Soil Organic Carbon Estimation Using NIRS and MIRS Spectroscopy with Machine Learning as a Statistical Tool in the Senegalese Peanut Basin: A Rapid Approach for Sustainable Soil Management

Abstract

Senegalese agriculture relies heavily on peanut cultivation, but agricultural intensification has led to soil degradation and a decline in fertility. Soil organic carbon (SOC) is a key indicator of soil quality, influencing its structure and fertility. However, conventional SOC analysis methods are costly and time-consuming. Infrared spectroscopy (NIRS and MIRS) offers a fast and non-destructive alternative, allowing SOC estimation based on the soil’s spectral properties. The study, conducted in the Senegalese Peanut Basin, involved the analysis of 240 soil samples at two depths (0–10 cm and 10–30 cm). Spectra were acquired using NIRS and MIRS, then calibrated with reference measurements obtained through CHNSO analysis. Various spectral preprocessing techniques (SNV, SG, MSC, etc.) and machine learning models (PLSR, SVM, Random Forest, XGBoost) were tested to optimize SOC prediction. The results show that the SVM and Random Forest models offer the best performance, particularly with NIRS spectra preprocessed using Savitzky-Golay, achieving a coefficient of determination (R²) above 0.8 and an RPD > 2, indicating sufficient accuracy for soil management applications. This study highlights infrared spectroscopy as a promising tool for the rapid and cost-effective mapping of SOC, contributing to improved agricultural soil fertility management.