The small spacings for bus make access on foot easier, and thus keep the out-of-vehicle travel time much lower (14 min, or 58% lower than metro). This result regarding route spacing is consistent with the findings of Tirachini et al (2010). BRT, which is faster than bus, has a slight edge in system cost because the travel time is 5 min shorter. LRT is able to be competitive for emissions because it has cost-optimal route spacing 50% higher than bus or BRT, so the amount of required infrastructure and number of vehicles in operation are lower. The emissions results in tables 5 and 6 are shown to no more than two significant digits because the quality of the data and the accuracy of the model do not justify greater precision. The section on sensitivity analysis provides further explanation.

Zoom In Zoom Out Reset image size

When β is quadrupled (figure 3—scenario 2), the costs and emissions increase for all modes. With the increase in value-of-time for users, the agency must improve service to minimize the system cost. The cost-optimal route spacings for bus and BRT are cut nearly in half and the route spacings for the other modes are also reduced. Along with the decrease in headways, the service improvements lead to a reduction in travel time for users at the cost-optimal point. While bus and LRT have lower cost-optimal emissions, BRT service has a lower cost and a greater travel time advantage than in the low β scenario.

Comparison of the remaining scenarios (figure 4) reveals similar patterns between low and corresponding high β cities. Additionally, we observe that, with an increase in β, bus systems have significant loss in cost advantage relative to the other modes. In the large cities the benefit of short access distances for bus is outweighed by the disbenefit of the slow speed and many stops that increase the travel time. Improved service is required to balance the increased impact of user costs as the number of users and their wages increase, and this service increases both the agency costs and emissions. As shown in table 6, the service improvements help to reduce the travel time for users compared to the low β cities (scenarios 1 and 3), by reducing headways and route spacing. Bus has the greatest headway reductions, the smallest route spacing reductions, and the smallest travel time reductions. The route and stop spacings appear to be as low as they can optimally be because of the tradeoff between access time and stop time in the vehicle. For metro we see travel time reductions of up to 20% as route spacing and headway reductions reduce out-of-vehicle time without increasing in-vehicle travel time. Scenario 4, which most resembles a high-density metropolis, is the only case where metro has the lowest travel time, but BRT still has a slight cost advantage that, however, may not be significant. We expect that the addition of bus feeder service will provide additional advantages to BRT, LRT and metro for larger, higher-density cities. With faster access to stations, the fast trunk modes can operate with larger stops spacings, thus reducing user in-vehicle travel time and agency costs (Sivakumaran et al 2013).

Zoom In Zoom Out Reset image size

Increasing the city size brings about a different set of system changes. The average travel distance is increased, and thus the in-vehicle travel time takes up a greater proportion of the trip. The costs and emissions increase by an order of magnitude because of the larger coverage area for service. Between scenarios 1 and 3, the cost-optimal route spacing and headways are relatively stable, but the stop spacing nearly doubles (table 6). This change increases the access time on foot, but also reduces the in-vehicle travel time by reducing the number of stops the vehicle makes. As the optimal results show, there is little benefit to the system in reducing headways, which remain around 8 or 9 min for all modes in scenarios 1 and 3. Decreasing headways for the large city scenarios would decrease travel times by up to 10% in the extreme case, but the extra vehicle costs and emissions would be prohibitive. BRT is the optimal system for both costs and is competitive for emissions in both large city scenarios (scenarios 3 and 4). The large cities require greater infrastructure mileage to serve the area, as well as higher speeds to reduce the travel times. BRT has the benefit of lower infrastructure costs than LRT and metro, which allows for greater coverage at a low cost, and higher speed than bus, which makes up for the slightly longer access time. Regardless, many of the emissions results are significantly different (> ± 10%) when compared using one significant digit.

The results of the parametric analysis suggest that a BRT system could be the best low-emissions option for many types of cities, and is also the lowest cost option. Bus is a low cost option in small cities, but BRT is lower cost when the wage rate and demand density are high. The optimal system attributes (p, s, H) vary between city types, so it is important that the analysis be repeated with the parameters of a specific city before a new system is designed. Metro, although cost competitive in the large cities, has emissions on the order of four times greater than any of the other modes for all scenarios. The emissions parameters used for metro, which were based on the BART system in the San Francisco Bay Area, were about an order of magnitude higher than any of the other modes. Modeling after a different metro system could potentially produce more favorable GHG emissions results.

Another way to present the optimization results is to show the change in travel time (including both in- and out-of-vehicle travel time) and GHG emissions as the carbon price is increased relative to the cost-optimal point on the Pareto frontier, where an optimal agency would operate. This carbon price analysis of optimal transit systems allows us to determine the economically efficient level of GHG reduction. By operating at the point on the curve where the shadow price is equal to the carbon price, the system can avoid investing more than the market value in achieving additional GHG reductions. As an example here, we look at scenario 2 and examine the changes that occur as the price of carbon is increased. In figure 5, the dashed lines show the percentage change in GHG emissions as the carbon price increases, and the solid lines show the percentage change in travel time with increase in carbon price. The vertical jog in the lines for some modes shows the point where the optimal route spacing multiplier increases by one. With the large increase in route spacing, the emissions are reduced slightly more, but the travel time for users increases significantly. In general, the GHG emissions reduce more quickly than the travel times increase. Bus has the smallest reductions in emissions, while BRT has the smallest increases in the travel time for the range of carbon prices. Figure 5 shows carbon prices up to $3000, much higher than seen in the literature ($5-$65, IWGSCC 2010; $115, Knittel and Sandler 2011), but which corresponds to a travel time increase of less than 10% for LRT. For a carbon price of $100/mt, the potential emissions reductions are at most 3% and the corresponding travel time increase is negligible (less than a minute) because the cost increase is spread among many users in the system. Similarly, for a carbon price of $30/mt, the potential emissions reductions are at most 1% and the travel time increase is even smaller. Achieving these small reductions in emissions will cause almost imperceptible service changes to the system. This result suggests that greater GHG emissions reductions beyond a realistic market value of carbon could be implemented without inducing users to shift to other modes.

Zoom In Zoom Out Reset image size

The carbon price analysis demonstrates that small changes in the values of the decision variables have a very small effect on the optimal cost, but a larger effect on the optimal emissions. To further examine this behavior of the model, we performed sensitivity analysis on the emissions parameters for BRT and metro for scenario 2. Increasing or decreasing each parameter by 50% caused small changes in the optimal cost results and produced the cost-optimal GHG emissions shown in table 7. For BRT, which requires little infrastructure investment relative to rail, changing the vehicle operations parameter (E_V) was the only change to cause greater than 5% changes in the total emissions at the cost-optimal point. The changes for metro were larger (up to 15%) for the two infrastructure parameters (E_I, E_S).

Although the optimal costs do not change with the sensitivity analysis, the emissions change with variation in E_V and cause significant changes to the shape of the Pareto curve (figure 6). This result suggests that improved bus technology, such as natural gas vehicles, could significantly improve the emissions for both bus and BRT. Additionally, the operating emissions for LRT and metro are highly dependent on the electricity mix, which is the Bay Area mix for the parameters used here. A transit system in Seattle, Washington would have lower operating emissions because of the large percentage of electricity coming from hydroelectric power. Since 2009, Muni and BART have switched to cleaner energy, including 100% hydroelectric power for Muni, which would lower GHG emissions by about 60% relative to the Bay Area electricity mix used in the calculations of emissions factors in Chester and Horvath (2009), Chester (2013). In Cleveland, Ohio, on the other hand, the electricity mix is more heavily dependent on coal power, so operating emissions would be significantly higher. Examining the source of the emissions for each mode (table 8), we see that infrastructure is the source of 30% of the emissions for metro, and no more than 4.5% for the other modes. This result further emphasizes the importance of vehicle technology and electricity mix for reducing transit emissions. Additionally, our scope is limited to GHG emissions, but there are additional tradeoffs when local criteria pollutants are considered. Electric vehicles have the advantage of keeping the emissions away from the transit users, but may disproportionately affect the vulnerable population that is located near the power source (Lobscheid et al 2012).

Zoom In Zoom Out Reset image size

The emissions results are uncertain, as demonstrated by the sensitivity analysis. The final emissions can vary by as much as 50% within reasonable ranges of parameter values. The vehicle operation parameters are based on outputs from Chester and Horvath (2009), who assessed them as being of relatively good quality. They based bus operating emissions, however, on a standard drive cycle, which cannot account for the nuance in the actual model drive cycle as the stop spacing varies. Emissions from the various transportation modes may vary based on the drive cycles these modes experience. For example, in Orange County, California, bus stop spacings will be larger than in New York City, likely resulting in better fuel economy for the buses. The BRT vehicle manufacturing and operating emissions are prorated from standard diesel bus emissions, and are therefore, less accurate. Infrastructure emissions are less certain than the vehicle emissions as they are based on US industry averages from EIO-LCA (CMU 2008) and do not account for local variations. The infrastructure emissions also have a much smaller impact on the total emissions, so variation in the parameter values will have a proportionately smaller impact.