Study On The Characteristics Of Soil Magnetic Parameters And Their Hyperspectral Inversion In Industrial Areas

With the acceleration of urbanization,while industrial zones promote urban economic development,pollutants generated by industrial activities directly or indirectly enter the soil of cities and surrounding areas,making soil environmental problems increasingly serious.The southwestern industrial zone of Minhang District is an important economic growth point in Shanghai.Carrying out environmental monitoring in the industrial zone can provide a scientific basis for soil pollution prevention and control.This article mainly focuses on the industrial area in the southwest of Minhang District,Shanghai as the research object,analyzes the statistical characteristics and spatial variability characteristics of soil magnetic-parameters in the industrial zone.The article model is established on the basis of differential transformation,and through correlation analysis,selecting feature bands with correlation above 0.01.using linear（multiple stepwise regression model and least square regression model）and nonlinear（BP artificial neural network）modeling methods to establish a hyperspectral inversion model of soil magnetic parameters.The best fitting model is selected,mainly considering the accuracy and stability of the model.It provides methods for quickly and effectively predicting changes in urban soil magnetic parameters,and provides technical support for urban environmental pollution monitoring.The main conclusions are as follows:（1）According to the analysis of the statistical characteristics of soil magnetic-parameters in the study area,shows that the soil magnetic minerals is relatively high in the industrial zone,and the magnetic properties of the soil are significantly enhanced.The soil magnetic minerals are mainly ferrimagnetic minerals.The soil magnetic mineral particles are mainly single domain（SD）and multi-domain（MD）of coarse ferromagnetic crystal grains,containing a small amount of superparamagnetic particles（SP）and they are mainly affected by man-made factors rather than soil-forming factors.（2）It can be concluded by semi-variance function:χ_lf,SIRM,HIRM,χ_arm/χ_lf,χarm/SIRM are better fitted with exponential models,andχ_fd%is better fitted with spherical models.χ_arm uses Gaussian model to fit better.The nugget coefficients are all greater than 25%,and the nugget coefficients ofχ_lf,SIRM,HIRM,χ_fd%,χ_arm,χ_arm/χ_lf,χarm/SIRM,and SIRM/χ_lf are 0.6,0.58,0.5,0.70,0.68,0.74,0.67,0.61.The shows a-moderate autocorrelation,indicating that the variation characteristics are caused by both structural factors and random factors.Through the spatial interpolation map of soil magnetic parameters,we can found thatχ_lf,SIRM,and SIRM/χlf have similar spatial variation characteristics.High-value areas appear in the northeast,where is near Luchun Road and Beidou Road in the Minhang Economic and Technological Development Zone.There are mainly power station equipment,medical industry,rail transit,zipper factory,heavy machinery factory.Industrial activities are relatively intensive,so the increase ofχ_lf is closely related to the production activities in the industrial zone.The low-value areas appear in the south,and the neighborhood is mainly residential areas.Residents produce less magnetic minerals in their activities,so soil magnetic parameters are low.（3）The spectral curves of the soil samples in the industrial zone are basically the same in the visible-near-infrared waveband,and they belong to the gentle slope type.In order to eliminate the"burr"noise in the spectrum and get the spectrum information closer to the real information.The soil spectrum curve is smoothed,and we can find that the frizziness of the spectrum curve is reduced and the curve is smoother.The slope of the spectral curve between 380-600nm the shape is steeper,and near 1400nm the increase is slower.In conclution,the spectral curve increases faster in the visible range and in the near-infrared range increases slowly.（4）In order to eliminate the influence of the original spectral background noise,effectively highlight the characteristic band,by first-order differential,second-order differential,first-order logarithmic differential,and second-order logarithmic curve on the spectrum curve of the study area Differential,reciprocal first-order differential,and reciprocal second-order differential transformation processing,we can find that the peak and valley information of the spectral curve in the visible light 350-520nm and the near-infrared region of the soil spectrum curve band increases more obviously.Especially near 1034nm,the peak and valley information of the band is the most obvious.In addition,the correlation between the spectral reflectance after differential transformation and the low-frequency magnetic susceptibility,saturated isothermal remanence,and non-hysteresis remanence has been significantly improved,and the characteristic band has been significantly enhanced.Through Pearson correlation analysis,it is found that the characteristic bands of low-frequency magnetic susceptibility,saturated isothermal remanence,non-hysteresis remanent magnetic susceptibility,and spectral reflectance appear in the near-infrared band,around1034nm-1408nm.（5）Using the multiple stepwise regression model to model and predictχ_lf,SIRM,and SIRM,we can find that:the modeling effect ofχ_lf is the best,and the optimal model involves the band mainly in the vicinity of the strong iron band of the spectrum.The modeling effect of the reciprocal first-order differential transformation of the spectrum and the saturation isothermal remanence is the best,and the spectrum of the optimal model is automatically.The variable is mainly located near the weak absorption peak of the iron band.The characteristic band of non-hysteresis remanence is mainly located near 829nm,which is affected by the soil organic matter,and the modeling effect is not ideal.（6）Using the BP artificial neural network model to model and predictχ_lf,SIRM,χ_arm,we can be concluded that SIRM andχ_lf have the best effect,and the model prediction error fluctuates between-5-5.The use of spectral logarithmic first-order differential transformation and saturation isothermal remanence modeling has the best effect,and the model prediction error is-10-2.The modeling effect of the first-order differential transformation of the spectrum and the non-hysteresis remanence susceptibility is the best,and the model prediction error fluctuates between-3-1.（7）Using Partial Least Squares Regression Model to model and predictχ_lf,SIRM,andχ_arm,it can be concluded that:The modeling effect of the first-order differential transformation andχ_lf,is the best,the determination coefficient R² is up to 0.58,and the root mean square error is also small,and the model fits well.By second-order differential transformation of the spectrum and the saturation isothermal remanence,the model effect is the best,the R² of the modeling set and the verification set are both high,and the stability and predictability of the model are relatively ideal.The non-hysteresis remanence magnetic susceptibility is mainly affected by the soil organic matter content,and the modeling effect is bad.（8）Comparing different modeling methods,it can be concluded that theχ_lf and SIRM using multiple linear stepwise regression model are less discrete than BP artificial neural network and partial least square regression.Model,so the use of multiple linear stepwise regression model to invert the soil magnetic parameters ofχ_lfand SIRM is more applicable.Compared with the multiple stepwise regression model and the partial least squares model,the BP artificial neural network model ofχ_arm has better fitting effects and lower degree of dispersion.Therefore,the applicability of using the BP artificial neural network model to invert the non-hysteresis remanence magnetic susceptibility is better.