Predicting soil organic carbon with ensemble learning techniques by using satellite images for precision farming

Area of Science:

• Agricultural Science
• Environmental Science
• Data Science

Background:

• Soil composition is vital for agriculture, with soil organic carbon (SOC) being central to soil health and sustainable practices.
• Accurate SOC assessment enables informed farming decisions, optimizing crop growth and resource management.

Purpose of the Study:

• To develop a machine learning (ML) system for evaluating soil organic carbon (SOC) using topographic, remote sensing, and climate data.
• To compare the performance of different ML algorithms, including XGBoost, Random Forest, and Stacking, for SOC prediction.

Main Methods:

• Utilized satellite-derived topographic variables, remote sensing indices, and climate data as predictor variables.
• Employed soil health card data as the dependent variable for model training and validation.
• Applied and evaluated XGBoost, Random Forest, and Stacking ensemble methods for SOC estimation.

Main Results:

• XGBoost demonstrated superior performance with the highest R-squared (0.95) and lowest RMSE (0.03) during training.
• For the testing dataset, XGBoost and Random Forest showed varying performance metrics, with Stacking effectively mitigating overfitting.
• The developed system aims to provide precise fertilizer recommendations, thereby enhancing crop yield.

Conclusions:

• Machine learning techniques applied to remote sensing data offer a robust approach for building decision support systems in precision agriculture.
• The proposed system can significantly assist farmers in optimizing agricultural inputs and improving overall farm productivity.
• Accurate SOC monitoring through ML is crucial for advancing sustainable agriculture and ensuring food security.