Estimation of Heavy-Metal Contamination in Soil Using Remote Sensing Spectroscopy and a Statistical Approach

Heavy-metal-contaminated soil is a critical environmental issue in suburban regions. This paper focuses on utilizing field spectroscopy to predict the heavy metal contents in soil for two suburban areas in the Jiangning District (JN) and the Baguazhou District (BGZ) in China. The relationship between the surface soil heavy metal contents and spectral features was investigated through statistical modeling. Spectral features of several spectral techniques, including reflectance spectra (RF), the logarithm of reciprocal spectra (LG) and continuum-removal spectra (CR), were employed to establish and calibrate models regarding to Cd, Hg and Pb contents. The optimal bands for each spectral feature were first selected based on the spectra of soil samples with artificially added heavy metals using stepwise multiple linear regressions. With the chosen bands, the average predictive accuracies of the cross-validation, using the coefficient of determination R², for estimating the heavy metal contents in the two field regions were 0.816, 0.796 and 0.652 for Cd; 0.787, 0.888 and 0.832 for Pb; and 0.906 and 0.867 for Hg based on partial least squares regression. Results show that better prediction accuracies were obtained for Cd and Hg, while the poorest prediction was obtained for Pb. Moreover, the performances of the LG and CR models were better than that of the RF model for Pb and Hg, indicating that LG and CR can provide alternative features in determining heavy metal contents. Overall, it’s concluded that Cd, Hg and Pb contents can be assessed using remote-sensing spectroscopy with reasonable accuracy, especially when combined with library and field-collected spectra.