Feasibility Study of PLS and Bagging-PLS Regressions in Predicting Some Soil Heavy Metals by VIS to NIR and SWIR Bands: Case Study of Hormuz Island Soils

Using visible (VIS), near-infrared (NIR), and short-wave infrared (SWIR) spectra can provide a quick, undisturbed, and inexpensive method for predicting deferent soil variables. This study aims to evaluate the ability of hyperspectral information in VIS-NIR-SWIR range to predict heavy metals concentrations including As, Pb, Sb, and Cr in Hormuz Island soils, with a total area of about 42 km 2 , south of Iran. Consequently, 57 samples were taken from the topsoil. The total As, Pb, Sb, and Cr concentrations were measured by ICP-OES apparatus. Soil sample’s spectral reflectance was quantified by an ASD field spectroradiometer. PLSR separately and coupled with Bootstrap aggregation (Bagging-PLSR) methods were employed to predict total concentrations of As, Pb, Sb, and Cr. The findings demonstrated that hyperspectral data in VIS, NIR, and SWIR could reasonably estimate the As, Pb, Sb, and Cr concentrations. The findings revealed that the PLSR method outperformed the bagging-PLSR to predict Pb, Sb, and Cr concentration. Based on the Ratio Performance Deviation (RPD), the prediction models have moderate accuracy (RPD = 1.15–1.79) for predicting heavy metals concentration. Finally, the result of this study revealed the feasibility of hyperspectral data (350–2500 nm) to predict heavy metals concentration on the little island of Hormuz. Nevertheless, it is fundamental to compare these methods with non-linear models for their sensitivity to a limited dataset, and also, for their predictive capability.