Taking the Beijiang River basin as the study area, a water cycle prediction model was constructed by coupling the Variable Infiltration Capacity (VIC) model with the downscaling results of 13 CMIP5 global climate models. Then the Standardized Evolutionary Stress Ratio (SESR) index was used to predict the main characteristics (frequency, duration, and intensity) of flash drought and the corresponding uncertainties in the Beijiang River basin under three emission scenarios (RCP2.6, RCP4.5, and RCP8.5) in the future (2021-2050). The results show that the VIC model has good runoff simulation accuracy (Nash efficiency coefficient>0.93) in the Beijiang River basin, and can better reproduce the interannual and intra-annual variations of runoff. Compared with the baseline (1971-2000), the frequency of flash drought in the future (2021-2050) is generally characterized by a spatial distribution of decreasing in the west and increasing in the east, and the increase in frequency tends to be larger under the higher scenarios (<80%). The northeast and southern regions of the basin are expected to encounter flash drought events with higher intensity and longer duration. The uncertainty of global climate models shows strong spatial heterogeneity, and the projection uncertainty of different flash drought characteristics varies greatly in space.