【作者】 乔璐；

【导师】 陈立新； 范文义； 王秀峰；

【作者基本信息】 东北林业大学 ， 土壤学， 2010， 硕士

【摘要】 有机质为土壤营养的重要来源,不仅为植物提供所需的各种营养元素,同时对土壤结构的形成、改善土壤物理性质、提高土壤保肥能力和缓冲性能有决定性作用。如何快速、准确地获得土壤信息是目前各国学者关注的焦点。高光谱遥感技术以其光谱分辨率高、波段连续性强等特点,能够及时、准确、动态地监测分析作物的健康状况和影响作物产量的土壤环境因素,这对农业生产、土壤质量监测、生态环境的维护与治理具有现实意义。通过GPS定位、野外采样分析和室内高光谱测定的方法,研究了土壤有机质和游离态氧化铁含量、土壤成分的高光谱曲线特征,以及利用matlab7.1对土壤有机质,氧化铁与土壤反射率的数学变换形式进行相关性分析。以光谱反射率(反射率倒数,反射率对数,反射率一阶微分等)作为自变量,有机质和游离态氧化铁作为因变量,运用多元线性回归、BP神经网络方法建立数学模型。研究结果显示：(1)不同土地利用类型(农用地、林用地、城市绿地、松花江沿岸流域、湿地)土壤有机质含量和游离态氧化铁含量差异较大,它们的变化幅度分别为12.24-274.24g／kg和0.65-10.59mg／kg。它们的平均含量排序分别为农用地>林用地>湿地>松花江沿岸流域>城市绿地和松花江沿岸流域>农用地>城市绿地>林地>湿地。(2)土壤颜色、机械组成,以及有机质含量、氧化铁含量的不同,影响土壤光谱曲线的差异。土壤随着黏粒、有机质含量的增加,土壤反射率将低。反之随着砂粒增加,颜色变浅,以及有机质含量的减少土壤反射率相对增加。(3)运用反射率对数,反射率倒数的对数,反射率对数的一阶微分,反射率倒数一阶微分等4种方法,分析了土壤光谱反射率与土壤有机质含量的关系,以及土壤光谱反射率与土壤游离态氧化铁含量的关系。发现对有机质含量做对数变换可提高与土壤反射率的相关性。采用多元线性回归统计法和BP神经网络法分别建立了预测土壤有机质含量和土壤游离态氧化铁含量的多元回归模型、BP神经网络模型。由BP神经网络法建立的土壤有机质含量模型预测结果较好,由反射率平方根的一阶微分建立的游离态氧化铁含量的BP神经网络模型的精度高于多元线性回归模型。(4)通过土壤光谱和土壤特性参数建立的反演模型,对TM遥感图像进行了土壤有机质和游离态氧化铁的含量的专题制图,建立较为精细的土壤参数空间分布图。更多还原

【Abstract】 Organic matter is an important source of soil nutrition. It can not only provide almost all necessary nutrient elements with plants, but also takes determinant roles in the formation of soil structure, and the improvement of soil physical properties and increased capacity of soil fertility conservation and buffering properties. Quick and accurate acquirement of soil information is widely paid more attention to by researchers. Hyperspectral remote sensing technique has special advantages of high spectral resolution and strong band continuity and so on. It can monitor and analyze crops vigor and soil environmental factors that affect crops production in timely, accurate and dynamical way. These possess practical significance for agricultural product and soil quality monitoring, eco-environment management and maintenance.The content of soil organic matter and free ferric oxide, the characteristics of hyperspectral curve of soil composition were studied by methods of GPS positioning, field sampling and analysis, hyperspectral measurement. Correlation analysis between soil organic matter, ferric oxide and soil reflectance was made in their mathematical transformation forms by Matlab 7.1. Mathematical models were built among spectral reflectance as independent variables and organic matter and ferric oxide as the dependent variable by multivariate linear regression and BP neural network. Research results showed:(1) For the content of soil organic matter and free ferric oxide, there were greater differences between different types of land use (farmland, forest land, urban green land, Song Hua Jiang River, wetland). They ranged from 12.24 to 274.24 g/kg, and 0.65-10.59 mg/kg, respectively. Their average value was farmland>forest land> wetland>Song Hua Jiang River>urban green land, Song Hua Jiang River>farmland>urban green land>forest land>wetland, respectively.(2) The differences in soil color, soil mechanical composition, the content of soil organic matter and ferric oxide influenced soil spectra curve. Soil reflectance decreased with increased content of soil clay and soil organic matter. Slight loss of soil color, increased content of sand particles and decreased content of soil organic matter resulted in increased soil reflectance, comparatively.(3) Using reflectance logarithms, the countdown reflectance logarithms, reflectance of one-order differential, the relationship between soil spectral reflectance and the content of soil organic matter, and between soil spectral reflectance and the content of free ferric oxide were analyzed. It was found that logarithm transformation of soil organic matter content could increase their correlation. The models of multivariate linear regression and BP neural network were established in order to predict their content of soil organic matter and free ferric oxide. The prediction result from BP neural network model was better. The prediction accuracy for the content of free ferric oxide from BP neural network model was higher than that from multivariate linear regression model.(4) Spatial distribution maps in detail were drawn according to inversion models established between soil spectrum and characteristic parameters, and the mapping for soil organic matter and free ferric oxide by TM remote sensing image.更多还原