Quantitative retrieval of soil total nitrogen content from airborne hyperspectral data combining spectral...

Soil total nitrogen (STN) plays a vital role in nutrient cycling and ecosystem restoration in mining areas. Accurate monitoring of STN in reclaimed soils is essential for assessing soil fertility and guiding ecological rehabilitation. Compared with chemical analysis, hyperspectral technology enables rapid and non-destructive prediction of soil physicochemical properties. However, the diversity of spectral transformation methods, spectral indices, and modeling algorithms poses challenges to determining the optimal combination for STN inversion. This study aimed to construct an optimal hyperspectral model for predicting STN concentration in the dumps of the Ha'erwusu open-pit coal mine in Inner Mongolia, China. Airborne hyperspectral data and 66 soil samples were collected. Four spectral transformation methods-multivariate scattering correction (MSC), standard normal variate (SNV), first derivative (FD), and second derivative (SD)-were applied to enhance spectral features. Six types of spectral indices, ratio index (RI), difference index (DI), product index (PI), sum index (SI), normalized difference index (NDI), and inverse difference index (IDI) were constructed from the transformed spectra, and a two-stage feature selection framework was implemented to select the optimal spectral indices. Four regression algorithms-back propagation neural network (BPNN), extreme learning machine (ELM), support vector machine (SVM), and random forest (RF)-were employed to build STN retrieval models. Results demonstrated that spectral transformations effectively strengthened the relationship between STN and spectra, among which SNV exhibited the highest correlation with STN. DI showed the strongest sensitivity to STN variations among all indices. Among modeling methods, RF achieved the highest accuracy. The combination of SNV transformation and RF modeling provided the optimal estimation performance (R_v² = 0.840, RMSE_v = 0.112, RPD = 2.516). These findings highlight the potential of hyperspectral technology for efficient and precise monitoring of STN content in mining reclamation areas, offering valuable technical guidance for soil quality evaluation.