Development and Application of a Dynamic Traffic Assignment Model for San Francisco
Gregory D. Erhardt, Parsons Brinckerhoff, Elizabeth Sall, Lisa Zorn, and Dan Tischler, San Francisco County Transportation Authority, Renee Alsup and Neema Nassir, Parsons Brinckerhoff and University of Arizona
San Francisco recently completed the development of a citywide dynamic traffic assignment (DTA) model. The approach combines a mesoscopic DTA model with an explicit representation of every hill, street, traffic signal, stop sign, and transit route in the city. The model was found to be highly sensitive to the details of intersection coding and signal timing, thus a key challenge was ensuring the accuracy and reliability of all model inputs. This challenge was met by developing an open-source toolkit to automate the conversion of networks and demand from the static assignment model and other existing data sources, minimizing the potential for human error in the application process. A second advantage to this approach is that consistent scenarios can be run through the existing activity-based demand model, and the resulting change in demand can be readily fed into the DTA model. The model was applied to forecast alternative scenarios, including a bus-rapid transit scenario and a cordon pricing scenario, and shown to produce very different responses than the comparable static assignment model. This presentation and the accompanying paper will focus on San Francisco’s experience developing and applying the DTA model, on lessons learned from the process, and on recommendations for future improvements.
For over a decade, the San Francisco County Transportation Authority (the Authority) has used an activity-based travel model, known as SF-CHAMP, for project evaluation. While SF-CHAMP has generally served the agency’s needs well, the lack of realism in the static assignment model it uses has been limiting for certain applications. Therefore, the Authority sought to develop a citywide dynamic traffic assignment (DTA) model.
At a project-level, planners have found DTA to be a useful post-process to static traffic assignment to (1) understand the effect of projects that can be measured at the mesoscopic level, and (2) serve as a tidier transition from static traffic assignment into a traffic microsimulation model. At a higher level, planners are interested in incorporating DTA’s ability to more accurately evaluate reliability, transit/auto interactions, and operational treatments into an integrated modeling framework that is capable of evaluating changes in travel behavior (an integrated activity-based demand model and dynamic network model).
This presentation details the development process and results from the Authority’s experience building and maintaining a citywide DTA model. A key challenge in developing a DTA model is that a mesoscopic simulation of a grid network is highly sensitive to network coding and decisions as small as how centroid connectors are placed. As such, the team collaboratively developed a toolset in order to accurately capture, manage, and translate information about every traffic signal, stop sign, roadway, and transit vehicle operating in the City of San Francisco. As much as possible, real data (as opposed to imputed data) was used to estimate traffic flow parameters. San Francisco’s hilly topography necessitated extensive field work in order to achieve this data-driven traffic flow parameter estimation. A series of scripts create each DTA scenario from the SF-CHAMP demand model inputs and outputs. Network geometry and demand issues found during calibration were addressed “at their source” either at the top of the model chain (i.e. road geometry) or where they appeared (i.e. destination choice models). This process made SF-CHAMP more accurate and ensures behavioral consistency between the demand and network models.
The DTA team had a second goal: to share the tools developed for the process with other cities and regions that might be interested in developing a DTA model. As a result, the software tools developed for this project to facilitate the creation of DTA networks and scenarios are available as an open source project called DTA Anyway (http://dta.googlecode.com). The development team designed the API before implementation with two primary objectives: making it (1) easy to use and (2) extensible. DTA Anyway consists of an object-oriented python library with objects that represent the dynamic network, nodes, links, movements, etc. While the San Francisco static network is in the Citilabs Cube format and the output dynamic network is in the INRO Dynameq format, the library design attempts to decouple the read/write methods from the data structures as much as possible, for easier future extension to other static and dynamic network formats.
Upon completion of the DTA model, it was used to evaluate hypothetical scenarios of interest to the Authority. The test applications run include a bus rapid transit (BRT) scenario in which a highway lane is remove to dedicate to busses, and a congestion pricing scenario in which vehicles are assessed a cordon charge to enter the downtown area. These scenarios show that the DTA model produces very different (and generally more realistic) patterns of traffic diversion than the static assignment model. In extreme scenarios, static models may allow much more volume on a link than it could possibly handle, while the DTA model will recognize that traffic will begin queuing and will shift to alternate paths. In addition to distributing flow changes more broadly, the DTA model predicts a more consistent response across the length of a roadway than the static model.
This presentation will include some of the lessons learned by the DTA Anyway team, and make recommendations for future developers to DTA models to maximize the efficiency and effectiveness of their work. These topics will include: recommended software practices, approaches to extracting demand matrices, challenges encountered in model calibration, and implications of the scenario test results for future model applications.
This work was completed collaboratively by Gregory D. Erhardt and Renee Alsup from Parsons Brinckerhoff, Elizabeth Sall, Lisa Zorn and Dan Tischler from the San Francisco County Transportation Authority, and Neema Nassir from the University of Arizona.
Association for European Transport