Simulation of Soil Organic Carbon and Inorganic Carbon in the Upper Yellow River Basin Based on Random Forest-Kriging Model and Their Driving Mechanisms

Abstract Soil organic carbon (SOC) and soil inorganic carbon (SIC) are fundamental components of the soil carbon pool, exerting critical influences on ecosystem carbon balance and climate regulation. Yet, their depth-dependent spatial distribution and controlling mechanisms remain insufficiently understood. In this study, we integrated SoilGrids 250 m version 2.0 data with a Random Forest–Kriging model to investigate SOC and SIC in the surface (0–30 cm) and subsurface (30–100 cm) layers of the Upper Yellow River Basin and to identify their dominant environmental drivers. Both SOC and SIC generally declined from southwest to northeast, with subsurface SIC hotspots shifting northward and low-value clusters concentrated in the southwest. In the surface layer, SOC and SIC were positively associated with elevation, soil total nitrogen, and mean annual precipitation (p < 0.001). These associations weakened with depth; notably, SIC lost its response to elevation and showed a negative correlation with cation exchange capacity (CEC) in subsurface soils. High-elevation, cool–humid conditions and nutrient accumulation promoted SOC and SIC in surface soils, whereas subsurface SIC was primarily regulated by leaching and mineral adsorption. The depth-dependent influences of soil moisture (WV0010) and CEC further highlighted the stronger pedochemical control of SIC in subsurface layers, while SOC drivers remained relatively stable with depth. These findings enhance the understanding of stratified soil carbon processes and provide a scientific basis for targeted carbon management and sequestration strategies in cold–arid regions.