|觀測反演（假設球對稱）的大氣折射率與真正的局地折射率，兩者間存在代表性的誤差。Sokolovskiy et al. (2005a,b)提出一向前模式算子，將模式局地與觀測折射率沿著掩星射線方向積分(此積分量稱為溢相值excess phase)，考慮了大氣水平梯度的效應，使代表性誤差得以減少，此同化算子稱為非局地算子(nonlocal operator)。本研究目的之一為發展非局地同化算子，並將之導入於WRF 3DVAR，同時利用2007年冬季和夏季各一個月的福爾摩沙衛星三號(FORMOSAT-3/COSMIC)掩星折射率資料，針對溢相值及折射率，進行觀測誤差統計，發現折射率之觀測誤差大約為溢相值的2 倍，而觀測誤差隨緯度變化，且以冬季低層最為顯著，在夏季則不明顯。 本研究針對2006 年的珊珊(Shanshan)颱風，探討此非局地算子同化能力的表現。非局地算子與局地算子，前者所產生的增量值稍大，且沿射線方向發展，呈橢圓形分布。當僅同化27 筆福衛三號折射率探空，WRF 模式的路徑及降雨模擬結果顯示，在初期二者較無同化實驗者略為改善，後期之降雨預報則顯示，使用非局地算子較使用局地算子略佳。以GPS 掩星觀測驗證預報的環境場，發現當同化傳統觀測資料時，若同時同化GPS 掩星觀測，對於初期預報有顯著改善（尤以5公里以下），而在後期亦有些改善，但使用局地或非局地運算子的效益則無明顯差異。若同時同化颱風虛擬渦旋，使颱風初始強度明顯增強，因而有較佳的路徑及降雨預報，其同化的影響使得其他觀測資料的同化效果相對不明顯。週期性同化實驗(Cycling Run)則顯示同化更多的觀測資料，使路徑及強度預報皆有較佳的表現。整體而言，GPS的同化效益使得模擬後期的大雨預報技術得分增加。 本研究另使用觀測系統模擬實驗(OSSE)，利用高解析度MM5模式模擬珊珊颱風，提供真實大氣條件，以確實了解資料同化對於颱風預報的影響。藉由二維射線追蹤方法，以及假設球對稱情形下，反演出非局地折射率的垂直探空。將此高解析度資料分為三區域，AREA1著重於颱風東北方之太平洋高壓區域，AREA2 主要以颱風東側之環境流場為主，AREA3 則是針對颱風暴風範圍之影響，同化的探空資料數分別為30、96及16筆。與真實大氣比較，使用不同的同化運算子時，初始的分析誤差在AREA1 及AREA3 有明顯的差異，而在AREA2 則相似，可能與後者的水平梯度較弱有關。若同化AREA1 中的30筆觀測資料，顯示所有同化算子之路徑預報，較無同化實驗之結果略為改善。當同化位於颱風範圍內(AREA3) 的16 點觀測資料，颱風路徑預報顯示，非局地算子較局地算子進一步改善模式初始場，進而得到較佳的路徑預報。然而，同化AREA2 的觀測資料，其路徑預報卻較無同化實驗差，原因可能來自於模式反演出的折射率探空與真實大氣有些微差異，而颱風路徑預報對於上游環境場之初始微幅差異較敏感所致。 GPS radio occultation (RO) refractivity was retrieved with the assumption of spherical symmetry, possessing representative errors for the true state. Sokolovskiy et al. thus suggested a nonlocal observable (excess phase, defined as the integrated refractivity along a fixed raypath) to account for the effect of horizontal gradients existing in the true state, and this forward model was called the nonlocal operator. One of the main goals of this study is to develop the nonlocal operator for implementation into WRF 3DVAR. In addition, we estimated the observational errors for both excess phase and refractivity data during two months in 2007, one in summer and the other in winter. The observational errors of excess phase are roughly one half of the refractivity errors, regardless of the winter or summer month. Both observational errors are dependent on latitude, but such dependence is stronger at the lower levels in the winter month and is less pronounced in the summer month. This study investigates the performance of the nonlocal operator in typhoon forecast for Shanshan (2006) using WRF. The model initial increments from the nonlocal operator are slightly larger and are somewhat extended along the raypath in elliptical contours as compared to those from the local operator. Typhoon track and recipitation forecasts show slight improvement for the first day when 27 refractivity soundings are assimilated into WRF. For local rainfall forecast at later time, the performance of the nonlocal operator is slightly better than the local operator. Verified against GPS RO observations for the environmental prediction, assimilation of GPS RO data shows improvement of early prediction (especially for the levels below 5 km) as the conventional GTS data are also assimilated; however, no significant improvement was found at later prediction for both nonlocal operator and local operator. As a bogus vortex is also combined into assimilation, initial typhoon intensity is enhanced and thus results in better track and rainfall prediction. The impact of bogus vortex overwhelms the impacts of all other data in this study. The cycling run shows that typhoon intensities and tracks can be further improved when more observational data can be assimilated at different times. In general, the impact of GPS RO data assimilation increases the skill score of prediction for local intense daily rainfall. Observing Systems Simulation Experiments (OSSE) were designed to explore the impact of GPS RO data assimilation on weather prediction using high-resolution MM5 model for the modeled atmosphere providing the true state. Vertical soundings of nonlocal refractivity as modeled by a 2-D raytracing operator under the assumption of spherical symmetry then were assimilated into OSSE. The nonlocal refractivity soundings were produced for three assimilated areas, AREA1 to the northeast of the typhoon over the edge of Pacific high, AREA2 to the east of the typhoon over the environmental flow, and AREA3 over the region of typhoon circulation, with 30, 96 and 16 soundings, respectively. Verified against the true state, the initial analysis errors in AREA1 and AREA3 are quite different but are similar in AREA2 for different operators, possibly due to weaker horizontal gradients existing in the latter. Slight improvement in track prediction is found for all the operators when the 30 soundings in AREA1 were ingested. The initial condition and the ensuing tracks were more improved for the nonlocal operator than local operator as the 16 soundings in AREA3 were assimilated. The track prediction with assimilation in AREA 2 becomes worse compared to that without assimilation. It appears that typhoon track is sensitive to initial tiny discrepancies in this upstream steering environment, which can be produced by the refractivity soundings as simulated by the model.