Case studies from the 2022 Statewide Modeling Peer Exchange
Statewide Travel Demand Modeling - A TPCB Peer Exchange Event was held in Columbus OH in June of 2022. Presentations, as well as a survey of participants, provided a good sense of the state of statewide models in the US and one Canadian province. Surveys were completed by representatives of California, Colorado, Delaware, Florida, Georgia, Iowa, Illinois, Kentucky, Massachusetts, Maryland, Michigan, North Carolina, Ohio, Ontario, Oregon, South Carolina, Texas, Virginia, Vermont, and Wisconsin. Although self-selected, these models are representative of those from across the US.
# Model Structure
There is not a standard statewide model, so structures can vary considerably depending upon needs of policy evaluation or project selection. Most statewide travel models are trip-based. Those states reporting activity-based model were California, Colorado, and Ontario.
External traffic is handled in a variety of ways including a distance halo, a county halo, a state halo, the nation, much of North America, and external stations with or without a halo. There is little discernable pattern as to the number of zones, which seems to be governed more by cost, computational difficulty, and geographic convenience than by population or area. Freight components tend to be limited to trucks in most models, but when rail is included, the traffic assignment step is often omitted. Delaware, Georgia, Mayland, Texas, Virginia, and Wisconsin indicated they did traffic assignment of commodities by rail. Most models included intercity transit as a standalone mode or combined with urban transit.
A handful of states included non-motorized travel to some degree, including all three states with activity-based models. Occasionally, statewide models include a land-use component, which would provide inputs to zonal-level demographic and economic forecasts. Land use forecasts are sometimes derived from regional or statewide master plans. Otherwise, land-use information is implied by demographic and economic forecasts.
# Model Uses
Models are used primarily for long-range forecasting, with horizon years largely within the 2040-2050 timeframe. All states reported using their models for statewide planning purposes. Most states also used their models for corridor planning, assistance to regional planning, freight planning, project prioritization, and project-level traffic forecasting. Some states also reported using their models for air quality conformity, economic development studies, toll studies, and traffic impact analysis. A few states use their models for safety studies. Statewide models are rarely used for operational-level studies.
# Sources of Demographic and Employment Data
Responses as to sources of current demographic and employment data varied considerably across states. Demographic sources included US Census, ACS, CTPP, PUMS, QCEW, state and local agencies, IRS, Data Axle, MPOs, NHTS, CEDDS (Woods & Poole), TREDIS/Moodyís, CBP, and university research centers.
Employment data at the workplace is always a challenge for travel modelers, since many of the databases are geographically coarse or incomplete. Cited employment data sources included LEHD, InfoUSA, QCEW, state agencies, ES202, IMPLAN, Data Axle, MPOs, and Woods & Poole.
# Sources of Origin-Destination Data
Passenger origin-destination data is not usually a direct input to a travel model, but it can assist model calibration and validation. Reported sources include Streetlight, CS LOCUS, Local OD surveys, NHTS, CTPP, AirSage, and LBS. Freight origin-destination data may or may not be a direct input to a travel model, depending upon the modelís sophistication, as freight components can vary greatly in complexity or may be omitted. Mentioned sources include FAF4, Transearch, Streetlight, TREDIS, INRIX, REMI, local ports, commercial truck survey (ON), CFS, and ATRI.
# Economic and Demographic Forecasts as Inputs
For the purpose of testing future scenarios, all statewide models require forecasts of economic activity, demographics, and land use, indirectly. For the most part, these forecasts originate outside the modeling agency. Choices of datasets are opportunistic, with modelers looking for least-cost options locally before purchasing commercial products.
States reported using economic forecasts from other state agencies, university research centers, internal economic model component, InfoUSA, REMI, EBP, Woods & Poole, MPOs, TREDIS, and Global Insight. Demographic forecasts come from both commercial and public sources, including other state agencies, university research centers, MPOs, Woods & Poole, REMI, TREDIS.
# Validation Statistics
Participants were asked about the quality of their validation results for highway traffic, specifically. While results varied considerably and while there were some inconsistencies in categories, a consensus emerged as to reasonable expectations of statewide models that are not dominated by a single large metropolitan area. The table, below, gives the validation results in percent RMSE for Florida, Michigan, and Virginia, which are near the middle of all the states surveyed as to reported quality.
|1 ~ 5000||92||82||84|
|5001 ~ 10000||60||50||46|
|10001 ~ 20000||44||38||34|
|20001 ~ 30000||34||28||23|
|30001 ~ 40000||30||24||25|
Statewide models do not validate as well as urban models, in general. Survey respondents also reported an overall RMSE in base year traffic volumes, but these data are difficult to compare, given the great variation in numbers of links at each volume category.