风云四号高光谱GIIRS水汽通道最优选择与变分同化方法研究

The quantitative application of water vapor channel data derived from hyperspectral atmospheric infrared sounder of geostationary orbit in Numerical Weather Prediction (NWP), with both high temporal and vertical resolution, was highly expected to improve the accuracy of high impact weather forecasting which was closely related to wet process, such as typhoon and rainstorm. And as one of the key technologies of effective assimilation, optimal channel selection for water vapor can reduce ill-posed problems of both assimilation and retrieval caused by redundant information of observed data from hyper-spectral sounder. Firstly, with a thorough understanding of the strong nonlinear characteristics of water vapor channel of Geostationary Interferometric Infrared Sounder (GIIRS) of FY-4A, optimal selection method suitable for water vapor channel was developed on the basis of entropy reduction(ER)-Shannon entropy and relative entropy and channel score index (CSI), and thererfore channel combination for East Asia Region was formed. Secondly, Considering the Non-Gaussian characteristics with the brightness temperature bias of water vapor channel, the commonly used M-estimators and new M-estimators, were coupled to the framework of the classical variational assimilation, so that researches on the non-gaussian generalized variational data assimilation method for GIIRS water vapor channel were carried on. Finally, the effect verification experiment of different channel combinations and new variational methods applied in the (3DVar/4DVar) assimilation and forecasting of GRAPES were given in this study. Comparative experimentation were designed, through which the impact of analysis filed and forecast filed, which provides a reference for the assimilation and application of GIIRS water vapor channel in numerical prediction.

静止轨道上高光谱红外探测仪水汽通道资料在数值预报中的定量应用，由于其高时间分辨率和高垂直分辨率特点，有望改进台风和暴雨等与湿过程密切相关的高影响天气预报，而水汽通道最优选择是有效同化的关键技术之一，能减少高光谱探测仪观测冗余信息所引起的同化和反演不适定性。本项目首先基于风云四号A星高光谱大气垂直探测仪GIIRS水汽通道非线性较强的特点，在“熵减法-香农熵”基础上，基于“相对熵”和“通道得分指数法”，发展适用水汽通道的最优选择方法，形成东亚区域通道组合；其次，针对水汽通道亮温偏差非高斯性，把常用M-估计和新构建M-估计法耦合到经典变分同化框架中，开展GIIRS水汽通道非高斯广义变分同化方法研究；最后，开展不同通道组合和新变分法在GRAPES-3DVar/4DVar同化及数值预报效果验证。设计对比试验，研究对分析和预报的影响，为GIIRS水汽通道在数值预报中的同化应用提供参考。

风云四号高光谱GIIRS资料在数值天气预报中的定量应用，有望改进高影响天气预报。本项目完成了FY-4A/GIIRS水汽通道最优选择与变分同化方法研究及GRAPES应用试验和相关算法推广应用。主要内容如下：（1）通道选择。在“熵减法-香农熵”的通道最优选择基础上，针对GIIRS水汽通道特点，采用了基于“相对熵”的非线性方法。进一步采用了改进的“通道得分指数法”-信号自由度方法。形成了GIIRS通道子集。本项目采用的方法均有效地改进了背景误差分布，尤其在中间层。（2）变分同化。基于观测误差重估计和M-估计构建了高光谱通道亮温非高斯广义变分同化方法。试验结果表明，基于Huber-估计的广义变分同化效果优于经典变分同化。进一步基于带约束项广义变分同化了GIIRS亮温，得到了较高的温度廓线反演精度。基于随机森林反演了高频次的温度廓线。并综合多源（或多维度）资料完成了台风暴雨高影响天气监测研究。（3）质量控制。在将GIIRS资料同化进GRAPES模式前，完成了亮温重构、异常值剔除、偏差订正和云检测等资料质量控制。并初步实现了将云参数等信息用于GIIRS云区亮温模拟；（4）GRAPES应用试验。在应用层面，完成了GIIRS资料在GRAPES变分同化及降水预报效果验证。结论为降水预报精度整体有所改进。进一步项目组实现了GIIRS资料在WRFDA中的同化应用。完成了对台风“玛莉亚”分析影响的试验；（5）算法推广应用。将本项目方法推广应用于FY-3D/HIRAS在GRAPES-4DVar台风路径数值预报，改进了预报精度；进一步完成了卫星红外资料（如，FY-4A/AGRI）反演降水等。本项目是卫星资料同化、数值天气预报、数学反问题、正则化约束、人工智能等多学科交叉融合算法研究与应用结合体。本项目目的实现FY-4A/GIIRS资料在区域部门业务定量化应用，以服务于气象防灾减灾。