While Minnesota has long considered the idea of a statewide travel demand model, new requirements to assess the GHG and VMT impacts of highway expansion projects tipped the scales to finally confirm the need for one. Updated transportation legislation in 2024 authorizes MnDOT to make a substantial investment in a statewide model over the next few years. This presentation will highlight the initial scoping of the statewide model conducted to help frame what the model might look like and how it can be applied to a host of transportation investment and policy questions in the years to come.

The scoping exercise included an extensive search of statewide modeling underway across the country, including a deep dive into models for Georgia, Oregon, and Wisconsin.  This was followed by an interactive workshop with a diverse group of stakeholders to gather their perspectives on desired characteristics and applications for a new statewide travel demand model.  A variety of statewide model scenarios were then developed and subjected to an evaluation process to assess how well they would accomplish MnDOT and stakeholders’ goals.

This work culminated in an implementation plan with a series of recommendations for MnDOT to invest in its statewide model capabilities.  The recommendations call for a novel approach that combines a strategic planning model with a tour-based travel model.  It was clear that a traditional travel model alone would not satisfy many of the expected use cases identified by stakeholders, particularly around behavior change initiatives and increased multimodal opportunities.  The proposed modeling suite provides opportunities to address a wider range of pressing policy questions, while continuing to support traditional transportation planning activities.

This presentation highlights the evolution and current capabilities of the Texas Statewide Analysis Model (SAM), focusing on the latest version completed in 2024. SAM has supported numerous projects for TxDOT, MPOs, and other partners, helping inform transportation planning efforts across Texas. As a critical tool for long-range planning and policy analysis, SAM provides a consistent statewide foundation for understanding travel patterns, evaluating infrastructure investments, and coordinating planning across regions.

While activity-based models offer richer behavioral detail, the choice of statewide model architecture depends on many practical factors beyond theoretical pros and cons, such as state size, available data, project timelines, and staff capacity. For Texas, a refined trip-based structure remains the best fit. SAM continues to use a four-step framework with major enhancements, including a multimodal network (highway, urban and intercity rail, and air), two complete model streams for passenger and freight, weekday and weekend scenarios, destination choice for selected passenger purposes, explicit modeling of long-distance trips, and improved freight mode choice using Transearch data. In addition to producing accurate forecasts, SAM emphasizes usability, delivering outputs through streamlined Excel and HTML reports as well as an ArcGIS Online dashboard for easy geographic comparisons and visualization.

The SAM experience shows that with thoughtful design, ongoing enhancements, and user-focused outputs, trip-based statewide models can remain highly relevant and practical. This philosophy is also being applied in other statewide efforts, such as Arkansas, where limited budgets and sparse data require pragmatic yet robust modeling solutions. The session will share practical lessons from SAM’s development that can guide similar statewide modeling projects.

The walking and biking traffic environment has been challenged due to increased number of vehicles, traffic speed, safety risks, insufficient facilities for active travels, and fierce competition for curbside activities. Current mode choice models on biking and walking, however, have two limitations. First, the trip-level information of alternative travel modes to bike and walk trips is often missing or limited. Second, the level of traffic stress (LTS) is a critical factor influencing travelers’ decisions on non-motorized modal choices but has not yet been incorporated into regional mode choice models in Maryland. LTS, which measures the traffic stress experienced by pedestrians and cyclists, has emerged as an important tool in the planning process for active travels by local transportation authorities. These limitations restrict the models' ability to accurately predict travel behavior and estimate travel demand for walking and biking. With the household travel survey and a comprehensive LTS dataset from Maryland, this study constructs a statewide mode choice model specifically considering biking and walking, along with public transit and private driving. We applied the Google Maps platform to acquire full trip-level information of all alternative travel modes for each trip, including routes, time, cost and distance. The results showed that LTS plays an important role in improving model performance. LTS has a significant and negative correlation with walking and biking. This study highlights that transportation planners and engineers should prioritize strategies, such as complete street measures, to create a low-stress traffic environment to encourage the use of active travel modes. In addition, LTS improvement projects should be strategically prioritized in specific locations and supported by targeted policies to maximize their impact. The resultant statewide non-motorized choice model can be applied to further develop strategies, policies and infrastructure planning for active transportation.

Long distance and visitor travel models are particularly important in statewide and other large area travel demand models.  However, they have suffered from lack of data for model development due their absence or rarity in the household travel surveys used to develop such models.  Third party origin-destination data (3POD) offer the opportunity to “fill in the gaps” but the details of this depends on both the nature of the data provided by the 3POD vendor as well as the other data available that needs supplemented.  To demonstrate, two case studies of recent statewide model development in Arizona and Ohio will be presented.  Both areas have existing long distance/visitor models of very different forms and have different data availability allowing a comparison of different techniques to leverage 3POD.

In Arizona, the former model involved an enumeration of state-to-state flows from the 2001 NHTS which were then disaggregated to TAZs.  The relative rarity of long-distance travel makes using the more recent small sample 2022 national NHTS surveys in this way problematic (less than 1000 long distance trips nationally).  ADOT also did not have a prespecified 3POD vendor, so an evaluation was conducted.  Replica was selected as the 3POD source which provides a modeled data set with synthetic travelers and disaggregate tours that allowed estimation of models in a manner analogous to traditional model development from household surveys, a key advantage where no other detailed travel data was available.  Aggregate models were developed for a variety of reasons, including that the modeled tour patterns in the data were not always internally consistent.  While the short distance resident models were updated using more recent NHTS data, models for these were also developed from Replica and compared to ensure that mixing and matching from such disparate data sources was reasonable.  Various challenges had to be overcome including the regional geographies used in the Replica data which limit its ability to analyze long distance travel due to the constraints of the boundaries.

In Ohio, the former model was developed from a bespoke long-distance survey.  This survey only captured the primary destination of long-distance trips so while the model developed was disaggregate, it did not capture intermediate stops nor complex tour patterns or trips made in the vicinity of the primary destination nor trips made by non-residents.  A new long-distance survey is available which was obtained via GPS capture, this included all trips over 50 miles (but not shorter trips on such long-distance tours) and only included trips for those HH members participating by GPS device which introduced additional challenges.  In this case, the state already had a 3POD data provider, specifically StreetLight.  Leveraging the ability of the survey to provide demographics with the high data density and inclusion of non-resident travel in the 3POD data, an aggregate model was specified that takes advantage of the strengths of each.  Reconciling these data as well as combining a StreetLight zonal analysis with an OD analysis to parse the trip matrices into resident/visitor and away from home components are the biggest challenges to overcome.

NCDOT is investing in a major update to the North Carolina Statewide Travel Model (NCSTM) to produce NCSTM version 5.  This update involves several basic updates to the model’s fundamentals, processing and analysis of multiple sources of big data, and five advanced innovations including nested long-distance destination choice models, disaggregate machine learning trip generation models, work from home choice modeling, truck route choice modeling, and CAV scenario testing functionality. 

The new NCSTM5 will include a revised network and zone system, updated socioeconomic data to reflect 2022, freight data from FAF5, and a fully rewritten codebase.  The model will also make use of multiple sources of big data including ATRI, Transography, and Streetlight. 

The first major innovation in version 5 of the NCSTM is the development of nested destination choice models for long-distance trips.  While the use of nested logit models has long been common for mode choice, the use of nesting in destination choice is relatively new in practice and the NCSTM is believed to be the first statewide model to adopt it.  The approach essentially divided the destination choice into an upper level choice of region and lower level choice of the specific zone in that region.  The approach produces much more accurate results allowing the model to far better replicate real flows between regions.   

The second major innovation is the incorporation of machine learning to more accurately predict trip generation.  The use of machine learning techniques such as decision trees have been shown to outperform traditional methods of trip generation in the Triangle Regional Model and have now been adopted by roughly half a dozen other agencies around the country.  The new trip generation models will also be disaggregate, person-based models and make use of a new synthetic population for the NCSTM.  The use of synthetic populations in statewide models has increased in recent years in part due to the development of the fast iterative proportional updating algorithm in TransCAD. 

NCSTM5 is also incorporating a new work from home model.  The module will be based primarily upon post-pandemic household travel survey data collected in the Triangle and Charlotte regions and calibrated to the Census Bureau’s American Community Survey (ACS) data for the state.  ACS data for the state shows that even though work from home has declined somewhat from its peak in 2020 and 2021, post-pandemic rates of work from home are still more than triple the pre-pandemic rates reflected in version 4 for the NCSTM. 

The NCSTM5 also incorporates a new truck route choice assignment for long-haul truck movements.  The method was first developed for FHWA’s FAF5 and was found to better reproduce observed truck routing than traditional methods.  The model will be calibrated to ATRI truck GPS data specifically for trucks traveling within, to, from, and through North Carolina.

The final innovation in NCSTM5 is the addition of functionality to allow exploratory modeling analysis and/or scenario analysis of connected and automated vehicles (CAVs).  The new functionality, based in part on recent research by NCDOT, will allow for a variety of changes in response to the new technology including induced trip-making, increased willingness to travel, mode shifts, time-of-day shifts (particularly for long-distance trips), and zero-occupant vehicle trips related to car-sharing and parking avoidance. 

The Ohio Department of Transportation performs annual evaluation and ranking of selected ky projects for project prioritization and funding approval overseen by the Transportation Review Advisory Council (TRAC).  This year they are changing the economic evaluation model to TREDIS (Transportation Economic Development System).  The travel demand data source is the Ohio statewide travel demand model which runs in the Cube modeling software.  An automation process was developed to extract the modeling output and to reformat it as input for the economic model.  The presentation will explanin the process of developing a user-friendly process of extracting the modeling data and show how the economic software is utilized in project evaluation.

The presentation will focus on elements of the analysis, especially unique features of the tools utilized:

• The Ohio statewide travel demand model is a mature activitiy-based model.
• The new process expands the modeling source pool from the statewide model to also include Ohio MPO models including the 3C models (Cincinnati, Columbus and Cleveland) and the smaller OMS (Ohio Medium Small) models.
• The TREDIS software calculates the economic impacts, benefits and costs of proposed projects, programs and policies.

The results of the process resulted in:

• A more transparent and explainable system which is used for important decision-making at the state level.
• Results can be generated at the statewide, regional or county level.
• The new software results are comparable to the old software results.