Potential Greenhouse Gas Emissions Reductions from Optimizing Urban Transit Networks

Potential Greenhouse Gas Emissions Reductions from Optimizing Urban Transit Networks
PI: Samer Madanat, madanat@berkeley.edu UC Berkeley; Co-PI: Arpad Horvath, UC Berkeley https://escholarship.org/uc/item/25x1b693

Main findings:

  • In large cities, hierarchical transit systems with mass transit modes (metro, for example) tend to be more cost-and emission-efficient. The dominance of auto emissions for the case of a large city means that large cities should rely more on transit in order to reduce emissions. Figure 4.14 shows the two extremes, where cities with varying sizes rely either on automobile or metro to cover the entire demand. As an auto-only city expands, the emissions increase dramatically due to the longer average driving distance and larger demand. For a city that runs on transit, the emissions also increase as a higher transit frequency is required. However, transit generates much lower emissions than automobile especially for large cities.
  • In small cities, trunk-only bus systems may be more favorable with regards to both costs and emissions savings.
  • Transit demand elasticity offsets transit emissions reduction efforts by causing additional automobile emissions due to demand shifting away from transit. Transit agencies should evaluate the demand elasticity in areas of interest before trying to reduce emissions through lowering transit LOS.
  • The process of transit cost minimization may also reduce GHG emissions.

Abstract: Public transit systems with efficient designs and operating plans can reduce greenhouse gas (GHG) emissions relative to low-occupancy transportation modes, but many current transit systems have not been designed to reduce environmental impacts. This motivates the study of the benefits of design and operational approaches for reducing the environmental impacts of transit systems. For example, transit agencies may replace level-of-service (LOS) by vehicle miles traveled (VMT) as a criterion in evaluating design and operational changes. Previous studies have demonstrated in an idealized single technology transit system the potential of reducing GHG emissions by lowering the transit level-of-service (LOS) provided to the users. In this research, we extend the analysis to account for a more realistic case: a transit system with a hierarchical structure (trunk and feeder lines) providing service to a city where demand is elastic. By considering the interactions between the trunk and the feeder systems, the study provides a quantitative basis for designing and operating integrated urban transit systems that can reduce GHG emissions and costs to both transit users and agencies.

The study shows that highly elastic transit demand may cancel emission reduction potentials resulting from lowering LOS, due to demand shifts to lower occupancy vehicles, causing unintended consequences. However, for mass transit modes, these potentials are still significant. Transit networks with buses, bus rapid transit or light rail as trunk modes should be designed and operated near the cost-optimal point when the demand is highly elastic, while this is not required for metro. We also find that the potential for unintended consequences increases with the size of the city. The results are robust to uncertainties in the costs and emissions parameters. The study also includes a discussion of a current transit system.

Since many current transit systems have not yet been optimally designed, it should be possible to reduce their GHG emissions without sacrificing the LOS. A case study of the MUNI bus system in San Francisco is used to validate this conjecture. The analysis shows that reductions in GHG emissions can be achieved when societal costs are reduced simultaneously. The cost-optimal MUNI bus system has a societal cost of 0.15 billion $/year and emits 1680 metric tons of greenhouse gases. These figures only amount to about half of the cost and a third of the emissions in the current MUNI bus system. The optimal system has a lower spatial availability but a higher temporal availability of bus service than the current system, which highlights the potential benefits of providing more frequent express bus services.

Final Report 

Click on the links below to download any of the quarterly progress reports already finished.


the Pareto curvesfor optimal transit system design by different trunk technologies for a small city with low demand (scenario 1). Compared with all the hierarchical transit systems, bus with no feeder access is the lowest-cost option for all values of the GHG emissions constraint. This is because the average travel distances are short for small cities, which eliminatesthe relative disadvantage of low cruising speedsof buses compared with the other modes. Furthermore, the low infrastructure cost of a bus network allows for smaller stop and route spacings, alleviating the need for incorporating the feeder access. The intra-modal transfer times and feeder emissions are avoided as a consequence. Among the hierarchical transit systems, BRT is the lowest-cost option for allvalues of the GHG emissions constraint. However, LRT and buseshave lower GHG emissions than BRT at their respective cost-optimal points. For cities similar to scenario 1 that aim to optimize only transit system costs, LRT or buses are better than BRT in reducing GHG emissions.Metro is not a competitive option in this context because its costs are higher than all the other modes for all values of the GHG emissions constraint. Moreover, metro has the highest GHG emissions at its cost-optimal point compared with all the other modes. For these hierarchical transit systems, the order of preference for the trunk technologies is consistent with the results of Griswold et al. (2013), where the feeder access is not provided. In summary, incorporating feeder access does not significantly change the comparisons between the trunk technologies for scenario 1, and the societal costs and emissions are even much higher compared with the trunk-only bus system.With the transit demand being doubled (Figure 4.2, scenario 2), the transit agency must improve the service level to minimize the societal cost. The emissions increase for all modes due to the improved LOS, along with the increased societal costs, due to the higher demand. Nevertheless, bus with no feeder access is still the lowest-cost option for all values of the GHG emissions constraint.As for the large cities, the average travel distances are longer, making metro more cost-competitive due to its higher cruising speed(by reducing travel time). Furthermore, incorporating the feeder access allows for larger stop and route spacings, saving the agency cost significantly by requiring a lower amount of the expensive metro infrastructure. This can be verified in Figures 4.3 and 4.4, where metro is the lowest-cost option for mostvalues of the GHG emissions constraint.Buses are the worst choice in this context due to high costs for all values of the GHG emissions constraint. The shortcoming of buses includes low cruising speeds magnified by long average travel distances in large cities, increasing travel time significantly. Meanwhile,the low capacity of buses requires a larger fleet of buses to cover all the transit demand, which also results in a significant increase in agency cost in this large-city scenario. LRT and BRT are better than buses but less cost-competitive than metro, yet both have lower GHG emissions than metro at their respective cost-optimal points.Although the GHG emissions associated with metro are high at the cost-optimal point, these emissions can be significantly reduced without causing large additional societal costs. Take scenario 3, for example, the GHG emissions of metro can be reduced by 30% from 2051 mt GHG/yr to 1435 mt GHG/yr, with only a 2% increment in the

Figures 4.5 through 4.9 show the change in total GHG emissions as transit travel time is increased from the cost-optimal value for a rangeof travel time elasticities. The figures are each shown with the same scale on the horizontal and vertical axes to allow for visual comparison. The bottom line (b= 0) shows the results for inelastic demand, where no users will change modes when the transit travel time increases (as a result of the reduced transit LOS). The top line (b= -1) shows the results where users are most sensitive to changes in the transit travel time.In the case of the trunk-only bus system (Figure 4.5), the elasticity values between -0.8 and -1 produce monotonically increasing emissions as the travel time increases from the cost-optimal value, implying that slight LOS reductions for a city with highly elastic transit demand would be harmful to both the transit users and the emissions. When the elasticity values are between -0.6 and -0.7, there is initially a small emissions benefit as the transit LOS is lowered. However, as the transit travel time approaches 39 min, the emissions start to increase. The elasticity values between0 and -0.3 produce monotonically decreasing emissions. The elasticity values between -0.4 and -0.5 produce slight or no emissions reductions over the travel time values shown. Nevertheless, the trunk-only bus system is still competitive for both emissionsand travel time, compared with all the other hierarchical transit systems in scenario 2.The case of the hierarchical transit systems (Figures 4.6through 4.9) shows the similar relationship between the total emissions and the travel time as in Figure 4.5. Ifind that, for buses, BRT and LRT, the elasticity values between -0.6 and -1 all produce increases in emissions over the travel time values shown. Thus for cities with highly elastic transit demand, transit systems with these trunk technologies should be designed near the cost-optimal points. However, greater reductions in the total GHG emissions are possible for low to moderate elasticities. In the case of

Figures 4.10 and 4.11show the results of this city-scale optimizationfor BRT and metro. The results for bus and LRT are similar to BRT and thus are not shown here. Increases in total emissions are avoided because they are constrained by Ein (4.1). Compared with Figures 4.7 and 4.9, the curves are truncated at the points where it is no longer possible to reduce the total emissions. These points represent the states where further reductions in transit emissions start to be overtaken by the increases in automobile emissions. For the case of the BRT system, the reductions in total emissions are not possible for high elasticity values, which is consistent with Figure 4.7. As a result, the curves of b= -1 and b= -0.9 are negligible. For the metro system, the reductions in total emissions are possible for all the values of elasticity.4.4Effect of city size on emissionsIn the previous section westudied the elastic case for a small city (scenario 2), and verified that for hierarchical metro systems, there exists a phase of decreasing total emissions as LOS is lowered, even for highly elastic transit demand. However,this is not necessarily true for larger cities where automobile emissions are much higher due to the longer average driving distance. To verify this conjecture, westudy the metro system for a large city (scenario 4) in this section.

Figure 4.12shows the relationship between the totalemissions and LOS for a large city (scenario 4). Compared withFigure 4.9, it is easy to observe the difference: with high elasticity, b= -1 for example, the total emissions increase monotonically as the LOS is lowered from the cost-optimal point. The situation is similar for middle elasticity, b=-0.5 for example, where the potential reduction in total emissions is negligible. As a result, for a large city with elastic metro demand, it is not recommended to lower the metro LOS for the sake of reducing the emissions. The underlying reason is that when the city size increases, the average driving distance also increases. These longer distances lead to significantly higher auto emissions. As the metro LOS is lowered, large auto emissions are generated as transit users shift to the automobile, completely overtaking the reductions in metro emissions. Figure 4.13shows the breakdown of the total emissions where we can see the difference between the small city and the large city scenarios.

The dominance of auto emissions for the case of a large city means that large cities should rely more on transit in order to reduce emissions. Figure 4.14shows the two extremes, where cities with varying sizes rely either on automobile or metro to cover the entire demand. As an auto-only city expands, the emissions increase dramatically due to the longer average driving distance and larger demand. For a city that runs on transit, the emissions also increase as a higher transit frequency is required. However, transit generates much lower emissions than automobile especially for large cities.

4.5 Sensitivity Analysis

Admittedly, the mode attributes, cost and emission factors for the trunk transit and feeder transit used in this study may vary in different cities and time windows. This has the potential of shifting the Pareto curves and affecting the comparisons of transit technologies. The changed parameters may also alter the shape of the curves in Figures 4.5 through 4.11hence affecting the discussion on elasticity. A set of sensitivity analyses has been conducted to address these concerns. The analyses help us evaluate the robustness of our conclusions to parameter uncertainties.For example, Figure 4.15shows the Pareto curves of metro, BRT and trunk-only bus for scenario 2after all the parameters are changed by ±30%. Table 4.3 shows an example of parameter changes for metro trunk in sensitivity analysis. The solid curves refer to those in Figure 4.2where the parameters were not changed. It should be noted that ±30% refers to the extreme cases where all the parameter changes contribute in the same direction to the changes in costs or emissions. In the real world, the contributions from different parameter changes often cancel out each other, and thus the Pareto curves are located between the boundaries provided in the extreme cases (Figure 4.15). It can be observed that the position of the Pareto curvesis of low sensitivity to the changes in the parameters. The shifted curves may affect the comparisons of BRT, LRT and bus due to the closeness between them in Figure 4.2. However, the advantage of trunk-only bus and the inferiority of metro remains obvious as in Figure 4.2. Similar observations are made for scenario 1, 3 and 4. It is also observed that the changes in the parameters slightly modifies the positions of the curve clusters in Figures 4.5 through 4.11. However, the shapes of the curves remain the same. Since the discussion on elasticity is largely based on the curve shapes, it is insensitive to the parameter changes.

Chapter 5Implementation: Transit Network StylizationIn the previous chapters, weanalyzed various hypothetical city scenarios and quantified the emissions saved when reducing the transit LOS. The model in thosescenarios utilizedan idealized rectangular grid transit network. Weused continuum approximation (CA) method to minimize the societal costs subject to an emissions constraint, identifying the Pareto curve of optimal transit system design.In reality, most current transit systems have not been optimized to achieve the lowest costs and emissions possible. Therefore, they do not necessarily fall on the Pareto curve. In this chapter, weextend themodelto analyze such a transit system, the MUNI bus network for the city of San Francisco. Webuild a mathematical model of the MUNI bus network to obtain a relatively accurate representation of the current network. Wequantify the societal costs and emissions for the current MUNI system, and where it falls relativeto the Pareto curve. The potentials for emissions reductions and the changes in LOS are then discussed for the MUNI system.As in our previous model, the continuum approximation method is employed to derive the costs and the emissions of the transit system. Since the method requires networks with simple geometric patterns, wechoose a stylized network to approximate the real MUNI network. The stylization procedure is described in section 5.1. The evaluation of the stylized network is presented in section 5.2. Based on the stylized network, the optimal bus system attributesare selected, which arediscussedin detail in Chapter 6.5.1Network stylizationTransit networks are often irregular in shapes. They are confined by the geographical layout of real cities. To use the continuum approximation method, it is necessary to stylize the complex network to the form of its closest geometric idealization. Figure5.1 shows the network of downtown San Francisco Muni bus system in real shape (a) and after stylization (b). Note that for simplicity, only the downtown portion of the Muni network is considered.The downtown area can be stylized as a combination of one trapezoid area and one rectangle area overlapping at an edge, Market Street. The other edges serve as the boundary of the area of interest in this case study (Table 5.1). Since the bus network in the downtown area is of grid structure ubiquitously (Figure 5.1 (a)), wedefine a stylized grid structure within the boundary (Figure 5.1 (b)). Within the trapezoid and rectangle, the bus routes are aligned orthogonally to maintain the grid feature. The stop spacing (s) and route spacing (r) are assumed to be the same throughout the network to filter out the complex network details.

Admittedly, stylization may bring in some errors due to possible oversimplification of the real network. To evaluate the influence of the stylization, it is necessary to compare the costs and emissions derived from the stylized network with those of the real network. The evaluation is provided in section of the user costsThe continuum approximation method is used in the derivation of the costs and emissions from the stylized network. The method enables us to formulate costs and emissionsas closed-form functions of basic system attributes. In this study, the attributes consist of i) supply-side parameters: stop spacing (!), route spacing (!), and headways (!); ii) demand-side parameters: demand density (!), value of time (!). The detailed derivations of these costs and emissions are described below.The cost to a transit user is measured by the time spent using the transit service. The user cost !!”#$is the sum of costs to all the transit users. The transit demand is divided into several categories to account for different O-D scenarios (Table 5.2). Similar to the previous chapters, since the values of the time spent in-and out of transit vehicles are different (Caltrans 2014), it is necessary to derive separately the in-vehicle travel time (IVTT) and the out of vehicle travel time (OVTT). The user cost !!”#$is therefore formulated as the sum of monetized IVTT and OVTT for different categories (5.1).–value of time in-and out-of transit vehicles ($/hr·pax).–Average IVTT and OVTT for the transit demand between –The sum of demands between !and !1.The formula derivations for demand and travel times are described as below.

Chapter 6 Implementation: Transit Network Optimization

In Chapter 5, weused the stylized network and the continuum approximation method to represent the real network. The decision variables (!, !, !!, !!) were assigned fixed values reflecting the current state of the real network. In this chapter, wesolve for the optimal values of these decision variables using the model formulation defined in (3.1).Particularly, thegoal is to solve for the values of the decision variables that minimize the total costs subject to a MUNI busemissions constraint.Similar to Chapter 3 and 4, a Pareto curve for optimal bus system design for MUNI is obtained(Figure 6.1), allowing for comparison between the system’s optimaland current stateof operations. Furthermore, the comparison of the decision variable values provides insights regarding design and operational modifications for the purposes of cost and emission minimization(Table 6.1).

Figure 6.1 shows the current state of the network and the Pareto frontier obtained by the network optimization. As conjectured earlier, the current MUNI bus system does not lie on the Pareto frontier.The relative position of the point of the current state with respect to the Pareto frontier demonstrates the potentials for cost and emission reduction via transit network optimization.The current state of the systemhas higher societal costs and GHG emissions compared with those of the cost-optimal state. In this context, transit agencies may achieve simultaneously the reductions in societal costs and GHG emissions simply by bringing the system from its current state to the cost-optimal state. The tradeoffs between the costs and emissions, as described in the previous chapters, only exists if onewantsto achieve further emission reductions starting from the cost-optimal state.

We compare the current state with the cost-optimal point on the Pareto frontier (Table 6.2). It isobservedthat significant reductions in total cost and emission may be achieved simultaneously by increasing route and stop spacing and decreasing headway. Therefore, a bus service plan with lower spatial availability but higher temporal availability may be more favorable. This suggests that there may be benefits to decreasing local bus service and providing more frequent express bus service. For a small city like San Francisco, the average walking distance is small. Local bus service (with small stop spacing) may cause unnecessary user cost by stopping too often. By increasing walking distance, which increases walking time, an express bus service (with large stop spacing) saves users’ IVTT significantly by stopping less frequently. Furthermore, the reduced headway saves users’ average waiting time at transit stops. As a result, we can observe that the cost-optimal state has lower user cost than the current state, as shown in Table 6.2.The overall transit LOS was not reduced. In fact, the cost-optimal state was found to have a higher transit LOS than the current state, as can be seen in the comparisons of the user costs in Table 6.2. In this case, transit agencies may not need to worry about the unintended consequences due to the potential demand shift.

Chapter 7 Conclusions and Recommendations

Quantifying the potential tradeoffs between level of service and emissions can help transit agencies select the trunk transit technology and optimal network attributes for hierarchical transit grid systems. In this research, we have developed an analytical model for hierarchical transit systems, and researched the economic and environmental competitiveness of different transit technologies in various city scenarios. The demand elasticity between transit and automobile, the demand split between transit and walking, and the demand split within the transit system have also been incorporated in the model. Compared to previous studies (Griswold et al. 2013, Chester and Horvath 2009, Chester et al. 2010, Sivakumaran et al. 2014), this study incorporated demand elasticity and the transit hierarchy simultaneously when assessing costs and emissions, providing a more complete picture that can better assist transit agencies in designing and operating transit systems.


The analysis in this research shows that, in small cities, trunk-only bus systems are more cost-and emission-competitive than hierarchical transit systems. On the other hand, in large cities, both societal costs and emissions reductions can be achieved by deploying hierarchical transit systems with mass transit modes such as metro as the trunk technology. Theseresults are different from earlier findings in the literature (Griswold et al. 2013), which were based on analyses of trunk-only transit systems. Theresults show that incorporating a feeder system not only increases the relative competitiveness of capital-intensive transit technology with respect to total costs,which had been already demonstrated in Sivakumaran et al (2014), but also emissions. Theanalysis also shows that, for cities that have metro as the trunk mode, it is not necessary to design or operate the system near the cost-optimal point as significant emission reductions are possible, without incurring large additional societal cost relative to the optimal cost. The sensitivity analysis shows that thesefindings regarding modes comparisons are robust to variations in the costs and emissions parameters.Transit demand elasticity offsets transit emissions reduction efforts by causing additional automobile emissions due to demand shifting away from transit. For metro, there exists a phase of decreasing total emissions as the LOS is lowered even for highly elastic transit demand. However, when transit demand is highly elastic for bus, BRT or LRT, reducing LOS will cause a net increase in city-wide emissions. These findings suggest that some cities may benefit from lowering transit LOS while others may not, depending on their trunk transit technology and transit demand elasticity. The size of thecity also matters when making such decisions: for a large city, it is not recommended to reduce the LOS of metro if the demand is highly elastic. Imposing an emissions budget on the entire city instead of the individual agencies is a safer course of action to avoid unintended emission backfire and achieve emissions reductions. However, transit demand elasticity is a key factor in determining the magnitude of such reduction potentials. In light of this, agencies should make a thorough investigation of transit demand composition (i.e., the fractions of captive and non-captive users) and flexibility before any LOS-emissions policy is passed.In reality,it is possible for many transit systems to reduce GHG emissions without sacrificingtransit LOS because manyof them are not yet efficiently designed to operate at the cost-optimal point. In this situation, Pareto-improving solutions are possible. The work presented in Chapter 5 and 6 demonstrated such a transitsystem, the MUNI bus network for the city of San Francisco.By optimizing the system’s design and operation to minimize societal costs, GHG emissions are significantly reduced.The cost-optimal MUNI bus system has a societal cost of 0.15 billion $/year and emits 1680 metric tons of greenhouse gases. Thesefigures only amount to about half of the cost and a third of the emissions in the current MUNI bus system.This may be an encouraging message for many current transit agencies that are improving transit systems to achieve minimum costs. General concerns regarding transportation cost minimizations often includes potential negative environmental impacts. Thefindings in this study suggest that this is not necessarily true for current transit systems. Urban transit systems that minimize societal costs may also reduce GHG emissions.Ouranalysis showsthat the cost-optimal state of the MUNI bus system has lower spatially availability but higher temporal availability than the current system. This suggests the potential benefits of decreasing local bus service and providing more frequent express bus service. The overall transit LOS was not reduced. In fact, the cost-optimal state was found to have a higher transit LOS than the current state.7.2Limitations and Future WorkIt isworthwhile to note that there are limitations in this research that might have hinderedmore insights inthe problem. The transit demand in thisstudy has been assumed to be distributed uniformly throughout the entire city. This might be close to reality when cities are small or we are solelyanalyzing small regions such as downtown areas. Yet for larger cities, a non-uniform demand distribution should be more realistic. Future work may address this issue by introducing various layers or spikes of transit demand to simulate downtown areas and large trans-mode stations. Four different trunk transit technologies are considered in this study. They are all assumed to be accessed by walking and feeder buses. In the real world, there are usually many more combinations of transit modes for completing daily travels. As a result, another extension of this study would be to include more potential transit mode combinations in the hierarchical transit system design. For example, many cities have both metro and buses as trunk transit in highly demanded regions. Transit users may have their own preferences when deciding which transit to take. Furthermore, the transit system design need not be “transit-exclusive” since transit users may consider non-transit modes for parts of their trips, such as the situation where people drive to the nearest metro stations. A more macroscopic transit design scheme with more balanced views of urban transportation compositions should benefit the society in the long term.The latter part of the research has utilized a stylized grid transit network for the case study. Even though the network has been evaluated to be a good representation of the real network regarding the cost and emission estimations, there are other concerns that might be of importance in future works. For example, cities that have ring and radial transit networks, such as Chicago and Paris, may require a different set of stylization procedures before applying the continuum approximation method. In this context, the Cartesian coordinate system needs to be replaced by the polar coordinate system. In the situations where cities have even more complex network structures, the stylization procedures should be done on a case by case basis.

25societal cost from $28.3 billion per year to $28.8 billion per year. Moreover, in scenario 4, metrois the lowest-cost option for all values of the GHG emissions constraint. It should be noted that in generating the Pareto curves of the trunk-only bus system in the large cities (scenario 3 and 4),wefind that the cost-optimal point is bound by the bus capacity constraint, which means the buses are already fully loaded in order to cover all the transit demand. With the binding capacity constraint, emissions cannot be reduced any further because that is achieved through lowering the operational frequency and spatial coverage, which will result in bus overloading (the same transit demand is covered by decreased bus service).As a result, the Pareto curve is essentially limited to the cost-optimal point, which can be observed in Figures 4.3 and 4.4. For the hierarchical bus system, wefind that the bus capacity constraint is also binding at the cost-optimal point in both scenarios. However, the Pareto curves are not limited to the cost-optimal points because the emissions can still be reduced from the feeder system. Wealso find that the capacity constraint for the feeder system is not binding because the servicearea for each feeder bus is small.4.3Elastic caseIn this section,weconsider the potential demand shift between transit and automobile, where some transit users may switch to automobile when the transit LOS is reduced,i.e. !is non-zero. To examine how the GHG emissions are affected by the changed demand due to the reductionsin service, curves showing the relationships between the total GHG emissions (!!”#$$#%&$,!+!!!”#$$#%&$,!”#$) and total transit travel time (!!”#$!%#) are developed for different trunk technologies. !!”#$$#%&$,!”!”denotes the marginal automobile emissions due to users switching from transit to cars.Previous research has established a range of reasonable travel time elasticity values for transit in major U.S. cities(Griswold et al. 2014),hence the value of !may differ for different scenarios and trunk technologies. To account for this variability, Iexamine the impact of !between 0.0 and -1.0 on the !−!curves.Webase the elastic case on scenario 2 (small city with high demand), a hypothetical city with characteristics similar to San Francisco.