Hyperspectral inversion of organic matter content in agricultural soils based on fractional-order derivative and ensemble learning

Abstract

Hyperspectral technology shows great potential for rapidly monitoring soil organic matter (SOM) in farmland. To improve prediction accuracy and model stability, this study proposes a new SOM inversion strategy based on the Stacking ensemble learning framework. By applying fractional-order derivative (FOD) processing to hyperspectral data, the spectral data quality is effectively enhanced. Feature selection was performed using the Boruta algorithm, and the modeling performance of six common machine learning algorithms, including Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), Random Forest (RF), Adaptive Boosting (AdaBoost), and Gradient Boosting Decision Tree (GBDT), was evaluated and compared. The results show that FOD outperforms traditional integer-order differential methods in modeling. Under the 1.8-order FOD processing, the prediction performance of all models improved significantly. Among them, the Stacking ensemble model combining SVR, RF, AdaBoost, and GBDT achieved the best performance on the test set, with an R² of 0.83, surpassing the R² of the best single model, GBDT (0.81). Additionally, SHAP value analysis further confirmed the effectiveness of combining FOD, Boruta, and Stacking strategies, greatly improving SOM prediction accuracy. This study offers a novel and effective technical approach for the rapid and accurate inversion of SOM content in farmland soil.