Prediction of Soil Properties Using Vis-NIR Spectroscopy Combined with Machine Learning: A Review

Area of Science:

• Agricultural Science
• Soil Science
• Spectroscopy

Background:

• Stable crop yields depend on accurate soil nutrient diagnosis (nitrogen, phosphorus, potassium).
• Traditional soil analysis is slow, complex, and lacks real-time data.
• Visible-near-infrared (Vis-NIR) spectroscopy provides a rapid, non-destructive alternative for soil analysis.

Purpose of the Study:

• To review the application of Vis-NIR spectroscopy for evaluating soil properties.
• To explore the potential of Vis-NIR for real-time field applications.
• To highlight the role of spectral preprocessing and machine learning in improving accuracy.

Main Methods:

• Utilizing Vis-NIR spectroscopy to estimate soil properties (water content, organic carbon, nutrients).
• Applying spectral preprocessing techniques to mitigate artifacts (noise, baseline drift, scatter).
• Employing machine learning algorithms (PLSR, SVMR) to enhance spectral data analysis and prediction accuracy.

Main Results:

• Vis-NIR spectroscopy, when combined with appropriate preprocessing and machine learning, can accurately estimate soil nutrient levels.
• Spectral preprocessing is crucial for addressing data artifacts and improving model performance.
• Machine learning models effectively capture complex patterns in spectral data for enhanced nutrient estimation.

Conclusions:

• Vis-NIR spectroscopy, enhanced by spectral preprocessing and machine learning, shows significant potential for real-time soil property evaluation.
• This approach facilitates more efficient and site-specific nutrient management.
• It contributes to the advancement of sustainable agricultural practices through improved soil analysis.