From Static Points to Dynamic Fields: Reconstructing Spatiotemporal Boundaries of Places via AutoEncoder and Mobility Big Data

Traditional Geographic Information Systems (GIS) treat places as static points or polygons, ignoring dynamic boundary changes over time. This study introduces an AutoEncoder-based AI framework to reconstruct the spatiotemporal boundaries of places using mobility big data. The model converts human mobility data in neighboring grid cells into a latent space embedding to identify functional homogeneity. Low latent space distances indicate similar synchronized mobility patterns, suggesting they belong to the same functional dynamic field; while high latent space distances signify behavioral divergence, marking the boundaries of a place. Validated with telecom data collected in Taipei, Taiwan, the framework’s effectiveness is demonstrated through three case studies: seasonal spatial extent dynamics around National Chengchi University, diurnal spatial extent dynamics around the Wenhu Station of Taipei’s rapid transit system, and event-driven spatial extent dynamics around the Taipei Arena. The results validate the efficacy of the proposed framework in delineating dynamic urban rhythms and uncovering fluid functional boundaries that characterize cotemporary urban life.