Study On Heavy Metals In Cultivated Soil—crops In Karst Mountain Areas Based On Hyperspectral Technology

The soil circle is in the junction of the atmosphere,hydrosphere,lithosphere and biosphere,and the flow of material and energy is complex.It is an indispensable condition for human beings to carry on normal production and life.As the saying goes:"food for the world",food is to maintain the basic functions of people's body functions,the root cause of food is soil,soil safety and soil fertility both have both to meet the human green,healthy and sustainable Development requirements.However,there is a certain gap between the ideal and the reality.The reality is that with the progress of science and technology,crop yields of cultivated land have been rising steadily.People's material life is abundant,but some farmland ecological ENVIronment is getting worse.Large amounts of pesticides,uncontrolled fertilizers,and unreasonable use of mineral resources have caused serious heavy metal pollution in the soil.Traditional chemical detection methods can no longer meet the development needs of modern agriculture.The development of hyperspectral technology provides us with new ideas for the key issues of large-scale,dynamic and rapid monitoring of soil heavy metal pollution and soil fertility.Due to the special landform and the soil forming ENVIronment in Karst area,the soil material,soil infertility and discontinuity in this area are very scarce.The contents of Mn,Ni,Zn,Pb,As and Cd in soils with limestone parental development.The background value of heavy metal elements is usually higher than that of other soil parent materials,which makes the agricultural health and sustainable development of karst mountainous soil seriously threatened.In this paper,soil spectral data of a farmland in Guiding County of Karst region were obtained by using portable land-based spectrometer(ASD)and airborne hyperspectral imaging system(GaiaSky-mini,GS)respectively for soil heavy metal pollution and soil organic matter content monitoring,The ASD spectral data were used to retrieve the heavy metal content of plant leaves.The main contents and conclusions are as follows:(1)Based on the first-order derivative(FDR)spectral data of ASD and GS data sources,a regression model of As,Pb,Cd and Cr four elements was established by Partial Least Squares Regression(PLSR)Low,did not model it).Among them,the best model was Cd,the relative analytical error(RPD)was 1.50 for GS spectra inversion,and the ability to roughly estimate Cd content in soil was the highest.The highest RPD was 1.95 in ASD spectra with quantitative estimation ability.Based on ASD spectral inversion,the AS model has a rough estimate of soil AS content with a RPD value of 1.45.The GS model based on the GS spectra only has the ability to distinguish between high and low AS levels,with a RPD value of 1.13.(2)Based on the original spectral(R),FDR and second-order derivative(SDR)spectral data of ASD and GS,the inversion model of soil organic matter(SOM)was established with PLSR.Among them,the regression model established by ASD spectral FDR transform is the best,the validation set determination coefficient(Rv~2)is 0.91 and the relative analytical error(RPD)is up to 2.68,which has a very good predictive ability.In the GS spectral SDR prediction model,Rv~2 is 0.77,RPD is 1.49,with a rough estimate.(3)The related research shows that SOM has certain adsorption capacity for heavy metal Cd.Based on the model between SOM and Cd and the predicted SOM content,the SOM content of hyperspectral data is used to indirectly retrieve soil Cd content.Within the range of 400~1000nm,where the GS spectral data source does not have predictability,its R~2is only 0.27.Comparing with the above-mentioned direct-regression based on spectral data,the accuracy of the heavy metal Cd content is R~2of 0.71,which is no doubt based on the direct return of spectral data to soil Cd content.ASD spectroscopy has good predictability,its R~2 is 0.82,relative analysis error(RMSE)is 0.028,compared with the above-mentioned spectral data based on the direct regression of heavy metal Cd content accuracy R~2 0.77,RMSE 0.025,comprehensively indirectly return to the soil with organic matter content The effect of Cd content is slightly better.(4)Based on the original,continuous removal and first-order differential transformation spectral data of eggplant leaf ASD spectral data,the Cd content estimation model of eggplant leaves was established by multiple stepwise regression(SMLR)and support vector machine(SVM)R~2 of 0.759 and RMSE of 0.0227;R~2 of SVM is 0.794 and RMSE is 0.0328.Taken together,the model established by SMLR is more suitable for the prediction of Cd content in eggplant leaves.