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.
58 CHAPTER SIX CASE STUDIES PORTLAND Portland, Oregon, has a long history of innovation in land use and transportation planning, and has consistently been a leader in the development of advanced models. Portland Metro is the MPO for the region, which had an estimated population of 2.2 million in 2008. In the 1990s, the well-known Land Use, Transportation, and Air Quality Connection (LUTRAQ) pro- gram provided the impetus for improved modeling in the region. Still working in a trip-based travel modeling paradigm, Metro developed models based on their 1985 household survey that incorporated land use effects and focused heavily on improving the treatment of non-auto travel. Parallel work on MetroScope, a land use and economic model, began during this time as well. It was expected from the outset that Metro- Scope and the travel model would eventually converge into a single model, a goal that has not been completely achieved, but in which considerable progress has been made. To date, Metro considers bringing land use explicitly into the regional travel modeling as its greatest contribution. Unique features such as legislated urban growth boundaries and transportation plan- ning rules have also influenced the modeling agenda. These and other sustainability issues have resulted in heightened interest in advanced models. Metroâs next steps in advanced modeling began in late 1993, when an expert panel recommended an investment in activity-based models. Metro embraced this view and began working to collect the necessary travel behavior data. This led to its 1994â1995 household travel survey program, which was expanded statewide in the following two years. These data were used for both updating the trip-based models and development of the activity-based model. In 1996, a perfect opportunity emerged to further the activity-based model development, in the form of an FHWA-funded demonstration project for road pricing. Only small changes to the survey were required to use it for estimating both trip-based and activity- based models. A stated preference survey was conducted in conjunction with it to obtain information about likely traveler responses to pricing. Keith Lawton, then manager of modeling at Metro, brought in a team consisting of Greig Harvey, Moshe Ben-Akiva, Cambridge Systematics (including John Bowman), and Mark Bradley to build the model. The model was based on Bow- manâs Masterâs thesis. A prototype of the model was used to evaluate a variable pricing scenario on OR 217 between I-5 in Wilsonville and US-26 in Portland. Preliminary reports were prepared in 1996â1997, and it was intended to adopt the STEP platform (Bowman et al. 1999) as the foundation for further model development. Unfortunately, Greig Harveyâs untimely passing in 1997 slowed progress on this front. The remaining team members built the model into what has since become known as the BowmanâBradley model. The resulting model was tested and used; however, diffi- culties arose with its joint model of destination and mode choice. The resulting elasticities looked excellent; however, the model suffered from a lack of base year calibration. It was slow (computationally heavy) and suffered from edge effects, which further slowed progress in model calibration. Despite these setbacks the team made remarkable progress. Unfortunately, the original funding was exhausted before cal- ibration of the model could be completed, which compounded the conceptual difficulties surrounding the calibration of the joint model. It was decided to use Metroâs four-step sequen- tial travel model to constrain the model, and to pivot the joint model off of its results. This approach yielded reasonable time- of-day changes in the model, lending confidence to its results. The joint model was converted into sequential destination and mode choice models to ease the calibration; however, recalibration was not completed before funding ran out. In 1999, Metro successfully applied to become an early deployment test site for TRANSIMS. Adequate funding and agency support was available for this effort, which allowed Metro to pursue its longstanding interest in network micro- simulation and DTA. It also allowed them to catch up on the supply side to where they had already reached on the demand side. In 2000, the testing focused initially on the router and microsimulation of traffic. A substantial amount of time and effort was devoted to testing the network, which expanded the knowledge about its performance in typical applications and helped guide the development of transit modeling capa- bilities. Its work confirmed that a highly detailed roadway performs no better than more abstract networks typically used with regional models, which makes the approach more tractable for most agencies. Work on the activity-based mod- eling component was carried out in 2003â2004, which made significant strides in destination choice and integration of the demand and supply sides of the model. This first phase of the project was considered a success; however, in 2004, reductions in TRANSIMS funding and unexpected heavy staff demands to support New Starts and Summit analyses
59 for transit proposals forced Metro to curtail work on the second phase (activity-based model development) before the microsimulation of mode choice was undertaken. Although officially dormant, further work on the project has not been undertaken in the past few years. Metro is continuing its evolution of activity-based travel models based on its collective experiences to date. An adap- tation of a Markov decision process formulation developed by Gliebe (2010) is underway. Known as DASH (for Dynamic Activity Simulator for Households), the model is expected to become operational within the next two years. The model is unique in that it assigns roles to individuals, and considers the interdependence of travel decisions among household members. As with many current formulations it treats time as a continuous variable rather than as periods of the day. Equally impressive progress has been made in parallel with MetroScope (Conder and Lawton 2002). Its roots date back to the LUTRAQ work, which served as the impetus for Metro to develop the model. It has been extensively used in planning and policy analyses over the past decade, and is a successful example of a locally developed model similar to the aggregate land use modeling approach developed by Alex Anas in New York City around the same time. The model started in spread- sheet form and has improved over time through linkages to the regional travel model and GIS. The current version is written in the R statistical language. It is now an integrated modeling application, and extensions are being added to calculate green- house gas emissions. More recently, the model has been ported to Salem, Oregon, which has demonstrated its portability. SAN FRANCISCO The SFCTA is in an unusual position as a developer of advanced models in that it is not an MPO organization or state DOT, but a county transportation authority and congestion management agency. The Authority was created in 1989 to administer a local half-cent transportation sales tax. The Metropolitan Planning Commission (MTC), the metro- politan planning agency serving the nine-county San Francisco Bay Area, has a long history of developing and applying trans- portation models. MTCâs existing model (BAYCAST-90) is a fairly standard trip-based model that served SFCTAâs needs throughout the 1990s. When the SFCTA began planning for the Doyle Drive seismic retrofit/replacement, it became clear that the MTC model in place at the time was limited in its ability to evaluate the types of alternatives in consideration for the project. Specifically, there was a strong interest in understanding the travel patterns and congestion effects by time-of-day, and the MTC model at the time was only a peak period/24-h model, and did not assign trips by time-of-day. Further, there was an interest in analyzing the results at a higher level of geographic detail than could be done with the zone system used in the MTC model. Given the desire for greater temporal and spatial resolu- tion, SFCTA chose to develop its own model and hired a con- sultant to do so. During the consultant selection process, one team proposed developing an activity-based model to gain all of the expected benefits of advanced models, specifically an enhanced ability to model time-of-day choices. SFCTA started down this path at a time when no activity-based models were being used in practice in the United States. The decision to push the envelope with a more advanced tool, rather than simply perform a subarea analysis with a trip-based model, can be attributed largely to the executive director of the agency, Jose Louis Moscovich, who pushed, and continues to push, to develop the best analytical tools possible. Having decided to become a pioneer in the modeling world, several factors combined to make the project successful. The first was the presence of an active and able champion within the agency. Joe Castiglione served as the client project man- ager of the model development project. Although that role was important, the more significant effort was in his shep- herding the model through its first few years of successful application. He was able to both learn the details of the model and its implementation and serve as a spokesperson and advo- cate for advanced modeling within the agency. Having some- one successfully play that role permitted the model to be inte- grated into the institutional flow of the agency, allowing the planners and managers at the agency to understand the bene- fits of using the advanced techniques and take advantage of it by asking more sophisticated questions. Further, as a cham- pion in that role, he was able to help the end users understand the limitations of the model and know when to invest in fur- ther refinements. Joe left the agency several years ago, but was replaced by Billy Charlton, an equally committed and qualified champion. A second major factor contributing to the success of the project was an intelligent phasing of the work. The approach was to first implement a basic activity-based model for just San Francisco County. Later, the activity-based model was extended to all nine Bay Area counties. As a third phase, func- tionality was added to better capture the response to pricing and further enhance the time-of-day models. The original implementation was a multi-tiered approach where there was a high level of geographic detail within San Francisco county and larger zones lining up with MTCâs zone system in the other eight Bay Area counties. San Francisco residents were modeled using the newly developed activity- based model. Those resident trips within the county were combined with nonresident trips and inter-county trips from the MTC model before assignment. Keeping the original implementation limited in geographic scope saved signifi- cant effort in model calibration and validation, because the effort did not need to be concerned about the full scale of heterogeneity in the region and complexities of inter-county travel including the bridges, tolls, ferry system, and long- distance commutes. Further, the initial implementation con- sidered only a limited number of tour purposes, and did not
consider such complexities as intra-household interactions, keeping the process as simple and manageable as possible. This approach of keeping the initial model simple was successful, as the model was developed, implemented, and calibrated within a timeframe of about two years. The initial development cost approximately $700,000 in consultant fees. The model was completed in 2001, making it the first activity-based model to be used in practice in the United States. A key aspect to making the project phasing successful, how- ever, is that it resulted in a working product. Because of this, the model could be applied for the Doyle Drive study, and the study could gain the benefits of the modelâs enhanced time-of-day abilities. This Doyle Drive application is actually a third factor contributing to the long-term success of the modelâby achiev- ing an early victory with a model application project, the tech- nical staff was able to build credibility with the management and planning staff at the agency. In turn, this added credibility allowed them the resources to further enhance the model in subsequent phases of work. The SFCTA staff noted that an important corollary lesson here is to make sure that you are spending resources to answer the plannersâ questions. Since 2001, the one-county model has been successfully applied to numerous transportation planning studies. Some examples include: ⢠Congestion Management Program ⢠Countywide Transportation Plan ⢠Yerba Buena Island Ramps Improvement Project ⢠19th Avenue Transportation Plan ⢠Central Freeway Replacement/Octavia Boulevard ⢠Geary Corridor Bus Rapid Transit ⢠Van Ness Avenue Bus Rapid Transit ⢠Market Street Study ⢠MissionâGeneva Neighborhood Transportation Plan ⢠Mission South of Chavez Neighborhood Transportation Plan ⢠Tenderloin/Little Saigon Neighborhood Transportation Plan ⢠Columbus Avenue Neighborhood Transportation Plan ⢠Bi-County Study Update ⢠Caltrain Oakdale Ridership Study ⢠Transbay Terminal Development ⢠Caltrain Electrification ⢠Transit Effectiveness Project ⢠Third Street Light Rail ⢠Central Subway New Starts Analysis. Throughout this broad range of applications, the modeling staff noted that the results consistently appeared reasonable. After careers working with four-step models, they have come to expect unusual results from the model at unexpected times, and were pleasantly surprised by how good the activity-based model results appear. With believable model results, staff noted that the results are easy to explain to planners and at public meetings, and the planning staff has become famil- 60 iar with these responses and come to place more faith in the model results. Staff further noted that the level of detail is impressive, and that for them time-of-day is important. They are not being asked what the volumes are on new freeways, but are instead being asked about pricing, traffic calming, bus rapid transit, and other non-brick-and-mortar projects, for which the model is ideally suited. In 2007, SFCTA received a grant from the FHWA to study congestion pricing within the city. The focus of the resulting MAPS was on evaluating the feasibility of charging vehicles a fee to drive in specific portions of the city when conditions are congested. The fees would seek to either dissuade some motorists from driving or instead entice them to drive in the less congested off-peak periods. Either way, peak period con- gestion would lessen, improving travel times for the remain- ing motorists and for the buses remaining on those routes. Furthermore, revenue could be generated for investment in improved transit service. Given these planning needs, it was clear that to model these congestion pricing alternatives well the model needed to do three things: 1. Appropriately capture travelersâ responses to pricing, 2. Reasonably model travelersâ willingness to shift times- of-day in response to pricing or in response to conges- tion, and 3. Treat San Francisco residents and nonresidents in a con- sistent manner given that a large number of nonresidents would be priced. Although the San Francisco activity-based model pro- vided a good framework to model congestion pricing, several enhancements were needed to do it well. To appropriately capture a travelerâs sensitivity to price, stated preference sur- veys were conducted to observe the sensitivity, and the mod- els were enhanced to use individual values of time within the synthetic population, rather than values specific to aggregate income groups. To capture the time-of-day shifts and peak spreading, the existing time-of-day models were enhanced to include the round-trip mode choice logsum for each time-of- day alternative as a descriptive variable. The mode choice logsum is a composite measure of impedance that weights both the travel time and the cost in a manner consistent with the travelerâs value of time. To treat residents and nonresidents in a consistent manner, the activity-based model was expanded to cover all nine Bay Area counties. The enhancements for the congestion pricing study consti- tuted phase 2 (extend to nine counties) and phase 3 (add func- tionality) of model development work. Again, the phasing was designed such that each phase resulted in a working product that could be used for planning purposes, such that as the plan- ning study progressed with the initial model results, the model- ers were working on the next phase of modeling. These two
61 phases of model development together took about 1.5 years, and cost approximately $250,000. This phasing approach again proved successful and served the needs of the study. This was a final, important lesson learned from the SFCTA modeling workâthat the activity-based model framework is adaptable enough that it can be enhanced to model very com- plicated and specific policies. All of the policies may not have been anticipated at the initial phase of model development; however, the framework is inherently more flexible than an aggregate trip-based model. It is therefore suitable for model- ing things such as congestion pricing and time-of-day shifting that would be very difficult to model in a traditional framework. Since that initial development, the SFCTA has continued to invest in its model. Over a five-year timeframe, it has spent approximately $300,000 in on-call consulting fees and model application assistance. Although the additional costs were not required, the SFCTA saw the value in further refinement and improvements. In addition, the SFCTA has generally maintained two staff positions that have been responsible for all model development work and all model applications. SACRAMENTO Early in this decade, SACOG embarked on an ambitious regional planning effort, which culminated in 2004 with the SACOG Board of Directors adopting the Sacramento Area Blueprint. The Blueprint promotes compact mixed-use devel- opment and more transit choices as an alternative to low- density development. As part of the Blueprint planning process, SACOG took advantage of a number of tools to assist decision makers and stakeholders in understanding the implications of the alter- natives under consideration. I-PLACE3S was used as a sketch- planning tool to assist the decision makers in assembling land use scenarios. Those land use scenarios were fed into the existing trip-based travel demand model to evaluate the transportation effects of such scenarios, and those results were fed into emissions models to understand the air quality effects. Other tools such as three-dimensional visual simu- lations were used to assist in visualizing different levels of density. The Blueprint process was successfully completed using a relatively simple set of analytic tools. It was however a set of tools that was appropriate to the process and available within a timeframe that could allow the planning process to progress on schedule. An important secondary effect of the Blueprint process was that it further established a culture where decision makers and planners need to provide relevant and insightful information, while still acknowledging the limitations of the existing analytical tools. This process helped to establish a relationship of trust with the technical staff, and assisted in integrating the role of technical information into the planning process. There is a key lesson to be learned here: because technical staff was focused on serving the planning needs of the agency to the best of its abilities, the process was successful. Rather than conceding or putting the planning process on hold while they developed a model that could, they sought to make the best of the situation, while acknowledging the limitations of their existing tools. By doing so, they both served the planning process and enhanced their own credibility. Following the completion of Blueprint, SACOG technical staff sought ways to further improve the available tools to better serve the planning process the next time. The model improvements focused on two areas: building an activity- based travel model (SACSIM), and building a land use model. The land use model is still in development; therefore, the remainder of this case study focuses on the activity-based travel model, which is currently being used successfully. Gordon Garry and Bruce Griesenbeck emerged as cham- pions of the new model. They found a high level of support from management and planning in doing so, largely because the goals of the project were to better serve the planning process of the agency, and the decision makers had come to understand the value of good information. Although the sup- port of management is important, it is also crucial to have onboard a technical champion who can communicate to man- agement the value of the product and take full advantage of the modelâs abilities in application. The primary motives for moving to an activity-based travel model were to improve the ability of the model to inform con- gestion pricing policies, and to improve the ability of the model to inform the planning process for Senate Bill 375 (SB 375) and Assembly Bill 32 (AB 32), the California greenhouse gas reduction bills. The latest version of the regional transportation plan includes the option of congestion pricing as a mitigation measure. As staff began to consider modeling this issue, it found in Bain and Plantagie (2004) and Bain and Polakovic (2005) that traditional models have done a poor job of pric- ing analysis for toll roads, even with so-called investment grade forecasts. They explored the possibility of an activity- based model and found that it offered the potential to do a better job of analyzing pricing by modeling individual trav- eler decisions, and potentially distributed values-of-time and disaggregate costs. Further, it offers the possibility of obtain- ing a consistent response to price across all components of the model system, including mode choice, destination choice, time-of-day choice, and tour generation. A second major factor influencing the decision was the pres- ence of SB 375, which is a common concern for all the major California MPOs. SB 375 requires MPOs to consider ways to to achieve greenhouse gas reductions. Therefore, it is important to understand how the built environment affects travel deci- sions, and how the location of households and jobs affects
VMT and emissions. Activity-based models offer an improved ability to do this analysis by eliminating NHB trips. Without NHB trips, the location choices of destinations can be better understood, and all travel can be traced back to individual households, allowing for an analysis of which households generate how many VMT. Also, SACSIM was specifically designed to understand the effects of urban form on travel, which it does by allocating households and jobs to the par- cel level, and modeling parcel-to-parcel travel, while aggre- gating to zones for assignment. This higher level of geo- graphic resolution allows the model to capture smaller-scale land use differences. SACOG completed the development of the activity-based model in 2007. It hired consultants to design and estimate the models, as well as develop software to implement the model. SACOG staff took on much of the model calibration and vali- dation work, running the model and tabulating results while continuing to engage the consultant team in resolving issues that arose during that process. Staff found this approach to be a mixed blessing. It noted that it was an excellent way to learn the behavior of the model system and be a part of the develop- ment process, while at the same time acknowledging that it was a challenge to be engaged in model development while still keeping up with day-to-day responsibilities. Bruce noted that, âIâve never worked as much overtime in my life as when we were calibrating those models.â In the end, they are pleased with the division of labor, because it puts them in a position to take advantage of the full power of the model system. SACSIM was also staged in a way to get a fully operational model up and running in a reasonable amount of time by implementing a model structure that had been used before, and is similar to that used by the SFCTA. The area in which SACOG pushed for new methods was in modeling travel at the parcel level, which was deemed important to understand- ing the travel implications of different land use scenarios. SACOG chose not to push for advanced pricing methods in the initial version, although they are interested in doing so as a next step. Also, SACOG chose to develop a land use model as a separate project, rather than have the first version closely tied to the activity-based model, and did not implement DTA in the first version, even though they have an interest in doing so. By separating out these other challenging components, SACOG was able to get the activity-based model up and run- ning in a period of about two years. This is another example of intelligent phasing of a model project, where the phase constitutes a large chunk of work, but not everything imag- inable or desirable in a model system. By getting the model up and running in this timeframe, it was available for use in the Placer Vineyards project for an early application. This project involved the review of a large proposed green field development that would incor- porate smart-growth concepts into the design. The project was proposed as containing 21,000 new households and 8,000 employees. Owing to concerns about the traffic that would be generated by such a large development, a less dense 62 version was also under consideration that would instead contain 14,000 households and 5,000 jobs. Because all trips could be traced back to the households in the Pacer Vineyards development, the model could be used to track the amount of VMT generated by those specific house- holds. The analysts considered that if this development were to be built less dense, the remaining 7,000 households and 3,000 jobs would go elsewhere in the region. To model this, 7,000 households were created with the same demographic and socioeconomic characteristics as those that would be located in Placer Vineyards. Those households were instead distributed throughout the region. The same was done with jobs. It was important to control for the demographic and socioeconomic characteristics of the households because if they were instead treated with the characteristics of the zones in which they were moved to, differences in income levels or household size could confuse the differences owing to geogra- phy. The model showed that the denser development produced fewer VMT than if the households were located elsewhere, and helped the project gain traction. It is interesting that this application helped the activity- based model gain favor among one stakeholder group in particularâdevelopers who are trying to show the benefit of infill projects. Although infill developments usually are beneficial from a standpoint of reducing automobile travel and greenhouse gasses, infill projects have neighbors, and thus often run into more community resistance than low-density suburban developments. Having a tool that can clearly quan- tify those benefits can help ease the political friction encoun- tered by infill development and help the region as a whole achieve a more sustainable future. A key lesson here for future implementers of advanced models is the value of finding an early success in model appli- cation to help build the credibility of the tool. It also reinforces the value of having one or more champions on the staff who can do a thoughtful job of applying the model to a complicated policy scenario, rather than just adding lanes to the highway network and pushing the ârunâ button. SACOG continues to maintain its old trip-based model for certain applications, but hopes to move away from it in the future. The trip-based model has been used successfully for New Starts analysis, and SACOG wants to maintain that option until the new model makes it successfully through a New Starts submittal process. Doing so is not expected to be a problem, and will involve adapting the procedures developed by SFCTA and the Mid-Ohio Regional Planning Commission (MORPC) to calculate user benefits for input to SUMMIT. Also, the trip-based model is currently used for air-quality conformity analysis. The emissions budgets were developed using that model; therefore, calculating emis- sions with the new model would lead to a result that is some- what inconsistent with the budgets. When the emissions budgets can be updated to be consistent with the new model,
63 the old model will be retired from use. SACOG staff view this approach as a logical way to manage risk with special appli- cations of the new model. They recognize, however, that the new model is capable of handling these tasks, and superior in other ways, and therefore seek to move all applications to the activity-based model in the future. LAKE TAHOE Activity-based models have generally been recognized as the most promising direction in the advancement of travel model- ing practice. However, at present, almost all activity-based models developed and applied in practice have been associated with large urbanized metropolitan areas with populations of 1 million or more (San Francisco, New York, Columbus, Atlanta, Denver, etc.). The model development process for these regions required significant time, budget, and data collection efforts. There is ongoing discussion about the applicability and transferability of activity-based model structures to smaller and less urbanized areas. Such areas actually constitute the majority of the planning organiza- tions in the United States. For the Lake Tahoe region, it was shown that it is possible to successfully transfer and apply an activity-based resident model originally developed for the MORPC in Columbus Ohio. The Lake Tahoe Region is located on the Californiaâ Nevada border between the Sierra Nevada Crest and the Carson Range. Development and urbanization of the basin occurred during and following the 1960 Squaw Valley Winter Olympics. According to the 2000 U.S. Census, the total year- round resident population in the Lake Tahoe Region was 63,448. More recently however it has been estimated that the year-round population has decreased to approximately 54,793, culminating from increasing home values and increases in second homeownership. Although the model was being transferred from a large metropolitan area to a comparatively small, non-urban area, the hypothesis was that the travel behavior of the region was similar enough that the explanatory variables and coefficients would capture the overall behavior. This was found to be true for most of the model components and therefore the transfer- ability of those components, such as tour generation, desti- nation choice, time-of-day choice, etc., was straightforward. Only minor parameter adjustments (i.e., distance coefficients and alternative specific constants) were made owing to regional characteristics and available data. Some of the other unique factors affecting travel in the region such as seasonal residents, the relatively large worker flows into and out of the region, and the seasonal variation of travel were not addressed and had to be added. The advantages to transferring a model are many. First, because the models were based on the MORPC structure time was not spent analyzing the household survey data and figuring out the model structure. Second, setting up estima- tion file sets and doing the estimation, which is a significant part of the model budget, was eliminated. Finally, the limited funds could be spent on other important tasks such as the user interface, the visitor model, and scenario development. The successful development and application of the proto- type model structure developed for MORPC in the Lake Tahoe Region leads to the important conclusion that an activ- ity-based modeling approach can be extended to smaller and less urbanized regions.
