過去研究顯示，影響臺灣梅雨季（5〜6月）降雨的機制並非單一。Yimetal. (2015)以多個可能影響 臺灣梅雨季降雨的機制為基礎，利用與這些機制相關的海表面溫度變化趨勢及地面溫度變化趨勢作為預 報因子，對1979〜2005年期間全球降水氣候學計畫(Global Precipitation Climatology Project;簡稱GPCP ) 觀測到的梅雨季臺灣降雨（簡稱GPCP-TWRI)之年際變化現象，建立了多組具物理意義之經驗預報方 程。在本研究中，我們以Yim et al. (2015)研究中所提出之表現最好的預報方程（即前置時間為0-month lead者）為基礎，針對以下議題進行探討：（1)是否可以在Yimetal. (2015)的基礎上，對1979〜2005年 期間地面測站觀測到的臺灣梅雨季降雨（簡稱CWB-TWRI)之年際變化現象，建立具物理意義的經驗 預報方程？ (2)是否可以透過預報因子的調整，找出對2006〜2015年期間的CWB-TWRI之年際變化現 象，更具預報能力的預報方程？（3)與CFSv2 (Climate Forecast System version 2)動力模式預報相比，議題 (2)中所找出的經驗預報方程，是否對CWB-TWRI更具預報能力？針對上述議題，本研究發現可以在Yim etal. (2015)的基礎上，對1979〜2005年期間梅雨季「CWB-TWRI年際變化」現象，建立具物理意義的 經驗預報方程；但若以此預報方程對2006〜2015年期間之CWB-TWRI進行預報，其預報能力表現不 佳。在預報方程的改善方面，我們發現若依海溫變化趨勢特徵，將預報因子中的「北大西洋海溫變化趨 勢之選取區域」進行調整後，所建立的預報方程對「CWB-TWRI年際變化」的預報能力可有效提高。而 在與CFSv2模式預報能力的比對方面，本研究所建立之物理一經驗方程較CFSv2對「CWB-TWRI年際 變化」更具預報能力。這些研究成果，有利於瞭解物理-經驗模式對梅雨季「CWB-TWRI年際變化」預 報的可應用性及改善方法。

This study is a follow up study of Yim et al. (2015), who developed physical-empirical models (PEMs) to predict the GPCP (Global Precipitation Climatology Project)-estimated interannual variation of Meiyu season (May and June) rainfall amount in Taiwan. Analyses of this study focus on three scientific questions, as listed follows. (1) Is it possible to establish a useful PEM to predict the Central Weather Bureau's station-estimated interannual variation of Meiyu season rainfall amount in Taiwan (denoted as CWB-TWRI), based on the prediction factors of 0-month lead PEM equation developed by Yim et al. (2015)? (2) Is it possible to improve the skill of PEM equation developed in the current study for predicting CWB-TWRI, through the adjustment of prediction factors? (3) Is the adjusted PEM equation in scientific question (2) has better skill than the CFSv2 (Climate Forecast System version 2) model for predicting CWB-TWRI? For scientific question (1), our analyses show that it is possible to establish a useful PEM equation for predicting CWB-TWRI, based on the prediction factors of 0-month lead PEM equation developed by Yim et al. (2015). For scientific question (2), our analyses show that through the adjustment of the prediction factor related to the “Atlantic sea surface temperature tendency”，the skill of PEM equation for predicting CWB-TWRI can be improved. For scientific question (3), our analyses show that the PEM equation developed in the current study has better skill than CFSv2 in predicting the CWB-TWRI. These findings provide a better understanding of the potential use and the further improvement of PEM in predicting CWB-TWRI.