Abstract Background

The relationship between transportation systems and land use is inherently interdependent, necessitating the integration of land-use models with travel demand models (TDMs) for precise forecasting. However, large-scale implementation encounters challenges such as data availability, geographic resolution, disaggregation, and computational efficiency. This project addresses the need for a robust, large-scale land-use model that seamlessly integrates with TDMs, thereby enhancing forecasting accuracy and planning capabilities.

Description of Abstract

This project introduces the Large-Scale Land Use Model (LS-LUM), a gravity-based land-use model designed to generate essential socioeconomic and land-use inputs for travel demand models. While microsimulation and agent-based land-use models (e.g., UrbanSim and MATSim) offer high accuracy, their extensive data requirements and computational demands limit large-scale applications. LS-LUM overcomes these challenges by employing a Genetic Algorithm (GA) for calibration, enhancing accuracy while reducing computation time.

LS-LUM forecasts population, household distributions, employment sectors, and land-use classifications at the Traffic Analysis Zone (TAZ) level from 2010 to 2050 in five-year increments. Its development was informed by a review of best practices in integrated land-use and transport modeling. Performance evaluation against established models such as the Gravity Land Use Model (G-LUM) and the Transportation Economic and Land Use Model (TELUM) demonstrated superior accuracy through R², Percentage of Good Prediction (PGP), and Mean Absolute Percentage Error (MAPE) measures.

After developing LS-LUM, its performance and accuracy were tested at different regional scales by various agencies, including Rural Planning Organizations (RPOs), Transportation Planning Organizations (TPOs), and Metropolitan Planning Organizations (MPOs) in Tennessee. Based on insights from these implementations, LS-LUM has now been fully integrated with TDOT’s statewide TDM V4 and adopted by various agencies to explore different planning scenarios.

To further enhance accessibility and usability, an interactive online tool was developed, allowing policymakers and planners to visualize and analyze forecasted demographic and socio-economic changes. The tool’s intuitive interface ensures that stakeholders can easily leverage the model’s insights for informed decision-making.

Statement on Why Abstract is Noteworthy

This project has produced a scalable, efficient, and accurate land-use forecasting tool, equipped with a user-friendly interface tailored for large-scale applications. It adeptly addresses the challenges of integrating land-use models with statewide travel demand models (TDMs). This powerful model is instrumental in assessing the cumulative and indirect effects of transportation projects, evaluating the economic impacts of various state and regional policies, and forecasting land-use changes driven by rapid shifts in travel behavior due to emerging technologies. Moreover, it enhances the accuracy of residential location choices by accounting for evolving preferences for greener and tech-savvy lifestyle. The addition of an interactive online tool further empowers decision-makers by facilitating scenario analysis, supporting data-driven policies, and enabling strategic planning efforts at both state and regional levels.

Project is Complete

The Southern California Association of Governments (SCAG) employs the Small Area Secondary Variables Allocation Model, or SASVAM, to disaggregate population and household data into secondary demographic variables, which support a wide range of planning applications, including travel demand modeling, long-range transportation planning, air quality analysis, and project evaluations. SASVAM primarily generates inputs for the population synthesizer (PopSyn), which feeds SCAG’s Travel Demand Model.

This study conducted by Cambridge Systematics (CS) and their project team assessed the effectiveness of SASVAM and provided implementable recommendations to enhance forecasting accuracy. The evaluation incorporated a literature review, interviews with peer agencies, and an empirical analysis of regions experiencing unique growth patterns. The findings revealed several key limitations of the existing tool, including reliance on historical data that may not reflect future conditions, inconsistencies in household size distributions, and population estimates that are incompatible with available survey data. Additionally, the model’s complexity presents a challenge, as it employs hundreds of coefficients, making validation, interpretation, and updates difficult.

To address these concerns, the project team recommended simplifying and improving the SASVAM tool. One improvement involves reducing the number of outputs by estimating age, race/ethnicity, and housing type at a mesoscale level, while maintaining household size and income estimates at the finer geographic scale with a streamlined income classification structure. Additionally, re-estimating the model to reflect the simplified structure and incorporate new data sources would enhance forecasting accuracy. Another recommendation focuses on optimizing computational performance by transitioning to SQL-based joins, improving processing efficiency. A further recommendation is converting SASVAM to Python, streamlining integration with other SCAG tools such as PopSyn and ensuring long-term maintainability.

Beyond these recommendations, two special cases have been identified which require additional refinements: Areas experiencing new development, where the base year population or households are zero but increase in future years, and areas experiencing rapid growth, defined as annual growth rate exceeding 150%, and inconsistencies in household size distributions. For new development and rapid growth areas, the introduction of typologies with predefined characteristics would improve the model’s ability to anticipate demographic shifts in newly urbanized regions. For cases where SASVAM-generated household size distributions are mathematically inconsistent with known population and household counts, implementing a household size disaggregation model would ensure accuracy.  

The proposed enhancements aim to improve the accuracy, usability, and efficiency of SASVAM while maintaining compatibility with SCAG’s existing demographic modeling framework. By streamlining the model structure, updating its estimation process, and optimizing computational performance, SCAG may produce more reliable small-area demographic forecasts, strengthening its ability to support data-driven regional planning efforts.

An important frontier in transportation planning is forecasting Vehicle Miles Traveled (VMT) under current and future transportation and land use scenarios. The most-used tool in this area, regional travel demand models, can provide geographically granular detail about expected future roadway volumes but are known to underestimate induced demand. Filling the gap are simplified tools based on long-run elasticity studies, but these are unable to make geographically specific predictions about VMT in response to network changes. Here we propose and test in multiple metropolitan regions an accessibility-based model, explaining household VMT with a combination of known demographic predictors and fine-scale access to opportunity data. At its core, our framework is simply supply-and-demand: demand (VMT) tracks supply (access).

Most previous models have been agnostic to land use, instead emphasizing household-level demographics: income, household structure, number of work commuters, and auto ownership. By not addressing the effects of the built environment on travel behavior, this research misses one of the “five D’s” explaining travel behavior: access to destinations, here called accessibility. The accessibility of a household’s location is a measure of the integrated land use and transportation opportunities at that place, a summary of the transportation supply in terms of what can be reached. Our analysis combines highly replicated, robust household daily trip data from state of the art surveys, with custom fine-scale access to opportunity measures in order to jointly model the demographic and spatial contexts of households to predict their expected VMT. 

We find a hierarchical modeling framework, where response to multiple types of accessibility is nested within household type (see Figure), is stable and makes reasonable predictions for the scale of impacts of accessibility on VMT per household. Local accessibility (by walk and bike) has a small, negative elasticity indicating that households do on average have lower VMT when they are located in a highly walkable area. Accessibility by public transport has a stronger negative elasticity, which helps explain why households in neighborhoods of high transit access have lower VMT. Regional accessibility, defined by the amount of destinations reachable by car between 20 and 60 minutes driving, has a much stronger, positive elasticity with VMT, providing further evidence that expansion of freeway access and longer-distance regional auto connections results in higher VMT from households, all else held constant. These effects vary by household characteristic, and by region in the study.

This modeling approach and initial results will be of broad interest to planners and modelers interested in how changes to land use and transportation networks affect VMT. Specifically for states under mandate to predict and mitigate future VMT from construction projects (including the Minnesota state department of transportation), the approach provides a middle ground between the travel demand models which offer specificity but low accuracy in induced demand, and the lane mile elasticity calculators which offer long-run estimates without geographic context.

Understanding how riders access transit stations is a critical challenge for transit agencies seeking to improve mobility and equity. The Maryland Transit Administration (MTA), in partnership with Foursquare ITP and WBA Research, conducted an innovative study using trip chain simulation to analyze first-and-last mile connections to transit stops. This study leveraged MTA’s on-board origin-destination (O-D) survey data and applied advanced trip chain simulation and validation techniques to enhance the accuracy of reported travel patterns.

Traditionally, the accuracy of on-board survey responses is limited to a respondent’s recollection of their own trips – respondents often misreport routes, provide incomplete descriptions of their trips, and inaccurately report travel times or walking distances. These human factors create challenges in for transit agencies trying to accurately assess how riders connect to and from transit stations using O-D survey data. The trip chain simulation and validation address these issues by systematically verifying whether reported trips are feasible within the transit system. The tool also provided MTA accurate geographic information on where the respondent traveled, beyond simple origin and destination data, such as trip routing and transfer points. By simulating actual travel patterns based on recorded survey responses, trip chain simulation and validation ensures that trip sequences align with transit schedules, network connectivity, and logical routing.

This advanced approach enabled MTA to conduct a much more detailed and reliable analysis of first-and-last mile access than was previously possible. By refining the survey dataset through trip validation, the study provided a clearer picture of how riders reach transit stations, including differences across need groups such as low-income, elderly, or transit-dependent populations. The validated dataset allowed for in-depth assessments of transfer locations, walk times, and multimodal connections, offering actionable insights to improve station access and inform transit planning.

Through this case study, MTA demonstrates how advanced travel simulation techniques can enhance survey-based transit research. By ensuring data integrity and enabling a deeper understanding of access patterns, this methodology provides transit agencies with a powerful tool to address first-and-last mile connectivity challenges and improve service for all riders.

Authors: 

Subid Ghimire, PhD, Transportation System Modeler, North Central Texas Council of Governments, sghimire@nctcog.org

Eleni Bardaka, PhD, Associate Professor, North Carolina State University, ebardak@ncsu.edu

Microtransit is an emerging mode of public transportation expected to significantly improve access to employment and other essential destinations for disadvantaged populations living in low-density areas. Although many transit agencies in the United States have implemented microtransit pilots in recent years, our understanding of these systems is very limited due to lack of empirical research. Utilizing trip data from microtransit systems in North Carolina, this study examines the relationship between the number of microtransit passengers originating from or arriving at a census block and the socioeconomic and built environment that generates or attracts these passenger trips. First, we estimate hurdle negative binomial models on the number of passengers traveling from or arriving at a census block in a specific hour of a weekday, aggregated over months for five different microtransit systems in North Carolina. To enhance our understanding of service characteristics that would contribute to greater demand, we combine data from the five different service areas to incorporate their distinct service characteristics as independent variables in our models. The models estimated separately for the five microtransit systems indicate that neighborhood socioeconomic attributes have a significant effect on demand. More specifically, areas with a higher percentage of females, younger populations between age groups 18-44, and carless populations contribute to greater demand across multiple service areas. In addition, the number of low-wage and healthcare jobs is also positively correlated with an increased number of passengers traveling from or arriving in a census block. For instance, a 1% increase in the number of low-wage jobs in a census block is associated with up to 0.109% greater probability of a passenger pick-up and up to 0.105% increase in the number of passenger pick-ups. The econometric analysis of the pooled data from all microtransit services reveals that, after controlling for temporal, socioeconomic, and built environment factors, service characteristics, such as fare, acceptable wait times, pick-up/drop-off locations, and monthly service hours, are important determinants of microtransit demand. Among these variables, service hours and acceptable waiting times set by the transit agency display the greatest effect (elasticity). For example, a 1% increase in acceptable waiting time is associated with a 0.188% decrease in the number of passengers pick-ups. Furthermore, the results also suggest that, if other factors are held constant, a microtransit service with door-to-door pick-up or drop-off service would lead to greater passenger demand. An important finding from this study is that microtransit use tends to be more sensitive to service reliability (waiting times and service hours) than fare levels, likely because, in small cities and low-density areas, it is often the only viable mobility option for low-income and carless populations. In sum, the findings of this study identify common socioeconomic and built environment factors that drive higher microtransit demand across different service areas and offer valuable insights into the service characteristics that would contribute to higher demand. The findings of this study will assist transit agencies in evaluating the suitability of microtransit for their jurisdictions and in designing service characteristics to enhance the efficiency of microtransit operation.