Development of a soil heavy metal estimation method based on a spectral index: Combining fractional-order derivative pretreatment and the absorption mechanism.

Visible and near-infrared (Vis-NIR) reflectance is an effective way to estimate soil heavy metal content. In this study, in order to magnify the spectral information of the soil heavy metals and solve the collinearity and redundancy of hyperspectral datasets, we aimed to explore the potential of the fractional-order derivative (FOD) spectral pretreatment method and the band combination algorithm in soil heavy metal estimation. A total of 120 soil samples were collected in Xuzhou city, Jiangsu province, China, and their heavy metal contents and spectra were measured. The FOD (intervals of 0.25, range of 0-2) and a new three-band spectral index which take into account the electronic transition of metal ions in the visible region and organic matter and clay minerals in the near-infrared region were utilized for the spectral pretreatment and the selection of characteristic bands, respectively. FOD with an order of 0.75 exhibited the best model performance for estimating Cr and Zn, yielding RP2 values of 0.74 and 0.81, respectively. As regards Pb, the highest estimation accuracy was achieved with the 0.5-order reflectance, yielding RP2 values of 0.56. The three-band spectral indices with the best performance were then combined for a better estimation. To improve the estimation accuracy and generalization, partial least squares (PLS), support vector machine (SVM), random forest (RF), ridge regression (RR), XGBoost and extreme learning machine (ELM) were used to estimate the heavy metals by incorporating multiple spectral indices, and it was found that ELM outperformed other counterparts (the highest RP2 = 0.77 for Cr, the highest RP2 = 0.86 for Zn, the highest RP2 = 0.63 for Pb). The main spectral absorption mechanisms and modes of heavy metals were also analyzed. This estimation method combining FOD and a three-band index will provide a reference to estimate soil heavy metals using Vis-NIR spectra over a large scale.