Rainfall-driven non-point source (NPS) pollution has become a critical issue for water environment management in urban watershed systems. However, single-model use is limited to fully represent the intricate processes of rainfall-correlated NPS pollution generation and dispersion for effective decision-making. This study develops a novel cross-scale, multi-factor coupled model framework to characterize hydrologic and NPS pollution responses to different rainfall events in Shenzhen, China, a representative worldwide metropolis facing challenges from rapid urbanization. The calibrated and validated coupled model achieved remarkable agreements with observed hydrologic (Nash–Sutcliffe efficiency, NSE > 0.81) and water quality (NSE > 0.85) data in different rainfall events and demonstrated high-resolution dynamic changes in flow and pollutant transfer within the studied watershed. In an individual rainfall event, heterogeneous spatial distributions of discharge and pollutant loads were found, highly correlated with land use types. The temporal change pattern and risk of flooding and NPS pollution differed significantly with rainfall intensity, and the increase in the pollutants (mean 322% and 596%, respectively) was much larger than the discharge (207% and 302%, respectively) under intense rainfall conditions. Based on these findings, a decision-support framework was established, featuring land-use-driven spatial prioritization of industrial hotspots, rainfall-intensity-stratified management protocols with event-triggered operational rules, and integrated source-pathway-receiving end intervention strategies. The validated model framework provides quantitative guidance for optimizing infrastructure design parameters, establishing performance-based regulatory standards, and enabling real-time operational decision-making in urban watershed management.