主办单位：中国气象局沈阳大气环境研究所
国际刊号：ISSN 1673-503X
国内刊号：CN 21-1531/P
气象与环境学报

气象与环境学报 ›› 2013, Vol. 29 ›› Issue (5): 43-48.doi:

• 论　文 • 上一篇    下一篇

双分辨率集合卡尔曼滤波同化方案及模拟试验

孙龙彧1  乔小湜2.3  蒋大凯2 陈力强2 吴曼丽2 梁寒2   

  1. 1.沈阳市气象局，辽宁 沈阳 110168；2沈阳中心气象台，辽宁 沈阳110016; 3. 南京信息工程大学，江苏 南京 210044
  • 出版日期:2013-10-31 发布日期:2013-10-31

The dual-resolution approach for ensemble Kalman filter and simulation tests

SUN Long-yu1 QIAO Xiao-shi 2.3   JIANG Da-kai2   CHEN Li-qiang2  WU Man-li2  LIANG Han2   

  1. 1. Shenyang Meteorological Service, Shenyang 110168, China; 2. Shenyang Central Meteorological Observatory, Shenyang 110016, China; 3. Nanjing University of Information Science & Technology, Nanjing, 210044, China
  • Online:2013-10-31 Published:2013-10-31

PDF (PC)

287

摘要： 高分辨率中尺度模式集合卡尔曼滤波实际应用的困难是集合预报会耗费大量的时间。而双分辨率集合卡尔曼滤波是由一组低分辨率样本提供同化所需的背景误差协方差矩阵，这种方法可以减少集合预报的时间。为了检验其有效性，文中利用模拟资料，与标准高分辨率集合卡尔曼滤波方法比较。结果表明：在第一个同化时次，两者对500 hPa水平风场和扰动位温场的分析增量场均与真实增量场的高低值中心位置一致，且结构与真实增量场接近，前者（高分辨率集合卡尔曼滤波）的增量值比后者（双分辨率集合卡尔曼滤波）的增量值更接近真实情况；在连续的预报—同化循环试验中，随着同化次数的增加，两种方法分析变量的均方根误差总体上都是下降的，均表现了很好的同化能力，但后者与前者相比仍存在一定的差距；在相同的运行环境下，后者的运行时间仅是前者的1/6。

关键词: 集合卡尔曼滤波, 分辨率, 预报误差协方差矩阵, 耗时

Abstract:  A huge computational cost in the ensemble forecast is primary challenge for the ensemble Kalman filter application into high-resolution mesoscale models under an operational environment. The dual-resolution ensemble Kalman filter algorithm could significantly save the time of computation because its covariance matrix of errors is provided by a group of low resolution samples. Using the simulated data, the dual-resolution ensemble Kalman filter algorithm was tested and it was compared with the high-resolution ensemble Kalman filter method. The results show that in the first assimilation cycle, the center locations of high and low values of simulated increments of horizontal wind field and disturbed potential temperature field at 500 hPa and their observed increment fields are same for both methods, and the structures are close to the observational increment field. The increment value from the approach of the high resolution ensemble Kalman filter is closer to the observational value than that from the dual-resolution method. In the forecast-assimilation cycle test, the root mean square errors from the two methods decrease generally with the increase of assimilation number and it suggests that both methods have compatible assimilation abilities. The result from the dual-resolution is poorer than that of the high resolution method, but the running time of the former is only 1/6 of that of the latter.

Key words: Ensemble Kalman Filter, Resolution, Forecast error covariance matrix, Computational time cost