Abstract: Spatial and temporal rainfall characteristics have an important impact on flood response in urban areas. To explore the characteristics of flood response in an urbanized watershed, a distributed hydrological model (GSSHA) was constructed in the Jiuxiang River basin, Nanjing, China. Using the stochastic storm transposition method, a large set of rainfall processes was generated and used to simulate flood responses under different underlying surface scenarios. By employing machine learning approach to identify key spatial-temporal storm characteristics, a basin-scale flood response constitutive relationship was identified. GSSHA accurately simulated the characteristics of the flood response in the Jiuxiang River basin, with the spatiotemporal properties of flood-producing storms being crucial in the flood response. Urbanization weakened the dependence of the flood response on storm characteristics. The basin-scale constitutive relationship, with its high prediction accuracy (average relative error of approximately 0.154), provides a mathematical expression for understanding flood response in a small urban watershed. These insights will offer scientific supports for efficient flood prediction and forecasting in urban areas.