Inversion Of Heavy Metals In Farmland Surface Soil Based On Vis-NIR Spectrum

For the monitoring of heavy metal content in the surface soil of farmland,traditionally mainly rely on physical or chemical analysis and measurement methods of ground sampling,and these discrete and local monitoring methods lack continuity in time and place,so it is difficult to achieve large-scale inversion prediction of dynamic changes of heavy metals in the surface soil of farmland.In recent years,with the continuous accumulation of heavy metal content in the surface soil of farmland,the problem of heavy metal pollution in farmland has intensified,and the inversion of heavy metal content in farmland soil has gradually developed from static to dynamic,from discrete to continuous,and from small-scale to large-scale monitoring.Direction and quantitative remote sensing is also meeting the needs of this development trend.In this study,farmland in the sewage irrigation area in the north of Longkou was determined as the study area.The contents of eight heavy metal elements Cr,Cu,Mn,Pb,Zn,As,Cd,and Hg in the surface soil of the farmland were selected as the research objects.Indoor hyperspectral data of the soil samples were collected to determine the corresponding types in the soil samples.The content of heavy metals is based on the measured content of heavy metal elements in the laboratory.The statistical characteristics of each element are described.The spatial interpolation analysis method is used to estimate the spatial distribution characteristics of the research objects in the surface soil of the study area.The spectral sensitive parameters of each element were extracted by combining spectroscopy and correlation analysis techniques.Using mathematical statistical methods such as principal component analysis and partial correlation analysis to analyze the correlation between heavy soil metals and soil components,and exploring the feasibility of using soil Vis-NIR spectral data to estimate the content of heavy metals;the fit of various spectral transformation forms is the smallest two-square regression method,geographic weighted regression method and sample classification local regression analysis method were used to construct the model,and the results of comparative analysis,modeling and verification were obtained to obtain the optimal spectral indexes and the best prediction models of the eight heavy metal element contents.The research results mainly include:(1)The average values of Cr,Cu,Mn,Pb,Zn,As,Cd,and Hg in the soil are 61.52,29.80,456.53,36.48,62.59,8.06,0.34,and 0.03 mg/kg.The coefficients of variation of the eight elements are sorted from large to small as Cu> Pb> Hg> Zn> Cd> As> Mn> Cr.Cu and Pb are unevenly distributed in the topsoil,and there are specific values due to human factors.The spatial distribution of heavy metals shows that the high value areas of most elements are concentrated in the northeast of the study area.Among the eight heavy metal elements,with the exception of Zn,the spatial distribution of the contents of the other seven elements showed spatial differences to varying degrees.(2)Correlation analysis between the content of each heavy metal and the spectral index,the maximum value of the Pearson correlation coefficient of the content of each element after the spectral preprocessing and the three spectral index data,compared with the maximum correlation with the original spectral reflectance The coefficients have increased significantly.By comparison,it is found that First derivative(FD)has better enhancement of the spectral characteristics of Cr,Mn,Zn,Cd,and Hg elements,while Second derivative(SD)has more obvious enhancement of the spectral characteristics of Cu,Pb,and As elements.Compared with the original spectrum,except for the Multiplicative scatter correction(MSC)transformation of Cu,Mn,Zn,Cd,and Hg elements,the maximum values of the correlation coefficients of other spectral transformation indexes have been significantly improved.(3)The correlation between the soil visible light-near infrared reflectance spectrum and the heavy metal elements in the soil is affected by the soil components with obvious spectral characteristics such as SOM and Fe elements to varying degrees,and the closeness of the correlation determines the accuracy of the prediction.The experimental results show that the prediction of Zn and Hg content mainly depends on the correlation with SOM,and the prediction of Cr,Mn and As content is mainly reflected in the correlation between the three and Fe element,while Cu,Pb,Cd can not determine its active soil composition.(4)Based on the results of modeling and verification of the inversion model of soil heavy metal element content,it was found that the partial least squares regression model has prominent advantages in modeling soil heavy metals Hg,Mn,and Zn in the study area,and the model predictive ability it is good.The inversion model established by geographic weighted regression has better stability than the other two models.The inversion prediction of soil heavy metals Mn,Hg,Zn,Cr,Cu,and Pb is better,and the RPD of the optimal model for most elements is More than 1.8.The coupled sample classification local regression model has the strongest ability to estimate various heavy metals,but the model stability is slightly lower than the established GWR model.The coupled sample classification local regression inversion model has the best prediction effect on soil As elements.The optimal model outside Cu element can meet the demand of inversion prediction of soil heavy metal element content.