Exploring soil multi-parameter stacking measurement through Raman and NIR dual-spectroscopy.

Area of Science:

• Agricultural Science
• Analytical Chemistry
• Spectroscopy

Background:

• Excessive fertilizer use leads to increased costs, soil degradation, and environmental contamination.
• Current soil testing methods are time-consuming, analyzing only one parameter at a time.
• A need exists for rapid, simultaneous multi-parameter soil analysis to optimize fertilizer application.

Purpose of the Study:

• To develop a novel method for simultaneous measurement of five key soil indicators: hydrolyzed N, available P, quick-release K, organic matter (OM), and pH.
• To create an optimized spectral feature fusion model using near-infrared (NIR) and Raman spectroscopy.
• To establish a robust stacking algorithm model for accurate soil parameter prediction.

Main Methods:

• Fusion of near-infrared (NIR) and Raman dual-spectral features.
• Exploration of seven preprocessing techniques and four feature extraction methods.
• Development of a stacking algorithm model integrating five base learners (RF, LR, SVR, ELM, PLS).

Main Results:

• The stacking model achieved high prediction accuracy for all five soil parameters, with R_p² values ranging from 0.9557 to 0.9966.
• Root Mean Square Error of Prediction (RMSEP) values were low, indicating precise measurements.
• The model demonstrated strong generalization, noise resilience, and robustness, outperforming mono-spectral and dual PLS models.

Conclusions:

• The developed stacking model offers a rapid, cost-effective, and precise solution for online soil physicochemical condition evaluation.
• This approach enables simultaneous soil parameter measurement, facilitating concurrent ploughing, surveying, and fertilizer application.
• The study provides a theoretical foundation for enhancing online monitoring systems in modern agriculture.