Satellite Soil Observation (SatSoil): extraction of bare soil reflectance for soil organic carbon mapping on Google Earth Engine

Accurate mapping of soil organic carbon (SOC) is essential for managing agricultural land and mitigating climate change. However, available bare soil reflectance extraction methods from satellite imagery are limited by vegetation, crop residues, and cloud cover. This study introduces SatSoil, an innovative, multi-temporal approach that uses optical remote sensing to isolate bare soil pixels. This approach combines two novel techniques: Consecutive Differential Series (CDS) and the Crop Residue Mitigation Index (CRMI). Based on the principle that soil reflectance increases with wavelength, CDS effectively isolates bare soil pixels from Landsat-8 imagery (2013–2023) over the study area (i.e., Germany). CRMI reduces interference from crop residues by analyzing differences in NIR and SWIR bands. Validation using Canonical Correlation Analysis revealed stronger correlations in the visible bands between satellite and laboratory measurements. The K-means train-test split was used to address the skewed distribution of SOC for stable predictive accuracy using Random Forest Regression (RFR). RFR models achieved R2 values of 0.90, 0.72, and 0.39 for LUCAS-2015, SatSoil, and GEOS3, respectively, with corresponding RMSE values of 17.84, 16.02, and 6.62 g/kg. SatSoil achieved 19.4% greater bare soil coverage than GEOS3, significantly improving satellite soil reflectance accuracy and enhancing SOC mapping for agricultural management.