Quantitative Estimating Of Soil Heavy Metals Pollution With Hyperspectral Remote Sensing In Shizhuyuan Mining Area

In mining production,the waste gas,waste liquid and waste residue carrying heavy metals that exceed the standard have caused serious soil pollution around the mining area,affecting the health of local residents,animals and plants.Therefore,analyzing and monitoring the soil pollution around the mining area is of great significance to ensure the sustainable development of the mining area and the quality of life of the people in the mining area.The use of hyperspectral remote sensing technology to monitor the content of heavy metals is a new soil monitoring method,which has the advantages of convenient data collection,low cost and fast analysis.It is a new developing technology to quickly monitor soil quality in recent years.In this paper,the Shizhuyuan nonferrous metal mining area in Hunan Province is used as the research area,the soil’s reflection spectrum was measured by SVC HR-1024i.On the basis of the study on the characteristics of heavy metals content and spectral characteristics of soil,linear and nonlinear methods,including Multiple Linear Regression（MLR）,Partial Least Squares Regression（PLSR）and Support Vector Machine Regression（SVR）,were utilized to build the inversion model for predicting soil heavy metal As,Pb,Cu,and Cd content in soil.Finally,the coefficient of decision（R²）and the root mean square error（RMSE）of the modeling set and the validation set are used as comprehensive evaluation indicators to evaluate the hyperspectral inversion method’s accuracy of the soil heavy metal element,and obtain the optimal prediction model of each heavy metal element.This research will provide technical support for the rapid and accurate acquisition of heavy metal content in soil,and provide a new ideas for the spectral monitoring of other soil heavy metal elements in soil.The main research results are listed as following:（1）The soil in the Shizhuyuan mining area is a weakly alkaline soil rich in heavy metals,which is affected by both organic matter and iron oxide.The degree of heavy metal pollution is Cd>Pb>As>Zn>Cu;（2）Four spectral transformation methods of first-order differential,standard normal variable transformation,continuum removal and reciprocal logarithm can effectively improve the ability of soil spectrum to characterize heavy metal content,among which the first-order differential transformation has the most significant effect.The characterization effect can be further enhanced by combining the difference and quotient operations of the bands with"synergistic effect".The combination characteristic bands with the best correlation between As,Pb,Cu,Cd and heavy metal content are B_FD（1530,684）,B_FD（410,2324）,B_FD（1266,2048）,B_FD（556,2325）,respectively.The absolute values of the corresponding correlation coefficients are 0.786,0.69,0.603and 0.733,respectively;（3）The characteristic bands of soil As in the study area are mainly distributed in the range of 540～558 nm,647～650 nm,669～674 nm,712～762 nm,1014～1202 nm,1288～1320 nm and 1528～1533 nm;Pb characteristic bands are mainly distributed in the range of 410 nm,555nm,653 nm,893～896 nm,1522～1526 nm,1953～1979 nm,2167～2173nm,2300～2304 nm,2322～2329 nm and 2385～2388 nm.Cu characteristic bands are mainly distributed in the range of 385 nm,1212～1218 nm,1264～1267 nm,1420～1463 nm,1570～1580 nm,1702～1707 nm,2040～2090 nm and 2250～2290 nm;Cd characteristic The bands are mainly distributed in the range of 400～515 nm,551～557nm,654 nm,890～950 nm,1655～1681 nm,1716～1761 nm and1996～2400 nm.The sensitive bands of soil Pb and Cd have a certain similarity,showing multiple overlapping or similar characteristic band ranges;（4）The three models of MLR,PLSR and SVR have better prediction effects on As,and the R²of the modeling set and prediction set are both greater than 0.85.In general,the best prediction model for As elements is the SVR model;MLR and PLSR are linear The model has poor prediction effect on Pb and Cu elements in the study area.The best prediction model for Pb and Cu elements is the SVR model.The modeling set R²reaches 0.74 and 0.68 respectively,and the prediction set R²reaches 0.57 and 0.74 respectively.The overall inversion prediction effect is relatively good.it is good.The prediction effect of the three models is better,and the best prediction model for comprehensive analysis of Cd elements is the MLR model.