Estimation And Spatial Distribution Mapping Of Heavy Metal In Mining Area Soils Using Remote Sensing Data

In order to predict the spatial distribution of heavy metals in mining area accurately and quickly,we analyzed the correlation between 8 spectral indices,3 terrain factors,Landsat8 OLI satellite image spectral reflectance and heavy metal content in 45 samples collected from the field.First,the correlation coefficient was selected by simple factor modeling modeling of three elements of copper,lead and arsenic;then the multivariate linear regression model of piecewise modeling algorithm based on M5 model tree,according to the 20% sample points are not involved in modeling,comparative evaluation of two modeling methods.Finally,a piecewise multiple linear regression model based on M5 model tree algorithm was used to retrieve the spatial distribution of three kinds of soil heavy metals.The main research results are as follows:(1)terrain factors affect the distribution of three metal elements.Therefore,terrain factors should be considered when modeling,and the correlation between the distribution of three metal elements and remote sensing data has obvious seasonal sensitivity.(2)using Landsat8 spectral information and other factors can predict arsenic,copper and lead contents in soil by establishing models,of which copper has the best prediction effect.(3)The remote sensing reflectance of each band,with separate modeling modeling,terrain factor spectral index model this three ways of modeling accuracy is increasing,that in the study area,Landsat image prediction of heavy metal content in soil can improve the accuracy of the model by introducing the spectral index and terrain factors based on modeling.Spectral reflectance modeling separately and combined with spectral index modeling modeling,terrain factors of three kinds of models can achieve significant correlation(P<0.05),but compared with the R2 value,P value,the total average error,root mean square error and average relative error values on the whole,the latter is better than the former.The R2 prediction model of three metal components increased by 40.1%,7.6% and 16.6%,respectively,compared with the prediction model using spectral reflectance.The results show that spectral index and terrain factors can effectively improve the accuracy of prediction models.(4)the piecewise linear multivariate regression model based on the M5 model tree has more advantages than the ordinary multiple regression model.Based on the root mean square error(RMSE),the prediction model of three metals including copper,lead and arsenic can be obtained through calculation.The prediction model based on M5 mode l tree has increased by 27.3%,24.6% and 20.9%,respectively,compared with the simple regression model.The results show that the regression model based on M5 model tree is better than the general linear regression model.In the research area,the regression model based on M5 model tree can better predict the spatial distribution of the three metals in the soil.(5)According to the spatial distribution map of the predicted values of three metal contents based on the three metal prediction models,we can draw the conclusion that the maximum spatial distribution of the three metals is in the middle mining area and valley.According to the field survey,the roads in the study area are mainly transported ore,and the roads are mainly concentrated in the valley,and the residents live mainly in the valley,and the main streams are also in the valley.The prediction results are basically consistent with the actual situation.