推荐内容
【2013博】卫星高光谱影像地表反射率一体化反演技术研究
【作者】 胡顺石 【导师】 童庆禧;王晋年;张立福 【 学位年度 】 2013 【论文级别】 博士 【关键词】 高光谱 大气校正 地表反射率 查找表 气溶胶光学厚度 水汽叶绿素荧光 6SV MODTRAN 【Key words】 Hyperspectral atmospheric correction surface reflectance look up table aerosol optical thickness water vapor fluorescenece 6SV MODTRAN 【中文摘要】 高光谱遥感影像容易受到大气产生的散射和吸收作用,从而使得高光谱传感器记录的光谱信息扭曲,不能反映地物的真实光谱。因此,必须去除大气效应以获得反映地物真实本质的反射率光谱。根据太阳辐射在大气中的传输原理和高光谱遥感影像的特点,本文对2001~2010年间7景EO-1 Hyperion高光谱影像进行了大气校正获得相应地表反射率立方体影像,主要完成了以下工作: 1. 对影响高光谱遥感影像大气校正的因素进行了敏感性分析,根据这些影响因素建立了通用化的多维大气参数查找表,对多维查找与插值计算进行了说明,多维大气参数查找表具有较为广泛的适用性; 2. 根据EO-1 Hyperion高光谱遥感影像特点,建立了大气校正前数据预处理技术流程,经过数据预处理后,高光谱遥感影像质量明显改善,减少了由于数据质量引入的误差;本文的数据预处理技术具有一定的借鉴意义,为高光谱遥感影像大气校正前数据预处理方法提供了相应的参考; 3. 利用改进暗目标法进行了大气气溶胶光学厚度反演,并对影响气溶胶光学厚度反演结果的大气模式、地表高程、水汽含量等因素进行分析,气溶胶光学厚度反演结果与AERONET网站观测结果/MOD04产品具有较高的线性相关性,2R为0.79,均方根误差为0.03074; 4. 采用940 nm水汽吸收带附近的多个水汽吸收通道和非水汽吸收通道反演大气水汽含量,水汽含量反演结果与MOD05水汽产品具有较高的线性关系,2R为0.973,均方根误差RMSE为0.0824 g/cm2;大气水汽含量反演结果受地表高程、水汽吸收通道中心波长偏移影响较大,受气溶胶光学厚度影响较小; 5. 根据气溶胶光学厚度与水汽含量反演结果建立了地表反射率一体化反演模型,地表反射率光谱曲线形状与地物的理论光谱形状较为相似。地物反射率光谱与FLAASH 反演结果相比具有较高的一致性,在相同输入参数的条件下,与FLAASH反演结果进行比较,地物反射率光谱的平均相关系数高于0.9,平均光谱卫星高光谱影像地表反射率一体化反演技术研究II角小于0.08弧度,平均均方根误差小于0.02; 6. 本文利用EO-1 Hyperion高光谱遥感影像所提取的太阳诱导叶绿素荧光强度在1~8 W/(m2∙sr∙μm)之间,这也与多数学者的研究接近;NDVI植被指数可用于指示地球表面植被的覆盖情况,可区分不同区域的植被覆盖率,但它与植被的光合作用和植物的生化参数关系较少;由于受到叶黄素脱环氧化的影响,植物冠层反射率在531 nm降低,而非植被区域的地表反射率不会受此影响,从而致使植被区域的PRI光化学指数值要比非植被区域的值偏低;2004-10-11、2004-08-31和2010-06-20三景影像中所提取的植被区域的荧光强度值依次升高;另外,这三景影像中高尔夫球场的草地所释放的荧光强度要高于其它植被区域,这或许与其植被的种类和冠层结构有较大的关系。
【Abstract】 Hyperspectral remote sensing image is susceptible to atmospheric scattering and absorption effect, which causes distortion in the spectral information of the surface recorded by hyperspectral sensor, leading to the fake spectral shape for the surface. The atmospheric effect must be removed in order to get the true surface reflectance. In this study, by following the principle of solar radiation transfer in the atmosphere and the characteristics of hyperspectral remote sensing images, the atmospheric effect within 7 EO-1 Hyperion images between 2001 and 2010 was corrected and the corresponding surface reflectance cube images were obtained. The main contents and results are summarized as follows: 1. A sensitivity analysis for hyperspectral remote sensing atmospheric correction was implemented and a general purpose multi-dimensional look up table was built up according to the sensitivity analysis results. As the look up table can be widely used in atmospheric correction, so the usage and interpolation for multi-dimensional look up table is described in this thesis. 2. Based on the features of EO-1 Hyperion image, a preprocess workflow before atmospheric correction was built up. The image quality was significantly improved after preprocessing and the retrieval error due to the quality of the data was reduced. The preprocess workflow in this thesis is useful as a reference for other hyperspectral remote sensing images. 3. The improved Dark Object method was applied to retrieve aerosol optical thickness and the affected factors, such as aerosol mode, surface elevation and water vapor content, were analyzed. These factors may introduce errors to aerosol optical thickness result. The aerosol optical thickness retrieval result showed good linear relationship with AERONET observed data/MOD04 aerosol optical product with R 2 equaled to 0.79, while root mean square error was found 0.03074. 4. The water vapor absorption and non-absorption channels near 940 nm water absorption region were chosen for water vapor retrieval. The water vapor retrieval result has a good linear relationship with MOD05 water vapor product. The R 2 is 0.973 and RMSE is 0.0824 g/cm 2 . The surface elevation and central band shift in water vapor absorption band has a greater impact on water vapor retrieval result than aerosol optical thickness. 5. An integrated surface reflectance retrieval model was built up under the support of aerosol optical thickness and water vapor retrieval result. The spectral shape of the retrieved surface reflectance is very similar with the theoretical one. The retrieved surface reflectance is comparable with the output of FLAASH software. Comparing to the FLAASH output, the retrieved result has an average correlation coefficient higher than 0.9, average spectral angle less than 0.08 radians and the root mean square error less than 0.02. Both the FLAASH output and retrieved result are run by the same input parameters.
6. The solar induced chlorophyll fluorescence intensity is extracted form EO-1 Hyperion remotely sensed images with the value of 1~8 W/(m 2 ∙sr∙μm), which is also close to the values calculated by other scholars. The NDVI index can be used to indicate and distinguish the Earth’s surface vegetation coverage but it has less relationship with biochemical parameters between vegetation photosynthesis and plant canopy reflectance. Due to the impact of lutein off epoxidation, the surface reflectance values of vegetate area are decreased while the non-vegetation area reflectivity will not be affected by this cause, resulting PRI photochemical index value of the vegetation area is lower than non-vegetation area. The fluorescence intensity of the vegetation area is increased successively in the following images, 2004-10-11, 2004-08-31 and 2010-06-20. The released fluorescence intensity from the golf course grass was found higher than other vegetation area, which might be
(责任编辑:黄长平) |