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

Southeast Michigan Council of Governments (SEMCOG) Travel Model Peer Review Report (2011)

3.0 Development of the Southeast Michigan Council of Governments TDFM

3.1 Introduction

This section of the report provides an overview of the development of the SEMCOG TDFM, including a description of the current version of the model, the current uses of the model, the updates made in response to the 2004 TMIP peer review, and the status of updates to the model that are currently underway. Figure 1 shows a chronology of the model versions that have been in use since 2001, when the "E series" of models was introduced, through the current version of the TDFM, Version E5, and the version of the TDFM that is under development, Version E6. If a transition to advanced models does takes place, SEMCOG expects that Version E6 might be one of the final model versions in this generation of trip-based models.

Figure 1: Progression of recent SEMCOG travel model improvements

The figure provides a summary of the progression of recent SEMCOG travel model improvements starting in 2001 and moving forward through 2012.

3.2 History of SEMCOG TDFM Version E5

3.2.1 Development of TDFM Version E1

The documentation for the TDFM [4] discusses the development path that led to Version E5. Starting in 1991, SEMCOG developed a multi-year plan to improve and maintain its TDFM, evaluated the modeling process, and developed a strategy for model improvement and associated data collection. SEMCOG conducted several important surveys, such as the 1994 household survey, the 1996 Detroit Department of Transportation (DDOT) transit on-board survey, the 1996 external station survey, and the 1999 commercial vehicle survey.. Some of the survey results were used to improve SEMCOG's TRANPLAN model.

The TRANPLAN model was principally a highway only model with three modules: trip generation, trip distribution, and highway assignment. The trip generation used cross-classification for trip productions and regression formulae for trip attractions. The trip distribution was a standard gravity model. The highway assignment was an equilibrium 24-hour one period assignment with 10-hour network capacity. Commercial vehicle movement was incorporated into the model trip table by factoring the expanded trips from the 1999 survey.

SEMCOG further improved its travel model in 2001, to create TDFM Version E1. A formal mode choice model and integrated highway and transit networks were introduced. The model platform was transformed from DOS based TRANPLAN to Windows based TransCAD. The project was completed in fall of 2002. As a result, SEMCOG upgraded its travel model from a highway only model to a four-step multi-modal travel demand-forecasting model.

3.2.2 Model Improvements from Version E1 to Version E4

Since the inception of TDFM Version E1 in 2002, SEMCOG implemented several improvements and updates as the model progressed to Version E4

3.3 Current SEMCOG TDFM: TDFM Version E5

SEMCOG TDFM Version E5 is thoroughly described in the model documentation[5]. The status of the model has recently been reviewed by the consultant Parsons Brinckerhoff as part of their engagement by SEMCOG to assess the agency's modeling program. This section of the report summarizes the presentation given by Rick Donnelly of Parsons Brinckerhoff during the peer review meeting.

3.3.1 Overall Structure

The model is a trip-based model which (for modeling person travel) includes the typical four-step model structure. In addition there is a three-step truck model to represent commercial vehicle movements and separate external trip generation and distribution components.

3.3.2 Data Engine

The model uses a collection of survey data and data inventories as its basis. These include national datasets such as NHTS and locally collected data such as:

3.3.3 Trip Generation

The trip generation model uses a standard approach, estimating and balancing trip productions and attractions at a traffic analysis zone (TAZ) level for six trip purposes (home based work, home based shopping, home based school, home based other, non-home based non-work, and non-home based other). It was re-estimated using 2005 survey data.

Productions are estimated using a cross classification model. Home based productions are a function of household size and the number of workers in a household, and auto ownership, while non-home based productions are a function of the number of jobs and households in a zone. Attractions are estimated using linear regression models. Productions and attractions are balanced in a standard manner, with home based work productions balanced to attractions and attractions for other trip purposes balanced to productions.

3.3.4 Trip Distribution

The distribution model uses a gravity model, which again conforms to typical practice. The model is calibrated to trip length frequency distributions using friction factors based on weighted average travel time and K-factors. The model's feedback loop provides updated weighted travel times during each model iteration. SEMCOG noted that the model produces a very good replication of the average trip lengths and trip length distributions and reasonable intrazonal trip percentages, but that it produces large errors in county to county work flows.

3.3.5 External Travel

The representation of external travel is described by Parsons Brinckerhoff as innovative but dated. The generation of external trips uses link types as a surrogate for trip purposes. Generated trips are subtracted from the trip generation model output. Attractions are a function of a zone's distance from the edge of the model region and internal attractions. External trips are distributed using a gravity model that employs friction factors. The external model overestimates trip lengths by between 7% and 62% by trip purpose. The model uses adjusted through trips from the Michigan statewide model. The temporal distribution of external trips is taken directly from counts.

3.3.6 Truck Travel

The representation of truck travel is also described by Parsons Brinckerhoff as innovative but dated. The model estimates travel by all commercial vehicles. The trip generation model was estimated for 248 districts but is applied at a TAZ level. Trips are generated for light, medium, and heavy trucks and are a function of the number of households and jobs, and area. The models contain statistically significant variables but the overall fit is mediocre, which is not uncommon. The truck trip distribution model is also a gravity model that uses friction factors. It reproduces regional average trip lengths and distributions. The time of day distribution is based on the 1999 Commercial Vehicle Survey and not truck counts. Through truck trips are taken from the Michigan statewide model.

3.3.7 Time of Day

This model component is established prior to mode choice, which is relatively uncommon. The time of day model conforms to typical practice. It factors the output from trip distribution by period and by directionality. It was originally based on the 1994 SEMCOG travel survey but has been updated using regional traffic pattern data.

3.3.8 Mode Choice

The mode choice model has a standard nested structure. The model is estimated using relatively aged data: the 1994 home interview survey, 1995 DDOT on-board survey, and the 1996 SMART on-board survey. The mode choice model uses a collapsed set of trip purposes: home based work, non-home based work, and non-home based.

3.3.9 Trip Assignment

The highway trip assignment model is described by Parsons Brinckerhoff as best practice. Highway assignment uses multi-class static user equilibrium. The classes include single occupancy vehicles, high occupancy vehicles with two occupants, high occupancy vehicles with three or more occupants, light trucks, medium trucks, and heavy trucks. The volume delay function is a Bureau of Public Roads function adjusted to match Highway Capacity Manual 2000 speed flow relationships. The generalized cost equation used to calculate the shortest path is a function of travel times, congestion, link length, and assumptions on value of time and vehicle operating costs.

Transit assignment conforms to typical practice. The model uses the TransCAD Pathfinder, with a generalized cost that is a function of travel time, wait time, and fare. Walk and auto access are available on both ends of a trip.

The feedback in the model uses Sheffi's MSA methodology, which is deemed best practice by Parsons Brinckerhoff.

3.3.10 Validation Summaries

The model produces a set of validation summaries covering both highway and transit outputs. For highway outputs, these include highway volumes by period and total daily results for various regions within the model area and for external stations, segmented by functional class and volume range. For transit outputs, these include boardings by line and daily trips by service provider.

3.4 SEMCOG 2004 TMIP Peer Review

SEMCOG's first TMIP peer review took place in 2004. The 2004 TMIP peer review report[6] summarizes the results of the two-day peer review meeting. During the 2004 TMIP peer review, SEMCOG requested that the peer review panelists examine its existing model and the agency's plans for future model improvement and enhancements, as recommended by its consultant (Cambridge Systematics, Inc.). SEMCOG asked for assistance with prioritizing near-term and long-term model enhancements and also for recommendations for transitioning to more advanced travel demand modeling methodologies.

After a day of SEMCOG presentations on the current travel demand model and plans for model improvements, the peer review panel met in private to discuss the model and make recommendations for model enhancements. The peer review panel felt that the existing model represented the "state of the practice." The panel felt that the model addressed time-of-day, commercial vehicle, and external trips particularly well.

The 2004 TMIP peer review report documents 20 specific recommendations, which SEMCOG has been working to address since 2004. Their status is as follows (the numbers in parentheses refer to the recommendation numbering using in the 2004 TMIP peer review report):


In progress

Not addressed

3.5 Status of SEMCOG TDFM Version E6

As outlined above, there are a set of model improvement recommendations made during the 2004 TMIP peer review that SEMCOG is currently working to address. SEMCOG's consultant, Cambridge Systematics, Inc., is currently developing a new version of the SEMCOG TDFM, Version E6, which will include those improvements. This section of the report summarizes the presentation describing the status of Version E6 given by Liyang Feng of SEMCOG and Tom Rossi of Cambridge Systematics, Inc. during the peer review meeting. In addition to implementing several of the remaining recommendations from the 2004 TMIP peer review, SEMCOG intends the transition to Version E6 to improve model components as needed to analyze key projects and policies and to reflect the most recent available data. The update work includes model estimation, application programming in TransCAD, and validation at a model component level. At the time of the peer review meeting, work was underway on each of the model components, with some being estimated and others awaiting final validation.

3.5.1 Data Sources

The model updates that are taking place rely on several data sources:

3.5.2 Trip Generation

Several elements of the trip generation model are being updated. The trip generation rates are being updated using the latest household survey data. Income segmentation has been introduced for some trip purposes (home based work, home based shopping, and home based other), primarily for environmental justice analysis. The home based school purpose was found to be insensitive to income and is instead a function of household size and the number of children in a household. Home based university was added as a trip purpose, with trip rates per person to the 25 largest colleges by type and distance. Home based university attractions are based on total enrollment minus group quarters population. Other updates include reclassifying employment types, adding factors to separate non-motorized travel, and updating the air passenger model. An area type adjustment factor that distinguishes between rural and non-rural areas has been calibrated, and the trip generation model is now awaiting final validation.

3.5.3 Trip Distribution

The gravity model parameters have been recalibrated by trip purpose to include the income segmentation used in the home based work, home based shopping, and home based other trip purposes. In addition, a destination choice model using a logit model form is to be estimated. This will be estimated using the most recent survey data and will be compared with the existing gravity model to test whether the destination choice model produces better results. If that proves to be the case, the destination choice model will be implemented; if not, the existing gravity model will be retained and validated using the model recent survey data.

3.5.4 Time of Day

New time periods have been defined that are useful for both highway and transit analysis. The day has been divided into five time periods and time of day factors have been re-estimated using the household survey data.

3.5.5 Transit Model

Work on the transit model has focused on the transit network parameters and the path building processes. The review of transit network parameters has considered travel times, fares, maximum access time, bus speeds, transfer rules, and mode choice related parameters. The updates have made use of new on-board survey data that allowed comparison between observed and modeled paths and supported adjustments to the path building settings to improve the match. SEMCOG compared modeled auto time and 2010 scheduled bus times for 145 routes operated by AATA, DDOT, and SMART, and used that analysis to adjust the relationship between bus travel times and modeled auto travel times. The work included developing operator specific dwell time assumptions. Transit walk access time was capped at 18 minutes in TDFM Version E5. Based on the on-board survey data, this has been increased to 36 minutes.

3.5.6 Mode Choice

The existing mode choice is being re-evaluated for several reasons:

Several nesting structures are currently being explored during model estimation. The revised mode choice model is intended to support the evaluation of new modes including Arterial Rapid Transit (ART), Bus Rapid Transit (BRT), Light Rail (LRT) (including on Woodward), and Commuter Rail (CRT) (from Detroit to Ann Arbor).

3.5.7 Commercial Vehicle Model

The commercial vehicle model is a three step model, with trip generation, trip distribution, and assignment steps. During this update, vehicle classification count data has been adjusted to account for growth and decline in the region, model parameters have been updated to reflect current data, and adjustments have been made to reflect changes in external station volumes.

3.5.8 System Calibration

The final part of the model update process will be to complete a system calibration. Individual model components are being validated as they are developed. Recent data is being used to see "what has changed" to enhance the short term forecasting capability. The overall aims of the calibration are to get the big picture correct, examine trouble spots from previous model versions, and ensure that forecasts are reasonable. The calibration work is expected to be complete in March 2012.

[4] Travel Demand Forecast Model: Version E5, June 2010, available at

[5] Travel Demand Forecast Model: Version E5, June 2010, available at

[6] Summary Report of the Peer Review Panel for the Southeast Michigan Council of Governments Travel Model Improvement Effort, available at

Updated: 04/28/2014
HEP Home Planning Environment Real Estate
Federal Highway Administration | 1200 New Jersey Avenue, SE | Washington, DC 20590 | 202-366-4000