基于最优化理论的大气气溶胶偏振遥感反演方法研究

Nowadays, the atmospheric aerosol polarized remote sensing inversion algorithm is a hot international research topic. In this project, fully considering the multi-angle and polarized observations information from the satellite and airborne sensors, combined with the constraints of prior knowledge as well as the bidirectional polarization distribution function (BRDF) and bidirectional reflectance distribution function (BPDF) model of surface, we try to use the Stokes and Jacobian information to set the optimization inversion modeling. Those Stokes and Jacobian information could be simulated by the unified linearized - vector radiative transfer model (UNL-VRTM), and then the inversion problem of aerosol remote sensing polarization could be transformed into an optimization problem, which can be iteratively solved by the L-BFGS-B optimization algorithm. Thus, for each retrieval parameters, the operational plan could be proposed, the joint inversion with aerosol and surface parameters could be completed, and the ill-posed inversion problem also could be effectively solved. Meanwhile, considering the sensors’ actual observation and specific source of errors, the observation information of observation and error transfer of the inversion model could be quantitative evaluated, which provides a useful reference basis for the inversion modeling and algorithm design. This study does not need to relay on a lookup table, and the real-time and dynamic joint inversion could be achieved flexibly. With the study in this project, the inversion accuracy could be effectively improved, and more aerosol information such as column volume concentration, particle size distribution, complex refractive index and the Angstrom index could be retrieved, further providing the technical support for the environmental remote sensing monitoring.

大气气溶胶偏振遥感反演是当前的一个研究热点，本项目充分考虑卫星和机载传感器多角度偏振观测的信息，结合先验知识的约束以及地表BRDF和BPDF模型，基于线性化的矢量辐射传输模型UNL-VRTM前向模拟的Stokes和灵敏度（Jacobian）信息，研究进行最优化反演建模，将气溶胶遥感偏振反演问题转化为一个最优化问题，进而利用L-BFGS-B优化算法进行迭代求解，针对各反演要素提出了可操作的实施方案，可实现气溶胶和地表多参数的联合反演，有效解决病态反演问题。同时，综合考虑实际观测和具体误差来源，对传感器的观测信息量及模型的误差传递进行定量分析和评价，为反演建模和算法设计提供可参考的依据。本研究不需要建立查找表，可灵活地进行实时和动态多参数的联合反演，预期可有效提高遥感反演精度，反演得到如体积柱浓度、粒子谱分布、复折射指数和Angstrom指数等更多的气溶胶信息，为环境遥感监测提供支撑。

大气气溶胶偏振遥感是当前的一个研究热点，而病态反演是其中不可避免的难题。本项目基于最优化理论，对大气气溶胶偏振遥感反演方法进行了系统研究。在遥感反演理论研究的同时，重点针对各反演要素提出了可操作的反演算法，建立了面向偏振遥感观测的最优化反演理论框架及信息分析方法。主要研究内容可分为3个部分：1）偏振卫星遥感观测前向模拟及参数敏感性分析研究；2）气溶胶偏振卫星遥感最优化反演建模及求解研究；3）最优化反演框架下的地气解耦方法及气溶胶关键光学参数反演结果地基验证研究。.本研究充分利用角度、光谱和偏振的观测信息，通过引入改进的地表双向反射分布函数（BRDF）和单参数地表双向偏振反射分布函数（BPDF）模型，发展了可进行气溶胶和地表多参数反演的最优化反演框架，理论上解决了多光谱、多角度标量和偏振遥感地气解耦的难题，算法上实现了气溶胶关键的微物理参数（分模态的体积柱浓度等）、光学参数（光谱的气溶胶光学厚度AOD和细粒子比FMF）及地表参数的联合反演，并利用信息量指标对可反演的关键参数进行了定量分析和验证，为我国新型偏振卫星传感器的广泛应用提供理论和算法等方面的关键支撑。.在最优化反演框架下，通过引入经验正交函数（EOF）进行多角度卫星遥感标量观测的地表反射估算；同时将所发展的“残差分组排序”评价函数与LOA官方算法相结合，可分别反演得到总AOD和细模态AOD，进而得到FMF的反演结果，并实现了基于PARASOL卫星观测的气溶胶关键光学参数区域反演。通过利用中国区域4个AERONET站点地基同步验证，反演获取的AOD和FMF参数均具有较高的精度，相比MODIS和LOA官方产品均有较大提升。该方法不受地表类型的限制，能够获得亮地表情况下的反演结果。本研究成果可经过进一步完善，后续应用到高分五号卫星DPC遥感数据的气溶胶反演研究上，对我国环境遥感监测提供有效支撑。