PDF files can be viewed with the Acrobat® Reader®
|Location||Date||Peer Review Host Agency||Peer Review Host Agency Contact||Peer Review Panelists|
|Detroit, Michigan||December 6-7, 2004||Southeast Michigan Council of Governments (SEMCOG)||Stephanie J. Taylor, SEMCOG, firstname.lastname@example.org||Ken Cervenka (Chair), North Central Texas Council of Governments (NCTCOG)|
|William A. Davidson, PBConsult, Inc.|
|Chris Forinash, U.S. Environmental Protection Agency (EPA)|
|Bill McFarlane, San Diego Association of Governments (SANDAG)|
|Mark Schlappi, Maricopa Association of Governments (MAG)|
Travel Model Improvement Program (TMIP)
Federal Highway Administration
Volpe National Transportation Systems Center
U.S. Department of Transportation
The content of this peer review report does not represent the opinions of FHWA nor does it constitute an endorsement, recommendation or specification by FHWA. The content of the report does not determine or advocate a policy decision/directive or make specific recommendations regarding future research initiatives. The report is based solely on discussions and comments during the peer review.
This report summarizes the results of the two-day peer review meeting conducted at the request of the Southeast Michigan Council of Governments (SEMCOG) through the Travel Model Improvement Program (TMIP) sponsored by the Federal Highway Administration (FHWA). SEMCOG requested that the panelists assess the current travel demand model and recommend both near-term and long-term model enhancements.
SEMCOG has a relatively new TransCAD-based four-step travel demand model. The model includes consideration of external, commercial vehicle, and transit trips. It relies on a variety of data sources including census information, a SEMCOG household travel survey, traffic counts, and a transit-on-board survey. 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 current model represents the "state of the practice." The panel felt that the model addresses time-of-day, commercial vehicle, and external trips particularly well.
For future model enhancements, the panel's recommendations include:
Several of these recommendations are included in SEMCOG's existing plans for improving its travel demand modeling.
The Southeast Michigan Council of Governments (SEMCOG) is the metropolitan planning organization (MPO) for the seven-county southeast Michigan region, encompassing the Detroit metropolitan area. The agency is responsible for developing the federally mandated long-range regional transportation plan (RTP) and the transportation improvement program (TIP) for the region. SEMCOG is also responsible for measuring and documenting on-road mobile source air pollutant emissions for both the air quality state implementation plan (SIP) and the regional air quality conformity analysis. The travel demand forecast model (TDFM) and its underlying theory is a key support tool for this core work.
Travel model operation and maintenance is critical to the quantitative forecasting methods that form the sound technical base for SEMCOG planning and policy decisions. Engineers and planners use the model results for traffic studies, master plan updates, and major transportation projects. The model is also important to local transit operators as it allows them to forecast the potential demand for various transit alternatives.
Over the past two years, SEMCOG has transformed its travel model from a DOS-based TRANPLAN platform to a Windows-based TransCAD® platform. The new model includes a transit model, a time-of-day (TOD) process, a multi-class highway assignment, and a feedback loop. The model was updated using 2000 as the base year. SEMCOG is currently using model Edition 3. The TransCAD highway network is essentially a translated TRANPLAN file with limited geo-location corrections. The new network development activities are based on the Michigan Geographic Framework (MGF). At this time, SEMCOG is updating its multi-year TDFM improvement plan and requested this peer review to assist them in this process.
A major challenge for the model is handling the movement of people and goods across jurisdictional boundaries. SEMCOG works with other transportation and planning agencies in the region to ensure that a truly regional planning process is always maintained. SEMCOG continues to look for ways to maintain an interface with the two regional sub-area models, Washtenaw Area Transportation Study and the St. Clair County Transportation Study, as well as the Michigan Department of Transportation's (MDOT) statewide travel demand forecast model.
SEMCOG's data collection efforts include several surveys conducted over the years with the assistance of various agencies: a 1994 household survey, a 1996 external survey, a 1998 Detroit DOT (DDOT) transit on-board survey, a 1999 commercial vehicle survey, and a 2002 regional transit on-board survey.
Having both roadway and transit forecasting capabilities allows for a comprehensive analysis of transportation alternatives, both in corridor-level studies and in the development of the RTP. Since the inception of the TransCAD model in 2002, SEMCOG has started several projects to further improve its travel model, including:
Over the long term, SEMCOG might consider advanced modeling techniques such as tour- or activity-based modeling.
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 the panel to assess their existing model and to help them prioritize near-term and long-term model enhancements. They also wanted recommendations for possible applications of more advanced travel demand modeling methodologies.
The SEMCOG planning region includes the following seven counties: Livingston, Macomb, Monroe, Oakland, St. Clair, Washtenaw, and Wayne. In the year 2000, the region had 4.83 million people, 1.85 million households, and 2.16 million employees.
In designing their new travel demand model, SEMCOG's objectives were to:
The model uses three types of data:
The new SEMCOG TransCAD model system is a traditional four-step, person trip-based model. The model was calibrated for a base year of 2000. Figure 1 shows the structure for the new SEMCOG travel model system. A description of the individual components follows.
The model area consists of the entire SEMCOG planning region. The model's zone system includes 1,442 internal zones and 63 external stations. The external stations are used to reflect trips that have at least one trip end outside the modeled area. This is the same zone system used with the original TRANPLAN model.
The model consists of a highway and transit network. When the previous TRANPLAN highway network was converted to a TransCAD network, the following changes were made:
The new TransCAD transit network is based on the highway network. It considers the maximum walk distances to and from transit stops. The transit network also considers transit park-and-ride lot nodes that serve as intermediate destinations for automobiles to park prior to the drivers and passengers boarding a transit line. The bus speeds used in the transit network are related to the highway network.
Figure 1: Structure of the SEMCOG Model
Four Steps of the Travel Model
The trip generation and distribution models of the new TransCAD model remain unchanged from the previous TRANPLAN model. The four steps are consistent with the traditional standard practice urban transportation planning system model. All the model steps have been validated, although the highway assignment will be subject to additional comparison checks as more traffic counts become available. The documentation for the model has been completed. The model has the following components:
The SEMCOG highway assignment uses a multi-class user equilibrium process. Trip tables are assigned for each time period for six vehicle types:
Speed flow curves were derived from a modified BPR approach. The highway assignment was validated using traffic counts.
The transit passenger trip tables are assigned using the TransCAD Pathfinder procedure. Walk access trips are subject to a maximum walk time of 18 minutes and auto access trips must go through a park-and-ride lot node.
Truck Travel Model
The truck modeling process creates trip tables for light, medium, and heavy trucks. It also creates TOD trip tables for external-internal and external-external truck trips. The data sources for the truck model development are Michigan's statewide travel model for external trips and the SEMCOG commercial vehicle survey (CVS) for internal trips. The internal trip generation is based on the CVS-developed trip rates applied to the socioeconomic data. The internal trip distribution is based on a gravity model formulation. External trips are obtained from the Michigan statewide travel model trip tables.
The TOD factors are derived from the household survey and applied after trip distribution. Time of day is modeled using four time periods:
The model feeds the highway assignment travel times back to the trip distribution module. The travel time skims are weighted for each purpose. The model uses the method of successive averages for feedback. Determination of convergence is based on changes in link volumes from run to run.
Since the first version of the TransCAD model, SEMCOG has updated the various model components. This update model is named SEMCOG Model E3. Most notably, the highway network was changed so that it is now fully compatible with MGF version 2.0. The network development is based on the MGF, which is the highway network translated from TRANPLAN. The consolidated highway network improved communication between databases and increased the number of centroid connectors from 1.6 to 1.8 per TAZ. Another improvement was the complete re-coding and re-development of the transit network based on the year 2000 published schedules from local transit providers.
The updated model also considered TOD calibration, mode choice model adjustment, external model adjustment, truck model adjustment, and highway assignment validation. It adopted the SEMCOG 2030 regional development forecast and used a zone system with 1442 internal TAZs and 63 external stations based on 1990 census zonal definitions. The individual calibrations of the updated model are:
For the SEMCOG planning region, the base year 2000 model run indicated approximately 133.3 million vehicle miles of weekday travel. The total number of weekday internal trips is 15.83 million and the total number of weekday linked transit trips is 182.8 thousand. The overall transit mode share is 1.1 percent.
The model improvement plan aims to evaluate SEMCOG's modeling needs and develop a list of necessary model improvements. The scope includes developing a multi-year model improvement plan and identification of data needs.
The model improvement plan includes a review of a number of issues relating to data collection, travel demand forecast modeling, and policy considerations.
Data collection is necessary for input and adjustment to the steps in the TDFM. One major data collection effort, the 2004-2005 Michigan statewide household survey, is near completion. In addition, SEMCOG collects data every year for its regional traffic count database. Once gaps in these databases are filled, model validation capabilities will improve. New census data will also be examined, as will information on performance measures and cross border travel.
The identification of short- and long-term needs is also a key priority. By identifying these needs, the improvement plan can help SEMCOG determine the appropriate balance between implementation of advanced techniques and the improvement of the basic model components.
On the policy side, changes in air quality conformity standards require that additional counties be included in SEMCOG's air quality conformity analysis. This requires a review of SEMCOG's air quality modeling-TDFM interface. Consequently, SEMCOG continues to maintain coordination with MDOT statewide modeling efforts. Additionally, SEMCOG is considering how to manage modeling issues such as high occupancy vehicle modeling and toll modeling. Other issues include development of the SEMCOG Year 2035 Regional Transportation Plan in 2009; project planning; microsimulation of households, trips, and traffic; intelligent transportation system applications; and staffing and resource availability.
The model improvement plan can be divided into short-term and medium-term modeling needs and long-term modeling possibilities.
Short-Term Modeling Needs
Medium-Term Modeling Needs
Long-Term Modeling Possibilities
As part of its continuing effort to update and improve its travel demand forecasting capabilities, SEMCOG has undertaken the following tasks:
The improvement plan has identified peer MPOs for SEMCOG and obtained documentation on their models. SEMCOG and its consultant have reviewed and documented the existing SEMCOG model and prepared a technical memo on modeling needs. SEMCOG and its consultants intend to complete the comparisons with peer MPOs and develop recommendations and a multi-year model refinement plan that incorporates the comments of the peer review panel.
Currently, the SEMCOG region and the contiguous county of Lenawee are in marginal nonattainment status for eight-hour ozone levels. Two of the seven counties in the SEMCOG region are in preliminary non-attainment status for fine particulate matter. SEMCOG is awaiting a final determination from the Environmental Protection Agency on these counties. Other air quality concerns exist in Wayne, Oakland and Macomb, which are in "maintenance" status for carbon monoxide.
The travel model inputs to the air quality model include all link-level data. The inputs are:
The local inputs to the EPA MOBILE6 model include age distribution of the light-duty vehicle fleet, local fuel sulfur data (gas and diesel), reid vapor pressure, and minimum and maximum temperatures. The model calculates emission factors for every two mile per hour (mph) speed increment from 5-65 mph. It calculates separate factors for freeways, arterials, and local roads. The air quality model post-processor calculates the seasonal VMT adjustment, temporal VMT distribution, vehicle type mix by hour and functional class, and the link-specific speed by hour and direction.
The panel was charged with two primary tasks:
On the first day of the meeting, SEMCOG and its consultants presented the status of the model and the model development plans. Following the presentation and subsequent discussions, the panelists convened in private to discuss the SEMCOG's model and identify recommendations for model development. The second day of the meeting was devoted to presenting and discussing these recommendations. The remainder of this section summarizes the panel's findings and recommendations presented to SEMCOG staff and representatives from other agencies (listed in Appendix II).
The panelists agreed that a four-step TransCAD-based "state-of-the-practice" model is now in place and ready for use on many transportation planning applications projects. The panel felt that the model addresses time-of-day, commercial vehicle, and external trips particularly well.
The panelists also agreed that there is room for improvement in the existing four-step model, and endorsed SEMCOG's desires to move forward with a multi-year travel demand forecast model improvement plan. Three areas for improvement were specifically noted:
The panel recognized that SEMCOG is already making progress on several data improvements. The 2004-2005 household survey data will be useful in improving the trip generation, trip distribution, mode choice, and TOD components of the model. SEMCOG must use this data to replace the 1994 household survey data carried over during the model translation. Census data will be updated from the 1990 data currently in use to data from the 2000 census. Maintaining and updating the agency's extensive traffic count database will provide a sound baseline for comparison against the travel model volumes and improve TDFM calibration and validation.
The transit mode choice model will be enhanced by incorporating data from the comprehensive 2002 transit on-board survey. Recent revisions to the transit network have improved network connectivity and representation of more of the local streets used by transit providers. Additional network editing will further enhance transit networks to reflect actual route configurations in model simulation.
By using a subset of the MGF base map to create the modeling network, the network geo-location and positional accuracy has already been improved. As refined versions of MGF become available, SEMCOG anticipates maintaining this compatibility through incremental updates. SEMCOG's GIS, planning applications, and alternatives analysis require this compatibility for data sharing. With a commitment toward integrated networks, SEMCOG is poised to make highway and transit networks fully integrated to allow more advanced planning analysis.
The panel's general recommendation is for SEMCOG to keep its near-term focus on important four-step model improvements and network and data enhancement activities that will keep the model's capability consistent with the model's primary purposes. However, SEMCOG must also recognize that the state of the practice for travel demand modeling is moving towards activity-based microsimulation modeling, and SEMCOG may eventually want to consider this type of modeling. Most of the panel's recommendations are useful for the current four-step model, but also provide a solid foundation for a future activity-based model implementation.
The following section summarizes the enhancements recommended by the panel. Unless otherwise stated, the panel was unanimous on its recommendations.
Although the original intent was to prioritize all recommendations, the panel felt SEMCOG staff could better perform this task since they have a better sense of their resource constraints. SEMCOG must now determine how to incorporate the following 20 recommendations into its multi-year travel demand forecast model improvement plan. The recommendations are organized under five categories: data collection/networks; model improvements; model validation and verification; model operations; and policy issues.
Recommendation 1. Existing Data Inventory
A data collection inventory will support efforts to make the most effective use of all traffic and transit data that already exists (e.g., for sub-area and screenline validations), and help determine what additional data are needed. The traffic count database must be updated. Also, the latest 2005 digital-ortho aerial photos must be reviewed to verify accuracy of current road alignments and existing number of lanes.
Recommendation 2. Vehicle Classification Counts
Additional vehicle classification counts are needed. Once these have been collected and analyzed, it may be necessary to update the existing commercial vehicle model. The classification counts will also be useful for air quality conformity analyses.
Recommendation 3. Network Coding and TAZ Structure
The panel endorsed SEMCOG's desires for an integrated multi-year network/database management structure. As SEMCOG considers a more refined zone structure, the level of roadway type coded in the database management structure needs to be kept consistent with current classifications. Transit access/egress coding should be considered in the TAZ development, along with the number of trip ends in a zone. Since interactions with Windsor, Canada are important for good regional modeling, the panel suggested consideration of a skeleton network/TAZ structure for Windsor rather than simple treatment of the bridges and tunnel to and from Windsor as external stations.
Recommendation 4. Land Use Modeling
The panel agreed on the importance of land use modeling and integration with travel demand modeling. The panel is aware that SEMCOG is completing some initial Washtenaw County tests using the URBANSIM software, and reminded SEMCOG that it must be mindful of the significant resources and multi-year time frame that will be needed for a full regional implementation of URBANSIM.
Recommendation 5. Additional Trip Purposes
Another consideration during the model updates is the additional market segmentation of home-based work trips for purposes of trip generation, distribution, and mode choice. Although the number of university trips represents a very small portion of all regional trips, they are nevertheless very important to some areas within the region (like Washtenaw County) and should be considered as a separate trip purpose.
Recommendation 6. Trip Generation and Distribution Review
The current model's purpose-specific trip rates, average trip lengths, and trip length frequency distributions need to be compared against the 2004-2005 household survey data. The home-based work trip patterns should also be compared against the year 2000 patterns from the 2000 Census Transportation Planning Package.
Recommendation 7. Non-Motorized Modes
When comparing SEMCOG's model to those in other areas, the panel supports the inclusion of the non-motorized mode as the "right thing to do" in good model practice, and noted the need to summarize the data from the 2004-2005 household survey. However, three of the five panel members felt that a full implementation is perhaps not the highest priority for the next near-term model update.
Recommendation 8. Trip Distribution
Two of the five panelists were concerned that gravity models are inadequate in modeling travel patterns for the trip distribution step. However, all five panelists agreed that a destination choice formulation should be developed and sensitivity-tested. The results should be compared against the current gravity model implementation to determine which of the techniques should be incorporated in a future updated model.
Recommendation 9. Mode Choice
SEMCOG can take advantage of the recent 2002 transit onboard survey data as well as the anticipated data from the 2004-2005 household survey to re-estimate and re-calibrate the mode choice models. However, the following tasks should be done in advance of the model re-estimation effort:
Recommendation 10. Traffic Assignment
The panel endorsed a time- and distance-based generalized cost traffic assignment as a good modeling approach. The panel also suggested that some of SEMCOG's reported validation results undergo a re-examination of the BPR-format volume-delay factors, especially for freeways. At a minimum, there should be some additional validation checks with updated traffic count data.
Recommendation 11. Air Quality Model Integration
To complete the air quality model integration, the panel felt that the class-specific TOD traffic volumes and speeds used in subsequent air quality analyses should be derived from the class-specific TOD traffic volumes from the traffic assignments.
Recommendation 12. Airport Modeling
SEMCOG should take advantage of the airport analysis in the model re-estimation work performed by PB (see Recommendation #9). Four of the five panelists felt that a special air passenger survey is probably not necessary.
Recommendation 13. Enhanced Freight Modeling
The panelists recommended that SEMCOG better integrate its truck model with the Michigan DOT statewide model. Although not under the control of SEMCOG, the panelists noted that the Michigan statewide model would be more useful for freight modeling if it included commodity flows for all modes of travel (e.g., truck and rail).
Recommendation 14. External Trips
SEMCOG's external trip modeling should be coordinated with the Michigan DOT statewide model to the greatest extent possible.
Recommendation 15. Uncertain Model Improvement Items
The panel identified three other model improvement areas that require further investigation:
Recommendation 16. Activity-Based Modeling
The panel recommended that a full activity-based model implementation be considered as a long-term model improvement goal. They felt that the best strategy for SEMCOG is to continue its pursuit of four-step model improvements, and await the results of rigorous activity-based model sensitivity tests performed in other regions.
Recommendation 17. Validations
The panelists felt that the year 2000 model results need to be examined in more detail. The high root mean square error for freeways may indicate a count problem, a volume-delay problem and/or upper-level model errors. The panel also supported the development of a year 2002 (or later) model validation that will provide further useful information on the quality of the current model (as well as any future model update).
Recommendation 18. Travel Speed Verification
The panel recommended additional highway speed (travel time) studies. These can be used to compare TOD assignment speeds against observed TOD speeds. The panelists discussed the related need to see how the average TOD modeled bus speeds compare to average TOD observed bus speeds.
Recommendation 19. Traffic Operation Tools
Overall, the panel recognized SEMCOG's interest in traffic operations tools for use with short-range traffic operations analysis (e.g., traffic signal improvements and/or changes to intersection geometries). The panel pointed out that the limiting constraint in pursuing traffic operations tools may be the resources available for both the four-step travel modeling and detailed traffic microsimulation (or dynamic traffic assignment). The panel agreed that there is a need for a bridge between regional modeling and detailed traffic operations analysis.
Recommendation 20. Travel Model Sharing
During the peer review, the panel was asked to consider how SEMCOG can share their travel model with other agencies. Workable options for travel model sharing include:
|Day One: December 6, 2004, Monday|
|8:30 am||Welcome and introductions - Carmine Palombo, Director, SEMCOG Transportation Programs|
|8:45 am||Overview of the peer review process and charge to the peer review panel - Ken Cervenka, NCTCOG|
|9:00 am||Overall schedule for the model and planning process - SEMCOG, Michigan Department of Transportation, Washtenaw Area Transportation Study, and St. Clair County Transportation Study|
|9:30 am||Presentation on current travel model - Li yang Feng, SEMCOG|
|10:15 am||Presentation on current travel models - (continued)|
|1:00 pm||Air Quality Conformity modeling - Joan Weidner, SEMCOG|
|1:15 pm||Presentation of the model improvement plan - Tom Rossi, Cambridge Systematics|
|1:45 pm||Questions and answers|
|3:15 pm||Panel deliberation (panel members only)|
|Day Two: December 7, 2004, Tuesday|
|8:30 am||Panel deliberation (panel members only)|
|10:30 am||Presentation of findings and recommendations - panel|
|11:00 am||Discussion of panel findings and next steps|
Terri Blackmore, WATS
Alex Bourgeau, SEMCOG
Tom Bruff, SEMCOG
Ken Cervenka, NCTCOG
Jim Cramer, FHWA
William Davidson, PB
Tiffany Draper, SEMCOG
Jennifer Evans, SEMCOG
Li-yang Feng, SEMCOG
Christopher Forinash, EPA
Jay Gardiner, SMART
Qiang Hong, SEMCOG
Tiffany Julien, SEMCOG
Shruti Mahajan, Volpe Center USDOT
Sayeed Mallick, SEMCOG
Chris Mann, SEMCOG
William McFarlane, SANDAG
Carmine Palombo, SEMCOG
Tim Roseboom, DDOT
Tom Rossi, Cambridge Systematics, Inc.
Jerry Rowe, SEMCOG
Mark Schlappi, MAG
Stephanie Taylor, SEMCOG
Joan Weidner, SEMCOG
Brad Winkler, MDOT
Supin Yoder, FHWA