Inversion Of Heavy Metals With GWR Model In Arable Land In Mining Area

As an important part of social development,Mining development has promoted the growth of economic.However,the mining and smelting of minerals exposes the original heavy metals buried in the soil to the surface and eventually flows into the surrounding arable soil environment,seriously threatening the growth environment of arable land food crops and ultimately endanger the safety of people’s lives.Therefore,using quantitative inversion techniques of remote sensing to monitor and evaluate the environmental quality of arable land soil in mining areas in order to better resolve the contradiction between mining development and the surrounding safe environment for cultivated land crops is of great significance to the society.This paper selects the soil around a mining area in Liuyang as the research object.Based on the theoretical basis of quantitative inversion for remote sensing,we take the soil hyperspectral data as the main research line.Firstly,the soil hyperspectral data is collected through experiments and the data is subjected to conventional preprocessing and combined transformation processing.Secondly,taking advantage of the heavy metal content measured by laboratory physicochemical analysis method to carry out correlation analysis to extract the characteristic bands of each heavy metal.Finally,utilizing the BP neural network and geographically weighted regression method to establish a hyperspectral quantitative estimation model for soil heavy metals in this study area and analysis and evaluation the prediction accuracy between this two models.The main research results of this paper include:(1)The soil of arable land in the study area is most severely contaminated by the heavy metal Cd.The spatial variability degree of heavy metal elements is:Cd>Cu>Pb,and there may be cluster symbiosis between Cu and Pb、Cu and Cd.(2)Compared to traditional preprocessing methods,the combination of differential transformation methods with reciprocal or logarithmic are the most sensitive ways to extreme differences of spectral data,which can effectively improve its correlation with heavy metals,and the model established by it also has better accuracy.(3)The BP neural network model established by Cu and Pb elements has good accuracy and prediction ability.The accuracy of the BP neural network model established by Cd elements is poor;The geographically weighted model(GWR)established by Cu,Pb,and Cd elements all have good accuracy.When the spatial variability of heavy metals is large,the modeling accuracy of the GWR model is particularly significant.This study analyzes the soil spectrum through combinational ways and respectively establishes the model based on the degree of variation.It provides new ideas and methods for modeling the hyperspectral inversion of arable soil,and it also helps to realize the change.of quantitative inversion technology of remote sensing from measured hyperspectral to multispectral imaging,which certainly has reference value for the promotion of pollution monitoring method with airborne satellite remote sensing technology in large area.