Predicting Soil Organic Carbon from Sentinel-2 Imagery and Regional Calibration Approach in Salt-Affected Agricultural Lands: Feasibility and Influence of Soil Properties

Mapping Soil Organic Carbon (SOC) at a regional scale is essential for assessing soil health and supporting sustainable land management. This study evaluates the potential of using Sentinel-2 imagery and regional calibration to predict SOC in salt-affected agricultural lands in Portugal while also assessing the influence of soil properties, such as texture and salinity, on SOC prediction. A per-pixel mosaicking approach was set to analyze the relationship of spectral reflectance indices linked to bare soil conditions with SOC. SOC prediction models were developed using linear regression (LR) and Partial Least Squares Regression (PLSR). Among the tested approaches, the combination of the maximum Bare Soil Index (maxBSI) with LR produced the most accurate SOC predictions, achieving moderate prediction performance (R2 = 0.52; RMSE = 0.16%; LCCC = 70%). This approach slightly outperformed the application of the 90th percentile of bare soil pixels (R90 reflectance) and the median approaches with PLSR. Notably, our findings indicate that soil salinity did not significantly affect SOC predictions within the observed salinity range of ECe between 1.2 and 10.4 dS m−1 in topsoil. However, further case studies are needed to validate this observation across diverse agricultural conditions. In contrast, soil texture and moisture content emerged as the dominant factors influencing soil reflectance. The combination of per-pixel mosaicking and regional calibration provides a practical, scalable, and cost-effective method for generating SOC maps using open access satellite imagery. To support wider adoption and improve model generalizability, future studies should incorporate a larger number of fields with a wider range of soil properties, crop types, and management practices.