The predictive capabilities of three nowcasting models (S-PROG, Extrapolation, and ANVIL) for extreme rainfall induced by tropical cyclones were evaluated using the 2017 Hurricane Harvey rainfall event as a case study. The results indicate that the ANVIL model shows the best performance in simulating the spatial distribution of rainfall, but systematically overestimates heavy rainfall intensity; the S-PROG model shows a lag in tracking rainfall intensity variations; and the Extrapolation model achieves the highest stability in short-term (0-2 h) nowcasting, accurately capturing the peak rainfall intensity and its temporal evolution. The fusion of high-resolution meteorological data and multi-source data can significantly enhance model performance, and the combination of radar extrapolation techniques and deep learning algorithms shows great potential.