Travel and Emissions Impacts of Highway Operations Strategies

Chapter 7. Conclusions and Recommendations

Conclusions

This study examined the effects that operations strategies have on demand and emissions, especially greenhouse gas emissions, in both the short and long terms. Many past studies have documented the positive effect of operations strategies immediately after implementation – the so-called “opening-day effect” – due to their ability to reduce delay at modest investment costs. The findings of this study reinforce that earlier work. However, a major reason for undertaking the study was to determine the extent to which opening-day emissions would endure potential increases in travel demand resulting from improved travel conditions.

Past studies of induced demand have not specifically addressed the effect of operations strategies. Rather, most have focused on the effect of how changes in a supply variable (e.g., lane-miles) result in changes in travel demand, the idea being that additional highway supply results in lower transportation costs by reducing travel times. However, adding supply (whether it be lane-miles or other form of effective capacity) will only lower travel times for an existing facility during times that the facility is congested, making highway supply a crude indicator of how travel times will change. Because of this indirect linkage, some studies have looked at the direct relationship between travel times and demand.

Past studies distinguish between short-term and long-term effects. Short-term effects include diverted trips (route, temporal, and destination shifts) and new or longer trips resulting from latent demand and mode shift. In the long term, relationship is more complex. Improvements in travel time lead to changes in development patterns, which in turn lead to changes in residential and commercial location choice, and car ownership. Many authors have argued that diverted trips are not true induced demand. In general the size of the long-term effect, in terms of the elasticity of demand with a supply variable, has been found to be higher than the short-term effect.

The applicability of past studies of induced demand to operations strategies is dubious for several reasons:

• Relationships based on areawide lane-mile additions versus VMT changes are too crude for judging operations strategies. Operations strategies are only going to be invoked when congestion is present on specific facilities, most often during peak periods, while it is impossible to tell where and when the areawide lane-miles in the studies were applied. There also is the problem of equating operations strategy effects to lane-miles, but this tractable.
• Relationships based on travel time changes versus VMT are based on travel times for an entire trip. Operations strategies are a concentrated on higher order facilities, and therefore a trip will only be partially exposed. This means that the travel time savings on the operations-improved facility is less than the overall trip travel time, and therefore an adjustment would have to be made. This is important because the study is concerned specifically with the long-term effects of a deployed operations strategy.

The issue of operations influence on travel-time reliability also is often cited as a reason that historical induced demand relationships do not apply. While it is true that operations strategies do improve reliability, it also is true that other types of highway improvements do as well. Do travelers respond differently to reliability changes than they do to changes in typical travel times? Recent research from the SHRP 2 program suggests that both typical travel time and reliability are components of total transportation cost (i.e., travelers’ utility) and that they respond similarly to changes in them. Accounting for reliability as an extra component of total travel cost would be a desirable feature not just for this project but for any analysis that encompasses traveler behavior.

In his 2001 review of induced travel, Cervero noted that traditional four-step travel demand models are ill-equipped to capture induced demand because of the lack of feedback to trip generation and land use. (Cervero, Robert, Induced Demand: An Urban and Metropolitan Perspective, paper prepared for: Policy Forum: Working Together to Address Induced Demand, March 2001.) Much has changed in the intervening decade with the advent of activity-based models, especially those that are linked to land use models. An exploratory analysis of the induced demand effect of operations was presented in Chapter 2 (White Paper #2). Elasticities were pulled from a previous study in 2005 of induced demand using the Portland, Oregon tour-based travel demand model. The effect of improved signal timing in a corridor was used to replicate the effect of arterial operations strategies (e.g., traffic adaptive control systems). The performance was worsened with the induced demand but is still better than the baseline conditions. A 3 percent increase in volumes worsens travel time performance by only 1.2 percent; even a 10 percent increase in through volumes has a better performance than baseline conditions with existing signal settings. The 3 percent volume increase represents the use of the tour-based model elasticities, which account for trip generation effects (as well as route and time-of-day shifts) but not longer term effects such as land use and car ownership shifts.

A second case study was undertaken using empirical data from the Atlanta metro area. The study was based on calculating facility travel times using continuously collected speed data from ITS sensors and automatic traffic recorders, in a before/after operational deployment setting with control sections. The results found that at several locations, ramp metering (one of the operations strategies used) did not have an appreciable effect on travel times. In locations where the operations strategies did improve travel times, no discernible increase in VMT occurred, based on an after period of more than one year. This finding corresponds to several studies of matched facility pair comparisons in the literature.

The results of the above case studies coupled with the fact that several agencies are now using advanced modeling frameworks, led the study to consider the use of a modeling framework as the primary way of getting at the demand implications of deploying operations strategies. Originally, the use of the SHRP 2 C10B modeling framework developed in Sacramento, CA was planned. This framework links an activity-based travel model with a mesoscopic simulation traffic simulation model. Such a framework provides more refined estimates of network performance than traditional traffic assignment procedures, but this framework treats land use as a fixed input. Due to delays in the SHRP 2 C10B schedule, it turned out that it could not be used for this project.

An alternative modeling framework was selected – the MTC travel model in use in the San Francisco, CA, Bay Area. This model is used for all of MTC’s travel forecasting needs, unlike the SHRP 2 C10B model which was the product of a research product and not in “production” mode yet. The MTC model links an advanced iterative land use simulation model (UrbanSim) with an activity-based travel model, so that a more comprehensive treatment of demand effects is possible. Its shortcoming is that it uses traditional traffic assignment procedures, which means the performance estimates are cruder. (A review of other advanced modeling frameworks around the U.S. revealed none fulfilled all the requirements for this project, namely, an integrated land use model, activity-based travel model, and dynamic traffic assignment based on mesoscopic simulation.)

One of the project’s original objectives was to account for the effect of reliability on demand. To this end, original research was conducted with the UrbanSim model and data from the Bay Area. The results found that development patterns are affected by changes in reliability in additional to typical travel times. This finding mirrors that of the SHRP 2 research that found traveler behavior also is influenced by both travel time and reliability. Essentially, reliability is an extra congestion-related cost that heretofore has not been accounted for in traveler behavior analyses. The relationships developed by the research were imbedded in a special version of UrbanSim for use in this project.

The enhanced UrbanSim version of the MTC modeling framework was used to conduct tests of deploying operations strategies. This framework includes feedback loops for travel time to the activity model and for both travel time and reliability to the land use model. Congestion in the network was high. Results show that the deployment of operations strategies increases regional VMT, and the increase is proportional to the travel time savings. For the network that was tested, which was significantly congested, for strategies that represent a reasonably high impact on congestion (e.g., bundles of strategies) the VMT increase does not fully erode the CO₂ emissions benefits of operations; small benefits remain after accounting for both short-term and long-term demand effects at the regional level. Strategies that have a lower congestion impact (e.g., ramp metering deployed alone), a marginal increase in CO₂ emissions was found.

The long-term demand increases observed in the MTC model were used to update the I-15 traffic simulation runs. Results showed that the increased demand runs showed less benefit than the original runs, but for the majority of cases, the emissions benefits were preserved. The demand adjustment procedure used was crude, but necessary given that an integrated model capable of estimating demand changes and refined speed/delay estimates currently is not available.

All of the review and analysis conducted in this study points to several overall conclusions:

1. Operations strategies have an effect on short-term and long-term demand patterns, based on the regional modeling conducted. Because operations strategies improve travel time, there is no a priori reason to expect them to behave any differently than capital expansion projects in this regard. However, the strategies tested in this study were all supply-related. Traveler information, which affects demand, was tested using a simulation model, but the results were deemed to be problematic. In the short term, traveler information may reduce demand on congested facilities by allowing travelers to make different choices for destinations, modes, or to forego a trip altogether. (Shifts in routes and departure times effected by traveler information are likely to have a negligible impact on demand.) However, in the long term, to the degree that traveler information has the global effect of reducing travel times, we would expect it to have similar demand characteristics of other strategies.
2. An empirical before/after analysis of operations deployment (ramp metering and incident management) revealed neither significant changes in travel time or demand. This may be due to relatively small decrease in travel times observed (compared to what would be achieved through capacity expansion or bottleneck removal), indicating that travelers require a significant change in travel time before they adjust their short-term behaviors.
3. Travel-time reliability affects land use decisions. Recent SHRP 2 research found that reliability affects traveler behavior and that, along with typical travel time, is part of the overall disutility associated with trip-making. This project has extended that finding to include the behavior household and business land use decisions. Because reliability is affected by many factors – including disruptions, demand, and their interaction with physical capacity – we expect that other improvements beyond operations would have a similar effect.
4. A microsimulation model, TransModeler previously calibrated for the Integrated Corridor Management Analysis in the I-15 corridor in San Diego, was used to gauge the effects of operations strategies. Individual vehicle trajectories were obtained from the model runs and converted to operating mode distributions for input the MOVES model to produce emissions estimates. This approach was deemed to be superior to using average speeds for MOVES input because it captures vehicle modal activity. However, there has been recent skepticism about the ability of microsimulation-based vehicle trajectories to replicate real-world trajectories. The reason is that the models have been internally calibrated to reproduce macrolevel performance, not individual vehicle performance. This discrepancy is a major concern for trying to obtain an absolute number for emissions; it is probably not as important for judging the relative differences in strategies, as done here.
5. Under the assumption that there is no short- or long-term change in demand, operations strategies produce emissions benefits at the corridor level, including the primary greenhouse gas, CO₂. The reductions in emissions range from two to nine percent, depending on the type of operations strategy deployed. These results are based on using a microscopic simulation model to develop trajectories for the MOVES model.
6. Accounting for demand changes created by the improved travel conditions resulting from operations, the emissions benefits at the regional level are less than at the corridor level. Regional emissions varied from -1.5 to +1.0 percent, depending on the strategy deployed. This result is based on the regional modeling framework used in this study.
7. Accounting for increased demand due to the original deployment of operations at the corridor level, emissions reductions are still present, although the reductions are not as great as if no demand increase is assumed (one to nine percent emission reductions). This result is based on using the demand shifts determined from the regional travel model and applied to the microscopic simulation/MOVES model framework. Because the simulations were unable to account for all of the additional VMT estimated by the regional modeling, we expect the emissions benefits to be overstated. Had the simulations accounted for all of the additional VMT, we believe that emissions benefits would have been either neutral or slightly positive.
8. Microsimulation models are excellent tools for assessing roadway performance in terms of travel time and delay. Our experience indicated that their handling of demand changes is more problematic. This manifested itself the most in the analyses where traveler information was implemented – some of the results appear to be counterintuitive. Also, trying to match roadway VMT targets by modifying the trip table based on a “select link” analysis is performed is a difficult task. Finally, even for routine scenarios the model’s shifting of demand makes it hard to compare the effects of one strategy versus another. VMT is an legitimate effect of network conditions, not a static input, but it is difficult to know if the model’s treatment of demand replicates reality.
9. The study stretched the limits of current modeling capability by stitching together results from one model (demand estimates from the MTC travel model) with another (speed estimates from the I-15 microscopic simulation model). The ideal modeling framework to study this problem would have a single model that has the land use and travel activity components of the MTC model with the traffic assignment portion replaced with mesoscopic simulation model. Even then, there is a question whether the vehicle trajectories produced by mesoscopic simulation adequately reflect real-world trajectories. In fact, the accuracy of microscopic simulation produced trajectories have been called into question. Until this issue is resolved, a good deal of uncertainty will remain in any modeling framework that is employed to study the long-term effects of operations strategies on emissions.

Recommendations

Based on our experience with this project, the team offers the following recommendations for future work.

• Fully integrated modeling frameworks with advanced features should be promoted in order to understand the supply demand implications for alternative investments. These features should include:
  • A land use model that is sensitive to changes in transportation network conditions.
  • An activity-based travel demand model.
  • Traffic assignment via simulation procedures (e.g., mesoscopic simulation) that employs dynamic traffic assignment.
• Travel-time reliability should be both an output of the modeling process as well as an input. Traditional travel demand and microsimulation models should produce reliability measures as output for assessing system performance. Research should be on alternative methods for doing so, including postprocessing and scenario-based analysis. Further, reliability should be part of the feedback process in the modeling chain, in the same way that typical (average) travel times currently are used. This study showed a method for incorporating reliability in land use projections; a similar effort should be undertaken to incorporate reliability into activity models, both as an adjunct to existing models and in the development of new ones.
• When operations projects are evaluated, demand changes over a short-term horizon should be included. Evaluations of completed projects is an important component of a performance management system. Before/after evaluations have traditionally focused on fairly short time periods. With the inclusion of reliability, the time periods must be at least one-year long. We recommend an even longer time horizon – perhaps two years – so that demand shifts can be observed and correlated with improvements in travel conditions. Challenges exist for conducting these studies, including the impact of diversion on facility traffic volumes and changes in the drivers of ambient demand such as economic fluctuations and fuel prices. These studies will add to the knowledge gained here in the Atlanta case studies.
• Emissions estimates derived from simulation model trajectory outputs should be investigated further. Specifically:
  • Comparison of emissions derived from simulated trajectories versus real-world trajectories.
  • Comparison of emissions derived from simulated trajectories versus the use of average speeds.

For these comparisons, determine the size of the differences and if adjustment procedures could be developed.