Evaluation of spectral data based soil organic carbon content estimation models in VIS-NIR

The precision agriculture is a key management tool for food security, requiring rapid and cost-effective field assessment methods to support agricultural decision-making. One solution is proximal sensors that collect spectral characteristics of soils. The objective of this study was to establish models for predicting soil organic carbon (SOC) content using VIS-NIR (400–2500 nm) spectroscopy as an alternative to destructive SOC measurement methods. For the modelling, 90 soil samples were collected from 0–0.20 m depth representing the most common soil types in the Northern Great Plain (Hungary), 60 soil samples were used for calibration, and 30 soil samples for model validation. The soil samples were evaluated both chemically and physically. The estimation models were fitted based on spectral indices, and the spectral bands used for indexing were identified by principal component analysis (PCA) of reflectance. Based on the PCA results, four SOC models were set up with moderately good coefficients of determination (R²=0.47–0.61). The results demonstrated that VIS-NIR spectroscopy (especially NIR) based organic carbon content estimation models are suitable for rapid estimation of soil SOC%. This can reduce sampling costs by optimizing the number of samples to be sent to the laboratory and by identifying heterogeneous patches in the study area.