Frequent extreme rainstorms driven by global climate change, together with the complex flow dynamics in mountainous river channels, have rendered flash flood early warning a persistent scientific challenge. While various hydrological and hydrodynamic models commonly adopted for flood prediction each have distinct focuses, none can effectively underpin small-scale flash flood early warning that demands high-resolution topographic data. Taking the Yintanxi Watershed as a case study, this study developed a coupled hydrological-hydrodynamic model based on high-precision topographic data to enhance the capacity of flash flood early warning and risk assessment, established a flash flood early warning indicator system for the watershed, and clarified the distribution of flash flood hazard zones, evacuation routes and emergency shelters. The results demonstrate that: the Xin'anjiang Model was applied to simulate four typhoon-induced flood events from 2015 to 2021, yielding relative errors of less than 20% for both runoff depth and peak discharge with coefficients of determination exceeding 0.7, which verifies its high applicability and superior performance in event-based flood simulation; coupled 1D and 2D unsteady flow simulations were conducted for the flood event on 21 July 2021 via the HEC-RAS Model, and grid cell discretization revealed that the primary flood inundation areas were low-lying river reaches and floodplains; a flash flood early warning indicator system for the Yintanxi Watershed was constructed using the water level (discharge) back-calculation method, and the results show that critical rainfall indicators increase with the extension of the warning period but decrease with higher initial soil moisture content; flash flood risk exhibits a gradual decline from the river channel outward, and regional differences exist in the distribution of flash flood hazard zones, evacuation routes and emergency shelters among villages during flood events due to variations in their topographic and hydrological conditions. Addressing the technical deficiencies of flash flood early warning in small watersheds, this study optimized the technical methodologies from three dimensions—model coupling, dynamic early warning and closed-loop disaster prevention—and developed a full-process technical system for flash floods in small watersheds of southern hilly and mountainous areas, which can provide technical support for disaster prevention and mitigation in similar watersheds across China.