Estimating Heavy Metal Concentrations In Suburban Soils With Visible And Near-infrared Reflectance Spectroscopy

Soil contamination by heavy metals is an increasingly important problem worldwide.Its task of governance,prevention and monitoring has also been a key target for the national government and related environmental protection agencies.With the rapid development of society,the process of urbanization is accelerating,the urban expansion trend is obvious,and the problem of environmental pollution in the process of urban development is also becoming more and more serious.Suburbs play an important role in the city and they are generally important vegetable supply plants for the central urban area.The widely use of chemical fertilizers,pesticides,etc.are inevitably to cause a certain degree of impact or pollution on the soil environment.Therefore,it is very necessary to pay more attention and do more research on heavy metals in suburban soils.Traditional soil heavy metal pollution monitoring methods have the disadvantages of high cost,long period,and unsuitable for large-scale monitoring.In recent years,visible and near-infrared reflectance spectroscopy(VNIRS)technology has been used because of its rapid,efficient,low cost,and large-scale,being widely used in precision agriculture,environmental monitoring and food testing.In this study,93 soil samples were collected from the suburbs of Wuhan City.The contents of heavy metals,such as Cd,Pb,As,Cr,Cu,and Zn in soil samples,were derived from chemical experiments,as well as the content of organic matter(SOM)and Fe;The reflectance spectra(350 nm～2500 nm)of the soil samples were determined by the ASD FieldSpec?3 portable spectroradiometers(Analytical Spectral Devices Inc.,USA).This study aimed to:(Ⅰ)compare the effect of different pretreatment methods on the soil spectra.The pretreatment methods include the transmission to absorption log(1/T),Abs,Savitzky-Golay smoothing(SG),standard nomal variate transformation(SNV),multiscatter scatter correction(MSC),1st derivative(FD),and 2nd derivative(SD);(Ⅱ)compare the performance of partial least squares regression(PLSR)with original and transformed spectra for predicting the concentrations of six heavy metals(i.e.,Cd,Pb,As,Cr,Cu and Zn)in suburban soils;(Ⅲ)explore the response mechanism for estimating heavy metals in suburban soils by using correlation analysis,partial correlation analysis,principal component analysis and biplot analysis from the perspective of statistics.The main contents and results are as follow:(1)different spectral pretreatment methods have different effects on soil spectra.The SG method can effectively reduce the noise carried by the spectral signal,and make the spectral curve more smooth;the spectral differentiation(FD and SD)makes the spectral characteristic absorption peak signal more obvious and enrich the spectral information;the MSC and the SNV method have the same purpose to reduce or even eliminate the difference of soil particle size and scattering caused by the uneven distribution of soil particles on the spectral reflectance,but the SNV method has a greater impact on the soil spectra;Abs method can effectively eliminate the impact of systematic errors,after treatment the spectral curve is similar in shape to the original spectral curve folded 180degrees along the x-axis.(2)In the process of predicting the concentrations of SOM,Fe and six heavy metals by using the PLSR modelling with original spectra,the model has best accuracy for organic matter,Fe and Cr.The prediction accuracy were R~2_p=0.81(RPD=2.22)and R~2_p=0.81(RPD=2.25),R~2_p=0.83(RPD=2.55),respectively;the medium prediction accuracies were for Pb(R~2_p=0.58,RPD=1.49),As(R~2_p=0.60,RPD=1.59),Cu(R~2_p=0.51,RPD=1.81),respectively;the worst predictons were for Cd(R~2_p=0.26,RPD=1.17)and Zn(R~2_p=0.06,RPD=1.04).(3)In the process of predicting the concentrations of SOM,Fe and six heavy metals by using the PLSR modelling with transformed spectra,the SG method outperformed other pre-treatments for building the estimation modelling.For SOM and Fe,the model still presented with excellent performance,and their prediction accuracy were R~2_p=0.93(RPD=3.31),R~2_p=0.88(RPD=2.89),respectively;For As,Cr,Cd,they can also be successfully predicted,and their prediction accuracy were R~2_p=0.71(RPD=1.81,respectively),R~2_p=0.89(RPD=2.92),R~2_p=0.58(RPD=1.49),respectively;Unfortunately,the models for Pb,Cu,Zn were unsatisfactory,and the prediction accuracy were R~2_p=0.25(RPD=1.15),R~2_p=0.32(RPD=1.16),R~2_p=0.22(RPD=1.11),respectively.In other words,the PLSR modelling in this study cannot be used to predict the concentrations of Pb,Cu and Zn in suburban soils.(4)The mechanism of estimating the concentrations of heavy metal in soils with VNIRS was mainly influenced by the relationship between heavy metals and organic matter and iron oxides.Generally speaking,the higher correlations means the better modeling performance.In addition,the degree of dependence of different heavy metals on organic matter and Fe is also different in the process of modelling.Cd is mainly affected by organic matter,As and Cr are mainly attributed to Fe content,and Pb,Cu,and Zn are jointly affected by SOM and Fe contents.