子計畫:梅雨期台灣地區中尺度降雨氣候與概念模式之建立-CFSv2的應用(II)

This project evaluates the capability of a physical-empirical (PE) model, which is constructed based on the three prediction factors (named WNPT, NATnew, and EAT) modified from Yim et al. (2015), in predicting the interannual variation of four different types of Meiyu season rainfall in Taiwan. In addition, we compare the PE model’s prediction skill with the CFSv2’s prediction skill (at lead time day 1 to day 45; hereafter LT1-45) to better understand the capability of CFSv2 in predicting the interannual variation of different types of Meiyu season rainfall in Taiwan. Our analyses show that the PE model, which constructed based on the prediction factors of WNPT, NATnew and EAT, only has skill in predicting the interannual variation of the characteristics (including amount and frequency) of “total rainfall” and “frontal convection (FC) type of rainfall” in Taiwan. However, for the interannual variation of the rainfall amount of diurnal convection (DC), typhoon convection (TC) and other southerly convection (SC), those PE models constructed using the prediction factors of WNPT, NATnew and EAT do not have good prediction skill. Further investigation is suggested in the near future to find out more suitable prediction factors for constructing the related PE models in predicting the interannual variation of DC, TC and SC types of rainfall amount in Taiwan. On the other hand, for the performance of CFSv2, our results show that overall CFSv2 is able to “qualitatively” depict the interannual variation of total rainfall amount, as well as the interannual variation of FC, TC and SC types of rainfall amount, in Taiwan at LT1~45 (i.e., all of the related temporal correlation coefficients between the CFSv2 predicted rainfall and the observed rainfall are significant at the 90% confidence interval). However, for the DC types of rainfall, the CFSv2’s prediction skill is not good. Quantitatively, the analyses of skill score (TS、BS) show that CFSv2 has a better skill in predicting the formation of FC types of rainfall compared to the other types (including DC, TC and SC) or the total rainfall amount, and the forecast skill is best for LT1~5 among LT1~45. These findings are important to understand the potential use of “CFSv2” and “PE model” in predicting the interannual variation of different types of rainfall formation in Taiwan during the Meiyu season.