Domain-Informed Bayesian Modeling of Place-Based Temporal Dynamics

This dissertation demonstrates how probabilistic modeling can bridge the gap between noisy, opportunistic data and longstanding concepts of place. By analyzing temporal activity patterns in the distributional domain, formalizing them through a conceptually-grounded domain-informed Bayesian Dirichlet regression framework, and implementing the approach in the Population Density Tables system, the work advances empirical, methodological, and applied foundations of human occupancy modeling. The result is both a new way to quantify concepts such as activity, affordance, and function, and a globally scalable capability for estimating hourly building occupancy—showing how data science can connect domain informed conceptual understanding with practice in service of pressing societal needs.