基于高光谱数据的土壤有机质含量反演模型比较

发布时间：2018-10-25 12:40

【摘要】：以土壤多样化的陕西省横山县为研究区域,比较了3种基于高光谱数据的土壤有机质含量反演模型,在实验室利用ASD Field Spec FR地物光谱仪对横山县野外采集的土壤样品进行光谱测定,并通过重铬酸钾氧化容量法测定土壤有机质含量。然后对原始光谱反射率的倒数进行微分运算获得其一阶导数光谱,将原始光谱反射率、一阶导数光谱分别与土壤有机质含量进行相关性分析,得到相关性系数r较高的特征波段的一阶导数光谱,直接建立基于一阶导数光谱的多元线性逐步回归分析(MLSR)模型。同时针对这些相关性系数较高的特征波段的一阶导数光谱进行主成分分析(Principal component analysis,PCA),利用主成分分析得到的结果分别建立BP神经网络反演模型(PCA-BP)和多元线性逐步回归分析模型(PCA-MLSR)。用上述3种方法进行土壤有机质含量反演,并对3种反演结果进行精度验证与比较。实验分析结果表明:在3种模型中,基于主成分分析结果构建的PCA-BP模型在土壤有机质含量反演中决定系数(R2)最高,为0.893 0,均方根误差(RMSE)为0.118 5%;其次为运用全部主成分PCA分析结果构建的多元线性逐步回归模型,R2为0.740 7,RMSE为0.161 3%;而采用一阶导数光谱反射率构建的多元线性逐步回归模型中,最佳反演模型R2仅为0.689 9,RMSE为0.171 0%。由此说明,PCA-BP模型有机质含量反演精度明显高于多元线性逐步回归模型,利用全部主成分进行多元逐步回归,其有机质含量反演精度优于仅用累计方差贡献率大于90%的主成分进行多元逐步回归的精度,可以更好地反演土壤有机质的含量。
[Abstract]:Taking Hengshan County of Shaanxi Province as the research area, three kinds of inversion models of soil organic matter content based on hyperspectral data were compared. The soil samples collected in Hengshan County were measured by ASD Field Spec FR ground object spectrometer and the content of soil organic matter was determined by potassium dichromate oxidation volumetric method. The first derivative spectrum is obtained by differential operation of reciprocal reflectance of original spectrum. The correlation between original spectral reflectance and soil organic matter content is analyzed respectively. The first derivative spectrum of the characteristic band with higher correlation coefficient r is obtained, and the multivariate linear stepwise regression (MLSR) model based on the first order derivative spectrum is established directly. At the same time, the first derivative spectra of these characteristic bands with high correlation coefficient are analyzed by principal component analysis (Principal component analysis,PCA). The BP neural network inversion model (PCA-BP) and the multivariate linear stepwise regression model (PCA-MLSR) are established by using the results of principal component analysis (PCA). The above three methods were used to invert the soil organic matter content, and the accuracy of the three inversion results was verified and compared. The experimental results show that, among the three models, the PCA-BP model based on principal component analysis (PCA) has the highest determining coefficient (R2) in soil organic matter content inversion. The root mean square error (RMS) is 0.893, the root mean square error (RMSE) is 0.118 5, the multivariate linear stepwise regression model based on the results of all principal component PCA analysis (R2 = 0.740 7), and the multivariate linear stepwise regression model based on the first derivative spectral reflectivity. The best inversion model R2 is only 0.689 9 and 0.171 0. It shows that the inversion accuracy of organic matter content in PCA-BP model is obviously higher than that in multivariate linear stepwise regression model, and multivariate stepwise regression is carried out by using all principal components. The inversion accuracy of organic matter content is better than that of multivariate stepwise regression with only the principal components whose cumulative variance contribution rate is more than 90%, and the content of soil organic matter can be retrieved better.
【作者单位】：同济大学测绘与地理信息学院;山东农业大学信息科学与工程学院;
【基金】：上海市科学技术委员会科研计划项目(13231203602)
【分类号】：S153.621

【相似文献】