對於颱風的預報，目前改進颱風的初始條件是研究的關鍵。所以和颱風相關的觀測資料將會於颱風的初始化中使用，以達到調整初始場的效果。本研究中利用MM5 4DVAR模式，同化德國CHAMP衛星及福衛三號的GPS 掩星觀測資料與加入虛擬渦旋，探討對於初始分析場的改進在颱風預報的影響。 文中選取2003年的杜鵑颱風個案與2006年的珊珊颱風個案，來進行各個實驗設計的研究。模擬實驗中利用NCEP/AVN全球分析資料做為初始的分析場，加入同化的資料，藉由四維變分(4DVAR)資料同化方法能把非傳統觀測資料結合進入模式裡，而且符合模式本身的動力及物理條件，使得模式的變數場和觀測資料之間的差距極小，達到藉由實際的觀測資訊去調整模式的變數場。研究結果顯示出加入GPS掩星折射率資料有改善初始場，只是調整的程度都不大，似乎僅對於調整外圍的環境場有幫助。而加入虛擬渦旋資料的方法，對於解析度不足的初始場則有很大幅度地調整，明顯地加強了颱風的結構特徵，可以提供較為接近實際且強勁的颱風。最後在結合GPS掩星折射率資料及虛擬渦旋的部份，因為GPS折射率資料沒有落在颱風範圍且修正量以溫濕度為主，所以顯示出虛擬渦旋導入氣壓場及風場大幅度地調整颱風本身結構，似乎比GPS折射率資料有更大的影響，而主導著預報的結果。 由於BDA改善了初始颱風渦旋的強度及結構，在中心路徑預報方面有明顯的助益，大大改善了其72小時路徑誤差。同時在雨量方面，其降雨分佈和累積雨量與觀測相較更接近許多。本文亦改善BDA方法，同時加入三維的梯度風平衡風場，結果顯示可以維持整層颱風渦旋，較原本BDA（只同化海平面）的結果更接近觀測，在路徑方面的改善最為明顯，其72小時路徑誤差大約只有60公里。 Improvement of the initial conditions of typhoon plays the crucial role in the prediction of typhoon. Therefore, observations helpful for analyzing a typhoon should be used to adjust initial fields. This study employs the MM5 4DVAR model not only with CHAMP and FORMOSAT-3 GPS radio occultation data but also with bogus data assimilation (BDA) to provide the optimal initialization that gives a positive impact on typhoon prediction. In this paper, two typhoon cases, Dujuan (2003) and Shanshan (2006), were chosen for such an impact study. In the simulation experiments, the NCEP/AVN global analysis is used as the initial fields. Then, we conduct 4-dimensional variational (4DVAR) data assimilation which incorporates GPS refractivity and bogus observations into the model. In such a way, with dynamical and physical constraints of the model, the deviation of the model state from the observations will be minimized and optimized. The assimilation results show that the GPS refractivity data improve the initial field only in the surroundings but with smaller magnitudes. On the other hand, the bogus vortex assimilation method adjusts the weak initial typhoon vortex significantly. It clearly strengthens the structure of the vortex reaching a more realistic and stronger typhoon. Finally, in the assimilation with both GPS refractivity data and a bogus vortex, the BDA provides much larger modifications on the typhoon and thus tends to dominate the impact of the observations on the typhoon prediction. Due to use of BDA improving both intensity and structure of the typhoon vortex, the track prediction has been improved as well, reducing the 72-h forecast error significantly. The predicted rainfall distribution and intensity for BDA are also greatly improved as compared to no-BDA run. In this study, we also revise the BDA method by including three dimensional gradient wind, which was found to well maintain the typhoon intensity in the whole layer. The results show that this revised BDA gives the best track prediction (with 60-km error on 72-h forecast) as compared to the no-BDA run and other BDA runs (assimilating the surface pressure and wind only).
