| 摘要: | 台灣位處於西北太平洋，若可以有較好的熱帶氣旋生成以及路徑預報，對災情預警及控制會有明顯的效益。由於熱帶氣旋生於海上，因此利用洋面上的觀測資料對於氣旋預報來說是非常重要的。而福衛七號GNSS掩星觀測系統提供相當充分的海上氣象資訊，可以大幅改善模式初始場的準確度，有助於熱帶氣旋生成及路徑預報。本篇研究使用WRF Hybrid資料同化系統，對2019以及2020年在西太平洋的七個颱風（利奇馬、米塔、哈吉貝、哈隆、哈格比、梅莎以及海神）進行模擬分析，從觀測生成時間的72小時前開始，分別進行同化GNSS RO加上GTS傳統觀測資料（為WR實驗）以及只同化GTS資料（為NR實驗）共120小時預報，探討同化GNSS RO折射率資料對氣旋生成預報的影響，同時挑選同化效益明顯的個案進行分析，進一步了解其中造成預報差異的動力因素。
模擬結果顯示在WR實驗，雖然在氣旋生成時間的平均誤差並無顯著改善，但氣旋生成位置誤差減少8.9公里，且被成功檢測出來的生成個數也較NR實驗多一些。氣旋生成時間誤差 介於24至48小時之間的改善較為明顯，且生成位置誤差亦較NR實驗小。在敏感度測試實驗上，顯示使用預報中心的作業實驗組有較好的表現，尤其是在使用分析場blending之後，使氣旋生成時間誤差改善3.6小時，而生成位置誤差改善93.9公里。針對利奇馬、哈吉貝以及梅莎颱風，使用最新估計的背景誤差時，整體來說模擬結果會變差，但若停用掩星資料稀化則模擬結果會變好。進一步分析同化效益明顯的個案(米塔與哈格比)，在12小時同化後的模式水氣增加，特別在中低層水氣呈現較大的增量，有助提升潛熱釋放效率。雖然掩星資料同化只直接影響到水氣以及溫度場，經由動力過程，也會改變風速以及渦度場，改善兩個案氣旋生成的掌握能力。由位渦收支分析證實，非絕熱加熱增加，增加正位渦趨勢，使垂直運動加強，同時也增加位渦垂直平流，促進氣旋的生成。;Taiwan is located at north-western Pacific and is affected by typhoons often in the summer season. Accurate predictions of tropical cyclogenesis and track will be helpful for the disaster preparedness. Since tropical cyclones form over oceans, it is important to make use of the observations over oceans. FORMOSAT-7 radio occultation provides high-resolution observations over lands and oceans, which can improve the accuracy of initial conditions, and is helpful for predicting tropical cyclogenesis and track forecast. In this thesis, WRF hybrid data assimilation system has been used to assimilate seven typhoon cases in 2019 and 2020 (Lekima, Mitag, Hagibis, Hagupit, Halong, Hagupit, Maysak and Haishen). The experiments start 72 hours before observed cyclogenesis time, with the experiment that assimilates both GTS and RO data (WR) and the other that only assimilates GTS data (NR), and both experiments are conducted for 120-h forecast. The experiment is expected to find out if the assimilation of GNSS RO reflectivity data will give positive impacts on cyclogenesis prediction, and also analyze the specific cases with larger positive impact, to understand the dynamic processes leading to the differences in forecasted cyclogenesis.
The model results show that, although the average time error is not improved significantly in the WR experiment, the average location error is reduced by 8.9 km, and the number of successful cyclogenesis detection is more than that from the NR experiment. Most of the improvement by RO assimilation is present for the time error between 24 to 48 hours that is smaller and grows slower. For sensitivity tests, the experiments with the operational ensemble prediction from CWB for the initial background has significantly larger improvement than that from the control experiment (CTL), in particular when performing the blending with the large-scale analysis, leading to a reduction on the time error by 3.6 h and location error by 93.9 km. Other sensitivity tests for Lekima, Hagibis and Maysak show that use of the updated background errors has increased the prediction error. On the other hand, the prediction is improved in time error than the CTL experiment when the default of data thinning is deactivated.
For the two specific cases (Mitag and Hagupit), water vapor mixing ratio increases after 12-h assimilation, especially in lower and middle levels helpful for latent heat release. Although RO data assimilation only changes the water vapor mixing ratio and temperature, it will also modify the wind and vorticity after the model integration. This makes the WR experiment improve the cyclogenesis prediction one day earlier than the NR experiment. Through analysis of potential vorticity budget, the increase of diabatic heating indeed enhances the positive PV tendency, and vertical motions, and thus facilitates the tropical cyclogenesis with the stronger PV vertical advection. |
