【摘要】 随着卫星遥感关键技术的突破,卫星光谱分辨率达到了分辨大气成分单个谱线的水平,研究人员开始了大量通道同时反演大气廓线和多种微量成分的研究。针对AIRS(Atmospheric Infrared Sounder)就红外高光谱资料反演大气水汽廓线的研究进展进行了评述,从训练数据、通道信息的提取及降维、反演算法和反演精度改进4个方面对反演晴空大气水汽廓线的研究现状进行了分析与讨论。AIRS资料反演大气水汽廓线的训练数据通常选用威斯康星大学提供的全球晴空反演训练样本集CIMSS(Cooperative Institute for Meteorological Satellite Studies,University of Wisconsin-Madison)和SARTA(Stand-Alone Radiative Transfer Algorithm)辐射传输模式模拟的亮温辐射值。归纳总结了2种通道信息的提取及降维方法:一是采用有效的方法来完成光谱信息压缩,对常用的主成分分析和独立分量分析方法进行了对比,认为独立分量分析更为可行。二是通道选择,即保留部分含有较多大气廓线信息量的通道,达到降维目的。在进行通道选择时要注意针对不同地区气候类型、下垫面、季节以及即时天气条件,选择不同的通道组合。介绍了3种反演算法:特征向量统计法、牛顿非线性迭代法和神经网络法。对比发现特征向量统计法简单易行,但精度不够理想;牛顿非线性迭代法精度虽高但计算耗时长,因此不适合业务使用;神经网络计算速度快、精度也能达到要求,具有很好的前景。对目前的几种样本分类方法及附加因子进行了对比分析,对反演算法精度的改进提出了一些有益的设想。最后对晴空辐射订正及云天大气水汽廓线反演进行了简要介绍,提出了该领域未来的一些研究方向。更多还原

【Abstract】 With the breakthrough of the satellite remote sensing key technology,satellite spectral resolution has reached to the level of distinguish between each spectral line,the researchers begin to research atmosphere profile and various micro constituent inversion using huge number of channels at the same time.This paper make a comment of the advances in Atmospheric humidity profile inversion using AIRS data,Analyzed and discussed the research status of clear sky atmospheric profiles from four aspects: training data,information extraction and dimensionality reduction of channels,inversion algorithm and accuracy improvement of Inversion.CIMSS and brightness temperature simulated used SARTA are usually chose to retrieve water vapor profile using AIRS data.Summarized two kinds of methods to do information extraction and dimensionality reduction,the first is spectral information compression,compared with PCA found that ICA is more practical.The second is channel selection,that is keeping part of channel contained more atmospheric profile information to achieve dimensionality reduction.It is worth mentioning that we need to choose different channel combination under different regional climate types,underlying surface,season,and real-time weather condition.Introduces three kinds of inversion algorithm: eigenvector regression algorithms(ERA),the Newton method(NM),and aritificial neural network algorithms(ANN).Compare the three methods found that ERA is simple,but the precision is not ideal;NW has high precision but it is not suitable for business application due to its long computing time;ANN has high computing speed and its precision can meet the requirements,thus,ANN has excellent foreground.Analyzed several samples classification method and additional factors,and give some suggestions to improve inversion algorithm.Finally,provide a brief introduction of infrared cloud-cleared and inversion atmospheric profile under cloud condition.更多还原