Synthesizing activity locations in the context of integrated activity-based models

Activity-based models are a powerful tool for transportation analysis and represent the future of the industry in terms of modeling techniques. However, the data-hungry aspect of these models makes them difficult and slow to build. This paper presents a set of methodologies to synthesize activity locations for U.S. cities, providing estimates of locations by land-use type in areas with limited available data. The methodology includes a regression method to estimate the number of locations by land-use type complemented by selective use of open data. Detailed information from the entire Southern California Association of Governments (SCAG) area, comprising more than 100,000 km2, is used to calibrate the model. A zero-inflated negative binomial (ZINB) regression is proposed to tackle the excess of zeros in the dataset. The model is estimated using a Bayesian approach that quantifies the coefficients' variability, uses information regarding prior beliefs, and estimates zero-inflated probabilities by zone. The main results suggest that the proposed methodological framework can be used to estimate locations in a fast and efficient way without the need for detailed land-use information. Transportation planners and policymakers can use the results and methods provided in this research to approximate activity location distributions in activity-based models.