衛星遙測資料三維變分同化對於颱風模擬的影響

Impact of Three-Dimensional Variational Assimilation of Satellite Remote Data on Typhoon Simulations

黃清勇(Ching-Yuang Huang) ； 王潔如(Chieh-Ju Wang)

《大氣科學》 36卷4期 (2008/12) Pp. 249-273

摘要

本文使用NCAR MM5三維變分同化3DVAR(Three-Dimensional Variational Data Assimilation)方法，將多種衛星觀測（GPS掩星、AMSU、QuikSCAT）及傳統探空觀測資料同化於中尺度模式中，以了解同化這些觀測資料對於改善數值天氣預報準確度的影響。選取兩個颱風個案，包括2003年8月中度颱風杜鵑(Dujuan)及2004年6月中度颱風敏督利(Mindulle)，每個颱風個案皆分成兩組模擬，一爲同化傳統探空觀測資料，另一組同時加入GPS掩星觀測及傳統探空觀測資料。
兩個颱風個案模擬皆使用3層巢狀網域，最高水平解析度爲5公里。在最外圍的網域，CHAMP衛星於模擬初始時間窗區內約有4~6個探空點，可提供溫濕度的資訊剖面。就Dujuan颱風個案，同化GPS掩星折射率資料及傳統探空觀測資料的模式初始場，所模擬出來的颱風路徑大致上與實際路徑相符，對於颱風移速的模擬也是相當不錯，但是模擬的中心氣壓強度則偏弱。Dujuan颱風掃過台灣南端，觀測的最大降雨量出現在東部地區，這與模擬的累積降雨量分布一致。就Mindulle颱風，同化GPS折射率資料的個案模擬，比只使用傳統探空觀測資料的模擬在登陸前48小時有將近40公里的路徑修正，颱風移動方向也更接近實際。由觀測累積降雨量來看，所模擬之颱風在登陸前後threat scores及root mean square errors都有不錯的表現。同時加入QuikSCAT資料（二維海平面風場）同化，模擬的環流更接近實際，而使降雨預報得到改善。另外，同時加入AMSU反演的三維風場資料同化，使颱風結構變得更紮實，因而在路徑模擬上有明顯的進步。

關鍵字

三維變分同化 ； GPS掩星折射率 ； AMSU

並列摘要

This study uses NCAR MM5 3DVAR (Three-Dimensional Variational Data Assimilation) to ingest various satellite observations (GPS radio occultation, AMSU, QuikSCAT) and conventional soundings into a mesoscale model for understanding the impact of these data on improvement of numerical weather prediction. Two typhoons were selected for simulation, including the August 2003 moderate Typhoon Dujuan and the June 2004 moderate Typhoon Mindulle. Numerical simulations consist of two comparative group experiments with assimilated data, one with the conventional soundings only, and the other with both GPS RO soundings and the conventional soundings.
Both cases apply three nested domains with the highest horizontal resolution of 5 km. In the outermost domain there are 4-6 soundings from the CHAMP satellite to provide temperature and moisture profile information. For Dujuan, in the run assimilating both GPS RO refractivity and conventional soundings the simulated typhoon track and moving speed are more consistent with the observed track and movement than those from the assimilation of conventional soundings only, in spite that the simulated central pressure intensity is still considerably lower than the observation. The simulated accumulated rainfall is also more consistent with the observation with a maximum over east Taiwan since Dujuan's circulation passed by the southern end of Taiwan. For Mindulle, the simulation with additional GPS refractivity gives an improvement of 40 km on track (together with a more coincident movement) 40 h prior to landfall when compared to the simulation with conventional soundings only. The threat scores and root mean square errors of simulated accumulated rainfalls are promisingly good compared with the observations, but additional QuikSCAT data (oceanic 2D near-surface wind speed) assimilation further improves the typhoon circulation and hence results in more improvement on rainfall prediction. On the other hand, additional assimilation of the 3D wind retrieved from AMSU makes the typhoon structure even more consolidated and thus leads to a significant improvement on track prediction.

並列關鍵字

3DVAR ； GPS RO refractivity ； AMSU