Abstract: Using a combination of Hubei weather radar network puzzle data and precipitation observations from Wuhan and surrounding ground-based encrypted automatic stations from March to September 2018, fast fusion and other techniques were used to establish dynamic Z-I relationships suitable for Wuhan localized optical flow method and TREC method to obtain high spatial and temporal resolution radar quantitative precipitation inversion data 2 h QPF in real time, and the data SWAN 1 h QPF was used to compared and analyzed. The results show that, in general, the optical flow method based on radar extrapolation and automatic station fast fusion technology is the best, and the TS integrated average is about 3 percentage points higher than that of SWAN. The three methods of QPF with 9-point or 25-point averaging can improve the relative and absolute errors, but have less contribution to improving the forecast accuracy, while the neighborhood method can improve the forecast accuracy significantly and also reduce the rate of empty and missed reports to some extent. In a comprehensive comparative analysis of the regional and local dispersion processes, the optical flow method is the best, the TREC method is the second best, and the SWAN method is the least effective in the magnitude below heavy rainfall. The greatest differences are mainly concentrated in the regional processes. At the heavy rainfall and heavy rainfall magnitudes, although SWAN has higher scores in the regional processes, it performs unstable in the local processes, while the optical flow method is relatively more stable in the local processes.

Key words: fusion technology, 2 h QPF, optical flow method, TREC method, neighborhood test