地表方向反射特性对AOD反演误差的影响研究  

Study on the Influence of Surface Non-Lambertian Reflection Characteristics on AOD Retrieval Error

作　　者：冀哲 李正强 马䶮 姚前 张鹏 陈震霆 JI Zhe;LI Zheng-qiang;MA Yan;YAO Qian;ZHANG Peng;CHEN Zhen-ting(State Environmental Protection Key Laboratory of Satellite Remote Sensing,Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100101,China;Key Laboratory of Remote Sensing and Digital Earth,Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100101,China;University of Chinese Academy of Sciences,Beijing 100049,China;Aerospace Information Technology University,Jinan 250200,China;School of Information Engineering,Kunming University,Kunming 650214,China)

机构地区：[1]中国科学院空天信息创新研究院国家环境保护卫星遥感重点实验室,北京100101 [2]中国科学院空天信息创新研究院遥感与数字地球重点实验室(中国科学院),北京100101 [3]中国科学院大学,北京100049 [4]空天信息大学(筹),山东济南250200 [5]昆明学院信息技术学院,云南昆明650214

出　　处：《光谱学与光谱分析》2025年第3期852-862,共11页Spectroscopy and Spectral Analysis

基　　金：国家杰出青年科学基金项目(41925019);对外技术合作科研项目(E3KZ03010F);国际大气环境遥感科研开放平台建设(E2D1030200);云南省李正强专家工作站(202205AF150031)资助。

摘　　要：分离地表贡献是反演气溶胶光学厚度(AOD)的关键步骤。传统针对单角度载荷开发的反演算法通常假设地表为均一朗伯体,忽略了地表的各向异性,从而带来了误差。然而,由于信息量不足考虑地表特性会产生病态反演问题,因此需要引入先验知识刻画地表各向异性参数。通常使用一定时间范围内均值作为地表参数的先验值,但地表情况随时间变化,使用平均值会造成地表贡献估计的不准确。为了评估使用平均值对反演精度的影响,该研究基于2022年的MODIS双向反射分布函数产品(MCD43C2)进行了分析,并结合MODIS地表分类产品(MCD12C1),对全球不同地表覆盖类型的各向同性散射核系数进行统计。结果显示,该参数在以林地为主要覆盖的场景下一般低于0.05,在草地、耕地以及城镇为主要覆盖的场景下有近90%的情况低于0.1。根据统计的经验值,模拟了不同地表类型下表观反射率可能出现的误差,并基于模拟结果反演AOD;经与预设的AOD“真值”对比发现,相对于植被覆盖较少的亮地表,在植被覆盖较高的地表处,以季度平均值作为该位置处的双向反射分布函数(BRDF)形状参数产生误差较少。即当太阳高度角为50°,AOD真值为0.4时,常绿阔叶林区域的平均反演绝对误差最大为0.053;近90%的草地、耕地以及城镇区域平均反演绝对误差最大分别为0.089/0.083/0.113。但在亮地表区域,AOD平均反演绝对误差显著增加,最大值达到0.145。这表明使用季度平均BRDF形状参数作为AOD反演中的地表约束的可靠性随着地表反射率的升高而降低,即使用地表信号占主导的标量卫星观测进行气溶胶的反演,存在不可避免的不确定性。因此,使用多角度偏振观测实现BRDF和AOD的协同反演是未来提高气溶胶反演精度的重要方向。研究结果可以对全球不同地表类型区域的BRDF核系数分布提供先验知识,且进一步探究了基于非Separating the surface contribution is crucial in aerosol optical depth(AOD)inversion.Traditional retrieval algorithms developed for single-angle sensors typically assume that the surface is homogeneous Lambertian,which ignores surface anisotropy and introduces errors.However,considering surface properties can create an ill-posed inversion,so it is necessary to introduce a priori knowledge to characterize the surface anisotropy parameters.Using the mean value over a certain time range as a fixed value for the surface parameter is a common practice.However,surface conditions vary over time,and averaging can lead to inaccuracies in estimating the surface contribution.This paper analyzes the anisotropic scattering kernel coefficients of various surface cover types to evaluate the effect of the mean value on inversion.The analysis is based on the MODIS bidirectional reflectance distribution function product(MCD43C2)for 2022 and the MODIS surface classification product(MCD12Q1).The statistical analysis indicates that when forested land is the dominant cover,the parameter is generally lower than 0.05.In nearly 90%of the cases where grassland,cropland,and towns are the dominant covers,the parameter is lower than 0.1.This paper presents a simulation of the potential errors in apparent reflectance under diverse surface types,intending to invert the AOD based on the simulation findings.A comparison of the results with the preset“true value”of the AOD revealed that the quarterly mean value of the BRDF shape parameter was less prone to producing errors in the location of the quarterly mean value of the BRDF shape parameter than the bright surfaces with less vegetation cover and the more highly covered surfaces.The shape parameter of the bi-directional reflectance distribution function(BRDF)at this location produces fewer errors than brighter surfaces with less vegetation cover.In other words,at a solar zenith angle of 50°and a true value of AOD of 0.4,the mean absolute error of retrievals is 0.053 for evergreen broadlea

关 键 词：气溶胶光学厚度 双向反射分布函数 辐射传输模型 MODIS 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]