基于表层土壤光谱的耕层土壤有机质间接估测

为解决遥感技术在监测耕层土壤有机质方面的应用问题,利用表层土壤光谱对耕层土壤有机质含量进行估测。以山东省济南市章丘区的表层、耕层各76个土壤样本为研究对象,首先对表层光谱数据进行小波变换去噪、剔除异常样本等处理,然后对处理后的光谱反射率进行一阶微分等10种数学变换,在对数倒数一阶微分和对数一阶微分变换后的反射率数据中选取43个与土壤有机质含量相关系数较高的波段,通过主成分分析以累计贡献率大于90%的标准选取5个主成分作为反演因子,利用BP神经网络(BPNN)、支持向量机回归(SVR)和多元线性回归(MLR)方法建立耕层土壤有机质含量间接估测模型。结果表明,耕层土壤与表层土壤有机质含量之间决定系数R~2达到0.839,显著性P0.01,存在着较强的相关性BPN估测模型的精度最优,决定系数R~2为0.845,平均相对误差为7.642%,RMSE分别为1.622g·kg~(-1)。研究表明,利用表层土壤光谱信息间接估测耕层有机质含量是可行有效的,为耕层土壤有机质的估测问题提供了一种新思路。

Indirect estimation of organic matter content in plough layer based on topsoil spectrum

In order to solve the application of remote sensing technology in monitoring the organic matter of plough layer, the topsoil spectrum was used to estimate the organic matter content of plough layer. Taking 76 soil samples from the surface layer and 76 soil samples from the plough layer collected in Zhangqiu District of Jinan city of Shandong Province as the research object. First, the surface spectral data was processed using wavelet noise reduction and eliminate abnormal samples. Then, 10 kinds of mathematical transformations such as first-order differential were performed on the processed spectral reflectance. In the data after the first-order differential and log-first differential transformation of the logarithm, 43 bands with high correlation coefficient with soil organic matter content were selected, and we used the principal component analysis to obtain 5 principal components as inversion factors that their cumulative contribution rate is greater than 90%. Finally, using topsoil spectral information, an indirect estimation model for the organic matter content of cultivated soil was established with BP neural network, SVR and MLR. The results showed that the R² between the plough layer soil and the surface soil organic matter content reached 0.839, and the significance P<0.01. There is a strong correlation between the plough layer soil and the surface soil organic matter content. The estimation accuracy of the BP neural network was higher. R², MRE and RMSE are 0.845, 7.642 % and 1.622g·kg^-1, respectively. The researchers showed that it was feasible and effective to estimate the organic matter content in plough layer indirectly using hyperspectral information of topsoil. It provides a new solution to the problem that estimating the organic matter content of plough layer.