龚建东. 2021. 集合预报误差在GRAPES全球四维变分同化的应用研究:I局地化方案设计与动力平衡特征分析[J]. 气象学报, (0):-, doi:10.11676/qxxb2021.006
集合预报误差在GRAPES全球四维变分同化的应用研究:I局地化方案设计与动力平衡特征分析
GONG Jiandong WANG RuichunNational Meteorological Center/Numerical Weather Prediction Center of CMA, Beijing, 100081
投稿时间:2020-04-24  修订日期:2020-09-15
DOI:10.11676/qxxb2021.006
中文关键词:  集合预报误差  水平与垂直局地化方案 地转分析平衡特征 GRAPES 4D-Var 数值试验
英文关键词:Ensemble  forecast error, horizontal  and vertical  localization, analysis  geostrophic balance, GRAPES 4D-Var, Numerical  experiment.
基金项目:
作者单位E-mail
龚建东 国家气象中心 gongjd@cma.gov.cn 
摘要点击次数: 46
全文下载次数: 54
中文摘要:
      [目的]本文研究了在GRAPES全球四维变分资料同化(4D-Var)中有效应用集合预报误差信息的方法,在大样本集合预报条件下,增加扩展控制变量时降低其空间维数增加,减少对高性能计算机存储和计算消耗,并对水平与垂直局地化对分析地转平衡的影响进行了分析。[资料和方法]本文利用高斯分布的谱滤波实现水平局地化、利用垂直正交EOF分解实现垂直局地化,并采用前8个主导特征模态来限制空间维数增加。[结果]引入20至180个集合预报样本,水平局地化后扩展控制变量增加了1.1到1.8倍,墙钟时间增加了约1.1到1.6倍。三维局地化方案扩展控制变量增加了约1.7到7.1倍,墙钟时间增加了约1.4到4.1倍。通过采用在非平衡分析变量上进行水平局地化,然后再将风压地转平衡关系重新叠加到非平衡分析变量上,所获得的分析保持了风压平衡关系,初始场地面气压倾向变化小。垂直局地化对分析平衡影响较大,由于目标函数中控制重力波的目标函数项的存在,分析场仍能保证动力平衡约束关系。[结论] 论文在GRAPES 4D-Var中所发展的引入扩展控制变量的方式具备实用性。
英文摘要:
      The research focuses on effectively utilizing ensemble forecast error information into GRAPES global four-dimensional variational data assimilation (4D-Var). Under a large number of ensemble members condition, the paper adds extra alpha control variable in 4D-Var with a limitation on control variable dimension increasing, to reduce the requirement on data storage and computation cost in high-performance computer system. The horizontal and vertical localization impact on dynamical balance of analysis field is also analyzed. The results show that, using Gaussian shape spectral filter for horizontal localization and vertical EOF decomposition for vertical localization, with only first 8 leading eigenvector mode to limit the alpha control variable dimension, with 20 to 180 ensemble members, for horizontal localization the extend alpha control variable size increases 1.1 to 1.8 times, the inner loop wall clock increases 1.1 to 1.6 times, separately. For three-dimensional localization they increase 1.7 to 7.1 times and 1.4 to 4.1 times separately. After horizontal localization applied on unbalanced analysis variable, and then the geostrophic balance is added back, the analysis field will keep geostrophic balance, the surface pressure tendency is smaller. But the vertical localization has large negative impact on balance, by companying with a constraint term in cost-function to suppress the gravity noise, the analysis also keep geostrophic balance characteristics well. The scheme developed by this paper is suitable for large ensemble members.
查看全文   查看/发表评论  下载PDF阅读器
分享按钮