土壤二向反射特性研究与标准光谱库的应用

发布时间：2018-06-14 21:23

本文选题：土壤二向反射 + Hapke模型　；参考：《华中农业大学》2015年硕士论文

【摘要】：自然状态下,凹凸起伏的地表高度变化会造成高光谱图像的像元尺度倾角不一,而二向反射研究能够将图像上的像元订正到相同的观测角度,从而提高定量反演的精度。同时,土壤作为植被的下垫面,研究土壤二向反射特性能够为研究植被冠层光谱提供背景参考;土壤反射率的方向性分布还潜在携带有土壤湿度、有机质含量、矿物含量等属性信息。因此,土壤的二向反射特性研究对土壤和植被定量遥感有着重要的理论意义和研究价值。目前,可见光-近红外光谱分析技术已成功应用于土壤关键属性预测,局部区域的土壤关键属性高光谱反演模型已经非常成熟;但由于各区域气候条件、成土母质的差异,局部模型难以适用于其他区域的土壤样本。随着大尺度土壤标准光谱库的出现,充分利用和挖掘大样本土壤光谱库中的有效信息,建立基于土壤光谱库的土壤关键属性高光谱预测模型,为解决以上问题提供了可能。本研究采集了三种典型土壤的二向反射率数据,分析了他们的二向反射特性和差异;进一步利用Hapke反演了平均单次散射反照率、粗糙度等土壤参数,分析了各参数与土粒组成的关系,在此基础上模拟了二向反射率分布。另外,本文针对全球土壤标准光谱库,采用模糊C均值聚类结合偏最小二乘回归的方法,提取了与研究区供试样本相似的光谱子样本集,进一步利用子样本集建立了土壤关键属性高光谱预测模型,并进行了模型不确定性分析。上述研究得出结论如下:1.三种典型土壤的二向反射率随观测角度的变化规律一致,反射率均随着观测天顶角的增加而增大,在前向散射方向达到最小,后向散射方向达到最大。原因是随观测角度的变化,土壤颗粒之间形成的阴影所占的比例会发生变化,导致探测器接收的光照组分有差异。2.Hapke模型各个参数对初始值的敏感性有差异。土粒组成相似的土壤单次散射反照率曲线形状相似。随着土壤粗颗粒(0.9mm以上)含量的增加,粗糙度参数增大,平均单次散射反照率反而减小。另外,Hapke模型能够很好地进行土壤二向反射率的模拟,但三种典型土壤之间的模拟精度有差别。3.通过模糊C均值聚类与偏最小二乘回归(PLSR)相结合的方法,可挖掘土壤标准光谱库中与研究区供试样本相似的有效光谱信息,建立的有机碳含量估算模型可用于研究区供试样本有机碳含量的粗略估算。本研究中,有机碳含量高光谱估算模型的预测能力主要与土壤样本的剖面层次有关,模型对下层样本的预测能力更好。
[Abstract]:In the natural state, the variation of the surface height caused by the ups and downs of the concave and convex surface will result in different pixel dips of hyperspectral images, while the bidirectional reflection study can correct the pixels on the image to the same observation angle, thus improving the accuracy of quantitative inversion. At the same time, as the underlying surface of vegetation, the study of soil bidirectional reflectance can provide background reference for the study of vegetation canopy spectrum, and the directional distribution of soil reflectance also potentially carries soil moisture and organic matter content. Mineral content and other attribute information. Therefore, the study of the bidirectional reflectance of soil has important theoretical significance and research value for quantitative remote sensing of soil and vegetation. At present, visible light near infrared spectroscopy (VNIR) has been successfully applied to the prediction of soil key attributes. The hyperspectral inversion model of soil key attributes in local areas is very mature, but due to the climate conditions in different regions, the soil-forming parent material is different. Local models are difficult to apply to soil samples in other regions. With the emergence of large scale soil standard spectral database, the effective information of large sample soil spectral database is fully utilized and the hyperspectral prediction model of soil key attributes based on soil spectral database is established, which provides the possibility to solve the above problems. In this study, the bidirectional reflectivity data of three typical soils were collected, and their bidirectional reflectivity characteristics and differences were analyzed, and soil parameters such as average single scattering albedo, roughness and other soil parameters were further retrieved by Hapke. Based on the analysis of the relationship between the parameters and the composition of soil particles, the bidirectional reflectivity distribution is simulated. In addition, aiming at the global soil standard spectral database, the method of fuzzy C-means clustering combined with partial least square regression is used to extract the spectral subsample set similar to the sample in the study area. Furthermore, the hyperspectral prediction model of soil key attributes was established by using the subsample set, and the uncertainty of the model was analyzed. The findings of the study are as follows: 1: 1. The bidirectional reflectivity of the three typical soils is consistent with the observed angle. The reflectivity increases with the increase of the zenith angle and reaches the minimum in the direction of forward scattering and the maximum in the direction of backscattering. The reason is that with the change of observation angle, the proportion of shadow formed between soil particles will change, which leads to the difference of light components received by detector. 2. The sensitivity of each parameter of Hapke model to initial value is different. The shape of soil single scattering albedo curve with similar soil particle composition is similar. With the increase of soil coarse particle content above 0.9mm), the roughness parameter increased and the average single scattering albedo decreased. In addition, the Hapke model can well simulate the bidirectional reflectivity of soil, but the simulation accuracy of the three typical soils is different. By combining fuzzy C-means clustering with partial least square regression (PLSRs), the effective spectral information similar to the sample in the soil standard spectrum library can be mined. The organic carbon content estimation model can be used to estimate the organic carbon content of the sample in the study area. In this study, the prediction ability of the hyperspectral estimation model of organic carbon content is mainly related to the profile level of soil samples, and the prediction ability of the model to the lower samples is better.
【学位授予单位】：华中农业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S127;S151.9

【参考文献】