Abstract: By using the latest Meiyu data developed based on national Meiyu monitoring criteria and taking Meiyu over the Yangtze River as an example, the multi-scale variation of Meiyu is analyzed and the predictability of Meiyu from the perspective of external forcing of sea surface temperature (SST) is investigated. The predictability of Meiyu anomalies on the interannual time scale is discussed by combining the variability of background SST and the hindcast of a predictive model. Results suggest that Meiyu over the Yangtze River shows a long-term decreasing trend and multi-scale quasi-periodic oscillations including 3-4-year, 6-8-year, 12-16-year, 32-year and 64-year. The 3-4-year quasi-periodic variation is the main component of Meiyu anomaly. Conversion between dry and wet phases of Meiyu is modulated by the 12-16-year quasi-periodic oscillation. The extreme flood Meiyu usually occurs simultaneously with the wet phase of the 12-16 year oscillation and the peak phase of the 3-4 year oscillation. Different components of Meiyu over the Yangtze River correspond to different SST external forcing background. Key regions of SST associated with interannual variation of Meiyu are located in the tropics. While Meiyu variations on longer time scales including inter-decadal and multi-decadal are related to SST in the middle and high latitudes. The SST signal of the 3-4-year quasi-periodic component converts from ENSO (El Niño-Southern Oscillation) in the preceding winter to IOD (Indian Ocean Dipole) in late spring and early summer. SST key regions of 6-8-year and 12-16-year oscillations are mainly located in the Pacific. 32-year and 64-year oscillations are influenced by multi-decadal changes of the North Pacific (Pacific Decadal Oscillation, PDO) and the Atlantic (Atlantic Multi-decadal Oscillation, AMO). The long-term changing trend of Meiyu is associated with both the global warming and the decadal change of SST especially PDO. Although the positive correlation between Meiyu anomaly and ENSO shows a decreasing trend, the predictability of the interannual variation of Meiyu anomaly has improved since the 1970s. Finally, a prediction model of Meiyu anomaly is established by combining models of multi-scale components. Independent samples test of Meiyu anomalies in the latest 5 years exhibits an encouraging hindcast performance, which verifies the stability of predictability of interannual component of Meiyu anomaly and the superiority of Meiyu prediction model based on multiple time scale separation.