Estimation Of As Content In Soil Based On Hymap Data

In this paper,the heavy metal As in the soil of Siping City,Jilin Province was taken as the object of study,the soil reflectance spectra extracted by Hymap hyperspectral image were used for continuum removal,first-order derivative,second-order derivative,inverse logarithms,and the correlation of soil As content measured by laboratory were combined to analyze the correlation between soil As content and spectral variables.By analyzing the correlation between soil As content and spectral variables,using multistepwise regression,partial least squares regression,and Bp neural network method to establish hyperspectral inversion model of soil As content,preliminary exploration directly using hyperspectral satellite image to estimate the feasibility of the soil As content.The main results are as follows:The As content of soil in the study area is 6.58~419.96mg/kg,and its average value was about 3.23 times of the soil background value,indicating that the soil in the study area has reached a certain degree of pollution.The correlation between soil reflectance which extracted from Hymap hyperspectral image and soil As content is very low,the maximum correlation coefficient is 0.177,after continuum removal,first-order derivative,second-order derivative and inverse logarithms transformation,the correlation between spectral variables and soil As content becomes stronger,among them,the correlation between first-order derivative spectrum variation and soil As content was the highest,the maximum correlation coefficient is 0.520,second-order derivative spectrum is the second and the maximum correlation coefficient was 0.499.The inversion model was established using the reflectance spectra of the soil and the transformed spectral variables and the measured values of the soil As content,the first-order derivative spectral data model is the best,the first-order derivative inversion model validation coefficient of determination R~2 of multiple stepwise regression,partial least squares and Bp neural network is 0.480,0.573 and 0.659,respectively.The order of three methods for estimating soil As content inversion model is BP neural network model>partial least squares model>multivariate stepwise regression model.The relationship between soil composition and soil spectrum is not a simple linear relationship,so the multivariate stepwise regression model and partial least squares model show some limitations,and the Bp neural network model can better deal with the nonlinear relationship between the spectral variables and the soil As content.According to the above conclusions,the spatial distribution map of soil heavy metal As content was obtained by inversion of the first-order derivative Bp neural network model and Hymap hyperspectral image.