The Study On Hyperspectral Quantitative Estimation Of Soil Heavy Metals Based On Combined Modeling Theory

With the strong support of the national policy for Xinjiang’s economy,the level of industry and agriculture in Xinjiang had been improved continuously,but the environmental issues shouldn’t be overlooked.As one of the important components of soil’s pollutants,heavy metals would not only caused damage to water and vegetation,but also caused human health problems.Therefore,it was great significance to monitored and give early warning of soil heavy metals in real time.Hyperspectral remote sensing data dued to it’s multiple and continuous spectral bands and speeds.The characteristics of rapid,low-cost and non-destructive provided basis for the practical application of remote sensing technology in soil heavy metal content monitoring.This paper used the hyperspectral remote sensing as a technical means to explored the hyperspectral remote sensing in the industrial zone on the northern slope of the Tianshan Mountains in the Zhudong region,including 2 cities and 2 counties of Midong,Fukang,Jimsar,and Qitai,the application potential of five soil heavy metals:As、Hg、Cr、Cu、Pb.Among them,the single quantitative estimation modeling method included Stepwise Multiple Linear Regression,Partial Least Squares Regression,and Error Back Propagation Algorithm;the combined quantitative estimation modeling method included multiple stepwise regression combination model,partial least squares combination model,and Error Back Propagation Algorithm combination model.To explored the feasibility of several modeling methods for heavy metal inversion in high spectral soil.The main findings are as follows:(1)The statistical results of heavy metal data indicated that the As,Hg,and Pb in the five soil heavy metals are higher than the background values??in Xinjiang;the Pb was only higher than the national background value and had a certain degree of accumulation.The average order of heavy metals in five soils was:Cr>As>Pb>Cu>Hg.According to the distribution pattern of the pollution load index,the high values were mainly concentrated in the Dahuangshan area and the southwestern Fukang area where the five heavy metal single factor pollution indexes were high.(2)The spectral reflectance curve of the soil in the study area was roughly an upwardly convex parabolic type.The spectral curves of 173 samples were similar in shape and approximately parallel.The spectrum of the 400-600 nm spectrum was steeper across the entire spectrum of the curve.The spectral shape of 600~1350 nm was relatively flat;near the curve centered at 600 nm,800 nm,900 nm,and 1000 nm was the reflection and absorption peak of the spectrum.(3)The R′、R〞、S(R)、(S(R))′、(S(R))〞、1/R、(1/R)′、(1/R)〞、LgR、LgR′、LgR〞、1/LgR、(1/LgR)′、(1/LgR)〞、Lg(1/R)、(Lg(1/R))′、(Lg(1/R))〞17 forms of transformation,correlation analysised of spectral reflectance after conversion and heavy metals,and the significant test of correlation coefficient at the0.01 level to extracted the characteristic band,compared with the original reflectance correlation coefficient.The correlation coefficient of the five kinds of heavy metals had been correspondingly increased,so it was necessary to understand the characteristic bands of the spectrum through differential transformation techniques.(4)Applied multivariate stepwise regression,partial least-squares regression,and error back propagation algorithm to construct an estimation model for five heavy metals,and used three kinds of evaluation indicators to test the predictability of the model.The results showed that three kinds of As are contained in five kinds of heavy metal elements.The prediction accuracy of the modeling method is the highest.Among three modeling methods,the Error Back Propagation Algorithm can obtain better results than the other two modeling methods.The estimation ability of the three models was not much different,but the coefficient of determination didn’t exceed 0.6.It showed the need for further combined forecasts.(5)Combined multiple stepwise regression and partial least squares regression used the improved simple average method,the Error Back Propagation Algorithm useed the Error Back Propagation Algorithm optimal weights to legally predict and obtained a combined multiple stepwise regression model(Z-SMLR),combined partial least squares regression model(Z-PLSR),combined Error Back Propagation Algorithm model(Z-BP).In addition to individualed elements,the combination of three kinds of accuracy evaluation indicators,compared with the single prediction model,the determination coefficient(R~2)is higher,the root mean square error(RMSE_C)and the prediction error ratio(RPD)effect is better than the other,compared to the remaining for the five estimation models,the prediction ability and stability of the Z-BP model were all optimal.Therefore,the Z-BP model was the best model for five heavy metals such as As、Hg、Cr、Cu、Pb with coefficient of determination of 0.67、0.65、0.54、0.60、0.63.