Optimization of UAV Hyperspectral Retrieval of Soil Copper (Cu) Content Using NRBO-XGBoost

To explore the feasibility of using low-altitude unmanned aerial vehicle (UAV) hyperspectral remote sensing to estimate soil copper (Cu) content in contaminated areas and to provide technical support for the rapid assessment and remediation of contaminated sites.

An abandoned non-ferrous metal processing plant in Yinzhou District, Tieling City, Liaoning Province, was selected as the study area. A quantitative inversion method for soil Cu content was developed by integrating spectral transformation, feature-band selection, and spectral indices. Three modeling strategies were compared: four two-dimensional spectral indices (2D), four three-dimensional vegetation spectral indices (3D SI), and four newly constructed three-dimensional spectral indices (3D BI). The Newton-Raphson optimized eXtreme Gradient Boosting (NRBO-XGBoost) algorithm was used to establish the Cu content inversion model.

Spectral transformation significantly improved the correlations between spectral variables and Cu content. The first-order derivative transformation yielded the strongest correlation, with ABI3 (850, 885, 520) exhibiting an r value of -0.570. Under the same preprocessing conditions, the 3D vegetation spectral indices and newly constructed 3D spectral indices showed better correlations with Cu content and better modeling performance than the 2D spectral indices. The NRBO-XGBoost model combined with the FD-3D SI dataset achieved the best performance (R² = 0.933, σ_RMSE = 75 mg·kg^-1), substantially outperforming the conventional XGBoost model (R² = 0.833, σ_RMSE = 116 mg·kg^-1).

The UAV hyperspectral estimation model based on NRBO-XGBoost can rapidly estimate soil Cu contamination and generate regional distribution maps. These results demonstrate the potential of UAV hyperspectral remote sensing for monitoring soil heavy metal contamination and provide an effective technical approach for the efficient assessment and remediation of contaminated sites.