Abstract: Flood forecasting in small and medium-sized rivers (SMRs) presents significant challenges due to inherently short concentration times, limited lead times, sparse monitoring networks, and insufficient flood control infrastructure. Traditional forecasting methods relying solely on ground-based rain gauge observations struggle to meet the demanding requirements of the "Four Pre-actions" (Prediction, Warning, Simulation, and Planning) within digital twin watershed frameworks. To address these limitations, this study proposes a novel real-time, multi-factor flood forecasting methodology and associated accuracy evaluation metrics, explicitly integrating information from the "three lines of flood defense"(3LFD) for rainfall and flood monitoring and forecasting. This methodology was implemented and rigorously tested in the digital twin Huanglei River Basin project. The "three lines of flood defense" framework is defined as follows: (1) The first line: Utilizes meteorological satellites and weather radar observations (measuring atmospheric precipitation) combined with Numerical Weather Prediction (NWP) models to provide extended lead-time rainfall forecasts. (2) The second line: Employs ground-based rain gauge networks to measure surface precipitation, providing accurate spatiotemporal rainfall data for driving hydrological models. (3) The third line: Relies on hydrological stations at critical control points, such as key river sections and reservoirs, to monitor real-time water levels and discharge, enabling dynamic state updating of hydraulic and reservoir operation models. The core innovation lies in the synergistic integration of these three data sources within a real-time rolling forecasting framework. Specifically, NWP rainfall forecasts (1st line) and ground-based rain gauge data (2nd line) are merged to construct a high-resolution, continuous rainfall sequence for driving hydrological models. The resulting runoff forecasts are then coupled with hydraulic routing and reservoir operation models, whose states are dynamically updated using real-time water level and discharge measurements from the 3rd line, to produce forecasts for key control points. The effective lead time for forecasting reservoir inflow peak discharge increased substantially from 3 hours (using only the 2nd line) to 30 hours with the incorporation of 1st line NWP forecasts. The effective lead time for the entire inflow hydrograph forecast was extended by imately 3 hours before the rainfall peak, and for reservoir water level forecasts by roughly 20 hours. Forecast accuracy improved markedly: the average relative error of reservoir water level and discharge forecasts was reduced to within permissible limits (20% of flood stage/flow amplitude). Incorporating 1st line rainfall forecasts substantially decreased average relative errors in discharge predictions for lead times beyond the basin concentration time (6 h, 12 h, 24 h), lowering peak errors from approximately 40% to 20%. Employing 3rd line measured water levels effectively corrected accumulated model biases, reducing peak average relative errors in short-term water level forecasts (3 h lead) from 12% to 4%. Operationally, the integrated forecasts provided critical support for reservoir management decisions. Based on 3LFD-informed forecasts, a controlled release of 2 m³/s was initiated from Huajiatuan Reservoir at 12:00 on August 28. This operation maintained the post-flood water level at 57.81 m, 0.4 m below the forecasted peak and safely under the flood limit (58.20 m), thereby ensuring basin safety. This study establishes a comprehensive technical framework for SMR flood forecasting grounded in the 3LFD concept. It demonstrates that the 1st line provides essential extended lead time, the 2nd line enhances rainfall input accuracy, and the 3rd line enables dynamic model state correction for superior short-term precision. The proposed real-time rolling forecast mechanism, combined with refined accuracy and lead-time evaluation metrics, effectively addresses the central challenges of SMR flood forecasting. The successful application in the digital twin Huanglei River Basin offers a valuable reference for implementing the 3LFD strategy to strengthen flood resilience in SMRs across China and comparable regions worldwide. Future work should emphasize improving the accuracy of satellite/radar quantitative precipitation estimation and forecasts (1st line) and advancing adaptable hydrological models for diverse precipitation inputs and evolving underlying surface conditions.