卫星红外高光谱反演大气水汽含量

A Study of Integrated Water Vapor Retrieval Using In-frared Hyperspectral Soundings

利用高光谱大气红外探测仪AIRS模拟及观测数据,发展基于主成分分析技术的多层前馈神经网络反演算法,进行大气中水汽柱总量(IWV)的反演计算、模拟及实测验证。首先,基于全球晴空大气廓线训练样本SeeBorV4.0,利用快速辐射传输模式CRTM进行了辐射传输模拟计算,得到全球高光谱分辨率模拟辐亮度;其次,利用主成分分析技术对模式模拟和AIRS实测高光谱数据进行降维、去噪及去相关处理,并采用多层前向神经网络算法反演大气水汽柱总量;最后,利用数值试验、AIRS实测L1B数据及其水汽产品,对反演算法进行了验证。通过与AIRS官方大气产品的统计分析,本算法反演均方根误差为0.387g/cm2,最大偏差为0.82g/cm2,空间分辨率保留了AIRS像素原分辨率(比AIRS官方大气产品高3倍)。

This paper mainly focused on some of the basic and critical questions of Integrated Water Vapor(IWV)inversion,including the radiative transfer simulation,hyperspectral data processing and the inversion method of the IWV.A new IWV retrieval algorithm based on the Principal Component Analysis PCA and Multi-layer Feed-forward back-propagation Neural Network(MFNN)is developed,which is referred as PCA-MFNN method.In this study,the radiative transfer simulation has been carried out with CRTM(Community Radiative Transfer Mode)combined with the latest global assimilated data which is from the CIMSS.The PCA is used to reduce the dimension,and eliminate the noise and correlation from the hyperspectral remote sensing data.A three layer forward neural network has been constructed,trained and tested,and then the IWV is retrieved using the optimized network.Compared with the AIRS L2 IWV product,the root mean square error of the retrieved integrated water vapor is 0.387g/cm2.