Soil organic carbon estimation using remote sensing data-driven machine learning.

Area of Science:

• Environmental Science
• Soil Science
• Remote Sensing

Background:

• Soil organic carbon (SOC) is vital for the global carbon cycle and ecosystem health.
• Understanding SOC distribution is crucial for effective land management and carbon balance.

Purpose of the Study:

• To assess surface SOC content in Shandong Province based on land use types.
• To explore the spatial distribution patterns and influencing factors of SOC.
• To compare the performance of machine learning models for SOC estimation.

Main Methods:

• Employed machine learning algorithms: Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Machine (SVM).
• Integrated diverse data: sample data, remote sensing, socio-economic, soil texture, topographic, and meteorological data.
• Validated model performance using coefficient of determination (R²), root mean square error (RMSE), and relative percentage difference (RPD).

Main Results:

• The XGBoost model demonstrated superior prediction accuracy (R² = 0.7548, RMSE = 7.6792, RPD = 1.1311).
• Elevation (21.74%) and clay content (13.47%) were the most significant factors influencing SOC.
• Higher SOC content was observed in mountainous forest regions compared to plains and coastal areas.

Conclusions:

• Machine learning models, particularly XGBoost, are effective for estimating surface SOC content.
• Elevation and soil clay content are key drivers of SOC spatial distribution in Shandong.
• Findings provide valuable insights for land use planning and SOC conservation strategies.