【作者】 王磊；

【导师】 田庆久；

【作者基本信息】 南京大学 ， 地理学（地图学与地理信息系统）， 2019， 博士

【摘要】 叶面积指数(LAI)是描述植被冠层几何结构的最基本参数,也是气候模型、地-气相互作用等模型的重要输入参数,作为全球重要植被类型之一的草地,在全球碳循环中具有重要作用和地位,准确开展大尺度的草地LAI遥感反演具有重要意义。然而,草地冠层物种丰富度高、结构复杂,物种个体在叶片和冠层尺度上均存在着较大的差异性,很大程度上影响了多物种构成的草地冠层光谱特征,进而为LAI的遥感反演工作带来较强的不确定性。高光谱成像数据以其丰富的谱段信息,能够获取植被冠层的大量参数信息,为冠层物种识别和丰度的估算提供了可能。利用无人机高光谱构建物种端元光谱库,基于Hyperion卫星高光谱数据开展冠层物种丰度估算,实现顾及草地冠层物种丰度的LAI反演工作,具有重要的应用价值。本文在内蒙古锡林郭勒草原选择典型研究区,在获取Hyperion卫星高光谱数据、无人机高光谱数据数据和地面同步草地观测数据的基础上,通过分析草地冠层物种构成及其丰度的光谱响应特征,提出了基于无人机高光谱数据,依据植被指数分级多次提取草地物种端元光谱的方法,进而面向草地物种丰度估算优选了不同混合像元分解模型,得到了研究区草地物种丰度估算结果,从而实现了顾及草地冠层物种丰度的LAI反演,获得了研究区草地LAI分布信息。论文的主要研究结论如下:(1)分析了草地冠层物种构成及其丰度的光谱响应特征。基于PROSPECT和PROSAIL模型分别从叶片到冠层尺度开展了草地冠层物种构成及其丰度的光谱响应特征分析。主要结论:在叶片尺度上,利用PROSPECT模型,分析了叶片结构、叶绿素、干物质、类胡萝卜素、水分含量等参数对叶片光谱特征的影响及敏感波长范围,进而将各物种的实测参数代入PROSPECT模型,获取了不同物种的模拟光谱,运用欧式距离和光谱角距离两个参数分别从反射率数值和曲线几何形态两个方面,评价了叶片尺度的可分性,表明不同物种叶片的光谱特征总体差异显著,但差异程度不同。在冠层尺度上,利用PROSAIL模型,对各物种在不同LAI下的冠层光谱进行了模拟,结果显示,各物种冠层的光谱特征随着LAI的增加,其区分度总体呈增强趋势,但在不同LAI下的差异性程度不同,当LAI<0.5时,其区分度较差,当0.5<LAI<4之间时,其光谱特征变化幅度显著增强,当LAI>4时,光谱特征的变化幅度有明显的减弱趋势;对相同LAI下的,不同物种的光谱特征分析表明,随着LAI的增大,各物种之间的光谱特征差异性总体呈增强趋势,当LAI<0.5时,其区分度较低,当0.5<LAI<4之间时,其区分度显著增强,当LAI>4时,不同物种的区分度有明显的减弱趋势。(2)提出了草地物种端元提取方法,并构建了端元光谱库。基于草地冠层物种构成及其丰度的光谱响应特征分析的基础上,提出了基于无人机高光谱数据,依据植被指数分级多次提取草地物种端元光谱的方法,并运用不同端元提取算法,构建了候选端元集,进而结合无人机高光谱成像过程中地面标识的物种参考端元光谱,综合欧式距离与光谱角距离优选了端元光谱库。(3)估算了研究区草地冠层物种丰度。分别运用完全约束最小二乘法(FCLS)和多端元线性混合光谱分解法(MESMA)2类丰度估算方法,基于Hyperion卫星高光谱数据对研究区进行了草地冠层物种丰度估算,通过精度评价,确定研究区草地冠层物种丰度的最优估算结果。(4)实现了顾及草地冠层物种丰度的LAI反演。在获取研究区草地冠层物种丰度估算的基础上,基于PROSAIL模型,根据实测参数值确定取值范围和步长,分物种构建查找表,利用光谱角和欧式距离结合作为光谱匹配算法,开展顾及草地冠层物种丰度的LAI反演研究,并开展精度评价与误差分析。结果显示,不同物种构成的反演精度存在一定差异性,羊草+糙隐子草的反演结果精度最高,平均绝对误差为0.31;而羊草+星毛委陵菜的误差相对其它群落类型较高,平均绝对误差为0.73;就整个研究区而言,草地LAI反演结果的总体平均绝对误差为0.43,达到了较高的反演精度。本文的主要创新点包括:(1)论文针对由多物种构成的复杂草地冠层的物种丰度估算问题,提出了基于无人机高光谱数据,依据植被指数分级多次提取草地物种端元光谱的方法,进而面向草地物种丰度估算优选了不同混合像元分解模型,得到了草地物种丰度估算结果,具有一定的技术方法创新性。(2)将草地冠层LAI遥感反演建立在草地冠层物种丰度估算的基础上,进行模型参数优化与约束,以期提高复杂物种构成的草地LAI反演精度的研究思路,具有一定的理论创新性。更多还原

【Abstract】 Leaf area index(LAI),the most basic parameter describing the vegetation canopy geometry,is also an important input parameter for models such as climate model and ground-air interaction.Steppe,one of the most widely distributed types of vegetation in the world,plays an important role in the global carbon cycle.Therefore,accurately carrying out large-scale steppe LAI remote sensing inversion is of great significance.However,the steppe canopy has high species richness and complex structure,and individual species show strong parameter differences at a leaf scale.Meanwhile,there are large differences in growth between species themselves and different species on an individual scale,which has great effects on the complexity of spectral composition at canopy scale,and then leads to strong uncertainty for remote sensing inversion of LAI.Hyperspectral data,with its rich spectral information,can obtain a large amount of parameter information of vegetation canopy,which provides a possibility for canopy species identification and abundance estimation.Utilizing the hyperspectral to construct a species end-spectrum library,based on Hyperion satellite hyperspectral data,the canopy species abundance estimation is carried out to realize the LAI inversion of the canopy species abundance,which has important application value.Based on the hyperspectral data of Hyperion satellite,hyperspectral data of drone and ground-synchronized steppe observation data in Xilin Gol Prairie of Inner Mongolia,a method based on the hyperspectral data of UAV was proposed to extract the endmember spectrum of grassland species based on vegetation index classification via analyzing the spectral response characteristics of steppe canopy species composition and its abundance.Furthermore,for the abundance of steppe species,different mixed pixel decomposition models were estimated,and the results of steppe species abundance estimation in the study area were obtained.Therefore,the LAI inversion of steppe canopy species abundance was realized,and the LAI distribution of steppe in the study area was obtained.The main conclusions of this study were as follows:(1)The spectral response characteristics of steppe canopy species composition and its abundance were analyzed.Based on the PROSPECT and PROSAIL models,the spectral response characteristics of composition and abundance by steppe canopy species were analyzed from leaf to canopy scale.On the leaf scale,the effects of leaf structure,chlorophyll,dry matter,carotenoids,moisture content and other parameters on the spectral characteristics of the leaves and the sensitive wavelength range were analyzed by using the PROSPECT model.Furthermore,the simulated spectra of different species were obtained by substituting the measured parameters of each species into the PROSPECT mode.Using the two parameters of Euclidean distance and spectral angular distance,the separability of the blade scale was evaluated from the two aspects of reflectance value and curve geometry,which indicated that the spectral characteristics of leaves by different species were significantly different though the difference degree was different.On the canopy scale,the canopy spectra of different species under different LAIs were modeled using the PROSAIL model.The results showed that the difference in the spectral characteristics of different spectral features increased with increasing of LAI,however,the degree of difference was different under different LAIs.The separability of LAI was poor in 0.5,the spectral characteristic variation amplitude was significantly enhanced in 0.5-4,and the variation range of spectral characteristics was obviously weakened in>4.The analysis of spectral characteristics by different species under the same LAI showed that with the increase of LAI,the spectral characteristics of each species generally increased.The discrimination degree was low in<0.5,the discrimination degree was significantly enhanced in 0.5-4,and the discrimination degree of different species had obvious weakening trend in>4.(2)An endmember extraction method for steppe species was proposed and an endmember spectral library was constructed.Based on the analysis of spectral response characteristics of steppe canopy species composition and its abundance,a method based on UAV hyperspectral data was proposed to extract the endmember spectral of grassland species based on vegetation index classification.Meanwhile,by using different endmember extraction algorithms,a set of candidate endmembers was constructed,and the endmember spectral library was optimized by combining the reference endmember spectrum of the terrestrial marker in the hyperspectral imaging process,and the integrated Euclidean distance and spectral angular distance.(3)The abundance of canopy species in the study area was estimated.Based on Hyperion satellite hyperspectral data,the full-constrained least squares(FCLS)and multi-terminal linear mixed spectral decomposition(MESMA)abundance estimation methods were used to estimate the canopy species abundance.In addition,the optimal estimation results of steppe canopy species abundance in the study area were determined via the accuracy evaluation.(4)LAI inversion taking into account the abundance of steppe canopy species.Based on the estimation of the abundance of steppe canopy species in the study area and the PROSAIL model,the range and step size were determined via the measured parameter values.Using the combination of spectral angle and Euclidean distance as the spectral matching algorithm,the LAI inversion study considering the abundance of grass canopy species and the accuracy evaluation and error analysis were carried out.There were some differences in the inversion precision of different species composition.The inversion result of Leymus chinensis+Cleistogenes squarrosa had the highest precision,and the average absolute error was 0.31.The error of L.chinensis+Potentilla chinensis was higher than other community types,and the average absolute error was 0.73.However,for the entire study area,the overall average absolute error of the steppe LAI inversion results was 0.43,achieving a high inversion accuracy.The main innovations of this paper include:(1)For the estimation of species abundance of complex turf canopy composed of multiple species,this paper proposes the method to extract the endemic spectrum of steppe species based on UAV hyperspectral data and vegetation index grading.Furthermore,different mixed pixel decomposition models were proposed for the estimation of steppe species abundance,and the estimation results of steppe species abundance were obtained,which had certain technical methods and innovation.(2)Based on the estimation of steppe canopy species abundance,the LAI remote sensing inversion of steppe canopy was carried out to optimize and constrain the model parameters,so as to improve the research idea of steppe LAI inversion precision composed of complex species,which had theoretical innovation.更多还原