MPAS為新一代全球模式，但是沒有接近真實場的模式初始場也無法讓數值天氣預報系統有良好的表現。為了提高更可靠的模式結果，本篇研究利用CWB_GSI同化系統將GPSRO之Bending angle及 Reflectivity資料同化至GFS模式的再分析場，以提供MPAS更良好的模式初始場。GSI資料同化系統解析度為(T320L40)，同化窗區為四天半，並將同化後的結果作為MPAS的初始場。而MPAS的模式解析度為60KM-15KM，模式高解析度(15KM)的範圍以台灣附近海域為中心，大致覆蓋西北太平洋及東南亞地區，並且進行5天的模擬。 本次實驗設計是將三種同化資料型態產出不同的分析場。第一種為GTS資料、第二種為Bending angle 加上 GTS 資料及第三種為Reflectivity加上 GTS資料。最後將三種不同的分析場做為MPAS的初始場進行比較，並分析初始時間及後期氣旋發展之差異性。本篇研究利用2個路徑迥異的颱風路徑做為研究，第一種為路徑穩定的颱風，個案為2015年8月的蘇迪勒颱風，同化起始點2015/07/30_00UTC，經過4天半的同化後於2015/08/03_12UTC產出GSI的模式分析場。第二種為路徑會明顯北轉的颱風，個案為2012年9月的杰拉華颱風，同化時間起始點2012/09/20_00UTC，經過4天半的同化後於2012/09/24_12UTC產出GSI的模式分析場。兩個個案除了會比較三組實驗的初始場差異外，也會進行五天模擬結果的分析。 由結果顯示，蘇迪勒颱風個案中，不論是BND或是REF，在路徑的模擬結果都較為為接近最佳化路徑資料；強度模擬方面，不論GTS、BND或是REF都試過度模擬的表現。而杰拉華方面同化GPS資料的路徑結果表現較單純同化GTS差。 ;Model for Prediction Across Scales-Atmosphere (MPAS-A) is the new generation global model developed by NCAR. However, the model prediction may not be improved without a good initial field. In order to improve the initial field, this research applies a hybrid system that combines two assimilation systems in which the background is from CWB_GFS, Grid-point statistical Interpolation (GSI) . The hybrid system assimilates the Global Positioning System Radio Occultation (GPSRO) data, including both bending angle and refractivity, in GFS to make a new analysis. The horizontal resolution in GSI-3DVar is T320 (approximately 37.5 km). The vertical resolution of both assimilation systems has 40 levels. Time window of assimilation in this research is 4.5 days, and the new analysis data are used as the initial field for MPAS-A. The horizontal resolution of MPAS-A is a variable resolution of 60-15 km. The higher resolution 15 km is centered is near Taiwan, the higher resolution domain covers East Asia and North Western Pacific. In this research, we design three kinds of experiments. The first experiment assimilates GTS data, the second experiment assimilates bending angle data (BND) and the third experiment assimilates refractivity data (REF). The first case is Typhoon Soudelor (2015). Time of assimilation is from 0000 UTC 30 July to 1200 UTC 03 August, and the simulation duration is five days from the end of data assimilation with three different analyses. The second case is Typhoon Jelawat with the simulation of three different analyses for five days. At last, we analyze the difference be the initial fields and simulations for these two cases, and discuss the data impacts on the simulations. The tracks with assimilation of BND or REF in Soudelor are better than that with GTS. However, all the experiments over-predict the typhoon intensity. In Jelawat, however, the tracks with both BND and REF are worse than that with GTS.
