物理滤波初始化在4DVar中的应用

发布时间：2018-05-26 15:01

本文选题：资料同化 + WRF模式　；参考：《中国气象科学研究院》2017年硕士论文

【摘要】：通常情况下,由于背景误差协方差的不确定、观测误差或模式本身误差等因素,资料同化过程所得到的结果不能在动力上达到较好的平衡性,因此为了减小或消除分析场中存在的虚假的高频振荡,气象学者引入了初始化的概念,近年来,数值滤波初始化(Digital Filter Initialization,DFI)被广泛应用于气象预报业务系统中,DFI方法是通过设置截断波数来进行滤波,但它的物理意义并不明确;另一种正在发展的在同化过程中阻尼高频噪音的方法是通过引入物理约束条件,如引入模式约束的同化方案(Model Constrained-3DVar),它在同化过程中极小化了模式变量的时间倾向来得到最优分析场。在本研究中,基于模式约束方案的一种物理滤波初始化方法(Physical Filter Initialization,PFI)应用于WRF模式四维变分同化系统中,分别进行单点试验和真实降水个例的同化和模拟,通过对比传统4DVar和PFI-4DVar试验结果的差异,来验证PFI方案的合理性和有效性。主要结论如下:1、传统4DVar同化过程中在加入了观测信息后,由于背景误差协方差的不确定性等因素,引起分析场存在高频振荡,这个高频振荡对模式积分预报会产生一定影响,并随积分时间不断向外扩散;PFI-4DVar在同化过程中,以模式的动力和物理过程作为弱约束条件,使得同化所得分析场中各变量之间有较好的平衡性和协调性,即有效抑制了同化所得分析场中出现的高频振荡。2、PFI-4DVar方法同化所得分析场中模式变量之间都能具有较优的流依赖特性,并能在积分过程中始终保持这种流依赖特征发展;初始时刻变量间相关性和协调性较好,并且在积分过程中,变量间始终保持相互协调,没有出现高频振荡随积分时间向外传播的现象。3、传统4DVar和PFI-4DVar同化所得分析场作为初始条件均能有效提高短时降水预报的准确性,而值得一提的是PFI-4DVar可以更有效预报前六小时降水,无论从降水落区或是评分效果看,前六小时的降水预报都有一定提高。这说明由于加入了模式作为弱约束条件,减小了分析场中的高频振荡,使得分析场变量间协调并合理,进一步导致模式积分过程中初始调整时间大大缩短,即减少了模式spin-up时间,从而提高了预报初期的准确性。4、在弱化BE矩阵作用和影响范围的前提下,在传统BE矩阵的作用下,观测信息和背景场误差存在不平衡地、较大范围地在背景场中向外传递的现象,导致分析场中变量之间不协调;同时初始时刻高度场和风场主要为气压梯度力和加速度的平衡,直至积分60min后高度场增量和风场增量之间开始逐渐建立起地转关系。初步说明了PFI-4DVar方法在传递背景场误差和观测信息时,能更加遵循模式中物理和动力模型,同时能更好地协调变量之间关系,进而得到一个平衡的分析场,从而缩短模式积分的spin-up时间。同时也初步证实了PFI方案有部分代替背景误差协方差矩阵,充当同化过程中传递观测信息的功能。
[Abstract]:Generally, because of the uncertainty of the covariance of the background error, the error of the observation or the error of the pattern itself, the results obtained by the data assimilation process can not achieve a better balance in the dynamics. Therefore, in order to reduce or eliminate the false high frequency oscillation in the analysis field, the meteorologist introduced the concept of initialization, in recent years, Digital Filter Initialization (DFI) is widely used in the weather forecasting business system. The DFI method is filtered by setting the truncated wavenumber, but its physical meaning is not clear; another developing method of damping high frequency noise in the process of assimilation is by introducing physical constraints, such as citation. In this study, a physical filtering initialization method based on schema constraints (Physical Filter Initialization, PFI) is applied to the WRF model four-dimensional variational assimilation system, in the assimilation process, which minimizes the time tendencies of the pattern variables in the assimilation process. Through the assimilation and Simulation of single point test and real precipitation case, the rationality and effectiveness of the PFI scheme are verified by comparing the differences between the traditional 4DVar and PFI-4DVar test results. The main conclusions are as follows: 1, after the observation information was added to the traditional 4DVar assimilation process, the uncertainties of the background error covariance were caused by the factors such as the uncertainty of the background error covariance. There is a high frequency oscillation in the analysis field. This high frequency oscillation will have a certain influence on the model integral prediction and spread out with the integration time. In the process of assimilation, PFI-4DVar takes the dynamic and physical processes of the model as a weak constraint condition, so that there is a better balance and coordination among the variables in the analysis field of assimilation, that is to say, it is effective. The high frequency oscillation.2 appeared in the analysis field of assimilation is suppressed. The model variables in the analysis field of the PFI-4DVar method assimilation can have better flow dependence, and can always keep the flow dependence in the integration process; the correlation and coordination between the initial time variables is better, and in the integration process, the variables are among the variables. It is always consistent with each other, and there is no phenomenon of high frequency oscillation spreading out with integral time.3. The traditional 4DVar and PFI-4DVar assimilation analysis field can effectively improve the accuracy of short-time precipitation prediction as the initial condition. It is worth mentioning that PFI-4DVar can be more effective in pre reporting the first six hours of precipitation, whether from precipitation area or evaluation. As a result, the precipitation forecast in the first six hours has been improved to a certain extent. This shows that the addition of the model as a weak constraint condition reduces the high frequency oscillation in the analysis field, makes the analysis of the field variables coordinated and reasonable, and further leads to a large reduction in the initial adjustment time in the model integration process, that is, reducing the time of the model spin-up and thus increasing the time of the model. The accuracy of the early prediction is.4. Under the premise of weakening the effect and the influence range of the BE matrix, under the effect of the traditional BE matrix, the observation information and the background field error are unbalance, and the phenomenon of the larger range in the background field leads to the disharmony between the variables in the analysis field; at the same time, the initial time height field and the wind field are mainly as a result. The balance between the pressure gradient force and the acceleration is gradually established between the increment of the height field and the increment of the wind field after the integration of 60min. It is shown preliminarily that when the PFI-4DVar method passes the background field error and the observation information, the physical and dynamic models in the model can be more followed, and the relationship between the variables can be better coordinated, and then the relationship between the variables can be better coordinated. To a balanced analysis field, the spin-up time of the model integral is shortened, and it is also preliminarily proved that the PFI scheme has partially replaced the background error covariance matrix, which serves as the function of transmitting observation information in the process of assimilation.
【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】