集合卡尔曼滤波（EnKF）岸基雷达资料同化对登陆台风数值模拟的影响研究

发布时间：2018-08-30 07:41

【摘要】：近几十年来,台风路径的预报误差逐年降低,而强度和降水方面的进步却很缓慢,这部分是由于台风数值模式对其强度和结构预报能力相对不足导致。由于台风系统非线性很强,初始场的微小偏差容易在积分过程中加倍放大,使得预报结果严重偏离实际天气状况。资料同化过程将模式背景场和观测资料基于一定的数学理论进行融合,获得理论上误差较小的分析场,能够为数值模式提供较为准确的预报初始条件,在一定程度上提高模式预报能力。岸基雷达能够探测登陆台风内核精细结构,合理地利用雷达资料对提升我国登陆台风数值预报很有意义。集合卡尔曼滤波利用一组集合成员构建具有“流依赖”特性的背景误差协方差,目前在国内雷达资料同化的台风研究中仍不多见。利用集合卡尔曼滤波同化我国岸基雷达资料来研究登陆台风的数值预报问题,是一个很有意义的工作。本文利用美国宾州州立大学开发的PSU WRF-EnKF同化系统结合我国岸基雷达径向风资料对近年来三个登陆台风“彩虹(1522)”、“莫兰蒂(1010)”、“威马逊(1409)”进行了数值试验。发现雷达资料集合同化技术能够显著地改善台风路径、强度、结构和降水模拟。通过循环资料同化,分析场台风位置逐渐向实测位置靠近,最终使得台风登陆点的模拟误差小于10km。三个台风平均来看,经过8h循环同化雷达资料后,台风路径误差相比同化前开始呈现正效果,同化窗内平均误差从未同化资料的60km减小到同化后的20km左右,多时次循环同化后路径误差可低于10km。台风强度从循环同化初期就有比较明显的改进效果,平均台风强度误差可以减小到10hPa以下,雷达资料同化能够比较显著地使得台风加强。在台风“彩虹”同化分析中还发现,随着同化资料次数增多,高层暖心结构明显加强、最大风速半径收缩、风眼收缩、对流非对称结构与实测接近。同化增量显示,随着循环次数增加,资料对模式背景场的修正逐渐集中于台风内核。资料循环同化一定程度上改善了台风降水的预报,并且同化资料时次越多,降水预报TS评分提高越多。进一步将雷达资料按距台风中心距离分成小于100km、100-200km和大于200km三部分分别进行同化,从“彩虹”试验结果来看,仅同化台风100km半径范围内资料在台风路径、强度、结构等方面均可以得到与同化所有资料基本相同的同化效果,而仅同化100-200km和200km以外范围内资料效果却不显著。同化内核100km范围内资料和同化全体资料在三个台风试验中均得到十分接近的效果,多时次循环同化后路径和强度误差分别可低于5km和5hPa。表明内核资料改善背景场的关键,这部分资料通常仅占总量的50%以下(依照台风个例而变化),但是却制约着同化效果。仅同化这部分资料可以得到与同化所有资料相同的效果,却可以将原本巨大的集合同化计算量减小超过一半,缩短计算机时。
[Abstract]:In recent decades, the forecast error of typhoon track is decreasing year by year, but the progress of intensity and precipitation is very slow, which is partly due to the relatively insufficient ability of typhoon numerical model to predict its intensity and structure. Due to the strong nonlinearity of the typhoon system, the small deviation of the initial field is easy to double magnify in the integral process, which makes the forecast result deviate from the actual weather condition seriously. In the process of data assimilation, the model background field and observation data are fused based on certain mathematical theory, and the analytical field with small theoretical error is obtained, which can provide more accurate prediction initial conditions for the numerical model. To a certain extent, the ability of model prediction is improved. Shore based radar can detect the fine structure of landing typhoon core. It is very important to make use of radar data to improve the numerical forecast of landing typhoon in China. Ensemble Kalman filter uses a set of members to construct background error covariance with "current dependence" characteristics, which is still rare in the study of typhoon data assimilation in China. It is significant to study the numerical prediction of landfall typhoon by using the ensemble Kalman filter to assimilate the land-based radar data in China. Three landfall typhoons Rainbow (1522), Moranti (1010) and Wieson (1409) have been numerically tested in this paper by using the PSU WRF-EnKF assimilation system developed by Penn State University and the radial wind data of land-based radar in China in recent years. It is found that the radar data set contract technique can significantly improve typhoon track, intensity, structure and precipitation simulation. Through the assimilation of the cyclic data, the typhoon position of the field is gradually approaching to the measured position, and the simulation error of the landing point of the typhoon is less than 10km. In the mean of three typhoons, after 8 h cyclic assimilation of radar data, the track error of the typhoon began to show positive effect compared with that before assimilation, and the 60km of the data never assimilated in the assimilation window was reduced to about the 20km after assimilation. The path error can be less than 10km after multi-time cycle assimilation. The typhoon intensity can be improved obviously from the initial stage of cycle assimilation, the mean typhoon intensity error can be reduced to below 10hPa, and the radar data assimilation can significantly strengthen the typhoon. In the analysis of typhoon "rainbow" assimilation, it is also found that with the increase of assimilation data, the structure of the upper layer warm center is obviously strengthened, the maximum wind velocity radius shrinks, the wind hole shrinks, and the convection asymmetric structure is close to the measured data. The assimilation increment shows that the correction of the model background field is gradually concentrated in the typhoon core with the increase of the number of cycles. The data assimilation improves the forecast of typhoon precipitation to some extent, and the more times of assimilation, the more the TS score of precipitation forecast is improved. The radar data are further assimilated into three parts according to the distance from the center of the typhoon, which are less than 100kmm2 100-200km and larger than 200km, respectively. According to the results of Rainbow test, only the data in the radius range of Typhoon 100km are assimilated in the track and intensity of the typhoon. The assimilation effect can be basically the same as that of all the data assimilated in structure, but the assimilation effect is not significant except in the range of 100-200km and 200km. The data and assimilation data in the 100km range of the assimilation core are very close to each other in the three typhoon experiments, and the error of track and intensity after multi-time cyclic assimilation is lower than that of 5km and 5HPA, respectively. It shows that the kernel data is the key to improve the background field, which usually accounts for less than 50% of the total data (varying according to the typhoon case), but restricts the assimilation effect. Only assimilation of this part of data can achieve the same effect as assimilation of all data, but it can reduce by more than half the calculation amount of the original huge set assimilation, and shorten the computer time.
【学位授予单位】：中国气象科学研究院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P456.7

【相似文献】