Inversion of heavy metals content with hyperspectral reflectance in soil of well-facilitied capital farmland construction areas

Abstract: Hyperspectral reflectance provides an alternative method to soil's physical and chemical analysis in laboratory for the estimation of soil properties in large range. In order to achieve rapid measurement of the soil heavy metal content in well-facilitied capital farmland construction areas, 154 soil samples at 0-30 cm depth were collected as research objects, which were from well-facilitied capital farmland construction areas in Xinzheng City, Henan Province. The raw hyperspectral reflectance of soil samples was measured by the standard procedure with a spectrometer of ASD Field Spec3 equipped with a high intensity contact probe under the laboratory conditions. Meanwhile, the contents of Cr, Cd, Zn, Cu, and Pb in these soil samples were analyzed. The 116 samples were used for building hyperspectral estimation models and the other 38 samples were used for model validation. In the next, the raw spectral reflectance of 400-2400 nm after multiplicative scatter correction and Savitzky-Golay was transformed to 2 spectral indices, i.e. first order differential reflectance(FDR) and second order differential reflectance(SDR). The correlation coefficient between the 3 kinds of spectral indices and Cr, Cd, Zn, Cu, Pb content was analyzed by Pearson correlation analysis. Then, the correlation coefficients (P<0.01) of the 3 spectral indices were got in significant test, which could be used to extract significant bands. At last, we used partial least squares regression (PLSR) method to build quantitative inversion models of soil heavy metal content based on significant bands for this study area, respectively. The prediction accuracies of these models were assessed by comparing determination coefficients (), root mean squared error (RMSE) and relative percent deviation (RPD) between the prediction and validation values. Based on these, the optimal models were selected. The spatial distribution map of Cr, Cd, Zn, Cu and Pb content was made by geographical interpolation. The results showed that, conducting the first order differential reflectance and second order differential reflectance transformation on raw soil spectral data, could highlight the hidden spectral reflectivity characteristics effectively. Among all of the 3 spectral indices based on PLSR model, the model of second order differential reflectance about Cr could obtain more robust prediction accuracies, its values of was 0.88，its values of RPD was 1.68; the model of the raw spectral reflectance (R) of 400-2 400 nm after multiplicative scatter correction（MSC）and Savitzky-Golay（SG）about Cd、Zn and Cu could obtain more robust prediction accuracies, their values of were 0.70, 0.88 and 0.99, their values of RPD were 1.50, 2.05 and 3.36 respectively; Pb could obtain more robust prediction accuracies, their values of was 0.93，its values of RPD was 3.16. The optimum model of soil heavy metal was used to interpolate the soil heavy metal content; the content of Zn was in accordance with the standard of soil environmental quality, and the contents of Cr, Cd, Cu and Pb met the soil environmental quality standard Ⅱ, but the contents in some well-facilitied capital farmland construction areas were more than the soil background value. This study provides a reference for the real-time monitoring of soil basic information in well-facilitied capital farmland construction areas by hyperspectral inversion model.