UC Santa Barbara

Activity-based models for travel behavior are an important tool in urban planning and transportation analysis because they simulate the lives of individual people minute-by-minute and mile-by-mile throughout urban space. These models produce realistic schedules that take into account each person’s work, school, and personal life while also abiding by constraints imposed by time, space, and the need to be in the same place at the same time as other people. While these models have made great strides in accurately representing the ways people and households schedule activities throughout the day, they do not do as good a job at understanding the interconnections between space / place, what activities people do, and when they do them. One factor that contributes to this shortcoming is a general mismatch between the spatial distribution of activities and opportunities that these models consider and the spatial units used in modeling. This dissertation seeks to improve this aspect of spatial choice models by using a network-distance variant of density-based spatial clustering methods to extract activity centers by clustering the locations of entertainment, food service, and retail businesses.

This sort of spatial clustering requires an accurate means of identifying neighboring points in addition to well-chosen clustering parameters. Various simplified methods for calculating network distance are compared for their accuracy at measuring distance and identifying neighbors, and the least simplified method is chosen. A range of clustering parameters are tested, their results compared, and a final clustering is chosen that balances the need to have small, discrete clusters while also capturing as many businesses as possible and matching most activity locations.

The types, timings, and durations of activities matched to different clusters are explored in order to assess the potential effectiveness of these clusters. This analysis identifies a set of center-level metrics that influence activity participation and timing. Finally, the spatial variability of activity durations and travel times is investigated using hierarchical models based on the clusters and spatially autoregressive models. These models indicate that much of the spatial autocorrelation of activity duration can be understood as primarily reflecting differences in the opportunities available at the level of individual centers.