[Study on the prediction of soil heavy metal elements content based on mid-infrared diffuse reflectance spectra]

The present paper analyzed the possibility of mid-infrared diffuse reflectance spectra for quick assessment of heavy metal element content in soil quickly. Soil samples were collected from Jiangning District and Baguazhou Island, and the numbers of sample were 103 and 58 separately. Jiangning District samples were used as calibration set while Baguazhou Island samples as validation set. To assess the utility of different pre-treatment process of MIR spectroscopy for soil heavy metal element content analysis, we used PLSR method to develop the calibration between spectral data and soil elements content. Three spectral pretreating techniques such as smooth, log(1/N), baseline correction, multiplicative scatter correction were used for promotion of predicting performance. The result showed that the progress of (log-BC-MSC) in turn achieved optimal calibration of MIR spectra and better prediction for ex-situ soils. Though the calibration data were treated by different pre-treating schema, the R2 of the 8 elements followed the same law: Ni > 0.8 > Cr, Cu, Zn, Pb, Hg > 0.6 > As, Cd. When we applied these calibrations to Baguazhou Island soils, (log-BC-MSC) treated data results in the smallest RMSEp-BGZ. We used the same calibration method to compare the predictive ability of MIR spectra to VNIR spectra. The R2 of 8 elements developed by VNIR spectral calibration are sometime larger than MIR's, but after we applied these calibrations to validation set, the RSME of MIR data for prediction of BGZ soil samples is 21% to 73% of VNIR's. This result showed us that for predicting ex-situ soils, MIR analysis substantially outperformed VNIR These results indicated that MIR spectra can be used to predict soil heavy metal content quickly and non-destructively.