Abstract: A flood forecasting method coupled with numerical weather prediction and multi-objective optimization for parameter calibration is presented to improve accuracy and lead time of flood forecasting. In this method, the epsilon-dominance non-dominated sorted genetic algorithm II (? - NSGA II) is used for multi-objective auto-calibration of the Distributed Hydrology Soil Vegetation Model (DHSVM), and ensemble averaging precipitation data from the European Centre for Medium-Range Weather Forecasts (ECMWF) is adopted to drive DHSVM for flood forecasting. The results demonstrate that for overall flow, forecasts of lead time within eight days are reliable and the relative mean error (RME) is within 20%. Compared with traditional deterministic forecasting, adopting ensemble precipitation forecasts can prolong the lead time of flood forecasting and provide an effective way for developing flood forecasting methods.