Possibilities of visible-near-infrared spectroscopy for the assessment of soil contamination in river floodplains

During the past decades, large amounts of diffuse cadmium (Cd) and zinc (Zn) contaminated soil material have been deposited in the floodplains of the river Rhine in the Netherlands. As spatial information on soil quality is required at different scale levels covering the whole area, characterisation exclusively based on soil sampling and analysis is time-consuming and very expensive. A quicker method is developed based on a multivariate calibration procedure using partial least squares (PLS) regression to establish a relationship between reflectance spectra in the visible-near-infrared (VNIR) region and spectrally active soil characteristics (organic matter and clay content) that are inter-correlated with concentration levels of Cd and Zn. Several spectral pre-processing methods (normalisation, multiplicative scatter correction (MSC), derivation, standard normal variate (SNV) transform) were employed to improve the robustness and performance of the calibration models. No pre-processing gave the best results for Cd and Zn with RMSECV equal to 0.676 and 80.97mgkg-1, respectively. Application of the calibration models for soil quality characterisation in river floodplains is promising. The future possibilities of multivariate calibration and pre-processing in remote sensing have to be explored.