Rapid Estimation of Soil Copper Content Using a Novel Fractional Derivative Three-Band Index and Spaceborne Hyperspectral Data

Rapid and large-scale monitoring of soil copper levels enables the quick identification of areas where copper concentrations significantly exceed safe thresholds. It allows for selecting regions that require treatment and protection and is essential for safeguarding environmental and human health. Widely adopted monitoring models that utilize ground- and uncrewed-aerial-vehicle-based spectral data are superior to labor-intensive and time-consuming traditional methods that rely on point sampling, chemical analysis, and spatial interpolation. However, these methods are unsuitable for large-scale observations. Therefore, this study investigates the potential of utilizing spaceborne GF-5 hyperspectral data for monitoring soil copper content. Ninety-five soil samples were collected from the Kalatage mining area in Xinjiang, China. Three-band indices were constructed using fractional derivative spectra, and estimation models were developed using spectral indices highly correlated with the copper content. The results show that the proposed three-band spectral index accurately identifies subtle spectral characteristics associated with the copper content. Although the model is relatively simple, selecting the correct fractional order is critical in constructing spectral indices. The three-band spectral index based on fractional derivatives with orders of less than 0.6 provides higher accuracy than higher-order fractional derivatives. The index with spectral wavelengths of 426.796 nm, 512.275 nm, and 974.245 nm with 0.35-order derivatives exhibits the optimal performance (R2 = 0.51, RPD = 1.46). Additionally, we proposed a novel approach that identifies the three-band indices exhibiting a strong correlation with the copper content. Subsequently, the selected indices were used as independent variables to develop new spectral indices for model development. This approach provides higher performance than models that use spectral indices derived from individual band values. The model utilizing the proposed spectral index achieved the best performance (R2 = 0.56, RPD = 1.52). These results indicate that utilizing GF-5 hyperspectral data for large-scale monitoring of soil copper content is feasible and practical.