Authors: Michael Onuogu, Heng Wang, Xueting (Sherry) Chen

Co-authors: Aaron Hekele, Joel Freedman, Hannah Carson, Ali Etezady, Rohan Sirupa, Edna Aguilar

Background to Problem

Accurately modeling airport ground access travel is critical for transportation planning, demand forecasting, and policy evaluation. To improve airport travel modeling, the Houston-Galveston Area Council (H-GAC) conducted a comprehensive airport ground access survey in 2023 at George Bush Intercontinental Airport (IAH) and William P. Hobby Airport (HOU) in Houston, Texas. The survey collected detailed data on passenger characteristics, access modes, and trip patterns, providing critical insights for enhancing travel demand models.

Description of Application

The primary objective of this project was to develop an improved airport ground access sub-model within ActivitySim, an open-source activity-based modeling (ABM) platform. This model builds on the framework initially developed by the San Diego Association of Governments (SANDAG) and has been customized for the Houston region. It provides a more detailed representation of airport access travel by incorporating residency status, trip purpose, and mode choice, enhancing the accuracy of demand forecasts.

The 2023 survey identified key travel patterns. Among travelers, 64% used private automobiles, 6% relied on transit, and 28% used shared auto modes such as taxis, ride-hailing services, or hotel shuttles. Trip purposes varied: 36% of trips were resident personal travel, 26% visitor personal travel, 18% visitor business travel, and 15% resident business travel. Travel times also differed, with 49% of passengers reaching the airport within 30 minutes and 43% taking between 30 minutes and one hour.

The activity-based airport sub-model was developed using this dataset and fully integrated with H-GAC’s Trip-Based Model (TBM) and Activity-Based Model (ABM). The model generates travel demand for airport passengers. Unlike traditional 4-step models, this disaggregate framework simulates individual travel behavior with greater accuracy, improving the representation of passenger trips to and from non-airport locations.

Statement on Why Application is Noteworthy

This sub-model allows planners to test various policy and operational scenarios, including changes in parking pricing, transit service expansions, ride-hailing regulations, and congestion management strategies. It provides insights into how shifts in parking costs or transit fares impact passengers’ mode choices, helping decision-makers optimize airport access and connectivity.

This presentation will summarize the findings of the 2023 H-GAC Airport Ground Access Survey and demonstrate how they informed the development of the activity-based sub-model. The presentation will provide an overview of the model’s structure, calibration results, and applications. By integrating advanced modeling techniques with robust survey data, this work enhances the precision of airport ground access forecasting and provides a valuable tool for transportation planning in the Houston region.

The project was completed in April 2025.

Authors: David Ory, Rebekah Straub, Zhujoun Jiang

The Ohio Department of Transportation is designing the next version of the activity-based modeling system used by their large MPOs, which is called the “3C” model (as it’s designed for Columbus, Cincinnati, and Cleveland, as well as other regions). One short-coming of the current 3C design is that it includes a large number of probabilistic models with a large number of indirect effects. For example, the work tour scheduling model includes coefficients on gender, age, occupation, and presence of a non-working adult in the household. Each of these variables serves as a proxy for some type of constraint on behavior. For example, a worker with a non-working adult in the household is more likely to have assistance with caretaking responsibilities. Nearly each step in the modeling system includes probabilistic models with similar indirect effects. 

There are two problems with this approach. First, it reduces the legibility of the modeling system. For example, it’s not always obvious what gender is serving as a proxy for in all of the models in which it is included. As gender roles evolve over time (or as scenario planning wants to understand the impacts of gender roles changing), model users are left to guess if the effect of gender in each model is likely to persist.

Second, approaching model design in this way reduces the computational efficiency of the modeling system, as it overstates — in many cases dramatically — and misplaces our uncertainty in behavior. For example, a non-trivial share of workers have a fixed work schedule (e.g., teachers, manufacturing workers, retail workers). In crafting a work activity or tour scheduling model, the first question we should ask is whether the subject worker is constrained in this choice by having a fixed work schedule. If the answer is yes, we need not consider the person’s gender or age or occupation or presence of a non-working adult. We simply need to assign this person one of the fixed work schedules we observe in the training data. This approach improves the computational performance of the system. 

In many cases the behaviors we are uncertain about are the constraints each person is under when making travel-related choices. Our travel modeling systems should reflect this by simulating constraints explicitly with probabilistic model forms. For example, the 3C version 2 modeling system will have an explicit “does this worker have a fixed work schedule” component. Downstream component forms, such as work activity scheduling in this example, can act on these constraints using, very often, deterministic or simple model forms. 

The challenge in implementing this approach is understanding and modeling constraints explicitly. This is the challenge ahead for the 3C version 2 implementation. 

Ohio DOT is in the early stages of implementing a version 2 model in Bentley’s AGENT platform. Work completed to date includes a design document, early implementation activities in AGENT, select prototype model components, and early data exploration of an on-going household travel survey. 

Topic

Travel demand model development

Background to Problem

Activity-based models (ABMs) commonly include a work location choice model.  Normally, the agents are individual employed persons given as part of residential demographics, and the choice set is spatial units with given numbers of employees, of various industries or types.  For model coherence, the number of workers modeled as choosing each workplace should equal the number of employees given at each.  Many ABMs model school location choice similarly.

It is well known that the maximum entropy solution is the logit model with endogenous alternative-specific constants, or “shadow prices.”  Its well-known iterative solution is reliable upon continuum models (classical trip distribution, iterative proportional fitting), but requires adaptation to work upon discrete simulations.  Those in common practice were developed by limited and mostly unpublished investigation.  One common method was also recently found to exaggerate stochastic variation (differences among runs with different random seeds). 

“Clearinghouse” matching models are not pursued, such as serial dictatorship (deducting each chosen job from succeeding agents’ choice sets) or the method in ActivitySim similar to rank maximal matching.  These are distinctly different models, not just alternative computational methods.   They have been promoted for computational expedience, with no claims as better representations of labor markets.

Description of Application

With an experimental focus, this study pursues both a clearer understanding of the problem of supply-constrained choice by discrete simulation, and methods to solve shadow prices for closer worker-job fit reliably and in minimal runtime.

Original code dedicated to a workplace location problem was developed to enable a wide range of comparable tests of alternative methods upon given data.  The methods vary by

·         Shadow-price update formula,

·         Monte Carlo vs. random-utilities methods of simulation,

·         Updating the shadow-prices after modeling portions, vs. the entirety, of agent set.

Comparisons are presented for various measures versus iterations or cumulative portions of the population modeled.  The measures compared are:

·         Agreement between jobs chosen and jobs given by location,

·         Stochastic variation of jobs chosen,

·         Stochastic variation of shadow-prices.

Some highlights from the analysis:

·         Common methods “bottom out” after which further iterations bring no improvement.

·         Some novel shadow-price adjustment formulas achieve closer and faster convergence than existing methods tested.

·         Simulation by frozen random utilities improves convergence over Monte Carlo simulation.

·         A signal-to-noise criterion successfully chose sufficient sample sizes from the population during runtime, to enable faster convergence.

·         Shadow-prices can be solved quickly with little stochastic variation by using conditional probabilities in place of discrete outcomes.  But single outcomes drawn afterwards have more than Poisson-distributed random error.

·         Closer fit reduces stochastic variation.

This study is the first to examine application of the supply-constrained problem by discrete simulation in depth, comparing both common and novel methods. 

The National Capital Region Transportation Planning Board (TPB), the federally designated MPO for the Metropolitan Washington region, has been developing its next-generation, activity-based regional travel model, known as the Generation-3, or Gen3, Model since late 2019. During the first two phases of the Gen3 Model development project, the model was developed using the ActivitySim platform with consultant assistance. Now, TPB staff are currently conducting the third and final phase of this project, which features usability testing of the model to ensure its readiness for production work.

Usability testing of the Gen3 Model includes sensitivity testing and simulating its applications in the development of the Visualize 2050 plan, including the Air Quality Conformity (AQC) Determination, mobile emissions analysis, and the performance analysis of the plan. Sensitivity tests were conducted in both the Gen3 Phase 1 Model and Gen3 Phase 2 Model, mainly to evaluate the reasonableness of model response to changes in model inputs, such as land use forecasts, planned transportation networks, and/or transportation policy.  During Phase 3, staff conducted additional sensitivity tests in support of the usability testing, aiming to: 1) Compare the Gen3 Model to TPB’s current production-use, trip-based Gen2 Model by examining the responses of the two models to the same or largely consistent set of changes to model inputs, and 2) Showcase the Gen3 Model’s unique capabilities for in-depth analysis with disaggregate data.

TPB staff conducted a series of hypothetical sensitivity tests and evaluated the model responses of both the Gen2 and Gen3 models. The first test looked at the impact of having a certain market share of autonomous vehicles in the metropolitan Washington region in a horizon year. The second test examined the regional impact of increased transit subsidy availability for full-time workers. The third test examined the model responses to a hypothetical cordon pricing scheme in downtown Washington, D.C. The fourth test evaluated the likely effects on travel of increased telecommuting in the metropolitan Washington region. TPB staff propose to present the results from these sensitivity tests and share findings and reflections on this modeling exercise with the conference attendees. Head-to-head comparisons of trip-based and activity-based models are not that common, so it is hoped that this study will be of interest to transportation planning practitioners.