火山碎屑物发育土壤有机质含量的高光谱预测模型研究

发布时间：2018-06-10 19:16

  本文选题：玄武岩 + 粗面岩　； 参考：《沈阳农业大学》2017年硕士论文

【摘要】：土壤有机质是土壤中含碳的有机化合物,是土壤的化学性质之一。目前,传统的土壤有机质含量测定方法主要有重铬酸钾-硫酸消化法、TOC分析仪法、元素分析仪法。虽然化学分析法测定精度较高,但室内化学分析费时费力。近来年,随着高光谱遥感技术的不断发展和数学多变量统计算法的深入研究,高光谱遥感以其光谱分辨率高、信息丰富等优势在土壤属性预测方面得到了较快的发展。通过室内光谱仪获取土壤样品的反射光谱数据省时省力,通过建立的模型进行估测,可以实现土壤有机质含量的快速测定。本研究采集东北地区分布的玄武岩质和粗面岩质火山碎屑物发育的土壤,对土壤有机质含量进行化学分析,并利用ASD FieldSpec4光谱仪获取土壤样品的高光谱数据。通过对原始光谱反射率进行平滑、去噪处理后结合连续统去除法、微分法和倒数对数法提取土壤光谱特征,采用多元逐步线性回归、偏最小二乘回归、主成分回归法建立两种岩性火山碎屑物发育土壤的有机质含量预测模型,并依据校正相关系数、预测相关系数、校正标准误差、预测标准误差、交叉验证均方差、预测均方差、预测偏倚差对建立的预测模型进行比较,分别获取两种岩性火山碎屑物发育土壤的有机质含量最优预测模型。玄武岩质火山碎屑物发育土壤的光谱曲线在400～1300nm范围内呈急剧增加趋势,在1440～1860 nm范围内缓慢增加;粗面岩质火山碎屑物发育土壤的光谱曲线在400～750 nm范围内呈急剧增加趋势,在750～1380 nm范围内增长缓慢,在1420～1880 nm范围内趋于平缓。玄武岩质火山碎屑物发育土壤的有机质光谱响应位于500nm、800 nm;粗面岩质火山碎屑物发育土壤的光谱响应位于405 nm、465 nm、575 nm、1105 nm。原始光谱进行一阶微分、二阶微分、反射率倒数对数、反射率倒数对数的一阶微分和反射率倒数对数的二阶微分处理后,分别与相应的玄武岩质和粗面岩质火山碎屑物发育土壤的有机质含量进行相关性分析,相关性均表现为显著增强。玄武岩质火山碎屑物发育土壤,除反射率倒数对数二阶微分外,最大相关系数的绝对值均在0.8以上,一阶微分的最大相关系数为-0.8898;粗面岩质火山碎屑物发育的土壤,一阶微分、反射率倒数对数的一阶微分和反射率倒数对数的二阶微分的最大相关系的绝对值亦在0.8以上,反射率倒数对数的一阶微分的最大相关系数为-0.9029。全谱范围内,采用多元逐步线性回归、偏最小二乘、主成分回归三种方法建立的两种岩性火山碎屑物发育土壤有机质含量的预测模型,均得到了较好的预测结果。其中,玄武岩质火山碎屑物发育土壤的有机质含量最优预测模型为基于光谱反射率倒数对数的一阶微分建立的多元逐步线性回归模型。模型自变量数为7,预测决定系数Rv2=0.9720,预测均方差RMSEP=2.0590,sig=-0.0030.01。粗面岩质火山碎屑物发育土壤的有机质含量最优预测模型为基于光谱反射率倒数对数的一阶微分建立的偏最小二乘回归模型。模型自变量数Pc = 5,建模相关系数Rc = 0.9872,决定系数Rc2= 0.9745,建模均方根误差RMSEC = 0.4821,校正偏差SEC = 0.4906,预测决定系数Rv2= 0.9702,预测均方根误差RMSEP = 0.9563,校正偏差SEP = 0.9711,预测偏倚差Bias=-0.0637。
[Abstract]:Soil organic matter is an organic compound containing carbon in soil and one of the chemical properties of soil. At present, the main methods of determining the content of soil organic matter are potassium dichromate - sulfuric acid digestion, TOC analyzer and elemental analyzer. Although the precision of chemical analysis is high, the indoor chemical analysis is time-consuming and laborious. The continuous development of spectral remote sensing technology and the in-depth study of mathematical multivariable statistical algorithms. Hyperspectral remote sensing has developed rapidly in the prediction of soil properties with its high spectral resolution, rich information and so on. The number of reflectance spectra obtained by the indoor spectrometer is saved and estimated by the established model. A rapid determination of soil organic matter content can be achieved. In this study, the soils developed from basalt and coarse rock volcanic debris distributed in Northeast China were collected, and the content of soil organic matter was analyzed by chemical analysis. The hyperspectral data of soil samples were obtained by using ASD FieldSpec4 spectrometer. After de-noising, the spectral characteristics of soil are extracted with continuous division method, differential method and reciprocal logarithm method. Multiple stepwise linear regression, partial least squares regression and principal component regression method are used to predict the organic matter content of two types of lithologic volcaniclastic soil, and the correlation coefficients are corrected and the correlation coefficients are predicted and the calibration criteria are corrected. The error, the prediction standard error, the cross validation mean square deviation, the prediction mean square variance, the prediction bias difference and the comparison of the predicted models were made to obtain the optimal prediction model of the organic matter content of the two kinds of lithologic volcaniclastic soils. The spectral curves of the basalt volcaniclastic soil increased sharply in the range of 400 to 1300nm. The trend is increasing slowly within the range of 1440~1860 nm; the spectral curve of the developed soil with coarse rock volcanic debris increases rapidly within the range of 400~750 nm, and grows slowly within the range of 750~1380 nm, and tends to be gentle in the range of 1420~1880 nm. The spectral response of the organic matter of the basalt volcaniclastic soil is located in 500nm, 8 00 nm; the spectral response of the coarse rock volcaniclastic soil is located at 405 nm, 465 nm, 575 nm, and 1105 nm. for the first order differential, the two order differential, the reflectivity reciprocal logarithm, the first differential of the reflectivity reciprocal logarithm and the two order differential of the reflectivity inversion logarithm, respectively, and the corresponding basalt and rough rock volcanoes. The correlation between the organic matter content of the detrital soil and the correlation analysis shows that the correlation of the basalt volcanic debris developed soil, except the two order differential of the reflectivity inversion number, the absolute value of the maximum correlation coefficient is above 0.8, the maximum phase relation of the first order differential is -0.8898, and the development of the coarse rock volcanic debris is developed. The absolute value of the first order differential of the logarithm of the logarithm of the reflectivity reciprocal logarithm and the two order differential of the reflectivity reciprocal logarithm is also more than 0.8. The maximum correlation coefficient of the first order differential of the reflectivity reciprocal logarithm is within the -0.9029. total spectrum range, and the multiple stepwise linear regression, partial least squares, and principal component regression are used for three kinds of squares. The prediction model of the soil organic matter content of two types of lithologic volcanics developed by the method has been well predicted. Among them, the optimal prediction model of the organic matter content in the developing soil of basalt pyroclastic is a stepwise linear regression model based on the first order differential of the reciprocal logarithm of spectral reflectance. The number of variables is 7, the prediction coefficient Rv2=0.9720, the predicted mean square variance RMSEP=2.0590, the optimal prediction model for the organic matter content in the developing soil of sig=-0.0030.01. rough rock volcaniclastic is based on the partial least squares regression model based on the first order differential of the inverse logarithm of spectral reflectance. The number of model independent variables is Pc = 5, and the correlation coefficient Rc = 0. of the modeling is Rc = 0.. 9872, the determination coefficient Rc2= 0.9745, the square root mean square error RMSEC = 0.4821, the correction deviation SEC = 0.4906, the prediction coefficient Rv2= 0.9702, the mean square root error RMSEP = 0.9563, the correction deviation SEP = 0.9711, the prediction bias difference Bias=-0.0637.
【学位授予单位】：沈阳农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S153.6
，

本文编号：2004305