基于星载红外高光谱观测用机器学习和物理的算法反演全天候大气参数

大气温度和湿度是天气预报和大气科学研究最基本的参数，全天候（晴空和有云）的高时间分辨率、高垂直分辨率、高精度三维大气温湿廓线的反演是十分重要而又迫切需要解决的科学问题。本项目将针对世界上首次搭载在静止轨道卫星平台上的高光谱红外干涉式探测仪GIIRS提供的高频次的辐射观测值开展全天候大气温湿垂直廓线反演算法研究，拟同时采用具有非线性处理能力的卷积神经网络的深度机器学习算法和物理意义明确的一维变分反演算法。研究内容包括：矩形面阵探元的观测误差分析；基于自身观测的双CO2吸收带云检测方法；最优反演算法的评估；反演结果的评价方法等。采用多种资料源检验大气温度、湿度垂直廓线反演精度，希望可满足业务数值天气预报资料同化系统的需要。本项目研究的反演算法可推广到FY-3D和未来国际静止卫星平台上同类红外高光谱探测资料的应用上。

The development of global climate and weather models requires accurate monitoring of atmospheric temperature and moisture profiles. Three-dimensional retrieval of atmospheric temperature and humidity profile with high temporal resolution, high vertical resolution and high accuracy under all weather is a scientific problem that is very important and urgently needed to be solved. Geostationary Interferometric Infra-Red Sounder (GIIRS) on Chinese FY-4A satellite is the first infrared high-spectral-resolution interferometric sounder uploaded on geostationary satellite platform in the world. The retrieval study of all-weather atmospheric temperature and humidity profile based on GIIRS high-frequency radiation measurements will be developed in this proposal. The deep machine learning algorithms, namely the convolutional neural network (CNN) technology with nonlinear processing ability and one-dimensional variational physical inversion algorithm will be tested at the same time in this research. The research contents include observation error analysis owing to GIIRS specific rectangular array detectors, detecting clouds located in different layers of the atmosphere just based on GIIRS measurements using combining double CO2 bands method, investigating the advantages and disadvantages of different inversion algorithms, the evaluation method of inversion results, and so on. Retrieval accuracy of atmospheric parameter profile will be evaluated by comparing with various sources, such as radiosonde observations and ECMWF (European Center of Medium-Range Weather Forecasts) analyses, et al. The improved retrieval quality which can satisfy with operational numerical weather forecast data assimilation system is expected. The developed retrieval algorithm in this proposal can be extended to level 2 atmospheric product developments in term of the similar infrared high-spectral-resolution interferometric sounder, uploaded on Chinese FY-3D polar-orbiting satellite and international geostationary satellite platform in future.