The invention provides a numerical weather prediction method based on a mixed ambient field. According to the method, a mixed data ambient field of numerical weather prediction is generated based on the existing data of the relevant weather data and forecast data; initial parameters required in establishing a WRF mode are set to complete initial establishment of the WRF mode; effective meteorological data on feature level grid of the WRF mode is obtained; in the WRF mode, conclusion parameters of the numerical weather prediction is obtained by calculation according to the effective meteorological data; the conclusion parameters are extracted and processed to obtain conclusion data diagram of the numerical weather prediction. The present invention provides the mixed data using the reanalyzed data and the predicted background field as the background data for driving the meso-scale numerical weather prediction for numerical weather prediction; as the reanalyzed data itself assimilates a large number of observation data and the accuracy is high, which can significantly improve the accuracy of numerical weather prediction and save a lot of computing resources.