系集卡爾曼濾波器(EnKF)的特性是利用系集之間發展的差異來代表與背景場動力相依(flow-dependent)的誤差統計特性，而其優勢的前提為系集對於誤差特性能有一定的代表程度。但在區域EnKF系統中因多半使用冷起始，因而需要數個同化循環後方能得到較可靠的誤差統計特性並有效的利用觀測資訊產生好的分析場。此也造成EnKF在劇烈天氣系統的模擬當中需要一段spin-up的時間來達到其分析同化之應有表現。即便有觀測資料，這2~3天的spin-up時間卻嚴重限制了模式能及早模擬及預測颱風發展的能力 本研究在WRF模式搭配局地化系集轉換卡爾曼濾波器(LETKF)資料同化系統的架構下使用由Kalnay and Yang(2010)所提出的”Running In Place(RIP)”方法來加速EnKF的spin-up，希望在較早的時間就能正確的預報颱風的路徑。在本研究中是首次針對2008年的颱風辛樂克進行真實個案的模擬。 實驗結果顯示，在使用RIP方法之後，分析場可以較好的掌握颱風的大小，但颱風的中心定位與強度則無明顯的改善。至於預報方面，颱風的三天預報路徑有顯著的改善，特別是當颱風還處於成長階段的時候。這是因為RIP方法可以更有效的利用觀測資料並且發揮這些有限資料的價值。而在颱風強度的預報方面，RIP方法的調整僅限於使用RIP的第一天。最後，在RIP的使用需注意當EnKF的背景誤差協方差已經可以有效的發揮觀測的效果時，應該停止RIP並回復到標準的資料同化循環以避免對觀測的過度擬合。The characteristic of Ensemble Kalman Filter (EnKF) is to use a set of ensemble to represent the flow-dependent dynamical error statistic. However, the prerequisite for optimizing the performance of EnKF is that the ensemble perturbations are representative to the error characteristic. However, for the regional EnKF, the cold starting procedure is normally used to initialize the ensemble members. In these cases, a spin-up period is required to construct a reliable background error covariance. After the spin-up, the observation can be correctly used and provide effective analysis corrections and the regional EnKF can then achieve its asymptotic level of performance. For typhoon assimilation and prediction, such issue of the EnKF’s spin-up becomes a serious concern in Taiwan since this limits our ability to predict the typhoon movement during the early stage of typhoon.In this study, the ‘Running In Place (RIP)’ method proposed by Kalnay and Yang (2010) is implemented in the framework of the Local Ensemble Transform Kalman Filter (LETKF) coupled with the Weather Research and Forecasting (WRF) model. The RIP method is used to accelerate the spin-up of EnKF, in order to make reliable typhoon track prediction at earlier time. The work with the 2008 Typhoon Sinlaku is the first real case study using the RIP method.Result shows that, the RIP method is beneficial for improving the environmental condition for the typhoon so that the size of the typhoon can be better estimated in the analysis fields. However, less impact on the inner structure of the typhoon and there is no significant improvement for the center position and intensity. For forecast verification, the track prediction has been significantly improved with RIP after 24-hr forecast hours. The improvement is especially significant during the rapid intensifying period of the Typhoon. It is because that with the RIP method, the limited observations can be used more effectively. We suggest that after the background error covariance has been spun-up and can reasonably extract the observation information, RIP method should be turned off and the assimilation should switch back to the standard LETKF cycle to avoid the overfitting.
