Digital mapping of soil inorganic carbon content and density in soil profiles after ‘Grain for Green’ program

Abstract

Soil inorganic carbon (SIC) is vital for terrestrial carbon reservoirs and the global carbon cycle. Understanding its spatial distribution is essential for environmental management and climate change mitigation. However, there remains a significant gap in predicting the spatial distribution of SIC content (SICC) and density (SICD), and our comprehension of the combined influences of natural factors and human activities on SIC is limited. This study in the Loess Plateau aimed to predict the spatial distribution of SIC content and density using data from 142 soil profiles and environmental covariates. We evaluated random forest (RF), support vector machine (SVM), and Cubist models for their predictive performance using metrics like coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). Landscape analysis revealed that land use significantly impacts both horizontal and vertical distributions of SICC and SICD, with leaching being a critical factor. Terrain attributes influenced these patterns by affecting sunlight exposure and hydrothermal conditions. Remote sensing technologies proved valuable for predictions. RF outperformed SVM and Cubist, yielding robust results for SICC (R²: 0.317–0.514, RMSE: 1.386–4.194 g/kg, and MAE: 1.045–2.940 g/kg) and SICD (R²: 0.282–0.490, RMSE: 0.220–1.069 kg m⁻², and MAE: 0.174–0.772 kg m⁻²). RF was used to estimate total SIC stocks at 286.92 × 10⁶ kg, with 49 % found in the 100–200 cm layer, underscoring the carbon sequestration potential of deeper soils. These insights are crucial for policymakers to understand SIC variability and inform sustainable land management strategies.