Environmental Controls of Soil Inorganic and Organic Carbon Revealed Through Multivariate Analysis and Machine Learning in the Eastern Black Sea Region of Türkiye

ABSTRACT

Understanding the controls of soil organic carbon (SOC) and soil inorganic carbon (SIC) in humid mountainous landscapes is challenging because of strong topo‐climatic and land‐use heterogeneity. This study assessed SOC, SIC, soil pH and the SIC/SOC ratio across 5964 km ² in the Eastern Black Sea region of Türkiye using 1310 topsoil samples collected at 0–30 cm depth. Soil observations were combined with climatic, topographic, hydrological, remote‐sensing, land‐use and legacy soil covariates. The study area was stratified into three physiographic datasets, and Random Forest (RF) was integrated with Principal Component Analysis (PCA), Structural Equation Modelling (SEM) and SHapley Additive exPlanations (SHAP) analysis to combine spatial prediction with process interpretation. SOC and SIC were governed by contrasting environmental controls. SOC was mainly associated with climate, vegetation, moisture and erosion–deposition gradients, whereas SIC was more strongly linked to soil pH, hydro‐climatic conditions and landscape position. Precipitation affected the SIC/SOC ratio nonlinearly, but threshold behaviour was region‐specific: significant breakpoints occurred for the complete dataset (~1426 mm) and Dataset III (~1434 mm), while transitions in the other datasets were not significant. SEM explained 23%–27% of SIC variability and 7%–15% of SOC variability, with the climate‐pH‐SIC pathway most clearly supported in Dataset III. RF achieved high validation accuracy for SOC ( R ²  = 0.89–0.93), SIC ( R ²  = 0.79–0.92) and the SIC/SOC ratio ( R ²  = 0.90–0.98), whereas pH predictions were weaker ( R ²  = 0.45–0.63). The integrated framework showed that SOC and SIC should be treated as distinct but complementary carbon pools in digital soil mapping and regional carbon management.