Thurston Regional Planning Council (TRPC) Travel Model Peer Review Report

Appendix D Overview of Thurston Regional Planning Council TDFM

The following text summarizes the current version of the TRPC model at the time of the review, along with data sources used in the development of the model.

D.1 Thurston Regional Planning Council Model Components

The following sections summarize models components as described in the current model documentation.[10] The trip-based model is made up of four primary modules with relatively sophisticated submodels and feedback as shown in Figure 2 and described below.

• Trip Generation
  • Inputs: TRPC Employment and Population Forecasting Data (single-family and multi-family dwelling units by TAZ) plus the base year housing occupancy rate reported by the Washington State Office of Financial Management (OFM) to factor DUs to households.
  • Cross-classifies the households by household size, income, and age of head of household.
  • Applies to the HH size-income-age classification a multinomial logit choice model to estimate households' number of workers by HH classification category.
  • Applies to the HH size-income-age-workers outputs a multinomial logit choice model to estimate the number of schoolchildren by HH classification category.
  • Applies to the HH size-income-age-workers outputs a multinomial logit choice model to estimate the number of vehicles by HH classification category.
  • Outputs: Applies trip production rates to generate productions by TAZ by HH classification category for the following trip purposes:
    • Home-Based Work (HBW)
    • Home-Based Shopping (HBShp)
    • Home-Based School (HBSch)
    • Home-Based College (HBC)
    • Other origin to other purpose (Other-Other or OO)
    • Work origin to other purpose (Work-Other or WO)
• Trip Distribution/Destination Choice
  • Inputs: Trip productions by purpose by TAZ from the Trip Generation component.
  • Applies to HBW, HBShp, WO, OO and Truck (see section D.2 below) multinomial logit destination choice models using the following explanatory variables variously in purpose-specific utility expressions:
    • daily average TAZ-to-TAZ auto travel time = (AMtTime + MiddaytTime + PMtTime)/3
    • number of retail employees in attraction TAZ
    • number of service employees in attraction TAZ
    • number of government employees in attraction TAZ
    • number of employees other than the above in attraction TAZ
    • number of households in attraction TAZ
    • Notes on trip distribution/destination choice:
      • WO and OO use production TAZ utilities to distribute total regional productions to production TAZs before applying attraction TAZ utilities to link WO and OO trips
      • average travel time is not used in the WO and OO production utility expressions
  • Links HBSch productions to attraction TAZs using school catchment boundaries defined by the school district within which the origin HH TAZ is located.
  • Links HBC productions to attraction TAZs using 1998 college enrollment data, singly balanced.
  • Outputs: trip tables by purpose in production-attraction (PA) format.
• Mode Choice
  • Inputs: trip tables by purpose in production-attraction (PA) format.
  • Applies a multinomial logit choice model to split trips by purpose into the following modes (except for the HBC purpose, see below) on a daily basis
    • drive-alone (DA)
    • drive with passengers (DP; essentially the driver of a shared-ride vehicle)
    • auto passenger (AP; essentially a rider in a shared-ride vehicle)
    • transit (TR)
    • bike (BK)
    • walk (WK)
  • The mode choice utility explanatory variables are (note that not all variables are used in all purposes):
    • Lcost = low-income household cost
    • Mcost = mid-income household cost
    • Hcost = high-income household cost
    • CV01 = no-car or cars<workers household
    • CV34 = cars=workers or cars>workers household
    • HH34 = 3-person or 4+person household
    • Lhh = low-income households
    • Tdist = trip distance
    • Tm = trip time
    • Trfare = transit fare
    • Twait = transit wait time
    • Twalk = transit walk time
    • Pkcost = parking cost
    • Em20tr = total employment accessibility within 20 minutes of transit time to TAZ
    • Em1 = total employment density within 1 mile of walking distance to TAZ
  • HBC purpose mode split on a daily basis is calculated using fixed proportions taken from the 1998/1999 household travel survey.
  • Outputs: trip tables by purpose and mode in PA format.
• Time of Day
  • Applies peak hour factors derived from the 1998 household survey to each mode/purpose/PA/AP combination to create AM, midday, and PM peak hour OD trip tables.
  • See Section D.2 below for a description of the truck and external trip handling.
• Assignment
  • Inputs: trip tables from the mode choice/time of day step for AM, midday, and PM by purpose by the truck, auto, and transit modes (bike and walk are not assigned).
  • Performs a multi-class vehicle assignment for all truck and auto modes in the AM, Midday, and PM peak hours.
  • Performs a multi-path transit assignment for the transit mode in the AM, Midday, and PM peak hours.
  • Outputs: auto travel time and distance matrices; transit travel time, wait time, boarding time, number boardings, and access/egress time matrices; auto and transit volumes on network links.
• Feedback
  • The following data are fed back into trip distribution and mode choice (note that there is no feedback to trip generation; see also the parenthetical note in the transit skim bullet):
    • Walk travel time skims
    • Bike travel time skims using the walk skims divided by four (assuming bikes are 4 times faster than pedestrians).
    • Auto travel time skims from the three peak hours modeled (AM, Midday, and PM) are averaged for use in the submodels described above.
    • Transit travel time, fare, wait, and access/egress time skims are fed back to mode choice (but NOT to trip distribution).
    • Intra-zonal travel times by mode are created assuming 0.5 or 0.75 of the value for the nearest zone-to-zone trip.
• The model makes three global iterations[11] through all components described above with auto assignments using 100 iterations, bgap=0.001%, ngap=0.001% as convergence criteria.

Figure 2: General Modeling Structure of TRPC Model[12]

D.2 Other Relevant Aspects of Thurston Regional Planning Council Model

D.2.1 Transit Modeling

• The model estimates walk-access to transit usage in the same three peak hours as it estimates other travel. It does not treat drive-access to transit, although the model network does include several Park-and-Ride lots as dummy zones.[13]

D.2.2 Commercial Travel

• Truck trip productions are calculated regionwide as a function of:
  • total retail employment
  • total service employment
  • total government employment
  • total other employment
  • total households
• Regional truck productions are allocated to TAZs based on truck production utilities
• Truck attraction trip ends are calculated using a multinomial destination choice model similar to the auto mode destination choice models.

D.2.3 External Travel

• Through trips are taken directly from the 1997/1998 vehicle classification counts and the 1997 I-5/SR-101 external origin-destination surveys.
• Outbound trips are derived from the same data as through trips but scaled and balanced to the internal trip productions for each modeled peak hour before being added to the drive-alone mode trip tables
• Inbound trips are derived from the same data as through trips but scaled and balanced to the internal trip attractions for each modeled peak hour before being added to the drive-alone mode trip tables.

D.3 Thurston Regional Planning Council Model Validation

• Comparison of observed-to-modeled volumes on links with traffic counts in the AM, Midday, and PM peak hours obtained R-squared results of 0.94, 0.95, and 0.95 respectively.
• Comparison of observed to modeled PM peak-hour auto volumes crossing 18 screenlines obtained percent differences ranging from 13% to -11% (with most percent differences in single digits). Directional daily volume comparisons on the same screenlines resulted in a range of -15% to 22% (again with most differences in the single digits).
• Total modeled daily transit person-trips totaled 97% of surveyed trips on routes surveyed by Intercity Transit in the base year (1998/1999). There was considerable variation between modeled and surveyed trips on individual routes, but this is not necessarily an issue given the known variability in route-specific findings from transit surveys and the fact that few regional models calibrate to the transit route level.

D.4 Thurston Regional Planning Council Current Model Data Sources

D.4.1 Household Survey

• Conducted in 1998-1999
• 1,537 complete responses
• 2-day travel diary with household, person, vehicle, and trip data
• Validated using 1990 CTPP

D.4.2 Demographic and Census Data

• 1990 Census
• 1990 CTPP

D.4.3 Transit Counts

• Intercity Transit Ridership Survey
• Conducted last quarter 1998/first quarter 1999

D.4.4 Traffic Volume Data

• WSDOT loop detector data on I-5
• Arterial traffic counts from various sources

D.4.5 Travel Time and Speed Data

• WSDOT loop detector data on I-5

D.4.6 Truck Data

• 1999 vehicle class counts
• 1997 I-5/US 101 OD Survey
• 1997 Reebie Freight data

D.4.7 External Surveys

• 1997 I-5/SR-101 External Origin-Destination Survey

D.4.8 GIS Data

• TRPC in-house GIS data
• Thurston County GIS clearinghouse

[10] Ren, Jin. Thurston Region Multimodal Travel Demand Forecasting Model Implementation in EMME/2. 15th International EMME/2 Users' Group Conference, Sept. 18, 2000.

[11] TRPC. Travel Model Improvement Program (TMIP) Peer Review Meeting--Review of Proposed TRPC Model Improvements. Presentation for the Peer Review. June 11, 2012. Slide 54

[12] Thurston Regional Planning Council. Thurston Region Multimodal Travel Demand Forecasting Model Development. 2000. p. 3

[13] Ren, Jin. Thurston Region Multimodal Travel Demand Forecasting Model Implementation in EMME/2. 15th International EMME/2 Users' Group Conference, Sept. 18, 2000. p. 28