Prediction of lithium content in typical mountainous clay in Xinjiang, China using fractional derivatives and feature extraction

Lithium (Li) is an ‘emerging’ environmental pollutant, especially in soil, which is a great concern because it can endanger human health through the food chain. Compared with traditional chemical analyses, hyperspectral techniques have achieved many exciting results in soil metal monitoring due to their advantages of being fast and non-destructive. However, insufficient attention has been paid to lithium in soil, and the feasibility of its estimation using hyperspectral techniques needs to be investigated. We studied 97 soil samples from clay-type lithium mines in the Ertanggou area of the East Tianshan Mountains of Xinjiang to explore the effects of spectral resolution, fractional order derivatives (FOD), and characteristic band selection on the estimation accuracy of clay Li content, to obtain a fast and effective method for estimating clay Li content. Finally, we developed a new method for rapid and non-destructive estimation of soil lithium content. We have obtained some important results from the study. Spectral resolution exerts a significant impact on model performance, and its reduction usually leads to a decline in model performance. For the full band, the models constructed with low-order derivatives were superior to those with high-order derivatives, and the best model was obtained at the 0.4-order derivative (coefficient of determination (R²) and relative predictive deviation (RPD) of 0.777 and 2.118, respectively). In the characteristic bands, the lower order is sensitive to the visible-near-infrared range, and the higher order is sensitive to the short-wave infrared range, and the model constructed with the higher-order derivatives outperforms the lower-order derivatives. In this study, the combination of FOD and Random Forest (RF) can significantly improve the model performance, with R², Relative Root Mean Squared Error (RRMSE), and RPD being 0.849, 1.526, and 2.574, respectively. Therefore, this research provides a theoretical basis and technical reference for imaging hyperspectral exploration of anomalous areas of clay-type Li resources.