本研究針對莫拉克颱風(2009)侵台帶來豐沛強降雨的時段，結合系集重新定位法(mean recentering scheme)與系集預報系統、以及WRF-LETKF雷達資料同化系統，探討系集重新定位法是否能用以改善對流尺度定量降水即時預報。實驗共分為兩部分，首先將系集重新定位法應用於單純系集預報系統，來評估其對定量降水即時預報的影響以及合適的實驗策略。接著將系集重新定位法應用於雷達資料同化系統，探討系集重新定位法能否進一步增進雷達資料同化系統的預報能力。 首先在單純系集預報系統實驗的部分，先以三小時累積雨量之空間相關係數選取最佳系集成員進行系集重新定位，另外也依據雷達徑向風、回波等不同變數選取最佳系集成員做為敏感度測試。所有實驗皆進行系集預報，並以PM (probability-matched)系集平均代替一般算數平均來表示系集的決定性定量降水預報。結果顯示在本研究的個案中，系集重新定位法可以有效地降低模式降雨預報在中央山脈南部的高估。而使用與降雨較相關的變數來選擇最佳系集成員，其降雨分布及機率定量降水預報與觀測較類似。另外使用幾個較好的成員平均當作最佳初始平均場，可以透過平均平滑掉一些個別成員的錯誤訊息，得到較穩定的改善。 而第二部分則將系集重新定位法應用於雷達資料同化系統。由標準雷達資料同化實驗的結果可知，不論有無同化回波，同化徑向風能幫助模式掌握到阿里山附近的強降雨帶，然而對於中央山脈南部高估的降雨預報則無明顯改善。而在結合系集重新定位法與雷達資料同化系統進行系集重新定位循環後，可以降低南部的總可降水量，進而改善標準雷達資料同化無法移除的過度預報。值得注意的是使用累積雨量來挑選最佳系集成員時，同化回波能讓選出的最佳初始場更具代表性，使系集重新定位法的改善更合理。另外在敏感度測試中我們發現，進行系集重新定位循環時，可使用較長的同化間距即可得到有效的改善，且其效果可維持到第六小時的定量降水預報。;In this study, the mean recentering (MRC) scheme was applied to the ensemble prediction system (EPS) and incorporated with the WRF-LETKF radar data assimilation system. The benefits of the MRC scheme to the convective-scale quantitative precipitation nowcasting (QPN) are investigated by a case study of Typhoon Morakot (2009). As a key step in the MRC scheme, the best member was selected by the spatial correlation coefficients (SCC) of the 3-hour rainfall accumulation and the performance of the precipitation nowcasting with the EPS was evaluated. In the deterministic QPN shown in probability-matched (PM) ensemble mean, the MRC scheme reduced the unrealistic excessive rainfall at the southern Central Mountain Range (CMR). Result suggests that the effect of the MRC scheme on QPN is sensitive to the choice of the best member. Selecting the best member base on the metrics related to precipitation helps improve the patterns of the QPN and the probability quantitative precipitation forecast (PQPF). In addition, taking the mean of a few good members as the best initial condition leads to more stable improvement. The MRC scheme was further incorporated with the WRF-LETKF radar data assimilation system. First, result shows that assimilating radar radial wind helped the model to capture the strong rainfall near Alishan (阿里山). However, the unrealistic excessive rainfall prediction at southern CMR still existed. After applying the MRC scheme, the amount of hydrometeor at southern Taiwan was reduced, and thus the unrealistic excessive rainfall could be alleviated. In order to make the MRC cycle effective with the metric of rainfall accumulation, it is suggested to assimilate both radar radial velocity and reflectivity. Finally, result from the sensitivity experiments suggests that the 1-hr interval is long enough for the MRC cycle to improve the 1-6 hr QPN.
