Abstract: Against the backdrop of frequent extreme rainstorm and flood events, improving the attribute comprehensiveness and spatial accuracy of urban rainstorm and flood early warnings is a critical approach to reducing urban flood disaster risks. Current early warning methods mostly adopt disaster-causing factors as flood warning indicators, which can only reflect the inherent hazard level of rainstorm floods and apply a single spatial scale for early warning. Responding to the practical demands of urban flood disaster early warning, this paper establishes a multi-dimensional flood disaster loss assessment method and a multi-scale loss early warning system by incorporating multi-dimensional loss attributes including economic losses, affected population, and traffic losses, as well as multi-scale spatial characteristics covering point (grid), line (road), and area (administrative district). A dynamic urban flood early warning paradigm of “flood simulation - loss assessment - multi-scale early warning - real-time update” is further proposed. The application results in Zhengzhou City show that under continuous return periods from 1-year to 100-year, flood losses present a growth trend of rapid initial increase followed by gradual stabilization; the spatial distribution of losses is highly uneven; there are significant differences in early warning information across different scales at the same time and at different times for the same scale. This framework provides important support for strengthening urban flood disaster risk management and enhancing urban adaptive disaster response capacity.