Verification results of different precipitation forecast products in the SWAN system and the GZMM model for 4 heavy rain cases in 2010 are discussed. The crossover point of CSI curves in forecast lead time where the model would perform better than the nowcast, is analyzed in particular. Firstly, the radar nowcast for 1hour precipitation forecast exceeding 0.1 mm starts with higher scores than the model, but the skill of the nowcast drops as the forecast lead time increases. The GZMM model performs at a fairly steady level over 3-6 hour period with lower scores at 1 and 2 hours. The CSI scores of 1hour QPF with 0.12°resolution in the SWAN at 1 and 2 hours are 0.504 and 0.442, respectively. At t=3 h, the two CSI curves cross, indicating the skill of the radar extrapolation has decreased to the same level as the model forecast. Secondly, the performance of 1hour precipitation forecast with 3 graded levels is documented. For the 0.1-2 mm rainfall, the CSI score of the model is about 0.23 over the 6hour period, and the GZMM model performs better than the SWAN. For rainfall between 2 and 10 mm, the SWAN product with 0.12°resolution performs better than the GZMM model over 0-3 hour period. At t=4 h, the skill of the radar extrapolation has decreased to the same level as the model forecast. For rainfall over 10 mm, both the SWAN and the GZMM perform not very good. But the skill of the radar nowcast is higher than that of the GZMM model over the 6hour period.