Urban spatial networks are complex systems with interdependent roles of neighborhoods and methods of transportation between them. In this paper, we classify docking stations in bicycle-sharing networks to gain insight into the spatial delineations of three major United States cities from human mobility dynamics. We propose novel timedependent stochastic block models, with degree-heterogeneous blocks and either mixed or discrete block membership, which (1) detect the roles served by bicycle-sharing docking stations and (2) describe the traffic within and between blocks of stations over the course of a day. Our models produce concise descriptions of daily bicycle-sharing usage patterns in urban environments. They successfully uncover work and home districts, and they also reveal dynamics of such districts that are particular to each city. When we look for more than two roles, we uncover blocks with expected uses, such as leisure activity, as well as previously unknown structures. Our time-dependent SBMs also reveal how the functional roles of bicycle-sharing stations are influenced by surrounding public transportation infrastructure. Our work has direct application to the design and maintenance of bicycle-sharing systems, and it can be applied more broadly to community detection in temporal and multilayer networks.