Estimation of an Activity-Based Model to evaluate Sustainable Mobility Policies in the Netherlands

Abstract

The transition towards greener mobility will play a significant role in decarbonizing the economy. Hence, policy makers need tools that allow them to test alternatives towards that goal. This drive has led to the development of increasingly accurate transport models, with the latest generation being activity-based transport demand models. These models, however, are hard to build, as they are very data-intensive and complex, therefore in this research project a methodology is conceived which attempts to use readily available data from the Dutch travel demand survey (ODiN) and open source software such as ActivitySim, originally developed as a package in Python to make activity-based models in the United States.

While a very time-intensive task, the data was able to be processed for use in ActivitySim. The data was mostly complete, but needed to be complemented with data from the \textcite{centraal_bureau_voor_de_statistiek_statline_2019}, and information about members of the household other than the survey respondent, and joint tours, is missing. The data, however, can be processed in a way that can be reapplied in the future, which lowers the barrier to develop a model with Dutch survey data.

Choice is modeled as logit discrete choice models, and the estimation of the parameters required is facilitated by ActivitySim, which has built-in functionality to support it, and with an integrated workflow the model parameters can be estimated with little effort. With this procedure, and choosing workplace and school locations in advance, a good degree of accuracy was achieved, but it was shown that the sampling method used to deal with the very large choice set introduced significant bias to the model output, as observed in the travel distances, which were shorter in the simulated output than in the observations in the survey data.

While ActivitySim has a sampling methodology to deal with large choice sets, an alternative method, Stratified Importance Sampling with activity spaces, is implemented based on the survey data, from where the sample is determined using the travel distances observed and which produces more accurate outputs when compared to the default sampling.

The result of this research is a framework to easily develop and estimate an activity-based transport demand model that is able to provide insights on the travel demand, and especially on how to influence individual choice behavior, which can facilitate the procurement of quality analysis for decision support in the arena of sustainable mobility, hopefully helping accelerate the mobility transition.

It was concluded that using ODiN data and ActivitySim presented as advantages an easy and replicable formulation, and the availability of data that can be used for sustainable mobility policy analysis; yet, this formulation fails to account for household interactions, something that activity-based models often promise to do, and the documentation provided by ActivitySim while extensive is still inadequate in some regards to understand how to process the data.

The resulting model is, however, highly accurate, despite needing some considerations and improvements. The model needs to sample destination choice alternatives, because otherwise its big size would bring the model to a halt, and it does so using a sampling method that is programmed into ActivitySim. This method was shown to introduce bias to the simulation output even if the choice model was properly estimated, and hence an alternative sampling method based on Stratified Importance Sampling was implemented, and the model output greatly improved as a result. Hence, we conclude that it is possible to obtain highly accurate and efficient activity-based models using available data such as ODiN and open source software such as ActivitySim.

It is argued that a formulation like this can be highly beneficial to the evaluation of sustainable mobility policies, as it lowers the barrier to obtain the accurate and detailed outputs that other models cannot produce, and it provides accurate destination choices that will then inform other submodels that are necessary to evaluate sustainability impact, such as mode choice, travel distance and travel time to evaluate emissions.

We continue by discussing the limitations stemming from the available data and the lack of information for other household members and joint travel, the trend-breaking nature of the COVID-19 pandemic and is impact on mobility in further years, and the possible untested bias of the newly proposed sampling method; and giving recommendations on how to effectively use of a model developed with this framework.

Finally, further research is proposed regarding remote work, the improvement of the choice models, and on the sampling method used.