Skip to content U.S. Department of Transportation/Federal Highway AdministrationU.S. Department of Transportation/Federal Highway Administration

Office of Planning, Environment, & Realty (HEP)
Planning · Environment · Real Estate

HEP Events Guidance Publications Glossary Awards Contacts

San Francisco County Transportation Authority (SFCTA) Travel Model Peer Review Report

2.0 Development of the San Francisco County Transportation Authority DTA Model

2.1 Introduction

This section of the report provides an overview of the development of the SFCTA DTA model, including a description of the current version of the model, its uses and latest updates. There is also a brief overview of SFCTA's travel demand model to provide some background. Towards the end, this section describes SFCTA's goals for the peer review.

The Authority's official travel demand forecasting tool is called San Francisco Chained Activity Modeling Process (known as SF-CHAMP). It is a state-of-the-art activity-based model that can be used to assess the impacts of land use, socioeconomic, and transportation system changes on the performance of the local transportation system. SF-CHAMP is different from traditional four-step model as it is tour-based and not trip-based. A tour is a chain of trips made by an individual that begins and ends at home with intermediate stops at locations other than home, whereas a trip is a single movement from an origin to a destination. The model structure as such is more complex and is sensitive to a broader set of factors that influence travelers' choices. This tool is used for many Authority planning studies and projects.

In the recent past, SFCTA has been using Dynamic Traffic Assignment (DTA) instead of the traditional Static User Equilibrium (SUE) assignment technique for a few projects in the northwestern section of the city. DTA provides planners with a better understanding of the how-and-why of traffic routing itself around San Francisco. DTA aims to represent the interaction between a time-varying network and demand in a behaviorally sound manner. In static assignment, the congestion or performance properties of a link in a network are described by a volume-delay function (VDF) which provides the relationship between the average or steady-state travel time on the link and the traffic volume of the link. The demand is loaded and routed on a set of shortest paths. Static assignment results in a volume-capacity (V/C) ratio for each link whose value may be well over one for congested links. On the other hand, DTA typically loads the demand as individual vehicles or packets of vehicles on shortest paths, consistent with the fundamental diagram of traffic flow and thus accounting for congestion and spillbacks. Overall, DTA gives planners a more fine-grained view of transportation system performance. Recently, the Authority obtained a grant from the Federal Highway Administration (FHWA) to expand the DTA model to cover the rest of San Francisco under a project nicknamed "DTA Anyway."

2.2 History of San Francisco County Transportation Authority DTA Model

San Francisco's tour-based travel demand model, SF-CHAMP, is capable of predicting precise changes in travel behavior in response to policies such as pricing but is limited in its ability to predict responses to operational improvements due to the use of SUE traffic assignment. Hence, to analyze strategies such as transit signal priority and network geometry changes, the city used traffic micro-simulation. However, linking the macro-behavioral sensitivity of demand models with the car-following behavioral sensitivity of traffic micro-simulation models has its own issues due to various reasons:

  1. The demand from a static traffic assignment model (or SUE assignment) can be high and simply feeding this into a microsimulation model may lead to unrealistic traffic congestion. This is because the demand level in the static traffic assignment model is calibrated using its own BPR-type volume-delay function, which is significantly different from the car following and meso-scopic traffic flow models used in DTA. The demand from static assignment does not represent real capacity constraints that should in reality result in switching to other modes, times of day, destinations, or routes. This often leads to the demand being reduced in an ad-hoc manner just to get the microsimulation model to run and produce results.
  2. The SUE assignment cannot represent operational improvements that will lead to induced demand or the return of latent demand. Thus, the demand fed into the traffic micro-simulation model does not increase enough and the benefits of these operational improvements are often over-predicted.
  3. When analyzing a subarea of the region, the context of the larger trip gets lost. In the case of an auto capacity-reducing project, many of the impacts are felt outside the modeled corridor and in the case of a grid network can be spread across the whole city. Most traffic micro-simulation models are only built for the limited corridor where the change occurs, and are not able to quantify the impacts to streets beyond it. It should however be noted that this is also an issue with subarea travel demand models in general.

The Authority needed a tool that could make robust predictions of the changes in traffic flow as a result of policy or network changes in order to evaluate the effects of various transportation projects being considered. Similar to the way that SF-CHAMP travel demand forecasting model provides a very detailed understanding of the intricate travel demand decisions of individuals for the entire region, DTA models are capable of providing insights into the behavior and flow of traffic and transit vehicles. DTA models incorporate transportation system performance details such as traffic signal timing, queue formation, and route choice decisions - important considerations when analyzing projects in San Francisco. DTA could be robustly used to measure both the local impacts of large, regional projects and also the regional impacts of local projects.

Based on its immediate needs, SFCTA developed a DTA network for the northwestern quadrant of the city. INRO's Dynameq software was chosen as the DTA package since it had a mature user interface and it also provided a highly detailed network representation including lane-based delays and explicit representation of transit vehicles and schedules. The DTA model was used for evaluating the following two projects and questions:

The following issues and observations arose from the use of DTA in the two projects specified above:

To gain further insight into network supply, the Federal Highway Administration (FHWA) awarded the Authority a grant via a Broad Agency Announcement (BAA) to implement dynamic traffic assignment in San Francisco, and use the DTA for analyzing proposed transit improvements, analyzing traffic diversions caused by those transit improvements, and comparing the effectiveness of roadway pricing alternatives.

2.3 Current San Francisco County Transportation Authority DTA Model

SFCTA's initial objective of the project was to have a working DTA model with results that make sense for the PM Peak period in San Francisco. Other key goals included:

The following items encapsulate the basic approach of the model development process:

It is about one year into the project now and SFCTA has developed an automated process that closely knits SF-CHAMP and the DTA model. The code, developed in Python, is capable of executing the following tasks:

The code base is named "DTA Anyway" and is hosted on Google and is open to all for download. All information about the project and its current status can be found at: http://code.google.com/p/dta/. Specific documentation on the DTA data preparation process code and APIs can be found at: http://dta.googlecode.com/git-history/dev/doc/_build/html/index.html#.

The DTA model covers the whole of San Francisco and has 976 TAZs with 22 external stations. As mentioned earlier, the model was developed in INRO's Dynameq software platform. The network has 1,115 signals and 3,726 stop controlled intersections. The DTA model was developed and calibrated for the PM peak period which is from 4:30 PM to 6:30 PM. A 1-hour warm-up time was used along with a 3-hour network clearing time. Due to the 1-hour warm-up period, demand from 3:30 PM to 6:30 PM is loaded at a uniform rate onto the network. The simulation period end 3 hours later at 9:30 PM. This demand includes 385,000 auto and 65,000 truck trips approximately. There are about 270,000 internal trips and about 180,000 trips involving external TAZs.

In adhering to the usage of as much actual data as possible in the development approach, the modeling team conducted a traffic flow survey to measure traffic flow parameters such as free flow speed, jam density, saturation flow rate, backwards wave speed, and driver response time. The parameters derived from the survey were reconciled with default parameters and parameters obtained from existing local traffic data sources such as Caltrans Performance Measurement System (PeMS) and San Francisco Municipal Transportation Agency (SFMTA) Speed Surveys data. Table 1 shows the various data sources used for the required traffic flow parameters.

Table 1: Data Sources for Parameters by Facility Type

Parameters\Facility type

Free-flow Speed

Saturation Flow

Response Time

Jam Density

Freeway

PeMS

PeMS

PeMS

Inferred from CBD arterials

Arterial

SFMTA speed surveys

CBD saturation headway observations

CBD queue dissipation observations

CBD arterial queue length observations

Local & Collector

Limited SFMTA speed surveys and supplemental observations

Mostly inferred from CBD arterials

Mostly inferred from CBD arterials

Mostly inferred from CBD arterials

The calibration process for the DTA model involved iteratively fixing network and supply issues in addition to making defensible adjustments to the model parameters. The following are some of the steps taken by SFCTA during calibration of the DTA model:

At the time of the peer review meeting, the modeling team at SFCTA felt that the model was reasonably calibrated and noted the following from the calibration exercise:

Further details about the model development, calibration, and validation process are provided in Appendix D of this document and the project webpage (http://code.google.com/p/dta/).

2.4 San Francisco County Transportation Authority Goals for Peer Review

SFCTA is the designated Congestion Management Agency of San Francisco County and is responsible for developing and maintaining San Francisco's official travel demand model: SF-CHAMP (San Francisco's Chained Activity Modeling Process). Over the past decade, the character of questions posed by San Francisco planners and decision-makers has shifted from "where should we add capacity?" to more nuanced questions revolving around managing capacity among users and modes. The SF-CHAMP model is an advanced activity-based travel demand model that analyzes travel behavior decisions across multiple dimensions, and is capable of evaluating a multitude of policies and investments based on how these policies and investments change various attributes of travel between two points in the region for a given time of day. However, the macroscopic static user equilibrium model currently used within SF-CHAMP needs improvement with respect to the nuances of congestion in the city. These needs resulted in the award of an FHWA grant to implement a DTA model in the city.

The tools developed as a part of this research project are open source, and the findings and lessons learned will be publicly available and open for practitioner discussion and research. Understanding the new territory that the Authority would be charting, the research proposal included a peer review panel partway through the project. The project team has identified the following areas of questioning and discussion with the peer review panel:

While the Authority and our research team have some experience in each of the above topics, they see this peer review panel as an opportunity to learn about strategies and techniques being employed in other areas across the country and world. Furthermore, they would like to understand how to make their investments in documentation and code more useful to the community at large.

2.5 Previous Peer Reviews

To the knowledge of this panel SFCTA has not previously held a formal TMIP DTA model peer review.

Updated: 03/28/2014
HEP Home Planning Environment Real Estate
Federal Highway Administration | 1200 New Jersey Avenue, SE | Washington, DC 20590 | 202-366-4000