Abstract
Regional travel demand models are an institutionalized element of the transportation planning process, requiring a multiyear investment from collaborating agencies that rely on model outputs to assist with project prioritization and community visioning. The purpose of this research is to identify ways in which Virginia might (1) alter existing travel demand models in order to consider the impacts of driverless vehicles and (2) use such models to inform questions of interest to regional planners. Because the behavioral impacts of DVs are not known, the paper examined how five sets of alternative futures regarding DVs could be incorporated into the regional model, using the Charlottesville-Albemarle travel demand model as a case study for both (1) potential model modifications and (2) alternative futures, and by extension, related policy questions that might arise.
An outreach exercise conducted with attendees at the Annual Meeting of the Virginia Association of MPOs suggested five particular alternative futures of interest. DVs may (1) alter capacity (reducing it based on operator comfort or later increasing it as platoons result); (2) increase privately owned zero-occupant vehicle (ZOV) trips as commuters seek to avoid parking fees; (3) alter transit’s mode share (decreasing it because DVs make auto travel more appealing by comparison or, alternatively, increasing transit’s mode share through shared DVs which reduce transit’s waiting time); (4) increase ZOV trips through non-familial sharing of DVs; and (5) increase travel by age groups with traditionally lower vehicle access. The regional model incorporated these impacts through altering the Charlottesville and Albemarle travel demand model as a case study, based on ranges of potential impacts of DVs as reported in the literature. (For example, because two sources had reported DVs might increase capacity by amounts of 30% and 100%, and because a third source had reported DVs might reduce capacity by 32%, scenarios were developed based on each of these values.) For each scenario, the impact on vehicle miles traveled (VMT) and vehicle hours traveled (VHT) was recorded as shown in table 11-17(the relative change rates were also recorded as shown in table 18-23), as well as performance measures of interest for each particular scenario. Examples of such measures include transit’s mode share (given stakeholders’ interest in transit) and impacts on oxides of nitrogen (NOx), a precursor to ground level ozone (given stakeholders’ interest in air quality).
For comparisons within a scenario, the results suggest that concerns about the alternative futures do not carry equal weight. For example, in Scenario 1, a capacity reduction attributed to DVs having lowered acceleration rates increases total travel time (vehicle hours traveled) by 46%. By contrast, a capacity increase attributed to DVs potentially having shorter headways of course reduces travel time—but only by 8%. As another example, within Scenario 3, DVs have the potential to increase transit’s mode share from about 0.26% to 3.36% of commute trips if they fully eliminate transit waiting time and render easier the ability to travel from the origin and destination to the transit line. (By contrast, if DVs can make auto travel more appealing, the changes in absolute shares were modest: drive alone, carpool 2, and carpool 3+ increased their mode share from 93.86% to 94.14 %.) Interestingly, the greatest impact for this latter portion of Scenario 3 was on nonmotorized modes: whereas transit trips decreased by about 5%, bicycle trips decreased by about 6%.
For comparisons across scenarios, the results can inform various policy initiatives. For example, the number of zero occupant vehicle trips may increase through a privately-owned DV self-parking (e.g., the owner sends the vehicle back home or to a lower cost parking area) or a shared DV traveling from one person’s destination to another person’s origin. Scenarios 2 and 4, respectively, suggest that while both situations may increase VMT, the former could increase VMT much more than the latter. (For the former, Scenario 2 increased commute-based VMT by 12% and NOx by 10.8%; for the latter, Scenario 4 increased VMT by between 2.3% and 7.3%, depending on the degree of geographical and temporal matching between a leading trip’s destination and a following trip’s origin, and these changes corresponded to NOx increases of 2.08% and 6.65%.) Such figures potentially inform a policy initiatives public support for sharing DVs (relative to individual ownership of DVs) if NOx reduction is a priority.
The ability to incorporate alternative futures into legacy regional planning models can help address some, but not all, questions of interest to MPOs. For example, for this region in particular:
• Planners in this region wanted to know about potential development impacts if parking was no longer needed. A sub-scenario within Scenario 2 examined how conversion of parking lots in the Central Business District to other land uses could affect travel conditions. The results indicate a 2% increase in VHT overall, and speed decreases of no more than 5 mph in the downtown area—and the GIS-based analysis showed substantial land development potential in the downtown areas.
• Concerns about the transition period during which DVs might result in a reduction in capacity are justified if one is concerned about VHT, which increases by 46% if capacity drops. However, the impact on emissions, if one is concerned about NOx, is actually positive—e.g., the reduction in speeds may be associated with a reduction in emissions owing to the parabolic relationship between emissions rates and speeds.
• Generally, induced travel will increase emissions, but not all types of induced travel are of equal concern. For example, an increase in travel by persons who presently do not have access to a vehicle increases NOx by 1.51%. If empty DVs were sent back to their origin rather than parked for all commuters; in that case, NOx increases by 10.8%--more if this behavior applies to other trip purposes. Finally, if longer trips become more feasible due to DVs offering increased comfort, then NOx increases by 21.65%. Thus, changes in behavior due to additional vehicle access increase NOx slightly (by less than two percentage points) by 1.51%—but longer term behavioral changes are much more problematic, with NOx increases that are more than ten times that amount. (A similar phenomenon is noted with VMT: additional travel by persons without access to a vehicle increases VMT by 1.7%; additional travel due to increased comfort increases VMT by 25.6%).
One caveat to these results: the sensitivity of the model to changes in travel impedance (such as capacity changes in scenario 1 or transit attractiveness in scenario 3) is influenced by whether the trip distribution step uses a singly or doubly constrained gravity model. The original Charlottesville/Albemarle travel demand model is doubly constrained such that forecast attractions and computed attractions are equal; this is not normally the case for a singly constrained gravity model. Assuming that one requires forecast productions to be equal to given productions after executing the trip distribution step, selection of the doubly or singly constrained version depends on the extent to which one has greater confidence in forecast attractions or transportation impedance. Both models were tested in this study and generally yielded similar trends; for example, in both cases, a complete elimination of waiting time for the local bus could increase transit’s mode share by about three percentage points in Scenario 3. There were a few cases, however, where the singly constrained gravity model showed a greater magnitude of change than the doubly constrained version. For example, redevelopment of parking lots in the CBD in Scenario 2 increased total VHT by 4% (for the singly constrained model) compared to the 2% as reported above for the doubly constrained version.
To place these results in the context of long range regional planning, they are not as important as key socioeconomic parameters that drive the model. For example, if the region’s population and employment doubled unexpectedly, VHT would increase by 102% and NOx would increase by 34.8%--easily dwarfing almost all of the other scenarios discussed herein. Further, because the results presented here are specific to the case study region, they are not necessarily generalizable to all other locations. However, the modifications to the travel demand model made here can indeed be replicated elsewhere in Virginia. Given that a sample of 11 regional travel demand models in Virginia shows an average age over eight years, the approaches suggested herein indicate one way that transportation agencies can begin to incorporate potential impacts driverless vehicles into their existing modeling efforts—just as those agencies periodically examine other types of unexpected changes in land development, regional growth, or the transportation network. Because of the uncertainty associated with DVs (e.g., will they cause us to take longer trips), the scenario-based approach used herein is one way to examine potential impacts relatively quickly.
