Estimation of soil organic carbon using NIR spectroscopy in soils cultivated with sugarcane (Saccharum officinarum)

Several studies have demonstrated the role of sugarcane-based systems soil organic carbon (SOC) dynamics and storage potential, positioning sugarcane cultivation as a strategic component in soil sustainability initiatives. From this perspective, agriculture not only fulfills a productive role but also offers key opportunities for the restoration of degraded soils and the transition toward resilient food systems, particularly in tropical regions subjected to intensive agricultural use and high edaphic variability. In this context, near-infrared (NIR) diffuse reflectance spectroscopy emerges as a rapid, non-destructive technique with high potential for SOC estimation. The objective of this study is to evaluate the certainty and performance of different spectral models for SOC estimation in agricultural soils of the Cauca River Geographic Valley, an important region for sugarcane production in Colombia. The study was conducted on Ingenio Risaralda farms, located in the departments of Valle del Cauca, Caldas, and Risaralda, following a standardized soil spectroscopy workflow for spectral preprocessing, modeling, and validation. The NIR spectra were processed and modeled using partial least squares regression (PLSR), Random Forest (RF), and Cubist. Preliminary results indicate superior performance of the RF model (R² ≈ 0.70, RMSE ≈ 0.16), with acceptable predictive ability in external validation, whereas PLSR showed overfitting and Cubist exhibited intermediate behavior. These findings highlight the importance of non-linear approaches for capturing the edaphic complexity of tropical sugarcane systems. The complete analysis of the planned samples will allow a robust assessment of the potential of NIR spectroscopy for SOC monitoring in tropical agricultural systems.