Monitoring soil organic carbon and pH in organic soils using VIS–NIR spectroscopy under field and laboratory conditions

Organic soils store a disproportionately large share of terrestrial soil carbon and play a key role in climate change mitigation. However, their high spatial variability and sensitivity to sampling and preparation procedures make routine monitoring of soil organic carbon (SOC) and related properties challenging. In this study, we evaluated the potential of visible–near infrared (VIS–NIR, 350–2500 nm) spectroscopy for the assessment of SOC and pH in organic soils under both field-moist and laboratory-dried conditions.

A dataset of more than 300 organic soil samples, including peat, muck, gyttja, and mineral–organic soils, was collected from reference soil profiles across Poland at two depth intervals (0–20 cm and 40–60 cm). Spectral measurements were acquired using a PSR-3500 spectroradiometer. Principal component analysis (PCA) was used to explore spectral variability among soil types, while partial least squares regression (PLS) and support vector machine (SVM) models were developed for SOC and pH prediction. Model performance was evaluated using independent validation datasets.

PCA revealed clear separation of major organic soil groups, reflecting differences in organic matter composition and degree of decomposition. SOC prediction accuracy was consistently higher for models developed on dried samples, while models based on field-moist samples showed reduced but still informative performance. Among the tested approaches, SVM generally outperformed PLS for SOC prediction, although model performance varied depending on soil type and calibration subset. Predictions of soil pH were less accurate than those for SOC but captured broad trends relevant for monitoring applications.

Overall, the results indicate that VIS–NIR spectroscopy provides a robust and non-destructive tool for SOC and pH assessment in organic soils, particularly under standardized (dried) conditions. While moisture effects remain a limitation for field-moist measurements, the approach shows strong potential for supporting soil carbon monitoring and digital soil assessment frameworks in natural and agroecosystems.