摘要 本文討論衛星資料改進數值預報的原因與方法，並將衛星資料應用於颱風個案的分析與預報，所採用的方法主要是變分反演與同化公式。衛星資料可改善數值預報，在南半球傳統資料缺乏的地區是肯定的，但對北半球傳統觀測密集地區有正面的頁獻，是因變分公式的發展。因此本文建立變分反演及資料同化程式，探討其在應用衛星資料可改進的空間，主要是以模擬實驗及真實的衛星資料反演來測試，在此主要有兩個重點，一是衛星資料的處理對衛星資料使用之重要性探討，二是當衛星資料與變分法結合時，衛星資料偏差的修正與變分公式中觀測誤差矩陣之求取。應用衛星資於颱風初始化的過程，我們採用衛星微波觀測反演的溫度及颱風環流，結合MM5三維變分的系統來完成，後續的模擬由MM5中尺度預報模式完成。實驗所使用的衛星資料，是以NOAA系列繞極軌道衛星，所酬載的觀測儀器為主。包括紅外線探空儀器HIRS與AMSU微波探空儀，由於AMSU是微波探測儀，對雲具有較高的穿透率，對劇烈天氣現象具較佳的掌握，因此研究以AMSU為主。 衛星資料對數值預報的影響，端看多少與天氣預報有關的資訊可由衛星資料中取出。透過建立模擬反演與同化的實驗，變分反演同化公式中背景場誤差與觀測誤差的設定，對反演/同化的結果是十分重要的，而其它反演參數(例如雲參數)的正確與否，對反演主要參數(溫度、水汽)的影響也是可觀的，由於變分反演分析公式的發展可將其它參數(例如雲參數，地放射率)與主要反演參數(溫度，水汽)一併求取，本文中以模擬實驗的方法，證明精確雲參數，對以HIRS反演溫濕剖面的助益，並於真實HIRS資料反演時，輔以AVHRR求取精確的雲參數進一步證實，並且討論真實衛星觀測中卷雲對反演結果的干擾程度進行評估。我們以此實驗為例，說明結合衛星資料於變分分析時，衛星資料前置處理的重要性。由於模擬實驗是假設背景場與觀測誤差都是無偏差的，然而在真實資料的情形下，二者都是有偏差(系統性誤差)，此外精確的估計觀測誤差對變分反演同化衛星資料而言是十分關鍵的因素，因此在本文中建立一套濾除觀測偏差(系統性誤差)及估計隨機誤差的方法，在許多計算觀測偏差與隨機誤差的方法中，常使用真實的衛星觀測值與模擬的衛星觀測值的差作為估計誤差的樣本，而模擬的衛星觀測是由短期預報為輸入值，經由輻射模式計算而得的。因此，如此得到的誤差樣本所計算出的觀測偏差與隨機誤差，二者都包含了短期預報的誤差在內而產生不正確的值，在本文的方法中我們發展出一個實用的方法降低短期預報誤差所產生的干擾，以改進AMSU觀測偏差修正及隨機誤差的估計，估算短期預報誤差所產生干擾的方法，我們以National Meteorological Center (NMC)方法計算短期預報誤差的溫溼剖面資料，經輻射模式轉到AMSU頻道所在的輻射量度空間，再計算出短期預報誤差在AMSU頻道所表現出的大小，然後將此效應自偏差修正與隨機誤差估計中濾除，而產生較正確的估計值，另外衛星觀測偏差與隨機誤差會隨掃瞄角度變化，因此所有的計算過程是分掃瞄角度不同的情形下完成，以考慮掃瞄角度的因素在內。在這次分析中，經過濾除短期預報(12小時)誤差後的AMSU頻道的隨機觀測誤差比背景場誤差於AMSU頻道量度值小很多，這表示AMSU頻道有機會提供正面的資訊修正背景場。在真實資料的反演實驗中，發現本文所建立的方法，確實有效的改進反演溫度剖面，較原先的方法進步甚多(反演成功收斂率達99%以上，在780bPa以上的地區改正背景場達到0.2至0.4K)。這個方法目前只用於一維變分反演中，但要推至三維或四維的公式是沒有困難的，而且建立這套方法後，可以偵測衛星資料的品質，在真實的作業環境是必須的。 將AMSU衛星資料應用於颱風的分析與預報的實驗，首先修改Zhu et. al.(2002) 所提出的方法，這個方法由反演得到的溫度場，及設定颱風地面氣壓後，在靜力平衡假設下，由下向上積分，求得各氣壓層的高度場，再經由梯度風平衡方程，求取颱風梯度風風場，另外以Omega方程與連續方程求出輻散風場，但是因為颱風的結構並非完全軸對稱，而在Zhu et al. (2002) 的方法中假設地表颱風氣壓以中心向外遞增的情形，這是不是十分合理的，因此我們採用假設颱風上層50hPa的高度場與環境的度場相異不大Kidder et. al. (2000)，因此可以環境場的50hPa高度場，為颱風上層的高度場，如此以靜力平衡假設下，求取各層的高度，可由上往下積分，而避免當颱風地面氣壓可能非軸對稱的情形下之誤差。一旦各氣壓層高度決定後，颱風在各層的旋轉風分量可經由平衡方程求得。由分析不同的颱風個案，可以發現颱風的非軸對稱特性是常常發生的，並且颱風中心隨高度傾側的現象也可由我們的分析方法所得到。對於颱風初始化的工作而言，由於洋面上缺乏適當的觀測，因此常需要加入人造的資料，這種方法常是根據經驗加入人造的渦漩，然而由於渦漩與其它的氣象場未能保持平衡，而使效果不佳，因此早期即有學者提出使用變分分析，在植入人造渦漩的同時加入觀測的溫度於分析之中。近年提出人造資料的變分同化，使高解析的預報模式產生更合理的颱風結構，在本文中我們使用AMSU反演的溫度與渦漩資料結合變分同化系統，由於我們的方法改進原先於反演渦漩時的人為地面氣壓假設，可以避開許多經驗值，而以觀測的資料結合變分同化系統，產生更接近於觀測而又量場平衡的颱風初始場。因此由本法所得到的三度空間溫度場及風場的資料，可提供數值模式預報颱風的初始場資訊，接下來的實驗以MM5三維資料同化系統，同化AMSU反演得到的溫度場與風場，由比對加入AMSU資料的初始場與反演所得到的風場，及原始的初始場，可以發現MM5三維資料同化系統確可有效的同化反演的參數，在後續的預報實驗中，也可發現AMSU反演所得的資料可以改進預報的路徑。 Abstract In this article we discuss the reason and method of using satellite data to improve numerical weather forecast. And the satellite data have been applied to the case study of typhoon forecast. In the pass decade the development of variational data assimilation method the positive impact of using satellite data on northern hemisphere forecasts have been established. To study the potential improvement of using satellite data in weather forecast a variational data assimilation system and related technique have developed. Using this system we achieve some experiments about using satellite data in the variational data assimilation system. Firstly, the important of satellite data pre-processing when using variational data assimilation system to assimilate satellite observation have been discussed. Secondly, the bias correction and error estimation of satellite data have been achieved. Finally experiments are case studies about typhoon forecast with satellite observation. This experiment uses MM5 model and its three-dimensional variational (3DVAV) data assimilation system. And the satellite observations are using the Advanced Microwave Sounding Unit (AMSU) data form NOAA satellite. The impact of satellite data on numerical weather forecast depend on how much weather information (ex: information about temperature profile/water vapor profile) could be extracted from satellite observation. The variational data retrieval/assimilation systems include not only weather parameters (ex: temperature profiles) but also others parameters (ex: cloud parameters or surface emissivity) in retrieval/assimilation process. In this situation the accurate estimation of others parameters is important for obtains accurate temperature/water vapor information. In our simulation study confirms this point. Using AVHRR data obtained accurate cloud parameters in HIRS data can provide same conclusion in real data retrieval experiments. The cirrus cloud effect has been discussed, as its presence generally degrades the quality of retrieval. Another key point for variational data retrieval/assimilation scheme is how to estimated observation errors. In practice, observation errors, both systematic and random, are often estimated from the difference between satellite observation and simulated satellite observation obtained from a radiative transfer operator with a 12-hr forecast as its input. Observation errors estimated by this approach may be contaminated by errors in the 12-hr forecast. This work describes a practical way to eliminate the 12-hr forecast error and improve the estimate of the observation error in the AMSU data. Following the philosophy of the National Meteorological Center (NMC) method (that derives the statistics of forecast error from the differences between pairs of forecasts at disparate ranges valid at the same time), in this study the pairs of forecasts at different ranges in the NMC method are first converted to brightness temperatures in the AMSU channels by a radiative transfer operator. The 12-hr forecast errors are then determined from the representations of these forecasts in radiance space spanned by the AMSU channels. Since most AMSU channels have beam position-dependent systematic observation errors, our procedure further takes into account this dependence by performing the statistics separately for sub-groups of data in each AMSU channel with different beam positions. In a case study, after eliminating the 12-hr forecast error obtained by this procedure from the total estimated observation error, the remaining random error of the satellite observation is shown to be smaller than the background error(provided by 12-hr forecasts of a numerical weather prediction model) in most of the AMSU temperature sounding channels. Using the error-corrected AMSU data in these channels, a retrieval experiment using a one-dimensional variational scheme shows an improvement of 0.2-0.4 K over the background error in the retrieved temperature profiles above 780 hPa. The observation errors estimated procedure has been test in 1DAVR retrieval experiment. But can be applied to the 3/4DVAR system. And it is useful for quality check for satellite data in an operational environment. The AMSU data is useful for retrieving the structures of typhoons due to the high resolution and the cloud-penetrating ability of the AMSU measurements. In this study, a procedure modified form Zhu et al. (2002) is established to first retrieve the three-dimensional temperature field from the AMSU data. A vertical integration of the retrieved temperature leads to the height field, which is then used to obtain the rotational wind component by solving a nonlinear balance equation. The vertical integration is performed downward from a uniform 50 hPa height at the top of a typhoon as estimated from the environmental mean value. This gives no restriction on the shape of the typhoon and allows the structure of the retrieved typhoon vortex to depart from axial symmetry. With this technique, the basic structures of several typhoons in the Eastern Pacific are successfully retrieved from the AMSU data. The retrieved temperature and velocity associated with the typhoon are further incorporated into the initial condition of a meso-scale numerical weather prediction model by using a three-dimensional variational data assimilation system. Forecast experiments show that the inclusion of the AMSU-based typhoon structures in the initial state has an overall positive impact on the forecast of the track of the typhoon.
