粮食主产区耕地土壤重金属高光谱综合反演模型

Hybrid Inversion Model of Heavy Metals with Hyperspectral Reflectance in Cultivated Soils of Main Grain Producing Areas

粮食主产区的耕地土壤重金属污染已成为影响粮食安全和人居环境安全的突出问题。高光谱遥感技术为快速高效获取土壤重金属含量提供了新的途径,也为土壤总金属污染的监测和防治提供了技术保障。以河南省粮食主产区新郑市为研究对象,共采集191个耕地土壤样品,利用Rank-KS法划分为144个建模集样本和47个验证集样本;在室内利用ASD Field Spec 3型地物光谱仪获取土壤高光谱数据,对耕地土壤样品在400~2 400 nm的光谱反射率与Cr、Cd、Zn、Cu、Pb 5种重金属元素进行相关性分析,筛选出5种重金属均通过P=0.01显著性检验的共用高光谱特征波段作为反演模型的自变量;采用基于OLS的固定影响变系数面板数据模型,对新郑市144个建模集样本的5种土壤重金属面板数据构建高光谱综合反演模型。结果表明:面板数据模型整体显著,拟合优度较高(R^2=0.993 7,F统计量为1 365.94)。模型精度检验Cu的相对分析误差为2.046,Pb的相对分析误差为3.432,都具有较好的预测精度;Cr、Cd、Zn的相对分析误差在1.4~1.8之间,具有一般的定量预测能力。面板数据模型通过一次建模综合反演多种土壤重金属,计算简便、速度快,可以用于新郑市耕地土壤重金属的高光谱快速监测。

The heavy metals pollution in cultivated soils of main grain producing areas has become a prominent problem affecting the safety of food and living environment. The hyperspectral remote sensing technology as the frontier technology in the field of remote sensing technology, provides a new approach to access to soil heavy metal data quickly and accurately, and also provides the technical support for monitoring and predicting. Taking Xinzheng City of main grain producing areas in Henan Province as the research object, the 191 cultivated soil samples collected were divided into 144 calibration set and 47 validation set by Rank - KS method. The hyperspectral reflectance of soil samples was measured by using ASD FieldSpec 3 spectrometer in laboratory experiments. The correlation analyses between row spectral reflectance in 400 - 2 400 nm and the content of heavy metals Cr, Cd, Zn, Cu, Pb were done, and the correlation coefficient by F significance test （P = 0.01 ） was selected which could be used to extract sensitive hyperspectral feature wavebands reflectance common to the above heavy metals as the independent variables of model. The hyperspectral inversion model was built by panel data model of fixed effect variable coefficient based on the ordinary least squares estimation method （OLS） , which was about the panel data of heavy metals Cr, Cd, Zn, Cu, Pb of 144 samples in Xinzheng. The results show that the panel data model is overall significant, with high goodness of fit （R^2 = 0. 993 7, F = 1 365.94）. The result of precision test indicated that models for Cu and Pb performed well in modeling and predicting with a good ability of quantificational prediction, with relative percent deviation （RPD） values of 2. 046 and 3. 432 separately; Cr, Cd,Zn could perform generally in modeling and predicting with a good ability of quantificational prediction, with RPD values range of 1.4 - 1.8. The panel data model can be used to calculate various heavy metals at the same time and rapidly monitor soil heavy metals with hyperspectral reflectance in Xinzheng, with simple and fast calculation.