利用臺灣中央氣象局(CWB)所提供的全球預報模式(GFS)以及美國國家環境預報中心(NCEP)的航空模擬預測系統(AVN)，來進行WRF颱風路徑預報。本研究使用臺灣CWB GFS和美國NCEP AVN之分析場和預報場作為WRF 模式的初始條件和邊界條件，進行92年杜鵑颱風和93年敏督利颱風WRF模式0～72小時颱風路徑之預報。對這兩個颱風個案，WRF模式對實際颱風路徑的預報能力良好。另一方面，將WRF模式預報輸出，利用經驗正交函數(EOF)分析單一時間點上環境流場垂直分佈之特徵向量，求得代表性駛流場，並配合風場的系集採樣深層平均，除去颱風渦旋求取其不對稱量交互驗證，探討颱風路徑預報誤差因素。利用EOF分析可得知，東西風及南北風方向駛流風場分量的貢獻量，此衍生之駛流方向與路徑變化趨勢頗為一致，亦與風場不對稱量大致吻合，但後者在敏督利颱風路徑向北轉折前滯留時並未顯示鞍型場之風場。本文EOF分析可以客觀地顯示環境流場變化之方向，可用於輔助對颱風路徑預報誤差之了解。

Using GFS forecast by Central Weather Bureau (CWB) of Taiwan and AVN forecast by National Center for Environmental Prediction (NCEP) of USA, we employ the Weather and Research Forecast (WRF) model to predict typhoon tracks. In this study, both analysis and forecast of CWB GFS and the NCEP AVN are used to provide initial and the boundary conditions for 72-h track prediction of Typhoon Dujuan (2003) and Typhoon Mindulle (2004). For the two typhoon cases, WRF predicts the typhoon tracks very well. On the other hand, we utilize empirical orthogonal function (EOF) to analyze the vertical distributions of characteristic vectors for environmental flow at various times, to obtain the principal variances, and we use the deep-layer mean of ensembles to retrieve asymmetry, by removing the typhoon vortex, for mutual verification for typhoon track prediction errors. The EOF analysis reveals that the directions of the principal variances of east-west wind and the south-north wind components are consistent with the tendency of typhoon track changes and also primarily agree with the asymmetry, but the latter (i.e., asymmetry) does not exhibit the saddle pattern of the wind when the stagnating Mindulle is going to turn northward. The EOF analysis of this study can provide objective information about the changing direction of the principal variances of environmental flow, which is helpful for understanding of typhoon track prediction errors.