颱風策略性觀測理論之特徵分析與驗證—駛流敏感共軛向量及系集變換卡爾曼濾波器

Characteristics and Validation of the Targeted Observations for Tropical Cyclones — ADSSV and ETKF

陳新淦(Shin-Gan Chen)

指導教授 ： 吳俊傑

國立臺灣大學/理學院/大氣科學研究所/博士(2011年)

https://doi.org/10.6342/NTU.2011.02997

摘要

颱風策略性（標靶）觀測為尋找影響颱風預報的敏感區域，以便制定最佳（飛行）觀測位置，以改進颱風預報，亦應用於闡釋颱風移動與綜觀環境間的動力分析與連結。本論文的研究為探討「颱風駛流敏感共軛向量（adjoint-derived sensitivity steering vector, ADSSV）」與「系集變換卡爾曼濾波器（ensemble transform Kalman filter, ETKF）」兩種策略性觀測方法的敏感性特徵與物理意義。
研究第一部分以2006年珊珊颱風北轉運動為個案，主要ADSSV敏感區域位於颱風東邊的副熱帶高壓邊緣以及上游的中緯度槽線，位渦診斷分析亦支持ADSSV表現的敏感特徵。為了進一步對ADSSV梯度敏感性進行探討與驗證，針對初始時間2006年9月15日0000 UTC的ADSSV敏感與不敏感區域，以初始渦度加以擾動方式設計一些實驗，評估初始擾動對模式的影響，包括路徑模擬、深層平均駛流變化與擾動垂直結構發展。路徑模擬結果顯示位於兩個敏感區域之擾動造成的18–48小時平均路徑偏移（相較於無擾動的控制實驗）比低敏感區域的變化來得大。經由Lanczos濾波方法得到大尺度平均環境場後，整體而言，敏感區域的擾動實驗相較於控制實驗之850–250 hPa深層平均風差異的訊號隨著綜觀系統演變逐漸傳送至驗證區域。此外，初始擾動經非線性積分造成在驗證區域的深層平均風差異與ADSSV線性計算的深層平均駛流變化呈現良好的一致性，而此一致性隨積分時間增加而降低。低敏感區域之擾動造成的深層平均風變化不論方向或量值皆與ADSSV的計算較不一致，原因可能與初始擾動與北邊槽線產生的非線性交互作用有關。有關槽線（副熱帶高壓）針對中高（中低）對流層的擾動實驗造成動能差異的垂直發展與演變以及進一步評估位於較高ADSSV敏感性區域的影響，亦在論文中有詳細探討。
第二部分研究建立於由National Centers for Environmental Prediction（NCEP）、European Centre for Medium-Range Weather Forecasts（ECMWF）、Canadian Meteorological Centre（CMC）模式以及結合三者（COMBINED）的系集預報輸出所計算ETKF敏感性，較過去文獻特別的是此研究提出移除每個系集成員之颱風分量的一個新計算方式，並以觀測位置遠近調整觀測誤差的方法試圖降低誤差協方差矩陣中長距離的假性相關（long-distance spurious correlation）。本研究以2008年辛樂克颱風為個案，探討ETKF敏感性在熱帶氣旋環境下呈現的特徵。結果顯示濾除颱風分量後的ETKF敏感性相較於保留原有系集的計算更能有效突顯出環境流場的敏感特徵，並減少因系集颱風路徑預報分歧造成接近颱風自身的敏感性。比較ETKF敏感性與其對應的系集變異發現兩者之間具有相異的分布特徵；不同模式系集間的ETKF敏感性亦呈現獨特不同的結構，主要為ECMWF模式主宰，原因與奇異向量方法建構初始擾動的特性與發展有關。此外，利用COMBINED系集探討加入位於不同層觀測變數的敏感性、颱風生命期不同階段、隨最佳化時間演變、以及預報時間長短的敏感性之議題中ETKF敏感性表現的特性，以瞭解ETKF敏感性如何掌握影響颱風運動或環境預報誤差之大尺度動力特徵。同時檢視ETKF策略性觀測理論預測訊息變異的能力，整體來說可以表現出NCEP模式三維資料同化方法造成模式訊息的近似分布，並提供投落送觀測導致訊息變異的可能傳播及變化。
本論文研究結果提供ADSSV與ETKF策略性觀測理論更多物理詮釋，以及與大尺度綜觀環境特徵作連結，進一步瞭解此兩種理論的實用性與表現，在作業預報或科學研究基礎上，此策略性觀測的研究除提供特殊飛機觀測之有用參考外，亦為改善颱風預報問題的重要一環。

關鍵字

策略性觀測 ； 共軛技術 ； 系集變換卡爾曼濾波器 ； 資料同化 ； 敏感性分析

並列摘要

The main concept of the targeted observation is to make extra observations in most sensitive regions, which are expected to reduce uncertainties in the initial condition and thus decrease errors in numerical forecasts. Targeted observing methods not only can be used to design optimal flight routes for the typhoon surveillance, but would also be applied to study the dynamical sensitivity between tropical cyclones (TCs) and the synoptic-scale environmental features. This research consists of two parts in which two targeted observing theories, adjoint-derived sensitivity steering vector (ADSSV) and ensemble transform Kalman filter (ETKF), are examined.
In the first part, the initial vorticity fields associated with the ADSSV signals are systematically perturbed to examine the impact on model simulations and validate the ADSSV gradient sensitivity based on the case of Typhoon Shanshan (2006). Two distinct ADSSV features associated with the midlatitude trough and the subtropical high that affect the recurving motion of Shanshan are identified and also well supported by the potential vorticity diagnostics. The result from track simulations indicates that the perturbations associated with the trough and the subtropical high lead to more track deflection (18–48-h mean) than the perturbations in the region with low ADSSV sensitivity. For experiments perturbed in high ADSSV sensitivity region, the main feature of the 850–250-hPa deep-layer-mean (DLM) wind difference relative to the control run propagates along with the synoptic system into the final verification area. In addition, the change in the area-average DLM wind over the verification areas is generally consistent with the steering change derived from the ADSSV. The resemblance between the ADSSV and the DLM wind change appears decreased with the integration time. The effect of perturbations specifically introduced within the sensitive layers and the evaluation from the region with higher sensitivity are explored.
The second part is to examine the characteristics of ETKF targets under the environment of TCs, including the comparison between the ensemble variance and the corresponding ETKF guidance, the sensitivity to hypothetical targeted variables at different levels and to the varying targeted times and forecast lead-times, as well as the evolution of ETKF guidance through the TC life cycle. In particular, a new way to calculate the ETKF signal variance reduction by removing the TC components in each ensemble member is proposed to effectively highlight the sensitivity to environmental features. Moreover, a method of scaling the observation errors according to the distance between the test probes and the ensemble mean TC center is adopted to reduce the likely spurious long-distance correlation in the error covariance matrix. The properties of ETKF guidance are explored based on the ensembles from National Centers for Environmental Prediction (NCEP)、European Centre for Medium-Range Weather Forecasts (ECMWF)、Canadian Meteorological Centre (CMC) and all three combined ensembles for Typhoon Sinlaku (2008) with scientific interests. ETKF guidance with the TC removal is found to correspond to the neighboring environment instead of the storm itself. The comparison between the variance and ETKF guidance shows the marked difference from each other. ETKF guidance using the combined ensembles is dominated by ECMWF ensembles, as a consequence of singular vector for constructing initial ensemble perturbations. The ability of the ETKF to predict signal variance is evaluated by comparing with the NCEP Global Forecast System (GFS) signals. Overall, the ETKF-predicted signal variance appears to capture the pattern of NCEP GFS signal as a first-order approximation.
As recommended in “The Seventh International Workshop on Tropical Cyclones” (IWTC-VII), the advanced science behind better understanding targeted observations is highly critical to improve the TC predictability. This study provides useful physical insights into the targeted observations and important steps in addressing this issue.

並列關鍵字

targeted (adaptive) observation ； adjoint technique ； ensemble transform Kalman filter ； data assimilation ； sensitivity analysis

參考文獻

陳占慧，2005：策略性颱風觀測―共軛模式之颱風駛流敏感向量。國立臺灣大學大氣科學研究所，碩士論文，79頁。

Aberson, S. D., 2002: Two years of operational hurricane synoptic surveillance. Wea. Forecasting, 17, 1101–1110.

_____, 2003: Targeted observations to improve operational tropical cyclone track forecast guidance. Mon. Wea. Rev., 131, 1613–1628.

_____, 2008: Large forecast degradations due to synoptic surveillance during the 2004 and 2005 hurricane seasons. Mon. Wea. Rev., 136, 3138–3150.

_____, and J. L. Franklin, 1999: Impact on hurricane track and intensity forecasts of GPS dropwindsonde observations from the first-season flights of the NOAA Gulfstream-IV jet aircraft. Bull. Amer. Meteor. Soc., 80, 421–427.