Visible-Near and short-wave infrared reflectance spectroscopy has the potential to become an important additional technique in the geosciences for soil classification, mapping and remote determination of soil properties and mineral composition. It is also becoming increasingly important to improve the spatial resolution of soil maps to better tackle localized issues such as soil contamination. Long-term spiked soils having a range of lead (Pb) concentrations from five different locations across Australia were analysed for a range of physio-chemical properties as well as their spectral reflectance between 400 and 2500nm. Spectral and chemical analyses were correlated using partial least squares regression (PLSR), to predict soil Pb concentration. While across all soils studied (n=31), the Pb content was weakly predicted from spectra, reliable correlations with the major spectrally active components were found in models of total carbon and iron, which were predicted much better than most other soil constituents. However, a good prediction of Pb concentration was found in two of the soil subsets which indicated that spectral reflectance analysis may require soils to be of the same type in order to be effective. For a long-term atmospheric smelter emission Pb contaminated soil, the correlations between Pb measurements and spectral reflectance in both calibration (Rc2) and validation (Rv2) modes reached 0.99 and 0.75 respectively with a calibration root mean square error (RMSEc) of 19 and validation root mean square error (RMSEV) of 345 and an acceptable performance of deviation RPD of 1.7. For a long-term spiked (LTS) soil, both Rc2 and Rv2 reached 0.99 and 0.96 respectively with a RMSEc of 58 and a RMSEV of 396 with an excellent RPD of 12.15. These results indicated that reflectance spectroscopy has the potential to rapidly determine Pb contamination in a homogeneous soil.
- Lead contamination
- Partial least squares regression
- Regression analysis