ETKF初值扰动方法中真实观测及扰动调节因子研究

目前国家气象中心业务GRAPES区域集合预报系统中集合变换卡尔曼滤波（ETKF）方法采用的是模拟观测信息,为进一步完善ETKF方法,拟对ETKF初值扰动通过引入真实探空观测资料,使扰动场能够代表真实观测的不确定信息,改善区域集合预报技巧。真实观测资料的引入会使得每日的观测数目和分布发生变化,这对ETKF方法而言可能会引起扰动振幅的不稳定,因此在引入真实观测资料的基础上设计了新的扰动振幅调节因子,通过格点空间中离散度和均方根误差关系来对初值扰动振幅进行自适应调整。从初值扰动结构、概率预报技巧以及降水预报效果等方面对比分析了基于模拟观测、真实观测以及真实观测结合新型调节因子的ETKF方案的差异,结果表明:真实探空资料能够有效应用于GRAPES区域集合预报系统中,真实观测资料与模拟观测资料相比较为稀疏,可以获得更大量级的初值扰动振幅;真实观测资料有助于提高区域集合的离散度,但对集合预报准确度以及概率预报结果的提高有限,对于降水预报效果提高也有限;新型的扰动振幅调节因子可以有效获得稳定的初值扰动振幅,并保持ETKF扰动结构,真实观测资料与扰动振幅自适应调节因子相结合,可以有效提高区域集合的概率预报结果,并有效提高降水预报效果。

Study of the Application of Real Observation Data and a Rescaling Factor in Ensemble Transform Kalman Filter Initial Perturbation Method

At present, ETKF (ensemble transform Kalman filter) method used in the operational GRAPES (Global/Regional Assimilation Prediction Enhanced System) regional ensemble prediction system of National Meteorological Center uses pseudo observation information, and the number and distribution of observations are fixed. To improve the ETKF method, real observation data are introduced into the ETKF process. The real observational radio-sounding data enable the perturbation field to represent uncertainty information about the observational state. Considering that the number and distribution of real observational data change daily, this can cause instability in the perturbation amplitude for the ETKF calculation. Therefore, the authors introduce a new self-adjustment amplitude rescaling factor. In this study, the authors analyzed ETKF schemes based on pseudo observations, real observations, and real observation plus the new rescaling factor and compared them in terms of their perturbation characteristics, ensemble verifications, and precipitation forecast skills. The results show that real sounding data can be effectively applied to the GRAPES regional ensemble forecasting system. Compared with pseudo observation data, real observation data is sparse, so large initial perturbation amplitudes can be obtained. The use of real observation data can help to improve the spread of the regional ensemble, but the improvements in the ensemble prediction accuracy and probability forecast skill are limited, as is the improvement in the precipitation prediction. The authors designed a new perturbation-amplitude rescaling factor to adaptively adjust the initial perturbation amplitude based on the spread and root-mean-square-error relationship in the grid space. Our investigation of the adaptive rescaling factor for adjusting the perturbation amplitude shows that this new rescaling factor can effectively obtain a stable initial perturbation amplitude and maintain the ETKF-generated perturbation structure. Since the real-observation-based ETKF scheme exhibits over-spread characteristics and limited improvement with respect to the pseudo-observation-based ETKF, the use of real observation data combined with the adaptive rescaling factor can effectively improve the skill of the regional ensemble in terms of the probabilistic forecast results while also effectively improving the precipitation forecasting skill.