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

Area of Science:

• Environmental Science
• Remote Sensing
• Soil Science

Background:

• Soil total nitrogen (STN) is crucial for nutrient cycling and ecosystem restoration in mining areas.
• Accurate STN monitoring in reclaimed soils is essential for assessing fertility and guiding ecological rehabilitation.
• Hyperspectral technology offers rapid, non-destructive soil property prediction, but optimal model selection for STN inversion remains challenging.

Purpose of the Study:

• To construct an optimal hyperspectral model for predicting STN concentration in reclaimed soils of the Ha'erwusu open-pit coal mine.
• To evaluate the effectiveness of different spectral transformation methods, spectral indices, and regression algorithms for STN estimation.

Main Methods:

• Airborne hyperspectral data and 66 soil samples were collected.
• Four spectral transformation methods (MSC, SNV, FD, SD) and six spectral indices (RI, DI, PI, SI, NDI, IDI) were applied.
• A two-stage feature selection framework and four regression algorithms (BPNN, ELM, SVM, RF) were used to build STN retrieval models.

Main Results:

• Spectral transformations enhanced the STN-spectra relationship, with SNV showing the highest correlation.
• The Difference Index (DI) demonstrated the strongest sensitivity to STN variations.
• The Random Forest (RF) algorithm achieved the highest prediction accuracy.

Conclusions:

• The optimal hyperspectral model, combining SNV transformation and RF modeling, achieved excellent STN estimation performance (R_v² = 0.840, RMSE_v = 0.112, RPD = 2.516).
• Hyperspectral technology shows significant potential for efficient and precise STN monitoring in mining reclamation areas.
• This study provides valuable technical guidance for soil quality evaluation in post-mining landscapes.