As transportation systems become more complex and travel behavior changes, agencies are facing new challenges in planning for the future. Post-pandemic travel patterns, emerging technologies, and shifting policy priorities are all testing traditional assumptions about how people move and how transportation investments should be evaluated.
Travel demand models are one tool agencies use to make sense of that uncertainty. But as David Hensle, a Senior Consultant at RSG, explains, the goal is not to predict the future with perfect certainty. It is to give agencies reliable, data-backed tools they can use to forecast different futures, understand how transportation systems might respond, and make better planning, policy, and investment decisions.
In our recent conversation with David, he shared what makes a model useful, how open-source collaboration through ActivitySim is helping agencies share innovations, and how RSG’s integration of data collection and modeling under one roof helps improve the quality and usability of the models agencies rely on.
Your PhD is in experimental nuclear physics. How does that background shape the way you think about developing travel models?
Believe it or not, the day-to-day work doesn’t look that different. I am still writing code and doing data analysis to try to model an observed process. From a problem-solving perspective, my background fits in quite well with travel modeling.
One concept from my physics education that has stayed with me goes back to my first physics class as a high school freshman. My teacher made us draw a box around our final answer and on each side of the box write “does” “this” “make” “sense.”
This idea has also served me well in travel demand modeling. I am always drawing that metaphorical box around the work that we are producing to make sure each model or development step is valid and appropriate.
Travel models can seem abstract to people outside the field. How do you explain what a good model does for an agency?
This is something we often talk about in our group. To me, a good model needs to do three things well.
First, it needs to answer the questions an agency is trying to ask at a policy level. That means it needs to be sensitive to the things the agency cares about, whether that is vehicle miles traveled analysis, transit ridership prediction, or something else.
Second, the model needs to be usable. It should run in a reasonable amount of time, be easy to update and maintain, be understandable to the people responsible for it, and be something that can be carried forward and enhanced over time.
Third, it needs to be grounded and validated against observed data. In travel demand modeling, that often means household travel survey data, along with validation data such as traffic counts and transit ridership.
If a travel model can do all three of these things well, it’s a great model!
You’ve played a major role in ActivitySim. What developments are you most excited about?
ActivitySim was built primarily to answer the following question: How can public agencies pool resources to create a modeling platform that works for everybody and is maintainable, usable, and extensible enough to handle different agencies with different questions and policy objectives?
Considering the consistent growth of the platform and consortium, I would say that the ActivitySim platform is succeeding in solving this problem.
One current development project that demonstrates the strength of the consortium is a new method for the park-and-ride lot choice model with parking capacity constraint. Original development for park-and-ride lot choice was funded by Oregon Metro, and our TransLink client in Vancouver also pitched in resources to help implement the capacity constraint part. The consortium is now supplying some funds to help pull this enhancement into the main ActivitySim codebase where ActivitySim consortium members like the Puget Sound Regional Council are using the tool.
This is a great example of ActivitySim’s strength: Individual agencies can invest in a new feature, and then the broader consortium can adopt it so others can use it.
I am very excited that we are continuing to see more agencies develop travel demand models using ActivitySim and collectively push the field forward.
How is travel modeling evolving to help agencies plan when the future is less predictable?
Travel demand models cannot predict the future. We do not know with one hundred percent certainty what will happen in 30 years.
What we can do is start from the behavior we observe today and build models with parameters that respond reasonably to inputs such as land use, network attributes, travel times, and costs. From there, planners and model application teams can develop scenarios, apply the model, and see how different policies or changes might affect the overall transportation system.
We also build specific submodels to explore potential future scenarios. For example, we recently did work for SANDAG where we built a model that answers the question: If a household owned an autonomous vehicle, where would that vehicle go throughout the day? It might stay with a worker downtown, or it might drop that worker off and return home, adding deadheading to the network and increasing VMT and congestion. This tool provides a way for planners to see how VMT and congestion changes in test scenarios where, say, parking costs are increased downtown or empty-vehicle driving is priced at a higher rate.
The goal of our travel demand models is not to predict the future exactly. Instead, the goal is to give agencies tools they can use to test different futures and understand how the transportation system might respond.
How has RSG’s integration of data collection and modeling under one roof benefited your work?
I am the primary architect of our survey-data-to-ActivitySim pipeline. That codebase takes the survey data we collect using rMove®, cleans it, codes it, creates tours and linked trips, and turns it into inputs and comparisons we can use to calibrate and validate travel demand models.
Having that work under one roof is invaluable. I can talk directly with the people who worked on a survey we administered and understand the details behind the data: what dwell-time cutoffs were used, when trips are created and when they end, how the data is cleaned upstream, what questions were asked, and how sampling was handled.
It is also important for weighting, which is an often-underappreciated part of survey data collection and is critically important to travel demand modeling. Conversely, one of the primary reasons to do a household travel survey is to collect data for travel demand models. Our modeling group helps to ensure the questions asked in the surveys and the weighting methodology are as useful as possible for our model development efforts.
Having the travel demand modeling group in constant communication with the survey data collection and weighting teams benefits both the models and the surveys with cross-pollination of ideas between the groups.
Looking ahead, what do you hope RSG’s travel modeling work helps clients do better over the next five to ten years?
One thing we have been working toward over the last couple of years, and are starting to see the benefits of now, is a more complete end-to-end platform where we can take survey data and directly output an updated travel demand model, which is not a trivial process.
The goal is to develop models more efficiently and meaningfully, with a strong understanding of the impacts along the way rather than putting everything into a blind pipeline.
The automated calibration tool we are working on with ActivitySim is part of that. So is the pipeline and integration with our household travel survey team. Ultimately, the goal is to reduce the cost of standing up, maintaining, and updating travel demand models for our clients.
If we can do the bread-and-butter work more quickly and cost-effectively, agencies will have more room to explore questions that might have been outside their previous budgets or time frames. This benefits everyone in the long run as planners are then able to forecast future scenarios and respond to changes in policies and behaviors more effectively.
···························
Travel behavior is changing, and agencies need models that can help them test uncertainty, communicate trade-offs, and make better planning and investment decisions. If you’re updating a travel demand model, considering what type of model your next one should be, evaluating new scenarios, or looking for ways to connect better data with more actionable forecasting, RSG can help. Contact us to learn how our data collection and modeling teams can support your next project from start to finish.
