考虑“流依赖”特征的风云三号卫星微波辐射资料同化和质量控制方法研究

At present, the application of satellite data is not sufficient in typhoon prediction in China. The microwave radiance soundings of FengYun-3 meteorological satellite provides vertical information for atmospheric temperature and humidity. During typhoon periods, the atmospheric systems appear complex, nonlinear and fast changing. And the assimilation of massive satellite data cost a lot of time. In this project, the Ensemble Reduced-Dimensional Variational method (ERDVar for short) is applied to the microwave radiance data of Fengyun-3 meteorological satellite. The SVD technique is used to extract the orthogonal basis vectors which can represent the spatial structure and temporal evolution of the analysis variables. This method shows a "flow-dependent" feature with the atmospheric dynamic changes and does not need to integrate adjoint mode, which improves the calculation efficiency of assimilation greatly. Offshore typhoon observation mainly relies on satellite data, and the existence of cloud and precipitation causes bias between microwave radiance observation and the model. The pixels polluted by cloud and precipitation should be filtered in assimilation preprocess. Cloudy pixels cann’t be directly identified because of the frequency settings of microwave temperature sounding on Fengyun-3 satellite, while the precipitation cloud cann’t be distinguished from the non-precipitation cloud through the alternative static threshold method. In this project, focus on cloud detection problem of satellite microwave radiance data, it is analyzed that the relationships among the atmospheric main characteristic vector derived from the ensemble forecast samples, typhoon observations from satellites as well as the microwave radiance observation increments. We will study the atmospheric satellite data quality control method appropriate to the main characteristics, to solve the detection for precipitation cloud or the non-precipitation cloud in Fengyun-3 satellite microwave data assimilation.

远海台风观测主要依靠卫星资料，目前卫星资料在我国台风预报中应用不充分，风云三号气象卫星微波辐射探测提供大气温度、湿度垂直信息；台风期间大气系统复杂、非线性、快变；海量卫星数据同化很耗时。本项目将集合降维变分方法应用于风云三号气象卫星微波辐射数据，利用奇异值分解SVD技术在四维空间的预报集合，提取能够表现分析变量空间结构、时间演变的正交基向量，具有“流依赖”性，不需积分伴随模式，提高了同化计算效率。云和降水造成微波辐射观测和模式的偏差,同化预处理需剔除受其影响的像元。风云三号微波温度计的探测频点不能直接判识有云像元，而静态阈值方法不能区分降水云和非降水云。本项目针对风云三号卫星微波辐射资料的云检测问题，分析集合预报样本的主要特征向量、台风系统卫星观测和微波资料观测增量三者之间的关系，重点研究与大气主要特征相适应的卫星资料质量控制方法，解决风云三号卫星微波资料同化中降水云和非降水云的判识问题。

本项目研究了全天空风云卫星微波资料同化方法和集合降维同化方法。本项目以风云三号微波成像仪资料和风云三号微波湿度计资料为研究对象。首先，针对全天空的观测误差不满足资料同化中观测误差为高斯分布的问题，基于云对称性假定，建立了风云三号卫星微波成像仪资料全天空观测误差模型，研究了微波资料晴空质量控制方法和全天空质量控制方法，选取超级台风“玛莉亚”和“利奇马”个例，完成全天空微波资料偏差特征分析。结果表明：经过全天空观测误差模型处理后，全天空观测偏差接近高斯分布，相比晴空同化增加43.9～54.63%可同化资料量，这些新增资料大多来自台风螺旋云带和云墙部分区域；其次，选取2018年7月的台风“玛莉亚”，利用WRF模式及其同化系统WRFDA，进行风云三号C星微波湿度计观测资料的全天空同化试验。与晴空同化试验相比，全天空条件下更多的云雨区域观测资料被有效利用，能够更好地模拟出台风“玛莉亚”核心区域的暖心和对称风速结构，有效改善湿度场的预报，台风路径预报误差平均降低约34~62%；第三， 基于全球数值预报模式T106L19，研究集合降维变分同化方法ERDVar，不需要求解切线性模式和伴随模式，不仅能减少同化计算量，而且能够提供“流依赖”的背景误差协方差矩阵。本项目提出用NMC初始扰动生成方法和分区同化方案，来解决初始扰动样本衰减问题和全球同化局地化问题，分别使预报误差降低10%和14%。