基于带约束项广义变分同化AIRS云影响亮温研究

Generalised Variational Assimilation of Cloud-affected Brightness Temperature of AIRS Data based on the Constraint Terms

经典变分同化基于误差服从高斯分布理论,在同化受云影响的红外探测器通道亮温时,需进行云检测或只同化权重函数峰值位于云顶之上的通道亮温。在云检测过程中需对亮温进行严格的质量控制以剔除"离群值",导致丢失大量有用数据。文中基于带约束项非高斯模型的广义变分对高光谱大气红外探测器(Atmospheric Infra-Red Sounder,AIRS)受云影响亮温进行了初步的同化研究。在执行过程中,首先,动态选择AIRS通道形成通道子集。其次,把云参数(有效云量和有效云顶气压)作为辐射传输模式输入变量参与变分同化极小化迭代并用于通道子集亮温模拟。同化试验结果表明,对于高云,基于Cauchy-估计的变分同化反演结果最好,而对于中云和低云,基于Huber-估计得到了较好的同化反演结果。然而,在反演模式高层温度时,基于Fair-估计反而得到了较差的结果,但其对于湿度反演效果较为理想,其结果可能与Fair-估计分布的固有特点有关。带约束项广义变分同化方法对受云影响亮温的同化效果比经典变分方法的好,但依赖于M-估计的选取。

Assimilated channel brightness temperature data from infrared sounders accounting for cloud effects has a positive effect on weather forecasting, especially in weather-sensitive areas. When cloud parameters inclu- ding effective cloud fraction and effective cloud top pressure are considered in the simulation on channel bright- ness temperature of the infrared sounders, the deviation of brightness temperature follows strong Non-Gaussian. The classical variational assimilation requires the observational errors to follow a Gaussian distribution to apply the least-square principle. The least-squares method is sensitive to outliers; if the analyzed data contain gross er- rors, the parameter estimation is inaccurate. When processing the cloud-affected brightness temperature, cloud detection or assimilating specific channel brightness temperature with weight function peaks above the cloud top were needed. Useful data were lost through the cloud detection process to eliminate the so-called outliers. The outliers are not always harmful, which may represent new information, such as weather phenomena. At present, the quality control is generally based on a certain threshold value if the subjective uncertainty is too strong. If out- liers persist after the quality control, the optimal parametric results obtained by the classical variational assimila- tion are meaningless. In this paper, Atmospheric Infra-Red Sounder （AIRS） brightness temperature channel which affected by cloud, were assimilated by generalized variation method from the constraint terms of Non- Gaussian model. It combines both the advantages of classical variational assimilation and robust M-estimators. Generalized one is coupled with quality control in the process of assimilation. The main idea is to use weighting factor of M-estimators to re-estimate the contribution rate of the observation items to the objective function in each process of objective function minimization based on the classical variational assimilation. The cost function contains the M-estimators to guarantee the robustness to outliers, thus to improve the assimilation results. Numeri- cal algorithm steps of the generalized variational assimilation are as follows ： Firstly, a channel set was formed by dynamically selecting AIRS channels based on the temperature Jacobian matrix in each field-of-view （FOV）. Secondly, generalized variational assimilation of cloud parameters, which were input variables in radiative trans- fer model （such as, RTTOV）, were involved in the variational minimisation iteration process, and were used to simulate AIRS brightness temperature of channel set. Assimilation experiment demonstrated that for high clouds, the Cauchy estimator inversion results are the best, whereas for the mid and low clouds, the Huber estimator pro- vides the best results. However, the inversion results for temperature in the high level of model is worse using the Fair estimator, and, on the contrary, the inversion results for humidity are good. These results may reflect an in- herent characteristic of Fair distribution. The assimilated results in cloud-affected brightness temperature are better by using generalized variation method than the classical method, but the former depends on selecting M-estimator weight functions. The preliminary results also demonstrated the potential application value of generalized varia- tional assimilation.