Human exposure to threats from natural hazards is generally estimated using a static approach with the fixed number of people located in hazard-prone zones; however, in reality this number varies due to population mobility. This study proposes a human–hazard coupled City Model (HazardCM) for accurately calculating city spatiotemporal dynamic exposure to different hazards. It includes four components: an urban environment module, agent-based model, city–hazard coupler, and dynamic exposure assessment. Rainfall-triggered natural hazards under extreme hydrometeorological events were modeled in Lishui, China. Scenarios covering different magnitudes, timings and locations, and return periods of hazards were investigated to derive the spatial distribution and evolution of human exposure. This model is the first that different natural hazards have been analyzed within a unified framework using a dynamic method and offers a new way to investigate exposure's space–time characteristics while considering the dynamic nature of both humans and hazards.