Inversion of soil properties in rare earth mining areas (southern Jiangxi, China) based on visible-near-infrared spectroscopy

Purpose Traditional measurement for soil properties is time-consuming and costly, while visible-near-infrared spectroscopy enables the rapid prediction of soil properties. In this study, visible-near-infrared spectroscopy was used to predict these four soil properties including OC (organic carbon) content, TN (total nitrogen) content, pH value, and clay content in rare earth mining areas based on different spectral transformation and calibration methods. Materials and methods A total of 232 soil samples were collected from unexploited, in situ leaching, and heap leaching mining areas in southern Jiangxi Province, China. The chemical properties and reflectance spectra of air-dried samples were measured. Spectral transformations including first-order derivative (FOD), continuum removal (CR), and continuous wavelet transform (CWT) were selected to improve the prediction accuracy of the model. Partial least-squares regression (PLSR), the support vector machine (SVM), and extreme gradient boosting (XGBoost) were used to construct prediction models. Results and discussion The highest prediction accuracies in terms of the coefficient of determination (R-2), root mean square error (RMSE), and relative prediction deviation (RPD) were obtained using CWT spectra with XGBoost for organic carbon content (R-2 = 0.89, RMSE = 0.24, RPIQ = 4.67), total nitrogen content (R-2 = 0.86, RMSE = 0.01, RPIQ = 4.14), and pH value (R-2 = 0.73, RMSE = 0.19, RPIQ = 1.66). The best prediction result for clay content was obtained using CWT spectra with the SVM (R-2 = 0.67, RMSE = 6.45, RPIQ = 2.75). Conclusions The CWT coupled with a non-linear model, such as XGBoost, is an effective method for the accurate prediction of soil properties in rare earth mining areas.