PDF files can be viewed with the Acrobat® Reader®
|Location||Date||Peer Review Host Agency Contact||Peer Review Panelists|
|Los Angeles, California||January 9-10, 2006||Deng Bang Lee, SCAG, firstname.lastname@example.org||Frank Spielberg (Chair), BMI-SG, a VHB Company|
|Mark Bradley, Mark Bradley Research Corporation|
|Chandra Bhat, University of Texas at Austin|
|Erik Sabina, Denver Regional Council of Governments|
|Chuck Purvis, Metropolitan Transportation Commission|
|Bruce Spear, Federal Highway Administration|
|Jim Ryan, Federal Transit Administration|
Travel Model Improvement Program
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.
The Southern California Association of Governments (SCAG), through the Federal Highway Administration's Travel Model Improvement Program, requested a peer review panel to assess its current travel demand model and its application, and to obtain comments on model components under development. The panel was asked to concentrate on updated model components, especially trip distribution, mode choice, and trip assignment. The panel was also asked to recommend near- and long-term model enhancements. This report summarizes the January 2006 two-day peer review meeting held in Los Angeles, which was the third of three peer reviews.
The panel felt that SCAG's current four-step model is consistent with the state-of-the-practice. The panel especially liked the plan for the freight model, the use of the "strategic work trip" purpose in trip distribution, and the use of four time periods in traffic assignment. It also felt that SCAG has done particularly well in data collection efforts. It recommended that SCAG make its extensive survey datasets available to other transportation planning agencies. The panel felt that the planned speed study was a good next step.
The panel's general recommendation was for SCAG to focus on near-term validation of the model. The suggested that SCAG assess the commuter rail nesting structure in the mode choice model as part of the mode choice validation process. SCAG should test its vehicle availability model, which might be too sensitive to the density variable. It recommended that SCAG consider a destination choice configuration for trip distribution. Finally, the panel said SCAG should consider adding model components for high-occupancy toll lanes and peak spreading.
The Southern California Association of Governments (SCAG) is the metropolitan planning organization for six Southern California counties: Los Angeles, Orange, San Bernardino, Riverside, Ventura, and Imperial. The region's population exceeds 16 million people over an area of more than 38,000 square miles.
To support the federally required Regional Transportation Plan and the Regional Transportation Improvement Program, SCAG maintains a travel demand model that predicts the impact of travel growth and evaluates potential transportation improvements. The current SCAG model follows the traditional four-step modeling structure. Three ancillary models feed into network assignment:
In 2002, SCAG initiated an effort to use new data to update and recalibrate its travel simulation model. In January of that year, the first peer review of SCAG's model was conducted. At this time, the panel felt that SCAG's model was at the leading edge of the state of the practice. The panel recommended changes including adding trip purposes, creating a vehicle availability model, and modifying the mode choice model. SCAG has implemented or begun implementing most of the recommendations. The second peer review was held in November 2003. Topics reviewed during this meeting included validation targets, revised vehicle availability model, trip generation, external trips, and the selection of variables for the mode choice model.1 Also in November 2003, Cambridge Systematics was awarded a contract to improve the current travel demand model.
During the third meeting of the peer review panel, documented in this report, SCAG requested that the panelists examine its current model and concentrate on assessing the model updates, including trip distribution, mode choice, and trip assignment. It also requested that the panelists assess the agency's plans for model improvements.
The main goal of the travel model improvement project was to improve policy sensitivity to refined behavioral characteristics and demographic characteristics, including:
The second important goal of the project was to increase network simulation accuracy through:
During the 1990s, SCAG embarked on the first comprehensive overhaul of its travel simulation capabilities since the mid-1970s. This included several major data collection activities and the development of new modeling components. The improvements included:
In 2004 and 2005, SCAG improved the following model components:
Additionally, the model's zone system was expanded to include a larger area and the modeling platform was converted from TRANPLAN to TransCAD. This included updating the network path building processes and trip assignment methods.
Improvement planned for the medium term include:
Proposed long-term model improvements include:
The SCAG model uses input from the following data sets:
Data were collected at eight sites in the city of Los Angeles. These included principal and minor arterials with 15-55K average daily traffic, 4-6 lanes, and 10-12 signals per mile. There are 54 directional segments between signals and four hours of observation for each direction. Two hundred sixteen hourly volume and speed observations were recorded for a volume-to-capacity range of 0.1-0.99 and a speed range 4-41 mph. Volume counts were recorded for signalized intersections at 15 minute intervals. Speed counts were recorded for global positioning system (GPS)-equipped floating cars at three to ten samples per hour.
A full-scale model validation is being conducted to check the accuracy of the model inputs and outputs, and to determine the model's sensitivity to specific variables. Additionally, some validation tests were completed in 2004 and 2005.
The accuracy check for the model will include validation of the following socioeconomic and parameter data:
The accuracy of the highway network will be validated by a review conducted by each county transportation commission. This will include the following activities:
The accuracy of the transit network will be validated by the following:
The accuracy of the outputs for each model will be validated as follows:
i. Trip Generation
The percentage of trips for each trip type and vehicle availability will be checked against the O-D survey
ii. Trip Distribution
Trip length frequency distributions by purpose will be compared to the distribution by market segment. The percentage flows of work trips-strategic3 plus direct-by county, regional statistical area, and community statistical area will be reviewed against the Census Transportation Planning Package (CTPP) data and O-D survey.
iii. Mode Choice
Trip end (production and attraction) percentage by mode will be compared against the O-D survey. Trip end percentage by mode by RSA will be compared against the O-D survey and CTPP. Sensitivity of mode split to changes in time and cost will be evaluated against expected ranges.
iv. Highway Assignment
Vehicle miles traveled (VMT) will be compared with the highway performance monitoring system database by air basin and county. Assigned volumes will be compared against the downtown cordon count report. Absolute and percent speed differences will be compared for the performance measurement system and the automated traffic surveillance and control database. Absolute and percentage volume differences by assignment group and mode will be compared against the screenlines. Modeled and count volumes will be compared based on:
The comparison will use:
v. Transit Assignment
Transit boarding counts by company and mode for the model output will be compared with actual bus ridership by company and with individual urban and commuter rail line ridership. Boarding counts will also be compared against the downtown cordon count report.
The proposed sensitivity analysis for the model will include the following steps:
This part of the presentation focused on the following elements of the SCAG regional model:
The trip generation, vehicle availability, household classification, trip distribution, and mode choice components were recently updated. The first three were reviewed in a previous peer review; the current peer review focused on trip distribution and mode choice. Although validation and sensitivity testing of all model components was discussed during the presentation, the panel concentrated on the validation and sensitivity testing of mode choice.
A key descriptive characteristic of the SCAG model is the number of vehicles either available to or owned by households in the region. Household vehicle availability-the total number of motor vehicles available for use by household members-is more closely related to the level of household mobility than the more limited household vehicle ownership measure. The three discrete choice model structures tested were a multinomial logit model, a nested logit model, and an ordered response logit model. The multinomial logit and nested logit models produced good results, but the ordered response logit model could not be estimated because of the structure of the SCAG dataset. After model estimation, it was recommended that SCAG use the multinomial logit model structure.
The primary data source for the model is the 2001 SCAG household travel survey. The independent variables are:
The independent variables that were tested but not included in the model are accessibility and ethnicity.
The results from the model estimation suggest the following trends for each variable:
The household classification model generates the percentage distribution of households for each trip purpose. To do this, it uses the Census Public Use Microdata Samples (PUMS) joint distribution of households to stratify the marginal distributions provided by SCAG. It estimates the following joint distributions of households:
The SCAG model has ten trip purposes:
These match the original nine trip purposes in the previous SCAG model, with home-based serve passenger added based on a recommendation from the second peer review panel.
Three income groups were used for home-based work trips. These were split into home-based work-direct, and home-based work-strategic. Home-based work-strategic represents all "journey to work" tours that have one or more intermediate stops, regardless of duration.
SCAG conducted a GPS survey from a sub-sample of households along with the household travel survey. Based on the GPS information, trip generation by trip purpose was adjusted based on age, income, and trip duration. This resulted in a 17.7 percent overall increase in the number of trips.
The trip production model for SCAG was developed using cross-classification. The following independent variables were used for the various trip purposes:
The trip attraction model was estimated by merging socioeconomic data with the SCAG household travel survey data by zone and then estimating linear regression models by trip purpose. Trip attractions were estimated at the zone, superzone, and regional statistical area level to understand the potential error or instability in the coefficients. The following independent variables were available at the zone level:
As part of the ongoing model improvement project, external trip models were developed and integrated into the SCAG model system. In the trip generation step, final trip generation rates include all trips (including internal-external and external-external). External trips made by residents-estimated from the external survey by trip purpose-are subtracted from the trip generation results. Zones near the border have more trips subtracted than other zones to reflect that fact that trips in these zones are more likely to have an external end than trips that originate elsewhere in the region. External trips are subtracted from the productions and attractions so that, when the remaining internal trips are distributed, there are fewer internal-internal productions and attractions at border zones than other zones.
After time-of-day models are processed (post-mode choice), all external trips are added back as vehicle trips, since trip purpose is not used in highway assignment. These are separated by external-internal, internal-external and external-external trip types and by vehicle occupancy groups (1, 2, and 3+ occupants). The current trip production and attraction models were updated to reflect the external trips, which account for 0.02 trips per household based on the SCAG household travel survey.
Trip distribution uses a gravity model. The model is applied for peak and off-peak conditions for each trip purpose and by income group for work trips. It was developed using the 2001 SCAG household travel survey. New friction factors were calibrated to validate observed trip length frequency distributions, district-to-district flows, and percent of intra-zonal trips. Trip distribution is applied to the same ten trip purposes used in trip generation for both peak and off-peak conditions. As a model enhancement, the gamma function was adjusted for the trip duration distribution by trip purpose. Also, the logsum from mode choice was used in home-based work-direct trips.
An intermediate-stop model is used to allocate home-based work-strategic trips to intermediate stops after mode choice. The intermediate stop model is a two-step process. It first estimates the distance from the home to the intermediate stop, and then estimates the location of the intermediate stop. The estimations are made using the SCAG household travel survey.
The distance to the intermediate stop is a log-linear distance decay function between the number of home-based work-strategic trips and the distance between home and the first intermediate stop. It is calculated using the formula:
ln(trips) = 6.009 - 0.026 x distance
The location of intermediate stops is calculated by using the formula:
Trips attracted at stop = 0.375 x retail employees + 0.120 x (education/health employees) + 4.651 x (other service employees) + 0.281 x (school enrollment) - 0.388 x (professional/management employees) + 0.232 x (public administration employees)
The mode choice model was originally estimated using the 1991 regional household travel survey for the home-based work, home-based school, home-based university, other-based other, home-based other, and work-based other. The first three trip purposes are nested logit models. The remaining two are multinomial logit models. The models each include the choice of motorized versus non-motorized modes. The auto mode is stratified by occupancy, and transit is stratified by mode of access (walk versus auto) and type of service (local versus express).
For the mode choice model, the six travel surveys were combined into a single estimation dataset. The weighted exogenous sampling maximum likelihood was used for estimation by mode and county. The following modes were included:
The model includes the following variables:
For mode choice model validation:
The trip assignment model was updated by the Caliper Corporation. Updated model components include:
The updated highway network is derived from the SCAG 2000 base year and 2003 networks. Caliper realigned the networks to match high-resolution aerial photos. All the TRANPLAN attributes were preserved and new network attributes were developed. The network has approximately 72,000 line segments. New network attributes added include parking restrictions, grade percentages, and median presence. The link topology was standardized (north or east-based) during the update. SCAG staff is currently refining the network, providing quality control and adding screenline counts.
TranStar, the region's automated transit trip planning service, provides the routes, stops, and schedules for the entire region's transit. The transit routes lie on top of realigned road network. Headways and transit times were developed from the schedule data. The bus transit times are a function of congested highway speeds.
Skims were developed from the new highway and transit networks. The congested highway and transit times were estimated from preliminary assignments carried out in two phases. Phase 1 was the initial user equilibrium highway assignment using the SCAG base model O-D matrix that was converted from 3000+ zone base model to 4000+ zones. In Phase 2, the user equilibrium highway assignment using 4000+ zone O-D matrix was developed from Cambridge Systematics' new trip generation, trip distribution, and mode split models.
In the highway skimming process, the travel costs are based on the congested peak and off-peak travel times in addition to the auto operating cost of 15.08 cents per mile. The free-flow speeds are based on functional class-area type tables with exceptions for certain classes and lane capacities. The congested costs are based on results from morning and mid-day highway assignment. SCAG conducted spot-checking of selected O-D pairs, which indicated reasonable congested travel times and paths.
Transit skims were developed using TransCAD's Pathfinder method. Most of the path parameters were inherited from earlier TRANPLAN models. Parking costs and an initial wait time function were added. The skims were provided to Cambridge Systematics for their final trip distribution and mode split model estimation.
A limited spot-checking by SCAG staff revealed some path anomalies, which were addressed by correcting data errors. SCAG plans to perform additional checking. Some Metrolink path problems still exist due to long initial wait times and higher fare costs
The Akcelik5 delay function was used for the arterial and freeway links. The delay parameters for each were estimated separately. Akcelik uses link length as an input parameter, so it is essential to ensure correct topology. For the centroid connectors, the unconstrained function was used.
The highway assignment is based on multi-class user equilibrium for the following user classes:
The Phase 1 assignments generally matched overall TRANPLAN results for VMT, vehicle hours traveled, delay, VMT by air basin, and volatile organic compounds by functional classification. The Phase 2 assignments are currently being evaluated. Screenlines and screenline counts are being coded by SCAG staff. The assignment aims to match daily and peak period counts.
Transit assignment is a production-attraction matrix on the peak and off-peak networks. The goal is to perform four-period assignments with correct park-and-ride accounting. The assignments will be compared with recently collected boarding count data.
Travel times were based on direct feedback of congested highway assignment results. The model uses the multiple successive average feedback technique. The number of feedback iterations will be determined as part of further calibration.
Reporting and post-processing macros need to be developed to replicate standard SCAG reports after the model run. Links to the TransCAD SUMMIT procedure and air quality models also need to be developed.
Cambridge Systematics presented an update of the HDT model. The topics covered during the presentation included:
The existing HDT model has three truck weight classes corresponding to trip purposes. The model has three main components for trip generation and distribution:
A review of the internal model highlighted the following areas for improvement.
SCAG plans to conduct a HDT travel survey for use in updating the HDT model. The following new trips types should be added:
The sampling frame is Department of Motor Vehicles registration data. The stratification of data will be conducted through weight class and business type (owner). The implementation of the survey will include:
A review of the external model highlighted the following areas for further improvement.
The improvements required for trip assignment include simplifying the PCE factors, adjusting the time-of-day factors based on new survey results, and modifying the volume-delay functions.
The peer review panel was asked to assess the current model and outline the strengths and weaknesses of the model and its application to transportation planning, concentrating on the updated model components including trip distribution, mode choice, and trip assignment.
The panel felt that SCAG has developed a "state-of-the-practice" four-step model. The panel highlighted several strengths, including the freight model, the strategic work trips, and the use of four time periods in assignment. The panel felt that SCAG has done a particularly good job in data collection, and that the planned speed study was a good next step. The following are specific recommendations on other model components.
SCAG needs to do additional work to test the sensitivity of the vehicle availability model. The model may be too sensitive to the density variable. If necessary, it should be reestimated.
The panel was concerned with the location of commuter rail in the nesting structure of the mode choice model. It recommended that SCAG reconsider this in combination with reestimation of the mode choice model. One panelist suggested that the use of auto access to transit might work as part of the mode choice nested structure. Two other panelists recommended exploring market segmentation, carried through further in the strategic trip patterns, and looking at the PUMS and CTPP datasets to examine commute time by income.
The panel recognized that SCAG has invested substantial resources in data collection and development of the mode choice model. However, the effort did not yield the mode choice form and structure that was expected. SCAG should consider reestimating the model using different structures. The mode choice model should be run for the year 2030 to ascertain if changes are reasonable.
The panelists were concerned that the gravity model is inadequate. As part of the long-term model improvement efforts, SCAG should consider developing and sensitivity-testing a destination choice formulation. The results should be compared against the current gravity model to determine which of the techniques should be used in the future.
The panel identified other opportunities for model improvement:
The panelists felt that SCAG should plan to devote significant resources to validation of the overall model. SCAG should plan to spend a year on this effort, which should be based on a carefully developed model validation scope of work and a plan for testing. The panel suggested that the scope address the following questions:
A good check of the transit network and impedances is assignment of a trip table developed from the on-board survey data.
SCAG staff noted that a constraint in carrying out the validation process is that the agency plans to use the models within six months for the RTP. The panel suggested that SCAG carry out validation as thoroughly and as promptly as possible. Therefore, the model run time needs to be reduced. SCAG should to install hardware that is fast enough to permit efficient use of the models.
SCAG has an impressive set of survey datasets. The panel felt that SCAG should make these survey datasets available to other agencies.
DATE: Monday, January 9, 2006
TIME: 8:30 a.m. to 4:30 p.m.
PLACE: 818 W. 7th Street, 12th Floor, Los Angeles, CA 90017 (213) 236-1800
|9:10||Welcome - Mark Pisano|
|9:15||Project Overview and Objectives - Keith Killough|
|9:30||Regional Model Improvement Project - Cambridge Systematics, Inc.|
|9:45||Review of Trip Generation Model - Cambridge Systematics, Inc.
|10:30||Trip Distribution Model - Cambridge Systematics, Inc.|
|11:30||Model Choice Model - Cambridge Systematics, Inc.
|1:00||Model Choice Model - Cambridge Systematics, Inc.
|2:00||Highway & Transit Assignment Model - SCAG & Caliper
|3:15||Model Structure & Implementation - Caliper
|4:00||Full Model Validation - SCAG|
|4:15||Current & Near Future Model Improvements - SCAG|
|4:30||Thank you - Day One Adjournment|
DATE: Tuesday, January 10, 2006
TIME: 9:00 a.m. to 12:00 p.m.
PLACE: 818 W. 7th Street, 12th Floor, Los Angeles, CA 90017, (213) 236-1800
|11:00||Panel Presentation of Findings and Recommendations|
|11:30||Discussion with SCAG Management|
|12:00||Thank you - Meeting Adjournment|
|Name||Company Name||Telephone No.||Email Address|
|Steve Smith||San Bernardino Associated Governmentsemail@example.com|
|Steve Pickrell||Cambridge Systematicsfirstname.lastname@example.org|
|Firooz Hamedani||Caltrans District email@example.com|
|Anup Kulkarni||Orange County Transportation Authorityfirstname.lastname@example.org|
|Shruti Mahajan||Volpe Centeremail@example.com|
|Chandra Bhat||University of Texas at Austinfirstname.lastname@example.org|
|Erik Sabina||Denver Regional COGemail@example.com|
|Tony Van Haagen||Caltransfirstname.lastname@example.org|
|Chuck Purvis||Metropolitan Transportation Commissionemail@example.com|
a VHB Company
|Deng Bang Lee||SCAGfirstname.lastname@example.org|
|Mark Bradley||Mark Bradley Research Corporation||805-564-3908||mark_Bradley@cox.net|
|Michael Fischer||Cambridge Systematicsemail@example.com|
|Maren Outwater||Cambridge Systematicsfirstname.lastname@example.org|
1 The reports from the first and second peer reviews can be found at http://tmip.fhwa.dot.gov/services/peer_review_program/status.stm#status
2 Much of the information presented in this section comes from "SCAG Travel Model Improvement Program Model Update Documentation" Cambridge Systematics, Inc. with Urban Analytics, Inc. July 2005.
3 A "strategic" work trip contains an intermediate stop between the origin and destination.
4 The numbers in parentheses indicate the number of categories in each variable.
5 See "Urban Arterial speed-Flow Equations For Travel Demand Models" Michael Dowling and Alexander Skabardonis, presented at the Innovations in Travel Modeling Conference 2006, sponsored by the Transportation Research Board, Austin, TX, May 21-23, 2006. http://trb.org/Conferences/TDM/papers/BS2C%20-%20Urban%20Arterial%20Speed-Flow.pdf
6 Much of this section is taken from "SCAG HDT Model Update Status Report" by Michael Fischer, Cambridge Systematics, presented to the Modeling Task Force, September 2005.