Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
At the beginning of the 21st century, clear indications of a paradigm shift in transportation modeling are apparent. A growing number of agencies across the United States are abandon- ing established traditional modeling techniques and exploring advanced practices in travel fore- casting. This synthesis report evaluates the benefits advanced models might offer, summarizes implementation and institutional issues that may form barriers to change, and distills lessons learned from those agencies that have invested in advanced modeling practices. The findings are based on narrative interviews with more than 30 agencies that have pioneered these mod- els, literature reviews, and practical experience gained by leaders in the field of advanced travel forecasting. Advanced transportation modeling is defined as those practices that go beyond the tradi- tional four-step travel demand modeling approach. Specifically, this includes five areas of modeling: tour- and activity-based models, land use models, freight and commercial move- ment models, statewide models, and dynamic network models. All of these advanced models, with the possible exception of dynamic network models, have been successfully used to address policy and investment options at urban and statewide levels. Several of these analyses simply could not have been credibly evaluated with traditional four-step models. Once advanced models were applied and implementation obstacles overcome, most agen- cies reported significant benefits from them. A frequently mentioned example is the elimina- tion of non-home-based trips in tour- and activity-based models. This trip purpose is the most uncertain one in traditional models, as neither origin nor destination is at the home and, there- fore, no socioeconomic data can be associated with or used to constrain these trips. In tour- based models, an individual may make several trips throughout a day, and their home loca- tion, work location, income, and modal availability are known for every trip simulated. Each trip can be attributed to single individuals or households, which allows analyzing; for exam- ple, vehicle-miles traveled generated by different neighborhoods or the impact of a toll road on low- and high-income households. Activity-based models allow for the splitting of time-of-day into much finer temporal units than traditional models that commonly differentiate at most four periods of the day. If the effects of congestion pricing are to be analyzed, activity-based models permit the tracking of how far different household types are willing to deviate from their preferred travel time to reduce or avoid a toll. These models further explicitly consider household interactions, such that if in a one-car household someone uses the car for a work trip other household members cannot use it for a different trip at the same time. Another important advantage of microscopic modeling approaches is its flexibility, in that the structure and internal relationships can usually be far more easily changed than in purely mathematical models. For example, an agency might wish to test the impact of only allowing vehicles with license plate numbers ending with odd or even digits within a congested area. The microscopic model simply extends the characteristics of vehicles by including the license plate numbers and is then ready to simulate such a policy. Dynamic network models are developed to keep track of single or small groups of vehi- cles on the network and therefore are able to define speeds and congestion with much higher SUMMARY ADVANCED PRACTICES IN TRAVEL FORECASTING
accuracy and precision than traditional static assignments. This allows for identifying bottle- necks in the network, as well as a much more precise estimation of traffic emissions. Land use models are implemented for two reasons. On the one hand, they allow the test- ing of land use policies, such as an urban growth boundary or transit-oriented development. On the other hand, they can be integrated with travel models to simulate the interaction between them. This interaction includes the effects that a new highway may trigger in land use patterns as well as new land use development that may worsen congestion. Freight and commercial movement models are implemented to account for their growing share of traffic congestion. Freight and commercial vehicles react quite differently to many transportation policies and network conditions. Depending on the commodities transported or the service provided these trips may be much more sensitive to changes in travel time or tolls, requiring models that are appropriately sensitive to such dynamics. Statewide models are implemented to analyze policies at the regional level. Although an additional highway may relieve congestion locally, it may alter long-distance trip routing sig- nificantly. Regional models, which in an ideal world are integrated with local travel models, reveal the impact of policies on the big picture, beyond the often artificial boundaries of a city or metropolitan planning organization. Table 4 in chapter three summarizes the benefits offered by these advanced practices in transportation modeling. The majority of agencies that decided to move toward advanced travel models were motivated by the need to address policy issues that go beyond simple traffic analysis. In a policy context where the questions asked are more complicated than âhow many lanes?,â the development of advanced models turned out to be more likely, as there was more support by decision makers to build models beyond the four-step travel model. As clear as the advantages of advanced travel models are for many agencies, implemen- tation and institutional issues have hindered their adoption in many cases. This is hardly sur- prising, as most paradigm shifts call for taking risks and overcoming difficulties with new approaches. It is interesting to note that the pioneers in advanced modeling mostly perceived changes associated with the paradigm shift to be gradual. By contrast, those who have begun the transition to advanced models more recently tended to view such changes as more radical and revolutionary. Several practitioners noted the perceived complexity of advanced modeling techniques as a significant barrier. They explained that such increased complexity pervaded all aspects of advanced models, including their structure, data requirements, and computational bur- den. However, it was pointed out that explaining an advanced modeling approach to deci- sion makers and the public may actually be easier, because simulated behavior is closer to reality and requires less abstract thinking than aggregate traditional approaches. Further- more, complexity is often necessary for policy analysis, such as having a time-of-day model that addresses peak spreading when peak period pricing is introduced versus a tra- ditional model with fixed time-of-day factors that is arguably simpler, but cannot answer the policy question. Model calibration becomes more challenging as more simulated detail equates to more out- put variables that need to be analyzed. Accepted standards on how to validate these advanced models is an open question. That said, it is reasonable to expect that advanced models would validate at least as well as traditional four-step models. Being in an early stage of development, very few advanced models have been trans- ferred from one location to another. The development costs are a significant issue, because no commercial standard software for activity-based models exists. Currently, Atlanta and 2
San Francisco are jointly developing an activity-based model and sharing the software devel- opment costs. Most activity-based and land use models developed to date have been based on open-source code that is further customized to conform to the particular agencyâs needs. In contrast, most dynamic network models are supported by commercial vendors. The hardware requirements are significantly greater for most advanced travel models than for traditional models. Even with clusters that consist of several computers, long run times remain a significant issue. Many agencies defined overnight runs (up to 16 h) as the upper limit for reasonably making use of an advanced model for policy advice. Data requirements are typically not more onerous for activity-based models than for tra- ditional four-step models. The methods required for travel surveys are quite similar, although larger sample sizes might be required if highly detailed travel markets are to be analyzed (as is also the case for traditional models). Land use models require additional zonal and regional data, although many of them are generally available with a reasonable level of effort. For freight modeling and dynamic network models, however, lack of data may be a serious imped- iment to their development, validation, and application. Most advanced models took more time to develop than anticipated. It was noted, however, that this undesirable outcome is hardly unique to the advanced models considered in this report. Meeting the schedule was revealed as a bigger obstacle than finding the necessary funding. In most cases, funding was provided by the metropolitan planning organization or state depart- ment of transportation. Although funds were generally available for model development, the same was often not true for education and training. Developing a model in phases, with well- defined milestones, has emerged as one effective method for reducing the risk of schedule delays or financial losses on such projects. Another frequently mentioned issue by those interviewed was the lack of sufficiently trained staff. Several of the agencies fortunate enough to have appropriate staff noted that they have to cover a wide range of assignments, often leaving them too little time to focus on model development and application. If a consultant delivers a model, the importance of extensive training throughout the life of the development work was emphasized several times by agencies interviewed during this study. Every interview was concluded by asking the respondent(s) what they would do differently if they had the chance to repeat the process. This provided interesting insights that are impor- tant for the profession at large to learn from. The respondents made it clear that the model must be designed to meet the needs of each agency. Planning departments that only report highway volumes at an aggregate level might not need to depart from traditional four-step models. However, in cases where complex policy and investment questions that transcend just trans- portation are being asked, the value of advanced models was readily apparent. The changing policy environment and current policy issues, most of them not anticipated when traditional models were developed, are pushing the development of advanced models forward. A large number of agencies reported that a multi-year travel model development plan was invaluable for justifying an investment in advanced transportation models. Such a plan was used to educate staff and decision makers as well as to justify funding. The written document guided the ongoing effort, reminded executives of the modeling vision agreed on, and estab- lished milestones and criteria for success. All successful advanced modeling projects reviewed here were guided by such a long-range plan. A champion to lead the modeling effort was identified as the key ingredient for success by those agencies that have moved toward advanced modeling. This champion was not neces- sarily the technical leader of the modeling team, but often someone who was closer to deci- sion makers and able to translate policy needs into modeling concepts. Having the support of 3
mentors or key executives strengthened the role of the champion. In a few cases, the role of the champion was fulfilled by a consultant. However, those agencies with an in-house cham- pion tended to be more successful in the long term, because the model was used in applica- tion after the initial development project was completed and the consultant contract finished. The critical importance of staff education and professional development was mentioned in many interviews, as technical skills alone were not sufficient to ensure success. Model developers being equally interested in using them in studies and application appears to be uncommon. In most cases the software development had to be outsourced, limiting the abil- ity to later make adjustments to the model with internal resources. No universally satisfying solution was found, underscoring the necessity of continuous education and training of staff members. Some of the most successful models analyzed in this report followed the Agile Develop- ment paradigm, which proposes to start with the simplest model possible and then continu- ally evolve it over time. This approach proved to be more successful than starting with the âbig design upfront,â which tries to build large complex models in one step. Interviewed agencies repeatedly brought up for discussion how much work could be out- sourced versus completed in-house. For some tasks it appeared to be more efficient to out- source the work, because special training of staff members would be required that could be applied only once. In other cases, however, outsourcing reduced the possibilities for agency staff members to further develop their own competence in advanced modeling, making the agency more dependent on external support. Overall, the study identified a large number of planning agencies that have implemented or wish to implement advanced transportation models. Although not every detailed method- ology is the right fit for every agency, the planning problems at hand as well as those expected in the near future could guide the selection of the appropriate approach. It was encouraging to see how many agencies are making significant contributions in answering challenging pol- icy questions with advanced travel modeling. Several important themes were identified throughout the report. The most significant is that the motivation and need for advanced models tends to follow the nature of the planning issues faced by the agency. Those agencies focused largely on expanding the capacity of the existing transportation system are inclined to employ modelers in search of the greatest accu- racy possible. Evidence could not be found at this time to demonstrate that advanced models are inherently more accurate or more capable of replicating observed traffic flows than their predecessors. Those agencies focusing on capacity expansion will likely find advanced mod- els of limited appeal. The users and proponents of advanced models, by contrast, reported that the benefits of advanced models are not in the incremental refinement of existing capabilities, but in their ability to answer a range of questions that could not even be asked of traditional models, and to provide a range of performance measures that could not be obtained from traditional mod- els. Agencies dealing with issues of system management, smart growth, pricing, and equity often found themselves compelled to develop advanced models so that they could respond to these issues in a timely and credible manner. Although numerous obstacles were overcome along the way, and even more so with subse- quent implementations, it is fair to say that tour- and activity-based models are a proven tech- nology that can succeed if supported by capable staff with adequate resources. Land use models have been used successfully for policy analyses. Freight and commercial models and dynamic network models are a few steps behind, and do not yet enjoy the same track record of success. They do, however, hold significant promise for those willing to push the practice forward. 4
